Accumulating (dis)advantage? Institutional and financial aid pathways of Latino STEM baccalaureates

PDF

Download

Open document

Flip pages

Summarize

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content ACCUMULATING (DIS)ADVANTAGE? INSTITUTIONAL AND FINANCIAL AID PATHWAYS OF LATINO STEM BACCALAUREATES by Lindsey Ellen Malcom A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (EDUCATION) August 2008 Copyright 2008 Lindsey Ellen Malcom ii DEDICATION To my parents iii ACKNOWLEDGEMENTS I would like to begin by acknowledging my dissertation chair, Alicia C. Dowd, and my committee members, Estela M. Bensimon and Harry Pachon. Thank you for your feedback and guidance. Your dedication to equity in educational access and outcomes is an inspiration. Thank you to Alicia for your support, encouragement and patience throughout this journey. Your thoughtful questioning has truly made this study better. I am grateful for the opportunities I have had to work with you and learn from you. Thank you to Estela for being a wonderful mentor and friend. From my first day at the Center for Urban Education, you have treated me like a valued colleague. Thank you for having confidence in me—even when I did not. The level of trust you have continually put in me has helped me grow as a person and as a scholar. We have come a long way since the Figaro Café! I am grateful for the guidance of Professors Marvin Titus, Rebecca Callahan, Tatiana Melguizo and Dominic Brewer. Thank you to my past and present colleagues at the Center for Urban Education. I have learned so much from each you. I am indebted to Arlease Woods who taught me how to show grace under pressure. I am also thankful for the support of Edlyn Vallejo Peña, Amalia Marquez, Dafne Espinoza, Frank Harris III, and Elsa Macias. Thank you to my “cohortians,” Jarrett Gupton, Sean Early, Margaret Sallee, Hyo Jin Lim, and Vicki Park. From Catalina Island, to cookie parties, to our dissertation support group meetings—you have made my doctoral experience unforgettable. I would also like to thank the Rossier School of Education, the Graduate School at the University of Southern California, and the ASHE/Lumina fellowship program for funding support as I completed my doctorate. iv Finally, I would like to thank my family for their unconditional love and support. To Sylvain—your reservoir of love, understanding and corny jokes has been a constant source of joy since the day we met. Thank you for having faith in me, and in “us.” Je t’aime. To Kelly—I am grateful for your patience during those days and weeks when I had no time to talk. Nonetheless, you were always there for me. Thank you. To Mimi and Daddy—for as long as I can remember, I have been in awe of your strength. You blazed this trail on which I walk. I could not have become who I am without your love, guidance, and wisdom. Thank you for believing in me. To Daddy, thank you for giving me the courage to follow my dreams. To Mimi, thank you for your endless support and invaluable advice. I dedicate this study to you. v TABLE OF CONTENTS Dedication ii Acknowledgements iii List of Tables vii List of Figures x Abstract xii Chapter 1: Introduction 1 Cumulative Advantage Theory 2 Cumulative (Dis)advantage and the Persistence of Educational 5 Inequity The Condition of Latinos in Higher Education 8 Furthering Disadvantage: Financial Aid and the Rise 11 of Student Loans Overview of the Study 14 The Importance of Increasing Latinos’ Participation in STEM 19 Organization of the Dissertation 24 Chapter 2: Review of the Literature 25 The Role of Context in College Decision-making: A Conceptual 28 Framework Selecting a Path through College: The Importance of Hispanic 37 Serving Institutions and Community Colleges Formulation of College Financing Strategies: Latino Students 52 and the Decision to Borrow Latinos Choosing and Succeeding in STEM: Factors in the 65 Institutional Environments of Community Colleges and Hispanic Serving Institutions The Decision to Attend Graduate School: Understanding the 79 Effects of Indebtedness Chapter Summary 92 Chapter 3: Research Design and Methodology 93 Research Questions 94 Research Approach 95 Data 97 Data Preparation and Management 104 Analyses 110 vi Chapter 4: Findings 130 Institutional Pathways to STEM Bachelor’s Degrees 130 for Latinos Financial Aid Pathways of Latino STEM Bachelor’s 145 Degree Holders The Effect of Indebtedness on Graduate School Enrollment 169 among Latino STEM Bachelor’s Degree Holders Summary of Key Findings 193 Chapter 5: Discussion 196 The Importance of Community College Transfer and Hispanic 196 Serving Institutions to Latino STEM Baccalaureates College Financing Strategies and Borrowing among 205 Latino STEM Bachelor’s Degree Holders 205 The Effect of Indebtedness on Graduate School Enrollment 210 Limitations of the Study 214 Directions for Future Research 216 Chapter 6: Conclusions and Implications 219 Revisiting Cumulative (Dis)advantage: A Tale of Three Students 219 Complicating our Understanding of ‘Disadvantaging’ Contexts 230 for Latinos in STEM Policy Recommendations 242 References 249 vii LIST OF TABLES Table 3.1: 2003 National Survey of Recent College Graduates (NSRCG) 100 Primary Variables of Interest Table 3.2: Demographic, Educational, B.S. Degree-granting Institutional, 108 and Financial Support Characteristics of Analytical Sample: Summary Statistics Table 3.3: Possible Responses for Cumulative Undergraduate Debt 116 Table 4.1: Institutional Pathways of Latino STEM Bachelor’s Degree 131 Holders Table 4.2: Baccalaureate-granting Institutional Characteristics of Latino 133 STEM Bachelor’s Degree Holders by Associate Degree Table 4.3: Location of Latino STEM Bachelor’s Degree Holders 135 by Associate Degree Table 4.4: Location of Latino STEM Bachelor’s Degree Holders by 136 Institutional Pathway Table 4.5: Institutional Pathways of Latino STEM Bachelor’s Degree 137 Holders by National Origin Table 4.6: Demographic Characteristics of Latino STEM Bachelor’s 139 Degree Holders by Associate Degree Table 4.7: Demographic Characteristics of Latino STEM Bachelor’s 141 Degree Holders by Institutional Pathway Table 4.8: Field of Study of Latino STEM Bachelor’s Degree Holders 142 by Associate Degree Table 4.9: Field of Study of Latino STEM Bachelor’s Degree Holders 144 by Institutional Pathway Table 4.10: Forms of Financial Aid Used by Latino STEM Bachelor’s 147 Degree Holders by Associate Degree Table 4.11: Forms of Financial Aid Used by Latino STEM Bachelor’s 150 Degree Holders by Institutional Pathway Table 4.12: Relative Debt Level of Latino STEM Bachelor’s Degree 153 Holders by Associate Degree viii Table 4.13: Relative Debt Level of Latino STEM Bachelor’s Degree 154 Holders by Institutional Pathway Table 4.14: Relative Debt Level of Latino STEM Bachelor’s Degree 155 Holders by Highest Parental Education Level Table 4.15: Forms of Financial Aid Used by Latino STEM Bachelor’s 156 Degree Holders by National Origin Table 4.16: Relative Debt Level of Latino STEM Bachelor’s Degree 158 Holders by National Origin Table 4.17: Conditional Probabilities of Using Financial Aid Forms 159 by Latent Class Table 4.18: College Financing Strategies of Latino STEM Bachelor’s 162 Degree Holders by Associate Degree Table 4.19: College Financing Strategies Used by Latino STEM 163 Bachelor’s Degree Holders by Institutional Pathway Table 4.20: College Financing Strategies Used by Latino STEM 165 Bachelor’s Degree Holders by Demographic Characteristics Table 4.21: College Financing Strategies Used by Latino STEM 168 Bachelor’s Degree Holders by Baccalaureate-granting Institutional Characteristics Table 4.22: Estimates of Average Treatment Effect of Relative 170 Indebtedness on Graduate School Enrollment, for Latino STEM Bachelor’s Degree Recipients, Using Logistic Regression Table 4.23: Logistic Regression Model of Graduate School Enrollment 171 including Financial Aid Indicator Variables for Latino STEM Bachelor’s Degree Holders Table 4.24: Logistic Regression Model of Graduate School Enrollment 173 including Latent College Financing Strategies, for Latino STEM Bachelor’s Degree Holders Table 4.25: Constrained Multinomial Probit Model of Relative 176 Indebtedness (by Level) for Latino STEM Bachelor’s Degree Holders ix Table 4.26: Cumulative Distribution of Estimated Propensity Scores 180 for Borrowing at a Low Relative Debt Level (Derived from Constrained Multinomial Probit Regression Model in Table 4.25) Table 4.27: Cumulative Distribution of Estimated Propensity Scores 182 for Borrowing at a High Relative Debt Level (Derived from Constrained Multinomial Probit Regression Model in Table 4.25) Table 4.28: Estimates of the Average Treatment Effect (ATE), Average 186 Treatment on the Treated (ATT), and Average Treatment on the Untreated (ATU) Effect of Relative Level of Indebtedness (Low, High) on 2000-01/2001-02 STEM Bachelor’s Degree Recipients Graduate School Enrollment, Based on Matching of Propensity Scores Derived from the Multinomial Probit Model in Table 4.25 Table 4.29: Estimates of the Treatment Effects of Relative Level of 190 Indebtedness on Graduate School Enrollment Table 4.30: Results of Mantel-Haenszel Bounds Sensitivity Analysis 192 Table 6.1: (Dis)advantaging Institutional Pathways for Latinos in STEM 240 x LIST OF FIGURES Figure 1.1: The Mechanism of Cumulative (Dis)advantage 4 Figure 1.2: The Intensification of Educational Inequities among Latinos, 2004 11 Figure 1.3: Latinos’ Share of the U.S. Population by Age, 2004 21 Figure 2.1: Perna’s Conceptual Model of Student College Choice 33 Figure 2.2: (Dis)advantaging Contexts Framework 36 Figure 2.3: Financial Aid Usage of Latino Undergraduate Students 57 by Institutional Sector Figure 2.4: Latinos’ Participation Rates in Student Loan Programs by 59 Institutional Sector and National Origin Figure 3.1: Pathways to the STEM Baccalaureate for Latino Students 111 Figure 3.2: College Financing Strategy Precedes Financial Aid Indicator Variables 113 Figure 3.3: Distribution of Relative Debt Dosage Level by Associate Degree 118 Figure 3.4: Model of Graduate School Enrollment for Logistic Regression Analysis 120 Figure 3.5: Covariates included in Constrained Multinomial Probit Model 125 of Relative Level of Indebtedness Figure 3.6: Illustration of Application of Propensity Score Matching 126 Techniques to Estimate the Effects of Undergraduate Loan Debt on Graduate School Attendance Figure 4.1: Conditional Probabilities of Use by Financial Aid Support Mechanism 160 Figure 4.2: Estimated Propensity Scores of Borrowing at a 184 Low Relative Debt Level, by Treatment Figure 4.3: Estimated Propensity Scores of Borrowing at a 185 High Relative Debt Level, by Treatment Figure 6.1: Inez’s Story: Accumulating Advantage and Disadvantage 234 Figure 6.2: Alex’s Story: Accumulating Advantage and Disadvantage 236 Figure 6.3: Eva’s Story: Accumulating Advantage and Disadvantage 238 xi ABSTRACT This study contributes to our understanding of the institutional and financial aid pathways traversed by Latina/o bachelor’s degree holders in fields related to science, technology, engineering, and mathematics (STEM), and the relationships between those pathways and selected outcomes of interest. Using the sociological framework of cumulative advantage/disadvantage and the statistical techniques of latent class analysis (LCA) and propensity score matching (PSM), data from the 2003 National Survey of Recent College Graduates (NSRCG) enhanced with institutional information drawn from the College Board Survey of Colleges and Universities, Barron’s Selectivity Index, and the Integrated Postsecondary Education Data System (IPEDS) were analyzed to describe the means through which Latina/o STEM baccalaureates accessed those degrees, characterize their college financing strategies and to understand how indebtedness, measured relative to typical levels of borrowing within the B.S. degree-granting institutional context, influences these students’ decisions to attend graduate school. The analytical sample reflects those Latinas/os who earned a bachelor’s degree in a STEM field during the 2000-01 or 2001-02 academic years from a postsecondary institution in the mainland United States. Three primary conclusions can be drawn from this study. First, there is an association among socioeconomic status, state policy context, and the pathways traversed by Latina/o STEM degree holders, particularly with respect to associate degree attainment and attendance at Hispanic-serving institutions. Second, this study finds that among Latina/o STEM bachelor’s degree holders, three college financing strategies were employed (self support, parental support, and balanced support), and that these strategies were associated with certain institutional pathways. Third, this study finds that low and high levels of indebtedness as measured relative to average borrowing at the B.S.-degree granting institution negatively affected graduate school xii attendance among Latina/o STEM bachelor’s degree holders, with the magnitude of the negative effect larger for those who borrowed at relatively high levels. 1 CHAPTER ONE INTRODUCTION “So he was nice,” I whispered to my mom as we walked down the quiet third floor of the Infinite Corridor. “Of course,” she responded. “All of my friends are nice.” I turned my head so she couldn’t see me roll my eyes. But for all of my exasperation with my mother, I was grateful that she brought me to Cambridge to meet Charles Vest, the president of my first choice college, MIT. The noise of the students hurrying to class grew louder as we descended from the top floor, which was dedicated to administrative offices. Moving my plastic MIT Coop bookstore bag to my other hand, I clutched the banister to steady myself as I teetered down the staircase in the 2-inch heels that my mother insisted I wear. Immediately after returning to home, I taped the physics major requirements and the felt MIT banner that I collected as souvenirs to the back of my bedroom door. In the weeks and months to follow, I gazed at these relics of my trip when I should have been doing my Calculus III homework. I was stricken with a serious case of ‘senioritis’ and was anxious to begin the next chapter of my life at MIT. My dad gave me the run down on the courses I would need to take: special relativity, statistical mechanics, quantum physics…the list went on. I was excited, but it all sounded so hard. When I shared my fears with my dad, he told me that he was sure I could do it. Perhaps seeing that I wasn’t satisfied with just verbal reassurances, he also said that though it had been awhile since he had taken those courses, I could call him for help if I had any questions on my problem sets. His concrete offer of assistance eased my anxiety, and in the years to come, I would cash in on it more times than I can remember. I often joke that I was conditioned from birth to be a scientist. My most memorable childhood Christmas gifts consisted of books about electricity, a chemistry set, and a microscope. Each Easter, my family and I sojourned to the Air and Space Museum instead of church. Science- themed birthday parties, during which my dad discussed surface tension as my friends and I blew bubbles, were an annual tradition. Today, long after I left science to study higher education policy, that MIT banner remains on the back of my bedroom door where it serves as a salient reminder of the advantages my parents imparted to me throughout my life in the form of social, cultural, and economic capital—particularly when I was making decisions about college. 2 I am the exception. Unlike many other underrepresented minority students who study the sciences, the advantages accumulated by my parents placed me into particular contexts that smoothed my pathway to study science, technology, engineering and mathematics (STEM). For example, I was aware from an early age what I needed to do to attend a selective institution because of the familiarity of the higher education system that my parents gained through their own college experiences. I was fortunate to attend well-funded public schools with wide-ranging curricula including Advanced Placement courses. While applying to college, my parents helped me to identify sources of financial aid, and were able to pay for costs not covered by scholarships so that I did not have to take out loans. I turned to my parents and their social networks to seek out advice regarding my major and undergraduate research opportunities. These experiences enabled me to attend graduate school at an equally selective institution, which will result in future advantages in terms of my career, professional and social networks, and earning potential. Cumulative Advantage Theory By sharing my personal story, I wish to illustrate the ways in which the advantaging contexts in which I was situated led to future advantages in terms of my college decision-making, experiences and outcomes. This process, in which “initial comparative advantages of trained capacity, structural location, and available resources make for successive increments of advantage such that the gaps between the haves and the have-nots…widen” (Merton, 1988, p. 106), is the basis of cumulative advantage theory. Though cumulative advantage theory was originally conceptualized to explain the hierarchal nature of the scientific community (Merton, 1968, 1988), the theory has been invoked to explain the persistent educational inequities among underrepresented minority students (Blau & Duncan, 1967; DiPrete & Eirich, 2006; Pollock, 2008). 3 In sociological applications of the theory, cumulative advantage is a mechanism that propagates and furthers inequality between groups over time. Due to historical and ongoing discrimination and denial of opportunity, certain groups maintain advantages in a variety of social metrics (e.g., wealth, educational attainment, health). These advantages result in rewards, which can then be turned into resources to acquire subsequent advantages (DiPrete & Eirich, 2006). This cycle continues, resulting in growing inequality between groups across generations, with the passage of time. In the sociohistorical context of the United States, race/ethnicity acts as a primary dimension along which inequality is perpetuated. Thus, generally, Whites maintain advantages over Latinos, Blacks, and Native Americans. These advantages grow over time, resulting in widening racial/ethnic inequities in educational and economic attainment. Sociological applications of cumulative advantage theory also introduce the idea of cumulative disadvantage (Blau & Duncan, 1967; DiPrete & Eirich, 2006) to reframe the increasing inequalities between whites and historically disenfranchised groups as a deliberate process during which minorities are placed in disadvantaging contexts that prevent the acquisition of rewards, thereby limiting access to resources. The lack of resources available to minorities confines them to these disadvantaging contexts causing them to accumulate further disadvantages over time (DiPrete & Eirich, 2006) and constraining future educational access and opportunity. The processes by which individuals accumulate advantage and disadvantage, referred to hereafter as cumulative (dis)advantage, are illustrated in Figure 1.1. 4 Figure 1.1. The Accumulation of Advantage and Disadvantage Educational researchers have used multiple theoretical lenses to understand the root of stratification in educational opportunity and outcomes. Like cumulative (dis)advantage theory, theories of human capital (Becker, 1993), social (Coleman, 1990; Portes, 1998) and cultural capital (Bourdieu, 1986) focus on access to resources as an explanatory factor for disparities in the economic and social outcomes of individuals. Human capital theory posits that differences in the education, ability, and information, i.e., human capital, possessed by individuals are responsible for differences in economic attainment (Becker, 1993). Social capital theory (Coleman, 1990; Portes, 1998) focuses on the role of resources available to someone through the relationships with members of his or her social networks in contributing to inequality (Portes, 1998). Cultural capital is complementary to social capital theory and looks to an individual’s Accumulation of Disadvantage Accumulation of Advantage Initial Disadvantages (Parents’ educational attainment, SES, social networks, etc.) Lack of Resources Disadvantages time Initial Advantages (Parents’ educational attainment, SES, social networks, etc.) Resources Awards/ Advantages time Source: Merton (1968, 1988) and DiPrete & Eirich (2006). 5 social status as signified by the non-economic goods, knowledge or skills he or she possesses as a contributing factor to inequality (Bourdieu, 1986). Though theories of human, social, and cultural capital differ in terms of the level of critique lobbed at the status quo, each theory claims that the distribution of capital determines the structure and function of the social world, with the maldistribution of economic, social and cultural capital acting as the source of social and educational inequality (Bourdieu, 1986). Though related, the concept of cumulative (dis)advantage differs from the more commonly invoked theories of human, social, and cultural capital. Cumulative (dis)advantage describes the process by which human, social, cultural, and other forms of capital function to further inequality over time. The forms of capital operate within the mechanism of cumulative (dis)advantage in that these forms of capital are resources available to the advantaged that are mobilized to bring about future advantages and increases in capital (DiPrete & Eirich, 2006; Elman & O’Rand, 2004). Thus, the constructs of human, social, and cultural capital are not competing frameworks for cumulative (dis)advantage theory; rather, the forms of capital are factors that fuel the process of cumulative (dis)advantage with which inequality persists. Cumulative (Dis)advantage and the Persistence of Educational Inequity Cumulative (dis)advantage theory predicts the persistence of educational inequity among racial/ethnic groups. While college enrollment rates of historically disenfranchised groups have increased over the past several decades, the gaps between underrepresented minorities and their White and Asian counterparts in terms of access, persistence, and degree attainment continue to grow (Bowen & Bok, 1998; Cabrera, Burkum, & La Nasa, 2005; Cabrera, La Nasa, & Burkum, 2001; Castellanos, Gloria, & Kamimura, 2005; Fry, 2002; Swail, Cabrera, Lee, & Williams, 2005). In science and engineering, these inequities are even more pronounced. For example, while Latinos constituted 11.5% of total undergraduate enrollment in 2004, they earned just 7.5% 6 of all bachelor’s degrees and 6.3% of bachelor’s degrees in science, mathematics and engineering awarded in the same year (National Science Foundation [NSF], 2005). The inequitable representation of historically disenfranchised groups among science, technology, engineering, and mathematics (STEM) majors and degree earners are attributable to a number of factors, but the framework of cumulative (dis)advantage provides a means for understanding the broader picture of the systematic exclusion of minorities, and Latinos in particular, from STEM fields. Rather than take the deficit-minded view that minorities are underrepresented in STEM fields because they lack certain social, economic, and cultural advantages, I posit that Latinos and other minorities accumulate disadvantages due to barriers constructed by discrimination, racism, an inequitable K-12 school system, racially biased college entrance requirements, and rising college costs, among other factors. For example, a student attending an urban, public high school might accumulate disadvantages due to the larger class sizes, limited access to certain resources, or the lack of availability of AP courses. These disadvantages can limit Latinos’ exposure to science at a young age, their participation in advanced-level science courses in high school, as well as access to higher education, and STEM fields in particular. By the time Latino students are college-aged, the disadvantages they have accumulated because of their lower position in society’s hierarchy and resulting context in which they are located reduces the chances of earning a STEM baccalaureate. The research literature reflects the stratified nature of postsecondary educational access provided to students of color, particularly Latinos (Cabrera, Burkum, & La Nasa, 2005; Cabrera, La Nasa, & Burkum, 2001; Fry, 2002, 2004; Swail, Cabrera, Lee, & Williams, 2005). Latinos are concentrated heavily in the community college sector and in lower status and less selective colleges and universities—institutional contexts considered to be disadvantaging (NCES, 2005). 7 Latinos are also underrepresented in STEM fields, which are considered to be higher status disciplines (Schwab, 1978) and are advantaging in terms of economic outcomes. Though previous research does not explicitly connect the mechanism of cumulative (dis)advantage to the underrepresentation of Latinos in STEM, I began to consider the ways in which cumulative (dis)advantage shape access for Latinos in STEM after my serendipitous meeting with Ericka 1 , a Latina electrical engineering student. Ericka, now working on her engineering degree at a highly selective research university, transferred to her current institution from an urban, Hispanic-serving community college. Growing up in a culturally-rich, but economically-poor urban area, she was an academically strong student who earned good grades in high school. Ericka was unable to enroll in her first choice college despite being accepted because her high school lost its accreditation. Left with no other choice but to go to community college, Ericka decided to work towards transferring to the nearby state university. She became interested in engineering after stumbling across a meeting of her community college’s Mathematics Engineering and Science Achievement (MESA) program. Ericka confessed that she did not know what engineering was until her involvement in MESA. Yet despite her initial unfamiliarity, Ericka’s interest in engineering grew and she eventually accumulated enough credits to transfer. She was extremely wary of applying to her current university due to her apprehensions about the cost, and only did so at the insistence of her MESA program advisor. Ericka’s transition to the university was extremely rough, and she continually questioned her decision to abandon the original plan of attending the less selective state university. She was caught off guard by the workload and the sheer amount of time she needed to spend studying. Ericka tried to adjust, studying for hours in between work and class, all the while suppressing her 1 Pseudonym is used to protect personal identity. 8 guilt for not spending much time with her family. Ericka struggled in large part because she was unable to find a support network after transferring and was forced to work on her own. She attributes much of her difficulty transitioning to her current institution to not knowing the “rules of the game”; she had to learn things that other students took for granted such as how to study, how to develop an educational plan, how to get an internship, or how to complete financial aid forms. Ericka overcame these difficulties and is currently one semester away from graduation. However, the pathway to her STEM degree was markedly different from the route that I traveled. While Ericka is certainly a success story, it is important consider the role that cumulative disadvantage played in her educational experiences. Ericka’s story is likely more typical than mine; yet much of the research on Latinos in STEM fails to consider the ways in which accumulated disadvantage situates these students in unique contexts that constrain their educational opportunities. Recent work (Contreras, Malcom & Bensimon, 2008; Stage & Hubbard, 2008) begins to focus on the linkages between cumulative disadvantage and Latinos’ access to STEM, but additional research is needed. The present study represents a first step in addressing this gap in the literature by investigating the institutional and financial aid pathways used by Latino STEM bachelor’s degree holders and exploring the role of contextual factors in shaping these students’ patterns of access, financing strategies and graduate school enrollment. The Condition of Latinos in Higher Education While Ericka’s story illustrates the ways in which cumulative disadvantage can constrain educational opportunity and shape college experiences, an examination of the patterns of educational access and outcomes of Latinos in higher education suggest that cumulative (dis)advantage is at work, contributing to growing inequality. It is certainly encouraging that Latino enrollment in colleges and universities more than doubled since 1990, going from 782,400 in 1990 to 1,809,600 in 2004 (National Center of Education Statistics [NCES], 2005). However, 9 this growth in college enrollment has not kept pace with Latinos’ growing share of the U.S. college-aged population. In 2004, Latinos constituted 18.6% of 18-24 year olds (U.S. Census Bureau, 2004), but only 11.5% of total postsecondary enrollment (NCES, 2005). While the continual addition of immigrants to the Latino population may account for some of this disparity, this factor cannot explain away the entire gap. The lower college matriculation rates among Latino high school graduates also account for the inequities in terms of enrollment. Fourteen percent of high school diploma recipients were Latino in 2004, but Latinos only constituted 9.5% of college first-time freshmen in fall 2005 (Sable & Garofano, 2007). Latino students enroll in college at rates lower than those of Whites, Asians, and African Americans, are more likely to delay college entrance, and are more likely to enroll part-time (Arbona & Nora, 2007; Castellanos, Gloria, & Kamimura, 2005; Fry, 2002; NCES, 2005; Swail et al., 2005), which have been shown to be detrimental to persistence and degree attainment (Adelman, 2005; Arbona & Nora, 2007; Cabrera, Burkum, & La Nasa, 2005; Cabrera, La Nasa, & Burkum, 2001). In sum, while Latinos’ access to postsecondary education has increased over the past few decades, disturbing inequities in enrollment and matriculation rates persist. Although more Latinos have gained access to some sort of postsecondary education, the nature of that access is also important to consider. Latinos are more likely to attend community college than individuals from other racial/ethnic groups (Adelman, 2005), and 58.3% of all Latinos enrolled in postsecondary education attend a community college (NCES, 2005). Those Latino students who are enrolled in four-year institutions are concentrated in Hispanic-serving Institutions (HSIs), which tend to be less selective, non-research colleges and universities (NCES, 2006). Although only about 8% of postsecondary institutions are HSIs, nearly half (48%) of Latino students enrolled in institutions of higher education attend HSIs (NCES, 2002; Brown & Santiago, 2004). While these institutions provide access to Latinos, they are considered to be of 10 lower status than more selective research universities. In this sense, community colleges and HSIs could be considered to be disadvantaging contexts, and may be contributing to the accumulation of disadvantage by Latino students. Disparities in the degree attainment rates of Latinos and whites also reflect the ongoing accumulation of disadvantage by this demographic group. The 6-year graduation rate of Latinos who began college in 1998 was 47.6%—nearly 12 points lower than that of Whites. These lower degree attainment rates reduce the pool of Latinos eligible to go on to graduate study, making the underrepresentation of Latinos even more severe among master’s, doctoral and professional degree holders. The figure below (Figure 1.2) employs the Center for Urban Education’s equity index (Bensimon, 2004; Hao, 2006) to illustrate the growing gaps between Latinos’ representation in the population and their share of a number of educational indicators, including STEM degree attainment. The equity index (Hao, 2006) is a measure of proportionality that establishes how far or how close a particular group (Latinos in this case) is from reaching representation on a particular indicator of attainment that is equal to their representation in a specified population pool. If the equity index equals one for a given indicator, Latinos have achieved equity for that educational outcome. An equity index below one indicates underrepresentation, and equity index greater than one indicates overrepresentation. The U.S. college-aged (18-24) population serves as the baseline population for the figure. Although Figure 1.2 was constructed using cross-sectional data, it clearly illustrates the underrepresentation of Latinos among high school graduates, four-year and four-year college students, STEM bachelor’s degree holders and STEM doctorates. Additionally, the figure demonstrates that the degree of the inequity becomes more severe with each subsequent educational outcome. 11 Figure 1.2. The Intensification of Educational Inequities in the Latino Population, 2004 Furthering Disadvantage: Financial Aid and the Rise of Student Loans The disadvantages experienced by Latinos are often exacerbated by the high cost of postsecondary education and the barriers these students face in securing financial aid. Changes in federal, state and institutional financial aid policies over the past two decades reflect the move away from the higher education as a public expenditure to a shared cost model (Heller, 2002; Price, 2004; St. John, 2003). While the availability of financial aid has significantly expanded during this period, much of this expansion has been driven by student loans (College Board, 2006). Unlike grant financial aid, which reduces college costs for students and their families, loans actually increase the price of college due to accruing interest. Though student loans are a 0.80 0.22 0.34 0.42 0.56 1.00 0 0.2 0.4 0.6 0.8 1 College-Aged Population High School Graduates 2-year College Enrollment 4-year College Enrollment STEM Bachelor's Degrees STEM Doctorates Educational Outcome Equity Index Equity = 1.00 Source: American Community Survey, 2004; National Center for Education Statistics, 2005; National Science Foundation, 2004. 12 common college financing strategy for many students, research suggests that the potential and actual burden of student loan debt may affect college choice, educational attainment and professional outcomes among minority and low-income students (Perna & Titus, 2004; Price, 2004; Zarate & Pachon, 2006). Federal appropriations for Pell grants and other grant programs have remained stagnant over the past decade, while financial aid via loans (subsidized and unsubsidized) has increased (College Board, 2006). At the state level, governors and legislatures continue to reduce funding to postsecondary institutions. These changes in federal and state policy occur against a backdrop of skyrocketing college tuitions for public and private institutions and rising student enrollments. Lost amidst these sweeping changes have been students and parents who are expected to shoulder an increasingly large share of college costs. The purchasing power of Pell grants and state grants has declined over the past few years. The percentage of tuition, fees, room, and board at the average public four-year college covered by the maximum Pell Grant declined from 42% in 2001- 02 to 33% in 2005-06 (College Board, 2006). While institutional grants have taken up some of the slack, in many cases students are left with no choice but to turn to loans to finance their college educations. The increased role of loan financing is evident when comparing grant aid and loan aid as a percent share of total aid for undergraduate students. In 1991-92, grants from all sources constituted 57% of total aid while loans represented 41% of total aid. By 2005-06, loans far outpaced grants in terms of share of total aid: loans comprised 52% share total aid compared to a 42% share of total aid for grants. Institutional policies often exacerbate students’ dependence on loans. As tuition rates continue to rise, loans have become of the centerpiece of financial aid packages offered to matriculated students, particularly at private institutions (College Board, 13 2006). Furthermore, as students progress through college, they are often required to take out larger loans each year as their institutional grant aid declines. The increased reliance on loans acts to limit access to higher education for low-income and minority students, particularly Latinos, as these populations participate in loan programs at lower rates compared to their White, middle-class counterparts. Latinos use loans to finance college costs less often than any other racial/ethnic group. Recent studies (Fry, 2002; Santiago & Cunningham, 2005; Zarate & Pachon, 2006) have shown that while Latino students feel that a college education is important, the debt levels associated with attending college significantly influence the educational experiences of these students (Burdman, 2005). Many researchers (Burdman, 2005; De La Rosa & Hernandez-Gravelle, 2007; Monaghan, 2001) have argued that Latino students, more so than students from other racial/ethnic backgrounds, elect not to use loans to finance their college educations, and may opt to alter their college choices to attend less expensive, and generally less selective colleges and universities (Kurlaender, 2006; Person & Rosenbaum, 2006). Indeed, Latino college students are concentrated in low-cost institutions: 40% of Latinos in higher education were enrolled in institutions with tuition and fees between $0 and $1,000 and 36% enrolled at institutions with tuition and fees between $1,000 and $5,000. These lower cost institutions are largely community colleges and less selective, non-research four-year institutions. The concentration of Latino students in these institutions may go on to affect graduate school enrollment and the labor market returns of postsecondary education (Burdman, 2005). In sum, shifts in financial aid policy at the federal, state and institutional level act to limit access to higher education for Latinos and may also hinder their educational attainment. As tuitions rise and the purchasing power of grant aid decreases, students are left with no choice but to turn to loans to finance college or attend less expensive, and often less selective institutions. In 14 light of Latinos’ low participation rates in student loan programs (Burdman, 2005; College Board, 2006; Dowd, in press; Price, 2004), the increased role of loans may widen the gaps in educational and economic attainment between Latinos and their White counterparts. Overview of the Study Latinos’ current level of participation and achievement in higher education reflect severe inequities. I have argued that cumulative (dis)advantage (Merton, 1968, 1988; DiPrete & Eirich, 2006) serves as a means for understanding the persistence of gaps in educational attainment and the underrepresentation of Latinos in STEM fields. Fortunately, cumulative (dis)advantage theory provides an “out,” through which the accumulation of disadvantages can be overcome. Cumulative (dis)advantage theory is not deterministic; Merton (1988) explains that “countervailing processes” can dampen the widening of the gap between the haves and have-nots (p. 606). Returning to the application of cumulative disadvantage to Latinos in STEM, despite numerous obstacles, there are Latinos who successfully earn undergraduate and graduate degrees in the sciences and engineering. Ericka is one such student. She was able to overcome her cumulative disadvantage via individual characteristics such as high aspirations (Adelman, 2005; Pascarella & Terenzini, 1991) and persistence (Arbona & Nora, 2007), and external factors such as institutional support (Dowd, Bensimon et al., 2006; Rendón, 1994; Stanton-Salazar, 2001; Valenzuela, 1999), and familial and peer support (Tierney & Venegas, 2006). Many researchers (Arbona & Nora, 2007; Bensimon et al., 2004; Cabrera, Burkum, & La Nasa, 2005; de los Santos, 2005; Fry, 2002; Gándara, 1998; Gándara, Larson, Mehan & Rumberger, 1998; Rendón, 1992, 1995; Rendón, Jalomo, & Nora, 2000; Swail et al., 2005) continue to study the factors that aid attainment among Latino college students, enabling them to overcome cumulative disadvantages, yet few studies focus on Latinos in the sciences, mathematics and engineering. 15 In this dissertation, I address the educational problem of the underrepresentation of Latinos in STEM by investigating the educational pathways utilized by Latino STEM bachelor’s degree holders to better understand the role of community colleges and Hispanic Serving Institutions (HSIs) in the production of Latino scientists and engineers. I focused on community colleges as the entry point into the STEM pipeline because Latinos are most likely to start their education in the two-year college sector compared to other racial/ethnic groups (Adelman, 2005; Cabrera, La Nasa & Burkum, 2001; Fry, 2002; Phillippe & González Sullivan, 2005). Further, Latinos are highly concentrated in community colleges that are designated as Hispanic Serving Institutions (HSIs) by the U.S. Department of Education. Nearly 60% of Latinos in the American higher education system are enrolled in a community college; and 40% of Latinos enrolled in four-year institutions attend HSIs. Recent data from NSF shows that nearly 44% of all STEM bachelor’s degree holders attend community college at some point in their career (Tsapogas, 2004), though it is unknown what proportion of these STEM degree holders actually transfer from the community colleges to four-year institutions. Community colleges and HSIs present an interesting challenge to our understanding of the ways in which Latino students accumulate advantage and disadvantage. For example, community colleges and four-year HSIs can be viewed as disadvantaging environments for Latino students because these institutional types are less selective (non-selective in the case of community colleges) and typically lower funded than more selective four-year institutions. However, Latinos who attend these institutions certainly benefit from that attendance in the form of increased earnings and job opportunity relative to those who do not pursue postsecondary education. Disentangling the various ways in which cumulative (dis)advantage are associated with particular institutional pathways, and the manner in which certain institutional pathways bring about additional advantage or disadvantage is a central aim of this study. 16 In this study, I focused on Latino STEM degree holders who receive an associate degree from the community college prior to earning the baccalaureate. While a large proportion of Latinos transfer without earning an associate degree, recent changes in state transfer policies may change this trend. Policymakers in Florida, and more recently Virginia, offer incentives to students who earn the associate degree prior to transferring (Florida Department of Education, n.d.; Virginia Community College System, 2008). These incentives include additional financial aid and guaranteed admission to public four-year institutions. From the perspective of the policy makers, increasing the number of students who earn the associate degree prior to transfer will enhance the efficiency of a state’s higher education system. However, the implications of such policies are unclear for Latinos in STEM. This study will contribute to our understanding whether the baccalaureate-degree granting institutional characteristics of Latinos who earn an associate degree differ from non-associate degree earners. I also examined the differences in educational outcomes of Latino STEM bachelor’s degree holders by institutional pathway. By investigating the differences in outcomes between those students who earned the associate degree and non-associate degree earners, and those that graduate from HSIs versus non-HSIs, I was able to understand the ways in which various institutional contexts advantage and disadvantage Latino STEM degree holders. In the present study, I investigated the college financing strategies employed by Latino STEM B.S. degree holders and their borrowing behaviors in an effort to critically examine the common characterization of Latino students as “debt averse” (Baker & Velez, 1996; Burdman, 2005; De La Rosa & Hernandez-Gravelle, 2007; ECMC Group Foundation, 2003; Monaghan, 2001; Nora & Horvath, 1989; Olivas, 1985, 1986). I also determined the effects of student loan debt on the graduate school enrollment of recent Latino STEM bachelor’s degree earners. Studying the institutional pathways and college financing strategies of Latinos who successfully 17 attained a bachelor’s degree in science, mathematics, engineering or a related field provides a better picture of the institutional characteristics associated with overcoming cumulative disadvantage. Similar to institutional types, student loan debt can bring about disadvantages and advantages to students. The mixed findings in the literature regarding the effects of debt (Dowd, 2008) underscore the complicated ways in which debt can be advantaging or disadvantaging for different students within various institutional contexts. This study aims to bring about more clarity to the nature of debt for Latinos in STEM fields. Further, investigating the effect of educational debt on graduate school enrollment for Latino STEM bachelor’s degree earners enriches our knowledge regarding debt as a potential barrier to advanced study in the sciences and engineering. In particular, I address the research questions below. Importance of Community College Transfer and Hispanic Serving Institutions to Latino STEM Baccalaureates (1) Nationally, what proportion of Latino STEM baccalaureate recipients earned associate degrees prior to transfer to the four-year sector? (2) What are the institutional characteristics of the four-year institutions from which Latino associate degree earners graduate, according to HSI status, selectivity, type (e.g. liberal arts or research university), and control (public/private)? (3) How do these characteristics compare with the institutions attended by Latino STEM bachelor’s degree holders who did not earn an associate degree? (4) How does the distribution of students across these institutional characteristics vary by state and region and by Latino ancestry for Mexican American/Chicano, Puerto Rican, and Cuban students? 18 (5) How do demographic characteristics of Latino STEM baccalaureate degree holders who earn associate degrees differ from non-associate degree earners? (6) And, what is the distribution of Latino associate degree holders across different STEM fields of study in comparison to those students who do not earn an associate degree prior to attending a four-year college? College Financing Strategies and Effects of Borrowing on Graduate School Enrollment (1) How does the use of various forms of college financial aid and financing strategies such as scholarships, loans (institutional and familial), college work study, earnings, and employer support vary by institutional pathway, national origin, and other demographic characteristics among Latino STEM bachelor’s degree holders? (2) What are the effects of debt on graduate school enrollment among Latino STEM bachelor’s degree holders? To address the research questions outlined above, I analyzed the 2003 National Survey of Recent College Graduates (NSRCG) 2 database enhanced with institutional-level data from the 2002-2003 College Board Annual Survey of Colleges and Universities, the Integrated Postsecondary Education Data System (IPEDS), and Barron’s Profiles of American Colleges and Universities. Incorporating multiple data sources enabled me to consider both individual-level and institutional-level variables in my analyses. The conceptual framework for the study draws upon sociocultural, social psychological and economic theories regarding college and financial aid decision-making. The statistical model includes variables related to enrollment in graduate school for Latino STEM bachelor’s degree holders at the institutional and individual level. I employed descriptive analyses and the 2 The use of NSF data does not imply NSF endorsement of the research, research methods, or conclusions contained in this report. 19 multivariate statistical techniques of latent class analysis, logistic regression analysis and propensity score matching (PSM) to address the research questions. At the descriptive level, chi- square tests were used to identify the differences in the educational pathways used by Latino STEM bachelor’s degree holders with different individual characteristics (e.g., national origin, gender, socioeconomic status), and the characteristics of four-year degree-granting institutions attended by Latinos who earn an associate degree from a community college compared to non- associate degree earners. I used latent class analysis in concert with descriptive techniques to characterize the college financing strategies of Latino STEM bachelor’s degree holders and assess the differences in the strategies used in particular institutional contexts. The logistic regression analysis allowed me to isolate the effect of cumulative undergraduate loan debt on the decision to enroll in graduate school among Latino STEM bachelor’s degree holders after controlling for the influence of other contextual factors. Finally, I used propensity score matching techniques to model the effects of educational debt on graduate school enrollment for Latino students by simulating an experimental design. I compared the results obtained using propensity score matching to the results of the regression analysis in order to evaluate whether PSM represents an improvement over standard estimation procedures. The Importance of Increasing Latinos’ Participation in STEM While Latinos, African Americans, and Native Americans are all severely underrepresented among STEM bachelor’s degree earners, the low degree attainment rate of Latinos in STEM fields is particularly significant due to the demographic reality of the country. Latinos are the fastest growing demographic group in the country and are projected to make up nearly a quarter of the entire U.S. population by the year 2050 (U.S. Census Bureau, 2004). The relatively low level of educational attainment among Latinos, particularly in high- demand fields such as science, technology, engineering and mathematics, is a multilayered 20 problem with many social and economic consequences. Not only does the current state of higher education reify long-standing inequities for Latinos, it threatens the economic health of the nation, particularly as Latinos stand to become a majority-minority in a number of southwestern states. Myers (2007) argues that as the baby boomers retire, the role of Latinos in the workforce becomes even more important. According to Myers, the continued prosperity of the nation depends on a social contract between younger, heavily Latino generations and older, predominantly White generations. While Latinos are certainly concentrated in the U.S. Southwest and California, nearly every state has experienced large gains in the proportion of Latino residents (U.S. Census Bureau, 2005). An age cohort analysis of the U.S. population clearly demonstrates that the U.S. is in the midst of a large demographic shift (see Figure 1.3). Figure 1.3 illustrates that Latinos’ share of the population increases with younger generations. If these patterns continue as projected, in the next fifty years, Latinos will constitute nearly 25% of the U.S. population. Clearly, these data demonstrate support Myers’ (2007) premise that the much of the burden for the future social and economic well being of the country will rest on the educational and occupational attainment of Latinos. 21 Figure 1.3. Latinos’ Share of the U.S. Population by Age, 2004 The U.S. economy’s growing dependence on Latinos is indeed a compelling reason to increase the Latino science and engineering workforce. However, the underrepresentation of Latinos in STEM is not merely a matter of economic importance. Increasing access and achievement for Latinos in STEM fields is also a social justice issue. Osei-Kofi and Rendón (2005) critique those who view the issue of opportunities in higher education for Latinos through the lens of capitalism. In their eyes, increasing access and attainment of Latinos is not just about fulfilling the “‘needs’ of the market” (Osei-Kofi & Rendón, 2005, p. 249), but is important to the economic and social emancipation of this historically disenfranchised group. Osei-Kofi and Source: U.S. Census Bureau, American Community Survey, 2005, Tables B01001 and B01001I. 22.3% 6.4% 7.8% 10.0% 14.6% 19.5% 18.7% 19.3% 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% Under 5 5-14 15-24 25-34 35-44 45-54 55-64 65+ Age Cohort Latinos' Percent Share of Cohort 22 Rendón (2005) warn researchers against reducing Latinos’ success in higher education to their “perceived economic and political ‘use-value’” (p. 253). Osei-Kofi and Rendón’s admonishment highlights the dual significance of the focus of the current study, as increasing access and achievement for Latinos in STEM fields is also a matter of social importance. The Heterogeneity of Latinos Although I use the term “Latino” to describe individuals of Mexican, Central American, Latin American, Dominican, Puerto Rican 3 or Cuban descent, I acknowledge the important intra- group differences among U.S. Latinos attributable to the wide-ranging historic, social, and economic diversity of this demographic group. The racial/ethnic categorization “Latino” includes recent immigrants, the children and grandchildren of immigrants, as well as those individuals whose ancestors resided in the present-day U.S. centuries prior to the country’s founding (Osei- Kofi & Rendón, 2005; Washington, 1993). As such, the unique sociohistorical and economic contexts from which Latinos of differing ancestry originate ought to be taken into account when conducting empirical analyses and developing and applying theory to members of these populations. National origin and generational status, as well as more common constructs such as socioeconomic status are important to include when theorizing or modeling the experiences of the Latino population. For example, the educational experiences and attainment of a middle-class Cuban-American student with college educated parents will not mirror that of a first-generation Chicana, despite the fact that both of these individuals are often classified using the umbrella designation “Latino.” For these reasons and to further understand intra-group differences in the educational experiences and attainment of Latino STEM B.S. degree holders, I investigated 3 My analysis neither includes Hispanic Serving colleges and universities located in the U.S. Commonwealth of Puerto Rico nor the students attending those HSIs due to the unique historical, social, and demographic context of Puerto Rico. The study does, however, include those individuals of Puerto Rican descent who reside and studied in the U.S. 23 differences in the institutional pathways and college financing strategies of Latinos by national origin to the extent that the sample sizes permitted. In addition to serving as the source of difference, the multiple histories of Latino groups in the U.S. have also shaped the geographical distribution of this population. Although Latinos are increasingly represented in the population throughout the country, there are particular areas in which Latinos are clustered. The Southwestern states of Arizona, New Mexico, Texas, and California are home to the majority of the U.S. Latino population, and these individuals tend to be of Mexican and Central American decent (e.g., Salvadorian, Guatemalan) (Hobbs & Stoops, 2002; U.S. Census Bureau, 2005). However, significant concentrations of Latinos are also found in Florida, New York, Nevada, Colorado, and Illinois, and these populations are often distributed based on national origin (U.S. Census Bureau, 2005). Thus, the geographical clustering of Latinos of particular national origins in historically and socially significant areas adds a layer of complexity to the understanding of the U.S. Latino population and their patterns of participation in higher education in general and the sciences and engineering in particular. Latinos’ increasing share of the U.S. population and the resulting reliance on this group for the nation’s continued economic prosperity signify the importance of addressing the problem of the underrepresentation of Latinos in STEM fields. Many (National Research Council, 1999; NSF, 2006) acknowledge that scientific innovation and technological advancement is central to the U.S.’s economic future. The participation of Latinos in the STEM workforce is vital to this advancement and fuels the demographic dynamism (Myers, 1999, 2007) capable of alleviating the educational and economic inequities between Latinos and their White counterparts. In order to further understand the problem of the underrepresentation of Latinos in STEM fields, it is necessary to investigate the nature of the various routes to the STEM for Latino B.S. degree 24 holders, the means by which these students finance their college education, and the effect of common financing strategies (e.g., student loans) on post-baccalaureate outcomes. Organization of the Dissertation In addition to this introduction, this dissertation consists of five chapters. Chapter two presents a review of the literature regarding the postsecondary educational experiences of Latino students in the sciences, mathematics and engineering. The literature review focuses on the institutional contexts and cultures in which Latinos STEM majors tend to succeed, drawing upon previous research on the importance of minority serving institutions to the production of minority scientists and engineers, and the ways in which financial aid, student loans in particular, impact the college experiences of Latino students and their educational outcomes. I also discuss the methodological and theoretical limitations of prior research on Latino student attitudes towards educational debt and the impact of indebtedness on educational outcomes. Chapter three details the research design, including the research questions, description of the sample for the study, the methods for addressing the research questions and the rationale for the chosen method. Chapter four presents the findings of the study, Chapter five discusses these findings, describes the limitations of the study and discusses directions for future research, and the final chapter offers implications for policy and practice. 25 CHAPTER TWO REVIEW OF THE LITERATURE In the previous chapter, I argue that accumulated disadvantage situates Latinos in particularized contexts that act to shape and often constrain postsecondary educational opportunity. This claim is bolstered by the fact that much of the literature on Latinos in higher education reveals the importance of context in understanding the cause of, and by extension, potential solutions for abating, the ongoing inequities in terms of educational access and academic success for this group. In trying to unpack the problem of the underrepresentation of Latinos in postsecondary education in general and among STEM bachelor’s degree holders in particular, previous research underscores the importance of considering the interactions between the environments in which Latinos are located and their decision-making regarding postsecondary enrollment, college financing, field of major and graduate school attendance (Padilla, 2007). Countless numbers of reports, statistical briefs, and research studies authored by educational researchers and policy experts explore the problem of the underrepresentation of Latinos in STEM fields (e.g., American Council on Education [ACE], 2006; Committee on Equal Opportunities in Science and Engineering [CEOSE], 2004; NSF, 2005). While the majority of these studies certainly acknowledge the relationship between the historical denial of opportunity of Latinos and their underrepresentation in STEM fields, often, these analyses do not consider the impact of the nested social, economic, higher education and financial aid policy, and institutional environments in which Latinos are located and their outcomes in STEM. Though the relationship between Latinos’ access to and success in STEM and their environments may seem intuitive, the nuances that the interaction of context and outcomes introduce are largely absent from previous research on Latinos in STEM. 26 For example, existing research provides limited information regarding the ways in which community colleges and Hispanic Serving Institutions (HSIs) serve as institutional pathways for Latino STEM majors, despite their pivotal role in educating Latinos. The literature also fails to address how these particular institutional contexts affect Latino students’ decision to major in a STEM field and their STEM degree attainment. Existing research on Latinos in STEM does not make linkages between the economic and financial aid policy contexts that act to shape the decisions of these students related to pathway selection and enrollment. Instead these two strands of research are separate in the literature, and questions of college financing strategies, borrowing, and the relationship between indebtedness and graduate education rarely intersect with studies of access and success for Latinos in STEM. These gaps in the literature on Latinos in the sciences, mathematics, and engineering are indicative of the myopic framework often brought to bear on studies of the underrepresentation of this particular group in STEM fields. Rather than focusing on the ways in which an individual’s educational decision-making results from interactions within his or her environment, research regarding the participation of Latinos in STEM tends to center on the characteristics of the individual (e.g., educational aspirations, interest in science, academic preparation/ability, and persistence). While these characteristics undoubtedly play a part in Latinos’ access to and success in STEM fields, an exclusive focus on the individual student without consideration of how her decision-making is shaped by her context is incomplete. Further, the failure to apply a holistic frame to research on Latinos in STEM has resulted in an incomplete understanding of the multiple barriers faced by Latino STEM majors and has prevented researchers from recognizing the ways in which nested social, economic, higher education and financial aid policy, and institutional contexts shape the educational progression of Latino STEM degree holders. It is particularly important to make these linkages when studying Latinos due to the way in which 27 accumulated disadvantage places them in unique contexts that operate synergistically to constrain their choices related to postsecondary education. The present study endeavors to broaden our understanding of the problem of the underrepresentation of Latinos in STEM fields by bridging the nested contexts in which Latinos are situated and STEM outcomes. I use a variety of quantitative methods to characterize the postsecondary institutional pathways followed by Latino STEM bachelor’s degree holders, the college financing strategies they employ, and the effect of indebtedness on their enrollment in graduate school. Applying a social ecological framework to guide my analysis, I focus on the ways in which key contextual variables shape the patterns of participation of Latino STEM bachelor’s degree holders. This study focuses on “endpoints” in the sense that I am looking backwards at those Latinos who successfully earned a STEM degree. While the data and methods used here do not permit me to delve into the everyday moments that lead to student success, studying endpoints and understanding the interactions between context and the educational progression of Latino STEM degree holders provides markers for further inquiry. In this chapter, I discuss the literature regarding the relationships between key contextual variables (e.g., socioeconomic status, higher education financial aid policy, and institutional type/mission) and the four decision-making processes at the center of the present study: (1) selection of institutional pathway into postsecondary education; (2) formulation of college financing strategies and the decision to borrow; (3) STEM major selection and degree completion; and (4) the decision to attend graduate school. These decision-making processes do not occur in a linear fashion or independently; however, for the sake of clarity of the discussion, I treat each one separately in the following sections. Before presenting the literature, I describe the conceptual framework of my study, which is based on Laura Perna’s (2006a) model of college decision-making. 28 The Role of Context in College Decision-making: A Conceptual Framework The Evolving Picture of Choice in Higher Education College decision-making – if and when to go, where to go, and how to pay – has been the subject of considerable research in the higher education literature (Hossler & Gallagher, 1987; Hossler, Schmit, & Vesper, 1999; McDonough, 1997; Paulsen, 1990; Paulsen & St. John, 2002; Perna, 2006a; Teranishi & Briscoe, 2006). As revealed by Perna’s (2006a)and Teranishi and Briscoe’s (2006) reviews of the literature on college choice, our view of college decision-making has widened since the first studies on the subject. Initially, researchers broached the issue from a human capital, rational choice perspective (Kinzie et al., 2004; Manski & Wise, 1983). Rational choice is grounded in economic theory and essentially posits that students assess the costs of higher education – both literal (i.e., monetary) and opportunity costs – and weigh those costs against the expected benefits of that education. Many researchers have attempted to use this economic analytical framework to understand how students make college enrollment decisions (see Hossler, Braxton, & Coopersmith, 1989). Critics of rational choice theory point out that this narrow economic framing of college choice makes the tacit assumption that all students have access to information about the true costs and benefits of higher education and make sense of that information in the same manner. Further, rational choice theory as applied to college decision- making fails to account for the disparate levels of access to information, resources and opportunity by race/ethnicity, socioeconomic status, and generational status (Teranishi & Briscoe, 2006). In an effort to overcome the inadequacies of the economic framing of college decision- making, Hossler and Gallagher (1987) conceptualized college choice as a three-stage process: predisposition; search; and choice. In the predisposition phase, students are exposed to the possibility or idea of attending college. During the search phase, students participate in behaviors 29 and activities that affect their chances of college enrollment, such as visiting college websites, gathering information about admissions requirements, and taking college entrance examinations. Finally, in the choice phase, students make decisions about the institutions to which they will apply. Though the three-stage framework offered by Hossler and Gallagher (1987) and later expanded upon by Hossler, Schmit, and Vesper (1999) allows us to better account for the role of information in shaping students’ college choice, it necessitates that all students follow the same decision-making process, regardless of differences in levels of accumulated disadvantage based on race/ethnicity, class, and generational status (Teranishi & Briscoe, 2006). While it is certainly true that the ascribed characteristics of race/ethnicity and SES have been included as explanatory covariates in the analyses of those employing Hossler and Gallagher’s (1987) framework, the larger implications of race, class, and parental education on the context in which students make decisions about college is not accounted for in the three-phase college choice model (Perna, 2006a; Teranishi & Briscoe, 2006). McDonough (1997) and others (e.g., Hurtado, Inkelas, Briggs & Rhee, 1997; Nora, 2004; Sewell, Hauser & Wolf, 1986; Teranishi & Briscoe, 2006) apply a sociocultural perspective to the exploration of college “choice,” focusing on the role of social and cultural capital and status attainment processes in shaping students’ educational aspirations and their college decision- making. Researchers using sociocultural theory aimed to understand the ways in which resources in the form of social networks, availability of information, and knowledge of admissions and financial aid application processes impact college enrollment. These studies also explore how disparate access to these resources within and between school and community contexts might explain educational inequities. This perspective was a valuable addition to the body of work on college choice in that it centered on the critical ways in which race/ethnicity and class intersect to perpetuate inequities in educational opportunity and attainment. In a sense, these studies’ 30 sociocultural perspective revealed that college “choice” is a misnomer, as low-income students and students of color often make constrained choices at best, and are commonly left with no choice. The evolving conceptualization of college choice from the economic theoretical perspective of rational choice, to a more nuanced view of choice as a multi-stage process (Hossler & Gallagher, 1987; Hossler, Schmit, & Vesper, 1999), to the most recent focus on the role of sociocultural constructs such as social and cultural capital (Lareau, 2003; Teranishi & Briscoe, 2006) and organizational habitus (McDonough, 1997), underscores the complexity of how and why certain decisions about college-going and college financing are made. Each of these theoretical perspectives facilitates our understanding of the factors that influence students’ college decision-making; yet none, taken on its own, provides a sufficient explanation of the ways in which individuals and their situated contexts interact to shape and constrain “college choice.” An Integrated Model of College Decision-Making For these reasons, Perna (2006a) proposes the bridging of the rational choice/economic and sociocultural perspectives so that we may understand the ways in which contextual factors at the institutional, higher education policy, and broader societal levels work in concert with individual characteristics to shape students’ college enrollment decisions. Perna (2006a) draws upon previous research to create an integrated conceptual model of college choice in which an individual’s perceptions of the costs and benefits of college and the decision-making process based on the weighing of those costs and benefits are shaped by four layers of nested contexts: (1) the individual’s habitus 4 ; (2) school and community context; (3) higher education context; and (4) social, economic, and policy context (Perna, 2006a). Perna also identifies potential factors in each 4 Habitus refers to an individual’s internalized thought patterns, behaviors, and tastes that reflect a his or her social class (Bourdieu, 1986; Bourdieu & Passeron, 1977) 31 level of context that act to shape the decisions a student makes regarding college enrollment. Perna identified the contextual factors shown in her model based what has been shown to impact college decision-making through empirical research. In the illustration of Perna’s (2006a, p. 117) proposed model shown in Figure 2.1, the innermost layer of context (Layer 1) pertains to the characteristics of the individual student, with each subsequent layer corresponding to a higher- order level of context, further removed from, and yet still affecting that individual. The first layer of the model, ‘individual habitus,’ includes the social and cultural capital possessed by a student and his or her demographic characteristics (i.e., race/ethnicity, gender, and socioeconomic status) that have been shown to influence college decision-making (McDonough, 1997; McDonough, Antonio, & Trent, 1997; Teranishi & Briscoe, 2006). School and community context (Layer 2) corresponds to the type of organizational habitus (McDonough, 1997) available to the student in the form of resources, institutional agents (Stanton-Salazar, 2001), and information about higher education. More specifically, the concept of organizational habitus refers to the college-going culture, or values that the school and community communicates around college attendance and the patterns of college-going that the culture creates (McDonough, 1997). This layer accounts for the ways in which the presence or lack of resources in the school and community setting can act to facilitate or hinder college decision-making (McDonough, 1997; Perna, 2006b; Stanton-Salazar, 2001). The third layer, higher education context, reflects the multiple ways in which the characteristics and policies of postsecondary institutions shape “college choice.” These could include the institution’s location, its mission, academic programs, and institutional reputation and prestige. Additionally, institutional policies, such as tuition levels and composition of financial aid packages offered by that institution are included in this contextual layer (Perna, 2004). The highest layer in the model contains the social, economic and policy contexts that act to shape college decision-making, for example, demographic changes, 32 shifts in the labor market, or policy changes at the state or federal level like the increasing reliance on student loans or decreasing grant aid (Perna & Titus, 2004). The arrows in Figure 2.1 indicate that within these layers of context, individual characteristics like educational aspirations and academic preparation, sociocultural factors captured by the concepts of social and cultural capital, economic factors such as perceived costs and benefits of higher education, and external factors like higher education policy, institutional characteristics, and broader social and economic considerations interact to directly and indirectly influence college enrollment decisions (Perna, 2006a). 33 Figure 2.1: Perna’s Conceptual Model of Student College Choice Social, economic, and policy context (layer 4) Demographic characteristics Economic characteristics Public policy characteristics Higher education context (layer 3) Marketing and recruitment Location Institutional characteristics School and community context (layer 2) Availability of resources Types of resources Structural support and barriers “Organizational habitus” Individual habitus (layer 1) Demographic characteristics Gender, Race/ethnicity Cultural capital Cultural knowledge Value of college attainment Social capital Information about college Assistance with college process Demand for higher education Academic preparation Academic achievement Supply of resources Family Income Financial aid Expected benefits Monetary Non-monetary Expected Costs College costs Foregone earnings College Choice Note. From “Studying College Access and Choice: A Proposed Conceptual Model,” by L.W. Perna, 2006, Higher Education: Handbook of Theory and Research, Vol. 21, p. 117. 34 In sum, Perna’s conceptual model of college choice provides a framework for understanding the role of context in college decision-making. Rather than offering a one-size-fits- all description of the process by which students make decisions regarding higher education, her model allows for, and even demands complexity based on the unique, nested environments in which a student in located. Further, the model acknowledges a student’s choice of college might be constrained due to his or her context. Although Perna’s framework was developed for the study of college enrollment decisions, her model is also applicable to other types of decision-making related to higher education. For example, concurrent to determining if and where to attend college, a student must make a decision about how to finance college. Findings from previous research reveal that multiple layers of context shape these college financing decisions as well. Similarly, a student must decide his or her field of study—a decision influenced by the nested environments in which he or she is situated. Also shaped by the layers of contexts identified in Perna’s (2006a) model is the determination whether to pursue graduate study. The notion of cumulative (dis)advantage that underpins the current study is complementary to the contextual model proposed by Perna (2006a). As I explain in the introductory chapter of this work, Latinos, as a demographic group, have accumulated disadvantage due to the historical and ongoing denial of opportunity. These disadvantages situate Latinos in particular contexts that lead to constrained educational decision-making, resulting in future disadvantages. Those contextual factors identified in each layer of Perna’s (2006a) model of college decision-making can exacerbate a student’s existing disadvantages by constraining his or her educational choices, leading to further disadvantage. For example, a low-income student who attended an underfunded, urban high school with no college-going culture might decide to delay entry to postsecondary education, which has been shown to lower the chances of attaining a 35 bachelor’s degree (Adelman, 1999). In this sense, the lack of organizational habitus within the student’s school context acts as a disadvantaging factor and leads the students to make an uninformed and constrained decision that will likely lead to future disadvantages. Alternatively, factors within the multiple contexts in which a student is located can lead to advantages in terms of educational decision-making and outcomes. For example, a middle- income student whose school incorporates preparing college applications into the curriculum and provided other resources to assist with the college enrollment process will benefit from his or her school’s organizational habitus and make more informed decisions regarding college. In this case, the culture and practices of the middle-class student’s school context is advantaging, and likely leads to future advantages in terms of college access. The two examples presented above, though simplistic, illustrate the ways in which the contextual factors identified in Perna’s (2006a) model of college decision-making can either constrain their educational choices leading to negative outcomes, or positively affect decision- making by enabling students to make informed choices that lead to more positive outcomes. In short, the factors in the four layers of context identified by Perna (2006a) can be disadvantaging or advantaging. Figure 2.2 illustrates the way in which I conceptualize the relationships between a student’s context and the four decision-making processes at the center of the present study. As reflected in the figure, four layers of context act in concert to shape Latino STEM baccalaureates’ institutional pathway, college financing strategy, field of study, and graduate school enrollment. In addition, as illustrated by the previous research, these four decision-making processes are often related to one another, as reflected by the connecting arrows in the figure. The framework of the study goes a step beyond Perna’s (2006a) model as it reflects the existence of disadvantaging and advantaging aspects within each layer of context. 36 Figure 2.2. (Dis)advantaging Contexts Framework The “disadvantaging contexts” framework illustrated above helped me to formulate research questions regarding the institutional and financial aid pathways of Latino STEM bachelor’s degree holders, to organize and synthesize the educational literature regarding the contextual factors relevant to Latinos’ patterns of participation in STEM fields, to determine relevant concepts and covariates to include in my analysis, and by serving as an interpretive lens through which I view my findings. In the third chapter of this work, I describe the ways in which I account for the layered contextual factors in my investigation of Latino STEM bachelor’s degree holders. However, in Social, Political and Economic Context Higher Education and Financial Aid Policy Context Institutional Context Individual Characteristics Institutional Pathway College Financing Strategy STEM Field of Study Graduate School Enrollment Note. Adapted from Perna (2006a). Disadvantaging Context Advantaging Context Legend 37 the remainder of the present chapter, I review the research literature regarding Latinos and the four key decision-making processes at the center of this study: (1) the selection of an institutional pathway through postsecondary education; (2) the development of a college financing strategy and the decision to borrow; (3) the decision to pursue and complete a STEM bachelor’s degree; and (4) the choice to attend graduate school. Selecting a Path through College: The Importance of Hispanic Serving Institutions and Community Colleges Over the past three decades, Latinos’ have seen a marked increase in their access to higher education, with higher proportions of Latinos enrolling in postsecondary institutions (NCES, 2002). However, there remain disturbing inequities in the patterns of this participation. The vast majority of Latinos find themselves at the lower portion of the severely stratified U.S. higher education system, concentrated in open-access community colleges and less-selective Hispanic serving institutions (NCES, 2002). Nearly 60% of Latinos are enrolled in a community college, and almost 40% of Latinos in four-year institutions attend HSIs (NCES, 2002). Previous research points to a multitude of factors that contribute to Latinos’ concentration in the community college and four-year HSI pathways. These factors, which operate within four different layers of context: (1) individual; (2) school/community; (3) higher education institutional; and (4) social, economic, and policy context, are discussed below. Contextual Factors Contributing to Latinos’ Concentration in Community Colleges Individual characteristics. Several researchers have attempted to understand why Latinos are so highly concentrated in the community college sector, and determine the reasons that lead Latinos to be more likely to begin postsecondary education in a two-year institution than any other racial/ethnic groups (Adelman, 2005). Some of these studies focus on student-level characteristics, such as academic preparation, educational aspirations, the possession of college 38 knowledge, and access to resources including social, cultural, and economic capital, as the cause of high-levels of community college attendance among Latinos. Community colleges are open-access institutions, without the battery of admissions requirements like standardized test scores and high school transcripts common to most four-year institutions. This has led some to posit that Latino high school graduates choose to enroll in community colleges because they lack adequate academic preparation to enter four-year colleges or universities (Adelman, 2005). While this seems to be a plausible explanation, researchers have found that among Latinos, academic preparation and high school achievement is not a predictor of the type of postsecondary institution in which students enroll (Kurlaender, 2006). Further, college-prepared Latinos are more likely to enroll in community colleges than students with similar levels of preparation from other racial/ethnic groups (Admon, 2006; Fry, 2004; Kurlaender, 2006). Because Latino students seem to be enrolling in community colleges despite being prepared to enter four-year institutions, it is reasonable to investigate their educational aspirations. Do Latinos who enroll in community colleges do so because they do not wish to earn a bachelor’s degree? Several researchers have sought to determine whether uncertainty in educational plans and goals (Grubb, 1991) lead to the high concentrations of Latinos in the community college sector. While it is true that Latinos tend to delay entry to postsecondary education (Rendón & Garza, 1996; Swail, Cabrera & Lee, 2004), this is not necessarily an indication of low aspirations. High proportions of Latino students who enroll in community college express a desire to earn a bachelor’s degree (Admon, 2006; Swail, Cabrera, & Lee, 2004; Kurlaender, 2006). Even among Latino students who have taken steps necessary to attend a four- year institution such as taking the SAT or ACT, community college enrollment is relatively common (Kurlaender, 2006). Though this does not constitute definitive evidence, previous 39 research indicates that Latino students enroll in community colleges at higher rates than students from other racial/ethnic groups despite being academically prepared, having high educational aspirations, and fulfilling admissions requirements for four-year institutions, such as taking college entrance examinations. Other researchers have focused on issues of class and socioeconomic status to understand the reasons that lead Latinos to enroll in community colleges in such high proportions. Socioeconomic status is typically correlated with the possession of college knowledge, including information about the advantages of attending four-year institutions, financial aid information, as well as access to financial resources (McDonough, 1997; Price, 2004; Tierney & Venegas, 2006; Venegas, 2006; Price, 2004). For example, transfer rates among all community college students are low (Dougherty & Kienzl, 2006; Grubb, 1991; Goldrick-Rab, 2006), which poses a potential barrier for students who wish to earn a bachelor’s degree. It is possible, however, that first- generation students might not be aware of the potential of being “diverted” from achieving their educational goals while at a community college (Goldrick-Rab, 2006). This has led some to posit that if Latino students, who tend to be first-generation students more so than their white counterparts (Stanton-Salazar, 2001), knew that direct entrance to a four-year institution was advantageous to their educational attainment, they might bypass the community college sector altogether. Community colleges are the lowest-priced postsecondary option available, with average tuition of $1,977, compared to an average in-state tuition of $3,400 for public four-year institutions (NCES, 2005). This cost difference between the two-year institutions and public, four-year colleges and universities has led researchers to explore the impact of price on college enrollment decisions (Paulsen & St. John, 2002; Perna, 2004; St. John, 2001; St. John, Paulsen, & Carter, 2005). Though most of these studies focus on all students’ enrollment decisions, they have 40 found that price is of great concern, particularly among Latinos (Immerwahr, 2003; Tomás Rivera Policy Institute, 2004; Zarate & Pachon, 2006). Kurlaender’s (2006) study of Latinos enrolling in community colleges revealed that while socioeconomic status was a predictor of community college enrollment, among Latinos, whites, and Blacks of the same SES, the Latino students were much more likely to enroll in a community college. Further, even Latinos from affluent households enrolled in community colleges more commonly than whites and Blacks from high SES backgrounds. Thus, while the low tuition levels of community colleges can account for the some of the attraction to the two-year sector, price does not fully explain why Latinos attend community colleges in such high proportions. The impact of price on the decision to attend community colleges ought not be divorced from a discussion of the well-documented informational barriers to financial aid access among Latino students (Admon, 2006; McDonough, 2004; McDonough & Calderone, 2006; Tierney & Venegas, 2006; Tomás Rivera Policy Institute, 2004; Tornatzky, Cutler, & Lee, 2002; Venegas, 2006). These barriers act to limit Latino students’ awareness of and access to financial aid that could defray college costs. Financial aid and perceptions of college costs go hand in hand, and Admon’s (2007) extensive study of the factors that lead to such high levels of community college attendance among Latinos illustrated that a low level of awareness of financial aid is associated with community college enrollment. The factors that lead to Latinos’ patterns of financial aid are discussed later in this chapter; however, the entanglement of access to information about financial aid and the decision to attend community college bears mentioning here. School and community context. In addition to exploring the individual-level characteristics that might lead Latino students to enroll in community colleges in such high proportions, researchers have studied factors in the school and community contexts that cause community college attendance. In particular, secondary school quality (Karabel, 1972), and 41 resources (Morrow & Torres, 1995) have been identified as two high school characteristics that can function to perpetuate and reproduce educational inequities among low-income students of color. Latinos on average, attend schools within districts that are underfunded and experience wide disparities in terms of instructional resources, college course offerings (e.g., Advanced Placement courses), quality of teachers, and class size. These disparities limit the students’ opportunities and exposure to college preparatory curricula and may serve to funnel Latino students into the community college sector (Goldrick-Rab, 2006; Kurlaender, 2006; Morrow & Torres, 1995; Person & Rosenbaum, 2006; Venezia, Kirst, & Antonio, 2003). Latino students in lower quality schools may also suffer the consequences of inadequate levels of college counseling, including incomplete or inaccurate information about college (McDonough, 1997; Tierney & Venegas, 2006). In urban high schools, which enroll large numbers of Latino students, student-to-counselor ratios are quite large. Thus, Latino students in these high schools, many of whom would be first-generation college students (Stanton-Salazar, 2001), simply cannot get the type of counseling necessary to inform them of the range of available college options and the advantages and disadvantages of each. Even those school environments within which counseling is available, the danger of the under-utilization of high school counselors remains. This phenomenon is particularly common among first-generation college students, who tend to rely on the advice of parents or other adults in the community to inform their college decision-making (Fuerstenau, 2002). As a result, first-generation Latino students are susceptible to making ill-informed decisions about college, which might be a contributing factor to Latinos’ high concentrations in the community college sector. Hallinan (2000) complicates the linkage between school quality and community college attendance further, by arguing that even those Latinos who attend schools with college preparatory curricula and higher levels of resources may not realize the benefits of those 42 resources due to tracking. Latinos and other students of color are disproportionately placed into non-academic tracks in high school and those students in non-academic tracks are more likely to attend community college than those graduating from academic tracks (Lavin & Hyllegard, 1996). Thus, the intersection of socioeconomic status and race within the high school context acts to disadvantage Latino students, increasing their chances of enrolling in the community college sector. McDonough’s (1997) notion of ‘organizational habitus’ is an important school/community contextual factor that has been shown to impact Latino students’ college decision-making. Student aspirations and expectations as they relate to college-going are largely shaped by the history of achievement in the high school and community context, or its organizational habitus (McDonough, 1997; Person & Rosenbaum, 2006). So for example, if previous graduates from a particular high school overwhelmingly attend community college, this trend of community college attendance will shape that high school’s students’ perceptions about their postsecondary options and constrain college decision-making. Latinos, who tend to be concentrated in underperforming urban high schools, are especially vulnerable to the negative effects of organizational habitus. The concept of organizational habitus also extends to the community in which Latinos are located. Admon (2006) and Person and Rosenbaum (2006) found that Latinos’ high concentration in community colleges could be linked back to the limited information about postsecondary education within their communities due to community members’ historical patterns of college-going. Person and Rosenbaum (2006) describe this as “chain enrollment,” by which Latino students select their own higher educational pathway based on where other community members have attended school. “Chain enrollment” is much more common among students with strong social ties to their communities, whereas students who lacked social capital within their community are more likely to seek out college information from 43 diverse, external sources on their own (Person & Rosenbaum, 2006). Thus, among those with strong community ties, previous patterns of participation in higher education within the school and community contexts serve to constrain Latino students’ college decision-making, thereby reproducing educational inequity, including high concentrations of Latinos’ in the community college sector. Higher education institutional context. Researchers have also looked toward the characteristics of community colleges and their environments in an effort to understand the relationship between higher education institutional contextual factors and high rates of community college attendance among Latino students. These factors, which I discuss below, include location (Admon, 2006; Olivas, 2005), costs (Paulsen & St. John, 2002; Perna, 2004; St. John, 2001; St. John, Paulsen, & Carter, 2005), marketing (Person & Rosenbaum, 2006) and program flexibility (Leigh & Gill, 2007). Several researchers have found a relationship between students’ college decision-making and institutional proximity to students’ home communities. Among Latino students, living at home, or attending college close to home are important factors when selecting a postsecondary institutional pathway (Grodsky, 2002, Person & Rosenbaum, 2006; Zarate & Pachon, 2006). This desire to remain in close proximity to home can be connected to Latinos’ high levels of community college enrollment due to the locations of these institutions. Community colleges have a wide geographic distribution, and some claim that they were intended to bring educational opportunity to the communities that they serve (Cohen & Brawer, 2003). In this sense, familiarity might be responsible in part for the heavy concentration of Latinos in community colleges. It is important to take a critical look at the intersection of place and college opportunity. As Olivas (2005) explains, “geography affects opportunity” (p. 182) and the relative ease of community college access juxtaposed with the physical and figurative distance of four-year institutions 44 should not be viewed as benign coincidence. Many scholars (Admon, 2006; Brint & Karabel, 1989; Person & Rosenbaum, 2006) have cautioned us that the convenience of community colleges has acted to constrain college choice among low-income students, Latinos and other underrepresented minorities, thereby furthering inequity and the accumulation of disadvantage. In addition to their location, there are other characteristics of community colleges that may make them particularly attractive to Latino students. These include program flexibility and breadth of offerings, both of which are often marketed by community colleges to encourage student enrollment (Leigh & Gill, 2007). In terms of program flexibility, community colleges allow for part-time attendance by offering courses that will fit an employed person’s schedule. Further, it is commonplace for students to move from community college to community college from one semester to the next, or even within the same semester (Adelman, 2005). The ease of the community college admissions process undoubtedly facilitates this type of lateral transfer, and allows students to go to the institution that offers the courses they need. Community colleges also have the unique ability to respond to business and “consumer” demands in terms of academic and vocational program offerings (Cohen & Brawer, 2003; Dougherty, 1994; Leigh & Gill, 2007). Thus, as a particular occupational or academic area becomes more desirable, community colleges can react quickly to offer related coursework, and even certificates and degree programs (Dougherty, 1994; Leigh & Gill, 2007). This ability to offer in-demand programs may attract Latino students who are looking to get immediate labor market returns to postsecondary education and the resulting upward economic mobility (Kane & Rouse, 1995). Economic and higher education policy context. The final layer in Perna’s (2006a) model of college decision-making pertains to those factors in the social, economic, and higher education policy context. In the specific case of Latinos enrolling in community colleges in high 45 proportions, there are a variety of economic factors and higher education policies that are related to this phenomenon. Though currently, there are a record number of bachelor’s degree holders in the U.S. (NCES, 2007), there remains a niche/demand in the American labor market for associate degree earners (Kane & Rouse, 1995; Leigh & Gill, 2007). This is evidenced by the fact that associate degree holders see labor market returns that place them at an advantage over individuals with just a high school diploma or equivalent (Kane & Rouse, 1995). This characteristic of the economic context might encourage Latinos to attend community college as these institutions can facilitate advances in the labor market on shorter time scales than it would take to earn a bachelor’s degree. For low-income Latino students, the prospect of near-term economic benefits resulting from community college attendance could be enough to lure them into the public 2-year higher education sector. State-level higher education policy also plays a large part in contributing to the large concentration of Latinos in the community college sector. Few open access public four-year institutions exist, which reflects many states’ decision to make the community college a heavily traversed pathway into higher education. Some states, many of which have large Latino populations, have provided a structure to smooth this pathway through policy. California, Florida, and New York, three states in which 45% of U.S. Latinos reside (U.S Census Bureau, 2007), have legitimated the community college as an entry point to the bachelor’s degree through their highly structured and articulated public postsecondary education systems (Cohen & Brawer, 2003; Wellman, 2002). These states’ purport to offer students a seamless route from the community college to four-year institutions via articulation agreements, and in some cases, guaranteed admission to public four-year institutions after transferring. For example, California’s 1960 Master Plan for Higher Education stipulates that the University of California (UC) and California State University (CSU) systems reserve 60% of its statewide undergraduate enrollment capacity 46 for community college transfer students. In Florida, the community college plays a vital role in the 2+2 higher education system, serving as the entry point to nearly 53% of juniors and seniors enrolled in the state’s universities (Florida Department of Education, 2005). Students who earn an associate degree in the Florida community college system are guaranteed admission to the public four-year university system. The structure and policies of Florida’s higher education system encourages community college attendance among all students, and Latinos in particular. In New York, community colleges are designated as campuses in the state’s two major university systems, the City University of New York (CUNY) and the State University of New York (SUNY). The inclusion of public two-year institutions in the CUNY and SUNY systems seeks to facilitate the transfer of community college credits toward a four-year degree (Wellman, 2002). In these three Latino-rich states, higher education policy acts to encourage community college attendance through the promise of transfer as indicated by the high concentrations of Latinos enrolled in California, Florida, and New York community colleges. Other states with large Latino populations lack longstanding agreements between community colleges and public four-year colleges and universities. Texas, home to 19% of the nation’s Latino population, does not have a statewide cooperative agreement between two-year and four-year institutions. Instead, Texas’ public higher education system is comprised of a combination of large-scale university systems, regional systems, and independent institutions. Community colleges were tasked with developing voluntary articulation agreements with each of these systems and institutions until the state’s higher education coordinating board developed a statewide transfer curriculum in the late nineties (Wellman, 2002). Further, there is no guaranteed transfer program mandated by state policy, though some institutions have developed such agreements on their own. Perhaps as a result from these policies, the rate of community college 47 attendance in Texas is lower than that of California and Florida. However, Latinos remain disproportionately represented in public two-year institutions (Admon, 2006). Recent state policy trends with regard to basic skills (i.e. remedial) education also act to impact community college attendance. In many states, basic skills education is increasingly relegated to the community college sector. For example, in California students are given one year to complete basic skills education in the state university system. If a student cannot successfully complete all basic skills requirements in that time frame, they are disenrolled from the four-year institution and forced to reverse transfer to a community college (California State University Office of the Chancellor, 1997). New York has taken further steps, opting to phase out all basic skills education from public four-year institutions. These policies disproportionately affect Latinos and other underrepresented students as they more commonly graduate from under- performing high schools in urban areas. Further, the advent of high school exit examinations (e.g., California High School Exit Examination [CAHSEE]) has sealed the community college’s fate as “remediator” of the failings of underperforming secondary schools. Funneling already disadvantaged students, including Latinos, into the community college sector can only be justified through the promise of transfer—that is, by providing access to a bachelor’s degree through the two-year sector. However, low transfer rates (Grubb, 1991) and the small proportions of Latino community college students who transfer to highly selective institutions (Dowd, Cheslock, & Melguizo, in press) indicate that state policies that serve to concentrate Latinos in community colleges hinder access to comprehensive institutions and severely reduces their chances of enrolling in highly selective universities. In addition to policies regarding the structure of postsecondary education systems and the function of the community college, higher education financial aid policy also contributes to the concentration of Latinos in the community college sector. Rising college tuitions and decreased 48 buying power of federal grants (e.g., Pell Grants) has led to an increase in the amount of debt students incur to finance college (College Board, 2006, 2007; Dowd, in press; Price, 2004; St. John, 2003). The policy decision to allow appropriations for federal grant programs to lag behind the growing number of students enrolled in postsecondary education and increased college costs has caused loans to become a central part of financial aid packages, particularly among low- income and minority students attending four-year institutions (Price, 2004; St. John, 2003). Further, cuts in higher education appropriations at the state-level and decreased availability in need-based state grant aid (St. John, 2003) has bolstered the reliance on student loans, particularly in the four-year sector. This financial aid policy environment coupled with the relatively low tuition at public two-year institutions leads to high Latino enrollment in the community college sector (Admon, 2006; Nora, 1990; Olivas, 1985; Person & Rosenbaum, 2006; Paulsen & St. John, 2002). Despite the decreased buying power of the Pell Grant, community college tuition rates are such that grant aid from federal and state sources can cover them in large part. This is evidenced by the fact that borrowing among community college students, and Latinos community college students in particular, are low compared to those in the four-year sector (Santiago & Cunningham, 2005). In this sense, going to community college in lieu of direct entry to a four- year institution may be an attractive option for those students who do not wish to borrow, which is increasingly necessary in the current financial aid policy context. Factors Related to Latino Student Attendance at Hispanic Serving Institutions Hispanic-serving institutions (HSIs) continue to be understudied in terms of their attributes that lead to high Latino enrollment and their uniqueness as an institutional type. Unlike other minority-serving institutions, HSIs inherit their designation based on the level of Latino student enrollment; beginning in 1992, any not-for-profit, accredited postsecondary institution whose full-time equivalent enrollment is at least 25% is designated as “Hispanic-serving” and 49 eligible to apply for earmarked funding under Title V of the Higher Education Act (U.S. Department of Education, 2007). Perhaps due to the newness of this institutional type, only a few researchers have attempted to understand what factors lead Latinos to enroll in these institutions in such high concentrations. Certainly this country’s evolving demographics are in part responsible for the growing number of HSIs; Latino enrollment in college has more than doubled since 1990 (NCES, 2002). However, it is important to consider what other contextual factors might account for the fact that HSIs, which comprise just 8% of U.S. postsecondary institutions, enroll nearly half of all Latinos in college (Stearns & Watanabe, 2002; Santiago, 2006). Further, the reader should understand that factors in the layers of context in which Latino students are situated shape their decision to enroll in certain institutions to the point that Latinos exceed 25% of those institutions’ enrollment, causing those institutions to be designated as HSIs. Individual characteristics. Although little research has been conducted regarding the characteristics of Latino students that are associated with attendance at an HSI, some of the findings regarding Latinos in community colleges may provide insight. For example, Hispanic- serving colleges and universities tend to be less selective institutions with lower admissions requirements (Contreras, Malcom, & Bensimon, 2008; Santiago, 2006; Stearns & Watanabe, 2002). Because race/ethnicity and class intersect in a manner that clusters the historically disadvantaged in underfunded, underperforming primary and secondary schools (Lareau, 2003; Price, 2004), Latinos, many of whom are first-generation college students, may not be as academically prepared as middle and upper-class white students (Lareau, 2003; Stanton-Salazar, 2001). The disadvantages accumulated over years of schooling likely prevent these Latino students from gaining access to highly selective institutions (Fry, 2004; Lareau, 2003), perhaps making less selective HSIs a more attainable college option. 50 Hispanic-serving institutions are on average, less expensive than other public four-year institutions. In 2003-04, the average in-state tuition at public four-year HSIs ($1,590) was less than half of that of all public four-year institutions ($3,400). In this sense, four-year HSIs are similar to community colleges in terms of affordability. Though empirical studies have not explored this question, it is possible that Latino students from lower socioeconomic backgrounds are attracted to HSIs due to lower tuition rates. Studies (Hurtado, Inkelas, Briggs, & Rhee, 1997; Swail, Cabrera, & Lee, 2004; Tornatzky et al., 2003) have shown that a high proportion of Latino students and their families express a strong desire to attend college close to their home communities. Hispanic-serving institutions are located in areas with large Latino populations; thus, the concentration of Latinos in HSIs may simply reflect the intersection of geography and educational access (Olivas, 2005). School and community context. Many of the factors in the school and community contexts related to Latinos’ high concentration in community colleges might also be applicable to their attendance at HSIs. For example, the importance of organizational habitus (McDonough, 1997) and chain enrollment (Person & Rosenbaum, 2006) should not be understated when attempting to describe school and community contextual factors that lead to high concentrations of Latinos in HSIs. If the social and cultural capital possessed by the high schools and communities in which Latinos are located is limited to certain institutions, these patterns of enrollment will repeat, leading to high concentrations of Latinos in those institutions. Unfortunately, researchers have yet to explore this question, and further inquiry into the school and community contextual factors that lead to high Latino enrollment at HSIs is needed. Institutional context. It is difficult to characterize the ways in which the institutional contexts of HSIs contribute to their large Latino enrollments. Latinos’ share of enrollment at HSIs range from the minimum 25% to more than 90% (Contreras, Malcom & Bensimon, 2008; 51 Santiago, 2006; Stearns, Watanabe, & Snyder, 2002). Undoubtedly, the institutional environments in the country’s HSIs vary widely; previous research has demonstrated a range of incorporation of the Hispanic-serving designation into institutional mission and artifacts (Contreras, Malcom & Bensimon, 2008). However, there are some commonalities among HSIs. The first of these is Title V funding. Title V of the Higher Education Act authorizes the Developing HSI Program, which seeks to “assist HSIs to expand educational opportunities for, and improve the academic attainment of Latino students. The Developing HSI Program also enables HSIs to expand and enhance their academic offerings, program quality, and institutional stability” (Title V, n.d.). The funding provided by Title V has helped many HSIs to develop programs intended to facilitate Latino student success. Though the effectiveness of Title V programs has yet to be illustrated on a large-scale (Contreras, Malcom & Bensimon, 2008), it is possible that special Title V programs and initiatives encourage Latinos to enroll in HSIs. Previous investigations regarding academic and student support programs available at HSIs (Kuh et al., 2003; Laden, 1999, 2000) demonstrate that some of these institutions attempt to be responsive to the needs of Latino students. It is not clear however, how much of a role these programs play in attracting Hispanic students to HSIs. The Hispanic-serving identity of many HSIs is largely invisible (Contreras, Malcom, & Bensimon, 2008). However, a limited number of HSIs have begun to market themselves as institutions specializing in providing access for and facilitating the success of Latino students. For example, California State University Long Beach (CSULB) has begun to distribute t-shirts imprinted with the mottos “Mi Casa: Mi Universidad, Hispanic Serving Institution” and “Tu Universidad es tu futuro.” Similarly the University of Texas—Brownsville has strategically marketed itself as an HSI, including the designation in key institutional materials and documents 52 (Kuh et al., 2003). These limited examples reflect an emerging trend of institutions promoting themselves as HSIs to encourage additional Latino student enrollment. Economic and social policy context. Much of the concentration of Latinos in HSIs is attributable to factors in the economic and social policy context. Educational opportunity is spatially dictated in large part, particularly among low-income students and students of color (Olivas, 2005). As a consequence, the residential patterns of Latinos and clustering in certain geographic areas affect the institutions in which they enroll. Tienda and Niu (2004) found that in Texas, the large extent of residential and secondary school segregation leads to clustering of Latinos in particular colleges and universities (e.g. University of Texas-El Paso, University of Texas-Brownsville, etc.), causing these institutions to become HSIs. Similar studies are needed to examine the relationship between geography and enrollment in HSIs in other states. However the patterns of Latino enrollment in HSIs, the locations of these institutions and demographic trends certainly suggest a strong connection. Community Colleges and Hispanic Serving Institutions in the Present Study Regardless of the factors that lead them to enroll in community colleges and HSIs, these two institutional pathways clearly serve large proportions of Latino students. However, it is unclear how much of a role HSIs and community college play in educating Latino scientists and engineers. Do these institutions serve as a pathway to the STEM baccalaureate for Latinos? The present study addresses this critical question and investigates the ways in which the pathways to STEM vary by Latino students’ context. Formulation of College Financing Strategies: Latino Students and the Decision to Borrow “[Debt] is the new paradigm of college funding” (Williams, 2006, p. 156). This poignant observation comes neither from a policy expert nor an educational researcher but an English professor who is also one of the millions of college graduates in debt due to student loans. 53 Williams (2006) goes on to detail the psychological and financial burdens of the student loan debt that he continues to bear nearly fifteen years after completing his education. “Debt” he writes, “is not just a check every month but colors the day-to-day experiences of my life” (Williams, 2006, p. 156). Indeed, the manner in which a student decides to finance college not only affects students’ educational experiences and outcomes, but also has longstanding consequences. Student loan debt has almost become a rite of passage for large proportions of college students. In 2000, nearly half of low- and middle-income undergraduates and 31% of high- income undergraduates borrowed (Horn, Wei, & Berker, 2002). Current borrowing rates, which are considerably higher than in the past, are attributable to a number of factors. Many researchers have documented the governmental and institutional policy changes that have led to increased borrowing (College Board, 2006, 2007; Long & Riley, 2007; Price, 2004; St. John, 2003). These changes are largely driven by the current perception of higher education as a private benefit as well as a public good (Long & Riley, 2007; Price, 2004; St. John, 2003). Though researchers have established links between the potential and actual burden of student loan debt and college choice, educational attainment and professional outcomes of minority and low-income students, uncertainty regarding the causality, direction and magnitude of these relationships remain. The potential consequences of students’ willingness to borrow and resulting debt burden underscore the importance of understanding the factors that shape students’ decisions regarding student loans. In this section of my review of the literature, I discuss the relationships between context and Latino students’ patterns of financial aid and in particular, student loan participation. Though some researchers (Burdman, 2005; De La Rosa & Hernandez- Gravelle, 2007; Monaghan, 2001) have characterized Latino students as “debt averse” or sought to identify “cultural barriers to incurring debt” (ECMC Group Foundation, 2003), the way in which cumulative disadvantage places Latinos in particularized contexts offers a different means 54 of understanding these students’ borrowing patterns. Before discussing the factors that influence decisions regarding college financing and borrowing, I now turn to a discussion of Latino college students’ patterns of financial aid. Paying for College: Latino Students’ Use of Financial Aid Latinos demonstrate relatively high levels of financial need compared to other racial/ethnic groups. In 2003-04, nearly 42% of Latino undergraduates had an expected family contribution, which is the federal government’s measure of a family’s ability to pay, of $1,000 or less (Santiago & Cunningham, 2005). Among all undergraduates, 30% had expected family contributions of $1,000 or less. This statistic reflects the concentration of Latinos in lower cost institutions and family socioeconomic status. Interestingly, this high level of need does not necessarily translate into greater usage of financial aid among Latinos. While almost 80% of Latino undergraduates applied for financial aid in 2003-04 and 63% of these students received some form of aid, the average award was lower than that of any other racial/ethnic group (Santiago & Cunningham, 2005). This may be a reflection of several factors, including Latinos’ tendency to enroll in institutions with lower tuitions (e.g., community colleges and HSIs), the below average rates at which Latinos receive state and institutional aid, and lower participation rates of Latinos in loan programs. Latinos at 2-year institutions. Nearly 43% of Latinos who attended public community colleges received financial aid of some sort during the 2003-04 academic year; a rate four percentage points below the average of 47% for all students in public two-year institutions (Santiago & Cunningham, 2005). The majority of these students’ aid came from the federal government in the form of Pell grants. Slightly less than 10% of Latinos in public community colleges received aid from the state, and about 11% received institutional aid. Latinos in community colleges were much more likely to receive grant aid than financial aid in the form of 55 loans. Only about seven percent of Latinos enrolled in a public two-year institution received loans. While more Latino community college students received grants than loans, the average grant award ($2,276) was smaller than the average loan ($3,273). Though the loan participation rate of Latino community college students was below the average of 12% for all community college students, Latinos did not borrow at the lowest rate. Both Asian students and Native Hawaiian/Pacific Islanders borrowed at a lower rate than Latinos (Santiago & Cunningham, 2005). Latinos at 4-year public institutions. The financial aid usage patterns among Latino students at public four-year institutions differed from Latino community college students. More than 75% of all Latino undergraduates at public four-year colleges and universities received financial aid. This rate is above the average of 68% for all undergraduates at public four-year institutions. Sixty-two percent of Latinos in public four-year institutions received federal financial aid, and about two-thirds of these students received Pell grants (Santiago & Cunningham, 2005). The large proportion of Pell grant recipients among Latino undergraduates at public four-year institutions reflects the relatively high number of low-income students in this population. Financial aid in the form of state aid and institutional aid were much more common among Latinos in public four-year colleges and universities with 28% and 26% receiving state and institutional aid, respectively. Grants from any source (i.e., federal, state, institution) were more commonly received by Latinos in four-year public institutions compared to loans. Sixty-two percent of Latinos enrolled in public colleges and universities received grants, while 41% received loans. The average grant size was significantly higher than the average loan ($7,103 versus $5,055). Interestingly, while Latinos in four-year public institutions borrowed at a rate lower than the average of 45%, Asian and Native Hawaiian/Pacific Islanders received loans at a much lower rate than Latinos (33 and 39% respectively) (Santiago & Cunningham, 2005). 56 Latinos at 4-year Private institutions. A larger proportion of Latino students at private, not-for-profit four-year institutions received financial aid compared to those in community colleges and public four-year institutions. Eighty-six percent of Latinos enrolled in private colleges and universities received financial aid, with nearly 85% of these students receiving federal aid. Half of Latinos attending four-year private institutions received Pell grants in 2003- 04. As one might expect due to their higher tuition rates, private colleges and universities awarded institutional aid to 37% of Latino undergraduates, primarily in the form of grants. About one quarter of Latinos in private four-year institutions received state aid, mostly as grants. Fifty percent of Latinos in private colleges and universities used student loans to pay for college, and the average loan size was $6,251. Latinos enrolled in private four-year institutions borrowed at a rate below the average of 56%, however, they borrowed more commonly than Asians, 49% of whom received loans (Santiago & Cunningham, 2005). Figure 2.3 provides a summary of the financial aid usage patterns among Latinos college students by institutional sector. Summary of differences in financial aid by institutional sector. Latinos enrolled in private four-year institutions received financial aid more commonly than Latinos in public four-year institutions or community colleges. Latinos in private four-year institutions were more likely to receive Pell grants, federal subsidized loans, private loans and institutional aid than those in public four-year institutions and community colleges. Latinos in community colleges borrowed at extremely low rates and less than half received financial aid from any source. The difference in the participation rates of Latino students is likely due to variations in costs of attendance by institutional types. 57 Figure 2.3. Financial Aid Usage of Latino Undergraduate Students by Institutional Sector Variations in Loan Program Participation among Latinos by National Origin. Among Latino undergraduates, there were considerable differences in loan program participation rates by national origin and institutional sector. On average, Latinos of Cuban origin borrowed more commonly than those of Mexican or Chicano descent and Puerto Rican descent. Nearly 35% of Latinos of Cuban origin borrowed compared to 29% of Latinos of Mexican/Chicano descent and 31% of those of Puerto Rican origin. However, this pattern does not hold for all institutional sectors. Among Latinos enrolled in community college, those of Cuban origin borrowed at a lower rate than students of Mexican/Chicano descent or Puerto Rican origin. Five percent of 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Total Aid Total Grants Total Loans Institutional Aid State Aid Pell Grants Federal Subsidized Loans Source of Aid Percent Receiving Aid from Source All Latinos Private 4-year Public 4-year Public 2-year Source: 2004 National Postsecondary Student Aid Survey (NPSAS) as cited in Santiago & Cunningham, 2005. Federal Unsubsidized Loans 58 Latinos of Cuban origin in community college received loans, while seven percent of Latinos of Mexican descent and 12% of Latinos of Puerto Rican descent enrolled in community college received loans. Among Latinos in public four-year institutions, 35% of students of Cuban origin borrowed, compared to 47% of students of Mexican descent and 25% of students of Puerto Rican descent. For those Latinos enrolled in private four-year institutions, 65% of students of Cuban origin borrowed, compared to 66% of students of Mexican descent and 37% of students of Puerto Rican origin (Santiago & Cunningham, 2005). Figure 2.4 summarizes the loan participation rates of Latinos by national origin and institutional sector. Although on average, Latinos of Mexican descent borrow at lower rates than those of Cuban and Puerto Rican descent, the patterns are very irregular when considering borrowing by institutional type. The average figures are likely due to the higher concentration of Latinos of Mexican or Chicano descent in the community college sector rather than any difference in attitudes toward borrowing. Extremely small percentages of students in community college receive loans and more than half (54%) of Latinos of Mexican descent enrolled in postsecondary education attend a community college, compared to 34% of students of Cuban origin. 59 Figure 2.4. Latinos’ Participation Rates in Student Loan Programs by Institutional Sector and National Origin Latino Student Attitudes towards Borrowing As the previous section illustrates, significant proportions of Latino students enrolled in four-year institutions use student loans to help finance college. Yet, because Latino college students borrow at lower rates than whites and African Americans there remains a narrative in the educational research literature (Burdman, 2005; De La Rosa & Hernandez-Gravelle, 2007; Monaghan, 2001; ECMC Group Foundation, 2003) and mainstream media (Guess, 2007; Vara- Orta, 2007) that Latinos hold negative attitudes, or an aversion, towards borrowing. According to those who make this argument, a combination of cultural reasons including distrust of lenders, economic circumstances such as fear of default, and informational barriers deter Latino students Source: 2004 National Postsecondary Student Aid Survey (NPSAS) as cited in Santiago & Cunningham, 2005. Percent of Students who Borrowed Institutional Sector 0% 10% 20% 30% 40% 50% 60% 70% All sectors Public 2-year Public 4-year Private 4-year National Origin Cuban Mexican Other Latino Puerto Rican 60 from using loans to finance their college educations (Burdman, 2005; De La Rosa & Hernandez- Gravelle, 2007). This is of concern because Latino students may alter their college-going choices by attending lower cost institutions, working while attending school, or stopping out, in order to avoid debt. A number of studies have demonstrated that Latinos do hold somewhat negative perceptions of student loans (McDonough, 2004; Tomás Rivera Policy Institute, 2004; Zarate & Pachon, 2006). In a study of California Latino youth, Zarate and Pachon (2006) found that while the students in their sample placed a high value on a college education, incurring debt and the inability to work were opportunity costs that, along with monetary costs, did not outweigh the benefits of college attendance. Additionally, Zarate and Pachon (2006) discovered that misinformation regarding the costs of attending California public colleges and universities, the availability of financial aid and the eligibility requirements for receiving such aid were all too common among Latino youth. Zarate and Pachon’s (2006) study revealed the ways in which the internal cost-benefit analysis regarding college attendance is a highly subjective process that can significantly impact college plans. Many researchers have aimed to understand the factors that influence the attitudes of Latino college-aged students toward financial aid and have identified a number of psychological, social, and informational barriers. In a study of 900 students, Trent, Lee, and Owens-Nicholson (2006) found that nearly half of the Latinos in their sample had extremely low debt tolerances. These students expressed an unwillingness to borrow to finance college, which the authors attributed to lower expectations of educational attainment and external attributions regarding academic achievement. Nora, Barlow, and Crisp (2006) identified the potential stress associated with loan repayment as an additional psychological barrier to debt tolerance among Latinos and other students of color. Certainly, fear of default negatively impacts debt tolerance, particularly if 61 students are uncertain about their ability to persist and complete a bachelor’s degree (Burdman, 2005). An unwillingness to borrow, however, does not necessarily result from some cultural predisposition shared by Latino students. Latinos have accumulated significant levels of disadvantage due to historic and ongoing opportunity denial. Cumulative disadvantage places these individuals within unique layers of context in which they are more likely to witness economic tenuousness and experience trepidation about the future. Among students in these circumstances, why would borrowing tens of thousands of dollars seem to be a safe bet? The connection between cumulative disadvantage, context, and Latino students’ perceptions of debt necessitates a more critical perspective to the study of the decision-making process with regard to student loans. Some researchers bring this critical perspective to the study of borrowing by invoking sociocultural theory in their attempts to understand the contributing factors to Latinos’ perceptions of debt and borrowing patterns. McDonough and Calderone (2006) present an alternative framework for understanding the perceptual differences in financial aid among low- income Latino students and their families. These authors posit that perceptions of financial aid by low-income students are socially constructed and are informed by their personal needs assessments and non-rational cost-benefit analyses that include ongoing mental balancing acts of perceived college costs and familial financial needs. McDonough and Calderone (2006) attribute the disconnect they observed between college counselors and low-income students and their families to the relative nature of affordability and cost, a lack of understanding of financial aid terminology (e.g. grants versus loans), and the parents’ perceptions of loans as “insurmountable debt” (p. 1714). McDonough and Calderone (2006) demonstrate the inapplicability of traditional rational choice models to the financial aid decision-making process among low-income students 62 and the ways in which sociocultural factors influence how students and their families make sense of college costs and loans. Their findings also underscore the relative nature of debt and the linkages between context and debt tolerance (McDonough & Calderone, 2006). St. John (2003) similarly cites the importance of the ways in which low-income Latinos make sense of money in decisions regarding their use of financial aid. Perna (2004, 2006b) elaborates on the often erroneous perceptions of college costs among low-income students, including Latinos, describing how these students and their families overestimate the true costs of college. These false perceptions significantly impact college choices, directing low-income Latino students to lower cost institutions, particularly community colleges (Admon, 2006; Perna, 2004, 2006b). Many scholars have cited the importance of providing access to information regarding financial aid and college costs to low-income students to alter the misconceptions of college costs and to overcome the negative attitudes towards debt. Tierney and Venegas (2006) highlight the importance of social networks as a means to gather information regarding financial aid and college access, while Venegas (2006) points out the potential of the Internet as a means to access information regarding financial aid. A national survey of college-aged Latinos and their parents conducted by the Tomás Rivera Policy Institute in concert with the Sallie Mae Fund provides further assessment of the level of awareness regarding financial aid among this population and the detrimental effects of a lack of information regarding financial aid on plans to attend college (Tomás Rivera Policy Institute, 2004). This study further reflects the lack of information regarding financial aid often endured by Latino college-aged students and their families. More than half of Latino parents and nearly 43% of the college-aged Latinos in the study could not name a single source of financial aid. This study also revealed the lack of awareness of loans as a source of aid: 80% of Latino parents and 74% of college-aged Latinos did not cite loans as a source of financial aid. The survey also revealed that lack of information about financial aid acts 63 as a barrier to college attendance; more than two-thirds of Latino families indicated that receiving information about financial aid before leaving high school was extremely important to plans to attend college, and three quarters of Latino college-aged respondents not enrolled in college indicated that they would have been more likely to attend if they had received information about college financing through grants, scholarships and loans. While the aforementioned studies offer empirical evidence of the perceptions of Latino college students with respect to financial aid and loans and describe some of the contributing factors to these perceptions, the link between the attitudes of Latinos towards student loan debt and college-going behaviors intended to avoid debt are not as well-established (Dowd, in press). It is certainly true that more than half of Latinos enrolled in postsecondary education attend community college, where tuitions are considerably lower than four-year institutions and borrowing to finance college costs is uncommon. However, very little empirical evidence exists to suggest that Latino students who enroll in community colleges do so to avoid loans. As explained in the previous section of this literature review, Latinos’ college decision-making processes are influenced by a number of contextual factors including socioeconomic status, parental educational attainment, generational status, geographical location, educational aspirations and expectations, access to information, as well as college costs and in some cases, the availability of financial aid. 5 Though there may be a relationship between the large portion of Latinos in community college and debt aversion, a confluence of social-historical, demographic, institutional, and economic factors are also contributing factors. Borrowing rates, although low among Latinos in community college, increase markedly for Latinos enrolled in public and private four-year 5 Although Kim (2004) and Arbona & Cabrera (2007) found that on average, Latinos’ college choices were not influenced by financial aid, this was not true of Latinos who were low-income. 64 institutions. Additionally and quite interestingly, Latinos enrolled in for-profit institutions borrow at rates greater than students at private not-for-profit colleges and universities (Santiago & Cunningham, 2005) despite the fact that a large proportion of Latino students who attend proprietary schools are low-income as evidenced by the proportion of Pell grant recipients (Santiago & Cunningham, 2005). This seems to suggest that institutional factors in addition to access to information, and attitudes towards debt influence the actual college financing strategies used by Latino students (Dowd, in press). In her extensive review of the literature, Dowd (in press) cautions researchers against accepting the debt aversion premise, and calls for additional empirical investigation of the financing strategies used by different types of students. Dowd’s point is underscored by the irregularity of patterns of borrowing among Latinos by national origin and institutional type. The present study contributes to the knowledge regarding Latino students and financial aid by determining the college financing strategies employed by Latino STEM bachelor’s degree holders, and assessing the differences in financing strategies among Latinos by national origin. Although previous researchers (Fenske, Porter & DuBrock, 2000) have studied the financial aid of underrepresented STEM majors, the findings described financial aid patterns at a single, large, public university and thus, the sample of Latinos was not large enough to disaggregate these students. Furthermore, the findings are limited in that they are tightly bound to the institutional context of the focus university. Other researchers (Redd, 2006) investigate the financial aid for minority students in STEM fields using national datasets, however the findings focus on minority students in graduate school and do not consider the specific financing strategies of Latino graduate students in STEM fields. The present work improves upon the current state of knowledge regarding college financing among Latino STEM majors, further informing the ongoing discussion regarding Latino student debt aversion. 65 Latinos Choosing and Succeeding in STEM: Factors in the Institutional Environments of Community Colleges and Hispanic Serving Institutions The success of Latino students studying science-related fields is attributable to many factors in the (dis)advantaging contexts framework used in this study. However, much of the research on the success of Latinos in STEM relies on theories of student success that focus on students’ preparation, involvement (Astin, 1997), academic integration (Tinto, 1993), and engagement (Kuh et al., 2005). These commonly invoked theories of student persistence and retention frame success as solely dependent on student characteristics, which are often beyond the control of the postsecondary institution (Bensimon, 2007). The problem of depressed participation and success rates of Latinos in the sciences is one of shared responsibility between the institution and the student. The success of certain programs and institutions in increasing the degree attainment of Latinos in STEM illustrates that contextual factors within institutional environments (e.g. institutional climate, faculty members, and structural support) make significant contributions to student outcomes. Very little research has focused on characteristics of institutional environments that positively impact the retention and persistence of Latino STEM majors. In particular, few studies have attempted to understand the ways in which the institutional contexts of Hispanic Serving Institutions and community colleges, the two types of institutions responsible for educating the vast majority of Latinos, contribute to or hinder the success of Latinos in STEM. In the present study, I investigate how the patterns of achievement of Latino STEM bachelor’s degree holders vary by institutional pathway. Do certain institutional pathways (e.g., HSI versus non-HSI) facilitate the success of Latino students in different STEM fields? What contextual factors might account for these differences? While factors in the three other layers of 66 context in Perna’s (2006a) conceptual model are indeed important, I am particularly interested in understanding the role of institutional context in degree outcomes of Latino STEM majors. In this portion of my review of the literature related to the present study, I discuss those characteristics of the institutional environments of community colleges and Hispanic-serving institutions that might facilitate the recruitment and success of Latinos in STEM fields, which have largely been ignored in the literature. Due to the lack of existing research on the contributions of Hispanic-serving Institutions, I draw upon the literature regarding the contribution of minority serving institutions (MSIs) (i.e., Historically Black Colleges and Universities (HBCUs) and women’s colleges) on the production of minority and female STEM degree holders. I also discuss how community colleges have acted as an entry point for Latino STEM majors via targeted programs such as the Puente Project and Mathematics Engineering and Science Achievement (MESA). Contributions of Minority Serving Institutions to STEM Degree Attainment Several studies (NSF, 2005; Pearson & Pearson, 1985; Perna, 2001; Redd, 1998; Redmond, Clinedinst, & O’Brien, 2000; Solórzano, 1995; Trent & Hill, 1994; Redd, 1998; Wolf- Wendel, 1998) on the baccalaureate origin institutions of minority and female scientists and engineers have found that MSIs award a disproportionately high percentage of degrees to African Americans and women, controlling for institutional size. Historically, HBCUs and women’s colleges have been top producers of African American and female scientists (Pearson & Pearson, 1985; Trent & Hill, 1994). This is not surprising due to the systematic exclusion of these students from predominantly white institutions (PWIs); however, the continued importance of these special mission institutions even after widespread enrollment by minority students at traditionally white institutions demonstrates that the success of HBCUs and women’s colleges is no coincidence. Indeed, attendance at HBCUs and women’s colleges leads to higher educational 67 outcomes for African American and female students in all fields, and particularly in science and engineering. Graduation rates of African Americans and women are significantly higher at HBCUs and women’s colleges, respectively (Ledman, Miller & Brown, 1995; NCES, 2006; Smith, Wolf, & Morrison, 1995). Additionally, HBCUs and women’s colleges award a disproportionately high share of degrees compared to their enrollments, illustrating their ability to foster success among their target populations. For example, HBCUs awarded 20% of bachelor’s degrees earned by African American students in 2003-04, though they enrolled just 12% of all African American students (NCES, 2005). Women’s colleges awarded 1.6% of bachelor’s degrees earned by females in 2003-04, though they enrolled just 0.8% of all female students (NCES, 2005). This trend of over-performance by HBCUs and women’s colleges continues for STEM fields. In 2000, HBCUs graduated 40% of all African American students who received a bachelor’s degree in biology, and more than 40% of black students who earned a degree in the physical sciences (NSF, 2002; Suitts, 2003). Spelman College, a historically black women’s college, is the top degree-granting institution for black undergraduate women in STEM fields (NSF, 2005). In fact, Ehrenberg & Rothstein (1993) found that HBCU attendance increases the chances of earning a bachelor’s degree for African American students, controlling for student background characteristics such as ability and socioeconomic origin. Female participation rates in STEM fields are higher at women’s colleges, as evidenced by the fact that they confer a larger proportion of bachelor’s degrees in traditionally male-dominated fields (mathematics, science, and engineering) than coeducational, private colleges do (Sebrechts, 2000). Furthermore, the percentage of majors in economics, math and the life sciences is higher in women’s colleges today than it is even for men at coeducational colleges (Sebrechts, 2000). 68 With HBCUs and women’s colleges’ high level of success in terms of conferring degrees to African American and female students in all fields, and STEM fields in particular, it is no surprise that these institutions also play a large role in educating students who go on to earn graduate degrees. Although Ehrenberg & Rothstein (1993) concluded that HBCU attendance does not increase the chances of graduate school enrollment among African Americans, several subsequent studies demonstrate the importance of MSIs as baccalaureate origin institutions for African American and women doctorates. Eight of the top ten baccalaureate origin institutions for black science and engineering doctorates are HBCUs (NSF, 2005), while three women’s colleges were among the top 50 baccalaureate origin institutions of female science and engineering doctorates (NSF, 2005). Solórzano (1995) found that 30 of the top 50 baccalaureate institutions of Black female doctorates in science and engineering were HBCUs, controlling for institutional size. These figures are particularly telling, considering that HBCUs and women’s colleges constitute just 3.5%, and 2% of all bachelor degree-granting institutions in the U.S., respectively (NCES, 2005). Through an in-depth analysis of the baccalaureate origin institutions of White, African American, and Latina female doctorates, Wolf-Wendel (1998) concluded that institutional sex and institutional race were more important than institutional selectivity to the production of successful graduates. Wolf-Wendel, who defined ‘productivity’ as the number of white female, African American female, or Latina doctorates per 1,000 students enrolled, found that for each of these groups, MSIs performed at higher levels than PWIs and co-educational institutions in graduating future doctorates. The importance of MSIs in terms of serving as a pathway to graduate study among women and African Americans is certainly connected to Wenglinsky’s (1996) conclusion that black students at HBCUs have higher educational aspirations and are more likely to plan to enroll in a graduate program in science and engineering. These findings are 69 further strengthened by Sax, Astin, Korn & Mahoney (1997) who found that 27% of freshmen attending HBCUs plan to earn a doctoral degree compared to 17% of all freshmen nationwide. In fact, HBCUs are important producers of African American faculty in science, mathematics and engineering (Perna, 2001), illustrating that HBCUs serve as vital entry points for future African American scientists and engineers. Women’s colleges have also been shown to increase educational aspirations among female students, however, it should be noted that women’s colleges tend to enroll students of higher socioeconomic status, which may also contribute to the observed high levels of aspiration (Wolf-Wendel, 1998). The above studies demonstrate that HBCUs and women’s colleges are more successful at facilitating degree attainment and fostering high educational aspirations among their target populations. But what about these institutions make them more “productive” in terms of granting degrees to historically underrepresented groups than their non-MSI counterparts? Educational researchers have taken several empirical approaches to addressing this question, including analyzing these institutions’ student enrollments (Allen, 1986, 1992; Allen, Epps, & Haniff, 1992), campus climates (Hurtado & Carter, 1997; Hurtado & Ponjuan, 2005), mission orientation (Contreras, Malcom, & Bensimon, 2008), resources (NSF, 1999) and faculty (Kim, 2002), using theoretical models of minority and female student success to guide their inquiry. I discuss these studies’ findings regarding the effectiveness of HBCUs and women’s colleges below. Doing More with Less: Fostering STEM Degree Attainment at HBCUs Historically black institutions illustrate that institutional factors can outweigh individual characteristics in the formula for student success. Students who attend HBCUs tend to have lower levels of academic preparation, lower GPAs and college entrance exam scores (Allen, 1992; Kim, 2002) and come from economically disadvantaged backgrounds (Freeman, Perna, & King, 1999; Perna, 2001), two factors that previous studies (Astin, 1993) have found to hinder degree 70 attainment. Yet, African American students at HBCUs earn STEM degrees at higher rates than their counterparts at non-HBCUs and attending an HBCU is positively associated with persistence and degree attainment (Astin, 1977, 1993). Additionally, HBCUs often have fewer financial and technological resources and research facilities than other colleges and universities (Kim, 2002; NSF, 1999; Redmond et al., 2000), offer STEM majors fewer opportunities for advanced studies (Allen, Epps, & Haniff, 1991), and have fewer faculty members with Ph.D.s (Kim, 2002), but still manage to be successful in producing African American STEM degree holders (Harvey & Williams, 1989). These characteristics of HBCUs and the students enrolled at these institutions demonstrate that HBCUs are able to “do more with less” (Redmond et al., 2000, p. 21). The success of HBCUs has been attributed to several factors related to how African American students experience the campus and classroom environment, and the nature of student interactions with faculty members who also act as mentors, or mirrors of students’ possible selves (Markus & Nurius, 1986). A telling empirical finding by several researchers is that African American students experience less social isolation, dissatisfaction and racism than African Americans at PWIs (Pascarella & Terenzini, 1991; Perna, 2001; Jones, Castellanos, & Cole, 2002). Additionally, HBCUs provide a social environment that is more caring, nurturing, and supportive than at non-HBCUs (Blackwell, 1998; Fleming, 1984; Nettles, Thoeny, & Gosman, 1986; Redd, 1998; Wagener & Nettles, 1998). African American students who perceive their campus environment to be welcoming and validating are more likely to persist to degree attainment (Arbona & Nora, 2007; Hurtado & Carter, 1996; Rendón, 1994). In fact, the cultural comfort enjoyed by African American students at HBCUs reduces student perceptions of hostile environments that can negatively impact academic adjustment, persistence, and classroom performance (Nora & Cabrera, 1996; Hurtado & Ponjuan, 2005). These theoretical links are 71 borne out in Pascarella and Terenzini’s (1991) finding that persistence and attainment rates are higher among African American students at HBCUs compared to their counterparts at PWIs. Scholars who attribute student success to in- and out-of-class engagement (Kuh & associates, 2005) have also investigated the differences between levels of engagement among African American students at HBCUs and PWIs. Several studies (Allen, 1986; Bohr, Pascarella, Nora & Terenzini, 1995; Cokley, 1999; Watson & Kuh, 1996) find that HBCUs offer more engaging learning environments for African American students, which may positively impact student outcomes. Due to the focus on serving African American students, chapters of science and engineering professional organizations intended for students of color and women (e.g., National Society of Black Engineers, Society of Women Engineers) are more commonly found at HBCUs (Brazziel & Brazziel, 1997), thereby increasing opportunities for community building and networking among African American STEM majors. African American STEM majors at HBCUs also benefit from more faculty interaction at these institutions (Allen, 1992), which is believed to positively impact student persistence and degree attainment (Astin, 1977, 1993). Student-faculty ratios are lower at HBCUs (Kim, 2002), which facilitates student-faculty contact. However, it is not just a question of numbers. Black students at HBCUs benefit from more nurturing student-faculty relations at HBCUs (Allen, 1987), perhaps due to the large percentages of African American faculty at HBCUs compared to the national average. Particularly for those students majoring in STEM fields, the “role models in excess” among faculty at HBCUs positively impact the educational aspirations and graduate school enrollment of African American students (Brazziel & Brazziel, 1997, p. 149). In sum, educational researchers attribute the success of HBCUs in fostering degree attainment among African American students, and STEM majors in particular, to their developmentally powerful and engaging academic environments (Watson & Kuh, 1996), 72 validating, supportive, and non-hostile campus climates (Hurtado & Carter, 1992; Rendón, 1994), and the frequency and quality of interactions with faculty, who are often African American themselves (Allen, 1992; Brazziel & Brazziel, 1997). As Allen (1987) explains, “where [African American] students are able to feel less ‘under siege’, the quality of their educational experiences and educational outcomes will be vastly improved” (p. 30). Women’s Colleges: Warming a ‘Chilly Climate’ for Female STEM Majors Like HBCUs, women’s colleges are characterized by high levels of success in terms of conferring STEM bachelor’s degrees to women, many of whom go on to earn graduate degrees in science and engineering. What accounts for the success of these institutions? The effectiveness of women’s colleges is best understood by considering the barriers typically standing in the way of women’s participation in STEM fields. Etzkowitz, Neuschatz, Kemelgor, and Uzzi (1994) describe three types of obstacles to entry into scientific careers for women that arise from (1) traditional socialization, (2) the structure of the academic system, and (3) discrimination. Females are socialized to believe that a scientific career is “antithetical” to being feminine and fulfilling ascribed functions such as childbearing and marriage (Ware & Lee, 1988). As Ware and Lee (1988) found, female colleges students who placed a higher value on having a family in the future were less likely to choose science as a major, illustrating that socialization places a high hurdle in the path of increasing the representation of women among scientists and engineers. Additionally, Holland and Eisenhart (1990) found that women science majors often abandon science-related careers due to the perceived dissonance between being a scientist and being a “feminine woman.” The lack of female role models among science and engineering faculty in academe (Chamberlain, 1988; Chesler & Chesler, 2002; Brainard & Carlin, 1998; Etzkowitz et al., 1994) also serves as a barrier to women’s participation in STEM fields. Finally, the “chilly,” often discriminatory, climate that is well documented in science classrooms and research environments (Chesler & 73 Chesler, 2002; Hall & Sandler, 1982), acts to deter women from entering scientific fields and lowers the persistence of women who initially choose to study science and engineering (Seymour & Hewitt, 1997). As Whitt et al. (1999) note, perceptions of the chilly climate among female students have a negative effect on cognitive outcomes. Much of the success of women’s college in the production of female STEM bachelor’s degrees is attributable to the ability of these institutions to remove the aforementioned barriers from their campuses. On women’s college campuses, the barriers constructed by socialization that can deter women from pursuing STEM fields are challenged by the very nature of the all- female student body. In a classroom surrounded by all women, the gendered nature of science and engineering is minimized. In essence, women’s colleges insulate female STEM majors from the wider social message that science is not ‘woman’s work,’ which contributes to the high levels of participation in science and engineering at these institutions. This message is reiterated to female STEM majors at women’s colleges by the presence of female faculty members in science and engineering fields. On average, women’s colleges have higher representations of female faculty members and administrators than coeducational institutions (NCES, 2005), thereby providing more female role models for women STEM majors and positively influencing academic success. As Tidball (1986) found, the proportion of female faculty has a significant, positive effect on the number of female students who pursue careers in science. This is not surprising considering the negative experiences interacting with male faculty often reported by female STEM majors (Belenky, Clinchy, Goldberger, & Tarule, 1986; Hall & Sandler, 1982). This gender-biased treatment is further exacerbated by the underrepresentation of female faculty or senior administrators in STEM fields (Seymour & Hewitt, 1997). Women’s colleges also remove barriers placed on the path to STEM by ‘chilly’ classroom climates and discriminatory practices. As many scholars have noted (Chamberlain, 74 1988; Holland & Eisenhart, 1992; Wolf-Wendel, 1998), women’s colleges, like their HBCU counterparts, offer more supportive educational environments, in which subtle discrimination (Chamberlain, 1988) is no longer a factor. The supportive environment of women’s colleges can eliminate many of the reasons often cited for attrition among female STEM majors, including “lack of self-confidence, poor advising, and not being accepted in their departments” (Brainard & Carlin, 1998, p. 374). Furthermore, the “warmed” classroom climates found on women’s college campuses act to increase in-class participation, foster high educational and career aspirations, boost self-confidence, and increase help-seeking behaviors among female STEM majors (Hall & Sandler, 1982). To summarize, women’s colleges are incredibly successful in terms of increasing degree attainment in STEM fields among female students. At women’s colleges, the participation rates in science and engineering are higher than at co-educational institutions, and all-female colleges produce a disproportionately high number of women who go on to earn a doctorate in science- related fields. The ability of women’s colleges to facilitate the success of female STEM majors is attributed to supportive academic environments, the large proportion of female faculty, and ‘warmer,’ less hostile classroom and campus climates. The very fact that these colleges are intended to serve women and shape their institutional and instructional practices to accomplish that aim enables their success in producing female scientists and engineers. Applying Lessons from HBCUs and Women’s Colleges to HSIs Though a significant amount of research has been done on the contributions of historically black institutions and women’s colleges in the production of scientists and engineers, Hispanic-serving institutions have not been the subject of much study. Due to the newness of the Hispanic-serving designation (Contreras, Malcom, & Bensimon, 2008; Laden, 2000, 2004) and the fluid nature of the demographically based label (Contreras, Malcom & Bensimon, 2008), 75 reliable statistics on the share of Latino STEM bachelor’s degrees conferred by these institutions are not widely available. Although the National Science Foundation (2005) has published data of the top baccalaureate origin institutions for Latino doctorates in science and engineering, these lists are compiled without considering Latino STEM degree holders schooled in Puerto Rico and those educated in the U.S. separately. This is problematic due to the differences in the cultural, social and economic contexts of Latinos in the U.S. and those in Puerto Rico. As such, the role of ‘mainland-based’ HSIs in conferring STEM degrees to Latinos remains unclear. In this study, I focus on the contributions of mainland-based HSIs in the production of Latino STEM bachelor’s degree holders in order to determine whether these institutions are comparable to HBCUs and women’s colleges in terms of acting as a pathway to STEM for their target populations. This study provides an empirical basis for the comparisons of HSIs to HBCUs and women’s colleges. Although Hispanic-serving institutions are often associated with other MSIs, any comparison between HSIs and HBCUs and women’s colleges ought to be made with caution due to the differing origins of these institutional types (Contreras, Malcom, & Bensimon, 2008; MacDonald, Botti, & Clark, 2007). Historically Black postsecondary institutions and women’s colleges were founded with the express mission of providing access to and fostering the success of their target populations (Brown & Davis, 2001). HSIs, on the other hand, earn their designation due to demographic factors 6 , rather than an institutional commitment to serving Latino students. In fact, many Hispanic serving institutions ‘closet’ their designations, omitting this from institutional mission statements and failing to produce equitable outcomes for Latinos, the very students they purport to serve according to the ‘HSI’ label (Contreras, Malcom, & Bensimon, 6 Three institutions were created prior to 1992 for the specific purposes of serving the educational needs of Latino students: Boricua College, an independent liberal arts college targeted towards Puerto Ricans in New York City; The National Hispanic University in San Jose, California, an independent four-year BA-granting institution; and Eugenio Maria de Hostos Community College, a two-year college in the South Bronx that is part of the City College of New York. 76 2008). In addition, although Latinos are represented among faculty in some HSIs at rates higher than the national average, a large proportion of these faculty members are foreign-born, and do not teach in STEM disciplines (NSF, 2005). Furthermore, the wide-ranging variability in Latinos’ share of student enrollment at four-year HSIs (i.e., from 25 to more than 90%) prevents any direct comparisons of their campus climates to the supportive campus environments at HBCUs and women’s colleges. It remains unclear whether HSIs play a comparable role to their HBCU and women’s college counterparts in terms of fostering Latino student success in STEM fields. The current study addresses this gap in the literature by analyzing a nationally representative sample of Latino STEM bachelor’s degree holders and degree-granting institutions. Enclaves of Success: The Contribution of Community College Programs to the Transfer of Latino STEM Majors Although community colleges are not designated formally as minority serving, they are de facto MSIs due to their high levels of minority enrollment. Community colleges are intended to provide access to higher education for those individuals who cannot or do not attend four-year institutions. While traditionally viewed as a vehicle for further educational opportunity by serving as an alternate, albeit longer, path to the baccalaureate degree, the community college sector is not fulfilling this promise (Chapa & Schink, 2006; Dougherty & Kienzl, 2006; Flores, Horn, & Crisp, 2006). Though there is emerging evidence that community college attendance does not disadvantage Latino students who successfully transfer to four-year institutions in terms of degree attainment (Melguizo, in press), transfer rates among those Latino students who aspire to earn a bachelor’s degree remain low (Adelman, 2005; Grubb, 1991). This suggests that this path to the baccalaureate through the community college is rife with obstacles. In many two-year institutions, targeted programs such as Mathematics, Engineering, and Science Achievement (MESA) and Puente aim to serve Latino students where the community 77 college falls short. Both MESA and Puente seek to facilitate the achievement of educationally disadvantaged students and increase their college participation and degree attainment by offering pre-college, community college and university programs (MESA Community College Program [MCCP], n.d.; Puente Project, n.d). The MESA Community College Program specifically targets those disadvantaged community college students interested in majoring in STEM fields by providing academic excellence workshops, career advising, college survival skills courses, and assistance through the transfer process (MCCP, n.d.). Though Puente is not limited to STEM majors, its aims are similar to the MESA program. Puente incorporates teaching, counseling, and mentoring to increase transfer and degree attainment among low-income and minority students in California community colleges. Although very few empirically sound large-scale evaluations of these programs have been done (Gándara & Bial, 2001; Swail, 2000; Tierney, 2002), many institution-specific tales of success are present in the literature (Kane, Beals, Valeau, & Johnson, 2004; Somerton et al., 1994). While these evaluations may be internally biased, they are often accompanied by data showing above average transfer rates for program participants (Haro, 2004; Kane et al., 2004; Malcom, 2006b, UCLA, 2004). Some researchers (Cooper, 2002; Gándara, 2002; Laden, 1999; Rendón, 2002) have attempted to understand why these programs might be effective in facilitating success for Latino students by examining practices as well as student experiences and making connections to existing theoretical frameworks regarding Latino student success. These studies aim to understand how MESA and Puente create enclaves of success within the same institutional environments that typically depress bachelor’s degree attainment and transfer rates among Latinos. While prior research on MESA and Puente is certainly limited in scope and depth, the few emergent themes illustrate that MESA and Puente are able to create validating environments 78 (Rendón, 2002) in which students receive institutional support (Grubb, Lara, & Valdez, 2002; Malcom, 2006b) to facilitate their successful transfer to four-year institutions (Cooper, 2002) and persistence to degree attainment (Rendón, 2002). As Rendón (2002) explains, students “need progressive and sustained assistance to ensure that they stay enrolled and graduate from college” (p. 642). MESA and Puente provide this assistance to educationally disadvantaged students by “promot[ing] involvement…[and]…affirm[ing] students as knowers and valuable members of the college learning community” (p. 645). The active offers of academic assistance, expectations of transfer communicated to students by counselors, and the opportunities to create social networks with peers and mentors common to both Puente (Cooper, 2002; Gándara, 2002; González & Moll, 2002; Grubb, Lara, & Valdez, 2002; Rendón, 2002) and MESA (Haro, 2004; Somerton et al., 1994) are just a few examples of the practices thought to contribute to the effectiveness of these programs. In a sense, these programs share some characteristics with the nurturing environments of HBCUs and women’s colleges that lead to success for their target populations. However, unlike HBCUs and women’s colleges, these environments have not been “institutionalized” and are limited to the confines of the programs. While the success of MESA and Puente in increasing the retention and persistence of minorities in STEM illustrates that institutional actions can make a difference, there remains a lack of knowledge regarding the ways in which the environments of community colleges external to these enclaves of success impact Latino STEM majors. Beyond MESA, Puente and other targeted programs, are community colleges acting as springboards to further educational attainment for Latino STEM majors? Unfortunately, existing literature provides few answers to this question. Among Latino STEM majors, the role that HSIs and community colleges play in terms of acting as a pathway to the bachelor’s degree remains unclear. Do they over-perform in a manner 79 similar to HBCUs and women’s colleges, or does the “accidental” way in which HSIs and community colleges have become Latino-serving mean that these institutions do not yet display the characteristics that allow HBCUs and women’s colleges to foster success among STEM majors in their target populations such as a Latino-friendly campus climate, a supportive academic environment and a large pool of Latino faculty who act as mentors? This study begins to address this question by describing the specific pathways used by Latino STEM bachelor’s degree holders and exploring how the patterns of achievement among these students vary by institutional pathway. The Decision to Attend Graduate School: Understanding the Effects of Indebtedness The final decision-making process addressed in this study concerns graduate school attendance. As with the selection of an institutional pathway, choices regarding borrowing to finance college, and the decision to major in a STEM field, factors from each layer in the ‘(dis)advantaging context’ framework (Figure 2.2) shape the decision to attend graduate school. Previous research has identified numerous individual characteristics, as well as factors in institutional, social, economic and higher education policy contexts that are associated with the decision to pursue graduate study. For example, class, race/ethnicity (Price, 2004), field of study (Bedard & Herman, 2005), institutional selectivity (Price, 2004), and labor market conditions (Bedard & Herman, 2005; Titus, 2007) have all been illustrated to be determinants of graduate school attendance. However, the relationship between debt accumulated through student loans and graduate school enrollment remains unclear. Characterizing the effects of indebtedness on graduate school enrollment is especially important given the rapid increase in debt accrued by college students, particularly among low- income and students of color (Price, 2004). Rising tuitions, the decline in the purchasing power of the Pell Grant, and slow growth in the appropriations for state and federal grant aid (College 80 Board, 2006; Dowd, in press; Heller, 2002; Hoxby, 2004; Hu & St. John, 2001) have caused loans to take center stage in the financial aid system. As a result, borrowing rates among all college students, particularly those attending private four-year institutions continue to rise, with greater numbers of students receiving loans from private lenders in addition to subsidized and unsubsidized loans via the federal government. Further, enrollment in postbaccalaureate study is central to increasing the representation of Latinos at STEM educational milestones beyond bachelor’s degree attainment (i.e., STEM doctorates, STEM faculty members). Thus, in the current study I determine the effects of debt on graduate school enrollment among Latino STEM bachelor’s degree holders. Below, I review the educational literature on the effects of debt on student outcomes in general and graduate school enrollment, in particular. These studies provide a murky picture of the impact of indebtedness on college outcomes due to mixed findings. I also address the methodological weaknesses of previous research and describe the ways in which the present study represents an improvement over traditional estimation techniques. The Effect of Loans on Undergraduate Student Outcomes Researchers attempting to estimate the effects of loans on persistence and degree attainment have used a variety of approaches and data sets at the national, state, and institutional levels. A review of the literature on the effect of financial aid, and loans in particular, on persistence and degree attainment reveals that the loans have been found to have positive, negative, and no significant effects on these educational outcomes. Several studies conclude that there is a positive relationship between loans and persistence and between loans and degree attainment (Dowd, 2004; Chen & DesJardins, 2006; Hu & St. John, 2001; Ishitani & DesJardins, 2002). Hu and St. John (2001) conducted a study of the effect of various financial aid packages (i.e., grant only, loans only, grants and loans, and other 81 packages) on the persistence of students enrolled in public universities in Indiana compared to those students who received no aid. By analyzing several years of data between 1990 and 1997, Hu and St. John hoped to capture the impact of financial aid on persistence in the context of the increased emphasis on loans for financing postsecondary education. Employing the method of logistic regression, Hu and St. John found that while financial aid had no significant effect on persistence in 1990-91, for Latino students in the 1993-94 cohort, loans, if they were paired with grants in financial aid packages, increased the probability of persistence compared to non-aid students. Interestingly, by 1996-97, each type of financial aid packages had a significant, positive effect on the persistence of Latino students compared to non-aid students, including loans. Hu and St. John’s study highlights the difficulty in assessing the effect of financial aid on persistence, as the multi-year analysis does not take into account the evolving nature of financial aid packages. Ishitani and DesJardins (2002) conducted an event history analysis of the effect of financial aid on persistence using the Beginning Postsecondary Students (BPS) national database in order to understand how these effects vary throughout a student’s tenure in college. They found that not only were students who received aid less likely to dropout, the size of the benefit to persistence provided by financial aid increased in the second and third years of college. While Ishitani and DesJardins’ study addresses the layer of complexity added to the discussion regarding the effects of aid on persistence by temporal considerations, limitations of the data used in the analysis (i.e., all types of financial aid were treated in the aggregate), prevented them from understanding how loans and grants might differentially impact this outcome. Dowd (2004) improved on this limitation in Ishitani and Desjardins’ (2002) study by including separate predictor variables for federal loans, grants (federal, state, and institutional need-based and non-need based grants), and work study. By considering different types of financial aid separately, Dowd (2004) was able to estimate the effects of loans on persistence 82 using the method of logistic regression. Loans were found to have a positive effect on persistence between the first and second years of college, but no significant effect on bachelor’s degree attainment (Dowd, 2004). Chen and DesJardins (2006) drew similar conclusions in their study of the effects of various types of aid on the persistence of students from low, middle, and high- income families. Loans were found to be negatively associated with student dropout (i.e., positively associated with persistence) for students in the BPS sample. Furthermore, Chen and DesJardins (2006) found these effects to be consistent across students in the three family income groups. In contrast, several other researchers (Braunstein et al., 2000; DesJardins et al., 2002; Titus, 2000 as cited in Dowd, in press; Stampen & Cabrera, 1988) have claimed that financial aid has no significant effect on persistence. Through their logistic regression analysis of students at Iona College, Braunstein, McGrath, and Pescatrice (2000) found that financial aid had no significant direct effects on the first-year to second-year persistence of freshmen at the institution, but the individual effect of loans was not considered. DesJardins, Kim and Rzonca’s (2002) findings supported the work of Braunstein et al. (2000), claiming that financial need had no significant effect on the odds of dropping out before completion of the first-year of college for students at the University of Iowa. However, like Braunstein et al. (2000), DesJardins et al. (2002) were not able to analyze the effects of loans on persistence separately, and did not consider the interaction of race and financial aid on persistence. Additionally, Braunstein et al. (2000) and DesJardins et al. (2002) consider the effects of aid on student persistence at one postsecondary institution, rather than focusing on a larger data universe such as the state or the nation. Titus (2000, as cited in Dowd, in press) estimated the effects of loans on college persistence at the state-level by analyzing the data from the University of Maryland system and 83 found that there was no significant relationship between the receipt of loans and persistence to the second year of college. An earlier study by Stampen and Cabrera (1988) also considered the effect of loans on persistence using statewide data from the University of Wisconsin system. These authors found that loans were insignificant to persistence; however, the fact that borrowers in their study tended to be from high-income families introduces a significant source of self- selection bias. Still other researchers find that loans have negative effects on college student persistence and degree attainment. In their study of the effect of loans on term-to-term persistence, St. John, Andrieu, Oescher, and Starkey (1994) found that students who received loans were less likely to return after the first-term. St. John and Starkey (1995) also found that financial aid in the form of loans was negatively associated with persistence, particularly among low-income students. Paulsen and St. John’s (2002) logistic regression analysis of the effects of financial aid on persistence by social class revealed that loans negatively affected persistence, particularly for low-income students. In an effort to do a more sophisticated analysis, DesJardins, Ahlburg, and McCall (2002) conducted an event history analysis to investigate the effects of loans on ‘stopout’ and to determine how the size of these effects vary temporally. The authors find that each $100 dollar increase in loan aid significantly increases the chances of stop out for college students, but that the size of this effect wanes with time (DesJardins, Ahlburg, & McCall, 2002). Interestingly, DesJardins, Ahlburg, and McCall (2002) found that loans had no direct effects on degree attainment, but posited that loans had negative indirect effects on degree attainment because loans increase the chances of stopout. The inconsistency of the findings regarding the effect of loans on persistence and degree attainment is also present in studies focusing specifically on students in community colleges. Cofer and Somers (2001) found that loan debt increased the persistence of community college 84 students, unless the level of debt was extremely high (greater than $7,000), in which case, the chances of persisting decreased. On the other hand, Dowd and Coury (2006) found that loans negatively affect the persistence of community college students, regardless of the level of debt. However, Dowd and Coury (2006) discovered no significant effect of loans on the degree attainment of community college students. Adding to the confusion is a prior study by Hippensteel, St. John and Starkey (1996), which found no significant effect of loans on persistence. The studies described above regarding the effect of loans on persistence and degree attainment of students in four-year colleges and universities and community colleges provide no definitive answer on whether the debt incurred from student loans has a positive, negative, or no significant effect on their term-to-term, or year-to-year re-enrollment. The mixed findings are likely due to differences in sample selection, the varying nature of the financial aid variables from study to study, and the many methodological limitations of the studies (Dowd, in press; Dowd & Coury, 2006). The Effect of Loans on Graduate School Enrollment Researchers have also attempted to understand the effect of loan aid on post- baccalaureate outcomes, such as graduate school enrollment directly after earning a bachelor’s degree. As more emphasis is put on loans to finance undergraduate education, it is important to consider how student indebtedness impacts student decisions to attend graduate school, particularly for Latinos in science-related fields as Latinos have been shown to graduate with higher levels of debt than their White counterparts (Price, 2004) and because graduate school is often essential for advancement in STEM fields. As with studies attempting to determine the effect of debt on persistence, those purporting to estimate the effect of borrowing on graduate school enrollment have mixed results. 85 Some researchers have found that debt has a positive effect on plans to attend graduate school (Bedard & Herman, 2005; Kim & Eyermann, 2006; Murphy, 1994), while others conclude debt deters graduate school enrollment (Tsapogas & Cahalan, 1996; Millett, 2003; Wilder & Baydar, 1991), and still others have found no significant relationship (Baum & Saunders, 1998; Millett, 2003; Nettles, 1989; Weiler, 1991, 1994). Murphy’s (1994) study of the High School and Beyond (HS&B) found that indebtedness had a small, but positive effect on students’ plans to attend graduate school. Kim and Eyermann (2006) measure the effects of borrowing on graduate school attendance prior to and after the 1992 Higher Education Act amendments that increased federal loan limits. Using logistic regression techniques, Kim and Eyermann (2006) found that prior to 1992, loan debt had no statistically significant effect on plans to attend graduate school. Subsequent to the 1992 changes in the HEA, Kim and Eyermann (2006) found that for students from low- and high-income groups, borrowing was not a significant factor in graduate school attendance. However, for middle-income students, undergraduate loan debt had a slight positive effect (0.4 percentage point increase in probability) on plans to attend graduate school. Bedard and Herman (2005) also conducted a study regarding predictors of graduate school attendance among STEM bachelor’s degree holders. Although their analysis included undergraduate debt as an independent variable, they were primarily interested in the effects of labor market indicators, such as unemployment rates. Nonetheless, Bedard and Herman (2005) found that those students with higher undergraduate loan balances were more likely to enroll in graduate programs, all else being equal. Other researchers (Tsapogas & Cahalan, 1996; Wilder & Baydar, 1991) have reported that undergraduate loan debt has a significant negative effect on application to or enrollment in graduate school. Wilder and Baydar (1991) found that among a sample of students who took the Graduate Record Examinations (GRE), undergraduate debt had a small negative effect on 86 applying to graduate school, but not on enrollment. Tsapogas and Cahalan’s (1996) analysis of the 1993 National Survey of Recent College Graduates reveals that loan debt had a significant negative effect on the chances of enrolling full-time in graduate school for males in the sample, but the amount of debt had only a small negative effect. Finally, in her logistic regression analysis of Baccalaureate and Beyond (B&B) data, Millett (2003) found that students with debt greater than $5,000 had lower odds of applying to graduate school compared to students with no loan debt. However, among students who applied and were accepted to graduate school, undergraduate loan debt had no statistically significant influence on enrollment in a graduate program. These three studies reveal that the relationship between borrowing and graduate school attendance is complicated, with differential impacts on sub-population of students (e.g., women), and varying directional effects depending on the how far along a student has progressed in the graduate school decision making process. In contrast, other studies (Nettles, 1989; Schapiro, O’Malley, & Litten, 1991; Weiler, 1991, 1994) failed to find a significant relationship between loan debt and graduate school enrollment. Nettles’ (1989) multiple institution study examined the relationship between entry- level debt burdens among White, Latino, and African American graduate students. Nettles found that Latinos and African Americans tended to have higher debt burdens than White students, and that those students who delayed graduate school entry had lower levels of undergraduate debt. However, Nettles (1989) found no significant relationship between debt and the decision to enter graduate school directly after baccalaureate attainment. Similarly, Schapiro, O’Malley, and Litten (1991) found that among students with high levels of debt ($10,000 or higher in 1982-84, and $12,500 or higher in 1989), undergraduate loan debt had no influence on plans to enroll in graduate school in the sciences. Finally, in two studies of the HS&B national dataset, Weiler (1991, 1994) found that the level of debt from undergraduate borrowing had no statistically 87 significant effect on graduate school attendance. Although these researchers conclude that loans do not influence graduate school plans, there is evidence that concerns about overborrowing may enter into students’ decisions to attend graduate school. For example, over 40 % of respondents to the 1997 National Student Loan Survey (NASLS) who did not attend graduate school stated that their level of undergraduate debt was “extremely important” or “very important” in their decision not to attend, while 28 % of these respondents reported that debt burden was not important to their decision (Baum & Saunders, 1998). As with the literature regarding the effect of loans on college student persistence, prior studies regarding the impact of undergraduate debt burden on graduate school enrollment provide no clear assessment. Again, the discrepancies found amongst the studies of the relationship between loans and graduate school attendance are likely due to the multiple datasets used, the differing contexts in which these studies were conducted (e.g., studies conducted prior to the prevalence of loans may no longer be applicable), the different treatments of debt (absolute versus relative), and severe methodological limitations. Furthermore, because most of these studies, with few exceptions, do not provide an estimation of the effect of loans on the graduate school attendance of Latino students, nor Latino STEM majors, the present study fills in a large gap in the existing literature by estimating the effects of loans on the graduate school enrollment of Latino STEM majors using a methodology intended to address the shortcomings of previous research. Methodological Limitations of Studies Estimating the Effect of Loans on Educational Outcomes With few exceptions, the studies that purport to estimate the effect of loans, or indebtedness, on educational outcomes such as persistence and graduate school enrollment suffer from a methodological flaw. The vast majority of these studies use single-stage logistic regression modeling to estimate the effects of the independent variable (i.e., having a loan, or the amount of 88 undergraduate loan debt), on the outcome of interest (e.g., persistence, degree attainment, graduate school enrollment), while controlling for a number of demographic, student and institutional characteristics. However, many researchers legitimately argue that this method is not capable of controlling for self-selection bias or dealing with issues of endogeneity (Heckman, 1979; Millimet, 2001; see also Dowd, in press; Titus, 2007). Because many factors contribute to a student’s decision to use loans as a means of financing college including but not limited to expected earnings, educational expectations, and self-efficacy, participation in a loan program is not a random phenomenon. In other words, students ‘self-select’ themselves into loan programs. This is a significant problem when trying to attribute differences in educational outcomes to participation in loan programs or to the amount of undergraduate debt, because the same student-level characteristics that may cause an individual to elect to use loans as a financing strategy, may also positively impact educational outcomes. Further, because these student-level characteristics that lead a student to participate in a loan program are often unobserved and not included in national datasets, differences in educational outcomes among respondents in the sample might be falsely attributed to participation in a loan program, when in reality, the unobserved characteristics (e.g., expectations, self-efficacy) that cause the differences in loan program participation rates are also responsible for the variation in educational outcomes (Dowd, in press). While logistic regression analytical techniques are able to “control” for other relevant variables (e.g., race, gender, age), the single- stage models often employed in research regarding the effects of loans on educational outcomes cannot account for self-selection, endogeneity, or selection on unobservables (Titus, 2007). As a result, the validity of the findings from prior research on the effect of loans on educational outcomes is severely undermined. 89 How can the non-random nature of participation in loan programs be accounted for so that self-selection bias no longer poses a methodological issue? The ideal means would be through experimental studies, in which students are randomly assigned to loan programs. Similar to medical trials, such a study would create ‘treatment’ and ‘control’ groups from which the effect of the ‘treatment,’ i.e., indebtedness, could accurately be estimated without problems of self- selection. Experimental studies are considered the “gold standard,” and are endorsed by the Institute for Education Science (IES) in the U.S. Department of Education (U.S. Department of Education, n.d.(a), n.d.(b)). However, while widely used in psychology, medicine, and increasingly used in K-12 education research, it is neither economically nor ethically practical to conduct such experimental designs in higher education research with respect to loans. As such, alternative approaches must be used to overcome problems of self-selection. Several researchers in higher education propose adopting quasi-experimental techniques from other disciplines designed to address problems of self-selection in situations where experimental designs are not possible (Agodini & Dynarski, 2002). Such techniques include natural experiments in which students with similar background characteristics are awarded very different aid packages due to financial aid policies at their home institutions. In other words, the differences in the aid packages between the groups are due to exogenous factors. These two groups of students can be thought of as a ‘treatment’ group and a ‘control’ group based on the type of financial aid awarded to them, and researchers can then analyze the differences in a particular educational outcome of interest to uncover the ‘treatment’ effect on that outcome. Bettinger (2004) used this technique to estimate the effects of Pell grants on student retention. Another quasi-experimental approach that has slowly found its way from economics to higher education research is propensity score matching (Rosenbaum & Rubin, 1983; see also Becker & Ichino, 2002; Caliendo & Kopeinig, 2005; Ham, Li, & Regan, 2006). Propensity score 90 matching (PSM) uses a counterfactual framework (Titus, 2007) to create ‘treatment’ and ‘control’ groups based on propensity scores, or the conditional probability of being assigned to treatment based on observed covariates. In essence, PSM reduces covariates to a single composite indicator (i.e., the propensity score) and ensures the creation of balanced comparison groups. Returning to the application to research on the effects of loans, using PSM will allow a researcher to calculate the propensity to borrow based on predictors (e.g., race, gender, SES, parental education, type and control of baccalaureate origin institution, etc.) using a probit or logit model. Next, cases are matched on the propensity score to form two groups: the ‘treatment’ group which did participate in a loan program, and the ‘control’ group which had the same probability of participating in the loan program as those in the treatment group, but elected not to take the ‘treatment.’ By comparing the outcomes of matched pairs in the ‘treatment’ and ‘control’ groups, the average treatment effect (ATE), average treatment effect on the treated (ATT), and average treatment effect on the untreated (ATU) can be calculated. Several educational researchers have used this technique to address problems of self- selection bias. Titus (2007) used propensity score matching techniques to estimate the effects of bachelor’s and master’s degree attainment on wage earnings. Melguizo, Kienzl and Alfonso (2007) and Kurlaender and Long (2007) employ PSM to estimate the effects of community college attendance on baccalaureate degree attainment. Propensity score matching has also been used to estimate the effects of various K-12 treatments such as English as a Second Language education (Callahan, Wilkinson, & Muller, 2008) and ability grouping (Wang, 2007). Quasi-experimental techniques, and propensity score matching in particular, are more desirable to the typical regression techniques employed by researchers attempting to understand the effects of loans on educational outcomes due to their ability to address problems of self- selection bias. When these problems are sufficiently addressed, the causal claims often 91 erroneously drawn from logistic regression modeling are much more sound. However, propensity score matching is not a cure-all to the methodological limitations of these studies. More times than not, there will be important “unobservables” that are not accounted for in national datasets, and thus, cannot be included in the analysis. Returning to the above example, if ‘unobservable’ characteristics are important in establishing an individual’s conditional probability of receiving ‘treatment,’ i.e., participating in a loan program, the propensity score estimates derived from the probit or logit model may not be accurate. So while the problem of self-selection is addressed via PSM, the remaining issue of ‘unobservables’ reduces the validity of any causal claims made. However, researchers can try to minimize this problem by including as many predictor variables as possible in the model to calculate the propensity scores of the individuals in the sample. In sum, propensity score matching and natural experiments are two techniques currently employed by educational researchers to identify the effect of various educational ‘treatments’ on outcomes. While many in higher education have begun to use these techniques to estimate how non-random educational treatments impact outcomes such as wages, degree attainment, and persistence, these techniques have not yet been employed to understand how undergraduate loan debt impacts graduate school enrollment. Natural experiments offer an interesting approach to dealing with issues of self-selection, yet one’s ability to use this approach is limited to her ability to exploit exogenous variation in students’ financial aid packages. This type of variation may not commonly occur in national datasets. Propensity score matching seems more promising than natural experiments because it can be conducted on a sample of students in a single national or institutional survey, provided that sample size is adequate. In many of the studies that employ PSM, both regression analyses and propensity score matching techniques are used to understand if PSM would result in more accurate point estimates. In theory, propensity score matching is advantageous to regression techniques due to the capability of decomposing treatment effects into 92 ATE, ATT, and ATU, and by addressing issues of endogeneity. However, selection on unobservables remains a potential problem in propensity score matching techniques and steps to test the robustness of PSM models ought to be undertaken. Chapter Summary My study of Latino STEM bachelor’s degree holders centers on the effects of key contextual factors on four decision-making processes: (1) selection of an institutional pathway; (2) formulation of a college financing strategy and the decision to borrow; (3) the selection of a STEM field of study; and (4) the decision to enroll in graduate school. As I argue in this chapter, Latino students are situated in contexts with potentially advantaging and disadvantaging aspects that shape and constrain educational decisions. The literature discussed in this chapter underscores the important role that students’ multiple environments play in shaping their patterns of educational access and outcomes. Similarly, students’ individual characteristics along with the institutional context, higher education and financial aid policy contexts, and social, political and economic contexts in which they are situated interact to affect Latino student achievement in STEM. Using the methods described in the following chapter, the present study critically examines the relationships between key contextual factors and the patterns of participation and attainment of Latino STEM bachelor’s degree holders. 93 CHAPTER THREE RESEARCH DESIGN AND METHODOLOGY This study explores the problem of the underrepresentation of Latinos among bachelor’s and graduate degree holders in STEM fields by characterizing the institutional pathways used by Latino STEM bachelor’s degree holders, examining the college financing strategies employed by these students and determining the effects of indebtedness on graduate school enrollment among this population. In particular, this study highlights the importance of the community college sector and Hispanic Serving colleges and universities, the two institutional types with the highest concentrations of Latinos, in the education of Latino STEM degree holders. The study also examines the patterns of financial aid and the underlying college financing strategies used by Latino STEM degree holders, and determines the relationship between student loan debt and graduate school enrollment. By tackling these three, interrelated questions, I describe the institutional and financial aid pathways used in by Latinos who have been successful in terms of earning the STEM B.S. degree. Much of the previous research on Latino STEM degree holders fails to systematically explore their institutional and financial aid pathways through the prism of the social and policy contexts in which these students are situated. Further, while researchers have previously attempted to determine the relationship between indebtedness and graduate school enrollment, these studies are flawed for several reasons: (1) they have operationalized debt in an absolute sense, neglecting the socially constructed, relative nature of debt (McDonough & Calderone, 2004); (2) they have failed to consider the interaction of indebtedness and race/ethnicity; and (3) they rely solely on regression analyses which are plagued by problems of self-selection. In light of the limitations of the existing research on the educational progression of Latino students summarized above and detailed in the previous chapter, I use the methods of propensity 94 score matching, logistic regression, latent class analysis and descriptive statistics to address the research questions below. Research Questions Importance of Community College Transfer and Hispanic Serving Institutions to Latino STEM Baccalaureates (1) Nationally, what proportion of Latino STEM baccalaureate recipients earned associate degrees prior to transfer to the four-year sector? (2) What are the institutional characteristics of the four-year institutions from which Latino associate degree earners graduate, according to HSI status, selectivity, type (e.g. liberal arts or research university), and control (public/private)? (3) How do these characteristics compare with the institutions attended by Latino STEM bachelor’s degree holders who did not earn an associate degree? (4) How does the distribution of students across these institutional characteristics vary by state and region and by Latino ancestry for Mexican American/Chicano, Puerto Rican, and Cuban students? (5) How do demographic characteristics of Latino STEM baccalaureate degree holders who earn associate degrees differ from non-associate degree earners? (6) And, what is the distribution of Latino associate degree holders across different STEM fields of study in comparison to those students who do not earn an associate degree prior to attending a four-year college? College Financing Strategies and Effects of Borrowing on Graduate School Enrollment (1) How does the use of various forms of college financial aid and financing strategies such as scholarships, loans (institutional and familial), college work study, earnings, and 95 employer support vary by institutional pathway, national origin, and other demographic characteristics among Latino STEM bachelor’s degree holders? (2) What are the effects of debt on graduate school enrollment among Latino STEM bachelor’s degree holders? Research Approach Social science methodological texts characterize quantitative inquiry as that in which an “objective, value-free” researcher seeks to make “generalizations leading to prediction, explanation, and understanding” (Creswell, 1994, p. 5). Schrag (1992) emphasizes the desire of the quantitative researcher to verify existing theories through the construction of models. This description of quantitative research reflects the positivistic tradition out of which this approach was born. In recent years, educational researchers have articulated a critical approach to quantitative inquiry (Price, 2004; Stage, 2007). Quantitative criticalists (Stage, 2007) use quantitative methods to explore or investigate critical questions in the interest of bringing about equity for typically underserved populations. While the methods employed by traditional quantitative researchers and critical quantitative researchers do not differ, the motivation, purpose and goals of such research are distinct. Critical quantitative researchers aim to “use data to represent educational processes and outcomes on a large scale to reveal inequities and to identify social or institutional perpetuation of systemic inequities in such processes and outcomes” (Stage, 2007, p. 9). In short, critical quantitative researchers use statistical methods to understand the educational experiences of those individuals who may not conform to existing models that were constructed based on the experiences of the majority. 96 I describe this study as critical quantitative inquiry because of the questions I pose and the way in which I carry out the analysis. The educational problem at the focus of the current study, the equitable representation of Latinos in STEM related fields, is an important issue not only because of its implications for the economic health of the U.S., but also because it pertains to the ongoing quest for social justice for historically disadvantaged and disenfranchised groups. Similarly, the specific research questions at the center of this investigation are informed by the unique, nested contexts in which Latinos are situated. By exploring the contribution of Hispanic Serving Institutions and community colleges to the production of Latino scientists and engineers, the college financing strategies employed by these students, and the relationship between indebtedness and graduate school enrollment, the current study problematizes the tacit assumptions of what we know about HSIs and community colleges in terms of providing access to Latino STEM bachelor’s degree holders and the borrowing habits of these students. The methods described in this chapter are applied to examine Latino STEM bachelor’s degree holders as a unique population. The educational experiences and outcomes of Latinos ought to be studied and understood in their own right, without constant comparisons to whites as the benchmark. Furthermore, the intra-group ethnic, socioeconomic and generational diversity of Latinos in the U.S. precludes treating this group as a monolith when seeking to understand their patterns of participation in higher education. For these reasons, my analyses focuses on Latinos distinctly to provide greater understanding of the educational progression of Latino STEM majors, but also how these experiences might vary based on socioeconomic status, first- generation status, and national origin. In sum, I seek to further the current understanding of the educational experiences of Latino STEM majors through rigorous quantitative analyses of a nationally representative database. In particular, I identify the institutional pathways most used by Latinos who earn 97 bachelor’s degrees in STEM fields and examine the role of community colleges and Hispanic Serving Institutions in the production of Latino STEM degree holders. I also characterize the college financing strategies used by these students and explore the impact of indebtedness on their graduate school enrollment. My motivation is not to simply confirm existing models or theories; rather, I aspire to understand how the unique historical, social, cultural and economic contexts of Latinos in the United States work to shape their educational experiences in the sciences and related fields. The remainder of this chapter is organized as follows: I begin by describing the primary and secondary data sources that I used in the current study, the key variables of interest, and the selection criteria for the sub-sample. I then turn to a discussion of the steps taken to prepare and manage the data, including the merging of the primary and secondary data sources. Third, I explain the descriptive statistical methods I used to address the first set of research questions regarding the institutional pathways of Latino STEM bachelor’s degree holders. Following this discussion, I describe the method of latent class analysis, which I used to characterize the college financing strategies employed by the population at the center of this study. I then discuss the means through which I modeled indebtedness as a policy treatment, conceptualizing debt as a relative construct, and the ways in which I employed the methods of logistic regression analysis and propensity score matching to determine the relationship between indebtedness and graduate school enrollment among Latino STEM B.S. degree holders. I conclude by describing two types of sensitivity analyses that I conducted to test the robustness of the model. Data To address the research questions outlined above, I analyzed the 2003 National Survey of Recent College Graduates (NSRCG) database enhanced with institutional-level data from the 2002-2003 College Board Annual Survey of Colleges and Universities, the Institute for College 98 Access and Success (TICAS), the Integrated Postsecondary Education Data System (IPEDS), and Barron’s Profiles of American Colleges. Incorporating these five data sources allowed me to consider both individual-level and institutional-level variables in the analyses. 2003 National Survey of Recent College Graduates The procedures for the analyses involved obtaining restricted data files from the Division of Science Resources Statistics (SRS) of the National Science Foundation. In an effort to compile information regarding the education, employment, and demographic characteristics of the nation’s scientists and engineers, the National Science Foundation (NSF) sponsors three national surveys of individuals who have earned an undergraduate or graduate degree in science, engineering or a science-related field. The National Survey of Recent College Graduates (NSRCG) is administered every two years to a nationally representative sample of individuals who have either earned a bachelor’s or master’s degree in the two academic years preceding the survey reference date. The 2003 NSRCG, which is used in these analyses, provides information on individuals who earned an STEM bachelor’s or master’s degree from a U.S. institution between July 1, 2000 and June 30, 2002. Sample design, data collection, and survey response rates. The 2003 NSRCG sample was identified in two stages: first, a nationally representative sample of 300 institutions was selected, with the top 85 STEM baccalaureate producers selected with certainty and the remainder selected from all institutions awarding STEM bachelor’s degrees with probability proportional to size 7 . In the second stage, a stratified sample of students who earned STEM baccalaureates from these institutions was selected. The sample of students selected from each institution was stratified by four domains: (1) the type of degree earned during the survey reference period (i.e., bachelor’s or 7 The composite measure of size is “a linear combination of desirable sampling rates for the domains of interest and the number of graduates reported for schools in each domain” (Jang & Lin, 2007, p. 1). 99 master’s degree); (2) race/ethnicity; (3) gender; and (4) field of study (Jang & Lin, 2007). A total of 18,000 graduates were surveyed. A 66 % (non-weighted) response rate yielded 10,831 respondents. This sample, weighted for non-response, represents a national population of STEM Bachelor of Science (B.S.) degree holders. NSRCG Variables of interest. Individuals in the 2003 NSRCG sample responded to survey items regarding their educational experiences and aspirations, employment situation, work-related experiences, and demographic background. This dataset represents a rich source of information regarding the characteristics of those individuals who have earned an STEM B.S. degree in the previous two years. Due to the frequency of survey administration, these data reflect recent trends in the educational experiences of the nation’s science degree holders and are ideal for statistical analyses, including that described herein. Several variables from the 2003 NSRCG dataset are of particular interest for addressing the research questions in the study and are listed in Table 3.1 below. 100 Table 3.1. 2003 National Survey of Recent College Graduates (NSRCG) Primary Variables of Interest Variable Name Description Student Demographic Variables Age (AGE) Age during survey reference week Gender (GENDER) F=Female; M=Male Race/ethnicity (RACETHM) 1=Asian, non-Hispanic ONLY; 2=American Indian/Alaska Native, non-Hispanic; 3=Black, non-Hispanic ONLY; 4=Hispanic, any race; 5=White, non-Hispanic ONLY; 6=Non-Hispanic Native Hawaiian/Other Pacific Islander ONLY; 7=Multiple Race Hispanic Origin (HISPANIC) Are you Hispanic [or Latino]? N=No; Y=Yes Hispanic Category (HISPCAT) Which of the following describes your Hispanic origin or descent? 1=Mexican, Mexican American; 2=Puerto Rican; 3=Cuban; 4=Other Hispanic; L=Logical Skip Mother’s education level (EDMOM) What is the highest level of education completed by your Mother [Stepmother or female guardian]? 1=Less than High School; 2=High School diploma or equivalent; 3=Some college, vocational, or trade school; 4=Bachelors degree (e.g., BS, BA, AB); 5=Masters degree (e.g., MS, MS, MBA); 6=Professional degree (e.g., JD, LLB, MD, DDS); 7=Doctorate (e.g., PhD, DSc, EdD, etc.); 7=Not applicable Father’s education level (EDDAD) What is the highest level of education completed by your Father [Stepfather or male guardian]? 1=Less than High School; 2=High School diploma or equivalent; 3=Some college, vocational, or trade school; 4=Bachelors degree (e.g., BS, BA, AB); 5=Masters degree (e.g., MS, MS, MBA); 6=Professional degree (e.g., JD, LLB, MD, DDS); 7=Doctorate (e.g., PhD, DSc, EdD, etc.); 7=Not applicable Educational Variables Bachelor’s degree- granting institutional code (BAINCD) Institution code (six-digit UnitID) for school awarding first bachelor’s degree Associate degree holder (ASDGRI) Do you have a 2-year associate degree? N=No; Y=Yes Community college attendance (COMCOLI) Have you EVER taken courses at a community college? N=No; Y=Yes Year of first bachelor’s degree (BAYR) Year of first bachelor’s degree 101 Table 3.1, Continued Variable Name Description Educational Variables (continued) Field of major (NBAMEMG) Field of major for first bachelor’s group [major group]: 1=Computer and mathematical sciences; 2=Biological, agricultural and environmental sciences; 3=Physical and related sciences; 4=Social and related sciences; 5=Engineering; 6=S&E-Related fields; 7=Non S&E fields Undergraduate GPA (UGGPA) Using a 4-point scale, what was your overall UNDEGRADUATE grade point average [GPA]? 1=3.75–4.00 [Mostly A]; 2=3.25–3.74 [About half A/half B]; 3=2.75–3.24 [Mostly B]; 4=2.25–2.74 [About half B/half C]; 5=Less than 2.24 [Mostly C]; 6=Have not taken courses for which grades are assigned STEM Graduate degree enrollment (TCDRG1) Toward what degree are you working? 0=No degree; 1=Bachelor’s; 2=Master’s; 3=Doctorate; 4=Professional; 5=Other; L=Logical Skip STEM Graduate degree(s) Obtained (DGRDG) Highest degree type: 1=Bachelor’s; 2=Master’s; 3=Doctorate; 4=Professional Financial Support/Aid Variables Earnings (UGFERN) Financial support for undergraduate degree from earnings: N=No; Y=Yes Employer Support (UGFEM) Financial support for undergraduate degree from employer: N=No; Y=Yes Grants (UGFGRN) Financial support for undergraduate degree from tuition waivers, fellowships, grants and scholarships: N=No; Y=Yes Loans from Government and Institutional Sources (UGFLN) Financial support for undergraduate degree from Loans from school, banks, and government: N=No; Y=Yes Loans from Parent, other family (UGFPLN) Financial support for undergraduate degree from Loans from parents or other relatives: N=No; Y=Yes Assistantship/Work Study (UGFAST) Financial support for undergraduate degree from assistantships or work study Other Source (UGFOT) Financial support for undergraduate degree from other source: N=No; Y=Yes Amount Borrowed to finance undergraduate degree (UGLOANR) TOTAL amount you have BORROWED to finance undergraduate degree: 1=Did not earn a degree at this level; 2=None; 3=$1–5,000; 4=$5,001–10,000; 5=$10,001–15,000; 6=$15,001–20,000; 7=$20,001–25,000; 8=$25,001–30,000; 9=$30,001–35,000; 10=$35,001 or more Source: NSF 2003 National Survey of Recent College Graduates (NSRCG). 102 Key variables derived from the NSRCG. The key dependent, or outcome, variable for the logistic regression and PSM analyses is graduate school enrollment in STEM fields, which was defined as studying full- or part-time towards a STEM graduate degree or having earned a master’s degree at the time of the NSRCG survey. This outcome variable was coded as a dichotomous variable (0=non-attendance; 1=graduate school enrollment/attendance). Key educational attributes derived from the NSRCG that served as covariates in the analyses included holding an associate degree and being of nontraditional student status (i.e., being age 25 or older at receipt of first bachelor’s degree). The NSRCG contains two variables, COMCOLI and ASDRGI, which were used to determine if a respondent earned an associate degree at a community college. For the purposes of my analyses, I defined community college students narrowly as those who attended community college and earned an associate degree (i.e., COMCOLI=1 and ASDRGI=1). This approach is necessitated by the fact that the variable in the NSRCG indicating community college attendance (COMCOLI) does not indicate the number of units earned. The approach used in the current study excludes some students who used the community college as a pathway to a STEM bachelor’s degree via transfer, as many students do not earn an associate degree before transferring, however, it also allows for the identification of community college students and transfers with greater certainty. The results of the present study, therefore, reflect the population of transfer students who earned associate degrees. Nontraditional student status was calculated by subtracting the year of birth from the year of bachelor’s degree conferment. If this figure was greater than or equal to 25, that respondent was assigned a value of 1 for the newly created variable, NONTRAD. Secondary Data Sources: Institutional-level Variables Several institutional characteristics that are important to the financial aid choices of students and students’ likelihood of graduate school attendance were added to each survey 103 respondent’s record in the 2003 NSRCG, thereby facilitating the consideration of individual and institutional-level factors related to the use of particular educational pathways, college financing strategies, and the likelihood of graduate school attendance. I used the UnitID code of the baccalaureate-granting institution in each individual record as indicated by the variable, BAINCD in the NSRCG, to match institutional characteristics of interest drawn from four secondary sources: the 2003 and 2004 College Board Annual Survey of Colleges and Universities, the Institute for College Access and Success (TICAS) Economic Diversity of Colleges dataset, Barron’s Profiles of American Colleges, and the Integrated Postsecondary Education Data System (IPEDS). I discuss these secondary data sources and the key variables I drew from each below. College Board Annual Survey of Colleges and Universities. The College Board Annual Survey of Colleges and Universities is distributed annually to U.S. colleges, universities, vocational/technical, and graduate schools, in order to compile information related to admissions, enrollment, financial aid, faculty, institutional resources, and academic programs of study. Key institutional variables from the College Board data included the mean per-borrower cumulative indebtedness for the 2001 and 2002 graduating classes, the percentage of students who borrowed, and the in-state and out-of-state tuition levels for the 2000-01 and 2001-02 academic year. The Institute for College Access and Success Economic Diversity data set. The Institute for College Access and Success’ Economic Diversity of Colleges (TICAS) website (TICAS, 2007) provides institutional-level data on student demographic characteristics, socioeconomic backgrounds, and patterns of student loan use for 3,000 accredited colleges and universities. The student loan usage variable, ‘mean cumulative per borrower indebtedness,’ for the 2000-01 academic year was used to supplement data in the case of institutions for which the College Board data was incomplete. The student debt data made publicly available by TICAS were derived from the Peterson’s Undergraduate Financial Aid and Undergraduate Databases. 104 Barron’s Profiles of American Colleges. Barron’s index of academic competitiveness for college admission describes the selectivity of U.S. four-year postsecondary institutions. Academic competitiveness is defined as a composite measure of the median entrance examination scores (SAT and ACT), high school class rank of entering freshmen, minimum GPA requirements for admission and the percentage of applicants to the freshmen class admitted. Barron’s academic competitiveness index has been widely used throughout the higher education literature as a standard definition of institutional selectivity (e.g., Dowd, Cheslock, & Melguizo, in press). The scale has ten categories of selectivity: non-competitive; less competitive; competitive; competitive plus; very competitive; very competitive plus; highly competitive; highly competitive plus; most competitive; and special 8 . Integrated Postsecondary Education Data System. The Integrated Postsecondary Education Data System (IPEDS) is the primary data collection system for the National Center for Education Statistics (NCES). It is from these data regarding Latino student enrollment that the federal government makes the determination if a postsecondary institution qualifies as a Hispanic Serving Institution (HSI). As such, institutions that qualify as HSIs are designated in the IPEDS data. Because IPEDS serves as the core source of postsecondary institution data, I determined whether a NSRCG respondent’s bachelor’s degree-granting institution was an HSI at the time of graduation using IPEDS. Data Preparation and Management Merging Institutional and Individual-Level Data To prepare the data for the analysis, I merged data from the four secondary data sources to the primary NSRCG database by matching on the six-digit institutional code (UnitID) for each 8 Institutions classified as ‘special’ are those with specialized programs of study whose admission requirements are generally based on non-academic criteria such as evidence of talent of special interest in the field (Barron’s Educational Series Inc., 2004). 105 respondent’s baccalaureate granting institution and adding the variables of interest. As stated in the previous section, these merged variables included institutional attributes (e.g., tuition, selectivity, Hispanic-Serving status) and financial aid indicators (e.g., mean per borrower cumulative undergraduate debt among 2001 and 2002 graduates, percent of undergraduates borrowing). Sample Selection Following the merging of institutional data with the NSRCG, cases that fulfilled the following selection criteria were indicated as selected by setting a newly created dichotomous variable, SELECTED, to a value of one: (1) Individual earned their Bachelor’s degree in the United States (i.e., earned B.S. from an institution within the 50 U.S. states or Washington, DC). (2) Individual record contained a valid institutional code for the baccalaureate-granting college or university. (3) Individual earned a Bachelor’s degree in the years 2000-01 or 2001-02 9 . Cases that failed to meet the above criteria were assigned a value of zero for the variable SELECTED. These unselected cases were not included in the analysis; however, they were not deleted from the merged database so that I could accurately account for design effects due to the complex sample of the NSRCG. Application of the selection criteria resulted in an analytical sample size of n=7,700 (71.1% of original cases). Latinos’ proportion of the analytical sample was 13.9% (n Latino =1,065), compared to their 14.5% proportion in the full NSRCG sample. 9 The NSRCG data contains records for individuals who earned a STEM bachelor’s or master’s degree in the years 2000-01 or 2001-02. Those who earned the B.S. before 2000-01 academic year are not considered in this analysis. Though the elimination of these cases reduced the analytical sample, I took this step to enable a more consistent interpretation of debt to college costs and financial aid policies at the time of undergraduate study. 106 Missing Data The cases in the analytical sample were complete with the exception of the variables indicating the 2001 and 2002 mean per borrower cumulative indebtedness of the baccalaureate- granting institution. These values were missing at random (MAR), meaning that they were not randomly distributed across the all observations, but were missing randomly within more than one sub sample of institutions (Allison, 2001). Due to the MAR condition, imputation of the missing values was an appropriate step to take. Missing values for the key variables ‘2001 mean per borrower cumulative undergraduate debt’ (17% missing) and ‘2002 mean per borrower cumulative undergraduate debt’ (29% missing) were imputed using random regression imputation (i.e., predictive regression imputation with the addition of a stochastic component through the residual terms). The imputation of the variable, 2001 mean debt, involved modeling this continuous variable as linear function of tuition, institutional selectivity, geographical location of the institution (i.e., state), and control: ε β + + + + = ) ( debt cumulative 2001 Control State y Selectivit Tuition i (1) I replaced the missing values with the predicted values plus a randomly selected residual term from those cases with non-missing values for 2001 mean debt. The variable 2002 average debt was imputed using the same model, with the addition of the 2001 average debt as a predictor: ε β + + + + + = ) debt cumulative ( debt cumulative 2001 2002 Control State y Selectivit Tuition i (2) 107 As with the 2001 values, I replaced missing values for 2002 mean debt with the predicted values plus a randomly selected residual term from those cases with non-missing values for the 2002 mean debt. In the table below, I provide the summary statistics (i.e., mean, standard deviation, minimum and maximum values) of key variables in the analysis for the Latino subsample (n Latino =1,065). 108 Table 3.2. Demographic, Educational, B.S. Degree-granting Institutional, and Financial Support Characteristics of Analytical Sample: Summary Statistics Variable Mean S.D. Minimum Maximum Demographic Variables Age at time of survey 26.97 5.66 21 63 Age at time of B.S. granting 25.18 5.63 19 62 Gender Female .58 .49 0 1 National Origin Mexican American .476 .50 0 1 Cuban .064 .24 0 1 Puerto Rican .126 .33 0 1 Other Latina/o .334 .47 0 1 Highest Parental Education Level Less than high school .159 .36 0 1 High school diploma or equivalent .200 .40 0 1 Some college, vocational, or trade school .265 .44 0 1 Bachelor’s degree .157 .36 0 1 Master’s degree .134 .34 0 1 Professional degree (e.g., JD, LLB, MD, DDS) .048 .21 0 1 Doctorate .035 .19 0 1 Educational Variables Associate degree holder .220 .41 0 1 Community college attendance .620 .49 0 1 Non-traditional student status .310 .46 0 1 Field of Study Computer and mathematical science .097 .30 0 1 Biological, agricultural, and environmental science .127 .33 0 1 Physical and related sciences .024 .15 0 1 Social and related sciences .573 .50 0 1 Engineering .100 .30 0 1 S&E-related fields .078 .27 0 1 Non S&E fields .001 .024 0 1 Undergraduate Grade Point Average 3.75–4.00 (Mostly As) .160 .37 0 1 3.25–3.74 (About half As/half Bs) .356 .48 0 1 2.75–3.24 (Mostly Bs) .351 .48 0 1 2.25–2.74 (About half Bs/half Cs) .119 .32 0 1 Less than 2.24 (Mostly Cs) .015 .12 0 1 Did not take courses for grades .0004 .02 0 1 Graduate Degree Enrollment/Attainment Earned graduate degree or currently enrolled in graduate school .330 .47 0 1 109 Table 3.2, Continued Variable Mean S.D. Minimum Maximum Baccalaureate-granting Institutional Characteristics Control Private .331 .47 0 1 Carnegie Classification Research University .399 .49 0 1 Doctoral University .138 .34 0 1 Master’s College or University .340 .47 0 1 Liberal Arts Institution .114 .32 0 1 Specialized Institution .009 .10 0 1 Selectivity Non-competitive .091 .29 0 1 Less Competitive .084 .28 0 1 Competitive .225 .42 0 1 Competitive Plus .044 .21 0 1 Very Competitive .231 .42 0 1 Very Competitive Plus .011 .11 0 1 Highly Competitive .128 .33 0 1 Highly Competitive Plus .019 .14 0 1 Most Competitive .163 .37 0 1 Special .004 .07 0 1 State California .340 .47 0 1 Florida .097 .30 0 1 Illinois .034 .18 0 1 New York .108 .31 0 1 Texas .118 .32 0 1 Other state .303 .46 0 1 Financial Support/Aid Variables Sources of Financial Aid Earnings .600 .49 0 1 Employer support .090 .28 0 1 Grants or scholarships .650 .48 0 1 Loans from Government, Banks & Institutional Sources .660 .47 0 1 Loans from Parents or other relatives .070 .26 0 1 Work Study .280 .45 0 1 Other source .020 .13 0 1 Amount Borrowed to Finance Undergraduate Degree None .279 .45 0 1 $1–5,000 .094 .29 0 1 $5,001–10,000 .113 .32 0 1 $10,001–15,000 .137 .34 0 1 $15,001–20,000 .137 .34 0 1 $20,001–25,000 .077 .27 0 1 $25,001–30,000 .074 .26 0 1 $30,001–35,000 .027 .16 0 1 $35,001 or more .063 .24 0 1 Source: NSF 2003 National Survey of Recent College Graduates (NSRCG), using final survey weight (WTSURVY) and Barron’s Profiles of American Colleges (2004). 110 Analyses I used descriptive statistical methods, latent class analysis, logistic regression analyses and propensity score matching to address the research questions. In order to ensure that the enhanced data file is nationally representative of the population of Latino STEM B.S. degree earners who met the aforementioned selection criteria, I weighted each case by the NSRCG sample weight and used procedures to account for the complex survey design in the analyses discussed below. In keeping with the quantitative criticalist approach described previously, I only included Latinos who met the selection criteria in my analysis. Rather than introducing race/ethnicity as a covariate in the analyses, I assumed that the processes explored through my analyses operate differently for Latinos and I sought to understand those processes for Latinos alone. Examining the Role of Community Colleges and HSIs in the Production of Latino STEM B.S. Holders using Descriptive Analysis Figure 3.1 illustrates the various pathways through which Latino students become STEM bachelor’s degree holders. To address the first research question of the study, I used descriptive statistics to document the proportion of Latino students who receive STEM bachelor’s degrees at HSI and non-HSI four-year colleges, distinguishing between those in the public and private sector and observing the total share contributed by HSI-intensive states and those states with the largest community college systems (i.e., California, Florida, Illinois, New York, and Texas). To characterize the role of community colleges in producing STEM bachelor’s degree holders, I also determined the share of Latino students who earned an associate degree at a community college. In this aspect of the analysis, I also compared the distribution of associate degree holders in specific STEM major fields of study in comparison with students who did not earn an associate degree before attending a four-year college/university. In order to account for the complex survey 111 design of the NSRCG, I used Stata’s survey (SVY) commands to tabulate the results, and interpreted the reported design-based F statistics to determine the statistical significance of inter- group differences. Figure 3.1. Pathways to the STEM Baccalaureate for Latino Students Determining College Financing Strategies among Latino STEM B.S. Degree Holders Using Descriptive Statistics and Latent Class Analysis The second set of research questions in the study pertains to the college financing strategies of Latino STEM bachelor’s degree holders and the differences in college financing strategy among Latinos by institutional pathway and demographic characteristics. This aspect of the study involved two types of analyses: (1) descriptive statistical analysis of the typical college financing strategies of Latinos, with a focus on the proportion of Latino students who borrow (comparing among Latinos of different national origins and between associate degree holders and non-associate degree holders); and (2) latent class analysis to characterize the underlying financing strategies employed by Latino STEM degree holders and the statistical significance of CC 4-year HSI CC 4-year non-HSI 4-year HSI 4-year non-HSI Latino Students Mexican American Puerto Rican Cuban Other 112 differences in the financing strategies used by students who utilize different institutional pathways (community college/four-year college; HSI /non-HSI to the STEM baccalaureate). In order to determine the differences in the patterns of financial aid among Latinos, I tabulated the descriptive statistics using Stata’s survey commands in order to account for the NSRCG’s complex survey design and interpreted reported design-based F statistics. I carried out these tabulations to understand how the use of eight forms of financial aid and support varied along the domains of interest, e.g., institutional pathway, state, and national origin. These eight forms of financial aid/support include: loans from a school, bank, or government; work study; scholarships/grants; earnings; employer support; parental/familial support not to be repaid; parental/familial loan; and any other source. These eight variables were mutually exclusive and dichotomous, with a response of one indicating that the respondent received college financing support through the specified mechanism, and zero indicating that no support was received through the specified mechanism. While the descriptive statistics enabled me to understand how the use of specific forms of financial aid among Latinos varied by the domains of interest (e.g., institutional pathway, national origin), this method alone could not characterize the underlying financing strategies indicated by the patterns of use of the eight forms of financial support. In order to address this question, I conducted a latent class analysis to categorize the NSRCG respondents’ patterns of financial aid usage into different types, or classes, of financing strategies. Latent class analysis (LCA) is an analytical technique that aims to uncover clusters of individuals who are similar with respect to a set of characteristics measured by categorical outcomes (Andersen, 1994; Goodman, 2002; McCutcheon, 2002) 10 . For the purposes of this 10 Latent class analysis (LCA) is similar to cluster analysis; however, LCA is a model-based approach in which it is assumed that the observed data are generated by a mixture of probability distributions. 113 application, the measured characteristics correspond to the observed, or manifest, responses indicating the eight forms of financial support, e.g., used loans to finance college/did not use loans to finance college, received a scholarship or grant to finance college/did not receive a scholarship or grant to finance college. My application of latent class analysis is based on the assumption that the observed financial aid indicator variables are associated because of an underlying, unobserved factor, i.e., the financing strategy. For example, an individual whose strategy is to finance college while avoiding debt will have a different pattern of responses on the eight financial support indicator variables than someone who seeks out financial aid from any source, including loans. In other words, the latent financing strategy is the antecedent to the manifest financial aid/support variables (Goodman, 2002, p. 17). This assumption is illustrated in Figure 3.2 below: Figure 3.2. College Financing Strategy Precedes Financial Aid Indicator Variables Loans (Y/N) Work Study (Y/N) Scholarship/Grant (Y/N) Earnings (Y/N) Parental/Familial Support (Gift) (Y/N) Parental/Familial Loan (Y/N) Employer Support (Y/N) Other Source (Y/N) COLLEGE FINANCING STRATEGY 114 It follows then that the relationship between any two manifest financial support indicator variables can be accounted for by the latent variable, college financing strategy (McCutcheon, 2002). A consequence of the above assumption is that individuals whose response patterns on the NSRCG questions regarding financial support resembled one another had a similar underlying college financing strategy. In the current study, I used the probabilistic parameterization of the latent class model which is characterized by the categorical variables: (1) latent financing strategy variable (X); and (2) the eight manifest financial aid indicator variables (A, B, C, D, E, F, G, and H), and the parameters: (1) latent class probabilities; and (2) conditional probabilities (McCutcheon, 2002). The latent class model is then expressed as a product of the latent class probabilities and conditional probabilities: X H ot X F nt X E mt X D lt X C kt X B jt X A it X t ABCDEFGHX ijklmnopt | | | | | | | π π π π π π π π π = , (3) where the latent class probability ( π t X ) is the probability that a randomly selected observation from the sample is located in latent class t, and the conditional probabilities (e.g., π it A|X ) are the probabilities that a member of the latent class t will be at a certain level of an observed indicator variable. For example, as I applied LCA in this study, the latent variable X t is a measure of the college financing strategy where X 1 might correspond to a balanced support strategy and X 2 might correspond to a loan averse strategy. The first manifest variable (A i ) indicates whether a respondent received a loan from a school, bank, or government to finance college (i=1, yes; i=0, no). The conditional probability π 11 A|X is the likelihood that a randomly selected respondent who 115 employed a balanced support strategy would report having received a loan from a school, bank, or government. As stated above, there are eight, dichotomous financial aid indicator variables, and 256 (2 8 ) possible response patterns. Presumably, each possible response pattern observed in the data could represent a latent class, however the probabilistic parameterization of the latent class model assumes that measurement error could cause some of the responses observed in the data to be the result of misclassification. Thus, the observed distribution of responses could be due to a smaller number of classes. In latent class analysis, the researcher is able to specify the number of latent classes used to fit the data and the resulting model can be tested to determine if it is an adequate representation of the observed data. For each financial aid indicator variable, the conditional probability of using that form of financial support is reported for each latent class. I graphed these conditional probabilities and formulated qualitative interpretations for the classes. Additionally, the LCA derived probabilities of class membership for each respondent in the NSRCG dataset. I assumed that each respondent was a member of the class for which they had the highest probability of membership. Subsequent to assigning the respondents the college financing strategy as indicated by their class membership, I conducted cross tabulations of the college financing strategies by domains of interest (e.g., institutional pathway, national origin) using Stata’s survey (SVY) commands and tested for statistically significant inter-group differences. Estimating the Effect of Indebtedness on Latino STEM B.S. Degree Holders’ Graduate School Enrollment using Logistic Regression and Propensity Score Matching In the final research question of the study, I wished to determine the effects of indebtedness of graduate school enrollment among Latino STEM bachelor’s degree holders. I conducted this analysis using two methods: logistic regression and propensity score matching 116 (PSM). I also compared the results obtained through use of these two different methods of analysis. I discuss each of these methods below, and describe the rationale for using propensity score matching. However, before discussing the methods, I explain the strategy I used for modeling debt as a treatment. Modeling Debt as a Treatment. In order to understand the way in which borrowing affects graduate school enrollment I used the methods of logistic regression and propensity score matching. The key variable of interest involved in modeling debt as a treatment is UGLOANR, or the cumulative undergraduate debt. UGLOANR is a categorical variable that indicated the amount borrowed (from any source) to finance the undergraduate degree. Possible responses are illustrated in the table below. Table 3.3. Possible Responses for Cumulative Undergraduate Debt Cumulative Undergraduate Debt 1. Did not earn a degree at this level 2. None 3. $1–$5,000 4. $5,001–$10,000 5. $10,001–$15,000 6. $15,001–$20,000 7. $20,001–$25,000 8. $25,000–$30,000 9. $30,001–$35,000 10. $35,001 or more Source: 2003 National Survey of Recent College Graduates (NSRCG), National Science Foundation. My strategy for modeling debt as a treatment involved developing three treatment categories for debt: high debt, low debt, and no debt. I defined the level of borrowing relative to a respondent’s peers at the baccalaureate-granting institution due to previous research that suggests debt is a socially constructed concept and perceptions of indebtedness are influenced by social location and context (Archer & Hutchings, 2000; McDonough & Calderone, 2006). In other words, the “dosage” of the treatment (i.e., borrowing), is defined in terms of how it 117 compares to the average cumulative undergraduate debt of the graduating class in the year of interest (i.e., 2000-01 or 2001-02). Because the cumulative undergraduate borrowing variable UGLOANR provided in the NSRCG dataset is categorical and the average cumulative debt from the College Board and TICAS datasets is a continuous variable, I had to decide how to define the “dosage” of the treatment. I decided to define the debt treatment level as a continuous variable with the following formula: Dose = ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ + ∗ debt UGLOANR debt UGLOANR val val max min 2 1 , (4) where UGLOANR min val and UGLOANR max val are the minimum and maximum value of the debt category indicated by the respondent and debt is the average cumulative debt of the graduating class of the baccalaureate-granting institution in the appropriate year. An alternative strategy, which I decided against, was to model the effects of the absolute magnitude of debt. I rejected this strategy because the literature of the effects of loans on enrollment choices suggests that there is not a linear relationship between these two factors. Instead, different borrowing strategies are at play at low and high levels of indebtedness. I acknowledge that decisions about college debt are not exogenous to decisions about institutional pathways and costs. My modeling approach implies a sequence of decisions regarding the institutional pathway (attend/not attend 2-year institution; more/less expensive 4-year institution) and initial financing strategy (borrow/not to borrow), followed by decisions about the amount of debt to incur each semester. 118 The definition of the treatment “dose” resulted in the following distribution of debt treatments levels among my sample as shown in Figure 3.3. The figure and the analyses make the distinction between those who have earned an associate degree and non-associate degree holders due to the expectation that associate degree earners who spend less time at the four-year institution borrow less. Thus, passage into the “high debt category” for an associate degree holder will occur at a lower dosage level compared to non-associate degree holders. Figure 3.3. Distribution of Relative Debt Dosage Level by Associate Degree As noted, the characterization of students’ decisions about borrowing is modeled within their chosen institutional pathway. Based on the distributions above, I defined the high debt category as shown below, where a dosage of 1.5 corresponds to borrowing an amount equal to one and a half times the average cumulative per borrower debt at the baccalaureate-granting Source: Analyses of the NSF 2003 National Survey of Recent College Graduates (NSRCG). Dosage Level Dosage Level Low debt High debt Associate Degree Holders (n=961) Cutpoint=1.00 0 2.0 4.0 20 0 40 80 100 60 Low debt High debt Cutpoint=1.5 Non-Associate Degree Holders (n=6,739) 0 100 200 300 0 2.0 4.0 Frequency 119 institution, and a dosage of 1.0 corresponds to borrowing at an amount equal to the average per borrower cumulative debt at the baccalaureate-granting institution. Non-Associate Degree Holders High Debt Category: Dosage ≥1.5 Low Debt Category: 0<Dosage<1.5 No Debt Category: Dosage = 0 Associate Degree Holders High Debt Category: Dosage ≥1.0 Low Debt Category: 0<Dosage<1.0 No Debt Category: Dosage = 0 For each respondent, I assigned the appropriate undergraduate debt dosage level based on the above definitions. The relative debt level was then used as a dependent variable in the logistic regression (in addition to other relevant covariates), and the treatment in the propensity score matching analyses. I discuss each of these methods below. Logistic regression. Regression techniques, including logistic regression, are typically used in higher education research to calculate the odds of attaining a particular outcome given a set of determining independent variables relative to a reference group (Peng, So, Stage, & St. John, 2002). In this study, I wished to measure the effects of undergraduate loan debt on graduate school attendance. The variables STEM graduate degree enrollment (TCDRG1), and STEM graduate degree(s) obtained (DGRDG), were used to determine if individuals in the sample were enrolled in graduate school during the survey reference week, or if they had already earned a 120 post-baccalaureate degree. The explanatory variable, Debt i , is a measure of the extent of a respondent’s participation in a student loan program to finance their undergraduate education (i.e., high debt, low debt, or no debt), and was determined as described in the previous section. The equation below specifies the model of graduate school attendance (GRADSCHL i ) as it depends on Debt i, and individual-specific covariates X i related to educational attainment and demographic characteristics, and institutional level characteristics, Z i . P(GRADSCHL i =1|X, Z) = z i x i loan i z i x i loan i Z X Debt Z X Debt e e β β β β β β β β + + + + + + + 0 0 1 (5) Figure 3.4 illustrates the individual- and institutional-level covariates included in the logistic regression model. Figure 3.4. Model of Graduate School Enrollment for Logistic Regression Analysis Graduate School Enrollment Individual-level Variables • Hispanic Origin Group (i.e., Mexican, Cuban, Puerto Rican, Other) • Gender • Undergraduate GPA • Non-Traditional Student Status • Highest Level of Parental Education • Associate Degree Holder • Field of Study • Forms of Financial Aid or College Financing Strategy • Relative level of indebtedness • Attended college in-state Institutional-level Variables • Control (public/private) • HSI status • Geographical location (i.e., state) 121 The individual-specific covariates include variables relating to national origin, gender, age, parental educational levels (which act as a proxy for socioeconomic origins), non-traditional student status, associate degree attainment, academic ability, field of major, and forms of financial aid/college financing strategy. The institutional-level characteristics in Z i include the geographic location (i.e. state), control, selectivity, Carnegie classification, and HSI status of the baccalaureate granting institution. Although many have acknowledged the shortcomings of logistic regression techniques (Agodini & Dynarski, 2004; DesJardins, Ahlburg, & McCall, 2006; Dowd, in press; Titus, 2007), these methods are commonly employed by researchers in an attempt to identify the independent effects of some antecedent condition on a given categorical educational outcome. For example, Malcom (2006a) used logistic regression to identify the impact of community college attendance on the sector of employment, and highest degree type for STEM B.S. degree earners. While other predictor variables (e.g., age, race, gender) are usually included in logistic regression analyses in an attempt to isolate the effects of independent variables of interest, logistic regression cannot avoid the problem of self-selection bias, or endogeneity. Endogeneity refers to the case when “predictors of an outcome are themselves associated with other unobserved or observed variables” (Titus, 2007, p. 489). This tends to be a problem because individuals do not randomly decide to participate in educational “treatments”, e.g., participate in a loan program. A variety of factors, some of which are unobservable, drive the student decisions to borrow to finance their education. Unfortunately, logistic regression techniques as they are typically employed in education research cannot account for self-selection bias and may produce biased results and inaccurate models. 122 Propensity Score Matching Analysis Self-selection bias tends to be more of a problem in higher education research due to the practical and ethical barriers to randomized experimental designs. Although the randomized double-blind trial is considered the “gold standard” and is increasingly promoted by the U.S. Department of Education’s Institute for Education Sciences (IES) (U.S. Department of Education, 2006a, 2006b), this type of design is not feasible for studies of the effects of programs and interventions in postsecondary institutions. Titus (2007) and others recently introduced propensity score matching to the field of higher education as a way to avoid the problems of self-selection bias. As applied in the social sciences, propensity score matching (Rosenbaum & Rubin, 1983) provides a strategy to estimate the counterfactual effect of a particular educational “treatment” (i.e., what would have happened to those who received treatment, if they did not receive treatment [and vice versa]) (Guo, Barth & Gibbons, 2004). PSM attempts to mimic an experimental design with randomized assignment of students to educational “treatment” conditions, by identifying cases in survey or administrative data that were subject to a treatment condition and comparing their outcomes to matched cases in the data. Unlike simple matching techniques, PSM matches on the propensity score, i.e., the probability of group membership (e.g., treatment vs. control group) based on observed predictor variables. A key assumption of the propensity score matching is the conditional independence assumption (CIA), which states that the selection into treatment is based solely on observable covariates and that all variables that affect assignment into treatment and the outcome variable of interest simultaneously are observable by the researcher (Becker & Caliendo, 2007; Caliendo & Kopeinig, 2005; Rosenbaum & Rubin, 1983; Rosenbaum, 2002). This is a strong assumption, and the validity of the assumption is heavily dependent on the quality of the data used in the PSM 123 analysis. While the CIA cannot be tested, there are ways to test the model’s robustness against violations of this assumption. For the purposes of this study, participation in a loan program is considered a “treatment,” and I wished to estimate the effect of the treatment on graduate school attendance among Latino STEM bachelor’s degree holders. Ideally, I would have disaggregated Latinos in the analytical sample by national origin, however the sample sizes were too small and the estimates resulted in unacceptably high standard errors. Similar to the approach used in the logistic regression analysis, I classified Latino STEM B.S. degree holders as having a high or low level of participation in borrowing compared to not borrowing at all. As described above, the levels of participation were defined in relation to the average per-borrower cumulative indebtedness for the graduating class of the student’s bachelor’s degree-granting institution during the appropriate year (2000-01 or 2001-02). Propensity scores, or the probability of participating in a loan program at a given level, were estimated using constrained multinomial probit regression 11 . Covariates known to be associated with borrowing were included in the probit model. Caliendo and Kopeinig (2005) warn that “only variables unaffected by participation [in the treatment] (or the anticipation of it) should be included” in the model used to predict propensity scores (p. 6, emphasis mine). Because previous research has found there to be strong relationships between borrowing and the use of other types of financial aid, neither the financial aid indicators variables nor the latent college financing strategies were included in the probit model. The multinomial probit model assumes the outcomes are derived from the following latent model, 11 Initially, I attempted to calculate the propensity scores with a constrained multinomial logistic model. However, due to violations of the Independence from Irrelevant Alternatives (IIA) assumption revealed by the Small-Hsiao test (Small & Hsiao, 1985), I used a multinomial probit model instead. The IIA restriction is not imposed on multinomial probit regression models due to the fact that the errors, which are assumed to be normally distributed, can be correlated across outcome categories (Long, 1997). 124 ) , 0 ( ~ , Ω + ′ + ′ = ∗ N z x y i ij j i j i ij ε ε γ β (6) with Σ ⊗ = Ω N I and ⎟ ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎜ ⎝ ⎛ = ′ = Σ MM M M i i E σ σ σ σ ε ε L O L 1 1 11 ) ( , (7) where y ij ∗ is an unobserved discrete variable, x ′ i is a matrix of individual-level covariates, z ′ i is a matrix of institutional-level covariates, β and γ are vectors of coefficients, ε i is a vector of errors following a multivariate normal distribution that is correlated across outcomes j (Davidson & McKinnon, 2004). Because this is a constrained multinomial probit model, γ 2 = γ 3 , where outcome j=2 is borrowing at a low level and outcome j=3 is borrowing at a high level. I applied these constraints because while institutional level covariates impact the likelihood of borrowing versus not borrowing, the covariates are not expected to have differential effects on the probability of borrowing at a low or high level once an individual is within the institutional environment due to the relative definition of debt employed in this study. Outcome category j is chosen if ∗ ij y is highest for j, that is: ⎪ ⎩ ⎪ ⎨ ⎧ = ∗ ∗ ∗ ∗ otherwise. 0 ) , , , max( = if 2 1 iM i i ij i y y y y j y K (8) The probability to choose outcome category j can then be expressed as: ) , , , , , ( ) | ( ) 1 ( ) 1 ( 1 ∗ ∗ ∗ + ∗ ∗ − ∗ ∗ ∗ > > > > = = iM ij j i ij j i ij i ij i i y y y y y y y y p x j y p K K (9) 125 I used Equations 6–9 to generate the propensity scores of participating in a loan program at a given level (i.e., high debt, low debt), given the individual-specific covariates x ′ i and the institutional-level covariates, z ′ i . Figure 3.5 illustrates the individual-specific and institutional-level covariates included in the multinomial probit regression model to calculate the propensity scores. Figure 3.5. Covariates included in Constrained Multinomial Probit Model of Relative Level of Indebtedness For each respondent in the sample, the probabilities of participating in a loan program at a high and low level were used to create a counterfactual framework, i.e., a ‘treatment’ group and a ‘control’ group. Members of the treatment group borrowed at either the high or low level, and members of the control group did not borrow at all. Predicted Probability of Borrowing at a High or Low Relative Level Individual-level Variables • Hispanic Origin Group (i.e., Mexican, Cuban, Puerto Rican, Other) • Gender • Highest Parental Education Level • Associate Degree Holder • Non-Traditional Student Status • Field of Study • Attended college in-state Institutional-level Variables • Geographic location (i.e., state) • Tuition and fees • Percent Students who Borrow • Control (public/private) • Type (liberal arts, comprehensive, research university) • Selectivity 126 I used specialized PSM software (the PSMATCH2 module in Stata) (Leuven & Sianesi, 2003; Sianesi, 2001) to create the ‘treatment’ and ‘control’ groups, check for balance in these two groups, match cases with similar propensities to borrow (e.g., high-high, and low-low), and compare the outcome of interest, i.e., graduate school attendance, to estimate the effects of the treatment on graduate school attendance. Figure 3.6 illustrates the creation of treatment and control groups and the matching of cases in PSMATCH2 (Leuven & Sianesi, 2003). As depicted in the figure, cases from the ‘treatment’ and counterfactual ‘control’ group are matched to estimate the treatment effects. Figure 3.6. Illustration of Application of Propensity Score Matching Techniques to Estimate the Effects of Undergraduate Loan Debt on Graduate School Attendance I obtained the average treatment effect (ATE), average treatment effect on the treated (ATT), and the average treatment effect on the untreated (ATU) through the use of PSMATCH2 Propensity Scores: p 0 < p j < p 1 Respondents who participated in a loan program at high level or low level Respondents who did not participate in a loan program Treatment Group Control Group increasing p j Matched Cases 127 in Stata (Leuven & Sianesi, 2003; Sianesi, 2001). The ATT represents the average impact of the treatment among those who have been exposed to it, and the ATU indicates the expected impact of treatment among those who have not been treated. The ATE corresponds to the average treatment effect for the entire population, whether or not a particular individual has been treated. The formulae for the ATT, ATU, and ATE are given below: [ ] [ ] [ ] T Y E T Y E T Y Y E ATT C i T i C i T i | | | − = − = (10) [ ] [ ] [ ] C Y E C Y E C Y Y E ATU C i T i C i T i | | | − = − = (11) [ ] [ ] ) ( | ) ( | ) ( ) ( C P C Y Y E T P T Y Y E C P ATU T P ATT ATE C i T i C i T i ∗ − + ∗ − = ∗ + ∗ = (12) PSM allows for a variety of matching methods (Becker & Ichino, 2002; Caliendo & Kopeinig, 2005) and the appropriate method ought to be selected by the researcher based on the characteristics of the data and her tolerance levels for variance in the estimates versus accurate matches (Caliendo & Kopeinig, 2005; Ham, Li, & Reagan, 2006). Because the propensity score derived using the constrained multinomial probit model is a continuous variable, the probability of observing the same value of the propensity score is zero. The matching methods proposed in the literature are designed to circumvent this issue by matching treatment and control cases with propensity scores within a certain interval (i.e., stratification matching), matching treatment cases with the control case with the closest propensity score (i.e., nearest neighbor matching), matching each treatment case with a control case whose propensity score falls in a predefined neighborhood of its propensity score (i.e., radius matching), and matching treatment cases with a weighted average of all control cases, where the weights are inversely proportional to the absolute value of the difference between the propensity scores of the treated and controls (i.e., kernel matching) 128 (Becker & Ichino, 2002). For the current study, I chose to use the kernel matching scheme in order to avoid matching cases with large differences in propensity scores, and to increase the quality of matches by identifying the best counterfactual (Titus, 2007). Kernel matching is thought to produce the most precise estimates of the treatment effects (Frölich, 2004). Once the ATE, ATT, and the ATU were calculated, I compared the results obtained using PSM with those from the logistic regression analysis in order to evaluate whether PSM represents an improvement over standard estimation procedures. Sensitivity Analyses While propensity score matching is widely preferred over standard regression techniques (Titus, 2007), PSM is not an infallible method of analysis. There are several pitfalls that can potentially undermine the validity of PSM results. Two such pitfalls are the quality of the model used to predict propensity scores, and the issue of unobserved variables that might affect the propensity score estimates and the outcome of interest. Below, I discuss two types of sensitivity analysis I conducted to bolster the validity of my findings. Testing the effects of additional and fewer covariates on the propensity score estimates. In order to establish the soundness PSM results, I tested the sensitivity of the calculated treatment effects to the modeled propensity scores. By changing the observed covariates included in the constrained multinomial probit model used to estimate the propensity scores (i.e., including additional covariates, and excluding some covariates from the original model), I was able to determine the change in treatment effects caused by these alterations. Other researchers using propensity score matching (Titus, personal communication, January 31, 2008) have used this technique to strengthen the validity of their results. Testing the robustness of treatment effect estimates against violations of the Conditional Independence Assumption. While PSM is capable of correcting for self-selection biases, it is 129 possible that unobserved characteristics or “unobservables” could simultaneously impact the outcome variable of interest (i.e., graduate school enrollment) and the propensity score (i.e., predicted probability of participating in a loan program at a given level), constituting a violation of the Conditional Independence Assumption (Caliendo & Kopeinig, 2005). The presence of such an unobserved determinant might create a hidden bias against which estimates of the treatment effects derived from PSM are not robust (Becker & Caliendo, 2007; Rosenbaum, 2002) undermining the validity of the PSM model. Though the conditional independence assumption cannot be directly tested (Caliendo & Kopeinig, 2005; Rosenbaum & Rubin, 1983), it is possible to test the robustness of the treatment effect estimates against hidden biases introduced by unobserved variables using the bounding approach first proposed by Rosenbaum (2002). Sensitivity analysis using the bounding approach tests how strongly ‘unobservables’ could simultaneously influence the propensity to participate in treatment and the outcome variable without undermining the results of the propensity score matching analysis. For example, if two individuals had identical observed covariates (e.g., national origin, gender, SES), they would have the same probability of receiving treatment (e.g., borrowing at a high level), and thus, the same propensity score. However, there could be some unobserved factor that altered the probability of borrowing at a high level so that the true propensity scores of the individuals are in fact different. The Stata module, MHBOUNDS (Becker & Caliendo, 2007), introduces different levels of bias that could be caused by unobserved variables and tests how large the hidden bias would have to be to render the PSM results insignificant. I used the MHBOUNDS module to test the robustness of the propensity score matching treatment effect estimates against unobserved variables. 130 CHAPTER FOUR FINDINGS The findings presented in this chapter relate to Latinos who earned a bachelor’s degree in STEM fields from a U.S. postsecondary institution in the 2000-01/2001-02 academic years. Institutional Pathways to STEM Bachelor’s Degrees for Latinos The first research question addressed in this study pertains to the institutional pathways traversed by Latino STEM degree holders. In this section, I describe the institutional pathways traversed by Latino STEM bachelor’s degree holders, focusing in particular on the role of the community college and Hispanic-Serving institutions (HSIs) for these students. Table 4.1 illustrates that nearly 20% of Latino STEM bachelor’s degree holders earned an associate degree from a community college. Slightly less than 20% of Latino STEM bachelor’s degree holders earned that degree from an four-year institution designated as ‘Hispanic-serving.’ In order to gain a better understanding of the institutional pathways Latino STEM degree holders used to gain access to that degree, four paths of interest have been identified: (1) associate degree completion at a community college followed by bachelor’s degree completion at a four- year Hispanic Serving Institution; (2) associate degree completion at a community college followed by B.S. completion at four-year non-HSI; (3) entrance to a four-year HSI without earning an associate degree; and (4) entrance to a four-year non-HSI without earning an associate degree 12 . Observed proportions presented 13 for Latino STEM bachelor’s degree holders using each pathway are illustrated in Table 4.1. 12 Due to the limitations of the NSRCG dataset, I was unable to determine which respondents may have transferred from a community college to a four-year institution without earning an associate degree. For this reason, the term “direct entrant” is not used to describe those individuals who did not follow pathways (1) or (2) as defined above. 13 Without controlling for other factors. 131 Table 4.1. Institutional Pathways of Latino STEM Bachelor’s Degree Holders Latinos Weighted Sample n=61,233 Associate Degree Earned an AA/AS at CC 19.8% Did not earn an AA/AS at CC 80.2% Hispanic Serving Institution B.S.-granting institution designated as HSI 19.8% B.S.-granting institution not designated as HSI 80.2% Institutional Pathway AA/AS at CC to 4-year HSI 6.4% AA/AS at CC to 4-year non-HSI 13.5% 4-year HSI 13.5% 4-year non-HSI 66.7% Source: Analyses of the NSF 2003 National Survey of Recent College Graduates (NSRCG), using final survey weight (WTSURVY). Differences in Degree-granting Institutional Characteristics by Pathway for Latino STEM Degree Holders The first research question posed in this study also asked what differences, if any, existed in the characteristics of baccalaureate-granting colleges/universities of Latino STEM B.S.-degree holders who earned an associate degree and those who did not earn an AA/AS. Table 4.2 shows that Latino STEM bachelor’s degree holders who have earned an associate degree graduate from publicly controlled four-year institutions in a higher proportion (p<.001) than those Latino STEM B.S. holders who did not earn an associate degree (83.0% and 62.9%, respectively). Differences in the Carnegie classification of baccalaureate-granting institution also exist between those Latino STEM B.S. holders who earned an associate degree and those who did not (p<.001) 14 . About 25% of Latino associate degree holders earned their STEM baccalaureate from a Research University compared to almost 44% of Latino STEM B.S. degree holders who did not earn an associate degree. Slightly over 16% of Latinos who were AA/AS degree holders earned their bachelor’s degree from a Doctoral University, compared to just over 13% of Latinos who 14 Design-based F statistics are used to determine the statistical significance of differences in proportions in order to account for the clustered (by undergraduate institution) and stratified (by race/ethnicity, gender, and field of study) nature of the National Survey of Recent College Graduates (NSRCG) sample design. 132 did not earn an associate degree. A larger proportion of associate degree holders graduated from institutions classified as Master’s Colleges/Universities (46.5%) compared to the proportion of non-associate degree holders attending this institutional type (30.9%). Table 4.2 also illustrates that a lower proportion of associate degree holders attended liberal arts colleges (7.8%) compared to non-associate degree holders (12.3%). Table 4.2 also shows that a larger proportion (p<.001) of Latino STEM bachelor’s degree holders who earned an associate degree graduated from a Hispanic-Serving institution compared to non-associate degree holders. Thirty-two percent of Latino AA/AS degree holders earned their bachelor’s degree from an HSI, compared to just below 17% of non-AA/AS degree holders. The distribution of Latino STEM bachelor’s degree holders across the ten categories of institutional selectivity defined by Barron’s shown in Table 4.2 illustrates that there are significant differences (p<.001) between associate degree holders and non-associate degree holders. Though a larger proportion of non-AA/AS degree holders attended institutions categorized as “non-competitive” (9.7%) than AA/AS degree holders (6.2%), more than half (58%) of associate degree holders graduated from less-selective institutions (those classified between ‘non-competitive’ and ‘competitive plus’). This is compared to 41% of non-associate degree holders who fall into these four, less selective categories. Additionally, Table 4.2 shows that just under 10% of Latino associate degree holders graduated from institutions in the highest category of selectivity (“most competitive”), compared to nearly 18% of non-associate degree holders. 133 Table 4.2. Baccalaureate-granting Institutional Characteristics of Latino STEM Bachelor’s Degree Holders by Associate Degree Institutional Pathway AA/AS @ CC No AA/AS @ CC Total Latino Weighted Sample n = 12,142 n = 49,092 n = 61,233 ** Control Uncorrected χ 2 =211.224 df=1 Public 83.0% 62.9% 66.9% Design-based F(1, 213)=17.093, p<0.001 Private 17.0% 37.1% 33.1% Carnegie Classification Uncorrected χ 2 =413.713, df=4 Research University 25.3% 43.5% 39.9% Design-based F(3.28, 689.83)= 6.932, p<0.001 Doctoral University 16.3% 13.2% 13.8% Master’s College/University 46.5% 30.9% 34.0% Liberal Arts College 7.8% 12.3% 11.4% Specialized Institution 4.1% 0.1% 0.9% Hispanic-Serving Institution Uncorrected χ 2 =170.842, df=1 Hispanic Serving Institution 32.1% 16.8% 19.8% Design-based F(1, 214)= 9.435, p<0.003 Not a Hispanic Serving Institution 68.0% 83.2% 80.2% Selectivity #/% of institutions † Uncorrected χ 2 =516.011 df=9 Non-Competitive (n inst =14, 5.2%) 6.2% 9.7% 9.0% Design-based F(6.42, 1374.51)= 4.662, p<0.001 Less Competitive (n inst =26, 9.6% ) 17.1% 6.3% 8.4% Competitive (n inst =82, 30.4%) 29.6% 20.7% 22.5% Competitive Plus (n inst =9, 3.3%) 5.1% 4.3% 4.4% Very Competitive (n inst =67, 24.8%) 25.2% 22.6% 23.1% Very Competitive Plus (n inst =11, 4.1%) 3.4% 0.6% 1.1% Highly Competitive (n inst =26, 9.6%) 2.6% 15.4% 12.8% Highly Competitive Plus (n inst =6, 2.2%) 1.0% 2.1% 1.9% Most Competitive (n inst =27, 10%) 9.8% 17.9% 16.3% Special (n inst =2, 0.8%) <1% 0.5% 0.4% Notes: Column proportions. ** May not sum to total due to rounding. † The number (n inst ) of institutions in each selectivity category and the proportion (%) of the total number of institutions in the NSRCG data universe are reported. Design-based F statistics are presented to account for the complex sample design of the 2003 NSRCG. Source: Analyses of the NSF 2003 National Survey of Recent College Graduates (NSRCG), using final survey weight (WTSURVY). 134 Differences in Institutional Pathways and Characteristics by State and Region Table 4.3 illustrates that significant differences exist (p<0.001) in the institutional pathways followed by Latino STEM B.S. degree holders by geographical region. Of those Latinos who earned the STEM bachelor’s degree from an institution located in the Northeast, less than 1% earned an associate degree. Just over 37% of Latinos who were awarded a STEM B.S. from an institution in the South Atlantic earned an associate degree. Though not as high as the South Atlantic region, over twenty percent of Latino STEM bachelor’s degree holders who attended an institution in the Mid-Atlantic, West North Central, and Pacific regions earned an associate degree. Table 4.3 also shows that significant differences in the proportion of associate degree holders and non-associate degree holders exist by state. The five states with the largest community college enrollments in the U.S. (California, Florida, Illinois, New York, and Texas) are included on the table; data for the other 45 U.S. states are reported in the aggregate. Nearly half of all Latinos who were awarded a STEM bachelor’s degree from an institution located in Florida earned an associate degree. In New York, just below 28% of Latinos who earned a STEM B.S. from an institution in the state earned an associate degree; in California, the figure was slightly lower at 22.2%. In Illinois, 16.3% of Latino STEM B.S. holders earned an associate degree. The proportion of Latino STEM B.S. holders who earned an AA/AS in Texas (9.5%) was closer to the figure for all other states (9.2%). 135 Table 4.3. Location of Latino STEM Bachelor’s Degree Holders by Associate Degree Institutional Pathway AA/AS @ CC No AA/AS @ CC Weighted Sample n = 12,142 n = 49,092 Region Uncorrected χ 2 =419.219 df=8 New England <1% 100% Design-based F(6.30, 1348.23)=4.227, p<0.001 Middle Atlantic 22.6% 77.4% East North Central 10.4% 89.6% West North Central 21.8% 78.2% South Atlantic 37.4% 62.6% East South Central 9.9% 90.1% West South Central 9.0% 91.0% Mountain 9.5% 90.5% Pacific 21.0% 79.0% Total Latino 19.8% 80.2% State Uncorrected χ 2 =651.432, df=5 California 22.2% 77.8% Design-based F(4.00, 856.36)= 11.831, p<0.001 Florida 49.5% 50.5% Illinois 16.3% 83.7% New York 27.9% 72.1% Texas 9.5% 90.5% Other State 9.2% 90.8% Notes: Row proportions. New England: Connecticut, Main, Massachusetts, New Hampshire, Vermont, and Rhode Island; Middle Atlantic: New Jersey, New York, and Pennsylvania; East North Central: Illinois, Indiana, Michigan, Ohio, and Wisconsin; West North Central: Iowa, Kansas, Minnesota, Missouri, Nebraska, North Dakota, and South Dakota; South Atlantic: Delaware, District of Columbia, Florida, Georgia, Maryland, North Carolina, South Carolina, Virginia, and West Virginia; East South Central: Alabama, Kentucky, Mississippi, and Tennessee; West South Central: Arkansas, Louisiana, Oklahoma, and Texas; Mountain: Arizona, Colorado, Idaho, Montana, Nevada, New Mexico, Utah, and Wyoming; Pacific: Alaska, California, Hawaii, Oregon, and Washington. Design-based F statistics are presented to account for the complex sample design of the 2003 NSRCG. Source: Analyses of the NSF 2003 National Survey of Recent College Graduates (NSRCG), using final survey weight (WTSURVY). Table 4.4 displays the geographical distribution of baccalaureate-granting institutions of Latino STEM bachelor’s degree holders who followed each of the four pathways (CC/HSI; CC/non-HSI; HSI; non-HSI). The table clearly reflects the concentration of HSIs in California, Florida, and Texas 15 , as significantly higher (p<0.003) proportions of students who earned a STEM bachelor’s degree in those three states attended an HSI (through entry after earning an associate degree community college or entry without earning an AA/AS). 15 In 2003-04, there were 184 Hispanic Serving Institutions (HSIs) within the 50 U.S. states. Sixty eight (37%) were located in California; ten (5.4%) were located in Florida; and 38 (20%) were located in Texas. 136 Table 4.4. Location of Latino STEM Bachelor’s Degree Holders by Institutional Pathway Institutional Pathway CC/HSI CC/non- HSI HSI Non-HSI Weighted Sample n = 3,892 n = 8,250 n = 8,240 n = 40,851 Region Uncorrected χ 2 =1190.495 df=24 New England <1% <1% <1% 100% Design-based Middle Atlantic 5.0% 17.6% 3.7% 73.7% F(8.79, 1880.52) = 2.442, p<0.01 East North Central 1.4% 8.9% 4.1% 85.5% West North Central <1% 21.8% <1% 78.2% South Atlantic 12.8% 24.6% 8.4% 54.2% East South Central <1% 9.9% <1% 90.1% West South Central 4.1% 4.9% 32.2% 58.8% Mountain 1.5% 8.0% 38.1% 52.4% Pacific 8.2% 12.8% 13.7% 65.3% Total Latino 6.4% 13.5% 13.5% 66.7% State Uncorrected χ 2 =1240.251, df=15 California 8.8% 13.4% 14.8% 63.0% Design-based Florida 19.2% 30.3% 12.7% 37.8% F(6.07, 1299.77)= 3.401, p<0.003 Illinois 2.7% 13.6% 7.8% 76.0% New York 7.2% 20.9% 5.3% 66.8% Texas 4.5% 5.0% 35.4% 55.1% Other State 0.3% 8.9% 7.2% 83.6% Notes: Row proportions. New England: Connecticut, Main, Massachusetts, New Hampshire, Vermont, and Rhode Island; Middle Atlantic: New Jersey, New York, and Pennsylvania; East North Central: Illinois, Indiana, Michigan, Ohio, and Wisconsin; West North Central: Iowa, Kansas, Minnesota, Missouri, Nebraska, North Dakota, and South Dakota; South Atlantic: Delaware, District of Columbia, Florida, Georgia, Maryland, North Carolina, South Carolina, Virginia, and West Virginia; East South Central: Alabama, Kentucky, Mississippi, and Tennessee; West South Central: Arkansas, Louisiana, Oklahoma, and Texas; Mountain: Arizona, Colorado, Idaho, Montana, Nevada, New Mexico, Utah, and Wyoming; Pacific: Alaska, California, Hawaii, Oregon, and Washington. Design-based F statistics are presented to account for the complex sample design of the 2003 NSRCG. Source: Analyses of the NSF 2003 National Survey of Recent College Graduates (NSRCG), using final survey weight (WTSURVY). Differences in Institutional Pathways and Characteristics by National Origin Table 4.5 illustrates that no significant differences in the institutional pathway to the STEM baccalaureate exist by Latino national origin (i.e., Mexican American, Puerto Rican, Cuban, and other). 137 Table 4.5. Institutional Pathways of Latino STEM Bachelor’s Degree Holders by National Origin National Origin Mexican American Puerto Rican Cuban Other Latino Weighted Sample n=29,160 n=7,714 n=3,904 n=20,454 Associate Degree Uncorrected χ 2 =36.120 df=3 Earned an AA/AS at CC 17.8% 26.6% 21.7% 19.8% Design-based F(2.83, 606.24)=0.774, p=0.502 Did not earn an AA/AS at CC 82.2% 73.4% 78.3% 80.2% Hispanic Serving Institution Uncorrected χ 2 =111.338 df=3 B.S.-granting institution designated as HSI 23.6% 8.8% 23.5% 17.9% Design-based F(2.28, 488.78)=1.310, p=0.272 B.S.-granting institution not designated as HSI 76.4% 91.2% 76.5% 82.1% Institutional Pathway Uncorrected χ 2 =309.131 df=9 AA/AS at CC to 4-year HSI 4.5% 6.1% 13.3% 7.8% Design-based F(5.43, 1163.02)=2.141, p=0.053 AA/AS at CC to 4-year non-HSI 13.4% 20.5% 8.4% 12.0% 4-year HSI 19.1% 2.7% 10.2% 10.1% 4-year non-HSI 63.1% 70.8% 68.1% 70.1% Notes: Column proportions. Design-based F statistics are presented to account for the complex sample design of the 2003 NSRCG. Source: Analyses of the NSF 2003 National Survey of Recent College Graduates (NSRCG), using final survey weight (WTSURVY). Differences in Demographic Characteristics by Institutional Pathway Table 4.6 illustrates that significant differences (p<.001) exist in some of the demographic characteristics of Latino STEM bachelor’s degree holders who did and did not earn an associate degree. A greater proportion of associate degree holders were non-traditionally aged students, i.e., earned their first bachelor’s degree at over the age of 24. Nearly 63% of AA/AS degree holders were non-traditional students, compared to 22.7% of non-AA/AS degree holders. Comparable proportions (p=0.053) of associate degree holders and non-associate degree holders earned their STEM B.S. from an institution in the same state in which they graduated high school (78.8% versus 76.1%). Significant differences in the highest level of education attained by any parent (p<.001) are also shown in Table 4.6. A larger proportion of associate degree holders had a parent that did 138 not graduate from high school compared to non-associate degree holders (29.5% and 12.5%, respectively). The highest reported parental education level was more commonly ‘some college, vocational, or trade school’ among Latino STEM B.S. holders who earned an associate degree compared to those who did not earn an associate degree (30.3% versus 25.6%). Non-associate degree holders were more represented than associate degree holders at all other parental education levels in terms of proportion. Table 4.7 illustrates that statistically significant differences in demographic characteristics also exist among Latino STEM bachelor’s degree holders by institutional pathway. A significantly higher proportion (p<0.001) of Latinos who followed the ‘CC/HSI’ and ‘CC/non- HSI’ pathway to the STEM bachelor’s degree were non-traditionally aged (73.2% and 58.0%, respectively). The high proportions of non-traditionally aged students who traversed these two pathways stood in stark contrast to the ‘non-HSI’ pathway, in which 19.1% of Latinos were non- traditionally aged. Comparable proportions (p=0.055) of Latino STEM bachelor’s degree holders who followed the four institutional pathways graduated from a four-year college/university in the same state in which they graduated high school. 139 Table 4.6. Demographic Characteristics of Latino STEM Bachelor’s Degree Holders by Associate Degree Institutional Pathway AA/AS @ CC No AA/AS @ CC Total Latinos Weighted Sample n = 12,142 n = 49,092 n=61,233 Non-Traditional Status Uncorrected χ 2 =881.470 df=1 Non-Traditionally Aged Student 62.8% 22.7% 30.7% Design-based F(1, 214)=55.962, p<0.001 Traditionally Aged Student 37.2% 77.3% 69.3% In-State College Attendance Uncorrected χ 2 =4.923, df=1 Attended In-State 4-yr College 78.8% 76.1% 76.6% Design-based F(1, 214)= 0.296, p=0.5873 Attended Out-of-State 4-yr college 21.2% 23.9% 23.4% Highest Parental Education Level Uncorrected χ 2 =333.280, df=6 Less than high school 29.5% 12.5% 15.9% Design-based F(4.87, 1041.79)= 4.474, p<0.001 High school diploma or equivalent 14.4% 21.4% 20.0% Some college, vocational, or trade school 30.3% 25.6% 26.5% Bachelor’s degree 11.6% 16.7% 15.7% Master’s degree 10.8% 14.1% 13.4% Professional degree (e.g., JD, MD) 1.8% 5.6% 4.8% Doctorate 1.6% 4.1% 3.6% Notes: Column proportions. ** May not sum to total due to rounding. Students were determined to have attended college ‘in-state’ if the baccalaureate-granting institution was in the same state from which they graduated high school. Design-based F statistics are presented to account for the complex sample design of the 2003 NSRCG. Source: Analyses of the NSF 2003 National Survey of Recent College Graduates (NSRCG), using final survey weight (WTSURVY). The highest parental education level also significantly varied (p<0.004) by institutional pathway among Latino STEM B.S.-degree holders. Parents of Latinos who followed the ‘CC/HSI,’ ‘CC/non-HSI,’ and ‘HSI’ pathways had lower levels educational attainment compared to those Latinos who followed the ‘non-HSI’ pathway. Nearly half of Latino STEM B.S. degree holders who followed the ‘CC/HSI’ pathway were first generation college students, i.e., they had parent(s) who did not attend any college, compared to 41.2% of students from the ‘CC/non-HSI’ pathway and 45.3% of Latino students from the ‘HSI’ pathway. Parents of Latino STEM B.S.- degree holders who followed the ‘non-HSI’ pathway achieved the highest levels of educational attainment. Slightly over 43% of Latinos who traversed the ‘non-HSI’ pathway attained a 140 bachelor’s degree or higher, compared to 27.0% of those in the ‘CC/HSI’ pathway, 25.1% of those in the ‘CC/non-HSI’ pathway, and 26.8% of those in the ‘HSI’ pathway. 141 Table 4.7. Demographic Characteristics of Latino STEM Bachelor’s Degree Holders by Institutional Pathway Institutional Pathway CC/HSI CC/non- HSI HSI Non-HSI Total Latino Weighted Sample n = 3,892 n = 8,250 n = 8,240 n = 40,851 n= 61,233 Non-Traditional Status Uncorrected χ 2 =1098.118 df=3 Non-Traditionally Aged Student 73.2% 58.0% 40.8% 19.1% 30.7% Design-based F(2.78, 594.37)=26.997, p<0.001 Traditionally Aged Student 26.8% 42.0% 59.2% 80.9% 69.3% In-State College Attendance Uncorrected χ 2 =118.396, df=3 Attended In-State 4-yr College 84.7% 76.1% 88.4% 73.6% 76.6% Design-based F(2.57, 550.52)= 2.676, p=0.055 Attended Out-of-State 4-yr college 15.3% 23.9% 11.6% 26.4% 23.4% Highest Parental Education Level Uncorrected χ 2 =575.888, df=18 Less than high school 28.0% 30.2% 18.8% 11.3% 15.9% Design-based F(10.10, 2160.38)= 2.598, p<0.004 High school diploma or equivalent 21.7% 10.9% 26.5% 20.4% 20.0% Some college, vocational, or trade school 23.3% 33.7% 28.0% 25.1% 26.5% Bachelor’s degree 16.0% 9.6% 18.6% 16.4% 15.7% Master’s degree 11.1% 10.7% 6.5% 15.6% 13.4% Professional degree (e.g., JD, MD) <1% 2.6% 1.6% 6.4% 4.8% Doctorate <1% 2.3% <1% 4.9% 3.6% Notes: Column proportions. ** May not sum to total due to rounding. Students were determined to have attended college ‘in-state’ if the baccalaureate-granting institution was in the same state from which they graduated high school. Design-based F statistics are presented to account for the complex sample design of the 2003 NSRCG. Source: Analyses of the NSF 2003 National Survey of Recent College Graduates (NSRCG), using final survey weight (WTSURVY). 142 Differences in Field of Study by Pathway Comparable proportions (p=.072) of Latino STEM bachelor’s degree holders who did and did not earn an AA/AS majored in each of the fields of study listed in Table 4.8. Table 4.8. Field of Study of Latino STEM Bachelor’s Degree Holders by Associate Degree Institutional Pathway AA/AS @ CC No AA/AS @ CC Total Latinos Weighted Sample n = 12,142 n = 49,092 n=61,233 Field of Study Uncorrected χ 2 =131.690, df=6 Computer Science/Mathematics 9.3% 9.9% 9.7% Design-based F(2.71, 579.47)= 2.417, p=0.072 Biological, Agricultural, and Environmental Science 7.9% 13.9% 12.7% Physical Science 0.9% 2.8% 2.4% Social and Behavioral Science 66.9% 54.9% 57.3% Engineering 5.2% 11.2% 10.0% Science & Engineering-related 9.8% 7.3% 7.8% Notes: ** May not sum to total due to rounding. Design-based F statistics are presented to account for the complex sample design of the 2003 NSRCG. Source: Analyses of the NSF 2003 National Survey of Recent College Graduates (NSRCG), using final survey weight (WTSURVY). Although there were no statistically significant differences in field of study between Latino STEM bachelor’s degree holders who did and did not earn an associate degree, Table 4.9 illustrates that significant differences (p<0.001) in field of study did exist between Latinos following the four institutional pathways to the STEM baccalaureate. Large proportions of Latinos who graduated from an HSI after earning an associate degree or without earning an associate degree majored in Computer or Mathematical Sciences (17.9% and 26.6%, respectively). These figures are strikingly higher than the proportions of Latinos in the ‘CC/non- HSI’ and ‘non-HSI’ pathways (5.3% and 6.5%, respectively). Latinos STEM B.S. degree holders who followed the ‘HSI’ pathway were more evenly distributed amongst the various STEM fields. Latinos following the ‘HSI’ pathway earned their degrees in computer science/mathematics, biological and related sciences, physical sciences, and engineering in the highest proportions, while those students in the ‘CC/HSI,’ ‘CC/non-HSI,’ and ‘non-HSI’ pathways were heavily 143 concentrated in the social and behavioral sciences. Only 35.2% of Latinos who traversed the ‘HSI’ pathway earned their bachelor’s degree in the social and behavioral sciences, compared to 60.3% in the ‘CC/HSI’ pathway, 70% in the ‘CC/non-HSI’ pathway, and 58.9% in the ‘non-HSI’ pathway. Interestingly, Latinos following the ‘CC/HSI’ pathway earned the bachelor’s degree in S&E-related fields (which include health-related fields) in the highest proportions (17.6%). 144 Table 4.9. Field of Study of Latino STEM Bachelor’s Degree Holders by Institutional Pathway Institutional Pathway CC/HSI CC/non- HSI HSI Non-HSI Total Latinos Weighted Sample n = 3,892 n = 8,250 n = 8,240 n = 40,851 n=61,233 Field of Study Uncorrected χ 2 =713.849 df=18 Computer Science/Mathematics 17.9% 5.3% 26.6% 6.5% 9.7% Design-based F(6.58, 1408.38) = 3.6532, p<0.001 Biological, Agricultural, and Environmental Science 2.5% 10.5% 16.0% 12.7% 12.7% Physical Science 0.6% 1.0% 3.1% 2.7% 2.4% Social and Behavioral Science 60.3% 70.0% 35.2% 58.9% 57.3% Engineering 1.1% 7.1% 13.9% 10.6% 10.0% Science and Engineering (S&E)- related 17.6% 6.1% 5.1% 7.8% 7.8% Notes: Column proportions. ** May not sum to total due to rounding. Design-based F statistics are presented to account for the complex sample design of the 2003 NSRCG. Source: Analyses of the NSF 2003 National Survey of Recent College Graduates (NSRCG), using final survey weight (WTSURVY). 145 Financial Aid Pathways of Latino STEM Bachelor’s Degree Holders The second research question of the present study turned to the forms of financial aid used by Latino STEM bachelor’s degree holders, the underlying financing strategies employed by this group, and the effect of various levels of indebtedness on their enrollment in graduate school. In this section, I describe the differences in the forms of financial aid used by Latino STEM B.S. degree holders by institutional pathway. I also present the results of the latent class analysis, which characterize the latent financing strategies derived from the observed, or manifest, financing mechanisms (e.g., loans, earnings, work study, etc.). The section concludes by examining differences in these unobserved financing strategies by institutional pathway, national origin, and other demographic and baccalaureate-granting institutional characteristics. Differences in Forms of Financial Aid Used by Latino STEM Bachelor’s Degree Holders by Institutional Pathway Table 4.10 illustrates that Latino STEM bachelor’s degree holders who earned an associate degree used various forms of financial aid at rates different from those non-associate degree holders. In particular, a lower proportion of associate degree holders used scholarships/grants to finance college than non-associate degree holders (59.2% versus 67.8%). Further, slightly over 37% of Latino STEM B.S. holders who earned an associate degree received parental support in the form of a gift to finance college; this is significantly lower than the over 60% of non-associate degree holders who used this form of financial support. A higher proportion of Latino STEM bachelor’s degree holders who earned an associate degree received support from their employer to finance college than non-associate degree holders (14.6% versus 7.4%). The proportion of associate degree earners who received work study to finance college was lower than that of non-associate degree earners (17.1% and 30.5%, respectively). 146 Associate degree earners and non-associate degree earners among Latino STEM bachelor’s degree holders borrowed from a school, bank, or the government, and/or from their parents or family at comparable rates. Additionally, there were no differences in the proportions of associate degree holders and non-associate degree holders among Latinos STEM baccalaureates who used earnings to finance college. 147 Table 4.10. Forms of Financial Aid Used by Latino STEM Bachelor’s Degree Holders by Associate Degree Institutional Pathway AA/AS @ CC No AA/AS @ CC Total Latinos Weighted Sample n = 12,142 n = 49,092 n=61,233 Loans Uncorrected χ 2 =38.766, df=1 Borrowed from school, bank, or government to finance college 59.2% 67.8% 66.1% Design-based F(1, 214)= 3.107, p=0.079 Did not borrow from school, bank, or government to finance college 40.8% 32.2% 33.9% Earnings Uncorrected χ 2 =0.060, df=1 Used earnings to finance college 59.7% 60.0% 59.9% Design-based F(1, 214)= 0.004, p=0.947 Did not use earnings to finance college 40.3% 40.0% 40.1% Scholarships/Grants Uncorrected χ 2 =49.679, df=1 Received scholarship or grant to finance college 57.3% 67.1% 65.2% Design-based F(1, 214)= 3.889, p<0.05 Did not receive scholarship or grant to finance college 42.7% 32.9% 34.8% Work Study Uncorrected χ 2 =103.931, df=1 Used work study to finance college 17.1% 30.5% 27.8% Design-based F(1, 214)= 9.277, p<0.003 Did not use work study to finance college 82.9% 69.5% 72.2% Parental/Familial Support (not to be repaid) Uncorrected χ 2 =251.637, df=1 Received parental/familial support not to be repaid to finance college 37.2% 60.3% 55.7% Design-based F(1, 214)= 23.125, p<0.001 Did not receive parental/familial support not to be repaid to finance college 62.8% 39.7% 44.3% Parental/Familial Loans Uncorrected χ 2 =0.313, df=1 Received loan from parents/family to finance college 7.2% 7.6% 7.5% Design-based F(1, 214)= 0.027, p=0.8688 Did not receive loan from parents/family to finance college 92.8% 92.4% 92.5% 148 Table 4.10, Continued Institutional Pathway AA/AS @ CC No AA/AS @ CC Total Latinos Weighted Sample n = 12,142 n = 49,092 n=61,233 Employer Uncorrected χ 2 =75.209, df=1 Used support from employer to finance college 14.6% 7.4% 8.8% Design-based F(1, 214)= 5.666, p<0.02 Did not use support from employer to finance college 85.4% 92.6% 91.2% Other Source Uncorrected χ 2 =91.549, df=1 Received support from some other source to finance college 4.6% 1.0% 1.7% Design-based F(1, 214)= 7.935, p<0.01 Did not receive support from some other source to finance college 95.4% 99.0% 98.3% 149 Table 4.11 illustrates the differences in the forms of financial aid used by Latino STEM bachelor’s degree holders in each of the institutional pathways of interest (i.e., ‘CC/HSI’, ‘CC/non-HSI’, ‘HSI’, ‘non-HSI’). Latino STEM bachelor’s degree holders who followed the ‘CC/HSI’ pathway used loans from a school, bank, or the government to finance college at a significantly lower (p<0.02) rate than students who followed the ‘CC/non-HSI,’ ‘HSI,’ and ‘non- HSI’ pathways. Similarly, about 30% of Latinos who followed the ‘CC/HSI’ pathway to the STEM baccalaureate received parental support in the form of a gift to finance college; a significantly lower (p<0.001) rate than students who traversed the ‘CC/non-HSI,’ ‘HSI,’ and ‘non-HSI’ pathways (41.1%, 53.3%, and 61.7%, respectively). The table below also indicates that Latinos who followed the ‘CC/non-HSI’ pathway more commonly (p<0.05) received support from their employer to finance their undergraduate education compared to students from the other three pathways. Latino STEM bachelor’s degree holders who followed the ‘CC/HSI’ pathway were the least likely (p<0.05) to have received work study to finance college (11.1%), while those in the ‘non-HSI’ pathway most commonly received work study (31.3%). 150 Table 4.11. Forms of Financial Aid Used by Latino STEM Bachelor’s Degree Holders by Institutional Pathway Institutional Pathway CC/HSI CC/non-HSI HSI Non-HSI Total Latinos Weighted Sample n = 3,892 n = 8,250 n = 8,240 n = 40,851 n=61,233 Loans Uncorrected χ 2 =214.476, df=3 Borrowed from school, bank, or government to finance college 48.5% 64.2% 69.0% 67.6% 66.1% Design-based F(4.75, 1016.73)= 3.042, p<0.02 Did not borrow from school, bank, or government to finance college 51.5% 35.8% 31.0% 32.4% 33.9% Earnings Uncorrected χ 2 =4.779, df=3 Used earnings to finance college 63.6% 57.8% 60.8% 59.9% 60.0% Design-based F(2.49, 533.87)= 0.100, p=0.939 Did not use earnings to finance college 36.4% 42.2% 39.2% 40.1% 40.0% Scholarships/Grants Uncorrected χ 2 =54.449, df=3 Received scholarship or grant to finance college 53.4% 59.1% 66.4% 67.3% 65.2% Design-based F(2.93, 627.72)= 1.488, p=0.217 Did not receive scholarship or grant to finance college 46.6% 40.9% 33.6% 32.7% 34.8% Work Study Uncorrected χ 2 =124.746, df=3 Used work study to finance college 11.1% 19.9% 26.7% 31.3% 27.8% Design-based F(2.69, 575.15)= 2.837, p<0.05 Did not use work study to finance college 88.9% 80.1% 73.3% 68.7% 72.2% Parental/Familial Support (not to be repaid) Uncorrected χ 2 =294.010, df=3 Received parental/familial support not to be repaid to finance college 28.9% 41.1% 53.3% 61.7% 55.7% Design-based F(2.95, 630.62)= 9.526, p<0.001 Did not receive parental/familial support not to be repaid to finance college 71.1% 58.9% 46.7% 38.3% 44.3% Parental/Familial Loans Uncorrected χ 2 =40.921, df=3 Received loan from parents/family to finance college 0.9% 10.1% 6.5% 7.8% 7.5% Design-based F(2.05, 438.57)= 2.193, p=0.112 Did not receive loan from parents/family to finance college 99.1% 89.9% 93.5% 92.2% 92.5% 151 Table 4.11, Continued Institutional Pathway CC/HSI CC/non-HSI HSI Non-HSI Total Latinos Weighted Sample n = 3,892 n = 8,250 n = 8,240 n = 40,851 n=61,233 Employer Uncorrected χ 2 =101.051, df=3 Used support from employer to finance college 9.4% 17.1% 8.7% 7.2% 8.8% Design-based F(2.51, 537.49)= 3.300, p<0.05 Did not use support from employer to finance college 90.6% 82.9% 91.3% 92.8% 91.2% Other Source Uncorrected χ 2 =125.369, df=3 Received support from some other source to finance college 7.3% 3.3% 0.4% 1.1% 1.7% Design-based F(1.79, 383.92)=3.873 , p<0.05 Did not receive support from some other source to finance college 92.7% 96.7% 99.6% 98.9% 98.3% Notes: Column proportions. ** May not sum to total due to rounding. Design-based F statistics are presented to account for the complex sample design of the 2003 NSRCG. Source: Analyses of the NSF 2003 National Survey of Recent College Graduates (NSRCG), using final survey weight (WTSURVY). 152 Differences in the Relative Level of Indebtedness by Institutional Pathway While the above tables reveal differences in the rate of borrowing among Latino STEM bachelor’s degree holders by institutional pathway, they do not indicate if and how the relative level of indebtedness resulting from borrowing differ by pathway. To address this question, I compare the relative debt levels (i.e., no debt, low debt, and high debt) 16 of Latino STEM bachelor’s degree holders below and examine the differences by associate degree and institutional pathway. Patterns of borrowing among Latino STEM bachelor’s degree holders who earned an associate degree and those that did not earn an associate degree were significantly (p<0.02) different. As shown in Table 4.12, associate degree holders were more likely to borrow at a high relative debt level (25.1%) than non-associate degree holders (19.8%). Interestingly, Latino STEM bachelor’s degree holders who earned an associate degree were more commonly in the ‘no debt’ category compared to non-associate degree earners (35.8% versus 25.9%). More than half of non-associate degree holders (54.3%) borrowed at a low relative debt level compared to 39.1% of associate degree holders. 16 For non-associate degree holders, low debt is defined as borrowing a non-zero amount less than one and a half times the mean cumulative indebtedness among all students in the graduating class at the B.S.-granting institution. High debt is defined as borrowing at least one and a half times the mean cumulative indebtedness among all students in the graduating class at the B.S.-granting institution. For associate degree holders, low debt is defined as borrowing a non- zero amount less than the mean cumulative indebtedness among all students in the graduating class at the B.S.-granting institution. High debt is defined as borrowing more than the mean cumulative indebtedness among all students in the graduating class at the B.S.-granting institution. 153 Table 4.12. Relative Debt Level of Latino STEM Bachelor’s Degree Holders by Associate Degree Notes: Column proportions. ** May not sum to total due to rounding. Design-based F statistics are presented to account for the complex sample design of the 2003 NSRCG. Source: Analyses of the 2003 College Board Annual Survey of Colleges and Universities and the NSF 2003 National Survey of Recent College Graduates (NSRCG), using final survey weight (WTSURVY). Table 4.13 illustrates that significant (p<0.02) differences exist in the levels of indebtedness among Latino STEM bachelor’s degree holders who followed each of the institutional pathways of interest. Those individuals in the ‘CC/HSI’ pathway were less likely to borrow at the low relative debt level and the high relative debt level than Latinos from the other three pathways. Latino STEM bachelor’s degree holders who followed the ‘non-HSI’ and ‘HSI’ pathways were similarly distributed amongst the no debt, low debt, and high debt categories. Latinos who followed the ‘CC/non-HSI’ pathway to the STEM B.S. degree were more likely to borrow at a high relative debt level compared to those individuals in the other three pathways. Institutional Pathway AA/AS @ CC No AA/AS @ CC Total Latinos Weighted Sample n = 12,142 n = 49,092 n=61,233 Relative Debt Level Uncorrected χ 2 =110.241, df=2 No Debt 35.8% 25.9% 27.9% Design-based Low Debt 39.1% 54.3% 51.3% F(1.94, 414.16)= 4.632, p<0.02 High Debt 25.1% 19.8% 20.8% 154 Table 4.13. Relative Debt Level of Latino STEM Bachelor’s Degree Holders by Institutional Pathway Institutional Pathway CC/HSI CC/non- HSI HSI Non-HSI Total Latinos Weighted Sample n = 3,892 n = 8,250 n = 8,240 n = 40,851 n=61,233 Relative Debt Level Uncorrected χ 2 =214.476 df=6 No Debt 52.0% 28.2% 23.8% 26.3% 27.9% Design-based Low Debt 30.3% 43.2% 52.6% 54.7% 51.3% F(4.75, 1016.73) = 3.042, p<0.02 High Debt 17.7% 28.6% 23.6% 20.0% 20.8% Notes: Column proportions. ** May not sum to total due to rounding. Design-based F statistics are presented to account for the complex sample design of the 2003 NSRCG. Source: Analyses of the 2003 College Board Annual Survey of Colleges and Universities and the NSF 2003 National Survey of Recent College Graduates (NSRCG), using final survey weight (WTSURVY). Differences in the Relative Level of Indebtedness by Parental Education Relative debt levels of Latino STEM bachelor’s degree holders varied significantly (p<0.001) by parental education level. As illustrated in Table 4.14, students whose parents had lower levels of educational attainment were most likely to borrow, and the majority of these students accumulated low relative debt levels. Students whose parents earned any graduate degree (master’s level or higher) were more likely to complete the bachelor’s degree with no debt. 155 Table 4.14. Relative Debt Level of Latino STEM Bachelor’s Degree Holders by Highest Parental Education Level Relative Debt Level No Debt Low Debt High Debt Weighted Sample n =17,064 n = 31,425 n = 12,743 Highest Parental Education Level Uncorrected χ 2 =399.388, df=12 Less than high school 28.0% 48.1% 23.9% Design-based F(9.72, 2079.23)=3.470, p<0.001 High school diploma or equivalent 16.5% 60.5% 23.1% Some college, vocational, or trade school 25.0% 48.9% 26.1% Bachelor’s degree 28.0% 58.1% 13.8% Master’s degree 35.8% 45.3% 18.9% Professional degree (e.g., JD, MD) 49.1% 45.7% 5.2% Doctorate 54.5% 32.8% 12.6% Total Latinos 27.9% 51.3% 20.8% ` Notes: Row proportions. ** May not sum to total due to rounding. Design-based F statistics are presented to account for the complex sample design of the 2003 NSRCG. Source: Analyses of the 2003 College Board Annual Survey of Colleges and Universities and the NSF 2003 National Survey of Recent College Graduates (NSRCG), using final survey weight (WTSURVY). Differences in Financial Aid Pathways and Relative Indebtedness by National Origin Very few differences exist in the forms of financial aid used by Latino STEM bachelor’s degree holders by national origin. As Table 4.15 illustrates, comparable proportions of Mexican Americans, Puerto Ricans, Cubans, and Latinos of ‘other’ national origin used loans, earnings, work study, employer support, and parental/familial support (gifts and loans) to finance college. There were significant differences (p<0.01) in the proportions of the aforementioned national origin groups using scholarships and/or grants to finance their undergraduate education: 76.5% of Puerto Ricans received scholarships and/or grants compared to 67.9% of Mexican Americans, 64.8% of Cubans and 57.0% of Latinos of ‘other’ national origin. There were no statistically significant differences in the patterns of indebtedness among Latino STEM bachelor’s degree holders by national origin (Table 4.16). In other words, similar proportions of Mexican American, Puerto Rican, Cuban, and ‘Other’ Latino STEM bachelor’s degree holders borrowed at the low debt and high debt levels. 156 Table 4.15. Forms of Financial Aid Used by Latino STEM Bachelor’s Degree Holders by National Origin National Origin Mexican American Puerto Rican Cuban Other Total Latinos Weighted Sample n = 29,160 n = 7,715 n = 3,904 n = 20,454 n=61,233 Loans Uncorrected χ 2 =61.322, df=3 Borrowed from school, bank, or government to finance college 68.5% 72.8% 61.4% 61.0% 66.1% Design-based F(2.82, 604.13)= 1.674, p=0.175 Did not borrow from school, bank, or government to finance college 31.5% 27.2% 38.6% 39.0% 33.9% Earnings Uncorrected χ 2 =71.360, df=3 Used earnings to finance college 61.4% 68.7% 62.9% 54.0% 60.0% Design-based F(2.93, 626.62)= 2.360, p=0.072 Did not use earnings to finance college 38.6% 31.3% 37.1% 46.0% 40.0% Scholarships/Grants Uncorrected χ 2 =134.975, df=3 Received scholarship or grant to finance college 67.9% 76.5% 64.8% 57.0% 65.2% Design-based F(2.95, 630.66)= 4.076, p<0.01 Did not receive scholarship or grant to finance college 32.1% 23.5% 35.2% 43.0% 34.8% Work Study Uncorrected χ 2 =36.820, df=3 Used work study to finance college 30.1% 21.4% 21.6% 28.3% 27.8% Design-based F(2.96, 633.01)= 1.263, p=0.286 Did not use work study to finance college 69.9% 78.6% 78.4% 71.7% 72.2% Parental/Familial Support (not to be repaid) Uncorrected χ 2 =3.668, df=3 Received parental/familial support not to be repaid to finance college 55.8% 58.2% 55.7% 54.6% 55.7% Design-based F(2.82, 602.82)= 0.102, p=0.952 Did not receive parental/familial support not to be repaid to finance college 44.2% 41.8% 44.3% 45.4% 44.3% 157 Table 4.15, Continued National Origin Mexican American Puerto Rican Cuban Other Latino Total Latinos Weighted Sample n = 29,160 n = 7,715 n = 3,904 n = 20,454 n = 61,233 Parental/Familial Loans Uncorrected χ 2 =3.668, df=3 Received loan from parents/family to finance college 8.1% 7.4% 11.7% 5.9% 7.5% Design-based F(2.89, 618.15)= 0.667, p=0.567 Did not receive loan from parents/family to finance college 91.9% 92.7% 88.3% 94.1% 92.5% Employer Uncorrected χ 2 =32.674, df=3 Used support from employer to finance college 10.7% 7.5% 4.6% 7.6% 8.8% Design-based F(2.50, 535.58)= 1.100, p=0.343 Did not use support from employer to finance college 89.3% 92.6% 95.4% 92.4% 91.2% Other Source Uncorrected χ 2 =110.160, df=3 Received support from some other source to finance college 0.4% 1.5% 0.3% 3.9% 1.7% Design-based F(2.12, 454.51)=7.728 , p<0.001 Did not receive support from some other source to finance college 99.6% 98.5% 99.7% 96.1% 98.3% Notes: ** May not sum to total due to rounding. Design-based F statistics are presented to account for the complex sample design of the 2003 NSRCG. Source: Analyses of the NSF 2003 National Survey of Recent College Graduates (NSRCG), using final survey weight (WTSURVY). 158 Table 4.16. Relative Debt Level of Latino STEM Bachelor’s Degree Holders by National Origin National Origin Mexican American Puerto Rican Cuban Other Total Latinos Weighted Sample n = 29,160 n = 7,715 n = 3,904 n = 20,454 n=61,233 Relative Debt Level Uncorrected χ 2 =78.107 df=6 No Debt 24.2% 25.8% 36.7% 32.2% 27.9% Design-based Low Debt 53.5% 50.7% 42.5% 50.1% 51.3% F(5.48, 1172.95) = 1.085, p=0.369 High Debt 22.3% 23.5% 20.8% 17.7% 20.8% Notes: Column proportions. ** May not sum to total due to rounding. Design-based F statistics are presented to account for the complex sample design of the 2003 NSRCG. Source: Analyses of the 2003 College Board Annual Survey of Colleges and Universities and the NSF 2003 National Survey of Recent College Graduates (NSRCG), using final survey weight (WTSURVY). College Financing Strategies Employed by Latino STEM Bachelor’s Degree Holders In order to characterize the underlying financing strategies employed by Latino STEM bachelor’s degree holders, I carried out a latent class analysis (LCA) of the manifest, or observed, financial aid variables in the NSRCG dataset. As explained in the previous chapter, latent class analysis is a multivariate technique that aims to uncover clusters of individuals who are similar with respect to a set of characteristics measured by outcomes. In this application, the outcomes include the forms of financial aid used by the respondents, i.e., loans, earnings, work study, scholarships/grants, employer support, parental/familial support (gifts), parental/familial loans, and other sources of support. The latent classes represent the financing strategies employed by the respondents. Three latent classes, or financing strategies, adequately (entropy=0.594) describe the eight dichotomous financial aid indicator variables in the NSRCG dataset. Results from the Vuong-Lo-Mendell-Rubin Likelihood Ratio Test (Muthén, 2001) for two (H 0 ) versus three classes indicates that two latent classes do not adequately describe the data (p<0.001), while the same test for four (H 0 ) versus three classes indicates that three classes are sufficient (p=0.539). 159 These three latent classes structure the NSRCG response cases with respect to the set of financial aid indicator variables. The conditional probabilities of using each of the eight forms of financial aid for each of the three classes are shown in Table 4.17 and graphed in Figure 4.1. Table 4.17. Conditional Probabilities of Using Financial Aid Forms, by Latent Class Latent Class Class 1 Class 2 Class 3 Form of Financial Aid Work Study 2.1% 3.3% 48.4% Loans (from school, bank, or government) 60.9% 26.2% 84.5% Earnings 53.2% 47.2% 67.9% Employer Support 23.5% 2.3% 6.2% Scholarships/Grants 45.6% 44.9% 87.2% Parental/Familial Support (not to be repaid) 14.9% 100.0% 63.2% Parental/Familial Loan 6.4% 8.4% 9.7% Other Source 5.8% 1.1% 1.0% Source: Analyses of the NSF 2003 National Survey of Recent College Graduates (NSRCG). 160 Figure 4.1 Conditional Probabilities of Use by Financial Aid Support Mechanism Source: Analyses of the NSF 2003 National Survey of Recent College Graduates (NSRCG). 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Financial Support Mechanism Conditional Probability of Use Class 1 2.1% 60.9% 53.2% 23.5% 45.6% 14.9% 6.4% 5.8% Class 2 3.3% 26.2% 47.2% 2.3% 44.9% >99.9% 8.4% 1.1% Class 3 48.4% 84.5% 67.9% 6.2% 87.2% 63.2% 9.7% 1.0% Work Study Loans (School, Bank, Gov't) Earnings Employer Scholarships/ Grants Parents/relatives not to be repaid Loan from Parents/ Relatives Other Source 161 Based on the patterns of conditional probabilities of using a particular form of financial aid, I interpreted the latent classes above to represent three, distinct college financing strategies: members of Class 1 financed college by being “self-supporters,” i.e., they received little to no parental support, instead relying primarily on loans, earnings, employer support and scholarships/grants. Members of Class 2 can be described as being “parentally supported”; i.e., they had a high probability of receiving parental/familial support not to be repaid, and mid-to-low probabilities of using loans, earnings, and scholarships/grants. Member of the final class, Class 3, financed college through “balanced support,” i.e., they had a high probability of using multiple forms of financial aid ranging from loans, work study, earnings, scholarships/grants, and parental/familial support (not to be repaid). For each respondent in the 2003 NSRCG dataset, conditional probabilities of belonging to each of the aforementioned latent classes were calculated based on his or her responses on the eight financial aid indicator variables. Each respondent was assigned to the class for which he or she had the highest probability of membership. Based on these class assignments (i.e., Class 1=Self-Support; Class 2=Parental Support; Class 3=Balanced Support), I was able to examine the differences in the college financing strategies employed by Latino STEM bachelor’s degree holders by institutional pathway, national origin, and other demographic and baccalaureate- granting institutional characteristics. These differences are discussed below. Table 4.18 indicates that significant differences (p<0.001) exist in the college financing strategies employed by Latino STEM bachelor degree holders who earned an associate degree and those who did not earn an AA/AS. Nearly 40% of Latino STEM B.S. degree holders who earned an associate degree were ‘self-supporters,’ compared to just below 15% of non-associate degree holders. A smaller proportion of associate degree holders were ‘parentally supported’ compared to non-associate degree holders (17.1% versus 30.0%). The largest proportions of 162 Latino AA/AS earners and non-AA/AS earners were classified as receiving ‘balanced support;’ however, Latino STEM bachelor’s degree holders who did not earn an associate degree more commonly fell into this ‘balanced support’ category than those who did earn an associate degree. Table 4.18. College Financing Strategies of Latino STEM Bachelor’s Degree Holders by Associate Degree Notes: Column proportions. ** May not sum to total due to rounding. Design-based F statistics are presented to account for the complex sample design of the 2003 NSRCG. Source: Analyses of the NSF 2003 National Survey of Recent College Graduates (NSRCG), using final survey weight (WTSURVY). Table 4.19 illustrates the college financing strategies employed by Latino STEM bachelor’s degree holders vary significantly (p<0.001) by institutional pathway. Nearly half of Latino STEM B.S. degree holders who followed the ‘CC/HSI’ pathway employed the ‘self- support’ financing strategy, compared to 34.3% of those in the ‘CC/non-HSI’ pathway, 17.9% of those in the ‘HSI’ pathway, and 14.2% of those in the ‘non-HSI’ pathway. Latino STEM bachelor’s degree holders who followed the ‘non-HSI’ pathway were the most likely to be ‘parentally supported;’ whereas similar proportions of Latinos following the ‘HSI’ and ‘non-HSI’ pathways used ‘balanced support’ as a college financing strategy (56.9% and 54.8%, respectively). Institutional Pathway AA/AS @ CC No AA/AS @ CC Total Latinos Weighted Sample n = 12,142 n = 49,092 n=61,233 College Financing Strategy Uncorrected χ 2 =450.207, df=2 Self-Support 39.2% 14.9% 19.7% Design-based Parental Support 17.1% 30.0% 27.4% F(1.84, 394.43)= 17.905, p<0.001 Balanced Support 43.7% 55.2% 52.9% 163 Table 4.19. College Financing Strategies Used by Latino STEM Bachelor’s Degree Holders by Institutional Pathway Institutional Pathway CC/HSI CC/non- HSI HSI Non-HSI Total Latinos Weighted Sample n = 3,892 n = 8,250 n = 8,240 n = 40,851 n=61,233 College Financing Strategy Uncorrected χ 2 =516.937 df=6 Self Support 49.7% 34.3% 17.9% 14.2% 19.7% Design-based Parental Support 16.3% 17.5% 25.2% 31.0% 27.4% F(5.11, 1092.76) = 7.479, p<0.001 Balanced Support 34.0% 48.3% 56.9% 54.8% 52.9% Notes: Column proportions. ** May not sum to total due to rounding. Design-based F statistics are presented to account for the complex sample design of the 2003 NSRCG. Source: Analyses of NSF 2003 National Survey of Recent College Graduates (NSRCG), using final survey weight (WTSURVY). Differences in the college financing strategies of Latino STEM bachelor’s degree holders by various demographic characteristics are shown in Table 4.20. Comparable (p=0.0915) percentages of Mexican American, Puerto Rican, Cuban and ‘other’ Latino STEM B.S.-degree holders utilized each of the financing strategies shown in the table below. There were, however, significant differences (p<0.001) in the strategies employed by non-traditionally aged and traditionally-aged Latino STEM bachelor’s degree holders. Non-traditional students employed the ‘self support’ financing strategy at a disproportionately high rate, relative to their share of the Latino STEM B.S. holder population. Further, non-traditional students were underrepresented among those students who employed the ‘parental support’ or ‘balanced support’ strategies. Table 4.20 shows that in-state and out-of-state college attendees were proportionately represented in each of the three identified college financing strategies (‘self-support,’ ‘parental support’, and ‘balanced support’). There were no significant differences in the distribution of Latino STEM bachelor’s degree holders that attended college in-state versus out-of-state. 164 Significant differences (p<0.001) exist in the college financing strategies employed by Latino STEM B.S. degree holders by the highest parental education level. Table 4.20 illustrates that individuals whose parents did not graduate from high school, or only have a high school diploma are overrepresented among Latino STEM bachelor’s degree holders who employed the ‘self support’ college financing strategy. Latinos whose parents earned a bachelor’s degree or higher were overrepresented among those who employed the financing strategy of ‘parental support.’ However, the ‘balanced support’ category saw roughly proportionate distribution of Latino STEM bachelor’s degree holders by highest parental education level. 165 Table 4.20. College Financing Strategies Used by Latino STEM Bachelor’s Degree Holders by Demographic Characteristics College Financing Strategy Self Support Parental Support Balanced Support Total Latinos Weighted Sample n = 3,892 n = 8,250 n = 8,240 n=61,233 National Origin Uncorrected χ 2 =121.301 df=6 Mexican American 45.8% 43.9% 50.2% 47.6% Design-based Puerto Rican 9.2% 10.2% 15.1% 12.6% F(5.44, 1164.54) = 1.864, p=0.0915 Cuban 9.1% 6.8% 5.2% 6.4% ‘Other’ Latino 35.9% 39.1% 29.5% 33.4% Non-Traditional Status Uncorrected χ 2 =1014.366 df=2 Non-Traditionally Aged Student 64.5% 16.6% 25.4% 30.7% Design-based F(1.96, 419.93)=44.516, p<0.001 Traditionally Aged Student 35.5% 83.4% 74.6% 69.3% In-State College Attendance Uncorrected χ 2 =16.326, df=2 Attended In-State 4-yr College 72.8% 78.5% 77.1% 76.6% Design-based F(1.98, 423.23)= 0.612, p=0.5410 Attended Out-of-State 4-yr college 27.2% 21.5% 22.9% 23.4% Highest Parental Education Level Uncorrected χ 2 =910.245, df=12 Less than high school 29.8% 4.4% 16.7% 15.9% Design-based High school diploma or equivalent 26.6% 11.5% 22.0% 20.0% F(10.32, 2208.35)= 7.668, p<0.001 Some college, vocational, or trade school 21.7% 26.3% 28.5% 26.5% Bachelor’s degree 11.3% 24.8% 12.7% 15.7% Master’s degree 6.7% 17.9% 13.6% 13.5% Professional degree (e.g., JD, MD) 3.6% 7.0% 4.2% 4.8% Doctorate 0.3% 8.1% 2.4% 3.6% Notes: Column proportions. ** May not sum to total due to rounding. Design-based F statistics are presented to account for the complex sample design of the 2003 NSRCG. Source: Analyses of NSF 2003 National Survey of Recent College Graduates (NSRCG), using final survey weight (WTSURVY). 166 Table 4.21 shows the distribution of Latino STEM bachelor’s degree holders using each of the three college financing strategies by baccalaureate-granting institutional characteristics. The table illustrates that over 73% of those Latinos who were categorized as ‘self-supporters’ attended public institutions – a disproportionately high share. Similarly, a disproportionately high share (72.6%) of Latinos using the ‘parental support’ financing strategy graduated from publicly controlled institutions. Among those Latinos who employed the college financing strategy of ‘balanced support’, 38.5% attended a private college/university; thus, private institution attendees were overrepresented among the ‘balanced support’ group. Statistically significant differences (p<0.002) also exist in the college financing strategies employed by Latinos attending institutions of various types. In terms of Carnegie classification, just below 27% of ‘self-supporters’ graduated from a Research University, despite the fact that about 40% of all Latino STEM bachelor’s degree holders graduated from these types of institutions. The proportion of ‘self-supporters’ attending all other institutional types (Doctoral University, Master’s College/University, Liberal Arts College, and Specialized Institutions) were higher than the overall proportion of Latino STEM B.S. holders in these institutional types. This clearly indicates that Latinos who graduated from a Research University were underrepresented among ‘self supporters,’ while Latinos attending other institutional types were overrepresented among ‘self supporters.’ Those Latino STEM B.S. holders who earned their degrees from a Doctoral University, Master’s College/University, Liberal Arts College, or Specialized Institution were underrepresented in the ‘parental support’ category, while those graduates of a Doctoral University and Specialized Institutions were slightly underrepresented among those who employed the ‘balanced support’ category. 167 Latino STEM bachelor’s degree holders who graduated from a Hispanic Serving Institution were overrepresented (p<0.03) among ‘self-supporters,’ and underrepresented among those employing the ‘parental support’ and ‘balanced support’ college financing strategies. Table 4.21 also shows that there were no significant differences in the college financing strategies employed by Latino STEM bachelor’s degree holders by state. 168 Table 4.21. College Financing Strategies Used by Latino STEM Bachelor’s Degree Holders by Baccalaureate-granting Institutional Characteristics College Financing Strategy Self Support Parental Support Balanced Support Total Latinos Weighted Sample n = 3,892 n = 8,250 n = 8,240 n=61,233 Control Uncorrected χ 2 =108.915 df=2 Public 73.5% 72.6% 61.5% 66.9% Design-based F(1.95, 414.87)=4.067, p<0.02 Private 26.5% 27.4% 38.5% 33.1% Carnegie Classification Uncorrected χ 2 =392.910, df=8 Research University 26.2% 45.8% 42.0% 39.9% Design-based F(6.08, 1277.53)= 3.742, p<0.002 Doctoral University 16.3% 11.4% 14.1% 13.8% Master’s College/University 39.1% 33.0% 32.6% 34.0% Liberal Arts College 13.8% 9.8% 11.4% 11.4% Specialized Institution 4.6% <1% <1% 0.9% Hispanic-Serving Institution Uncorrected χ 2 =86.084, df=2 Hispanic Serving Institution 28.2% 16.1% 18.6% 19.8% Design-based F(1.93, 413.18)= 3.889, p<0.03 Not a Hispanic Serving Institution 71.7% 83.9% 81.4% 80.2% State Uncorrected χ 2 =181.265, df=10 California 37.1% 34.3% 32.7% 34.0% Design-based F(9.04, 1935.08)= 1.583, p=0.1143 Florida 10.9% 10.4% 8.9% 9.7% Illinois 2.7% 2.4% 4.2% 3.4% New York 12.1% 5.8% 12.9% 10.8% Texas 5.6% 16.7% 11.5% 11.8% Other State 31.5% 30.4% 29.7% 30.3% Notes: Column proportions. ** May not sum to total due to rounding. Design-based F statistics are presented to account for the complex sample design of the 2003 NSRCG. Source: Analyses of NSF 2003 National Survey of Recent College Graduates (NSRCG), using final survey weight (WTSURVY). 169 The Effect of Indebtedness on Graduate School Enrollment among Latino STEM Bachelor’s Degree Holders The final question addressed in the present study pertains to the effect of indebtedness on graduate school enrollment among Latino STEM bachelor’s degree holders. As described in the previous chapter, this question was addressed using two analytical methods: logistic regression and propensity score matching (PSM). Findings using these two methods are presented and compared below. Results of Logistic Regression Analyses Tables 4.22 through 4.24 present the results of two design-based logistic models of graduate school enrollment. In the first model, I controlled for seven of the eight, dichotomous financial aid indicator variables (e.g., scholarships/grants, earnings, work study, etc.) as well as various individual-level characteristics and social, economic, financial aid policy, and institutional-level contextual variables in order to understand the effect of high debt and low debt on graduate school enrollment, relative to no debt. The eighth dichotomous financial aid variable, loans (1=Yes, 0=No) was dropped from the model due to collinearity. In the second model, in addition to individual-level and other larger contextual variables, I included the latent college financing strategies (i.e., self-support, parental support, and balanced support) in lieu of the eight manifest financial aid variables. This second model allowed for the determination of the effect of high debt and low debt on graduate school enrollment, relative to no debt, controlling for the college financing strategy and other relevant factors. Results from the first logistic regression model, which included the financial aid indicator variables, reveal that controlling for all other variables of interest (listed in Table 4.23), Latinos who completed the STEM bachelor’s degree with a low relative level of indebtedness, i.e., low debt, were 8% (marginal effect = −0.110, p<0.05) less likely to attend graduate school relative to 170 those Latino students who graduated with no student loan debt. The first model does not indicate that Latino STEM bachelor’s degree holders graduating with a high relative level of indebtedness, i.e., high debt, experience a change in probability of enrolling in graduate school relative Latinos who did not borrow to finance college. The full model results using the eight financial aid indicator variables are shown in Table 4.23. An F-adjusted mean residual goodness-of-fit test applied to the design-based logistic regression model using the Stata module svylogitgof (Archer, Lemeshow & Hosmer, 2007) suggested no evidence of lack of fit (p=0.407). Table 4.22. Estimates of Average Treatment Effect of Relative Indebtedness on Graduate School Enrollment, for Latino STEM Bachelor’s Degree Recipients, Using Logistic Regression (1) (2) Logistic Regression Model with Financial Aid Indicator Variables Logistic Regression Model with Latent Financing Strategy Variables Relative Level of Indebtedness (Reference is ‘No Debt’) Marginal Effect S.E. † Marginal Effect S.E. † High Debt − 0.095 0.275 * − 0.080 0.299 Low Debt − 0.110 0.217 * − 0.096 0.242 Notes: *** p<0.001/k, ** p<0.01/k, * p<0.05/k, where k is the number of comparison categories for the Bonferroni adjustment. †robust standard errors of the beta coefficients (not shown) which take into account the clustered (by undergraduate institution) and stratified (by race/ethnicity, gender, and field of study) sample design of the NSRCG. Source: Analyses of the 2003 NSF NSRCG, using final survey weight (WTSURVY). Table 4.22 indicates that the second logistic regression model, in which the latent college financing strategy variables were included, does not indicate that low or high levels of indebtedness result in a change in the probability of attending graduate school among Latino STEM bachelor’s degree holders. Full model results for the second logistic regression model are listed in Table 4.24. An F-adjusted mean residual goodness-of-fit test applied to the design-based logistic regression model using the Stata module svylogitgof (Archer, Lemeshow & Hosmer, 2007) suggested no evidence of lack of fit (p=0.335). 171 Table 4.23. Logistic Regression Model of Graduate School Enrollment including Financial Aid Indicator Variables for Latino STEM Bachelor’s Degree Holders Marginal Effect S.E.† Social, Economic and Policy Contextual Variables State California −0.066 0.205 *** Florida 0.033 0.290 *** Illinois −0.131 0.507 *** New York −0.012 0.331 *** Texas −0.060 0.361 *** Other State (reference) National Origin Mexican American 0.032 0.193 *** Puerto Rican −0.023 0.294 *** Cuban 0.026 0.387 *** ‘Other’ Latino (reference) Gender Female −0.025 0.182 *** Male (reference) Field of Study Computer Science −0.129 0.235 *** Mathematics −0.040 0.336 *** Biological, agricultural, and environmental science 0.249 0.329 *** Physical Science 0.318 0.485 *** Engineering 0.141 0.259 *** Social Science 0.169 0.263 *** Science & Engineering-related (reference) Higher Education Financial Aid Contextual Variables Forms of Financial Aid Work Study −0.009 0.183 *** Did not receive Work Study (reference) Earnings −0.032 0.187 *** Did not use Earnings to finance college (reference) Parental/Familial Support (not to be repaid) −0.107 0.158 *** Did not receive Parental/Familial Support (reference) Scholarships and/or Grants 0.032 0.192 *** Did not receive Scholarships and/or Grants (reference) Parental/Familial Loans 0.008 0.296 *** Did not receive Parental/Familial Loan (reference) Employer Support 0.053 0.318 *** Did not receive Employer Support to finance college (reference) Other Source −0.060 0.589 *** Did not receive financing help from another source (reference) Relative Debt Level High Debt −0.095 0.275 *** Low Debt −0.110 0.217 *** No Debt (reference) B.S.-granting Institutional Contextual Variables Control Private −0.016 0.219 *** Public (reference) Hispanic Serving Institution HSI −0.060 0.237 *** Non-HSI (reference) 172 Table 4.23, Continued Marginal Effect S.E.† Individual-Level Characteristics Associate Degree Associate Degree Holder 0.057 0.236 *** Non-Associate Degree Holder (reference) Non-Traditional Student Status Non-Traditionally Aged Student −0.089 0.206 *** Traditionally Aged Student (reference) Undergraduate GPA 3.25 −3.74 (About half A’s/half B’s) −0.128 0.260 *** 2.75 −3.24 (Mostly B’s) −0.227 0.255 *** Less than 2.75 −0.232 0.327 *** 3.75 −4.00, Mostly A’s (reference) Notes: *** p<0.001/k, ** p<0.01/k, * p<0.05/k, where k is the number of comparison categories. † Robust standard errors of the beta coefficients (not shown) which take into account the clustered (by undergraduate institution) and stratified (by race/ethnicity, gender, and field of study) sample design of the NSRCG. Source: Analyses of the 2003 NSF NSRCG, using final survey weight (WTSURVY). 173 Table 4.24. Logistic Regression Model of Graduate School Enrollment including Latent College Financing Strategies, for Latino STEM Bachelor’s Degree Holders Marginal Effect S.E.† Social, Economic and Policy Contextual Variables State California − 0.063 0.214 *** Florida 0.039 0.284 *** Illinois − 0.128 0.519 *** New York − 0.002 0.336 *** Texas − 0.053 0.359 *** Other State (reference) National Origin Mexican American 0.034 0.200 *** Puerto Rican − 0.025 0.292 *** Cuban 0.019 0.391 *** ‘Other’ Latino (reference) Gender Female −0.017 0.174 *** Male (reference) Field of Study Computer Science − 0.140 0.236 *** Mathematics − 0.048 0.337 *** Biological, agricultural, and environmental science 0.258 0.325 *** Physical Science 0.315 0.476 *** Engineering 0.147 0.264 *** Social Science 0.174 0.269 *** Science & Engineering-related (reference) Higher Education Financial Aid Contextual Variables College Financing Strategy Self Supported 0.092 0.250 *** Parental Supported −0.041 0.237 *** Balanced Supported (reference) Relative Debt Level High Debt −0.080 0.299 *** Low Debt −0.096 0.242 *** No Debt (reference) B.S.-granting Institutional Contextual Variables Control Private −0.014 0.205 *** Public (reference) Hispanic Serving Institution HSI −0.06 0.246 *** Non-HSI (reference) Individual-Level Characteristics Associate Degree Associate Degree Holder 0.055 0.235 *** Non-Associate Degree Holder (reference) Non-Traditional Student Status Non-Traditionally Aged Student −0.084 0.204 *** Traditionally Aged Student (reference) 174 Table 4.24, Continued Marginal Effect S.E.† Undergraduate GPA 3.25-3.50 (A’s and B’s) −0.125 0.261 *** 2.75-3.25 (Mostly B’s) −0.228 0.262 *** Less than 2.75 −0.233 0.325 *** 3.75 or higher, Mostly A’s (reference) Notes: *** p<0.001/k, ** p<0.01/k, * p<0.05/k, where k is the number of comparison categories. † Robust standard errors of the beta coefficients (not shown) which take into account the clustered (by undergraduate institution) and stratified (by race/ethnicity, gender, and field of study) sample design of the NSRCG. Source: Analyses of the 2003 NSF NSRCG, using final survey weight (WTSURVY). As explained the previous chapter, the above logistic regression parameter estimates are potentially biased due to endogeneity, i.e., self-selection bias. Thus, in order to check for biased estimates of treatment effects, I used the method of propensity score matching to determine the effect of low debt and high debt on graduate school enrollment. The results of the PSM analysis are presented below. Results of Propensity Score Matching Analysis Propensity score matching involves three distinct steps: (1) the calculation of propensity scores, that is, the predicted probability of undergoing the treatment(s) of interest; (2) the creation of a counterfactual framework by matching untreated and treated cases with the same, or similar propensity scores; and (3) the calculation of treatment effects, i.e., the average treatment effect (ATE), the average treatment effect on the treated (ATT), and the average treatment effect on the untreated (ATU). Propensity scores. In this section, I describe the constrained multinomial probit regression model used to generate the predicted probabilities of borrowing at the low debt level and the high debt level. These predicted probabilities were then used as propensity scores. Table 4.25 illustrates that among Latinos, being a non-traditionally aged student increased the probability (marginal effect = 0.080, p<0.01) of having ‘low debt’. Earning an associate degree decreased the likelihood (marginal effect = –0.152, p<0.01) of having ‘low 175 debt’. Latinos who had a parent with a graduate degree of any type had a lower probability (marginal effect = –0.178, p<0.01) of borrowing at a low relative debt level. Table 4.25 also reveals that gender and non-traditional student status have statistically significant effects on the probability of borrowing at a high level among Latinos who earned a STEM bachelor’s degree. Being female negatively influences the probability (marginal effect = – 0.043, p<0.05) of having a high relative debt level among Latinos, while being a non-traditional student increases the probability (marginal effect = 0.061, p<0.01) of finishing college with a high cumulative relative debt level. 176 Table 4.25. Constrained Multinomial Probit Model of Relative Indebtedness (by Level) for Latino STEM Bachelor’s Degree Holders Low Relative Debt High Relative Debt Marginal Effect S.E.† Marginal Effect S.E.† Social, Economic and Policy Contextual Variables State California 0.052 0.224 *** 0.022 0.224 *** Florida 0.049 0.296 *** 0.021 0.296 *** Illinois 0.031 0.481 *** 0.013 0.481 *** New York 0.079 0.274 *** 0.033 0.274 *** Texas 0.074 0.289 *** 0.032 0.289 *** Other State (reference) National Origin Mexican American 0.046 0.193 *** 0.044 0.224 *** Puerto Rican 0.031 0.299 *** 0.045 0.317 *** Cuban −0.132 0.309 *** 0.038 0.311 *** ‘Other’ Latino (reference) Gender Female −0.038 0.154 *** −0.043 0.185 *** Male (reference) Highest Parental Education Level Less than High School −0.095 0.314 *** 0.085 0.326 *** High School Diploma or equivalent −0.004 0.234 *** 0.089 0.249 *** Some College, Trade School, or Vocational −0.110 0.202 *** 0.121 0.231 *** Graduate Degree (Master’s, Professional, or Doctorate) −0.178 0.222 *** −0.002 0.224 *** Bachelor’s Degree (reference) In-State College Attendance Attended college in-state 0.114 0.234 *** −0.014 0.251 *** Attended college out-of-state (reference) 177 Table 4.25, Continued Low Relative Debt High Relative Debt Marginal Effect S.E.† Marginal Effect S.E.† Social, Economic and Policy Contextual Variables (cont’d) Field of Study Computer Science −0.008 0.274 *** 0.101 0.281 *** Mathematics −0.074 0.316 *** 0.028 0.293 *** Biological, agricultural, and environmental science 0.085 0.311 *** −0.002 0.318 *** Physical Science −0.028 0.390 *** 0.037 0.387 *** Engineering 0.008 0.258 *** −0.008 0.247 *** Social Science 0.072 0.281 *** −0.038 0.245 *** Science & Engineering-related (reference) B.S.-granting Institutional Contextual Variables Control Private 0.152 0.442 *** 0.064 0.442 *** Public (reference) Tuition Tuition (Dollars) −0.000001 0.000022 *** −0.000001 0.000023 *** Borrowing Rate Percent of Undergraduates who Borrow 0.002 0.005 *** 0.001 0.005 *** Carnegie Classification Research University 0.134 0.995 *** 0.056 0.995 *** Doctoral University 0.168 1.007 *** 0.074 1.007 *** Master’s College/University 0.002 1.029 *** 0.001 1.029 *** Liberal Arts 0.055 1.028 *** 0.023 1.028 *** Other (reference) Selectivity Less Competitive −0.087 0.350 *** −0.035 0.350 *** Competitive 0.018 0.352 *** 0.007 0.352 *** Competitive Plus 0.167 0.453 *** 0.075 0.453 *** Very Competitive −0.094 0.359 *** −0.038 0.359 *** Very Competitive Plus 0.132 0.507 *** 0.058 0.507 *** Highly Competitive −0.066 0.385 *** −0.027 0.385 *** Highly Competitive Plus 0.168 0.561 *** 0.075 0.561 *** Most Competitive −0.0002 0.419 *** −0.0001 0.419 *** Special −0.181 0.467 *** −0.070 0.467 *** Non-Competitive (reference) ** 178 Table 4.25, Continued Low Relative Debt High Relative Debt Marginal Effect S.E.† Marginal Effect S.E.† Individual-Level Characteristics Associate Degree Associate Degree Holder −0.152 0.212 *** 0.012 0.235 *** Non-Associate Degree Holder (reference) Non-Traditional Student Status Non-Traditionally Aged Student 0.080 0.162 *** 0.061 0.204 *** Traditionally Aged Student (reference) Wald χ 2 138.26 Notes: *** p<0.001/k, ** p<0.01/k, * p<0.05/k, where k is the number of comparison categories. † Robust standard errors of the beta coefficients (not shown) which take into account the clustered (by undergraduate institution) and stratified (by race/ethnicity, gender, and field of study) sample design of the NSRCG. Source: Analyses of the 2003 NSF NSRCG, using final survey weight (WTSURVY). 179 As stated above, the predicted probabilities of having low debt and high debt generated by the constrained multinomial probit model were used as propensity scores. Tables 4.26 and 4.27 show the cumulative distributions of propensity scores for low debt and high debt. These tables illustrate that many non-borrowers (i.e., those with no debt) have similar propensity scores to those Latinos who borrowed at a low relative level and high relative level. Box plots (Figure 4.2 and Figure 4.3) illustrate this overlap in propensity scores, or, common support area, of the treated and untreated groups. Within the respective common support areas, borrowers with high relative debt and low relative debt (treated cases) and non-borrowers (untreated cases) were matched on the propensity scores and the differences in graduate school enrollment between the treated and untreated cases were interpreted as the treatment effect by relative level of indebtedness. 180 Table 4.26. Cumulative Distribution of Estimated Propensity Scores for Borrowing at a Low Relative Debt Level (Derived from Constrained Multinomial Probit Regression Model in Table 4.25) Low Relative Debt No Debt Propensity Score Proportion Sample Size Proportion Sample Size 0.10 0.000 0 0.004 1 0.12 0.002 1 0.004 1 0.14 0.004 2 0.004 1 0.15 0.004 2 0.007 2 0.16 0.005 3 0.011 3 0.17 0.005 3 0.015 4 0.18 0.011 6 0.015 4 0.19 0.011 6 0.022 6 0.20 0.013 7 0.026 7 0.21 0.016 9 0.033 9 0.22 0.016 9 0.037 10 0.23 0.016 9 0.055 15 0.24 0.018 10 0.063 17 0.25 0.020 11 0.066 18 0.26 0.026 14 0.066 18 0.27 0.029 16 0.081 22 0.28 0.029 16 0.100 27 0.29 0.036 20 0.107 29 0.30 0.044 24 0.129 35 0.31 0.051 28 0.151 41 0.32 0.053 29 0.162 44 0.33 0.071 39 0.173 47 0.34 0.077 42 0.188 51 0.35 0.088 48 0.229 62 0.36 0.102 56 0.255 69 0.37 0.106 58 0.280 76 0.38 0.128 70 0.303 82 0.39 0.148 81 0.328 89 0.40 0.173 95 0.339 92 0.41 0.182 100 0.369 100 0.42 0.192 105 0.387 105 0.43 0.219 120 0.410 111 0.44 0.239 131 0.424 115 0.45 0.257 141 0.450 122 0.46 0.281 154 0.472 128 0.47 0.299 164 0.509 138 0.48 0.330 181 0.546 148 0.49 0.367 201 0.583 158 0.50 0.407 223 0.624 169 0.51 0.432 237 0.661 179 181 Table 4.26, Continued Low Relative Debt No Debt Propensity Score Proportion Sample Size Proportion Sample Size 0.52 0.456 250 0.675 183 0.53 0.485 266 0.697 189 0.54 0.520 285 0.712 193 0.55 0.544 298 0.734 199 0.56 0.586 321 0.756 205 0.57 0.626 343 0.779 211 0.58 0.651 357 0.808 219 0.59 0.682 374 0.830 225 0.60 0.697 382 0.867 235 0.61 0.721 395 0.878 238 0.62 0.746 409 0.897 243 0.63 0.776 425 0.919 249 0.64 0.797 437 0.934 253 0.65 0.827 453 0.945 256 0.66 0.845 463 0.963 261 0.67 0.869 476 0.970 263 0.68 0.903 495 0.978 265 0.69 0.914 501 0.982 266 0.70 0.925 507 0.989 268 0.71 0.940 515 0.989 268 0.72 0.943 517 0.989 268 0.73 0.956 524 0.996 270 0.74 0.971 532 0.996 270 0.75 0.982 538 0.996 270 0.76 0.989 542 0.996 270 0.77 0.996 546 1.000 271 0.79 0.998 547 1.000 271 0.80 1.000 548 1.000 271 Source: Analyses of the 2003 NSF NSRCG, using propensity scores generated from a constrained multinomial probit model and weighted data (WTSURVY). 182 Table 4.27. Cumulative Distribution of Estimated Propensity Scores for Borrowing at a High Relative Debt Level (Derived from Constrained Multinomial Probit Regression Model in Table 4.25) High Relative Debt No Debt Propensity Score Proportion Sample Size Proportion Sample Size 0.03 0.004 1 0.004 1 0.04 0.004 1 0.007 2 0.05 0.008 2 0.022 6 0.06 0.016 4 0.044 12 0.07 0.016 4 0.070 19 0.08 0.016 4 0.092 25 0.09 0.033 8 0.137 37 0.10 0.049 12 0.166 45 0.11 0.061 15 0.214 58 0.12 0.065 16 0.255 69 0.13 0.081 20 0.314 85 0.14 0.122 30 0.362 98 0.15 0.167 41 0.410 111 0.16 0.195 48 0.465 126 0.17 0.224 55 0.491 133 0.18 0.264 65 0.535 145 0.19 0.341 84 0.576 156 0.20 0.411 101 0.627 170 0.21 0.459 113 0.649 176 0.22 0.484 119 0.683 185 0.23 0.520 128 0.738 200 0.24 0.565 139 0.775 210 0.25 0.614 151 0.819 222 0.26 0.650 160 0.841 228 0.27 0.683 168 0.863 234 0.28 0.711 175 0.893 242 0.29 0.728 179 0.926 251 0.30 0.764 188 0.934 253 0.31 0.801 197 0.945 256 0.32 0.821 202 0.948 257 0.33 0.850 209 0.967 262 0.34 0.882 217 0.970 263 0.35 0.902 222 0.974 264 0.36 0.911 224 0.978 265 0.37 0.915 225 0.982 266 0.38 0.923 227 0.989 268 0.39 0.943 232 0.989 268 0.40 0.955 235 0.993 269 0.41 0.959 236 0.996 270 0.42 0.972 239 1.000 271 183 Table 4.27, Continued High Relative Debt No Debt Propensity Score Proportion Sample Size Proportion Sample Size 0.43 0.976 240 1.000 271 0.44 0.980 241 1.000 271 0.45 0.980 241 1.000 271 0.46 0.980 241 1.000 271 0.47 0.984 242 1.000 271 0.48 0.988 243 1.000 271 0.49 0.992 244 1.000 271 0.50 0.992 244 1.000 271 0.51 1.000 246 1.000 271 Source: Analyses of the 2003 NSF NSRCG, using propensity scores generated from a constrained multinomial probit model and weighted data (WTSURVY). 184 Figure 4.2. Estimated Propensity Scores of Borrowing at a Low Relative Debt Level, by Treatment low debt no debt . 0.80 0.60 0.40 0.20 0.00 . Source: Analyses of the 2003 NSF NSRCG, using propensity scores generated from a constrained multinomial probit model and weighted data (WTSURVY). No Debt (Untreated Group) Low Relative Debt (Treated Group) Propensity to Borrow at a Low Relative Debt Level 185 Figure 4.3. Estimated Propensity Scores of Borrowing at a High Relative Debt Level, by Treatment high debt no debt . 0.60 0.50 0.40 0.30 0.20 0.10 0.00 . Source: Analyses of the 2003 NSF NSRCG, using propensity scores generated from a constrained multinomial probit model and weighted data (WTSURVY). No Debt (Untreated Group) High Relative Debt (Treated Group) Propensity to Borrow at a High Relative Debt Level 186 Treatment effects derived from Propensity Score Matching. The average treatment effect (ATE), treatment effect on the treated (ATT), and treatment effect on the untreated (ATU) resulting from the PSM analysis using the kernel matching technique (bandwidth = 0.06) are shown in Table 4.28. The standard errors were calculated using bootstrap repetition. The percentage of cases that were matched, i.e., fell within the common support area, are also listed. Table 4.28. Estimates of the Average Treatment Effect (ATE), Average Treatment on the Treated (ATT), and Average Treatment on the Untreated (ATU) Effect of Relative Level of Indebtedness (Low, High) on 2000-01/2001-02 STEM Bachelor’s Degree Recipients Graduate School Enrollment, Based on Matching of Propensity Scores Derived from the Multinomial Probit Model in Table 4.25 Low Relative Debt High Relative Debt ATE ATT ATU ATE ATT ATU Kernel Matching Estimates −0.141 *** −0.138 *** −0.156 *** −0.185 *** −0.171 **** −0.198 *** S.E. 0.0403 * 0.0370 * 0.0351 * 0.0507 * 0.0445 * 0.0383 * Percent of Cases Matched 100 * 99.2 Notes: *** p<0.001, ** p<0.01, * p<0.05. Standard errors were generated using bootstrap replications. Source: Analyses of the 2003 NSF NSRCG, using propensity scores generated from a multinomial probit model and weighted data (WTSURVY). The results illustrate that the effect of indebtedness on graduate school enrollment is statistically significant and negative for both low levels of debt and high levels of debt. As shown in Table 4.28, the size of the ATT effect of borrowing at a low level is smaller than the estimate of the ATE. This indicates that among Latinos who did borrow at a low level, the negative effect of that low relative debt burden on the likelihood of enrolling graduate school (change in probability = –13.8%) is slightly smaller than the effect experienced by the entire sample (change in probability = –14.1%). This suggests that Latino STEM B.S. holders who have a higher propensity of borrowing at a low level experience a smaller ‘penalty’ on the chances of enrolling in graduate school than those with a lower propensity to borrow at a low level. The size of the ATU effect of borrowing at a low level is larger than the estimated average treatment effect. This demonstrates that if those Latinos who did not borrow at all had borrowed at a low 187 level, they would have experienced a larger negative effect on their probability of enrolling in graduate school (change in probability = –15.6%). This implies that the ‘penalty’ on the chances of enrolling in graduate school is greater for Latinos with a low propensity to have a low relative debt burden. For relatively high levels of debt, the size of the ATT effect is smaller than the estimate of the ATE. This means that Latino STEM bachelor’s degree holders who did borrow at a high level experienced a smaller negative effect of that high relative debt burden (change in probability = –17.1%) on their likelihood of enrolling in graduate school than that experienced by the entire sample (change in probability = –18.5%). Further, the ATT effect suggests that Latinos with a higher propensity to borrow at a high level experience a smaller ‘penalty’ than those with a lower propensity to graduate with high relative debt. The ATU effect of borrowing at a high relative debt level exceeds the ATE. This indicates that if those Latinos who did not borrow at all had a high relative level of indebtedness, they would have experienced a larger negative effect (–19.8%) on their probability of enrolling in graduate school than that of the entire sample. This result implies that the negative effects of high relative debt are higher for those Latinos with lower chances of borrowing at a high level. Comparison of Propensity Score Matching and Logistic Regression treatment effect estimates. Treatment effect estimates from the propensity score matching analysis differ with those obtained through the logistic regression analyses shown in Table 4.22. The logistic regression analyses revealed that a low relative level of indebtedness has a statistically significant negative average treatment effect (ATE) [–11% (–0.110, p<0.05)] on the probability of graduate school enrollment when financial aid indicator variables were used. The estimated ATE of –14% (–0.141, p<0.001), ATT effect of –14% (–0.138, p<0.001), and ATU effect of –16% (–0.156, p<0.001) derived from the propensity score matching are larger than the logistic regression 188 estimate 17 . This indicates that the calculated marginal effect size from the logistic regression model using financial aid indicator variables has an overall bias 18 of +0.031 (–0.110–(–0.141)) and reveals a self-selection bias of +0.028 (–0.110–(–0.138)). Contrary to the logistic regression model in Table 4.22 which indicated that the effect of low debt on graduate school enrollment was not statistically significant when latent college financing strategy variables (i.e., self-supported, parentally-supported, balanced-supported) were used in the model, the propensity score matching found highly significant negative effects of low debt on graduate school enrollment. Further, the logistic regression model results in Table 4.22 showed that high debt was not found to have a statistically significant relationship with graduate school enrollment in either model (with financial aid indicator variables or latent financing strategy variables). This stands in contrast to the statistically significant negative effects of high debt on the probability of enrolling in graduate school found via the propensity score matching analysis. The inconsistencies between the logistic regression models and the results of the propensity score matching underscore previous assertions by researchers (Attewell, Lavin, Domina & Levey, 2006; Ichino, Mealli, & Nannicini, 2006; Rosenbaum, 1984; Titus, 2007) that logistic regression models do not appropriately correct for self-selection bias, resulting in biased parameter estimates. Sensitivity Analyses While propensity score matching is widely preferred over standard regression techniques (Titus, 2007), PSM is not an infallible method of analysis. There are several pitfalls that can 17 In order to determine if the differences in treatment effect estimates using logistic regression and propensity score matching were due to factors related to weighting, the logistic regression was run without weights as a sensitivity analysis. The marginal effect estimates for the effect of low relative debt and high relative debt on graduate school enrollment derived from this sensitivity analysis were statistically significant and negative (-0.11 and -0.14, respectively), but were still smaller in magnitude than the PSM treatment effect estimates. 18 Bias estimates cannot account for the possibility of selection on “unobservables”. 189 potentially undermine the validity of PSM results. Two such pitfalls are the quality of the model used to predict propensity scores, and unobserved variables that might affect the propensity score estimates and the outcome of interest. Below, I discuss two types of sensitivity analyses I conducted to bolster the validity of my findings. Testing the sensitivity of estimated treatment effects to modeled propensity scores. In order to establish the soundness PSM results, I tested the sensitivity of treatment effect estimates to the modeled propensity scores. By changing the observed covariates included in the constrained multinomial probit model used to estimate the propensity scores (i.e. excluding selected covariates from the original model), I was able to determine the change in treatment effects caused by these alterations. Other researchers using propensity score matching (Titus, personal communication, 1/31/08) have used this technique to establish the validity of their results. Table 4.29 shows treatment effect estimates (ATT, ATU, and ATE) resulting from the propensity score calculations using four different constrained multinomial probit models (Models 1-4) to predict the probability of treatment (low debt, high debt). The covariates removed from Model 1 through Model 4 are indicated in the second column of the table. The resulting predicted probabilities from each model were interpreted as propensity scores and treatment effects were calculated using kernel matching techniques within the Stata module PSMATCH2 (Leuven & Sianesi, 2003). Results from the final model are also shown in the table for comparison purposes. 190 Table 4.29. Estimates of the Treatment Effects of Relative Level of Indebtedness on Graduate School Enrollment Low Relative Debt High Relative Debt Removed Covariates ATE ATT ATU ATE ATT ATU Model 1 State −0.154 *** −0.150 *** −0.161 *** −0.182 *** −0.162 **** −0.200 *** Model 2 Field of Study −0.148 *** −0.146 *** −0.154 *** −0.174 *** −0.158 **** −0.189 *** Model 3 National Origin −0.155 *** −0.152 *** −0.162 *** −0.184 *** −0.165 **** −0.201 *** Model 4 Tuition and Percent Borrowers at B.S.- granting Institution −0.151 *** −0.147 *** −0.160 *** −0.186 *** −0.168 **** −0.203 *** Final Model −0.141 *** −0.134 *** −0.156 *** −0.185 *** −0.171 **** −0.198 *** Notes: *** p<0.001, ** p<0.01, * p<0.05. Source: Analyses of the 2003 NSF NSRCG, using propensity scores generated from a multinomial probit model and weighted data (WTSURVY). As the table illustrates, the removal of covariates from the probit model used to predict propensity of incurring high and low levels of debt alters the treatment effect estimates derived from PSM. However the direction of the effect remains negative across all models, and range of the treatment effect estimates in Models 1-4 lies within one standard error of the final model’s treatment effect estimates. Testing the robustness of treatment effect estimates against violations of the Conditional Independence Assumption. While PSM is capable of correcting for self-selection biases, it is possible that unobserved characteristics or “unobservables” could simultaneously impact the outcome variable of interest (i.e., graduate school enrollment) and the propensity score (i.e., predicted probability of participating in a loan program at a given level), constituting a violation of the Conditional Independence Assumption (Caliendo & Kopeinig, 2005). The presence of such an unobserved determinant might create a hidden bias against which estimates of the treatment effects derived from PSM are not robust (Becker & Caliendo, 2007; Rosenbaum, 2002) undermining the validity of the PSM model. Though the conditional independence assumption cannot be directly tested (Caliendo & Kopeinig, 2005; Rosenbaum & Rubin, 1983), it is possible to test the robustness of the treatment 191 effect estimates against hidden biases introduced by unobserved variables using the bounding approach first proposed by Rosenbaum (2002). Sensitivity analysis using the bounding approach tests how strongly ‘unobservables’ could simultaneously influence the propensity to participate in treatment and the outcome variable without undermining the results of the propensity score matching analysis. For example, if two individuals had identical observed covariates (e.g., national origin, gender, SES), they would have the same probability of receiving treatment (e.g., borrowing at a high level), and thus, the same propensity score. However, there could be some unobserved factor that altered the probability of borrowing at a high level so that the true propensity scores of the individuals are in fact different. The Stata module, MHBOUNDS (Becker & Caliendo, 2007), introduces different levels of bias that could be caused by unobserved variables and tests how large the hidden bias would have to be to render the PSM results insignificant. I used the MHBOUNDS module to test the robustness of the propensity score matching treatment effect estimates against unobserved variables. In the table below (Table 4.30), Γ represents the odds of differential assignment due to unobserved factors. Thus, Γ=1 corresponds to the assumption of no hidden bias, Γ=1.5 assumes a hidden biases that affects the odds ratio of treatment assignment to differ between the treatment and comparison groups by 1.5. The second and third columns of the table contain the significance level (p-value) of the low debt and high debt treatment effect under the assumption of the specified bias level ( Γ), respectively. Bolded p-values are not significant. The results of the sensitivity analysis testing the robustness of the treatment effect estimates against bias due to selection on ‘unobservables’ indicate that for certain levels of Γ, the treatment effect estimates become insignificant. More specifically, hidden biases that affect the odds of selection into the low debt category by 1.5 to 2.4 render the results insignificant. For Γ=2.5 or higher, the low debt treatment effect becomes significant again. The treatment effects of 192 high debt become insignificant for hidden biases that affect the odds of selection into the high debt category by 1.4-2.6. For Γ=2.7 or higher, the high debt treatment effect is again significant. While the results of the sensitivity analysis procedure described above cannot directly test the conditional independence assumption, they indicate that the model is somewhat sensitive to selection on ‘unobservables’ and that careful interpretations of the PSM-derived treatment effects should be made. Table 4.30. Results of Mantel-Haenszel Bounds Sensitivity Analysis Low Relative Debt Treatment High Relative Debt Treatment Γ (Assumed Bias Level) p-value of Treatment Effect Estimate p-value of Treatment Effect Estimate 1 <.0001 .0003 1.1 .0003 .0020 1.2 .0023 .0079 1.3 .0103 .0237 1.4 .0336 .0566 1.5 .0836 .1125 1.6 .1682 .1929 1.7 .2854 .2939 1.8 .4225 .4068 1.9 .4992 .5216 2 .3687 .4437 2.1 .2571 .3436 2.2 .1699 .2573 2.3 .1068 .1868 2.4 .0642 .1319 2.5 .0370 .0907 2.6 .0206 .0610 2.7 .0111 .0401 2.8 .0058 .0259 2.9 .0029 .0164 3 .0015 .0103 3.1 .0007 .0063 3.2 .0003 .0401 3.3 .0002 .0023 3.4 .0001 .0014 3.5 <.0001 .0008 3.6 <.0001 .0004 3.7 <.0001 .0002 3.8 <.0001 .0002 3.9 <.0001 .0001 4 <.0001 .0001 Source: Analyses of the 2003 NSF NSRCG, using propensity scores generated from a multinomial probit model and weighted data (WTSURVY). 193 Summary of Key Findings The key findings of this study are summarized below. Institutional Pathways to STEM for Latinos • Latino STEM bachelor’s degree holders who earned an associate degree prior to transferring to a four-year institution graduated from less selective, publicly-controlled, non-research Hispanic-serving colleges and universities more commonly than non- associate degree earners. • Clear state-determined patterns existed in the pathways followed by Latino STEM bachelor’s degree holders, with community colleges playing a large role in Florida, California and New York, and HSIs serving as a large pathway for Latinos in Texas, Florida, and California. • National origin was not a significant factor by which Latinos’ institutional pathways varied. • Latino STEM bachelor’s degree holders who earned an associate degree were more commonly non-traditionally aged and first-generation college students than non-associate degree earners. • Nearly half of Latino STEM B.S. holders who traversed the community college/HSI or HSI pathways were first-generation students, compared to 30% of those in the non-HSI pathway. • Latinos who graduated from Hispanic-serving Institutions majored in non-social science STEM fields, i.e., Computer Science/Mathematics; Biological, agricultural, and 194 environmental sciences; Physical Science; and Engineering, in higher proportions than those who graduated from non-HSIs. College Financing Strategies of Latino STEM Bachelor’s Degree Holders • Significant differences in borrowing exist by institutional pathway: More than half (52%) of Latino STEM B.S. holders who followed the CC/HSI pathway did not borrow to finance college compared to 28%, 24%, and 26% of Latinos in the CC/non-HSI, HSI, and non-HSI pathways, respectively. Latino STEM bachelor’s degree holders in the CC/non-HSI pathway most commonly accumulated a high relative level of debt compared to students in the other pathways. Latinos in the HSI and non-HSI pathways most commonly accumulated low relative levels of debt (53% and 55%, respectively). • Three major college financing strategies were employed by Latino STEM B.S. degree holders: ‘self-support,’ ‘parental support,’ and ‘balanced support.’ o ‘Self-supporting’ Latino STEM B.S. degree holders tended to be non- traditionally aged, first-generation college students, and constituted nearly half of all Latinos who followed the CC/HSI institutional pathway. o ‘Parentally supported’ Latino STEM B.S. degree holders tended to be traditionally aged, had at least one parent with a four-year college degree (bachelor’s or higher), and did not earn an associate degree. Non-HSIs served as the largest pathway for ‘parentally supported’ students. o Latino STEM B.S. degree holders who received ‘balanced support’ were more likely to be traditionally aged students, had parents with a range of education levels, and were slightly overrepresented in the HSI pathway. 195 • Mexican American/Chicano, Puerto Rican, Cuban and Latinos of ‘other’ national origin borrowed at similar rates and employed similar financing strategies. The Effect of Debt on Graduate School Enrollment • High and low levels of relative debt negatively affected the likelihood of graduate school enrollment among Latino STEM bachelor’s degree holders. However, the size of the negative effect of high debt was larger than that of low debt. • If Latino STEM bachelor’s degree holders with low probabilities of incurring student loan debt had borrowed, they would have experienced a greater negative effect of borrowing on their chances of graduate school enrollment than those who did borrow. • Logistic regression techniques underestimated the negative effects of borrowing on graduate school enrollment among Latino STEM B.S. degree holders. 196 CHAPTER FIVE DISCUSSION In this chapter I summarize and discuss the key findings of the study, describe the study’s limitations and discuss directions for future research. The Importance of Community College Transfer and Hispanic Serving Institutions to Latino STEM Baccalaureates Much of the literature on the underrepresentation of Latinos in STEM fields neglects to consider the ways in which these students typically access postsecondary education, i.e., via community colleges and Hispanic-serving institutions. Three institutional pathways -- (1) community college to four-year HSI; (2) community college to four-year non-HSI; and (3) entry to four year HSI -- were of particular interest to this study given that they are seen as ‘unconventional’ routes to STEM, despite the frequency with which all Latinos employ these pathways. This investigation revealed that community colleges and four-year HSIs do in fact serve as institutional pathways for Latino STEM bachelor’s degree holders. Nearly one-fifth (19.8%) of Latino STEM bachelor’s degree holders earn associate degrees 19 at community colleges prior to completing the baccalaureate, and about 20% of Latino STEM bachelor’s degree holders earn the B.S. degree from HSIs. However, the proportions of Latinos who use the community college and HSIs as a pathway to the STEM B.S. degree is considerably lower than that of all Latinos. As stated in the second chapter of this work, more than half of all Latinos begin their postsecondary education in the community college, and HSIs are responsible for awarding nearly 40% of all bachelor’s degrees conferred to Latinos. 19 This figure does not include the 42% of Latinos in the sample who attended community college at some point, some of whom may have earned a year or more of credit. 197 The findings of the present study suggest that these trends are not mirrored in the population of Latino STEM B.S. degree holders. Instead, community colleges and HSIs, while playing a role, do not seem to act as large as a pathway for Latino STEM degree holders as these institutions do for all Latino students. Recognizing that my study employs a particularly narrow definition of community college attendance due to limitations in the NSRCG data source, even employing the ‘looser’ definition of attendance as enrollment in a community college at any point for any reason, Latino STEM bachelor’s degree holders attend community college in lower proportions than all Latino undergraduates (62.0% versus 79.3%) (NCES, 2006). It seems as if, however common among all Latino students, institutional pathways through community colleges and Hispanic Serving Institutions remain ‘unconventional’ in the sense that the patterns of access among Latino STEM bachelor’s degree holders differ from those of all Latino students. My analysis also focused on the ways in which the institutional pathways used by Latino STEM bachelor’s degree holders varied based on contextual factors identified in the conceptual framework of this study (Perna, 2006a). Beginning with demographic and higher education policy environments, the highest levels of context identified in the conceptual framework of this study, clear trends exist. Hispanic-serving institutions played the largest role in Texas, Florida and California – the three U.S. states with the largest Latino populations (U.S. Census Bureau, 2006). Forty percent of Latino STEM B.S. degree holders who attended college in Texas received their degrees from HSIs, compared to 32% in Florida and about 24% in California. The ‘Hispanic- Serving’ designation is demographically driven, thus, the relationship between the size and proportion of a state’s Latino population and the percentage of Latino STEM bachelor’s degree holders using HSIs as a pathway is not surprising. Further contributing to the relationship between demographic context and the size of the ‘Hispanic-Serving’ institutional pathway is the large degree of residential segregation that occurs in these states (Olivas, 2005; Tienda & Nu, 198 2006). Previous research has illustrated that geographic proximity is an important concern among Latino students and their families (Tornatzky et al., 2003) thus one would expect that proximity preferences and residential segregation intersect to contribute to the concentration of Latinos in HSIs (Olivas, 2005; Tienda & Nu, 2006). The findings from this study support this notion as Latino-intensive states generally had larger HSI pathways (CC/HSI or HSI) than those states with smaller Latino populations. Factors within states’ higher education policy context also shaped Latinos’ patterns of access with regard to the role of the community college in serving as an institutional pathway to STEM. In the current study I was interested in the extent to which Latino STEM bachelor’s degree holders earned associate degrees at community colleges prior to attending their baccalaureate-granting institution. Clear state patterns emerged from my analysis of the data. Half of Latino STEM B.S. degree holders who graduated from institutions in Florida earned an associate degree. The numbers were lower in California and New York; in California, 22% of Latino STEM bachelor’s degree holders earned an AA/AS degree and 28% earned AA/AS degrees in New York. In Texas, the numbers were considerably lower; just 9.5% of Latino STEM bachelor’s degree holders earned associate degrees from a community college. These findings are indicative of the differences in these states’ higher education system structures and policies. As noted in my review of the literature, Florida, California, and New York have highly articulated, stratified systems of postsecondary education in which community colleges are intended to act as a means to access four-year institutions. Through systemwide articulation agreements, Florida, California and New York facilitate the transfer of credits to the four-year institution for those students who attend community college. California has policies in place that guarantee transfer admission to a public four-year institution provided that students successfully complete the necessary requirements. Florida’s 2+2 higher education system goes 199 steps further than both California and New York by providing incentives for students to earn the associate degree prior to transfer in the form of guaranteed admission for only AA/AS degree holders. The large number of Latino STEM bachelor’s degree holders who earn associate degrees at community colleges in Florida, California, and New York were able to successfully navigate the path from the community college to four-year institutions by using the policies and structures in place. The considerable difference in the proportion of Latino STEM bachelor’s degree holders who earn associate degrees in Florida, California and New York likely results from Florida’s emphasis on associate degree attainment prior to transfer. In contrast, Texas’ postsecondary landscape consists of loosely coupled state and regional university systems, independent colleges and universities, and community colleges. Though many Texas community colleges have established articulation agreements with a number of four-year institutions, the state’s higher education governing body only recently began to facilitate the transfer process through formal structures (Wellman, 2002). Thus, Texas’ nascent transfer policies do not go nearly as far as those of Florida, California, and New York in terms of paving the way from the community college to four-year institutions for Latinos in the sciences. This is certainly borne out by the relatively small percentage of Latino STEM bachelor’s degree holders who graduate from institutions in Texas and earn an associate degree. I was able to identify key individual characteristics along which the institutional pathways of Latino STEM bachelor’s degree holders differed. In particular, I wanted to understand the characteristics of Latino students who traversed community college and HSI pathways to the STEM baccalaureate. The findings from the present study reveal that these ‘unconventional’ pathways to the STEM bachelor’s degree were primarily used by nontraditionally aged and first generation Latino students. Sixty-three percent of Latino STEM bachelor’s degree holders who earned an associate degree were non-traditionally aged, i.e., they 200 were 25 years or older at the time of the awarding of the baccalaureate. This overrepresentation is quite striking considering that 30.7% of all Latino STEM bachelor’s degree holders were non- traditional students. This finding mirrors the patterns observed regarding all Latino students, wherein non-traditionally aged students are more likely to attend community college. Parental education was also an important contextual factor associated with Latino STEM bachelor’s degree holders’ institutional pathways. Nearly 44% of Latino STEM bachelor’s degree holders who earned an associate degree were first generation college students, compared to 34% of those STEM B.S. degree holders who did not earn an associate degree. Just below three quarters of Latino STEM B.S. holders who earned associate degree had parents who did not earn a bachelor’s degree versus 60% of non-associate degree holders. This indicates that students whose parents were less familiar with postsecondary education more commonly used the community college as a pathway to the STEM baccalaureate than those students whose parents earned at least a bachelor’s degree. Though not a perfect proxy variable, parental education levels are commonly associated with socioeconomic status and access to various types of capital (e.g., economic, social, and cultural). The findings of this study suggest that Latino STEM bachelor’s degree holders that use community college pathways tend to be more disadvantaged. Previous research on educational opportunity indicates that accumulated disadvantage acts to limit the ways in which Latinos access college, resulting in the highly stratified enrollment patterns of these students. The findings from this study suggest that these patterns also apply to Latinos who earn degrees in science and related fields. While relatively high proportions of all Latino STEM bachelor’s degree holders are non-traditionally aged and first generation students, those who earn associate degree at a community college prior to attending a four-year institution are more likely to possess these characteristics. 201 Compounding the initial disadvantages that Latino students who attend community colleges possess, previous research tells us that community college attendance further disadvantages students in terms of a variety of educational outcomes (Adelman, 2005) though there is some debate as to whether this applies to Latino students (Melguizo, in press). From a sociocultural theoretical perspective, one could suggest that because community colleges are perceived to be lower status institutions, community college attendance limits the acquisition of various forms of capital, thereby furthering the accumulation of disadvantage. In this study, I sought to understand whether there were any differences in the characteristics of Latino STEM B.S. degree holders’ baccalaureate-granting institution by associate degree attainment. My findings reveal that Latino STEM bachelor’s degree holders who earned associate degrees went on to attend less selective, public non-research institutions. Thirteen percent of associate degree earners graduate from institutions classified as highly competitive, highly competitive plus, or most competitive, compared to 35% of non-associate degree earners. Seventeen percent of associate degree earners attended private four-year institutions, while 37% of non-associate degree earners did. Just 25% of Latino STEM bachelor’s degree holders who earned an associate degree graduated from research universities, compared with nearly 44% of non-associate degree earners. Interestingly, associate degree earners graduated from four-year HSIs in greater proportions than non-associate degree earners (32.1% versus 16.8%). It is possible that unobserved individual characteristics account for the differences in baccalaureate-granting institutional characteristics between Latino STEM bachelor’s degree holders who earned an associate and those that did not. However, clear trends exist and it is important to examine the implications of these trends. Many educational researchers (Bensimon, 2004; Fry, 2004) have argued for the importance of gaining access to selective research institutions among Latinos and other students 202 of color. Various studies (Brewer, Eide & Ehrenberg, 1999; Dale & Kruger, 2002; Eide, Brewer, & Ehrenberg, 1998; Monks, 2000) have shown that there are many educational and economic benefits to attending selective institutions for low-income and minority students, including increased chances of enrolling in graduate school and higher earning potential. Further, attendance at selective research universities is associated with increased opportunity to engage in scholarly research as an undergraduate, which increases retention, faculty interaction, and graduate school aspirations among underrepresented students in the sciences (Foertsch, Alexander, & Penberthy, 1997; Kardash, 2000; Kinkead, 2003; Sabatini, 1997; Seymour, Hunter, Laursen & Deantoni, 2004). The fact that Latino STEM bachelor’s degree holders who earn an associate degree at a community college are less likely to graduate from highly selective research universities than non-associate degree earners suggests Latinos who followed the community college pathway may be missing out on opportunities to engage in research and the associated benefits. Latino STEM bachelor’s degree holders who earn the AA/AS are also more likely to graduate from Hispanic-Serving Institutions, which tend to be less selective non-research institutions. The arguments outlined above suggest that graduating from an HSI would be disadvantageous because it could deny these students those benefits of attendance at selective institutions with more research opportunities. However, very little empirical evidence exists regarding the benefits or detriments of HSI attendance (Contreras, Malcom & Bensimon, 2008). While other minority serving institutions have been shown to be beneficial to their target populations, HSIs are not analogous to HBCUs and women’s colleges due the unique, demographically-driven way in which they earn the ‘Hispanic-serving’ designation. In order to understand whether HSI attendance is associated with Latino STEM bachelor’s degree holders’ outcomes, I examined differences in the field of study by institutional 203 pathway. My findings reveal that the field in which Latino STEM B.S. holders earned their degrees varies based on the pathway that they followed, without controlling for other factors. Latino STEM bachelor’s degree holders who followed the HSI or community college to HSI pathways earned their degrees in the so-called “hard” sciences in greater proportions than those Latinos who graduated from non-HSIs. Nearly 65% of Latino STEM bachelor’s degree holders who followed the HSI pathway earned their degrees in “hard sciences,” i.e., computer science/mathematics; biological, agricultural, and environmental science; physical science; engineering; and health/medical science. Just 41% of Latinos in the non-HSI pathway earned their degrees in the “hard sciences.” Among associate degree earners, the proportion of Latinos who earned the STEM bachelor’s degree in “hard” sciences was lower than that of non-associate degree earners, but the trend by HSI status remained. Nearly 40% of Latino STEM bachelor’s degree holders who traversed the community college to HSI pathway earned their degree in a ‘hard science,’ compared to just 30% of those in the community college to non-HSI pathway. Though the patterns of degree attainment in the various STEM fields discussed above reflect mere associations, not causal relationships, the findings suggest that there might be something unique about the institutional environments of HSIs that could lead Latinos to pursue and succeed in non-social sciences. Previous research on other minority serving institutions demonstrates that women’s colleges and HBCUs are more effective at facilitating the success of women and African Americans in the sciences, granting a disproportionately high share of STEM degrees to their target populations. These previous studies have established clear linkages between characteristics of the institutional environments and student success in STEM. The findings of the current study suggest that, in this respect, HSIs could possibly be comparable to other minority serving institutions. However, this conclusion requires further investigation using quantitative and qualitative methods. 204 Additional work ought to be done to establish if these patterns remain after controlling for student characteristics and other relative institutional characteristics. It is not clear from the methods used in the current study whether the association between HSI attendance and degree attainment in a “hard” science is indicative of an underlying causal relationship or if there is some sort of self-selection process occurring. Do HSIs award more degrees to Latinos in the “hard” sciences due to characteristics of their institutional environments, or are there unobserved factors that lead those Latino students more apt to pursue and succeed in non-social sciences to self- select themselves into HSI pathways? Though the answer remains unclear, the findings of the current study suggest that institutional environment is an important contextual factor to consider when understanding Latino student outcomes and attainment in particular STEM fields of study. The final key finding regarding the institutional pathways of Latino STEM bachelor’s degree holders pertains to the similarities in patterns of access across national origin. Mexican American, Puerto Rican, Cuban and other Latino STEM bachelor’s degree holders earned associate degrees at community colleges and graduated from Hispanic Serving Institutions at statistically similar rates. It is possible that the absence of statistically significant differences in the patterns of access among Latino STEM bachelor’s degree holders of various national origins was due in part to each group’s small sample size and the resulting larger standard errors. Though national origin did not seem to be a vital factor in this exploration of institutional pathways, Latinos ought to continue to be studied in a way that accounts for the diversity of this ethnic group. Questions of intra-group variability should remain central to subsequent studies of Latino STEM bachelor’s degree holders as national origin remains an important contextual factor capable of shaping educational opportunity and outcomes. 205 College Financing Strategies and Borrowing among Latino STEM Bachelor’s Degree Holders The second research question in this study centered on the college financing strategies of Latino STEM bachelor’s degree holders and their patterns of borrowing. Latino STEM bachelor’s degree holders employed one of three strategies to finance college: (1) self-support; (2) parental support; and (3) balanced support. ‘Self-supporters’ relied on loans, earnings, employer support and scholarships/grants to pay for college and received little to no parental support. Students who employed the ‘parental support’ strategy primarily relied on money from their parents or other relatives to help pay for college and used loans, earnings, and scholarships/grants as a secondary funding source. The ‘balanced support’ strategy involved paying for college through multiple forms of financial aid including loans, work study, scholarships/grants, and money from parents or other relatives. The findings of this study revealed that certain financing strategies were more common among Latino STEM bachelor’s degree holders with specific characteristics and within certain institutional pathways. The majority of self-supporting Latino STEM bachelor’s degree holders were non-traditionally aged and first-generation college students. Self-supporting students were also overrepresented in the community college (CC/HSI and CC/non-HSI) institutional pathways and at public institutions and were underrepresented at research universities. ‘Parentally supported’ Latino STEM bachelor’s degree holders, on the other hand were overwhelmingly traditionally-aged and had college-educated parents. These Latino STEM B.S. degree holders were overrepresented in the non-HSI institutional pathway, slightly underrepresented in the HSI pathway, and severely underrepresented in the two institutional pathways involving community colleges (CC/HSI and CC/non-HSI). ‘Parentally supported’ Latino STEM B.S. degree holders were less likely to graduate from privately controlled institutions and were more likely to 206 graduate from research universities. Latino STEM bachelor’s degree holders who employed the ‘balanced support’ college financing strategy generally were reflective of the entire population of Latino STEM degree earners. Though traditionally-aged students were overrepresented among ‘balanced supporters,’ the extent of this overrepresentation was extremely small. The distribution of parental education levels of Latino STEM bachelor’s degree holders in the ‘balanced support category nearly mirrored that of all Latino STEM B.S. degree holders. Those students who employed a ‘balanced support’ financing strategy were overrepresented in the HSI and non-HSI institutional pathways, and were underrepresented among associate degree earners. Latino STEM bachelor’s degree holders who received ‘balanced support’ were overrepresented in private institutions and research universities. Previous research (DesJardins, Ahlburg, & McCall, 2002; Paulsen & St. John, 2002) has illustrated that decisions regarding institutional pathways and college financing occur simultaneously, making it difficult to fully understand the relationships between these two decision-making processes. Further complicating matters is the fact that students often are required to reconsider their financing decisions multiple times, as the composition of their financial aid packages change from year to year and college costs rise. Though the findings of this study show clear associations between Latino STEM bachelor’s degree holders’ individual characteristics and their financing strategies, the data and methods employed in this study were not able to disentangle the relationships between college financing strategies and institutional pathways. For example, though the findings illustrate that ‘self-supporting’ Latino STEM baccalaureates were overrepresented among associate degree earners, it is not clear whether ‘self supporting’ students were able to execute such a financing strategy due to the cost savings resulting in community college attendance, or if they selected a community college pathway in 207 order to stick to their preconceived ‘self support’ financing strategy. Similarly, while the findings illustrate that ‘self supporting’ students were overrepresented in lower-priced HSIs, I am unable to discern whether these students chose to attend HSIs so that they could self-finance college or if they were able to self-finance because, on average, HSIs cost less to attend than non-HSIs. The findings also reveal associations between the ‘parental support’ and ‘balanced support’ strategies and institutional characteristics. However, due to the cross-sectional nature of the data and limitations in the methods I employed, these relationships cannot be fully specified. Many researchers (DesJardins et al., 2002; Paulsen & St. John, 2002; Perna, 2000) interested in the relationships between institutional context and college financing decisions have noted the difficulty of addressing these questions. Despite these challenges, the findings of the current study illustrate interesting trends and provide several markers for further inquiry using qualitative methods and other quantitative techniques with longitudinal data. The second research question also centered on borrowing and the resulting relative debt levels among Latino STEM bachelor’s degree holders. To address this question, I conceptualized debt as a relative construct such that the definition of ‘low debt’ and ‘high debt’ were dependent upon average debt levels at the baccalaureate-granting institution. As previous studies illustrate, decisions regarding college financing and borrowing are not rational processes. Debt tolerance and borrowing behavior are shaped by many contextual factors, and perceptions of debt vary among students based on their context (Archer & Hutchings, 2000; Christie & Munro, 2003; McDonough, 2004; McDonough & Calderone, 2006; Trent, Lee, & Owens-Nicholson, 2006). Though the literature provides an unclear picture of borrowing among Latino college students, research illustrates that the decision to borrow and tolerance for debt are highly subjective. For these reasons, I defined debt relative to the average borrowing at the baccalaureate-granting institution and relative to the institutional pathway taken, i.e., time spent in a community college. 208 Thus, the ‘high debt’ and ‘low debt’ categories as defined in this study can be considered more reflective of a student’s willingness to borrow or debt tolerance within their institutional environment and eliminates the differences in debt level that may be attributable to cost differences between institutions. Interestingly, the findings of this study reveal significant differences in the relative level of indebtedness among Latino STEM bachelor’s degree holders by institutional pathway that were not attributable to the cost differences of these four pathways. Latino STEM bachelor’s degree holders who earned an associate degree and then graduate from an HSI were considerably more likely to graduate with no debt than the average Latino STEM B.S. holder. However, associate degree earners who later graduated from non-HSIs accumulated high relative debt levels more commonly than students in any other institutional pathway (CC/HSI, HSI, or non-HSI). Latino STEM bachelor’s degree holders who graduated from an HSI without earning an associate degree were slightly overrepresented among Latinos with ‘low debt’ and ‘high debt.’ Non- associate degree earners who graduated from non-HSIs were overrepresented among those Latino STEM bachelor’s degree holders with ‘low debt.’ Due to the relative definition of debt employed in this study, these differences in the level of accumulated loan debt are not attributable to the differences in cost of the four institutional pathways. Instead, institutional pathway may be confounding the effects of student-level characteristics such as socioeconomic status (as measured by parental education) on debt tolerance and willingness to borrow. The findings of this study illustrate that first generation college students were overrepresented in the ‘high debt’ category, indicating that they borrowed large amounts relative to the average student at their baccalaureate-granting institution. Further, students whose parents were high school graduates were also overrepresented in the ‘low debt’ category. These first-generation students were also more likely to use the CC/HSI, CC/non-HSI 209 and HSI pathways. Thus, the overrepresentation of first-generation students in the community college and HSI pathways likely contributes to the aforementioned differences in debt level by institutional pathway. The descriptive techniques used in this study cannot unpack the relationship between institutional pathway and students’ levels of indebtedness; however, the constrained multinomial probit model of indebtedness employed in the propensity score matching analysis indicates that parental education level has a significant effect on the level of borrowing among Latino STEM bachelor’s degree holders, with students who have at least one parent with a graduate degree less likely to acquire debt. The findings of this study also indicate that certain contextual factors are not associated with the college financing strategies and borrowing behaviors of Latino STEM bachelor’s degree holders. Notable factors along which differences in college financing strategy and borrowing did not exist include the state in which a student graduated and Latino national origin. Instead, factors in Latino STEM bachelor’s degree holders’ immediate context such as parental education, and individual characteristics such as non-traditional student status, and associate degree attainment were associated with borrowing behaviors. Though it appeared that institutional pathway was also strongly associated with college financing strategy, the exact nature of this relationship was not discernable from the results of the study. Do students select certain institutional pathways that will allow them to carry out their preconceived college financing strategies, or do students develop college financing strategies based on their institutional contexts? The answer to this question remains unclear. Additionally, the findings of this study reveal that institutional context is associated with certain patterns of borrowing among Latino STEM bachelor’s degree holders; however a student’s institutional pathway may confound the effects of individual characteristics such as socioeconomic status and age on borrowing behaviors. Additional variables not available for this study including psychological and cognitive factors that have been shown previously to 210 affect willingness to borrow would likely help to disentangle the relationship between institutional pathway and college financing. The findings of this study illustrate that Latino STEM bachelor’s degree holders incorporate student loans into their college financing strategies. The findings further show that though some previous research characterizes Latinos as debt averse (Burdman, 2005; De La Rosa & Hernandez-Gravelle, 2007; Monaghan, 2001), the vast majority of Latino STEM bachelor’s degree holders borrow as a means for financing higher education. Additionally, those Latino STEM bachelor’s degree holders of lower SES backgrounds (as measured by parental education) borrow at similar or higher rates than Latino STEM bachelor’s degree holders from higher SES backgrounds. Even more surprising, Latino STEM bachelor’s degree holders who are first generation students graduated with high levels of indebtedness relative to the average cumulative debt at the B.S.-granting institution more often than Latino STEM bachelor’s degree holders whose parents held at least a four-year college degree. The findings of the current study reveal the complexity of borrowing among Latino STEM bachelor’s degree holders and underscore the need to critically examine the broad-brush characterizations of Latinos as debt averse previously offered in the literature. The Effect of Indebtedness on Graduate School Enrollment The final question in this study centered on the effect of indebtedness on graduate school enrollment among Latino STEM bachelor’s degree holders. This question was pursued using two quantitative techniques: logistic regression and propensity score matching. Results from the propensity score matching analysis indicated that low and high levels of relative indebtedness had a negative effect on the probability of graduate school enrollment among Latino STEM bachelor’s degree holders. Latino STEM bachelor’s degree holders who accumulated low levels of relative debt were, on average, 14% less likely to pursue graduate study than those who 211 finished the STEM bachelor’s degree with no debt. Interestingly, if those Latino STEM bachelor’s degree holders who did not borrow at all had borrowed and accumulated a low relative debt level, their likelihood of enrolling in graduate school would have dropped by nearly 16%. This indicates that the negative effect on the likelihood of enrolling in graduate school associated with borrowing at a low level is larger among Latino STEM bachelor’s degree holders with a low propensity to borrow. High levels of relative indebtedness had a larger negative effect on the likelihood of graduate school enrollment among Latino STEM bachelor’s degree holders than low levels of debt. On average, high debt reduced the probability of enrolling in graduate school by nearly 19%. The findings from the propensity score matching analysis indicate that Latino STEM bachelor’s degree holders with a higher propensity to accumulate a high relative debt level upon completing the B.S. degree experienced a smaller penalty on the probability of graduate school enrollment than those students with a lower propensity to accumulate high debt. Further, if a Latino STEM bachelor’s degree holder who did not borrow had accumulated a high level of relative debt, she would have experienced a larger negative effect on graduate school enrollment than her counterparts who did borrow at a high level. To summarize, the findings of this study indicate that Latino STEM bachelor’s degree holders who are willing to borrow and accumulate more debt than their peers within their institutional environment are less likely to attend graduate school within two years of earning the bachelor’s degree, while those students who are unwilling to borrow and do not accumulate debt are more likely to attend graduate school within this timeframe. Results from this study agree with some of the previous research on the direction of the effect of debt on graduate school enrollment, but also demonstrate that previous research has underestimated the size of the 212 negative effects of debt. There may be several reasons for the negative relationship between debt and graduate school enrollment. Students with debt may wish to pay down that debt as early as possible, and could feel that the opportunity costs in foregone earnings of attending graduate school are too large to bear. Multiple studies have illustrated that higher levels of educational debt are associated with higher initial wages (Minicozzi, 2005; Fosnacht & Santos, 2008), though over time borrowers experience lower growth rates in wages (Fosnacht & Santos, 2008). Students with high debt may wish to take advantage of this initial earning power in the labor market to pay down what they perceive to be high levels of debt. Perhaps these students are aware that their educational debt burden may force them to delay future borrowing or face higher interest rates (Flyer, 1997; Orazem & Mattila, 1991) due to the effects of this debt on credit worthiness. In the eyes of students with higher levels of debt, the long-term benefits of graduate school in terms of a steeper career trajectory and the accompanying increased earning power may be outweighed by these negative consequences of educational debt. The support for this interpretation of the relationship between debt and graduate school enrollment would be bolstered if it were possible to delineate between debts accrued through subsidized versus unsubsidized student loans. Because students are able to defer repayment on subsidized loans without penalty until after they complete graduate school, the effect of debt from subsidized student loans may be different. An alternative interpretation for the negative effect of low and high debt on the likelihood of graduate school enrollment is that borrowing behavior at the undergraduate level and resulting relative debt levels could be associated with long-term educational plans. Those students who borrow at higher levels may have initially planned to earn only a bachelor’s degree and thus, have borrowed heavily as undergraduates. Latino STEM bachelor’s degree holders who accumulate lower levels of debt or do not borrow at all may have planned to earn a graduate degree prior to 213 entering college or early on in their undergraduate tenure, and thus desired to spread out their borrowing across college and their anticipated post-baccalaureate careers. This borrowing strategy would result in lower borrowing as undergraduates. These competing hypotheses ought to be explored in future work. Despite my uncertainty in understanding the cause of the negative relationship between indebtedness and graduate school enrollment among Latino STEM bachelor’s degree holders, the findings do illustrate that both low and high levels of relative debt have larger negative effects on graduate school enrollment among individuals with lower propensities to borrow. This suggests that if these Latino students are in fact, averse to borrowing, this debt aversion has positive benefits in the sense that they avoid experiencing the negative effects of debt on graduate school enrollment. Much of the literature on educational borrowing frames debt aversion as a negative characteristic of students that constrains educational options. Researchers who have discussed debt aversion among Latino students attribute these behaviors to a lack of information about student loans or cultural barriers to accumulating debt. These researchers have pointed out several potential drawbacks to avoiding debt included being forced to attend lower cost institutions, working while in school, or even stopping out to save more money to pay for school. While an unwillingness to borrow to finance college may results in these negative effects, few researchers have explored the possible benefits of not borrowing. The findings of this study suggest that perhaps students who opt not to accumulate high relative debt levels unknowingly benefit in terms of graduate school enrollment. The final key finding of my study pertains to quantitative methods. I used both logistic regression and propensity score matching techniques to understand the relationship between debt and graduate school enrollment. Comparing the results of these two methods illustrated that logistic regression methods underestimated the size of the negative effects of debt on graduate 214 school enrollment. This finding underscores the need to address problems of self-selection when trying to assess the effects of various educational treatments on outcomes. As previous researchers have argued, experimental research designs tend not to be feasible in postsecondary education, leading researchers to rely heavily on ex-post facto data. However, because students are not randomly assigned to treatment conditions, particularly treatments related to financial aid, endogeneity remains a considerable problem in higher education research. This study suggests that propensity score matching can serve as an alternative strategy to standard regression techniques. Limitations of the Study This study has several limitations, the first of which is the primary data source, the National Survey of Recent College Graduates (NSRCG). The NSRCG is limited in the sense that there are key classes of variables that are likely to simultaneously impact the outcome variable of interest (i.e., graduate school enrollment) and the propensity score (i.e., predicted probability of participating in a loan program at a given level) that are not included in the data set. Some of these omitted variables include students’ aspirations, indicators of academic and social integration involvement, socioeconomic status, and high school grades and course taking patterns. Further, the variables related to financial aid usage are less than ideal. For example, the NSRCG does not provide aid amounts for non-loan financial aid. Instead, students simply indicate whether they received the specified forms of financial aid. Additionally, the NSRCG does not provide the students’ debt levels disaggregated by loan type (e.g., subsidized versus unsubsidized) which prohibited me from determining whether the effects of debt vary based on the type of loan. Finally, the NSRCG data only provides respondent’s indebtedness as a categorical variable. This means of operationalizing debt neglects its threshold nature; ideally, the relative level of indebtedness ought to be provided as a continuous variable. 215 A second limitation of the study pertains to my focus on associate degree earners. This approach, which was necessitated by the available variables in the NSRCG, certainly excluded those students who used the community college as a pathway to a STEM bachelor’s degree via transfer without earning an AA/AS, as many students do not earn an associate degree before transferring. Though this represents a limitation in that the population under study is a subset of all community college transfer students, the current higher education policy environment in which heavily incentivized or mandated associate degree attainment is being explored as a means to increase the efficiency of public higher education system makes this population particularly interesting. As state policy makers consider making the associate degree a necessary step for more students, it is important to understand how associate degree holders differ from those students who do not earn the AA/AS in terms of educational outcomes (e.g., baccalaureate- granting institutional characteristics, field of study, and patterns of participation in financial aid). There are also limitations to the methods I used in the current study. While PSM is capable of addressing the issue of self-selection bias, it is possible that unobserved characteristics or “unobservables” simultaneously impact the outcome variable of interest (i.e., graduate school enrollment) and the propensity score (i.e., predicted probability of participating in a loan program at a given level). Although all relevant observed variables were included in the analyses, I was unable to account for “unobservables”. As such, while PSM represents an alternative to regression methods, both techniques are subject to the effects of unobservables. An equally serious limitation was related to the small sample sizes of Latino STEM bachelor’s degree holders in the NSRCG sample. Higher education literature on financial aid and the borrowing of Latinos illustrates that there are significant differences among Latino students based on U.S. nativity and national origin (Tornatzky, Cutler, & Lee, 2002; Tomás Rivera Policy Institute, 2004; Santiago & Cunningham, 2005; Zarate & Pachon, 2006). In order to capture these 216 differences, I disaggregated the Latino students in the sample by national origin. However, this resulted in sub-samples too small to draw statistically significant conclusions. I was also forced to treat the Latino students in the NSRCG sample in the aggregate to compensate for small sample sizes while conducting the PSM analysis. Small sample sizes also prohibited me from conducting separate propensity score matching analyses on associate degree earners and non-associate degree earners within the sample of Latino STEM bachelor’s degree holders. Thus it is possible that the effects of debt on graduate school enrollment vary among Latino STEM bachelor’s degree holders who earned an AA/AS from a community college and those who did not earn an AA/AS degree. Directions for Future Research While the present study significantly contributes to our knowledge regarding the institutional pathways and college financing strategies of Latino STEM bachelor’s degree holders and the effect of debt on their graduate school enrollment, the findings raise many new questions that ought to be pursued in future research. These questions are discussed below. In what ways do those Latino STEM bachelor’s degree holders who do not earn an associate degree but still attend community college use the 2-year institutions? The primary data source for this study, the NSRCG, only indicates if respondents earn an associate degree and whether they attend community college at any point for any reason. Thus, I was unable to determine the ways in which Latino STEM bachelor’s degree holders who attended community college but did not earn an associate degree used the two-year institution. Do these students simply transfer from the community college without earning the associate degree? Or, do these Latino STEM bachelor’s degree holders take courses for credit ‘a la carte’ via reverse transfer or “swirling,” i.e., concurrent enrollment at two-year and four-year institutions (McCormick, 2003)? These two functions are very different, and it is important to characterize the role of community 217 colleges in the educational experiences of Latino STEM bachelor’s degree holders. Unfortunately, data available through the National Science Foundation, which compiles the most information on STEM degree holders, do not include these indicators. The questions raised by this study underscore the importance of collecting additional data on community college experiences of STEM degree holders. The study also points to the potential benefits of employing national and state-level data sources with detailed transcript data to examine the ways in which Latino STEM bachelor’s degree holders use the community colleges. Such data sources will allow researchers to further understand the function of community colleges in the education of Latino STEM bachelor’s degree holders. Why do Latinos who graduate from HSIs earn STEM bachelor’s degrees in the “hard” STEM disciplines more commonly than their counterparts at non-HSIs? Historically Black colleges and universities (HBCUs) and women’s colleges have been shown to increase STEM degree attainment among their target populations. The levels of success among HBCUs and women’s colleges have been attributed to their supportive institutional environments. However, Hispanic-serving institutions are unlike other MSIs in that they are designated as minority serving institutions based on enrollment rather than mission orientation. The findings of this study illustrate that Hispanic-serving institutions award higher proportions of STEM degrees in the “hard” sciences to Latinos than non-HSIs, which raises many questions about HSIs and their institutional environments. Do HSIs award more degrees in the “hard” sciences because they employ institutional policies and practices aimed at fulfilling the responsibilities implied by the Hispanic serving designation? Contreras, Malcom and Bensimon’s (2008) study of a ten HSIs revealed that these institutions only quietly acknowledge the ‘Hispanic-serving’ designation, if at all. Do the findings of this study reflect a change in institutional attitudes toward the HSI designation? Alternatively, Hispanic-serving institutions’ 218 success in awarded STEM bachelor’s degrees in the fields in which Latinos are most severely underrepresented may reflect the benefits of a “critical mass” of Latino students rather than purposeful action on the part of the institutions. An additional explanation could be related to self-selection processes by which Latino students who are more apt to earn a degree in a “hard” science field ‘self-select’ themselves into HSIs. These questions point to areas for future research. Do unsubsidized and subsidized student loan debts differentially affect graduate school enrollment among Latino STEM bachelor’s degree holders? While the findings of this study characterize the effect of relative debt on graduate school enrollment, the relationship between debt and graduate school enrollment warrants further study. I was unable to determine whether debt from unsubsidized loans affects graduate school enrollment differently from debt from subsidized loans. Future studies that explore this question using appropriate quantitative methods would increase our understanding of the effects of debt. Educational researchers interested in the effects of debt on educational outcomes would benefit from rich data sets containing multiple attitudinal, aspirational, and cognitive measures so that uncertainty in the estimated treatment effects introduced by selection on “unobservables” can be reduced. In addition, researchers can better understand the effect of indebtedness on graduate school enrollment by using methodologically sophisticated quantitative techniques capable of “adding a timing light” (DesJardins, McCall, Ahlburg & Moye, 2002, p. 555) to their analysis. Techniques such as structural equation modeling and latent growth analysis will enable researchers to understand how the accumulation of debt over time alters student aspirations regarding graduate school throughout the undergraduate experience. 219 CHAPTER SIX CONCLUSIONS AND IMPLICATIONS The purpose of this study was to characterize the institutional pathways traversed by Latino STEM bachelor’s degree holders, understand the college financing strategies employed by these students, and determine the effect of indebtedness on graduate school enrollment. In this final chapter of the present work, I offer interpretations of the key findings using the lens of cumulative (dis)advantage theory. I conclude by discussing implications for policy and practice. Revisiting Cumulative (Dis)advantage: A Tale of Three Students I began my dissertation by arguing that persistent race-based educational inequities, including the underrepresentation of Latinos in STEM fields, can be linked to the disadvantages Latinos have accumulated throughout previous generations of opportunity denial. I sought to uncover the ways in which these disadvantages and their associated context shape the patterns of institutional and financial aid access and outcomes of Latino STEM bachelor’s degree holders. This study revealed the complex ways in which context interacts with and shapes these students’ institutional pathways, college financing strategies, STEM degree outcomes and graduate school enrollment. The findings suggest that there are multiple unexpected means through which Latinos in STEM accumulate advantages and disadvantages within their pathways to STEM, often simultaneously. The study offers a counternarrative to existing educational literature by suggesting that many contexts commonly thought to be disadvantaging are capable of offering benefits to Latinos in STEM, provided that students’ initial disadvantages are mitigated. To better illustrate the ways in which Latino STEM bachelor’s degree holders simultaneously accumulate advantage and disadvantage in unexpected ways, I present fictional composites of three students, each traversing a different pathway to STEM. The composites are 220 informed by this study’s findings and reflect the common patterns of access and financial aid participation revealed by the present analyses. Following the three composites, I discuss the implications of the findings. Inez: The Story of a Non-traditionally Aged First Generation College Student Inez’s pathway to STEM. Immediately after Inez graduated from high school with a B grade point average, she found a job in order to help pay for her father’s medical bills. Inez turned her love for math into a profession and handled the bookkeeping at a small, neighborhood grocery store. After working at the store for a couple of years, Inez’s boss suggested that she think about going to college. Neither Inez’s father nor mother went to college, so she was not sure where to apply or how to pay for it. Upon soliciting advice from her friends and family, Inez decided to enroll in Los Angeles Community College (LACC), which was a 15 minute bus ride from her job. Inez continued to work during the day, while taking classes in the evenings. She squeezed in studying each morning before work and on weekends. After two and half years of taking classes, Inez earned her associate degree in mathematics. In her final semester, Inez attended LACC’s transfer fair to learn more about colleges and universities from which she could earn her bachelor’s degree. Inez wanted to major in math and applied to multiple California State University (Cal State) campuses and UCLA. Unfortunately, Inez was not admitted to UCLA because math was an “impacted” program. She decided to attend Cal State LA—her second choice institution and an HSI. Inez applied for financial aid and received a Pell Grant, but was faced with the dilemma of borrowing $1,000 dollars each year to help pay for her tuition, fees, and books. Her parents did not have the financial means to help with college costs and Inez did not feel comfortable borrowing, so she decided to continue working part-time at the store while attending Cal State 221 LA. She was able to further reduce her college costs by living at home and taking the bus to campus. During her first semester at Cal State LA, Inez excelled in her math courses and became involved in the Society of Hispanic Engineers (SHPE). After learning more about engineering through SHPE, Inez switched her major from mathematics to electrical engineering. Many of the students in SHPE were also electrical engineering majors and they formed study groups in which Inez participated. Her classmates were a vital source of support as she continued to make progress toward her bachelor’s degree. Inez’s SHPE faculty advisor suggested she think about going to graduate school and try to get some research experience. He explained to her that while the research opportunities for undergraduates were limited at Cal State LA, she should consider applying for an internship at a local company; he even volunteered to call some of his contacts in the engineering industry. Though the idea of an internship appealed to Inez, both of the companies that were interested in giving her a position were located in the Inland Empire—too far for Inez to travel and make it back to Los Angeles in time for her shift at work. Because the companies could not pay her, Inez ultimately lost out on these opportunities because she was unable to quit her job at the store. During her final semester as an undergraduate, Inez applied to the master’s of Engineering program at the University of Southern California (USC) and Cal State Long Beach. Unlike many of her peers, Inez had no student loan debt and thus, did not feel pressured to get a job right after graduation. She was almost certain she would not get into her first choice graduate program at USC because they placed a high value on research experience and she did not have any. In the end, Inez enrolled in the master’s program at Cal State Long Beach after being denied admission at USC. 222 Advantages and disadvantages in Inez’s story. Inez’s story, though fictional, illustrates the complex ways in which context shapes students’ accumulation of disadvantages and advantages along the pathway to STEM. Inez was a first-generation, non-traditionally aged college student. She delayed entry to postsecondary education, entering community college a few years after graduating from high school. Inez’s non-traditional status and her need to continue working while in school compounded her limited organizational habitus (McDonough, 1997), leading her to enroll in community college. Though there remains some uncertainty about the diversionary effects of community college attendance, particularly among Latinos (Melguizo, in press), previous literature demonstrates that community college reduces the likelihood of attaining a bachelor’s degree (Adelman, 2006; Pascarella & Terenzini, 2005). Thus, a potential disadvantage looms for Inez in the institutional context of the community college. In the above example, Inez is able to mitigate this disadvantage and successfully completes her associate degree in mathematics. However, she quickly encounters an additional disadvantage resulting from her community college attendance as she is unable to transfer to her first choice institution due to its limited capacity to admit transfer students to oversubscribed academic programs. In California, this is a common barrier faced by community college transfer students who wish to major in a STEM field at the most selective University of California institutions (University of California, n.d.). Thus, community college attendance causes Inez to accumulate an additional disadvantage in the form of constrained college options. Inez ends up attending Cal State LA, which is a Hispanic-serving, non-research university close to her home community. Within the institutional environment of Cal State LA, Inez simultaneously accumulates advantages and disadvantages along the pathway to her bachelor’s degree. The first advantage Inez accumulates pertains to college financing. The relatively low cost of attendance at Cal State 223 LA enabled Inez to execute a ‘self-support’ college financing strategy. Though her parents were unable to help her pay for college, Inez was able to live at home, lowering her cost of living. Her university’s close proximity to her neighborhood also allowed her to continue working her job at the neighborhood grocery store part-time. This income enabled Inez to avoid borrowing to finance college. As reflected by the findings of this study, Inez’s likelihood of attending graduate school was increased because she completed the bachelor’s degree without debt. Another key advantage provided to Inez by the environment of Cal State LA involved academic support. Cal State LA is a Hispanic-serving institution and had an active chapter of the Society of Hispanic Engineers (SHPE). Inez became involved in SHPE and was exposed to the field of electrical engineering. Although this was not her originally intended major, Inez’s participation in SHPE gave her access to a network of Latino students with which she could identify, socially and academically. Her involvement in the Latino professional organization also enabled Inez to interact with the SHPE faculty advisor, who provided Inez with advice about graduate school and the importance of undergraduate research. Inez’s academic support network likely facilitated her persistence in electrical engineering, an academically challenging “hard” science field in which Latinos are severely underrepresented. The institutional environment of Cal State LA also caused Inez to accumulate disadvantages. Because Cal State LA, like the majority of HSIs, is not a research institution, undergraduate research opportunities were extremely limited. Though her SHPE faculty advisor recognized the value of engaging in undergraduate research, the institution was unable to provide Inez with these opportunities. Undergraduate research has been illustrated to be positively associated with graduate school attendance, particularly among underrepresented minorities in STEM fields (Foertsch, Alexander, & Penberthy, 1997; Kardash, 2000; Kinkead, 2003; Sabatini, 1997). Though her SHPE faculty advisor encouraged Inez to look for internships off-campus 224 within the engineering industry, barriers in terms of location, transportation, and time prevented Inez from doing so. In this instance, Inez’s initial disadvantage in terms of socioeconomic class compounded a disadvantaging aspect of her institutional context (i.e., lack of undergraduate research opportunities). Inez enrolled in graduate school; however, she was denied admission to the master’s program at a highly selective private university because she did not have research experience. Thus, while the lack of research opportunities at her bachelor’s degree granting HSI did not prevent Inez from attending graduate school, it disadvantaged her in terms of the quality of her graduate institution. This may lead to future disadvantages in terms of earnings (Brewer et al., 1998; Dale & Kruger, 2002; Monks, 2000). In sum, Inez’s institutional pathway to her STEM bachelor’s degree and her college financing strategy resulted in the simultaneous accumulation of advantages and disadvantages. Her initial disadvantage due to her first-generation and non-traditional student status constrained her college options, resulting in her enrollment in community college. While the community college served as a low-cost pathway to a four-year institution via transfer (an advantage), the community college institutional context also disadvantaged Inez by limiting her ability to be admitted to a highly selective four-year institution. Inez’s enrollment in a local, four-year HSI resulted in simultaneous accumulation of advantage and disadvantage. The institutional context of the HSI advantaged Inez in three ways. First, due to the institution’s low college costs, Inez’s income from a part-time job was enough to cover her unmet financial need and helped her to avoid debt, which raised her likelihood of attending graduate school. Second, the institution’s proximity to her neighborhood allowed Inez to live at home with her parents, further lowering her college costs. Third, the institution’s environment provided Inez with sources of social and academic support within her major that facilitated her persistence, major retention, and degree 225 attainment. However, Inez’s attendance at the HSI also disadvantaged her by limiting her opportunities to engage in undergraduate research, thereby reducing her ability to be admitted to a selective graduate program and lowering her subsequent earning potential. Inez’s story reflects the complex ways in which a student’s initial disadvantages, institutional pathway, college financing strategy, and resulting student loan debt operate to simultaneously advantage and disadvantage that student. Alex: A Traditionally Aged, Second Generation College Student Alex’s story. Alex, a second generation college student decided to go to community college after being denied admission to USC, his first choice institution. He had heard from some of his classmates that if he joined the honors program at Santa Monica Community College and maintained a high GPA, he would have a better chance of gaining admission to USC as a transfer student. Alex attended community college for two years and completed all of the requirements to transfer to the neuroscience program. Only at the behest of his parents did Alex fill out the form to be granted his associate degree. Though Alex applied for many scholarships and received state and institutional grant aid, he needed to take out nearly $5,000 in loans each year to pay the high cost of attending USC. Alex’s parents encouraged him to borrow because they knew how important it was for Alex to attend USC and they viewed loans as a means for him to avoid working while in school. After arriving at USC, Alex had a difficult time connecting with the other students in his major. He was the only Latino in the majority of his courses, and was intimidated by his professors. Already self-conscious about being a transfer student, Alex did not ask questions in class, even though most of the material was going over his head. Alex’s time in his residence hall amongst his suitemates was his only respite from the isolation he felt in his classes. By the middle of the semester, Alex had stopped going to class. He subsequently failed the midterm in two of 226 his classes, and went on to drop them. Though Alex tried to refocus and worked hard to pass his three remaining neuroscience courses, he had already fallen too far behind. After his first semester, Alex was placed on academic probation. Upon his parents’ advice, Alex switched his major from neuroscience to psychology. Because he was missing some of the prerequisite courses for his new major, Alex spent a total of three years at USC. He graduated with about $17,000 in debt—more than most of the other community college transfer students he knew. Eager to pay off of his loans, Alex decided to get a job. He attended career fairs on-campus and landed several interviews during the spring semester of his final year at USC. Shortly after graduating, Alex accepted a job offer from an advertising agency owned by a USC alum. Advantage and disadvantage in Alex’s story. The fictional story of Alex illustrates the ways in which certain institutional pathways and college financing strategies can advantage and disadvantage Latino students in terms of STEM outcomes. Alex transferred from a community college to a non-Hispanic serving, highly selective research institution. He elected to attend Santa Monica College in a strategic move and had well-defined academic goals. Thus, Alex avoided many of the disadvantages reportedly associated with the community college institutional context. However after transferring, Alex quickly began to accumulate disadvantages from the institutional context of his non-Hispanic serving research university. The first disadvantage related to Alex’s field of major. Though Alex entered USC as a neuroscience major, he was unable to find sources of academic and social support within his department. As the only Latino and community college transfer student in his courses, Alex felt extremely isolated. These feelings led to poor academic performance, and Alex switched from the “hard” science field of neuroscience to psychology, a social science field. This is a disadvantage because the “hard” science disciplines in which Latinos are most severely underrepresented tend to be more high 227 demand and lead to higher earning potential (NSF, 2005). Further, Alex’s change of major required him to spend an additional year at USC, beyond what he planned initially. Alex’s second disadvantage pertained to college financing. Due to the high cost of attending USC, Alex took out several thousands of dollars in loans for each of the three years he spent at USC. This caused him to graduate with a significant amount of debt relative to other community college transfer students at the university. His relatively high level of debt deterred Alex from attending graduate school because he wanted to begin paying off his student loans as soon as possible. Attending a highly selective institution certainly helped Alex to accumulate advantages. He secured a well-paying job immediately after graduation, and was able to tap into the extensive, well-connected alumni network. Alex’s employers were certainly impressed with his college pedigree, which played a large part in his hiring. There were many other ways in which the institutional context of USC could have benefited Alex (e.g., undergraduate research, involvement in academic organizations); however, Alex’s initial academic difficulties prevented this from occurring. In sum, Alex garnered simultaneous advantages and disadvantages while earning a STEM bachelor’s degree. By attending community college, Alex was able to lower his initial college costs and gain admission to a highly selective institution via transfer. However, after enrolling at the non-Hispanic serving, highly selective research university, Alex began to quickly accumulate disadvantages. He did not connect with the students in his initial major, neuroscience, and developed feelings of isolation. He fell behind in his classes, and ended up switching to majors to psychology, a field not as “high demand” as neuroscience. Alex’s initial academic difficulties required him to remain at his four-year institution for an extra year and to borrow more than he initially anticipated. Alex graduated with a higher level of debt than his peers who 228 also transferred from community college, and decided against graduate school so that he could begin to pay down his loans. While the institution’s prestige advantaged Alex during his job search, his decision to forgo graduate school would act as disadvantage to him in the long run in terms of earning potential (Becker, 1993; Titus, 2007). Eva: A Traditionally Aged, Second Generation College Student Eva, the daughter of a nurse, graduated from a magnet high school for students interested in science. She applied to multiple universities, eventually settling on UCLA. Eva wished to major in biology, and eventually wanted to go to medical school. Eva applied for financial aid, but her package consisted primarily of loans due to her middle-class socioeconomic status. Realizing the amount of debt she would later incur in medical school, Eva’s mother discouraged her from borrowing heavily as an undergraduate. Her mother, who also had to care for Eva’s younger brother on a single income, took on extra shifts at the hospital so that she could keep Eva’s borrowing to a minimum. Eva ended up borrowing about $1,000 in her first year, and slightly more each subsequent year. In her sophomore year, Eva found a part-time job as an assistant in a genetics lab on campus to help pay college costs and to gain research experience. She felt it was important to build her resume in order to increase her chances of being accepted into medical school. Because Eva spent most of her time in the lab, in class, and in the library studying, her circle of friends was small, consisting of a few women from her residence hall. Eva graduated from UCLA with a cumulative debt level of about $5,000—an amount much lower than that of her peers. Eva was accepted to medical school during her final semester at UCLA and was chosen to participate in a special summer program for underrepresented minorities at the medical school. However Eva declined this invitation and deferred her entry until the fall semester so that she could work during the summer and pay off her loans. 229 Advantages and disadvantages in Eva’s story. This fictional story of a Latina, second- generation college student demonstrates the ways in which institutional pathways and college financing strategies that are considered to be advantaging can also result in the accumulation of disadvantage. Prior to college, Eva possessed initial advantages; she attended a magnet high school for science and her mother was employed in a science-related field. These advantages “smoothed” her path to attend a highly selective research university. Though Eva’s mother was not wealthy, her middle-class socioeconomic status enabled Eva to limit her borrowing and incur a low level of debt relative to other students at her institution. Eva benefited from her mother’s experiences within the medical profession as she advised Eva to limit borrowing as an undergraduate because of the future debt she would accumulate in medical school. The institutional context of the highly selective research university from which Eva graduated simultaneously advantaged and disadvantaged her. In terms of advantages, Eva was able to engage in undergraduate research in an area related to her major, biology. The wide availability of research opportunities is a characteristic unique to the environments of research universities (NSF, 1996; Strassburger, 1995). Additionally, the highly selective nature of her undergraduate institution acts as a form of capital, facilitating her admission to graduate school (Eide, Brewer & Ehrenberg, 1998). However, Eva was also disadvantaged by her institutional context. Due to the higher costs of attending highly selective, non-Hispanic serving institutions, Eva was required to borrow to finance her education, though she was able to keep the level of debt relatively low. As the findings of this study show, both relatively high and relatively low levels of debt disadvantage Latino STEM majors by lowering the probability of graduate school enrollment. In this particular vignette, Eva did enroll in medical school. However, her indebtedness caused her to defer her enrollment by one semester. 230 The stories of Inez, Alex and Eva illustrate the ways in which initial disadvantages shape Latinos’ patterns of institutional access and financial aid participation, and the complex manner in which the resulting institutional pathways and college financing strategies cause Latino STEM degree holders to simultaneously accumulate disadvantage and advantage. I turn to a broader discussion of these implications below. Complicating our Understanding of ‘Disadvantaging’ Contexts for Latinos in STEM Much of the higher education research on Latinos focuses on these students’ patterns of institutional access and achievement, and the barriers posed by college costs and access to financial aid. Conventional wisdom, borne out of empirical research, frames our understanding of the institutional and financial aid contexts that disadvantage Latino students. Such dominant narratives include: community college is seen as a less desirable entry point to higher education than direct enrollment at a four-year institution; attendance at a higher-status, selective research university is thought to be advantageous compared to enrollment in a lower-status, less selective, Hispanic serving institution; and, debt aversion, or an unwillingness to borrow to finance college, limits postsecondary educational access for Latino students. Previous research lends credence to these dominant narratives in the literature. However, the findings of this study necessitate that we complicate our understanding of the relationships between the aforementioned institutional and financial aid pathways and the accumulation of disadvantage among Latino STEM bachelor’s degree holders. The findings of this study reveal that institutional pathways and financing strategies conventionally thought to be ‘disadvantaging’ can simultaneously benefit Latinos in terms of STEM outcomes. As illustrated in the three preceding vignettes, these advantages pertained to Latinos’ field of study within STEM, levels of indebtedness, and graduate school enrollment. 231 Latinos are most severely underrepresented in the “hard science” fields, e.g., biology, physical science, and mathematics. The findings of this study illustrate that attending an HSI, even after earning an associate degree from a community college, is positively associated with majoring in a “hard science” field. In this sense, a Latino student who attends an HSI receives an advantage from that attendance. However, HSIs tend to be less selective, non-research institutions; a fact that may act as a detriment to Latino STEM majors because they have fewer opportunities to engage in undergraduate research which has been shown to be associated with graduate school enrollment (Foertsch, Alexander, & Penberthy, 1997; Kardash, 2000; Kinkead, 2003; Sabatini, 1997). Thus, the findings suggest that for a Latino student, the decision to attend an HSI simultaneously advantages and disadvantages that student in terms of STEM outcomes. A second example of simultaneous accumulation of advantage and disadvantage pertains to Latino STEM bachelor’s degree holders’ institutional pathways and college financing strategies. Community college attendance is thought to have a diversionary effect on educational aspirations and attainment, though these effects have recently been shown to be overestimated among Latino students (Melguizo, in press). In this sense, the institutional context of community colleges disadvantages students by reducing their chances of earning a bachelor’s degree. However, the findings of this study illustrate that those Latino STEM bachelor’s degree holders who earned an associate degree were more likely to enact a ‘self-support’ college financing strategy, resulting in lower levels of relative debt than non-associate degree holders. With the negative effects of debt on graduate school enrollment revealed in this study, the community college context can be beneficial to students by allowing them to implement a ‘self-support’ financing strategy and avoid incurring high levels of relative debt. The final example of the simultaneous accumulation of advantage and disadvantage can be drawn from the findings regarding college financing and the effect of debt on graduate school 232 enrollment. Previous literature illustrates that an unwillingness to borrow to finance college can result in constrained college choice and limit access to pricier, private institutions, and several researchers (Monaghan, 2001; De La Rosa & Hernandez-Gravelle, 2007) have attributed the smaller percentages of Latino college students who borrow to debt aversion. Though the present study reveals that Latino STEM bachelor’s degree holders do, in fact, borrow and that Latino STEM B.S. holders from lower SES backgrounds accumulate higher levels of relative debt more commonly than those of higher SES, it is true that Latino students, on average, borrow less in absolute dollars than students from other racial/ethnic groups. Lower debt tolerances act as a disadvantage in terms of access to certain institutions. However, the findings of this study illustrate that low and high levels of indebtedness, defined relative to the average borrowing at a student’s institution, negatively affect the likelihood of attending graduate school among Latino STEM bachelor’s degree holders. In other words, not borrowing is advantageous when considering the educational outcome of graduate study. The decision to borrow to finance college is another example of simultaneous advantage and disadvantage. Initially, a willingness to accrue debt advantages Latino students in that student loans widen the range of college options; but, the resulting debt later acts as a disadvantage for Latino STEM bachelor’s degree holders, reducing the likelihood of attending graduate school. The findings of this study reveal that the relationships between Latino STEM bachelor’s degree holders’ context, institutional pathways, and college financing strategies commonly result in the simultaneous accumulation of advantage and disadvantage. The present study complicates our understanding of cumulative (dis)advantage and disputes the direct linkages between initial disadvantage and future disadvantage. The figures below (Figures 6.1–6.3) depict the three vignettes from the beginning of this chapter and illustrate the multiple ways in which certain institutional pathways and college 233 financing strategies can be both beneficial and detrimental to Latinos in terms of STEM outcomes. In the figures, gray rectangles represent what are typically considered to be disadvantaging contexts, and white rectangles represent advantaging contexts. In the diagrams, a plus (+) sign indicates a positive association or increased probability of exhibiting the specified behavior or achieving the indicated educational outcome, and a minus (–) sign represents a negative association or decreased probability of exhibiting the behavior or achieving the outcome. Figure 6.1 illustrates the ways in which Inez’s initial disadvantages led to future disadvantages and at the same time, advantages. Inez, a first generation college student, is more likely to be non-traditionally aged and to attend community college—two disadvantages in terms of bachelor’s degree attainment. The findings from this study show that if a Latino student successfully overcomes those disadvantages and transfers with an associate degree as Inez did, she is more likely to end up attending an HSI. This attendance at an HSI results in the advantages of having lower college costs (Santiago & Cunningham, 2005) and as my findings show, accumulating less relative debt and earning the STEM bachelor’s degree in a “hard science” area. My findings also reveal that these advantages garnered by Latinos who follow community college to HSI pathway (i.e., lower relative debt and an increased likelihood of earning a STEM B.S. degree in a “hard” science discipline) positively affect graduate school enrollment, a desirable outcome. However, the literature also demonstrates that HSIs typically offer Latino STEM majors fewer opportunities to engage in undergraduate research—a disadvantage in terms of graduate school enrollment (Foertsch, Alexander, & Penberthy, 1997; Kardash, 2000; Kinkead, 2003; Sabatini, 1997; Seymour et al., 2004). 234 Figure 6.1. Inez’s Story: Accumulating Advantage and Disadvantage + Inez: Latina First- Generation College Student + Non-traditional Student Status Community College Attendance STEM Bachelor’s Degree Attainment Fewer Undergraduate Research Opportunities STEM Degree in “Hard” Science Discipline Low Levels of Absolute Debt ‘Self Support’ College Financing Strategy No Debt from Borrowing Graduate School Attendance Attendance at less selective HSIs Lower College Costs + + – + + + + + + + + + – – Legend Advantaging Context : positive association, or increased probability Disadvantaging Context : negative association, or decreased probability + – 235 Figure 6.2 illustrates the advantages and disadvantages accumulated by Alex along his pathway to STEM. Relative to Inez, Alex possessed initial advantages because his parents attended college. However, Alex entered a disadvantaging context by attending community college. Alex successfully transferred to a highly selective research university, which is conventionally considered to be an advantaging environment. However, the institutional context of the university acted as a source of disadvantage for Alex by preventing him from earning his bachelor’s degree in a “hard” STEM field. Alex’s ‘balanced support’ college financing strategy and the financial aid context of the research university led to further accumulation of disadvantage for Alex. He accrued a relatively high level of debt at the institution, which prevented him from enrolling in graduate school. 236 Figure 6.2. Alex’s Story: Accumulating Advantage and Disadvantage Alex: Latino Second-Generation College Student Community College Attendance STEM Bachelor’s Degree Attainment STEM Degree in Social Science Discipline High Levels of Absolute Debt ‘Balanced Support’ College Financing Strategy High Relative Debt Level from Borrowing Graduate School Attendance Attendance at Highly Selective Research University Higher College Costs + – – – + – – + + + + Legend Advantaging Context : positive association, or increased probability Disadvantaging Context : negative association, or decreased probability + – 237 Eva’s story is illustrated in Figure 6.3. Like Alex, Eva possessed initial advantages relative to Inez because her mother attended college and was employed in a science-related field. Further, Eva’s high school was endowed with a relatively high level of organizational habitus (McDonough, 1997). Eva’s initial advantages led her to enroll in a highly selective research university. Though my findings reveal that the institutional context of non-HSI research universities is negatively associated with degree attainment in a “hard” STEM field, Eva’s participation in undergraduate research allowed her to avoid accumulating this particular disadvantage. Her undergraduate research experiences acted as a further advantage, as participation in research is positively associated with graduate school attendance (Foertsch, Alexander, & Penberthy, 1997; Kardash, 2000; Kinkead, 2003; Sabatini, 1997; Seymour et al., 2004). Eva employed a ‘parental support’ college financing strategy, which allowed her to accumulate a relatively low level of debt due to borrowing. However, this debt still acted as a disadvantage by causing Eva to defer her enrollment in graduate school. 238 Figure 6.3. Eva’s Story: Accumulating Advantage and Disadvantage Eva: Latina Second-Generation College Student Undergraduate Research Opportunities STEM Degree in “Hard” Science Discipline High Levels of Absolute Debt ‘Parental Support’ College Financing Strategy Low Relative Debt Level from Borrowing Graduate School Attendance Attendance at Highly Selective Research University Higher College Costs + + + – + – + + + + + Legend Advantaging Context : positive association, or increased probability Disadvantaging Context : negative association, or decreased probability + – 239 The findings of this study reveal that there are a multitude of ways in which contextual factors shape Latino students’ decisions about institutional pathways and college financing strategies, which result in advantages or disadvantages in terms of educational outcomes. In order to achieve the desirable outcome of graduate school enrollment (and graduate degree attainment), a Latino student must successfully navigate the complex webs shown in the preceding figures (Figures 6.1–6.3), overcoming their accumulated disadvantage while amassing advantages as opportunities to do so present themselves. Institutional actors and policymakers can help Latino students to mitigate the accumulation of disadvantage through strategic interventions in the form of policies and practices. Moving Toward a General Model of STEM Postsecondary Outcomes for Latinos? The current study and the findings that have emerged underscore the difficulty in establishing a model of disadvantaging and advantaging contextual factors that affect Latino students’ outcomes related to STEM. As this study has illustrated factors within the layers of context in which Latino students are located can have simultaneous positive and negative effects. Through the three vignettes presented above I have argued that the “net” effect of those factors depends heavily on whether disadvantages can be mitigated through specific policies, practices, or even interactions (i.e., student-student, student-faculty, etc.) within those contextual layers. Because of the complexity uncovered in this study, I am hesitant to put forth a model of STEM outcomes for Latino students based on this dissertation alone. Rather, I offer the following table that summarizes the ways in which specific institutional pathways can advantage and/or disadvantage Latinos students who desire to pursue STEM fields. Each of these advantages and disadvantages are discussed in the three vignettes presented above, however the table provides a succinct summary of the benefits and detriments of the four identified institutional pathways to the STEM baccalaureate for Latinos. 240 Table 6.1 (Dis)advantaging Institutional Pathways for Latinos in STEM Institutional Pathway Advantaging Factors Disadvantaging Factors Associate Degree at CC to HSI HSI attendance is positively associated with degree attainment in “hard” (non-social) science Largest proportions of Latino STEM bachelor’s degree holders graduating with no debt follow this pathway, which helps them to avoid the negative effects of debt on graduate school enrollment Largest proportion of Latino STEM baccalaureates enacting a “self support” financing strategy follow this pathway Pathway is located within many Latino communities in urban areas and in Latino-rich states (i.e., FL, CA), making it more accessible Community college attendance may decrease chances of B.S. degree attainment Perceived to be less prestigious pathway which may have implications for graduate school admissions Fewer research opportunities available at community colleges and HSIs Associate Degree at CC to non-HSI Significant proportion of Latino STEM baccalaureates enacted a “self support” financing strategy in this pathway Largest proportion of Latino STEM bachelor’s degree holders majoring in social science fields followed this pathway Largest proportion of Latino STEM bachelor’s degree holders finishing with high relative debt burden were in this pathway, which negatively impacts graduate school enrollment Perceived to be less prestigious pathway which may have implications for graduate school admissions Fewer research opportunities available at community colleges 241 Table 6.1, Continued. Institutional Pathway Advantaging Factors Disadvantaging Factors HSI HSI attendance positively associated with degree attainment in “hard” (non-social) science Largest proportion of students enacting a “balanced support” strategy, i.e., receiving a wide-range of financial aid, followed this pathway Lowest cost option for Latinos entering four-year institutions directly leading to lower absolute debt levels (Santiago & Cunningham, 2005) Higher proportions of Latino students located within HSI institutional environments (by definition) may provide academic and social benefits to Latino STEM majors On average, HSIs have higher proportions of Latino faculty (Contreras, Malcom, & Bensimon, 2008), which may provide academic benefits to Latino STEM majors Perceived to be less prestigious pathway because HSIs are less selective, which may have implications for graduate school admissions Fewer research opportunities available at HSIs as they tend to be non-research institutions Non-HSI More research opportunities available at many non-HSIs (research and doctoral institutions) Tend to be more selective and offer more prestigious pathway to the STEM degree for Latinos Large proportion of Latino STEM bachelor’s degree holders majoring in social science fields Largest proportion of Latino STEM bachelor’s degree holders enacting a “parental support” strategy (i.e., may not be taking advantage of full range of sources of financial aid) Large proportion of Latino STEM baccalaureates complete with some debt which negatively affects graduate school enrollment 242 Policy Recommendations The ways in which Latinos simultaneously accumulate advantage and disadvantage within various institutional and financing contexts is particularly important when determining how best to address the problem of the underrepresentation of Latinos among STEM degree holders, faculty, and in the science and engineering workforce. In order to abate the underrepresentation of Latinos in STEM, steps ought to be taken to eliminate the sources of disadvantage in students’ institutional and financing contexts, while maintaining the advantaging aspects of these contexts. Below, I offer two policy proposals to accomplish this aim. The first proposal is intended to allay the disadvantages associated with the institutional context of HSIs. The second proposal is intended to mitigate the disadvantages associated with indebtedness in terms of graduate school enrollment. Undergraduate ‘Research Study’ Program As the findings of this study illustrate, Latino STEM bachelor’s degree holders commonly graduate from less selective, Hispanic-serving institutions. This is particularly true for associate degree earners, non-traditionally aged students, and first-generation college students. The findings of this study suggest that HSIs are more effective in fostering degree attainment in “hard” science fields than non-HSIs. For this important STEM educational outcome, HSIs represent an ‘advantaging’ institutional context for Latinos. However, HSIs are also considered to be lower status institutions and may constitute a ‘disadvantaging’ context due to their lower levels of prestige, selectivity, and resources (Fry, 2004). The federal government has recognized the importance of HSIs in educating Latino students, and established the Title V program to funnel more resources to these institutions. However, despite Title V funds, additional disadvantages remain in the institutional environments of HSIs, particularly for STEM majors. 243 A primary disadvantage pertains to the opportunities provided to STEM majors to engage in undergraduate research. Many HSIs are teaching institutions that lack the undergraduate research opportunities and grant monies to fund student research widely available to Latino STEM majors at research universities. The lack of research opportunities at HSIs is problematic because undergraduate research experiences have been shown to increase retention and graduate school enrollment among underrepresented minority STEM majors (Foertsch, Alexander, & Penberthy, 1997; Kardash, 2000; Kinkead, 2003; Sabatini, 1997; Seymour et al., 2004). The next question, then, is how to mitigate the disadvantage posed by the lack of research opportunities at HSIs, while maintaining the advantaging aspects of their institutional environments, (i.e., effectiveness in facilitating degree attainment in “hard” science fields)? The establishment of a federal ‘research-study’ program, akin to the existing ‘work-study’ program, would enable Latino STEM majors to garner the benefits from research opportunities while maintaining the advantages offered by Hispanic-serving institutions in terms of STEM degree outcomes. The primary goal of the program would be to increase the participation, retention, and graduate school enrollment of Latinos in high-demand STEM fields. Program details. The proposed federal ‘research study’ program would award up to $3,000/year to first, second, third, and fourth year Latino students majoring in biological, physical, or computers sciences, mathematics, or engineering at a not-for-profit, four-year mainland-U.S. HSIs. To be eligible, students would be required to maintain at least a 3.0 GPA. The funds, which would be dispensed to students by their Title V-eligible institution, would be used to support participation in STEM-related research at the student’s home institution, or off- campus at other universities, non-profit research institutes, or companies within the science and engineering industry. 244 Similar to the existing federal work-study program, eligible students would be paid hourly for part-time research positions (e.g., research assistant, lab technician, intern) related to their STEM program of study. Participants in the STEM federal research study program would be able to use these funds during the academic year or summer months to maximize opportunities to engage in research outside of their local community. The proposed research study program would require that HSIs facilitate the identification of on- and off-campus research opportunities and establish agreements with off-campus entities. HSIs with eligible students would receive supplemental funding to help defray the costs of administering the research study program. The present study revealed that an average of about 3,500 Latinos were awarded bachelor’s degrees in the identified fields by HSIs each year. Based on these findings, I estimate that between 15,000 and 18,000 Latino students would be eligible for the program in a given academic year. This constitutes an annual cost between $45,000,000 and $54,000,000, not including administrative costs. Program benefits. The proposed federal research-study program offers many potential benefits to Latino STEM majors and HSIs. First, the program would enable STEM majors to gain undergraduate research experience, mitigating a primary disadvantage associated with attending an HSI. Second, the proposed program introduces an additional source of non-loan financial aid for Latino STEM majors at HSIs. Finally, the research-study program will facilitate the formation of partnerships between HSIs and the science and engineering research community, which can lead to future collaborative opportunities for these colleges and universities. The proposed program represents an improvement over the somewhat similar National SMART [Science and Mathematics Access to Retain Talent] grant program (U.S. Department of Education, n.d.). SMART grants provide $4,000 each year to third and fourth year students who 245 are eligible for Pell Grants, are majoring in high demand fields, and have at least a 3.0 GPA. Unlike the SMART grants, the proposed program would award funds to students who participate in STEM-related research. Further, the proposed research-study program would provide aid to eligible students during all years of college, whereas SMART grants provide aid to students relatively late in their college experience. While allowing lower and upper division Latino STEM majors to receive research-study funds will make the proposed program more costly, previous research (Seymour et al., 2004) has shown that engaging undergraduate research early on in the college career increases retention in STEM fields among underrepresented minority students and helps these students to establish clear, educational and professional goals. Though the research study program as proposed would certainly have high levels of participation because HSIs would identify eligible students and automatically award those funds to Latino STEM majors who meet the criteria, students ought to be advised of the opportunities afforded to them by the program prior to enrolling in college. The research study program offers potential benefits to Latino students who may be wary of enrolling in an HSI due to a perceived lack of research opportunities. If Latino students and their parents are advised of the potential benefits of attending an HSI in the form of this new proposed source of financial aid, students who aspire to major in a STEM field may be attracted to these institutions. Debt Forgiveness Program for Latinos in STEM who Attend Graduate School The findings of this study reveal that many Latino STEM bachelor’s degree holders borrow to finance college. However, the debt that they accumulate negatively affects the likelihood of enrolling in graduate school. In order to mitigate the negative effects of debt on graduate school enrollment, I offer the following policy proposal aimed at “forgiving” portions of undergraduate debt upon the completion of educational milestones in graduate school. Increasing STEM graduate degree completion among U.S. Latinos is a national imperative. With the 246 economy becoming increasingly science and technology-based and Latinos constituting a greater share of the U.S. population, barriers to graduate school enrollment among Latino STEM bachelor’s degree holders ought to be removed. The proposed policy would act as an incentive to encourage Latino STEM bachelor’s degree holders to earn graduate degrees. Program details. In the proposed program, Latinos who enrolled in graduate school within a year of completing a STEM B.S. degree would have ten percent of their total Federal Direct Loan balance forgiven after the completion of specified educational milestones in the graduate school process. To be eligible for the debt forgiveness program, students would have to maintain a 3.0 GPA, and be enrolled in a master’s or doctoral program in mathematics, engineering, biological, physical, or computer science at an accredited, not-for-profit institution. For students enrolled in master’s programs, debt forgiveness would occur at two milestones— completion of coursework and degree conferral. Students in doctoral programs would have ten percent of their federal loan balance (principal and interest) forgiven at three milestones— completion of coursework, advancement to candidacy, and degree conferral. Eligible students would be enrolled in the proposed program automatically. Training would be provided to financial aid and academic counselors at the undergraduate level to raise students’ awareness of the program as they begin to make decisions about graduate school enrollment. Upon the completion of the specified milestones, students would be required to submit official certification from their graduate institution to the U.S. Department of Education. The findings of this study revealed that an average of 13,060 Latinos earned the bachelor’s degree in the eligible STEM fields each academic year. These students completed the baccalaureate with an average cumulative indebtedness between $10,000 and $15,000. If each of these students enrolled in a doctoral program and successfully completed the doctorate, between $2,710 and $4,065 of debt relief would be provided to each Latino student on average. This 247 corresponds to a maximum total program cost between $35,400,000 and $53,100,000 for each cohort of Latino STEM degree earners. Program benefits. The proposed policy would act to remove the disadvantaging effects of debt on graduate school enrollment among Latino STEM bachelor’s degree holders. Recognizing the barriers that debt can pose to borrowers, the U.S. Congress put income contingent repayment and debt forgiveness policies in place for degree earners who pursue public service careers in 2007 (U.S. Department of Education, 2008). In existing debt forgiveness programs, the remaining loan principal and interest balance is forgiven for individuals in qualified careers who have made 120, or ten years worth, of payments (U.S. Department of Education, 2008). Though the premise of this existing debt forgiveness program is positive, there are several reasons why this approach is not effective at removing the barriers that debt poses to graduate school enrollment. First, the existing debt forgiveness program only benefits students after they have spent many years in the labor market. Because the existing program forgives debt after a period of ten years, it would not lessen the psychosocial burdens associated with debt (Trent et al., 2006) that may prevent Latino STEM bachelor’s degree holders from entering graduate school. Second, because students must opt into the existing debt forgiveness program and inadequate training has been provided to graduate financial aid counselors, the program is being underutilized. A final drawback to the existing debt forgiveness program is that the forgiven loan balance is considered to be taxable income, leading to an increased tax liability for program participants. The proposed program overcomes these shortcomings by removing portions of Latino STEM degree holders’ debt burden while they are enrolled in a master’s degree or doctoral program, thereby serving as an incentive for graduate study. In addition, forgiven debt in the 248 proposed program would not be subject to tax in order to avoid introducing potential liabilities to program participants. The proposed program is specifically targeted at removing the disadvantages associated with student loan debt in terms of graduate school enrollment. Latino students will also benefit from a widespread information campaign targeted toward them as well as financial aid counselors, faculty, and academic support personnel. In order for the proposed debt forgiveness program to be effective in mitigating the disadvantages associated with debt, the availability and benefits of the program would be publicized using a variety of media (e.g., direct mail, internet, advertising in newsletters of STEM professional organizations, advertising at science-related conferences) prior to and during the time in which students make decisions about graduate school. As such, those practitioners who are often involved in students’ graduate school decision-making process would be targeted in addition to Latino STEM majors in order to increase the visibility of the proposed program. The above implications for policy and practice emerged from this investigation of the educational journeys of Latino STEM bachelor’s degree holders. Though cumulative disadvantage may act to constrain college decision-making, this study permits us to understand the institutional and financial aid pathways traversed by Latinos who successfully earn the STEM baccalaureate and the ways in which these pathways can be simultaneously disadvantaging and advantaging in terms of educational outcomes. Finally, the findings also point to several ways in which postsecondary institutions and policymakers can mitigate the disadvantaging aspects of institutional and financial aid policy contexts to address the problem of the underrepresentation of Latinos in STEM. 249 REFERENCES Adelman, C. (2005). Moving into town—and moving on: The community college in the lives of traditional-age students. Washington, DC: U.S. Department of Education. Adelman, C. (2006). The toolbox revisited: Paths to degree completion from high school through college. Washington, DC: U.S. Department of Education. Admon, N. (2006). Hispanic students and the decision to attend community college. Dissertation Abstracts International, 67 (09), 3355A. (UMI No. 3235692) Agodini, R., & Dynarski, M. (2004). Are experiments the only option? A look at dropout prevention programs. The Review of Economics and Statistics, 86(1), 180–194. Allen, W. R. (1986). Gender and campus differences in Black student academic performance, racial attitudes, and college satisfaction. Atlanta, GA: Southern Education Foundation. Allen, W.R. (1987). Black colleges vs. White colleges: The fork in the road for Black students. Change, 19, 28–34. Allen, W. R. (1992). The color of success: African-American college student outcomes at predominantly white and historically black public colleges and universities. Harvard Education Review, 62(1): 26–44. Allen, W. R., Epps, E. G., & Haniff, N. Z. (Eds.). (1991). College in black and white: African American students in predominantly White and in historically Black public universities. Albany, NY: SUNY. Allison, P. (2001). Missing data. (Quantitative Applications in the Social Sciences). Thousand Oaks, CA: Sage. Allison, P. (2002). Multiple imputation for missing data: A cautionary tale. Sociological Methods and Research, 28, 301–309. Alon, S. (2005). Model mis-specification in assessing the impact of financial aid on academic outcomes. Research in Higher Education, 46(1), 109–125. American Council on Education. (2006). Increasing the success of minority students in science and technology. Washington, DC: Author. Andersen, E. B. (1994). The statistical analysis of categorical data (3rd. ed.). Berlin: Springer- Verlag. Arbona, C., & Nora, A. (2007). The influence of academic and environmental factors on Hispanic college degree attainment. The Review of Higher Education, 30(3), 247–269. 250 Archer, K. J., & Lemeshow, S., & Hosmer, D. W. (2007). Goodness-of-fit tests for logistic regression models when data are collected using a complex sampling design. Computational Statistics & Data Analysis, 51(9), 4450–4464. Archer, L., & Hutchings, M. (2000). “Bettering yourself”? Discourses of risk, cost and benefit in ethnically diverse, young working-class non-participants’ constructions of higher education. British Journal of Sociology of Education, 21(4), 555–574. Astin, A. W. (1977). Four critical years. San Francisco: Jossey Bass. Astin, A. W. (1993). What matters in college? Four critical years revisited. San Francisco: Jossey Bass. Attewell, P. A., Lavin, D. E., Domina, T., & Levey, T. (2006). New evidence on college remediation. Journal of Higher Education, 77(5), 8886–924. Baker, T. L., & Velez, W. (1996). Access to and opportunity in postsecondary education in the United States: A review. Sociology of Education, 69(SI2), 82–101. Baum, S., & Saunders, D. (1998). Life after debt: Results of the national student loan survey. Braintree, MA: Nellie Mae Foundation. Becker, G. S. (1993). Human capital: A theoretical and empirical analysis with special reference to education, (3rd Ed.), Chicago: University of Chicago Press. Becker, S. O., & Caliendo, M. (2007). Sensitivity analysis for average treatment effects. Stata Journal, 7(1), 71–83. Becker, S. O., & Ichino, A. (2002). Estimation of average treatment effects based on propensity scores, Stata Journal, 2(4), 358–377. Bedard, K., Herman, D. A. (2005). Who goes to graduate/professional school? The importance of economic fluctuations, undergraduate field, and ability. Santa Barbara, CA: University of California, Santa Barbara. Unpublished Manuscript. Belenky, M. F., Clinchy, B. M., Goldberger, N. R. & Tarule, J. M. (1986). Women’s ways of knowing: The development of self, voice, and mind. New York, NY: Harper Collins. Bensimon, E. M. (2004). The diversity scorecard: A learning approach to institutional change. Change, 36(1), 45–52. Bensimon, E. M. (2007). The underestimated significance of practitioner knowledge in the scholarship of student success. Review of Higher Education, 30(4), 441–469. Bettinger, E. (2004). How financial aid affects persistence. (Working Paper 10242). Cambridge, MA: National Bureau of Economic Research. 251 Blackwell, J. E. (1998). Faculty issues: The impact on minorities. Review of Higher Education, 11(4): 417–434. Blau, P. M., & Duncan, O. D. (1967). The American occupational structure. New York: Wiley. Bohr, L., Pascarella, E. T., Nora, A., & Terenzini, P. T. (1995). Do Black students learn more at historically Black or predominantly White colleges. Journal of College Student Development, 41, 381–394. Bourdieu, P. (1986). The forms of capital. (R. Nice, Trans.). In J.G. Richardson (Ed.), Handbook of theory and research for the sociology of education (pp. 241–258). New York: Greenwood Press. Bourdieu, P., & Passeron, J.-C. (1977). Reproduction in education, society, and culture. London: Sage. Bowen, W. G., & Bok, D. (1998). The shape of the river. Princeton, NJ: Princeton University Press. Brainard, S. G., & Carlin, L. (1998). A six-year longitudinal study of undergraduate women in engineering and science. Journal of Engineering Education, 87(4), 369–375. Braunstein, A., McGrath, M., & Pescatrice, D. (2000). Measuring the impact of financial factors on college persistence. Journal of College Student Retention, 2(3), 191–203. Brazziel, W. F., & Brazziel, M. E. (1997). Distinctives of high producers of minority science and engineering doctoral starts. Journal of Science Education and Technology, 6(2), 143–153. Brint, S., & Karabel, J. (1989). The diverted dream: Community colleges and the promise of educational opportunity in America, 1990–1985. New York: Oxford University Press. Brewer, D., Eide, E., & Ehrenberg, R. (1999). Does it pay to attend an elite private college? Cross-cohort evidence on the effects of college type on earnings. Journal of Human Resources, 34, 104–123. Brown, M. C., & Davis, J. E. (2001). The historically Black college as social contract, social capital, and social equalizer. Peabody Journal of Education, 76(1), 31–49. Brown, S. E., & Santiago, D. (2004, December 27). Latino students gravitate toward HSIs. Hispanic Outlook in Higher Education, 15(7), 21. Burdman, P. (2005). The student debt dilemma: Debt aversion as a barrier to college access. Berkeley, CA: Project on Student Debt, The Institute for College Access and Success, Inc. 252 Cabrera, A. F., Burkum, K. R., & La Nasa, S. M. (2005). Pathways to a four-year degree: Determinants of transfer and degree completion. (pp. 155–214). In A. Siedman (Ed.), College student retention: A formula for student success. Westport, CT: ACE/Praeger Series on Higher Education. Cabrera, A. F., La Nasa, S. M., & Burkum, K. R. (2001). Pathways to a four-year degree: The higher education story of one generation. State College, PA: Center for the Study of Higher Education, The Pennsylvania State University. Cabrera, A. F., & Nora, A. (1994). College students’ perceptions of prejudice and discrimination and their feelings of alienation. Review of Education, Pedagogy, and Cultural Studies, 16, 387–409. Cabrera, A. F., Nora, A., & Castañeda, M. B. (1993). College persistence: Structural modeling of an integrated model of student retention. Journal of Higher Education, 64, 123–139. Caliendo, M., & Kopeinig, S. (2005). Some practical guidance for the implementation of propensity score matching. Discussion paper No. 1588. Retrieved on February 6, 2007 from http://ssrn.com/abstract=721907 California State University, Office of the Chancellor (1997, February). Executive Order No. 665: Determination of Competence in English and Mathematics. Long Beach, CA: California State University. Callahan, R., Wilkinson, L., & Muller, C. (2008). Social context and the effect of ESL placement on Mexican-origin adolescents’ achievement. Social Science Quarterly, 89(1), 177–198. Carter, D. F. (1999). The impact of institutional choice and environments on African-American and White students’ degree expectations. Research in Higher Education, 40(1), 17–41. Castellanos, J., Gloria, A. M., & Kamimura, M. (2005). The Latina/o Pathway to the Ph.D. Sterling, VA: Stylus. Chamberlain, M. K. (1988). Women in academe: Progress and prospects. New York: Russell Sage Foundation. Chapa, J., & Schink, W. (2006). California community colleges: help or hindrance to Latinos in the higher education pipeline? New Directions for Community Colleges, 133, 41–50. Chen, R., & DesJardins, S. L. (2006). Exploring the effects of financial aid on the gap in student dropout risks by family income. Ann Arbor, MI: University of Michigan. Unpublished Manuscript. Chesler N., & Chesler M. (2002). Gender-informed mentoring strategies for women engineering scholars: On establishing a caring community. Journal of Engineering Education, 91(1), 49–55. 253 Christie, H., & Munro, M. (2003). The logic of loans: Students’ perception of the costs and benefits of the student loan. British Journal of Sociology of Education, 24(5), 621–636. Cofer, J., & Somers, P. (2001). What influences student persistence at two-year colleges? Community College Review, 29, 56–76. Cohen, A. M., & Brawer, F. B. (2003). The American community college. San Francisco: Jossey Bass. Cokley, K. (1999). Reconceptualizing the impact of college racial composition on African American students’ racial identity. Journal of College Student Development, 40, 235– 246. Coleman, J. S. (1990). Foundations of social theory. Cambridge, MA: Belknap Press. College Board. (2006). Trends in student aid. (Trends in Higher Education Series). Washington, DC: Author. College Board. (2007). Trends in student aid. (Trends in Higher Education Series). Washington, DC: Author. Committee on Equal Opportunities in Science and Engineering. (2004). Broadening participation in America’s science and engineering workforce. (CEOSE 04–01). Washington, DC: Author. Contreras, F. E., Malcom, L. E., & Bensimon, E. M. (2008). Hispanic-serving institutions: Closeted identity and the production of equitable outcomes for Latino/a students. In M. Gasman, B. Baez, & C. S. V. Turner (Eds.), Understanding minority serving institutions (pp. 71–90). Albany, NY: SUNY Press. Cooper, C. R. (2002). Five bridges along students’ pathways to college: A development blueprint of families, teachers, counselors, mentors, and peers in the Puente project. Educational Policy, 16(4), 607–622. Creswell, J. W. (1994). Research design: Qualitative and quantitative approaches. Thousand Oaks, CA: Sage. Dale, S., & Krueger, A. (2002). Estimating the payoff to attending a more selective college: an application of selection on unobservables. Quarterly Journal of Economics, 117, 1491–1528. De La Rosa, M. L., & Hernandez-Gravelle, H. (2007, March). The strategy of debt aversion: How Hispanic students pay for college. Paper presented at the meeting of the American Association of Hispanics in Higher Education, Inc., Costa Mesa, CA. 254 de los Santos, A. G., & de los Santos, G. E. (2005). Latina/os and community colleges: A pathway to graduate studies? In J. Castellanos, A.M. Gloria, & M. Kamimura (Eds.), The Latina/o Pathway to the Ph.D. (pp. 37–53). Sterling, VA: Stylus. DesJardins, S. L., Ahlburg, D. A., & McCall, B. P. (2006). An integrated model of application, admission, enrollment, and financial aid. Journal of Higher Education, 77(3), 381–429. DesJardins, S. L., Kim, D., & Rzonca, C. S. (2002). A nested analysis of factors affecting bachelor’s degree completion. Journal of College Student Retention, 4(4), 407–435. DesJardins, S. L., McCall, B. P., Ahlburg, D. A., & Moye, M. J. (2002). Adding a timing light to the “toolbox”. Research in Higher Education, 43(1), 83-114. DiPrete, T. A. & Eirich, G. M. (2006). Cumulative advantage as a mechanism for inequality: A review of theoretical and empirical development. Annual Review of Sociology, 32, 271– 97. Dougherty, K. J. (1994). The contradictory college: Conflicting origins, impacts, and futures of the community college. Albany, NY: SUNY Press. Dougherty, K.J., & Kienzl, G.S. (2006). It’s not enough to get through the open door: Inequalities by social background in transfer from community colleges to four-year colleges. Teachers College Record, 108(3), 452–487. Dowd, A. C. (2004, May 12). Income and financial aid effects on persistence and degree attainment in public colleges. Education Policy Analysis Archives, 12(21) Retrieved April 4, 2007, from http://epaa.asu.edu/epaa/v12n21/ Dowd, A. C. (in press). Dynamic interactions and intersubjectivity: Challenges to causal modeling in studies of college student debt. Review of Educational Research. Dowd, A. C., Bensimon, E. M., Gabbard, G., Singleton, S., Macias, E., Dee, J., et al. (2006). Transfer access to elite colleges and universities in the United States: Threading the needle of the American dream. Boston and Los Angeles: University of Massachusetts Boston and the University of Southern California. Dowd, A. C., Cheslock, J., & Melguizo, T. (in press). Transfer access from community colleges and the distribution of elite higher education. The Journal of Higher Education. Dowd, A. C., & Coury, T. (2006). The effect of loans on the persistence and attainment of community college students. Research in Higher Education, 47(1), 33–62. ECMC Group Foundation. (2003). Cultural barriers to incurring debt: An exploration of borrowing and impact on access to postsecondary education. Santa Fe, NM: Author. 255 Ehrenberg, R.G., & Rothstein, D.S. (1993). Do historically black institutions of higher education confer unique advantages on black students: An initial analysis. Working Paper No. 4356. Cambridge, MA: National Bureau of Economic Research. Eide, E., Brewer, D., & Ehrenberg, R. (1998). Does it pay to attend an elite private college? Evidence on the effect of undergraduate college quality on graduate school attendance. Economics of Education Review, 17, 371–376. Elman, C., & O’Rand, A. (2004). The race is to the swift: Socioeconomic origins, adult education, and wage attainment. American Journal of Sociology, 110, 123–160. Etzkowitz, H., Kemelgor, C., Neuschatz, M., & Uzzi, B. (1994). Barriers to women’s participation in academic science and engineering. In W. Pearson & A. Fechter (Eds.), Who will do science? Baltimore, MD: Johns Hopkins University Press. Fenske, R.H., Porter, J.D., & DuBrock, C.P. (2000). Tracking financial aid and persistence of women, minority, and needy students in science, engineering, and mathematics. Research in Higher Education, 41(1), 67–94. Fleming, J. (1984). Blacks in college. San Francisco: Jossey Bass. Flores, S.M., Horn, C.L., Crisp, G. (2006). Community colleges, public policy, and Latino student opportunity. New Directions for Community Colleges, 133, 71–80. Florida Department of Education. (n.d.). Frequently asked questions: Community colleges. Retrieved March 12, 2008, from http://www.fldoe.org/faq/default.asp?Dept=135& ID=65#Q65 Florida Department of Education. (2005). Access challenges and Florida’s community college and workforce education system. Tallahassee, FL: Author. Flyer, F. (1997). The influence of higher moments of earnings distributions on career decisions. Journal of Labor Economics, 15(4), 689–713. Foertsch, J., Alexander, B. B., & Penberthy, D. (1997). Summer research opportunity programs (SROPs) for minority undergraduates: A longitudinal study of program outcomes, 1986– 1996. Madison: The Lead Center, University of Wisconsin-Madison. Fosnacht, K. J., & Santos, J. L. (2008, March). The influence of debt burden on career sector choice: An exploratory analysis. Paper presented at the annual meeting of the American Educational Research Association, New York. Freeman, K.E., Perna, L.W., & King, N. (1999). UNCF Statistical Report: 1998. Fairfax, VA: Frederick D. Patterson Research Institute. Frölich, M. (2004). Finite sample properties of propensity-score matching and weighting estimators. The Review of Economics and Statistics, 86(1), 77–90. 256 Fry, R. (2002). Latinos in higher education: Many enroll, Too few graduate. Pew Hispanic Center: Washington, DC. Fry, R. (2004). Latino youth finishing college: The role of selective pathways. Pew Hispanic Center: Washington, DC. Fuerstenau, C. (2002, April). Valuing context: The impact of social capital on students’ college choice. Paper presented at the annual meeting of the American Educational Research Association, New Orleans, LA. Gándara, P. (2002). A study of high school Puente: What we have learned about preparing Latino youth for postsecondary education. Educational Policy, 16(4), 474–495. Gándara, P., & Bial, D. (2001). Paving the way to postsecondary education: K–12 intervention programs for underrepresented youth. Washington, DC: National Postsecondary Education Cooperative, U.S. Department of Education. Gándara, P., Larson, K., Mehan, H., & Rumberger, R.W. (1998). Capturing Latino students in the academic pipeline. CLPP Policy Report. University of California, Berkeley: Chicano/Latino Policy Project. Goldrick-Rab, S. (2006). Following their every move: An investigation of social-class differences in college pathways. Sociology of Education, 2006, 79(2), 61–79. González, N., & Moll, L. (2002). Cruzando el Puente: Building bridges to funds of knowledge. Educational Policy, 16(4), 623–641. Goodman, L. A. (2002). Latent class analysis: The empirical study of latent types, latent variables, and latent structures. In J. A. Hagenaars & A. L. McCutcheon (Eds.), Applied latent class analysis (pp. 3–55). Cambridge, UK: Cambridge University Press. Grodsky, E. S. (2002). Constrained opportunity and student choice in American higher education. Dissertation Abstracts International, 63 (08), 3008A. (UMI No. 3060591) Grubb, W.N. (1991). The decline of community college transfer rates: Evidence from national longitudinal surveys. Journal of Higher Education, 62, 184–217. Grubb, W.N., Lara, C.M., Valdez, S. (2002). Counselor, coordinator, monitor, mom: The roles of counselors in the Puente program. Educational Policy, 16(4), 547–571. Guess, A. (2007, May 30). Student aid, opportunity and Latinos. Insider Higher Ed. Retrieved September 20, 2007, from http://www.insidehighered.com/news/2007/05/30/finance Guo, S., Barth, R., & Gibbons, C. (2004). Introduction to propensity score matching: A new device for program evaluation, Society for Social Work Research. New Orleans. 257 Hall, R. R., & Sandler, B.R. (1982). The classroom climate: A chilly one for women. Project on the Status and Education of Women. Washington, DC: Association of American Colleges. Hallinan, M. T. (2000). On the linkages between sociology of race and ethnicity and sociology of education. In M. T. Hallinan (Ed.), Handbook of the Sociology of Education (pp. 65–84). New York: Kluwer Academic/Plenum. Ham, J.C., Li, X., & Reagan, P.B. (2006, February). Propensity score matching, a distance-based measure of migration, and the wage growth of young men (IEPR working paper 05.13). Los Angeles, CA: University of Southern California, Institute of Economic Policy Research. Hao, L. (2006). Assessing equitable postsecondary educational outcomes for Hispanics in California and Texas. Dissertation Abstracts International, 67 (10), 3732A. (UMI No. 3237151) Harvey, W.B., & Williams, L.E. (1989). Historically black colleges: Models for increasing minority representation. Education and Urban Society, 21, 328–340. Haro, R. (2004). Programs and strategies to increase Latino students’ educational attainment. Education and Urban Society, 36(2), 205–222. Heckman, J. (1979). Sample selection bias as specification error. Econometrica, 47(1), 153–161. Heller, D. E. (Ed.). (2002). Condition of access: Higher education for lower income students. Westport, CT: American Council on Education and Praeger Publishers. Hippensteel, D. G., St. John, E. P., & Starkey, J. B. (1996). Influence of tuition and student aid on within-year persistence by adults in two-year colleges. Community College Journal of Research and Practice, 20, 233–242. Hobbs, F., & Stoops, N. (2002). Demographic trends in the 20th century. (Census 2000 Special Reports, Series CENSR–4). Washington, DC: U.S. Government Printing Office. Holland, D.C., & Eisenhart, M.A. (1992). Educated in romance: Women, achievement, and college culture. Chicago: University of Chicago Press. Horn, L., Wei, C. C., Berker, A. (2002). What students pay for college: Changes in net price of college attendance between 1992–93 and 1999–2000 (NCES 2002174). Washington, DC: National Center for Education Statistics. Hossler, D., Braxton, J., & Coopersmith, G. (1989). Understanding student college choice. In J. C. Smart (Ed.), Higher education: Handbook of theory and research (Vol. V, pp. 231– 288). New York: Agathon Press. Hossler, D., & Gallaher, K. S. (1987). Studying college choice: A three-phase model and the implications for policy-makers. College and University, 2, 207–221. 258 Hossler, D., Schmit, J., & Vesper, N. (1999). Going to college: How social, economic, and educational factors influence, the decisions students make. Baltimore, MD: Johns Hopkins University Press. Hoxby, C. M. (Ed.). (2004). College choices: The economics of where to go, when to go, and how to pay for it. Chicago: University of Chicago Press. Hu, S., & St. John, E.P. (2001). Student persistence in a public higher education system: Understanding racial and ethnic differences. The Journal of Higher Education, 72(3), 265–286. Hurtado, S. (2001). Linking diversity and education purpose: How diversity affects the classroom environment and student development. (pp.187–203). In G. Orfield (Ed.), Diversity Challenged: Evidence on the Impact of Affirmative Action. Cambridge, MA: Harvard Education Publishing Group. Hurtado, S. & Carter, D.F. (1997). Effects of college transition and perceptions of the campus racial climate on Latino students’ sense of belonging. Sociology of Education, 70(4), 324–345. Hurtado, S., Inkelas, S., Briggs, C., & Rhee, B. (1997). Differences in college access and choice among racial/ethnic groups: Identifying continuing barriers. Research in Higher Education, 38(1), 43–75. Hurtado, S., & Ponjuan, L. (2005). Latino educational outcomes and the campus climate. Journal of Hispanic Higher Education, 4(3), 235–251. Ichino, A., Mealli, F., & Nannicini, T. (2006). From temporary help jobs to permanent employment: What can we learn from matching estimators and their sensitivity. IZA Discussion Paper No. 2149. Bonn, Germany: Institute for the Study of Labor. Immerwahr, J. (2003). With diploma in hand: Hispanic high school seniors talk about their future. (National Center Report #03–2). San Jose, CA: National Center for Public Policy and Higher Education, and Public Agenda. Ishitani, T. T., & DesJardins, S. L. (2002). A longitudinal investigation of dropout from college in the United States. Journal of College Student Retention, 4(2), 173–201. Jang, D., & Lin, X. (2007). Evaluation of the National Survey of Recent College Graduates (NSRCG) postsecondary institution sample. Washington, DC: Mathematica Policy Research, Inc. Jones, L, Castellanos, J., & Cole, D. (2002). Examining the ethnic minority student experience at predominantly White institutions: A case study. Journal of Hispanic Higher Education, 1, 19–39. 259 Kane, M., Beals, C., Valeau, E. J., & Johnson, M. J. (2004). Fostering success among traditionally underrepresented student groups: Hartnell College's approach to implementation of the Math, Engineering, and Science Achievement (MESA) Program. Community College Journal of Research and Practice, 28, 17-26. Kane, T. J., & Rouse, C. E. (1995). Labor-market returns to two- and four-year college. American Economic Review, 85(3), 600–614. Kardash, C. M. (2000). Evaluation of an undergraduate research experience: Perceptions of undergraduate interns and their faculty mentors. Journal of Educational Psychology, 92(1), 191–201. Karabel, J. (1972). Community colleges and social stratification. Harvard Educational Review, 42, 521–562. Kim D. (2004). The effect of financial aid on students’ college choice: Differences by racial groups. Research in Higher Education, 45(1), 43–70. Kim, D., & Eyermann, T.S. (2006). Undergraduate borrowing and its effects on plans to attend graduate school prior to and after the 1992 Higher Education Act amendments. NASFAA Journal of Student Financial Aid, 36(2), 5–21. Kim, M.M. (2002). Historically Black vs. White institutions: Academic development among Black students. Review of Higher Education, 25(4), 385–407. Kinkead, J. (2003). Learning through inquiry: An overview of undergraduate research. New Directions for Teaching and Learning, 93, 5–17. Kinzie, J., Palmer, M., Hayek, J., Hossler, D., Jacob, S., & Cummings, H. (2004). Fifty years of college choice: Social political and institutional influences on the decision-making process. Indianapolis, Indiana: Lumina Foundation for Education: New Agenda Series. Kuh, G. D., Kinzie, J., Schuh, J. H., Whitt, E. J., & Associates. (2005). Student success in college: Creating conditions that matter. San Francisco: Jossey Bass. Kurlaender, M. (2006). Choosing community college: Factors affecting Latino college choice. New Directions for Community Colleges, 133, 7–16. Kurlaender, M., & Long, B.T. (2007, April). Do community colleges serve as a viable pathway to a baccalaureate degree? Paper presented at the Annual Meeting of the American Educational Research Association, Chicago, IL. Laden, B.V. (1999). Socializing and mentoring college students of color: The Puente project as an exemplary celebratory socialization model. Peabody Journal of Education, 74(2), 55–74. 260 Laden, B. V. (2000). Hispanic-serving two-year institutions: What account for their high transfer rates? Paper presented at the annual meeting of the Association for the Study of Higher Education, Sacramento. Laden, B. V. (2004). Hispanic-serving institutions: What are they? Where are they? Community College Journal of Research and Practice, 28, 181–198. Lareau, A. (2003). Unequal childhoods: Class, race, and family life. Berkeley, CA: University of California Press. Lavin, D. E., & Hyllegard, D. (1996). Changing the odds: Open admissions and the life changes of the disadvantaged. New Haven: Yale University Press. Ledman, R.E., Miller, M., & Brown, D.R. (1995, October). Successful women and women’s colleges: Is there an intervening variable in the reported relationship?, Sex Roles: A Journal of Research, 489–97. Leigh, D. E., & Gill, A. M. (2007). Do community colleges respond to local needs? Evidence from California. Kalamazoo, MI: W.E. Upjohn Institute for Employment Research. Leuven, E., & Sianesi, B. (2003). PSMATCH2: Stata module to perform full Mahalanobis and propensity score matching, common support graphing, and covariate imbalance testing [Computer software]. Retrieved December 11, 2007, from http://ideas.repec.org/c/boc/ bocode/s432001.html Long, B. T., & Riley, E. K. (2007). Financial aid: A broken bridge to college access? Harvard Educational Review, 77(1), 39–63. Long, J. S. (1997). Regression models for categorical and limited dependent variables. Thousand Oaks, CA: Sage. MacDonald, V.M., Botti, J.M., & Clark, L.H. (2007). From visibility to autonomy: Latinos in higher education in the U.S., 1965–2005. Harvard Educational Review, 77(4), 474–504. Malcom, L. (2006a, November). Exploring the effects of community college attendance on the educational, professional, and economic outcomes of scientists and engineers. Paper presented at the Annual Meeting of the Association for the Study of Higher Education, Anaheim, California. Malcom, L. (2006b). Moving students forward: The Pasadena City College MESA Program director as an institutional agent. Los Angeles, CA: University of Southern California. Unpublished Manuscript. Manski, C. F., & Wise, A. D. (1983). College choice in America. Cambridge, MA: Harvard University Press. Markus, H., & Nurius, P. (1986). Possible selves. American Psychologist, 41(9), 954–969. 261 McCormick, A. C. (2003). Swirling and double-dipping: New patterns of student attendance and their implications for higher education. New Directions for Higher Education, 121. 13–24. McCutcheon, A. L. (2002). Basic concepts and procedures in single- and multiple-group latent class analysis. In J. A. Hagenaars & A. L. McCutcheon (Eds.), Applied latent class analysis (pp. 56–89). Cambridge, UK: Cambridge University Press. McDonough, P. M. (1997). Choosing colleges: How social class and schools structure opportunity. Albany, NY: SUNY Press. McDonough, P. M. (2004, July). The impact of advice on price: Evidence from research. Retrieved March 15, 2007, from http://www.teri.org/researchstudies/financial-aid- impact.asp McDonough, P. M., Antonio, A. L., & Trent, J. W. (1997). Black students, Black colleges: An African American college choice model. Journal for a Just and Caring Education, 3, 9– 36. McDonough, P. M., & Calderone, S. (2006). The meaning of money: Perceptual differences between college counselors and low-income families about college costs and financial aid. American Behavioral Scientist, 49(12), 1703–1718. Melguizo, T. (in press). Are community colleges an alternative path for Hispanic students to attain a bachelor's degree? Teachers College Record. Melguizo, T., Kienzl, G.E., & Alfonso, M. (2007, April). Evaluating the Effect of Community College Attendance on Educational Attainment: Application of Propensity Score Matching Methods. Paper presented at the Annual Meeting of the American Educational Research Foundation, Chicago, IL. Merton, R.K. (1968). The Matthew effect in science. Science, 159(3810), 56–63. Merton, R.K. (1988). The Matthew effect in science, II: Cumulative advantage and the symbolism of intellectual property. Isis, 79(4), 606–623. MESA Community College Program. (n.d.). Mesa Community College Program. Retrieved April 4, 2007 from http://www.ucop.edu/mesa/staff/Staff%20Only/ MESA_Community_College_Program.pdf Millett, C.M. (2003). How undergraduate loan debt affects application and enrollment in graduate or first professional school. The Journal of Higher Education, 74(4), 386–427. Millimet, D. (2001). What is the difference between 'endogeneity' and 'sample selection' bias? Retrieved September 16, 2002, from http://www.stata.com/support/faqs/stat/bias.html 262 Minicozzi, A. (2005). The short term effect of educational debt on job decisions. Economics of Education Review, 24(4), 415–430. Monaghan, P. (2001). Researchers note correlation between students' race and willingness to take on college loans. Retrieved April 17, 2001, from http://chronicle.com Monks, J. (2000). The returns to individual and college characteristics: evidence from the national longitudinal survey of youth. Economics of Education Review, 19, 279–289. Morrow, R., & Torres, C. (1995). Social theory and education. Albany, NY: SUNY Press. Murphy, S. H. (1994). The impact of student debt on post-baccalaureate decisions: A focus on educational choice. Dissertation Abstracts International, 55 (05), 1158A. Muthén, B. O. (2001). Second-generation structural equation modeling with combination of categorical and continuous latent variables: New opportunities for latent class/latent growth modeling. In L. M. Collins & A. G. Sayer (Eds.), New methods for the analysis for change (pp. 291–322). Washington, DC: American Psychological Association. Myers, D. (2002). Demographic dynamism in Los Angeles, Chicago, New York, and Washington D.C. In M.J. Dear (Ed.), From Chicago to L.A.: Making sense of urban theory (pp. 21– 53). Thousand Oaks, CA: Sage. Myers, D. (2007). Immigrants and boomers: Forging a new social contract for the future of America. New York: Russell Sage. National Center of Education Statistics. (2002). Digest of education statistics, 2000. Washington, DC: Author. National Center of Education Statistics. (2005). Digest of education statistics, 2005. Washington, DC: Author. National Center of Education Statistics. (2006). Placing college graduation rates in context: How 4-year college graduation rates vary with selectivity and the size of low-income enrollment (NCES 2007161). Washington, DC: Author. National Center of Education Statistics. (2007). Digest of education statistics, 2006 (NCES 2007017). Washington DC: Author. National Research Council. (1999). Harnessing science and technology for America’ economic future: National and regionally priorities. Washington, DC: National Academies Press. National Research Council. (2005). Enhancing the community college pathway to engineering careers. Washington, DC: National Academies Press. 263 National Science Foundation. (1996). Shaping the future: New expectations for undergraduate education in science, mathematics, engineering, and technology (NSF 96-139). Washington, DC: Author. National Science Foundation. (1999). What is the level of federal science and engineering support to historically black colleges and universities? Issue Brief (NSF 99–356). Arlington, VA. National Science Foundation. (2002). Science and engineering degrees, by race/ethnicity of recipients: 1991–2002 (NSF 02–329). Arlington, VA. National Science Foundation. (2005). Women, minorities, and persons with disabilities in science and engineering: 2004 (NSF 04–317). Arlington, VA. National Science Foundation. (2006). Investing in America’s future (NSF 06–48). Arlington, VA. Nettles, M. T. (1989). Comparing the backgrounds, educational experiences and outcomes of Black, Hispanic, and White doctoral students. College Park, MD: National Center for Postsecondary Governance and Finance. Nettles, M.T., Thoeny, A.R., & Gosman, E.J. (1986). Comparative and predictive analyses of black and white students’ college achievement and experiences. Journal of Higher Education, 57, 289–328. Nora, A. (2002). A theoretical and practice view of student adjustment and academic achievement. In W. Tierney & L. Hagedorn (Eds.), Increasing access to college: Extending possibilities for all students (pp. 41–56). Albany, NY: SUNY Press. Nora, A. (2004). The role of habitus and cultural capital in choosing a college, transitioning from high school to higher education, and persisting in college among minority and non- minority students to college. Journal of Hispanic Higher Education, 3(2), 180–208. Nora, A., Barlow, L., & Crisp, G. (2006). Examining the tangible and psychosocial benefits of financial aid with student access, engagement, and degree attainment. American Behavioral Scientist, 49, 1636–1651. Nora, A., & Cabrera, A.F. (1996). The role of perceptions of prejudice and discrimination on the adjustment of minority students to college. Journal of Higher Education, 67(2), 120–148. Nora, A., Cabrera, A.F., Hagedorn, L.S., & Pascarella, E. (1996). Differential effects of academic and social experiences on college-related outcomes across different ethnic and gender groups at four-year institutions, Research in Higher Education, 37(4), 427–452. Nora, A., & Horvath, F. (1989). Financial assistance: Minority enrollments and persistence. Education and Urban Society, 21(3), 299–311. Olivas, M. A. (1985). Financial aid packaging policies: Access and ideology. Journal of Higher Education, 56(4), 462–475. 264 Olivas, M. A. (1986). Financial aid and self-reports by disadvantaged students: The importance of being earnest. Research in Higher Education, 25(3), 245–252. Olivas, M. A. (2005). Higher education as “place”: Location, race, and college attendance policies. Review of Higher Education, 28(2), 169–189. Orazem, P. F., & Mattila, J. P. (1991). Human capital, uncertain wage distributions, and occupational and educational choices. International Economic Review, 32(1), 103–122. Osei-Kofi, N. & Rendón, L. (2005). Latinos, higher education and the “needs” of the market. Latino Studies, 3, 249–260. Pachon, H. P., & Zarate, M. E. (2005). Perspectives of college costs, benefits, and financial aid among California Latino youth. Los Angeles, CA: The Tomás Rivera Policy Institute and the Sallie Mae Fund. Pascarella, E., & Terenzini, P. (1991). How college affects students: Findings and insights from twenty years of research. San Francisco: Jossey Bass. Pascarella, E., & Terenzini, P. (2003). How college affects students: A third decade of research. San Francisco: Jossey Bass. Paulsen, M. (1990). College choice: Understanding student enrollment behavior. ASHE-ERIC Higher Education Report No. 6. Washington, DC: George Washington University, School of Education and Human Development. Paulsen, M. B., & St. John, E. P. (2002). Social class and college costs: Examining the financial nexus between college choice and persistence. Journal of Higher Education, 73(2), 189– 236. Pearson, W., & Pearson, L.C. (1985). Baccalaureate origins of Black American scientists: A cohort analysis. Journal of Negro Education, 54(1), 24–34. Peng, C.J., So, T.H., Stage, F.K. & St. John, E.P. (2002). The use and interpretation of logistic regression in higher education journals: 1988–1998. Research in Higher Education, 43(3), 259–293. Perna, L. W. (2000). Differences in the decision to attend college among African Americans, Hispanics, and Whites. Journal of Higher Education, 71(2), 117–141. Perna, L. W. (2001). The contribution of historically Black colleges and universities to the preparation of African American for faculty careers. Research in Higher Education, 42(3), 267–294. 265 Perna, L. W. (2004, July). Impact of student aid program design, operations, and marketing on the formation of family college-going plans and resulting college-going behaviors of potential students. Boston, MA: The Education Resources Institute, Inc. (TERI). Perna, L. W. (2006a). Studying college access and choice: A proposed conceptual model. In J.C. Smart (Ed.), Higher education: Handbook of theory and research (Vol. XXI, pp. 99– 157). Dordrecht, The Netherlands: Kluwer Academic Publishers. Perna, L.W. (2006b). Understanding the relationship between information about college prices and financial aid and students’ college-related behaviors. American Behavioral Scientist, 49(12), 1620–1635. Perna, L. W., & Titus, M. A. (2004). Understanding differences in the choice of college attended: The role of state public policies. Review of Higher Education, 27(4), 501–525. Person, A. E., & Rosenbaum, J. E. (2006). Chain enrollment and college enclaves: Benefits and drawbacks of Latino college students’ enrollment decisions. New Directions for Community Colleges, 133, 51–60. Phillippe, K.A., & Gonzalez Sullivan, L. (2005). National profile of community colleges: Trends and statistics, 4 th edition. Washington, DC: Community College Press. Pollock, M. (2008). Because of race: How Americans debate harm and opportunity in our schools. Princeton, NJ: Princeton University Press. Portes, A. (1998). Social capital: Its origins and applications in modern sociology. Annual Review of Sociology, 24, 1–24. Price, D.V. (2004). Borrowing inequality: Race, class, and student loans. Boulder, CO: Lynne Rienner. Puente Project. (n.d.). Welcome to the Puente project. Retrieved April 4, 2007 from http://www.puente.net/ Redd, K.E. (1998). Historically Black colleges and universities: Making a comeback. New Directions for Higher Education, 102, 33–43. Redd, K.E. (2005). Data sources: Financial aid for minority students in STEM fields. CGS Communicator, August/September, 4–5. Redmond, C., Clinedinst, M., & O’Brien, C. (2000). Education the emerging majority: The role of minority-serving colleges and universities in confronting America’s teacher crisis. Washington, DC: Alliance for Equity in Higher Education, Institute for Higher Education Policy. Rendón, L.I. (1992). From the barrio to the academy: Revelations of a Mexican American “scholarship girl.” New Directions for Community College, 20(4), 55–64. 266 Rendón, L.I. (1994). Validating culturally diverse students: Toward a new model of learning and student development. Innovative Higher Education, 19(1), 33–51. Rendón, L. I. (1995, March). Facilitating retention and transfer for first-generation students in community colleges. Paper presented at the New Mexico Institute: Rural Community College Initiative, Espanola, NM. (ED 383 369) Rendón, L.I. (2002). Community college Puente: A validating model of education. Educational Policy, 16(4), 642–667. Rendón, L. I., & Garza, H. (1996). Closing the gap between two- and four-year institutions. In L. I. Rendón & R.O. Hope (Eds.), Educating a new majority: Transforming America's educational system for diversity (pp. 289–308). San Francisco: Jossey Bass. Rendón, L. I., Jalomo, R. E., & Nora, A. (2000). Theoretical consideration in the study of minority student retention in higher education. In J. Braxton (Ed.) Rethinking the departure puzzle: New theory and research on college student retention (pp. 127–156). Nashville, TN: Vanderbilt University Press. Rosenbaum, P. R. (1984). From association to causation in observational studies: The role of tests of strongly ignorable treatment assignment. Journal of the American Statistical Association, 79(385), 41–48. Rosenbaum, P. R. (2002). Observational studies. New York: Springer. Rosenbaum, P.R., & Rubin, D.B. (1983). The central role of the propensity score in observational studies for causal effects. Biometrika. 70, 41–55. Sabatini, D. A. (1997). Teaching and research synergism: The undergraduate research experience. Journal of Professional Issues in Engineering Education and Practice, 123, 98–102. Sable, J., & Garofano, A. (2007). Public elementary and secondary school student enrollment, high school completions, and staff from the common core of data: School year 2005–06 (NCES 2007–352). U.S. Department of Education. Washington, DC: National Center for Education Statistics. Santiago, D. A. (2006). Inventing Hispanic-serving institutions (HSIs): The basics. Washington, DC: Excelencia in Education. Santiago, D. A., & Cunningham, A. F. (2005). How Latino students pay for college: Patterns of financial aid in 2003–2004: Excelencia in Education and the Institute for Higher Education Policy. Sax, L.J., Astin, A.W., Korn, W.S., & Mahoney, K.M. (1997). The American freshman: National norms for fall 1997. Los Angeles, CA: Higher Education Research Institute, UCLA. 267 Schapiro, M. O., O'Malley, M. P., & Litten, L. H. (1991). Progression to graduate school from the 'elite' colleges and universities. Economics of Education Review, 10(3), 227–244. Schrag, F. (1992). In defense of positivist research paradigms. Educational Researcher, 21(5), 5–8. Schwab, J. J. (1978). Science, liberal education, and curriculum: Selected essays. Chicago: University of Chicago Press. Sebrechts, J. (2000, Spring/Summer). Nurturing scientists at women’s colleges. Trinity Magazine, Retrieved March 27, 2007 from http://www.trinitydc.edu/news_events/mags/ winspr00/2nursci.html Sewell, W. H., Hauser, R. M., & Wolf, W. C. (1986). Sex, schooling and occupational status. American Journal of Sociology, 86, 551–583. Seymour, E., & Hewitt, N.M. (1997). Talking about leaving: Why undergraduates leave the sciences. Boulder, CO: Westview. Seymour, E., Hunter, A., Laursen, S. L., & Deantoni, T. (2004). Establishing the benefits of research experiences for undergraduates in the sciences: First findings from a three-year study. Science Education, 88, 493–594. Sianesi, B. (2001, May). Implementing propensity score matching estimators with STATA. Prepared for UK Stata Group, VII Meeting: University College London and Institute for Fiscal Studies. Small, K. A., & Hsiao, C. (1985). Multinomial logit specification tests. International Economic Review, 26, 619–627. Smith, D.G., Wolf, L.E., & Morrison, D.E. (1995). Paths to success: Factors related to the impact of women’s colleges. The Journal of Higher Education, 66, 245–266. Solórzano, D. G. (1995). The baccalaureate origins of Chicana and Chicano doctorates in the social sciences. Hispanic Journal of Behavioral Sciences, 17(1), 3–32. Somerton, W.H., Smith, M., Finnell, P., & Fuller, T. (1994). The MESA Way: A Success Story of Nurturing Minorities for Math/Science-Based Careers. San Francisco: Caddo Gap Press. St. John, E. P. (2006). Contending with financial inequality and academic success: Rethinking the contributions of qualitative research to the policy discourse on college students. American Behavioral Scientist, 49(12), 1604–1619. St. John, E. P., Andrieu, S., Oescher, J., & Starkey, J. B. (1994). The influence of student aid on within-year persistence by traditional college-age students in four-year colleges. Research in Higher Education, 35(4), 455–480. 268 St. John, E. P. (2003). Refinancing the college dream: Access, equal opportunity, and justice for taxpayers. Baltimore, MD: Johns Hopkins University Press. St. John, E. P., Paulsen, M. B., & Carter, D. F. (2005). Diversity, college costs, and postsecondary opportunity: An examination of the financial nexus between college choice and persistence. Journal of Higher Education, 76(5), 545–569. St. John, E. P., & Starkey, J. B. (1994). The influence of costs on persistence by traditional college-age students in community colleges. Community College Journal of Research and Practice, 18, 201–213. St. John, E. P., & Starkey, J. B. (1995). An alternative to net price: Assessing the influence of prices and subsidies on within-year persistence. Journal of Higher Education, 66(2), 156–186. Stage, F.K. (Ed.). (2007). Using quantitative data to answer critical questions: New Directions for Institutional Research, 133. Stage, F. K., & Hubbard, S. M. (2008, March). Producers of future women and minority scientists: Comprehensive and doctoral institutions. Paper presented at the annual meeting of the American Educational Research Association, New York, NY. Stampen, J. O., & Cabrera, A. F. (1988). The targeting and packaging of student aid and its effect on attrition. Economics of Education Review, 7(1), 29–46 Stanton-Salazar, R. D. (2001). Manufacturing hope and despair: The school and kin support networks of U.S. Mexican youth. New York: Teachers College Press. Stearns, C., & Watanabe, S. (2002). Hispanic-serving institutions: Statistical trends from 1990– 1999. National Center for Education Statistics (NCES 2002–051). Washington, DC: U.S. Department of Education. Strassburger, J. (1995, May). Embracing undergraduate research. American Association of Higher Education (AAHE) Bulletin, 47(9), 3–5. Suitts, S. (2003). Fueling education reform: Historically black colleges are meeting a national science imperative. Cell Biology Education, 2, 205–206. Swail, S.W. (2000). Preparing America’s disadvantaged for college: Programs that increase college opportunity. New Directions for Institutional Research, 107, 85–101. Swail, W. S., Cabrera, A. F., and Lee, C. (2004). Latino Youth and the Pathway to College. Washington, DC: Educational Policy Institute, Inc. Swail, S. W., Cabrera, F. A., Lee, C., & Williams, A. (2005). Latino students and the educational pipeline: Pathways to the bachelor's degree for Latino students: Educational Policy Institute. 269 Teranishi, R., & Briscoe, K. (2006). Social capital and the racial stratification of college opportunity. In J.C. Smart (Ed.), Higher education: Handbook of theory and research (Vol. XXI, pp. 591–614). Dordrecht, The Netherlands: Kluwer Academic Publishers. The Institute for College Access and Success (2007). Economic diversity of colleges. Retrieved January 11, 2008, from http://www.economicdiversity.org/ Tidball, M.E. (1986). Baccalaureate origins of recent natural science doctorates. Journal of Higher Education, 57(6), 606–620. Tienda, M. & Niu, S. X. (2006). Capitalizing on segregation, pretending neutrality: College admissions and the Texas top 10% law. American Law and Economics Review, 8(2), 312–346. Tierney, W.G. (2002). Cautionary tales: Evaluation and college preparation programs. In W.G. Tierney & L. Hagedorn (Eds.), Extending their reach: Strategies for increasing access to college. Albany, NY: SUNY. Tierney, W. G., & Venegas, K. M. (2006). Fictive kin and social capital: The role of peer groups in applying and paying for college. American Behavioral Scientist, 49(12), 1687–1702. Tinto, V. (1993). Leaving college: Rethinking the causes and cures of student attrition (2nd ed.). Chicago: University of Chicago Press. Titus, M. A. (2000, May). The financing of success in higher education: Student aid, expectations, and the persistence of first-time, full-time freshman. Paper presented at the Association for Institutional Research, Cincinnati, Ohio. Titus, M. A. (2007). Detecting selection bias, using propensity score matching, and estimating treatment effects: An application to the private returns to a master's degree. Research in Higher Education, 48(4), 487–521. Tomás Rivera Policy Institute. (2004). Caught in the financial aid information divide: A national survey of Latino perspectives on financial aid. Los Angeles, CA: Author. Tornatzky, L.G., Cutler, R., & Lee, J. (2002). College knowledge: What Latino parents need to know and why they don't know it. Los Angeles, CA: The Tomás Rivera Policy Institute. Tornatzky, L.G., Lee, J., Mejia, O., & Turant, S. (2003). College choices among Latinos: Issues of leaving home (Policy Brief). Los Angeles, CA: Tomás Rivera Policy Institute. Trent, W., & Hill, J. (1994). The contributions of historically black colleges and universities to the production of African American scientists and engineers. In W. Pearson, Jr., & A. Fechter (Eds.), Who will do science? Educating the next generation, (pp. 68–80). Baltimore, MD: The Johns Hopkins University Press. 270 Trent, W. T., Lee, H. S., & Owens-Nicholson, D. (2006). Perceptions of financial aid among students of color: Examining the role(s) of self-concept, locus of control, and expectations. American Behavioral Scientist, 49(12), 1739–1759. Tsapogas, J. (2004). The role of community college in the education of recent science and engineering graduates. (NSF 04–315). National Science Foundation: Arlington, VA. Tsapogas, J., & Cahalan, M. (1996, May). Incidence of and factors related to progression to graduate school among recent science and engineering bachelor's degree recipients: Results from a national study. Paper presented at the annual meeting of the Association of Institutional Research, Albuquerque, NM. University of California. (n.d.). Transfer Q&A: Campus selection practices. Retrieved March 30, 2008 from http://www.universityofcalifornia.edu/educators/counselors/adminfo/ transfer/advising/answers/practices.html University of California, Los Angeles, (2004). Academic Achievement Program Review. Los Angeles, CA. U.S. Census Bureau. (1990). 1990 Census of Population and Housing, Summary Tape File 3; Matrices generated by Lindsey Malcom using American Factfinder http://factfinder.census.gov. U.S. Census Bureau. (2004). U.S. interim projections by age, sex, race, and Hispanic origin. Retrieved July 15, 2005, from http://www.census.gov/ipc/www/usinterimproj/ U.S. Census Bureau. (2005) American Community Survey; Matrices generated by Lindsey Malcom using American Factfinder http://factfinder.census.gov. U.S. Census Bureau. (2007). American Community Survey; Matrices generated by Lindsey Malcom using American Factfinder http://factfinder.census.gov. U.S. Department of Education. (2006a). What Works Clearinghouse: Study design classification. Washington, DC: Institute of Education Sciences. U.S. Department of Education. (2006b). What Works Clearinghouse: Study review standards. Washington, DC: Institute of Education Sciences. U.S. Department of Education. (2007). Title V developing Hispanic-serving institutions program: Definition of an HSI. Retrieved January 13, 2008, from http://www.ed.gov/ programs/idueshsi/definition.html U.S. Department of Education. (2008, Winter). Loan forgiveness for public service employees. Retrieved April 17, 2008, from http://studentaid.ed.gov/students/attachments/ siteresources/LoanForgivenessMarch18.pdf 271 U.S. Department of Education. (n.d.). The National Science & Mathematics Access to Retain Talent grant (National SMART Grant). Retrieved April 17, 2008, from http://studentaid.ed.gov/PORTALSWebApp/students/english/SmartGrants.jsp Valenzuela, A. (1999). Subtractive schooling: U.S. Mexican youth and the politics of caring. Albany, NY: SUNY Press. Vara-Orta, F. (2007, January 31). Most Latino students spurn college loans. Los Angeles Times, p. A1. Venegas, K. M. (2006). Internet inequalities: Financial aid, the internet, and low-income students. American Behavioral Scientist, 49(12), 1652–1669. Venezia, A., Kirst, M., & Antonio, A. (2003). Betraying the college dream: How disconnected K–12 and postsecondary education system undermine student aspirations. Final Policy Report from Stanford University’s Bridge Project. Virginia Community College System. (2008). Transfer list. Retrieved April 10, 2008, from http://www.vccs.edu/Students/Transfer/tabid/106/Default.aspx Wagener, U., & Nettles, M.T. (1998). It takes a community to education students. Change, 30(2), 18–25. Wang, H. (2007, April). Combining Propensity Score Matching and Hierarchical Modeling to Evaluate the Effects of Ability Grouping. Paper presented at the Annual Meeting of the American Educational Research Association, Chicago, IL. Ware, N.C., & Lee, V.E. (1988). Sex differences choice of college science majors. American Educational Research Journal, 25(4), 593–614. Watson, L.W., & Kuh, G.D. (1996). The influence of dominant race environments on student involvement, perceptions, and educational gains: A look at historically Black and predominantly White liberal arts institutions. Journal of College Student Development, 37, 415–424. Weiler, W. C. (1991). The effect of undergraduate student loans on the decision to pursue postbaccalaureate study. Educational Evaluation and Policy Analysis, 13(3), 212–220. Weiler, W. C. (1994). Transition from consideration of a college to the decision to apply. Research in Higher Education, 35(6), 631–646. Wellman, J. V. (2002). State policy and community college-baccalaureate transfer. (National Center Report #02–6). San Jose, CA: National Center for Public Policy and Higher Education, and The Institute for Higher Education Policy. 272 Wenglinsky, H.H. (1996). The educational justification of historically Black colleges and universities: A policy response to the U.S. Supreme Court. Educational Evaluation and Policy Analysis, 18(1), 91–103. Wilder, G. Z., & Baydar, N. (1991). A study of decisions about graduate education: A two-year longitudinal study of GRE takers. Paper presented at the annual meeting of the American Educational Research Association, Chicago. Williams, J. (2006). The pedagogy of debt. College Literature, 33(4), 155–169. What Works Clearinghouse: Study design classification. (2006). U.S. Department of Education Institute of Education Sciences. What Works Clearinghouse: Study review standards. (2006). U.S. Department of Education Institute of Education Sciences. Whitt, E.J., Edison, M.I., Pascarella, E.T., Nora, A., & Terenzini, P.T. (1999). Women's Perceptions of a "Chilly Climate" and Cognitive Outcomes in College: Additional Evidence. Journal of College Student Development, 40(2), 163–177. Wolf-Wendel, L.E. (1998). Models of excellence: The baccalaureate origins of successful European American women, African American women, and Latinas. The Journal of Higher Education, 69(2), 141–186. Zarate, M.E., & Pachon, H.P. (2006). Perceptions of College Financial Aid Among California Latino Youth. Policy Brief. Los Angeles: Tomás Rivera Policy Institute. Retrieved January 28, 2007 from http://www.trpi.org/PDFs/Financial_Aid_Surveyfinal6302006.pdf

Abstract (if available)

Abstract This study contributes to our understanding of the institutional and financial aid pathways traversed by Latina/o bachelor's degree holders in fields related to science, technology, engineering, and mathematics (STEM), and the relationships between those pathways and selected outcomes of interest. Using the sociological framework of cumulative advantage/disadvantage and the statistical techniques of latent class analysis (LCA) and propensity score matching (PSM), data from the 2003 National Survey of Recent College Graduates (NSRCG) enhanced with institutional information drawn from the College Board Survey of Colleges and Universities, Barron's Selectivity Index, and the Integrated Postsecondary Education Data System (IPEDS) were analyzed to describe the means through which Latina/o STEM baccalaureates accessed those degrees, characterize their college financing strategies and to understand how indebtedness, measured relative to typical levels of borrowing within the B.S. degree-granting institutional context, influences these students' decisions to attend graduate school. The analytical sample reflects those Latinas/os who earned a bachelor's degree in a STEM field during the 2000-01 or 2001-02 academic years from a postsecondary institution in the mainland United States.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Learning from our global competitors: a comparative analysis of science, technology, engineering and mathematics (STEM) education pipelines in the United States, Mainland China and Taiwan

Asset Metadata

Creator Malcom, Lindsey Ellen (author)

Core Title Accumulating (dis)advantage? Institutional and financial aid pathways of Latino STEM baccalaureates

School Rossier School of Education

Degree Doctor of Philosophy

Degree Program Education

Publication Date 07/30/2008

Defense Date 05/08/2008

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag financial aid,Latinos,OAI-PMH Harvest,postsecondary education,STEM

Language English

Advisor Dowd, Alicia C. (committee chair), Bensimon, Estela M. (committee member), Pachon, Harry (committee member)

Creator Email malcom@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m1457

Unique identifier UC1100059

Identifier etd-Malcom-20080730 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-102568 (legacy record id),usctheses-m1457 (legacy record id)

Legacy Identifier etd-Malcom-20080730.pdf

Dmrecord 102568

Document Type Dissertation

Rights Malcom, Lindsey Ellen

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu