University of Southern California Dissertations and Theses

No place like home: a three paper dissertation on K-12 student homelessness & housing affordability

(USC Thesis Other)

No place like home: a three paper dissertation on K-12 student homelessness & housing affordability

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content No Place Like Home: A Three Paper Dissertation on K-12 Student Homelessness & Housing Affordability by Tasminda Kaur Dhaliwal A Dissertation Presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY URBAN EDUCATION POLICY August 2021 Copyright 2021 Tasminda Kaur Dhaliwal ii Dedication This dissertation is for the giants on whose shoulders I stand: Nani, Bebe, Nana, Dada, Dadi, Mum & Papa. iii Acknowledgements There are so many people that made up my village of family, friends, and mentors that helped me complete this dissertation. First and foremost, I must thank my family for their unwavering support and belief in me. Mum, Papa, Gavin, and Jasmin you never doubted my decision to do something crazy like leave a job with a good salary and retirement benefits to pursue a PhD with limited employment prospects. Your unwavering belief in me kept me going through all of the difficult times when I had lost faith in myself. Mum, you reminded me of whose blood I had running through my veins—the determination, bravery, and work ethic of my ancestors. Papa, you offered encouragement when I was down and always reminded me that I could do anything I wanted to do. Gavin, you helped me sort through my feelings in a methodological way, bringing reason and past evidence to conversations when I was spiraling with anxiety and fear. Jasmin, you were always there to build me up, make me laugh, and cheer me up when I was down (which often included hugs and ice cream). Achieving a PhD was only possible for me because of the hard work, sacrifice, and love of my extended family. Bebe and Nani and Dadi and Dada who raised me, believed in the possibilities, and sacrificed so much for my parents, siblings, and me. You are the giants whose shoulders I stand on. None of this would be possible without you. The staff, faculty and mentors at the Rossier School of Education also made it possible for me to write this dissertation. Laura Romero is one of the most amazing program administrators I have ever met. Her care and attentiveness to students is unparalleled and the PhD program is very fortunate to have her. Along with Laura, I want to thank the amazing program staff including John, Evan, Patrick and Alex. Being part of the #lastObamacohort was one of the greatest gifts of graduate school. All of my cohort mates (Adrian, Ashley, Christina, David, Martin, Neha, Neil, iv Liane, Sarah, Shira, and Theresa) greatly enriched my experience at USC, offering thoughtful perspectives in class and community. I feel blessed to be among such smart, generous, and fun colleagues. My colleagues on the 9 th floor deserve a special shout out, especially Shira, Sarah, Martin, Neha, Paul, Ayesha, and Taylor. Sarah, we spent many late night together in WPH. Shira, you were always an amazing listener, had the best advice, and the most generous. Martin, you were bar-none the best at gassing me up and reminding me that I’ve got this. Neha, your move into the office was such a treat. I loved our conversations and catch ups on work and life and have missed them dearly since you left. Ayesha, you have been such a generous mentor and all-around role model. Paul, thank you for your willingness to bounce around ideas, mull over methods questions, and be an econ homework collaborator. Taylor, I’ve enjoyed making ideas come to life in manuscripts with you. I also must thank Shira and Rachel who provided community and support that was vital to my final year in the PhD and being on the job market. Connecting with these scholars on a weekly basis kept me going through a long and, at times, lonely and anxiety-filled year. I could not have finished without them! I also must thank the faculty of the K-12 program. Julie Marsh was my first advisor at USC and someone who always treated me as a co-collaborator, even when I did not think my ideas were worth much. She continued to nurture me and serve as a trusted mentor even after I switched advisors. Adam Kho was such a wonderful addition to the program. I thoroughly enjoyed all of our conversations on the 9 th floor and so appreciated that your door was always open to me. Tricia Burch always made a point to check in with me and I appreciated her support throughout the program. Dave Quinn who brought me onto one of his projects, funded me for my work, and always was open to discussing my research ideas- thank you! v My dissertation would not have been possible without my advisor Katharine Strunk. Katharine’s high expectations made me a better scholar. She balanced tough love with a genuine care for me as a person. Katharine provided amazing feedback on earlier drafts of this dissertation (and anything I ever have her review) and asked tough questions when this dissertation was in its nascent stages. Her feedback is not for the faint of heart, but it always made my work stronger. She is someone who truly makes excellence a habit. I am thrilled that I will get to work with her at Michigan State University and have an opportunity to continue to learn from her. I look forward to having her as a trusted colleague, in addition to a trusted mentor. Gary Painter and Ann Owens were integral members of my dissertation committee and wonderful mentors and co-authors. Working with both of them allowed me to do what Jean Anyon (2014) suggests: “rethink what ‘counts’ as educational policy” (p. 3). They both provided great feedback on my dissertation papers and were thoughtful co-authors. Ann Owens provided extensive feedback on other parts of my dissertation and fellowship materials. I would like to thank the Michigan Department of Education and the Los Angeles Unified School District for providing the data used in this dissertation. The data from these districts tells the story of students and families living at the margins—those that are experiencing homelessness. I am humbled to share their stories of hardship and resolve in this dissertation. Similarly, I would like to thank the American Education Research Association and National Science Foundation Dissertation Grants program for funding this work. The ability to work exclusively on my own projects in this final year provided the space to dig deeper into my dissertation and research agenda. Finally, I would like to thank my husband, Erik. He has had a front-row seat to the PhD journey. There are too many times to count where Erik lifted me up, eased my anxieties, and helped me keep things in perspectives. You are my rock. vi Table of Contents Dedication ....................................................................................................................................... ii Acknowledgements ........................................................................................................................ iii List of Tables ................................................................................................................................ vii List of Figures .............................................................................................................................. viii Abstract .......................................................................................................................................... ix Introduction ..................................................................................................................................... 1 Paper 1: Putting Homelessness in Context: The Schools and Neighborhoods of Students Experiencing Homelessness ............................................................................................................ 6 Paper 2: Promoting Resiliency? The Role of Schools and Neighborhoods in Supporting the Success of Students Experiencing Homelessness ......................................................................... 25 Paper 3: A Lever for Improving Student Success? The Causal Effect of LIHTC on Underserved Students ......................................................................................................................................... 54 References ..................................................................................................................................... 85 Tables .......................................................................................................................................... 102 Figures ......................................................................................................................................... 119 Appendix ..................................................................................................................................... 127 vii List of Tables Table 1.1. School & Neighborhood Characteristics of Homeless & Nonhomeless Students .…102 Table 1.2. Prevalence and Direction of Mobility………………………………….…….………103 Table 2.1. Factor Loadings for School & Neighborhood Indices…………………………….…104 Table 2.2. Homeless and Non-homeless Student Characteristics………………………….……105 Table 2.3. Schools and Neighborhoods of Students Experiencing Homelessness………..……..106 Table 2.4. Relationship between Homelessness and Outcomes……………………..……..……107 Table 2.5. Relationship between School Resource Indices, Attendance, & Academic Achievement…………………………………………..…...………….……………108 Table 2.6. Relationship between Neighborhood Resource Indices, Attendance, & Academic Achievement…………………………………...…………...………….……………109 Table 3.1. Descriptive Statistics for Non-QCTs & QCTs…………………...…………….……110 Table 3.2. The Effect of QCT Status on LIHTC Investment…………………...……….………111 Table 3.3. The Effect of LIHTC Investment on Student Homelessness, Mobility, & Attendance… ……………………...……………………...………….……………112 Table 3.4. Title I & MEP Eligibility: The Effect of QCT Status on LIHTC Investment….……113 Table 3.5. Title I & MEP Eligibility: The Effect of LIHTC Investment on Student Homelessness, Mobility, and Attendance …………………………….……………114 Table 3.6. High Rent Burden Census Tracts (Above State Median) …………………….……..115 Table 3.7. Housing Effects for Non-Title I & MEP Eligible Students: The Effect of LIHTC Investment on Student Homelessness, Mobility, and Attendance.………116 Table 3.8. Spillover Effects for Non-Title I & MEP Eligible students: The Effect of LIHTC Investment on Student Homelessness, Mobility, and Attendance..……..117 Table 3.9. Composition Effects for Non-Title I & MEP Eligible Students: The Effect of LIHTC Investment on Student Homelessness, Mobility, and Attendance..……..118 viii List of Figures Figure 1.1. Distribution of Homeless Students in LAUSD Schools and Neighborhoods, 2016-2017………………………………………………………………………………………119 Figure 1.2. Comparing Average School and Neighborhood Disadvantage When Mobile and Not Mobile…………………………………………………………………..……………..120 Figure 2.1. Factors Related to Resilience………………………………….………….…………121 Figure 2.2. Summary of Promotive, Risk, Protective and Vulnerability Factors Across Outcomes………………………………………………………………………………….……122 Figure 3.1. Conceptual Model of How LIHTC Investments Impacts Student Outcomes………123 Figure 3.2. Probability of QCT Status on Running Variables…………………………………..124 Figure 3.3. Distribution of Running Variables………………………………………………….125 Figure 3.4. LIHTC Investment by Centered Binding Score…………………………………….126 ix Abstract Researchers have acknowledged for some time that housing matters a great deal for students’ performance in schools. I present three dissertation papers that examine how issues related to housing affordability impact students and whether current policy solutions create conditions that promote student learning. In the first paper, my co-authors and I examine the school and neighborhood contexts of homeless students in the Los Angeles Unified School District (LAUSD). To do so, we use a combination of descriptive analyses and geospatial techniques. In the second study, I examine whether schools and neighborhoods moderate the relationship between homelessness and academic and behavioral outcomes using a restricted state administrative dataset from Michigan and a fixed-effects model. In the third study, I examine the effects of the federal government’s largest affordable housing program, the Low-Income Housing Tax Credit (LIHTC) program, on student homelessness, mobility, and attendance. I leverage a discontinuity in the LIHTC funding formula and administrative educational data from Michigan, along with publicly available data, to estimate the causal effects of this program. Jointly, these studies elevate our understanding of how housing affordability impacts students by focusing on symptoms of housing affordability, student homelessness, and a housing solution, LIHTC. 1 Introduction Researchers have acknowledged for some time that factors outside the classroom matter a great deal for students’ performance in schools. From food to healthcare access, out-of-school factors create the conditions that enable (or constrain) learning in schools (Anyon, 2014; Amy E. Schwartz, Ellen, Voicu, & Schill, 2006). Housing is one of the most consequential factors for education. For the vast majority of students, where students live dictates where they attend school and the conditions of their housing and neighborhoods can influence how they perform in schools (see Leventhal & Newman (2010) and Newman & Holupka (2015) for reviews). One dimension of housing that has implications for education is affordability. Problems with housing affordability can lead to families living in substandard and crowded housing, and the financial burden of allocating a significant portion of monthly pay to housing can create stress and crowd out investment in children’s educational development (Leventhal & Newman, 2010; Newman & Holupka, 2015). Additionally, shortages in affordable housing have been a main contributor to the growth in family homelessness, which has implications for schools since they are mandated to identify and support students experiencing homelessness (McKinney-Vento Homeless Assistance Act, 2015). The growth in family homelessness has resulted in schools educating more homeless students than ever before (Institute for Children Poverty & Homelessness, 2019; United States Interagency Council on Homelessness, 2018b). Resolving the affordable housing shortage and its associated impacts on students will require evidence-based practices and policy solutions. Education leaders and policymakers searching for new ways to account for the affordability-related challenges their students are facing find little direction from the research base. For instance, while many studies have identified a large number of correlates of homelessness (see Miller, 2011 for a review), little attention is paid to where students 2 are located, how neighborhood and school resources could affect their educational outcomes, or to whether existing large-scale affordable housing programs reduce the incidences of homelessness. Housing policy solutions are also critical to solving affordability challenges. Most of the existing literature has focused on two policy solutions for affordable housing: vouchers and public housing (e.g., Chetty et al., 2016; Horn, Ellen, & Schwartz, 2014; Jacob, 2013). The influence of the federal government’s largest affordable housing development program, the Low-Income Housing Tax Credit (LIHTC), on students and schools has been underexamined. The development of affordable housing could benefit students by reducing homelessness and creating conditions that are important for academic success (e.g., increased stability and attendance). These outcomes, however, have not been considered by existing studies on LIHTC. I present three dissertation papers that are designed to address gaps in the literature on how issues related to housing affordability impact students and whether current policy solutions create conditions that promote student learning. The first two papers of this dissertation focus on students experiencing homelessness. In the first paper, my co-authors and I examine the school and neighborhood contexts of homeless students by asking: 1) What are the characteristics of homeless students’ schools and neighborhoods? 2) How are homeless students spatially distributed in LAUSD schools and communities? 3) How do the characteristics of students’ schools and neighborhoods change during and after becoming homeless? Using a rich student-level administrative panel (2008–09 to 2016–17 school years) from the country’s second largest school district, the Los Angeles Unified School District (LAUSD), and publicly available data on schools and neighborhoods, we use a combination of descriptive analyses and geospatial techniques to answer the above research questions. We find that students 3 experiencing homelessness tend to be clustered in lower achieving schools that enroll higher proportions of students of color, FRL eligible, and English Learner (EL) students, and they live in disadvantaged neighborhoods. Despite the goals of the McKinney-Vento Act, we find that homeless students have high rates of school and residential mobility, and these rates are higher in the years that they are homeless. In general, school mobility tends to result in moves to more advantaged schools, and this is even true for students who move while homeless. But while mobility to advantaged schools is consistent with McKinney-Vento’s emphasis on moves in the best interest of students, some research suggests that high levels of school mobility may disadvantage students, regardless of their destinations. We conclude with policy implications for the implementation of the McKinney-Vento Act. In the second paper, I examine whether these contexts have implications for students’ academic and attendance outcomes. I ask the following research question: 1) Do school and neighborhood resources serve as promotive or protective factors for academic and behavioral outcomes (i.e., math and ELA standardized assessment scores and school attendance) of students experiencing homelessness? I draw on nearly 11 million observations from approximately 2 million unique students enrolled in public K-12 Michigan schools from the 2011-2012 to 2017-2018 school years. I focus on four dimensions of school and neighborhood resources that are positively associated with achievement and attendance: 1) school peer advantage, 2) teacher qualifications and experience, 3) neighborhood advantage, and 4) neighborhood educational and occupational attainment. Using a fixed effects approach that compares students during periods when they are experiencing homelessness to times when they are housed, I first examine how being homeless is associated with attending schools with higher levels of these school and neighborhood resources. 4 When students are homeless, they attend schools and live in neighborhoods with similar resources to times when they are not homeless. I also find that school and neighborhood resources have varied implications for students’ attendance and academic outcomes. While teacher qualifications and experience and neighborhood advantage typically function as promotive or protective factors across outcomes, peer advantage and neighborhood educational and occupational attainment can operate as either promotive, protective, and risk factors. I conclude with a discussion of why these factors may have such varied impacts. Lastly, I turn to a specific policy intervention that has been used to increase the supply of affordable housing, the Low-Income Housing Tax Credit (LIHTC) program, and employ a quasi- experimental research design: 1) Does the development of affordable housing via LIHTC funds affect student homelessness, residential and school mobility, and attendance? 2) Does LIHTC have differential impacts for students who are also targeted by educational programs (e.g., Title I, Migrant Education Program)? Drawing on educational and housing data from the state of Michigan, I exploit a discontinuity in the LIHTC funding formula that results in additional tax credits for projects and, consequently, additional units being built in qualified census tracts (QCTs). I use the quasi-experimental variation created by this discontinuity to estimate the causal effects of the program on student homelessness (including those who live temporarily with others, “doubled up”, that HUD homeless counts exclude), residential and school mobility, and attendance. While I find that QCTs receive additional LIHTC investment, I do not find evidence of LIHTC’s effects on student homelessness, mobility, or attendance. Nor do I find LIHTC effects for students who are already considered underserved by federal educational programs (i.e., Title I, Migrant Education 5 Program). There does, however, appear to be small reductions in neighborhood and school mobility for students who are relatively more advantaged, which may be primarily driven by students who live in the same census block as LIHTC investment (a proxy for housing effects) or live in the vicinity of LIHTC investment (a proxy for spillover effects). My dissertation extends the existing research base in several ways. Paper 1 provides the first evidence of where students live and attend school and how these contexts change over time. The findings from this study can inform more targeted support practices that meet students where they are. Paper 2 provides new evidence on which factors promote student success and protect students experiencing homelessness from the deleterious effects of housing loss. Despite the potential for neighborhoods and schools to serve as protective factors, there remains little evidence examining whether this is or is not the case. Finally, Paper 3 provides among the first evidence of whether the federal government’s strongest tool for creating affordable housing affects student homelessness, mobility, and attendance. Affordable housing is critically important for ending homelessness, and this study will contribute to the evidence base on LIHTC, the largest affordable housing development program (United States Interagency Council on Homelessness, 2018a). Jointly, these studies elevate our understanding of how housing affordability impacts students by focusing on symptoms of housing affordability, student homelessness, and a housing solution, LIHTC. 6 Paper 1: Putting Homelessness in Context: The Schools and Neighborhoods of Students Experiencing Homelessness With Soledad De Gregorio, Ann Owens and Gary Painter The number of K–12 students experiencing homelessness has grown since the Great Recession. Recent federal counts find more than 1.3 million K–12 students experienced homelessness in the 2016–17 school year, a 7 percent increase from the 2014–15 school year (National Center for Homeless Education, 2019). Education data are uniquely positioned to capture trends in homelessness and housing insecurity because federal education policy, the McKinney-Vento Homeless Assistance Act (McKinney-Vento Act), requires schools to identify and support homeless students. The McKinney-Vento Act defines homelessness as lacking a fixed, adequate, and stable nighttime residence, including living doubled up (i.e., the far majority of K– 12 homeless students who are temporarily living with others due to housing loss or economic hardship), and provides some funds to support these students (McKinney-Vento Homeless Assistance Act, 2015; National Center for Homeless Education, 2019). Other agencies, such as the Department of Housing and Urban Development (HUD), have a more limited charter and, therefore, do not include doubled-up families in their homeless counts. The inclusion of doubled- up students (and their families) means education data provide a broader view of housing insecurity than other agencies. School districts around the country are concerned about rising homelessness for many reasons, including the direct negative consequences of homelessness on students’ academic and behavioral outcomes (e.g., Cowen 2017; De Gregorio et al. 2020; Fantuzzo et al. 2012; Rafferty and Shinn 1991). Policy-makers and advocates have long recognized the potential of schools to serve as a source of support and stability for homeless students by providing resources (e.g., free 7 and reduced price lunch [FRL] program) and social connection (Masten et al. 1997). The McKinney-Vento Act requires schools to remove barriers to school enrollment (e.g., proof of address, immunization records), provide transportation to and from school, promote stability in school enrollment, and, in cases where school mobility is desirable, ensure moves are in the best interest of students (McKinney-Vento Homeless Assistance Act, 2015). Despite recognition that schools may be especially important community institutions for homeless children, little existing research has systematically examined the types of schools that students experiencing homelessness attend, the communities in which they live, and how these change with transitions into and out of homelessness. Our research contributes to such knowledge by analyzing a rich student-level administrative panel from the country’s second largest school district, the Los Angeles Unified School District (LAUSD), and publicly available data on schools and neighborhoods from the 2008–09 to 2016–17 school years to address the following research questions: 1) What are the characteristics of homeless students’ schools and neighborhoods? 2) How are homeless students spatially distributed in LAUSD schools and communities? and 3) How do the characteristics of students’ schools and neighborhoods change during and after becoming homeless? We find that students experiencing homelessness tend to be clustered in lower achieving schools that enroll higher proportions of students of color, FRL eligible, and English Learner (EL) students, and they live in disadvantaged neighborhoods. Despite the goals of the McKinney-Vento Act, we find that homeless students have high rates of school and residential mobility, and these rates are higher in the years that they are homeless. In general, school mobility tends to result in moves to more advantaged schools, and this is even true for students who move while homeless. But while mobility to advantaged schools is consistent with McKinney-Vento’s emphasis on 8 moves in the best interest of students, some research suggests that high levels of school mobility may disadvantage students, regardless of their destinations. We conclude with policy implications for the implementation of the McKinney-Vento Act. Background Schools, neighborhoods, and homelessness Cowen (2017) uses data from Michigan to provide detailed information on where students who experience homelessness go to school. Cowen (2017) finds that homeless students attend schools with more marginalized groups than their nonhomeless peers. Homeless students attend schools with larger shares of students of color, FRL and special education (SPED) eligible students, EL students, and lower-achieving students than students who are not homeless. A number of studies have identified neighborhood characteristics associated with homelessness. However, none of these studies draws on education data, thus they exclude doubled- up students. Homelessness is positively related to median rent and is negatively related to the presence of low-cost rental housing, while the relationship between homelessness and other housing and income-related variables (e.g., rental tenure, residential mobility, poverty, and unemployment rates) is less consistent (Culhane, Lee, & Wachter, 1996; Fargo, Munley, Byrne, Montgomery, & Culhane, 2013; Quigley, Raphael, & Smolensky, 1999). The inconsistent correlations between homelessness and neighborhood characteristics are perhaps due to the immense heterogeneity in the homeless population (e.g., unaccompanied youth, homeless families, single homeless individuals) and because of differences in the unit of analysis (e.g., city vs. neighborhood characteristics). Existing evidence suggests that homelessness is higher in cities and clustered in select neighborhoods within them, including neighborhoods outside historic “skid-row” districts (Lee & 9 Price-Spratlen, 2004). We may expect that homeless students are less concentrated than other homeless populations because the majority are doubled-up in another persons’ home. As a result, homeless students may be less concentrated in areas with homeless services (i.e., “skid row”). In this study, we examine the characteristics of homeless students’ neighborhoods and schools. Previous research shows that the socioeconomic, network, and institutional resources available in one’s neighborhoods and schools matter for children’s academic, social, and health outcomes (Coleman et al., 1966; Jencks & Mayer, 1990; Sharkey & Faber, 2014). Homeless students are particularly at risk for negative academic and behavioral outcomes, so it is important to understand the available supports in their neighborhoods and schools. We also provide among the first evidence on whether students experiencing homelessness are clustered in schools, which could limit schools’ ability to provide support if the need is too great. School and residential mobility One of the goals of the McKinney-Vento Act is to promote school stability by allowing homeless students to attend their origin school even if they leave the school attendance boundary. Reducing school mobility is desirable because most studies find negative effects of changing schools on students, including studies of structural moves (i.e., changing schools due to the completion of all available grade levels, such as elementary to middle school transitions) and nonstructural moves (i.e., changing schools due to other reasons such as a change in residence) (see Welsh 2017). These negative effects appear to be larger for low-income students and students of color (Xu, Hannaway, and D’Souza 2009; Hanushek, Kain, and Rivkin 2004). Nonmobile students can also be indirectly harmed by attending high-mobility schools (E. A. Hanushek et al., 2004). Some studies find the initial negative effects of mobility eventually lead to positive benefits for students in the long run but these results are sensitive to the type of move made (i.e., structural 10 vs. nonstructural) and the timing of the move (i.e., mid-year vs. over the summer) (Swanson and Schneider 1999; Schwartz, Stiefel, and Cordes 2017). Studies examining moves to schools with higher test scores find mixed evidence of the effects of “upward” school mobility, which suggests that the benefits of upward mobility may not always be realized (Cordes, Schwartz, & Stiefel, 2019; Cordes, Schwartz, Stiefel, & Zabel, 2016). The effects of changing residences or neighborhoods on students’ outcomes are mixed. Moving to a more advantaged neighborhood is associated with improved academic and life outcomes (Chetty, Hendren, and Katz 2016; Burdick-Will et al. 2011), while moves to similar neighborhoods have no discernable effects on academic outcomes (e.g., Jacob 2004). Most studies, however, fail to disentangle the effect of changing neighborhoods or residences from the effects of changing schools. In one exception, Cordes, Schwartz, and Stiefel (2019) examine residential moves that do not include school moves. They find that short-distance residential moves result in higher academic achievement. Meanwhile, long-distance residential moves result in lower academic performance, perhaps due to changes in neighborhood social capital or longer school commute times. Overall, the literature on mobility suggests that McKinney-Vento’s emphasis on stability is a potentially important support for students experiencing homelessness. We contribute to the existing research by examining whether homelessness is associated with neighborhood and school mobility. Students entering a new school may have to adapt to a new curriculum and a new school environment with its own culture, processes, and expectations, which may pose transition costs (Kerbow, 1996; Rumberger, Larson, Ream, & Palardy, 1999). Similarly, a loss of school- and neighborhood-based social capital in the form of social ties and relationships may stunt participation and academic performance (Coleman, 1988). In addition to housing instability, school and neighborhood moves produce multiple instabilities that could 11 exaggerate the negative academic and behavioral outcomes for homeless students. Homeless student identification and district context LAUSD uses identification tools, staff training, and a system of monitoring to identify and support homeless students. Consistent with best practices, the district collects homeless status and nighttime residence through a student residency questionnaire that is distributed annually during school registration to parents/guardians. The questionnaire can be updated throughout the year if housing status changes. In addition, the district’s homeless education program trains teachers and administrators on how to identify and support homeless students. Each school also has a homeless liaison (typically a school counselor or attendance administrator) who undergoes additional training and is responsible for monitoring the school’s identification and support of homeless students (Gonzalez, 2016). The district’s homeless education program monitors school compliance and is accountable to the California Department of Education (CDE). LAUSD enrolls proportionally fewer homeless students than the overall county rate and other large urban school districts. In the 2016–17 school year, 3.3 percent of LAUSD students were homeless, compared to 4.7 percent of students in Los Angeles County school districts (CDE and LACOE, 2018). In New York City Public Schools, the nation’s largest district, and Chicago Public Schools, the third largest district, homelessness rates were 9.8 and 4.7 percent, respectively (U.S. Department of Education, 2020). Variation in student homeless rates likely reflect differences in the actual incidences of homelessness and differences in how well students are identified as homeless in the data. 1 1 There are a few potential reasons why LAUSD’s homeless rates differ from other districts in the county and across the country. LAUSD serves a unique demographic including more undocumented students, who are less likely to share their housing status, than other schools in the county and other districts. There are also local policies that make disclosing homeless status more appealing in other cities, such as New York City’s right to shelter mandate, which could explain between district differences. 12 Methods Data and measures We draw on a student-level administrative data panel from LAUSD that spans the 2008– 09 to 2016–17 school years. We observe 5.33 million student-year observations and 1.24 million unique K–12 students. The administrative dataset includes a homeless indicator for whether the student was identified as homeless at any point in the school year and whether the student was doubled-up or resided in another type of unstable housing (e.g., shelters, hotel/motel, car, unsheltered), which is collected through the student residency questionnaire described above. The dataset includes variables related to demographics (i.e., students’ race/ethnicity, FRL eligibility, and whether a student is an immigrant), learning needs (i.e., SPED and EL status), and measures of academic performance and behavior (standardized English/Language Arts [ELA] and math achievement, suspensions, and attendance rate). We generate measures of school mobility, which includes all structural and nonstructural school changes. Notably, the panel contains student addresses. Students with multiple addresses are assigned to the address they occupy for the longest duration in that school year and addresses are geocoded to census tracts. We supplement the administrative data with publicly available data from the CDE and LAUSD on school type (i.e., traditional public, magnet, SPED, and alternative school). Additionally, we use geocoded data on LAUSD school locations and district boundaries from the City of Los Angeles’ GeoHub. The American Community Survey (ACS) five year-estimates provides data on neighborhoods (operationalized as census tracts). 2 These data are used to create indicators of neighborhood concentrated disadvantage, which we describe in the following sections. 2 Because yearly estimates are not available at the census tract level, we assign each ACS five-year estimate range to its midpoint year. For example, the ACS 2013 five-year estimates (i.e., reflect estimates from 1/01/2009 to 12/31/2013) are assigned to the 2010–2011 school year. 13 Student sample On average, 2.1 percent of students in LAUSD are identified as homeless each year between 2008–09 and 2016–17. The vast majority of students experiencing homelessness (81 percent) are observed as homeless in three or fewer consecutive years. We only observe about 8 percent of homeless students exit homelessness and re-enter in a later year. Throughout this study, we draw comparisons between the schools and neighborhoods of homeless and nonhomeless students. Consistent with existing studies, results from t-tests show that, in any given year, homeless students differ from nonhomeless students along a number of dimensions: homeless students are disproportionally Black (19 vs. 9 percent), FRL eligible (94 vs. 79 percent), mobile between schools (38 vs. 28 percent) and addresses (46 vs. 21 percent), and have lower academic performance (–0.3 SD). We also find that homeless students are more likely to exit the district (17 vs. 15 percent) and be immigrants (14 vs. 10 percent), two findings that are new additions to the literature. When we compare doubled-up homeless students to other homeless students, we find a smaller proportion of doubled-up students are Black (18 percent vs. 20 percent), and a greater proportion are Latinx (77 percent vs. 73 percent), or immigrants (15 percent vs. 13 percent). Doubled-up students are also less likely to exit the district than other homeless students (16 vs. 19 percent). See Table A1.1 in the online appendix for complete results. School and neighborhood disadvantage Because many of the school and neighborhood variables are highly correlated, we create indices of school and neighborhood disadvantage. We create a school-level index within each year using exploratory factor analysis of the following variables: percent of underrepresented students of color (Native American, Black, Filipino, Latinx, and Pacific Islander), FRL students, EL students, and SPED students. We retain one factor and create a weighted sum using standardized 14 beta weights for each variable in the factor score (Acock, 2013). Supplemental Table A1.2 shows the rotated factor loadings for the 2016–17 school year; factor loadings and standardized beta weights are consistent across school years. Because student composition shapes the economic, social, and cultural resources available (e.g., funding, teacher quality), the school concentrated disadvantage index is interpreted as a proxy for the level of material and immaterial resources present at the school (Owens & Candipan, 2020). The index scores have a mean of 0 and a SD of 1 in each year. We use the same process to create an index of neighborhood concentrated disadvantage. A student’s neighborhood of residence is the single tract in which the student resides. The index consists of the following variables informed by the literature (e.g., Owens 2010; Sastry 2012): percent female-headed households, employed, people and families whose income in the past 12 months is below the poverty level, Black residents, high school graduates, college graduates, employed adults with professional occupations, median log household income, severe rent burden (50 percent or more of income on rent), and severe overcrowding (more than 1.5 occupants per room) (Appendix Table A1.2, bottom panel, provides rotated factor loadings). The level of neighborhood disadvantage is a proxy for the neighborhood’s available economic, institutional, and relational resources. Analysis plan First question. We compare homeless (including those who are doubled up) and nonhomeless students’ school and neighborhood characteristics, within a given year, using t-tests. School characteristics include: peer demographics (race/ethnicity, FRL eligibility, EL, homeless status, SPED eligibility, and whether they are an immigrant), behavior (suspension and attendance rates), school mobility rate, academic achievement (standardized ELA and math achievement), 15 school type (traditional, magnet, SPED, and alternative), and the school disadvantage index. We also test differences among the neighborhood-level variables that compose the neighborhood disadvantage index listed above along with the index itself. Second question. We examine the spatial distribution of homeless students across schools in several ways. First, we map school locations, varying the size of the school marker by the proportion of homeless students in the 2016–17 school year. To examine the correspondence between neighborhoods and schools, we overlay school locations on census tracts shaded by the neighborhood disadvantage index in 2016. Second, we estimate a homeless isolation index, ! = #$% ! ! " &% ! ! # ! &', for the district to measure whether homeless students are overrepresented in some schools. 3 I describes the average proportion of homeless students in a homeless student’s school— the extent to which homeless students are “exposed” to other homeless students. An I of 1 indicates exposure only to other homeless students and 0 indicates exposure only to nonhomeless students. Third question. We examine trends in students’ school and neighborhood disadvantage before, during, and after homelessness using descriptive statistics and a fixed effects model. For these analyses, we limit our sample to homeless students. For students who exit and re-enter homelessness, we use the first homeless incidence. 4 Additionally, we examine neighborhood characteristics only for doubled-up students. We do so because interpreting the neighborhood characteristics of other homeless students, such as shelter users, is complicated by the fact that the constraints and preferences of service providers govern siting decisions. Focusing on doubled-up 3 The isolation index, I, is estimated as ! = #[(& "/$ )(& "/% ! )] where & " is the number of homeless students in school i , ) is the number of homeless students in the district, and * " is the number of students enrolled in school i (Massey & Denton, 1988). 4 Only 8 percent of homeless students exit homelessness and re-enter in a later year, thus, we retain complete data for the majority of homeless students. 16 students means that we can detect changes for families that are not “pulled” into neighborhoods by the siting decisions of service providers. First, we generate descriptive statistics on the frequency of school and residential address moves in the three time periods (years before, during, and after homelessness). Within each time period, we examine the extent to which students move to schools or neighborhoods in higher, lower, or relatively similar concentrated disadvantage deciles. Next, we examine the mobility patterns of homeless students using the fixed effects model: ()*!+,-. $# =/ % +/ ' 1 (# +/ ) 2 (# +/ * 32 (# +/ + 1 (# ∗2 (# +/ , 1 (# ∗32 (# +θ ( +6 - + 7 ($# (2) Where ()*!+,-. is the concentrated disadvantage index for either school or neighborhood s in time t. These outcomes are estimated as a function of 1 (# , which equals 1 if the student moved to a different school in year t compared to t-1 or if a student lived at a different address in year t compared to t-1, when estimating school and neighborhood disadvantage, respectively. 2 (# is an indicator for whether student i is homeless at time t, and 32 (# is an indicator for whether student i is no longer homeless at time t (i.e., students have exited homelessness). The years before a student’s first homeless incidence serves as the reference category. Interacting the mobility indicator with each period indicator (1 (# ∗2 (# and 1 (# ∗32 (# ) measures whether students move to different types of neighborhoods or schools during and after homelessness, versus before. / + captures the marginal difference in school or neighborhood concentrated disadvantage for moves made while homeless, while / , captures the marginal difference for moves after exiting homelessness. The model includes student fixed effects, : ( , to account for time-invariant student characteristics that predict neighborhood or school disadvantage (e.g., race/ethnicity) and grade fixed effects, 6 - , to account for the fact that middle and high schools have higher concentrated disadvantage than elementary schools. Standard errors, 7 ($# , are clustered 17 at the student level. For ease of interpretation, we generate predicted means from this model and report results from significance tests. The results from this regression model describe homeless students’ schools and neighborhoods. These results provide useful evidence of the types of schools and neighborhoods that homeless students experience and how the student’s trajectory into and out of homelessness relates to where they live and attend school. However, we do not interpret these results as causal evidence of how homelessness impacts where students go to school and where they live. Results School and neighborhood characteristics of homeless students In Table 1.1, we show that the schools and neighborhoods of students experiencing homelessness are more disadvantaged than nonhomeless students’ along a number of dimensions. Because all differences are statistically significant, we call attention to those that are largest in magnitude, although many of these differences are modest. Panel A shows that homeless students attend schools with more students of color (93 vs. 86 percent), FRL students (86 vs. 79 percent), EL students (34 vs. 29 percent), and other homeless students (5 vs. 2 percent). Mean achievement is 0.2 to 0.3 SD lower in homeless students’ schools, and homeless students are more likely to attend traditional public schools (85 vs. 81 percent) and less likely to attend magnet schools (2 vs. 4 percent). Panel B shows that homeless students live in neighborhoods with higher poverty rates (25 vs. 21 percent), more Black residents (12 vs. 9 percent), and lower median income (by $4,942). It follows that homeless students attend schools and live in neighborhoods with higher concentrated disadvantage (0.3 SD higher) (school and neighborhood disadvantage are correlated at r=0.67). These differences are broadly similar when comparing the school and neighborhood contexts of doubled-up and nonhomeless students. 18 Spatial distribution of homeless students Geospatial analysis shows that students are concentrated in several clusters. Figure 1.1 maps each school with a circle proportional to its school homelessness rate. Census tracts within the LAUSD boundary are color coded based on the concentrated neighborhood disadvantage level. Schools enrolling substantial shares of homeless students tend to be clustered geographically and tend to be clustered in neighborhoods with greater disadvantage. 5 For example, box A captures census tracts with the highest levels of disadvantage in the district, those in downtown and south central LA, that also have schools enrolling larger proportions of homeless students. Yet there are exceptions to these associations. For example, schools located in relatively less disadvantaged communities, such as those in the west and the north-west areas of the district (see boxes B and C), enroll considerable proportions of homeless students. There are also examples of schools located relatively close to each other with varying proportions of homeless students, which may be an artifact of school attendance boundaries that pull from different neighborhoods (see box D). Roughly half of the district’s schools (47 percent) enroll about the same or more homeless students than the district as a whole in 2016–17 (3.3 percent), and almost a quarter of schools have homeless rates above 5 percent. We present the share of schools by homeless student enrollment rates in detail in Table A1.3 of the appendix. The fact that a relatively large number of schools have homeless rates well above the district average raises concerns about whether these schools have sufficient resources to support students adequately. The isolation measures confirm that homeless students experience some isolation, but they are less isolated than racial groups within the district. In 2016–17, the homeless isolation index was 0.05, which means that the average homeless student attends a school that is 5 percent 5 School homeless rates are correlated with school concentrated disadvantage at r=0.48 and neighborhood concentration disadvantage at r=0.37. 19 homeless—if there were no segregation, this figure would be 3.3 percent, the district rate. Comparisons to Black-white or Latinx-white isolation reveal that homeless students are far less isolated than Black and Latinx students (I=0.73 and 0.93, respectively). Isolation measures are sensitive to district composition, so a lower isolation index for students experiencing homelessness is expected, given their smaller population compared to Black or Latinx students. Changes in schools and neighborhoods for homeless students Finally, we explore how changes in homeless status are associated with neighborhood and school mobility and changes in disadvantage. Table 1.2 shows how frequently students experiencing homelessness move between schools (panel A) and how frequently doubled-up students move between addresses (panel B). Even before becoming homeless, students have high rates of school and residential mobility—about one-third changed schools or addresses. In contrast, roughly a quarter of nonhomeless students change schools or addresses during their entire schooling in LAUSD. While experiencing homelessness, mobility nearly doubles, with roughly 60 percent of students changing schools and addresses. In years after homelessness, students experience their lowest school and address mobility rates (27 and 16 percent, respectively). Aside from the fact that students exit homelessness to stable housing, lower mobility rates after homelessness may be due to the fact that roughly 30 percent of homeless students exit the district (and our data panel) while homeless. Sample attrition could influence our findings if those who exit are different from the students who remain. When we compare exiting and remaining homeless students (see Table A1.4 of the appendix), exiting students are more mobile, as evidenced by higher school and address mobility, than those who remain in the district. Consequently, we may observe mobility decreasing after homelessness because the students remaining in our sample are generally more stable. However, attrition does not seem to account 20 for differences in the distribution of moves (upward, similar, downward), as exiting and remaining students exhibit remarkably similar patterns before students exit. 6 Homeless students attend more disadvantaged schools (0.2 to 0.4) and live in more disadvantaged neighborhoods (0.7 to 0.8) than the average LAUSD school or neighborhood, although school and neighborhood disadvantage are the lowest after they exit homelessness. To assess more meaningful contextual changes, we compare the share of students that experience a significant upward move (2 or more decile reduction in disadvantage), similar move (same decile of disadvantage), and a significant downward move (2 or more decile increase in disadvantage). Regardless of homeless status, more than a third of school moves are significant upward moves, while a smaller share of moves is to similar schools (17 to 19 percent) or significantly downward moves (15 to 18 percent). 7 Our findings for homeless students are consistent with research showing that school mobility more often leads to significant upward than downward moves in school quality (Cordes et al. 2016). However, students are most likely to move to similar neighborhoods (45 to 50 percent). We test whether differences detected in the descriptive data are statistically significant using a fixed effects model. Predicted means from the fixed effects model are displayed graphically in Figure 1.2. 8 Starting with school mobility (top panel), when students change schools they move to less disadvantaged schools, regardless of homeless status (p<.001). Contrary to moving to disadvantage when homeless, we find that mobility results in similar (upward) moves during homelessness as before and after homelessness. While school mobility is upward, homeless 6 Results available upon request. 7 The predominance of upwards moves may suggest a “floor” effect where students move upwards because they are already located in the most disadvantaged schools. While homeless students are disproportionately located in more disadvantaged schools, these distributions are not skewed enough to guarantee an upwards move and therefore do not seem to account for our results (see Table A1.5 in online appendix). 8 Regression results are available in online appendix Table A1.6. 21 students still attend schools that are roughly 0.3 SDs more disadvantaged than the average school, so gains made by mobility are relatively small. Doubled-up students make similar residential moves before and during homelessness, and they move to slightly more disadvantaged neighborhoods after exiting homelessness (p<.01) (bottom panel). Changes in neighborhood disadvantage after exiting homelessness, however, are marginal—0.02 SDs more disadvantaged. Overall, doubled-up students move among neighborhoods that are much more disadvantaged than the district average. 9 Discussion Our results deepen the research community’s understanding of K–12 homeless students by making several contributions: 1) We describe the schools and neighborhoods of students experiencing homelessness; 2) We assess the spatial clustering of students in schools; and 3) We document homeless students’ residential and school mobility and explore whether their contexts change during and after homelessness. Overall, our findings show that, compared to nonhomeless students, students experiencing homelessness are clustered in lower achieving schools that enroll higher proportions of students of color, FRL, and EL students, and they live in neighborhoods with higher concentrated disadvantage. We find that homeless students are a highly mobile population and, despite the goals of the McKinney-Vento Act, residential and school mobility is higher in the years they experience homelessness. That said, school mobility tends to be upward, even during homelessness, consistent with McKinney-Vento’s emphasis on facilitating moves that are in students’ best interest. Whether this upward mobility is beneficial for students is unclear. The school mobility literature suggests moves are generally detrimental, but some studies find benefits of moving to significantly higher- 9 The influence of sample attrition on these results is unclear since those who leave our sample tend to live in schools and neighborhoods that are generally similar to those who remain (see online appendix Table A1.4). 22 quality schools. The types of schools that homeless students move to in our study are still very disadvantaged compared to the average district school, which may portend little benefit or even harm for students’ academic outcomes. Advocates have described the potential for schools to be a place of “refuge” for students experiencing homelessness (National Association for the Education of Homeless Children and Youth, 2010). However, homeless students’ schools serve a large number of other homeless and low-resourced students, and these schools are located in disadvantaged neighborhoods. If schools are overwhelmed by serving higher concentrations of marginalized student groups, this may limit their ability to serve as sources of support for homeless students. Future research should examine schools that disproportionately serve homeless students across other dimensions including those related to teacher quality, funding, and school climate. Our results show that students experience multiple instabilities when homeless, including changes in schools and neighborhoods. McKinney-Vento is unsuccessful in maintaining enrollment in origin schools for the majority of homeless students in our study. Qualitative research is needed to understand how homeless families understand and use McKinney-Vento. For example, future research should address questions such as: What services provided by schools do homeless families access? How do homeless families understand McKinney-Vento’s school mobility protections? Why do homeless families change schools and what guidance do families receive when they move schools? The answers to these questions could illuminate why McKinney- Vento falls short of ensuring stability for students experiencing homelessness. Implications Our results suggest a few implications for school districts and practitioners. First, the identification of homeless students could be improved. Variation in student homelessness rates 23 between districts (and schools) that educate students with similar demographics could mean some districts (and schools) are better at identifying homeless students than others. The McKinney- Vento Act does not specify how districts should identify students, but best practices suggest that a system of tools (e.g., a questionnaire), training, and monitoring is useful. Additionally, homeless liaisons should frequently and proactively confirm residency status, rather than waiting for families to do it. Because our results suggest higher mobility while homeless, districts should ask students to update their residency status each time a student moves schools or changes addresses. Additionally, students may not feel comfortable disclosing their homeless status because they lack caring and trusting relationships with school personnel and a positive school racial climate (Edwards, 2020). Aside from technical fixes in the identification process, building caring relationships and attending to racial disparities in academic expectations and discipline could also improve identification. Second, because homeless students are more likely to change schools or exit the district when homeless, administrators should ensure that families know their rights to remain in their school via information campaigns, follow-up after McKinney-Vento screening, and connections with transportation. School transfers of homeless students should be flagged for district and school homeless liaisons so that they can follow-up with families to make sure they are aware of their rights. When school transfers are requested by families, district and school staff should work with families to make sure the move is in the best interest of the student. Third, the increased likelihood of homeless students exiting the district calls for a regional approach to identifying and supporting homeless students. County offices of education (or their equivalent) could assist in forming data-sharing agreements and collaboratives to help track and share educational records for homeless students who move within and between districts. 24 Researchers should also make use of state-level or regional datasets to formulate a more comprehensive picture of how frequently and how long students are homeless. Finally, the clustering of homeless students in schools provides an opportunity to concentrate support services in schools and neighborhoods. Schools that serve high numbers of homeless students should receive McKinney-Vento funds that commensurate with their level of need. There is also an opportunity for schools that disproportionately serve homeless students to develop and share best practices in identifying and supporting homeless students. 25 Paper 2: Promoting Resiliency? The Role of Schools and Neighborhoods in Supporting the Success of Students Experiencing Homelessness 10 Homelessness is largely associated with negative academic and behavioral outcomes for K12 students. For example, students experiencing homelessness have lower academic achievement (Cowen, 2017; De Gregorio, Dhaliwal, Owens, & Painter, 2020; Fantuzzo et al., 2012; Herbers et al., 2012), lower attendance (Cowen, 2017; Fantuzzo et al., 2012), and increased likelihood of grade retention (Rafferty & Shinn, 1991) and dropping out of high school (Masten et al., 1997). But there is also variation within these outcomes. Some students continue to achieve academically despite experiencing homelessness, including performing above average on standardized assessment (Obradović et al., 2009) and graduating from high school (Edwards, 2019). Students like these are labeled as “resilient” because they show patterns of normal social and academic development despite living in adverse circumstances, including homelessness, abuse, and neglect (Masten, 2001; Masten, Cutuli, Herbers, & Reed, 2009). Given the documented challenges that come with homelessness for many students, it is useful to understand why some students experiencing homelessness show resilience. Existing studies find that resiliency is supported by promotive (i.e., factors generally beneficial for all students) and protective factors (i.e., factors that are especially beneficial when students are experiencing adversity), while resiliency is impeded by risk (i.e., factors generally detrimental for all students) 10 This research result used data structured and maintained by the MERI-Michigan Education Data Center (MEDC). MEDC data is modified for analysis purposes using rules governed by MEDC and are not identical to those data collected and maintained by the Michigan Department of Education (MDE) and/or Michigan’s Center for Educational Performance and Information (CEPI). Results, information and opinions solely represent the analysis, information and opinions of the author(s) and are not endorsed by, or reflect the views or positions of, grantors, MDE and CEPI or any employee thereof. This research was supported by a grant from the American Educational Research Association which receives funds for its "AERA Grants Program" from the National Science Foundation under NSF award NSF-DRL #1749275. Opinions reflect those of the author and do not necessarily reflect those AERA or NSF. 26 and vulnerability factors (i.e., factors that are especially detrimental when students are experiencing adversity). Although factors that support resiliency are thought to exist within schools and communities, up until this point empirical studies have only identified individual and familial characteristics that support resiliency, such as a child’s executive functioning or emotional regulation, or whether their caregiver engages in responsive parenting (Buckner, Mezzacappa, & Beardslee, 2009; Cutuli & Herbers, 2014; Masten et al., 2012; Miliotis, Sesma, & Masten, 1999). The large literatures on neighborhood and school effects suggests that these contexts matter a great deal for students’ educational outcomes (see Johnson, Jr., 2012 and Sastry, 2012 for reviews), including when students experience adversity. In particular, school resources related to teacher quality and peers could be important factors for resilience, as could resources related neighborhood composition and educational and occupational attainment (e.g., Chetty, Friedman, & Rockoff, 2014b; Coleman, 1988; Jencks & Mayer, 1990; Sacerdote, 2011). In this study, I examine whether school and neighborhood resources influence the academic and behavioral outcomes of students experiencing homelessness in the state of Michigan. I ask: Do school and neighborhood resources operate as factors related to resiliency (i.e., promotive, risk, protective, and/or vulnerability factors) for the attendance and achievement of students experiencing homelessness? To answer this question, I draw on a seven-year data panel (2011-2012 to 2017- 2018 school year) of students enrolled in public K-12 Michigan schools and I focus on four dimensions of school and neighborhood resources that might matter for resilience: 1) school peer advantage, 2) teacher qualifications and experience, 3) neighborhood advantage, and 4) neighborhood educational and occupational attainment. I apply a series of student and year fixed effect regressions to this data to capture whether school and neighborhood resources support (or 27 undermine) resiliency. I also descriptively explore the school and neighborhood resources of students experiencing homelessness to identify the types of resources that students have access to. Overall, I find school and neighborhood resources have varied relationships with student outcomes. Depending on the outcome examined, at times they function as promotive and protective factors and less frequently they serve as risk and vulnerability factors. Beginning with the factors that benefit educational outcomes, I find neighborhood advantage is a promotive and protective factor for student attendance and a promotive factor for math achievement. Students also live in more advantaged neighborhoods in years they experience homelessness, perhaps because shelter or temporary housing is located in these communities. Additionally, peer advantage is a promotive and protective factor for student attendance, but I also find that students attend schools with slightly fewer advantaged peers in years they are homeless. Finally, teacher qualifications and experience and neighborhood educational and occupational attainment are promotive factors for academic achievement; although, when homeless, students attend schools with slightly lower levels of qualified and experienced teachers and they live in neighborhoods with slightly higher levels of educational and occupational attainment. The evidence of risk and vulnerability factors is scant, with nominal negative associations that are not practically significant. In what follows, I describe the conceptual framework and background that motivates this study and guides variable selection. Next, I describe the data, how I generate the school and neighborhood measures, and the analytical methods I use to detect whether these measures support the educational resilience of students experiencing homelessness. Finally, I present the results and I conclude by describing the limitations of this study and discussing the results. Conceptual Framework & Background 28 The existing evidence shows that homelessness is negatively associated with students’ educational outcomes, including academic achievement (Cowen, 2017; De Gregorio et al., 2020; Fantuzzo et al., 2012), attendance (Cowen, 2017; De Gregorio et al., 2020; Fantuzzo et al., 2012), graduating from high school (Masten et al., 1997), and being promoted to the next grade level (Rafferty & Shinn, 1991). We also know of some of the mechanisms that drive these associations, including the school instability produced by homelessness (Fantuzzo et al., 2012) and heightened levels of stress for parents and children (Masten, Miliotis, Graham-Bermann, Ramirez, & Neemann, 1993). Here, I introduce a conceptual framework and related literature on the factors that support resiliency in educational outcomes, which might mitigate some of the effects described above. I begin by describing the processes through which resiliency operates, including how concepts related to resiliency are defined and which factors have been identified for the resilience of students experiencing homelessness. Drawing on the neighborhood and school effects literature, I then provide background on which school and neighborhood resources may promote resiliency for students impacted by housing loss. Resilience Processes and Factors While adversity is hard for all students, some are more affected than others. Students who show patterns of positive development during significant adversity are labeled as resilient (Masten et al., 2009). Positive development is defined as positive social behaviors and academic achievement, and significant adversity includes things like abuse, extreme poverty, and homelessness. Resilient students are thus students that continue to be successful in school despite the fact that their adverse circumstances are generally negatively associated with behavioral and academic outcomes. The factors that support resiliency for children living in challenging 29 circumstances are typical every day, routine occurrences (e.g., healthy relationships with friends, responsive parenting from caregivers)—what Masten (2001) refers to as “ordinary magic”. As shown in Figure 2.1, research on resilience has identified four types of factors that are related to positive development: 1) promotive factors, 2) risk factors, 3) protective factors, and 4) vulnerability factors. Promotive and risk factors are those that relate to child development regardless of whether children are experiencing adversity (Masten et al., 2009). Promotive factors are assets that are generally beneficial for the positive development of children, not just those facing adversity. Risk factors do the opposite and are defined as characteristics that are generally detrimental to the positive development of all children. Protective and vulnerability factors are similar to promotive and risk factors, except that their associations are heightened when students experience adversity. Protective factors have a greater benefit for children facing adversity, while vulnerability factors are a greater detriment for children facing adversity (Masten et al., 2009). There could be cases where factors are both promotive and protective, meaning, that they are generally beneficial for all children and even more beneficial when students experience adversity. The same could be the case for factors that are both risk and vulnerability factors. Most documented examples of these factors are individual or familial characteristics. For instance, executive functioning (Cutuli & Herbers, 2014), ability to control emotions and behaviors (Buckner, Mezzacappa, & Beardslee, 2003; Buckner et al., 2009; Obradović, 2010), and responsive parenting (Miliotis et al., 1999) have all been identified as promotive and/or protective factors, including for students experiencing homelessness. Meanwhile, parent and student substance abuse and mental illness are examples of risk and/or vulnerability factors (Masten et al., 2009), including for students experiencing homelessness. 30 Yet, there are important factors outside of the self or family that may also be important for resilience. These are particularly important to identify because they may allow for non-familial actors to supply the necessary supports to help students facing adversity succeed, especially during times when parents and families may have less time and capacity to supply these supports themselves (e.g., when families are experiencing housing instability). Schools and neighborhoods may be even more important for supporting students in times of adversity because they provide environments of continued interaction for students. Research on resilience has acknowledged the importance of these environments but, to the best of my knowledge, there is no empirical evidence examining whether aspects of these environments support resilience. To understand which aspects of schools and neighborhoods might matter most, I turn to the research on how schools and neighborhoods shape student success next. School-based Factors for Resilience Schools provide important resources that shape student behavior and academic achievement, and these could also promote resiliency for students experiencing homelessness. I focus on two school-based resources that could impact student outcomes as promotive and/or protective factors: 1) quality and experience of their teachers (e.g., Chetty Raj, Friedman John, 2013; Jackson, Rockoff, & Staiger, 2014), and 2) peer composition (e.g., E. A. Hanushek, Kain, Markman, & Rivkin, 2003; Sacerdote, 2011). Teacher quality is one of the most important in-school factors shaping achievement. How effective educators are in teaching content, creating supportive learning environments, and developing relationships with students has direct implications for students’ achievement, well- being, and later life outcomes (Chetty et al., 2010, 2014b; Rucinski, Brown, & Downer, 2017). However, there remains intense debate over how to measure quality and which teacher 31 characteristics matter most for student performance. The most commonly used measure of teacher quality within education research is teachers’ contributions to student achievement growth, or value-added measures (VAMs) (Chetty, Friedman, & Rockoff, 2014a). Aside from VAMs, experience level is one of the most consistent predictors of academic achievement with some evidence suggesting that teachers improve the most early in their careers while other research suggests they continue to improve as they approach ten years of experience (Clotfelter, Ladd, & Vigdor, 2007; E. Hanushek, Kain, O’Brien, & Rivkin, 2005; Kraft, Papay, & Chi, 2020). There is mixed evidence about the extent to which other measures, like teacher evaluation ratings, teacher credential, and education level, imply teacher quality (Clotfelter et al., 2007; Grossman, Cohen, Ronfeldt, & Brown, 2014; E. Hanushek et al., 2005; Kraft et al., 2020). Regardless of how teacher quality is measured, there is a consensus that teachers matter in a myriad of ways for student success (Chetty et al., 2010, 2014b; Rucinski et al., 2017). Students’ peers are also important school-based resources. The peer effects literature finds that students can benefit academically if their peers are higher achieving (Hanushek et al., 2003, see Alexander & Eckland (1975) for the reverse), higher income (H. Schwartz, 2010), and less mobile (E. A. Hanushek et al., 2004). The influence of peers operates through “externalities that spill over from peers’ actions or family background” (Sacerdote, 2011, p. 250). As Sacerdote (2011) describes, attending school with more advantaged peers could benefit achievement through a variety of mechanisms, including through the actions of peers and their parents. For example, peers could motivate students to achieve by setting norms around achievement. The actions of peers could also have spillovers to teachers who may adjust their expectations of students based on these norms (Jencks & Mayer, 1990; Rumberger & Palardy, 2005). The actions of parents could also improve teacher and leader quality, for instance, if parents push for teacher accountability or 32 if they are involved in recruiting a highly effective school leader. Aside from peer effects, peer composition is also associated with how well-resourced schools are. Schools that serve more advantaged student groups tend to have more financial resources, higher quality teachers, and more rigorous curriculums and course offerings (e.g., Owens & Candipan, 2020). In summary, peer composition can benefit students through peers’ spillover effects on school culture, teacher expectations, and the quality of teachers and leaders. Additionally, peer composition can also be positively related to educational outcomes because schools with more advantaged peers have more financial resources, more rigorous curriculum and more extensive course offerings—all of which are important for student learning (e.g., Jackson, Johnson, & Persico, 2015; Warne, 2017). During a period of extreme adversity, teacher quality and peer composition could function as important protective factors for students like those experiencing homelessness. The consistency of schools, teachers, and peers could provide students experiencing homelessness with stability at a time of intense instability in their lives. Federal policy recognizes the importance of schools as a source of support for homeless students and offers policy protections that remove barriers to maintaining school enrollment (although students still change schools frequently when experiencing homelessness, see Dhaliwal, De Gregorio, Owens, & Painter, 2021). Teachers and peers could buffer students from the adversity of their circumstances by providing care, social connection, and immaterial and material support to students and their families. Thus, teachers and peers may matter even more for students experiencing homelessness, however, existing research has not empirically examined whether this is the case. Neighborhood-based Factors for Resilience Aside from schools, neighborhoods provide consequential resources that matter for students’ educational outcomes and these resources could also promote resiliency for students 33 experiencing homelessness. 11 I focus on two consequential neighborhood resources that could impact student outcomes as promotive and/or protective factors: 1) community resources and 2) social ties. One of the reasons neighborhoods matter for students is because they provide material and immaterial resources. For example, as articulated by institutional resource theory, neighborhood-based resources (e.g., schools, community centers, childcare, libraries, parks, medical facilities, employment opportunities) provide engaging learning and social environments that can influence academic preparedness (Jencks & Mayer, 1990; Leventhal & Brooks-Gunn, 2000). The presence of these neighborhood resources can lead to higher academic outcomes because, for example, students with access to more extra-curricular activities have spaces to engage in out-of-school learning and social activities (Jencks & Mayer, 1990; Leventhal & Brooks-Gunn, 2000). The neighborhood composition and characteristics of residents also serves as a proxy for neighborhood resources due to a legacy of policies that have systematically underinvested in poor communities and communities of color (e.g., redlining) (Rothstein, 2017). Neighborhood residents can also act as relational resources that shape students’ academic experiences through social ties. More advantaged neighbors can provide social networks that are useful for school performance because these networks could enforce dominant norms that schools expect students to follow and because these networks can provide role models for students (e.g., Bourdieu, 1973; Coleman, 1988; Wilson, 1996). For example, advantaged neighbors could promote and enforce school-dominant norms (which typically align with White, middle-class norms) on how to behave and engage in schools, including how to dress, speak and act (e.g., 11 Because most students attend schools based on where they live, there’s considerable overlap between school and neighborhood resource levels. Still, studies have shown that school-related characteristics continue to matter even once neighborhood variables have been accounted for, and vice versa (Jargowsky & El Komi, 2009; Owens, 2010; Rendón, 2014). As a result, neighborhoods and schools independently (and jointly) shape students’ educational outcomes and thus both environments are examined in this study. 34 Bourdieu, 1973). Additionally, neighbors that have graduated from high school or college may also promote high expectations around educational attainment for the students living in that community and even act as role models (Wilson, 1996). Social ties and the ability to enforce norms are weaker in neighborhoods that have high residential turnover, high poverty and low collective efficacy (Sampson, 1997, 2012). As such, children living in neighborhoods with higher socio- economic status may be able to draw on these relational resources for school success, while others do not. Again, during a period of extreme adversity, neighborhood resources like those mentioned above could function as important protective factors for students experiencing homelessness. For example, neighborhood resources and connections to more advantaged social networks could help families secure emergency resources and shelter. Additionally, when families experience homelessness parents may have less time and capacity to care for students because they are preoccupied with working to secure housing and services. Thus, neighbors could supplement the level of care that students receive by enforcing norms, offering encouragement, and serving as role models. While researchers studying resilience have speculated that neighborhoods could be important promotive and protective factors, the empirical evidence establishing this association is not evident. Contributions In this study, I determine whether dimensions of schools and neighborhoods support resilience in the academic achievement and attendance for students experiencing homelessness. While the theoretical link between schools, neighborhoods, and resiliency for students experiencing homelessness is apparent, an empirical link has not been established. I provide new 35 evidence on which – if any – school and neighborhood factors promote resilience in outcomes and protect students experiencing homelessness from the generally negative effects of housing loss. Methods The main objective of this paper is to examine whether school and neighborhood resources support resilience in the academic (i.e., math and ELA standardized assessment scores) and attendance outcomes of students experiencing homelessness. To do so, I draw on several data sources on students, schools, and neighborhoods, and use a fixed effects analytical approach. In this section, I begin by describing the data sources utilized before explaining how I generate school and neighborhood measures. Before delving into the analytical approach used to answer the research question, I describe the sample of students used in this analysis. Data Sources I draw on student and school-level administrative data from the state of Michigan. At the student level, I utilize nearly 11 million observations from approximately 2 million unique students enrolled in Michigan schools between the 2011-12 and the 2017-18 school years. At the student- level, this administrative dataset includes an indicator for whether the student was identified as homeless within the school year. Under the McKinney-Vento Act, school districts in Michigan are required to identify students that “lack a fixed, regular, and adequate nighttime residence” as homeless, including the vast majority of students who live “doubled up” or temporarily with others due to housing loss or economic hardship (McKinney-Vento Homeless Assistance Act, 2015). The Michigan Department of Education monitors compliance with the McKinney-Vento Act by requiring districts to document their efforts in identifying and supporting homeless students. Homeless populations are generally challenging to identify because they are a transient and stigmatized population. The state’s monitoring of identification provides some assurance that 36 districts are using a common set of definitions and procedures to consistently identify homeless students (Cowen, 2017). In addition to the homeless indicator, I observe students’ addresses, school enrollment, grade level, free- or reduced-price lunch (FRL) eligibility, special education eligibility, and English learner (EL) status. I use address and school enrollment data to generate an indicator for whether the student changes census blocks at the end of the year (i.e., the student is in a different census block in time t+1 compared to time t) and whether the student changes schools at the end of the year (i.e., the student is in a different school in time t+1 compared to time t) 12 . Because I am interested in school resources, I also leverage school-level data related to teacher qualifications and experience, including average years of teaching experience, share of teachers with a masters or doctoral degree, and share of teachers rated as highly effective on the state’s educator effectiveness evaluation system. 13 I supplement these data with data on neighborhood characteristics available at the census tract-level from the American Community Survey’s (ACS) five-year estimates for 2013 to 2019. The five-year estimates are generated by collecting data over a five-year period for census tracts, so I assign each of the estimates to the midpoint year of its respective five-year range. 14 I draw on data related to neighborhood advantage and education including the census tract’s share of female- headed households, Black residents, White residents, employed, college graduates, high school dropouts, working in professional occupations, living in non-rent burdened conditions (30% or less of monthly income spent on rent), living in non-overcrowded conditions (less than 1.5 12 I ignore moves that are “structural” or moves that occur after completing the last available grade level in the school such as moving from elementary to secondary school. 13 I do not have access to teacher VAMs nor am I able to generate such a measure because I cannot link students to teachers. 14 For example, the 2013 ACS estimates span over 2009 to 2013. I assign the 2013 ACS estimates to the 2011-12 school year because the midpoint year in 2011. 37 occupants per room), along with the share of children not living in poverty and median household income. I draw on these variables because existing literature has found that these measures are proxies for community resources and social ties, which are important for students’ educational outcomes (e.g., Owens, 2010; Sampson, Sharkey, & Raudenbush, 2008). School & Neighborhood Measures The existing literature surfaces a few important school and neighborhood-based dimensions that may promote resilience: advantaged peers (e.g., Sacerdote, 2011), qualified teachers (e.g., Chetty et al., 2014b), and advantaged neighbors (e.g., Coleman, 1988; Wilson, 1996). As a result, I draw on variables related to these dimensions to create a measure of school resources and neighborhood resources. Specifically, I produce measures of school resources using the following school variables: share of peers that are not FRL-eligible, share of peers that are not homeless, share of peers that are stable in their school, share of peers stable in the neighborhood, average teacher years of experience, share of teachers new to the profession, share of teachers with a graduate degree, and share of teachers rated highly effective. I produce measures of neighborhood resources using the same variables described in the neighborhood data section. Because many of the variables listed above are highly correlated, I use exploratory factor analysis to condense school and census tract characteristics into indices. I do this in three steps. First, I use factor analysis to understand how many distinct factors surface from the school and neighborhood characteristics. I enter school characteristics jointly and neighborhood characteristics jointly. Following conventional practice, I retain factors with eigenvalues greater than one, although most factors have eigen values of at least two, which results in four retained factors. 38 Second, for each retained factor, I create indices using regression-based factor scores where the factor score for each student is predicted based on the observed values for items and their respective factor loadings. 15 Regression-based factor scores maximize validity more so than other methods (e.g., taking an average of values weighted by respective factor loadings) because they generate scores that provide the highest correlations between the factor score and the corresponding latent factor (DiStefano, Zhu, & Mîndrilǎ, 2009). I refer to these factor scores as indices. Lastly, to facilitate interpretation, indices are standardized across years so that the mean is zero and the standard deviation is one. Using this process, I generate four distinct indices: 1) peer advantage; 2) teacher qualifications and experience; 3) neighborhood advantage; and 4) neighborhood educational and occupational attainment. Table 2.1 includes each of the indices, the variables that contribute to each index, and each of the variables’ factor loadings. Positive factor loadings suggest that the variable is positively related to the underlying construct, while negative loadings suggest that the variables is negatively related. Higher values on these indices suggest higher levels of the construct (e.g., more peer advantage). Beginning with school-based characteristics, I generate a measure of students’ peer advantage, referred to as the peer advantage index, that includes the share of the students’ school peers who are not FRL-eligible, not homeless, who remain in the same school (unless their moves are “structural,” meaning that they move after completing the last available grade level in the school, such as moving from elementary to secondary school), and who remain in the same neighborhood (see panel A for factor loadings). The peer advantage index is interpreted as a measure of material and immaterial resources available at the school because schools that 15 The factor score is predicted by taking sum of the values of each student item multiplied by its unique regression coefficient, which is the product of its factor loadings and the inverse of the observed variable correlation matrix (DiStefano et al., 2009). Regression-based factor scores are created using Stata’s factor postestimation command. 39 educate more advantaged student groups are better funded and staffed. Additionally, other students benefit from the social ties and norms that are established and maintained if their peers are more stable and more advantaged. The next index describes teachers’ qualifications and experience, and it draws on average years of experience, and the share of teachers who are new to the profession (i.e., those with 1-2 years of experience), have a masters or doctoral degree, and were rated as highly effective on the state’s educator evaluator system (see panel B for factor loadings). I interpret this factor as the level of teachers’ experience and qualifications at the school. The third index describes neighborhood advantage, and the index utilizes the census tract’s share of female-headed households, residents who are Black, White, employed, living in non-rent burdened conditions, and living in non-overcrowded conditions, along with the share of children not living in poverty (see panel C for factor loadings). More advantaged neighborhoods should have stronger mechanisms to enforce norms (including school-dominant norms) and potentially more relational resources to share amongst neighbors (Coleman, 1988; Sampson, 2012; Sampson et al., 2008). The final index is neighborhood educational and occupational attainment, which consists of the share of the census tract residents that are college graduates, high school dropouts, working in professional occupations, and the tract’s median household income (see panel D for factor loadings). Like neighborhood advantage, areas with higher educational and occupational attainment may also have stronger mechanisms to enforce norms and more relational resources; but they could be distinct from neighborhood advantage because they may also have higher educational expectations, more role models for students that have completed high school and 40 college, and more knowledge and skills around how to successfully navigate educational institutions. Student Sample Past studies have found that students experiencing homelessness differ from those who are stably housed along a number of dimensions. I examine whether there are differences between students who are ever identified as homeless in the panel (“ever homeless”) and those that are never identified as homeless (“never homeless”). Table 2.2 shows average values for students that were ever and never homeless. Consistent with existing studies, I find that a smaller share of ever homeless students is White (56% vs. 68%), Asian (0.5% vs. 3%), and English Learners (ELs, 4% vs. 6%) while a larger share is Black (26% vs. 18%), Hispanic (10% vs. 7%), and receive special education services (22% vs. 13%). Ever homeless students have higher suspension rates (6% vs. 2%) and lower attendance rates (89% vs. 94%), and they are more likely to change schools (31% vs. 15%). All differences between these groups are statistically significant at p<.001. Analytical Approach To answer my research question, I start by descriptively investigating the types of schools and neighborhoods students are located in, including how this changes when students become homeless. These results provide important context about what types of schools and neighborhoods students have access to. I then explore whether school and neighborhood resources support resilience in attendance and academic achievement using a series of fixed effect regression models. Schools and Neighborhoods of Students Experiencing Homelessness. Using t-tests, I compare the averages for each of the four school and neighborhood indices for students that were ever homeless and never homeless. In addition, I also examine whether ever homeless students are located in schools and neighborhoods with lower levels of resources when they were identified as 41 homeless. To do so, I limit my sample to ever homeless students and then compare the school and neighborhood indices between times these students were identified as homeless and times these students were not identified as homeless. Together, these results provide relevant information about the school and neighborhood resources for students that ever experience homelessness, including whether students attend more, less or similarly resourced schools when identified as homeless. Educational Outcomes, Homelessness, and Factors Associated with Resilience. I use a series of student and year fixed effects regression models to explore my research question which allow me to make within-student and within-year comparisons. In the first model, I establish the relationship between homelessness and the outcomes I examine. In the second model, I explore whether school and neighborhood resources operate as promotive or risk factors for students. In the final model, I explore whether these resources operate as protective or vulnerability factors for students experiencing homelessness. Beginning with associations between homelessness and the outcome, the first model is specified as the following: ; ($.# =/ % +/ ' 2<=->-* * ($.# +/ * ? /012 +@ / +A 2 +ε ($.# (1) Where the outcome ; is the attendance rate, standardized math assessment score, and standardized ELA assessment score for student i in school s and neighborhood h at time t. 2<=->-** is an indicator for whether student i in school s and neighborhood h is homeless at time t. / ' provides the association between being homeless and each respective outcome compared to times the student is not homeless. As such, it provides a baseline measure of how negatively related homelessness is to the outcomes examined. These coefficients are estimated while controlling for ?!"#$, a vector of 42 time-varying student characteristics, that may influence outcomes including grade level and whether the student was FRL eligible, special education eligible, and an EL. I also estimate these coefficients within-student (@ / ) and within-year (A$) to account for characteristics related to students (e.g., race) and year trends (e.g., economic downturn) that may influence homelessness and student outcomes. Standard errors are clustered at the student level. I then examine whether school and neighborhood resources serve as promotive or risk factors that either are positively or negatively related to outcomes regardless of whether students are experiencing adversity. I specify the following model: ; ($.# =6 % +6 ' 2<=->-* * ($.# +6 ) EFGHIJKFG+6 * ? /012 +@ / +A 2 +ε ($.# (2) All covariates follow from equation 1 with the exception of EFGHIJKFG. EFGHIJKFG represents either a vector of MKN EFGHIJKFG 02 : 1) peer advantage index and 2) teacher experience and qualifications, or a vector of OFPQN EFGHIJKFG 12 : 1) neighborhood advantage and 2) educational and occupational attainment. I do not include both school and neighborhood indices jointly because of high correlations between these indices. 16 The coefficients on these resource indices, represented by 6 ) , provides information on whether the index is a promotive factor or a risk factor. If these coefficients are positive and significant, then they are promotive factors because they are positively related to student outcomes regardless of whether students are homeless. If the reverse is the case (i.e., they are negative and significant), then they are risk factors because they are negatively related to student outcomes regardless of whether students are homeless. Standard errors, 7, are clustered at the student level. 16 School peer advantage index is correlated with the neighborhood indices at r>0.65. The school resource indices are correlated at r=0.33 and the neighborhood resource indices are correlated at r=0.57. Models with both neighborhood resource indices have variance inflation factors within the normal range (below 10) (Hair, Anderson, Tatham, & Black, 1995). 43 The final model provides evidence of whether neighborhood and school resources serve as protective factors that are more positively related to outcomes when students are homeless, or vulnerability factors that are more negatively related to outcomes when students are homeless. I specify the following model: ; ($.# =/ % +/ ' 2<=->-* * ($.# +/ ) EFGHIJKFG+/ * 2<=->-* * ($.# ∗EFGHIJKFG+/ + ? /012 +@ / +A 2 +ε ($.# (3) All covariates follow from equation 2 with exception of the interaction term, 2<=->-* * ($.# ∗ EFGHIJKFG. This interaction term represents interactions between the homeless indicator and either the vector of school indices, 2<=->-* * ($.# ∗ MKN EFGHIJKFG 02 or the vector of neighborhood indices, 2<=->-* * ($.# ∗ OFPQN EFGHIJKFG 12 . The coefficients on these interaction terms, represented by / * , capture any additional association of school or neighborhood indices for students when they are homeless compared to times when they are not homeless. The interaction terms, along with the coefficients on the main effects of the school and neighborhood resource indices, provide information on whether the index is a protective or vulnerability factor. School or neighborhood resources are protective factors when the interactions terms are significant and positive (/ * >0) and the cumulative association of the main effect and interaction is positive (/ ) +/ * >0). A positive interaction term and cumulative association would indicate that the index is especially beneficial for student outcomes at times they are homeless compared to times when they are not homeless. Resources are vulnerability factors when they operate the opposite way: interactions terms are significant and negative (/ * <0) and the cumulative effect of the main effect and interaction is negative (/ ) +/ * <0). A negative 44 interaction term and cumulative association would indicate that the index is especially detrimental for student outcomes when they are homeless. Standard errors are clustered at the student level. Results I begin by describing the school and neighborhood resource levels for students who experience homelessness using descriptive statistics. Then, I turn to the series of fixed effects models I estimate, which 1) establish the relationship between homelessness and the outcomes examined (attendance and academic achievement), 2) show whether school and neighborhood resources serve as promotive or risk factors, and 3) show whether these resources serve as protective and vulnerability factors for students experiencing homelessness. Schools and Neighborhoods of Students Experiencing Homelessness Table 2.3 includes the descriptive results comparing school and neighborhood resource indices between students that were ever homeless and never homeless (panel A) and, for ever homeless students, between times they are and are not identified as homeless (panel B). Students that were ever homeless attend schools and live in neighborhoods with fewer resources when compared to never homeless students. To begin with, when compared to never homeless students, students that experience homelessness at some point attend schools with less advantaged peers (- 0.62 SD) and less qualified teachers (-0.16 SD) (see panel A). They also live in neighborhoods with lower levels of advantage and educational and occupational attainment (-0.28 and -0.49 SD, respectively) than students who are never homeless. All these differences are statistically significant (p<0.001). Overall, these differences show that ever homeless students are located in schools and neighborhoods that are far below the average for these measures (zero represents the index mean), while never homeless students are located in schools and neighborhoods with near average levels of resources. 45 There are small changes in school and neighborhood resource levels when I compare indices between times an ever homeless student is and is not identified as homeless (see panel B). When students become homeless, they are still located in schools and neighborhoods with resources lower than the average for each index. They attend schools with slightly lower resource levels (peer advantage= -0.06 SD and teacher qualifications and experience= -0.04 SD) and they move to slightly more resourced neighborhood environments than when they were not homeless (neighborhood advantage= 0.10 SD and neighborhood educational and occupational attainment= 0.02 SD). Students neighborhood resources might change because they move into these areas to secure temporary housing (e.g., shelters, permanent supportive housing, living doubled-up with other families). The overall trends in within student differences in schools and neighborhoods are consistent with Dhaliwal et al. (2021). Educational Outcomes, Homelessness, and Factors Associated with Resilience School and neighborhood resources could matter a great deal for the educational outcomes of students experiencing homelessness. To examine if this is the case, I turn to the next set of regression models. I begin by examining how homelessness is associated with attendance and academic achievement by reporting results from equation (1). Table 2.4 shows results from equation (1) with attendance in column 1, standardized math scores in column 2, and standardized ELA scores in column 3. Then, I examine whether school and neighborhood resources function as promotive or risk factors for educational outcomes by reporting results from equation (2). Finally, I report results from equation (3) which show whether these resources function as protective or vulnerability factors when students experience homelessness. Results from equations (2) and (3) are split across two tables: Table 2.5 includes models using school resource indices and Table 2.6 includes models using neighborhood resource indices. 46 In Table 2.4, I find that when students are homeless, they have lower attendance (column 1; -1.2 percentage points), math scores (column 2; -0.02 SD), and ELA scores (column 3; -0.01 SD) than times when they are not homeless. These models provide an understanding of how becoming homeless relates to the outcomes of interest and they are consistent with existing studies that use fixed effects (e.g., Cowen, 2017). School Indices & Factors Related to Resilience. Depending on the outcomes examined, I find that school resources either serve as promotive or risk factors (see Table 2.5). I begin by exploring the instances when school indices are promotive factors, followed by cases where these indices are risk factors. I find that peer advantage functions as a promotive factor for attendance rates, while teacher qualifications and experience operate as a promotive factor for math and ELA scores (columns 1, 3 and 5). For instance, beginning with peer advantage and attendance, a SD increase in the level of peer advantage is associated with a 1.1 percentage points higher attendance (column 1). The same increase in teacher qualifications and experience is related to a 0.02 SD increase in math scores (columns 3) and a 0.01 SD increase in ELA scores (column 5). Teacher qualifications and experience are a stronger predictor of achievement (both ELA and math) than peers, while the reverse is true for attendance. Interestingly, the teacher qualifications index is negatively associated with attendance—although the magnitude of the coefficient is near zero (less than a hundredth percentage point lower attendance) so its practical significance is nominal. School resources could take on a heightened role when students experience adversity, so I look for evidence of whether these resources operate as protective or vulnerability factors next (columns 2, 4, and 6). Overall, I find that peer advantage operates as a protective factor for student attendance and as a vulnerability factor for ELA achievement, although, in the case of ELA achievement, the size of its association is nominal. There is not statistically significant evidence 47 of protective or vulnerability factors for other outcomes. Beginning with student attendance (column 2), when students are homeless, a SD increase in peer advantage is associated with a 0.4 percentage point greater marginal increase in attendance rate compared to times when they are not homeless. Cumulatively, a SD increase in peer advantage is associated with a 1.5 percentage point increase in attendance rates when students are homeless. The positive interaction and cumulative association show that peer advantage serves as a protective factor. In contrast to its relationship with attendance, peer advantage appears to works as a vulnerability factor for ELA achievement when students are homeless: a SD increase in peer advantage is associated with a 0.007 SD greater decrease in ELA achievement for students when they are homeless compared to times when they are not homeless (column 6). Cumulatively, the association of a SD increase in peer advantage is negative—0.008 SD lower ELA achievement at times when students are homeless—however, this represents less than a hundredth of a SD unit. As such, the practical significance is nominal. Neighborhood Indices & Factors Related to Resilience. Neighborhood indices function as promotive factors across attendance and achievement outcomes. As shown in Table 2.6, I find that neighborhood advantage functions as a promotive factor for attendance rates and math achievement, and neighborhood educational and occupational attainment is a promotive factor for ELA achievement (columns 1, 3, and 5). When it comes to student attendance, both neighborhood indices operate as promotive factors. A SD increase in neighborhood advantage is associated with a 0.7 percentage point increase in attendance, while the same increase in neighborhood educational and occupational attainment is associated with a 0.1 percentage point increase in attendance (column 1). Neighborhood resources are also promotive factors for math and ELA achievement but there are differences in which neighborhood resource is positively related to which outcome. 48 Neighborhood advantage is positively related to math achievement while educational and occupational attainment has the same relationship with ELA achievement. In both cases the magnitude of this relationship is small, less than a hundredth of a SD (columns 3 and 5, respectively). As such, the relationship between academic achievement and neighborhood indices is not practically significant. There is little evidence that neighborhood resources function as protective factors and no evidence that they function as vulnerability factors (columns 2, 4, and 6). Attendance is the only outcome where neighborhood resources, specifically neighborhood advantage, is associated with larger increases in attendance when a student is homeless compared to times when they are not homeless (column 2). Specifically, when a student is homeless, a SD increase in neighborhood advantage is associated with a 0.1 percentage point marginally greater attendance rate compared to times a student is not homeless. Cumulatively, a SD increase in neighborhood advantage is associated with 0.8 percentage point higher attendance rate for students when they are homeless. In summary, I find that school and neighborhood resources work in varied ways depending on the outcome examined, with examples of these resources working as promotive factors and fewer examples of these resources working as protective factors. I also find examples of resources working as risk or vulnerability factors, although the practical significance of these relationships is nominal. Figure 2.2 shows which resources function as promotive or risk factors and which serve as protective of vulnerability factors. A few patterns emerge across outcomes. First, teacher qualifications and experience operate as promotive factors for achievement outcomes as do neighborhood-related resources. Second, there are few examples of protective factors. The exception being peer and neighborhood advantage, which serve as protective factors for student 49 attendance. These results suggest that school and neighborhood resources work in nuanced ways to shape resiliency. Limitations The results presented in this study are not causal impacts of homelessness nor are they causal impacts of school and neighborhood resources on students’ attendance and academic outcomes. As such, I present these results as associations. In the fixed effects regression models, I am able to account for important student-related characteristics and secular time trends that may jointly influence predictors and outcomes by making within-student and within-year comparisons. However, there may be other omitted variables that I do not control for that could be driving the relationships I find. While these results cannot be interpreted as causal, they do still account for some major sources of selection bias by incorporating student and year fixed effects. Discussion Experiencing homelessness is largely associated with negative educational outcomes; and yet, there is also variation in outcomes with some students continuing to achieve academically despite housing loss. Research on resilience provides a framework for understanding why some students continue to progress academically, revealing the role of promotive, protective, risk, and vulnerability factors in shaping students’ outcomes. Using this framework, I develop school and neighborhood resource measures and analyze data from the state of Michigan to make a few important contributions. First, I analyze the school and neighborhood environments of students experiencing homelessness in a large and heterogeneous state. Second, I determine whether important dimensions of schools and neighborhoods serve as promotive or risk factors (i.e., generally beneficial or detrimental for outcomes) and/or protective or vulnerability factors (i.e., 50 especially beneficial or detrimental when students are homeless) for academic achievement and attendance. School and neighborhood resources have varied relationships with student outcomes, at times functioning as promotive and protective factors and less frequently serving as risk and vulnerability factors. Beginning with the factors that benefit educational outcomes, I find neighborhood advantage is a promotive and protective factor for student attendance and a promotive factor for math achievement. Students also live in more advantaged neighborhoods in years they experience homelessness, perhaps because shelters and temporary housing is located in these communities. Additionally, peer advantage is a promotive and protective factor for student attendance, but I also find that students attend schools with slightly fewer advantaged peers in years they are homeless. Finally, teacher qualifications and experience and neighborhood educational and occupational attainment are promotive factors for academic achievement; although, when homeless, students attend schools with slightly lower levels of qualified and experienced teachers, and they live in neighborhoods with slightly higher levels of educational and occupational attainment. The evidence of school and neighborhood resources serving as risk and vulnerability factors is scant, with nominal negative associations that are not practically significant. Overall, these findings raise questions about why certain resources operate as promotive and protective factors for some outcomes but not others. The promotive factors identified largely align with existing literature. As suggested by other studies, peer and neighborhood advantage may be promotive factors for attendance and academic achievement because, for example, being surrounded by more advantaged peers and neighbors likely promote norms around school achievement and attendance (e.g., Sacerdote, 2011). Teacher qualifications and experience along 51 with neighborhood educational and occupational attainment are another set of promotive factors, but these are only positively associated with academic achievement. More experienced and qualified teachers promote student learning through effective instruction which would have a greater link to student achievement than to student attendance (although higher teacher VAMs have also been linked to greater attendance, see Jackson (2012) and Ladd & Sorensen (2015)) . Similarly, highly educated neighbors promote expectations around educational attainment and provide resources that support learning (Coleman, 1988). These same expectations and resources might not be as strongly related to attendance. Student attendance is the only outcome for which protective factors are identified. Peer and neighborhood advantage are especially beneficial for student attendance when students are homeless. This might be because schools and neighborhoods with higher advantage provide better transportation access. Transportation is a major barrier to school attendance for students experiencing homelessness, especially if temporary housing means students are located farther away from schools (Mawhinney-Rhoads & Stahler, 2006). Schools are required by law to provide students with access to transportation if they move outside of the neighboring community but families experiencing homelessness still cite transportation as a common obstacle for school attendance (Pavlakis, 2014). For example, students experiencing homelessness are often provided with city bus passes to get to and from school. However, parents may be concerned about their student’s safety using public transportation. Areas with more advantaged peers and residents could have more comprehensive and safer school and public transportation systems that students rely on to get to school while they are homeless. Consequently, peer and neighborhood advantage could be a protective factor for attendance because of its association with transportation access. 52 Transportation access, however, may not translate to higher achievement, which explains why these same measures are not protective factors for academic achievement. I purposefully chose school and neighborhood characteristics that other studies have identified as beneficial resources for educational outcomes. Yet, the school resource measures selected sometimes operated as risk and vulnerability factors. In both cases, the negative associations of these school resources with the outcomes examined are small. For instance, a one SD increase in teacher qualifications and experience is associated with less than a hundredth percentage point reduction in student attendance and the same increase in peer advantage is associated with less than a hundredth decrease in standardized ELA scores. Thus, the practical significance of these associations is trivial. Future Research Future research should examine whether there is evidence of other school- and neighborhood-based resources operating as protective factors for students experiencing homelessness. Scholars of resilience point out that seemingly ordinary processes and relationships may protect students living through adversity (e.g., Masten, 2001), including many of those not captured by this paper. For instance, schools could serve as protective factors through the presence of strong and consistent relationships with teachers and other students, positive school climate, presence of support staff like social workers and counselors, and curriculum that teaches social emotional skills for managing stressful situations. Some of these school-based resources, like support from close friends and caring teachers, have been documented by qualitative researchers as beneficial to the success of students experiencing homelessness (Edwards, 2019). Similarly, neighborhoods could be protective factors through social ties and characteristics related to safety 53 and the presence of social services. Future research should examine whether these other aspects of schools and neighborhoods make a difference for students experiencing homelessness. Completely eliminating the relationship between homelessness and students’ educational outcomes would require solving homelessness itself, which demands actions outside the purview of educational systems. One way to improve the educational success for students experiencing homelessness within the confines of schools is to invest in protective and promotive factors that are associated with better outcomes for these students. This study provides important evidence towards this end by identifying which school and neighborhood resources support the resiliency of educational outcomes for students experiencing homelessness. Many of the factors that support the success of students experiencing homelessness (e.g., teacher quality, peer advantage, neighborhood composition) are also resources that students experiencing homelessness have less of in their schools and neighborhoods. School and neighborhood-based resources could help support resilience, but targeted investments are needed to support to get there. 54 Paper 3: A Lever for Improving Student Success? The Causal Effect of LIHTC on Underserved Students 17 Improving educational outcomes for underserved students continues to be a primary goal of education leaders and policymakers. While education policies focus on improving students’ experiences within the confines of the school system (e.g., improving curriculum, teacher quality), there remains a firm understanding that students’ learning is impacted by factors that extend beyond the schoolhouse walls (e.g., housing, poverty, health). In particular, housing intersects with education in ways that are consequential for a variety of student outcomes. Shortages in affordable housing and other housing-related factors (e.g., housing quality, neighborhood contexts) are drivers of homelessness and mobility (i.e., changing residences and, consequently, schools), which are negatively correlated with students’ educational outcomes (see Miller (2011) and Welsh (2017) for reviews). Affordable housing may also be beneficial for underserved students in other ways that matter for educational outcomes (Leventhal & Newman, 2010), such as increasing family economic resources available for educational investments (Becker & Tomes, 1986) and reducing stress (Conger, Ge, Elder, Lorenz, & Ronald, 1994; Mistry, Vandewater, Huston, & McLoyd, 2002). Given how consequential housing is for student well-being, providing students with both education and housing supports may be a more effective way of promoting academic and behavioral success than focusing on schools alone. In fact, many of the families targeted for 17 This research result used data structured and maintained by the MERI-Michigan Education Data Center (MEDC). MEDC data is modified for analysis purposes using rules governed by MEDC and are not identical to those data collected and maintained by the Michigan Department of Education (MDE) and/or Michigan’s Center for Educational Performance and Information (CEPI). Results, information and opinions solely represent the analysis, information and opinions of the author(s) and are not endorsed by, or reflect the views or positions of, grantors, MDE and CEPI or any employee thereof. This research was supported by a grant from the American Educational Research Association which receives funds for its "AERA Grants Program" from the National Science Foundation under NSF award NSF-DRL #1749275. Opinions reflect those of the author and do not necessarily reflect those AERA or NSF. 55 affordable housing supports, such as low-income, homeless, and housing insecure families, are the very same families that are targeted by education programs for underserved communities. Federal education programs are designed to provide additional funding and supports for low-income students (e.g., Title I), students that are mobile due to labor migration (e.g., Migrant Education Program) 18 , and for homeless students (e.g., McKinney-Vento Homeless Assistance Act). Students that benefit from a combination of programs like federal educational supports along with access to affordable housing programs may be best positioned to succeed academically because jointly these programs create conditions necessary for student learning, such as housing security, stability, and school attendance (Childs & Lofton, 2021; Leventhal & Newman, 2010; Welsh, 2017). In this study, I examine whether the largest federal affordable housing development program, the Low-Income Housing Tax Credit (LIHTC) program, affects student homelessness, school and residential mobility, and attendance. LIHTC increases the availability of affordable rental units by offering real estate developers tax credits in exchange for constructing or rehabilitating rental units and offering them at below market rents to low-income households. Although there have been several studies that assess the impact of LIHTC on neighborhood poverty and inequality (Baum-Snow & Marion, 2009; Deng, 2007; Ellen, Horn, & O’Regan, 2016; Freedman & McGovak, 2015), less is known about how the program might affect students enrolled in K-12 public schools, including through direct housing and indirect spillover effects. Households with children who move into affordable units may benefit from housing effects that could reduce homelessness, diminish residential and school mobility, and thereby improve educational outcomes. The program may also impact students who live close by to units through documented spillover effects of LIHTC development, such as reduced crime, changes in property values, and 18 Students are eligible for the Migrant Education Program if their parent is a migratory worker in the agricultural, dairy, lumber or fishing industries and if they have moved school districts at any point in the past three years. 56 changes in the composition of residents, that could have varied impacts on the outcomes examined (Diamond & McQuade, 2019; Freedman & Owens, 2011; Amy E. Schwartz et al., 2006). I bring together student-level administrative data from the state of Michigan along with data from the Department of Housing & Urban Development (HUD), Michigan State Housing Development Authority, and the Census Bureau to answer two research questions: 1) Does the development of affordable housing via LIHTC funds affect student homelessness, residential and school mobility, and attendance?; and 2) Does LIHTC have differential impacts for students who are also targeted by educational programs (e.g., Title I, Migrant Education Program)? I exploit a discontinuity in the LIHTC funding formula that results in additional tax credits for projects and, consequently, additional units being built in qualified census tracts (QCTs). I use the quasi- experimental variation created by this discontinuity to estimate the causal effects of the program on student homelessness (including those who live temporarily with others, “doubled up”, that HUD homeless counts exclude), residential and school mobility, and attendance. While I find that QCTs receive additional LIHTC investment, I do not find evidence of LIHTC’s effects on student homelessness, mobility, or attendance. Nor do I find LIHTC effects for students who are already considered underserved by federal educational programs (i.e., Title I, Migrant Education Program [MEP]). There does, however, appear to be small reductions in neighborhood and school mobility for students who are relatively more advantaged (i.e., not identified by Title 1 & not identified by MEP), which may be primarily driven by students who live in the same census block as LIHTC investment (a proxy for housing effects) or live in the vicinity of LIHTC investment (a proxy for spillover effects). In what follows, I provide background on the LIHTC program and outline a conceptual model that explains why LIHTC could have implications for student mobility and attendance. 57 Next, I summarize the data and methods used to generate causal estimates, before describing the study results and sensitivity to robustness checks. I conclude with a discussion of results, including policy implications. Background & Conceptual Framework I begin by describing the LIHTC program, including its structure and requirements. Next, I introduce a conceptual framework that shows why LIHTC investments might matter for student outcomes, while drawing on existing literature to make connections between LIHTC and homelessness, residential and school mobility, and educational outcomes. The LIHTC Program Established by Congress in 1986, LIHTC incentivizes the construction of affordable rental units by offering real estate developers tax credits in exchange for constructing and leasing units at below-market rates. LIHTC is the largest program in U.S history to provide property-based rental subsidies-- outstripping the production of public housing and HUD- assisted, privately-owned housing (Khadduri, Climaco, Burnett, Gould, & Elving, 2012). The program is also a significant contributor to the overall construction of multi-family rental housing, with a third of all construction between 1987 and 2006 receiving LIHTC investment (Eriksen & Rosenthal, 2010). As of 2015, there are roughly 2 million LIHTC units offered at below-market rates nationally (Scally, Gold, Herman, Gerken, & DuBois, 2018). The federally-funded program is administered by the U.S. Department of Treasury in partnership with state housing finance agencies (SHFA). Each SHFA receives tax credit allocations based on the state’s population size. Real estate developers proposing projects in the state apply directly to the SHFA for tax credits. The SHFA awards tax credits through a competitive process that adheres to the state’s qualified allocation plan (QAP), a publicly 58 available document that discloses special preferences (e.g., preferences for projects that target certain populations like seniors, people with special needs, or families) used in tax credit allocation decisions. 19 In exchange for tax credits, developers must guarantee that a certain share of units funded by LIHTC are offered to low-income tenants at below-market rates determined by HUD for a 30-year compliance period. 20,21 In practice, however, the tax credit incentives produce projects where more than 90% of units are rented at below-market rates for low-income tenants (Ellen, O’Regan, & Voicu, 2009). Research on LIHTC tenants confirms that low-income (i.e., those with incomes between 60% to 31% of Area Median Gross Income [AMGI]) and, to a lesser extent, extremely low-income tenants (i.e., those with incomes at or below 30% of AMGI) occupy these units (Buron, Nolden, Heintzi, & Stewart, 2000; O’Regan & Horn, 2013). Extremely low-income tenants are more likely to be served through other HUD housing programs, such as public housing and housing vouchers, because the below-market rents in LIHTC properties may still be too expensive for these tenants (McClure, 2019; O’Regan & Horn, 2013; Williamson, 2011). 22 Particularly relevant for this study, which is located in Michigan, LIHTC residents in the state of Michigan appear to be more economically disadvantaged than the national average 19 The LIHTC program offers two kinds of taxes credits: the 9% tax credit and the 4% tax credit. This study examines only the 9% tax credit program because these credits are administered by SFHAs through a competitive process. The 4% tax credits are administered through a non-competitive process by the federal government. 20 At least 20% of households that occupy constructed rental units (i.e., occupy the apartment building) must have incomes below 50% of area median gross income (AMGI) or at least 40% or more of households that occupy constructed units must have incomes below 60% of AMGI. 21 An analysis of the earliest funded LIHTC properties found most properties continue to offer below market rental rates once the compliance period ends because these properties were awarded additional tax credits for rehabilitating rental properties (Khadduri et al., 2012). Thus, LIHTC properties continue to be affordable even after the compliance period ends. 22 LIHTC tenants can be served by multiple programs so they could also receive a housing voucher, which they use to pay for rent in LIHTC properties. Thus, extremely low-income tenants can afford rents in LIHTC properties with the use of other income supports like housing vouchers. 59 LIHTC resident, with the majority of residents making at most 30% of AMGI in 2012 and a median yearly income of $14,256 (Hollar, 2014). Additionally, a majority of households nationally in LIHTC properties (54%) are families with at least one child under 18 living in the rental unit (Buron et al., 2000), although, in Michigan, only a third of units are occupied by families with children (Hollar, 2014). Because housing assistance is not available to all eligible households, the housing supplied by LIHTC does not meet the demand for affordable housing. 23 As such, despite the prevalence of LIHTC development, a relatively small share of low- income families that qualify for below-market units receive subsidized housing. Developers can reap up to 30% higher tax credits by proposing a project in a QCT. Census tracts are deemed “qualified” as long as they meet one of the two criteria: 1) have a poverty rate of at least 25% (referred to as poverty rate eligibility); or 2) have at least 50% of households with incomes that are 60% of AMGI (referred to as income share eligibility). 24 Until recently, QCT status was generally redesignated every 10 years with the release of decennial census data (Hollar & Usowski, 2007). 25 National studies of LIHTC development confirm that the incentives to build in QCTs result in more affordable rentals being built in areas that are just above the QCT threshold than those that are just below it (Baum-Snow & Marion, 2009; Brummet & Bartalotti, 2016). Conceptual Model 23 For example, in 2014 there were roughly 11 million extremely low-income households (incomes at or below 30% of the area median income), not to mention millions more low-income households. In this same time period, there were roughly 2 million LIHTC units (Getsinger et al., 2017). 24 Census tracts that are eligible based on at least one of these two criteria may be disqualified if the total population of all QCTs in a metropolitan statistical area (MSA) exceed 20 percent of total MSA population, referred to as the population cap. Because few tracts in Michigan are impacted by the population cap, I drop the three census tracts impacted by the population cap criteria. 25 Beginning in 2013, HUD transitioned to using the American Community Survey 5-year estimates for some data fields used to calculate QCT (Federal register, 2011). From 2015 onward, QCT status is redesignated every year rather than every decade (Federal register, 2014). 60 There are several mechanisms through which LIHTC could impact a variety of student outcomes. Figure 3.1 distills these mechanisms in a conceptual diagram. On the left-hand side of Figure 3.1, LIHTC investment in the form of tax credits is a “treatment” that leads to the production of new below-market rate rentals or the rehabilitation of existing rental units. The treatment generates housing and spillover effects that could ultimately influence student outcomes. Housing Effects LIHTC investment generates direct housing effects for households that move into the below-market rate rentals. Living in a new or remodeled below-market rate unit reduces rent burden, or the share of monthly income allocated to housing costs, (Buron et al., 2000; O’Regan & Horn, 2013; Williamson, 2011) and increases housing quality (e.g., new amenities in building, more security features, new fixtures and appliances) (Buron et al., 2000; O. Jackson & Kawano, 2015). The development of LIHTC units may also reduce overcrowding because affordable shortages are associated with families living in overcrowded conditions (Carter, 2011). The changes produced by housing effects may have benefits for reducing homelessness, residential and school mobility, and improving students’ educational outcomes. For instance, rent burden is a strong predictor of homelessness, even after controlling for individual risk factors (e.g., persistent poverty, health issues, unemployment) (Early & Olsen, 2002; Hanratty, 2017). Rent burdened-families have fewer reserve funds to deal with everyday emergencies; thus, when they do arise (e.g., job loss, unexpected medical bills), these families are at-risk of falling behind on rent and losing their housing (Shinn & Gillespie, 1994). Families living in subsidized units may be able to establish savings that could buffer them from housing loss 61 when they experience a negative economic event. Additionally, families that are experiencing homelessness could also secure these units as long-term subsidized housing with the help of a housing voucher, which has been shown to be an effective intervention for homeless families (e.g., Gubtis et al., 2010). Two existing studies find that LIHTC may reduce homelessness. Jackson and Kawano (2015) use a quasi-experimental design and find statistically significant reductions in homelessness at the county level, but not the census tract level, while Kim and Sullivan (2020) find reductions in homelessness in a Continuum of Care area (a HUD-defined area similar to a county or metro area). While informative, the homeless data used in these studies suffers from limitations and they do not specifically examine family or student homelessness. 26 Affordable housing development could also change residential and school mobility. Census Bureau data suggest that housing-related factors such as affordability explain almost half of residential moves (U.S. Census Bureau, 2017). In the context of LIHTC, perhaps initially, there may be increases in residential mobility as families move into newly available LIHTC units. However, after this transition period, families should be more likely to stay in their units due to rent increase restrictions and higher quality housing stock. Low vacancy rates in LIHTC properties suggest this is indeed the case, although residential turnover still exists (Collinson & Winter, 2010). Because the majority of school moves are due to residential moves, reductions in residential mobility should produce similar results for school mobility (de 26 Jackson and Kawano (2015) refer to their data as 2000 Census “homeless count”. However, the Census Bureau withheld 2000 homeless counts and instead published counts of those in “other non-institutionalized group quarters” (Holmes, 2001), which includes homeless observed in shelters, in the streets, and other locations (excluding those residing doubled up—the far majority of homeless K-12 students) during one night, along with medical workers staying in dormitories in general and military hospital, among others. As a result, it is unclear exactly who is recorded in these counts. In the case of Kim and Sullivan (2020), the authors use counts collected from trained volunteers, much like the Census counts of those observed homeless described above, which also exclude doubled up individuals and families. 62 la Torre & Gwynne, 2009; Rumberger, 2003). Students’ educational outcomes may also benefit from reductions in rent burden, residential mobility, and increases in housing quality. 27 To that end, homeless students and students who change schools, respectively, have lower academic achievement and, in the case of homelessness, lower attendance (see Miller (2011) and Welsh (2017) for reviews). If LIHTC development reduces homelessness and mobility, then academic achievement and attendance gains might follow. Additionally, LIHTC may directly impact academic achievement if reductions in rent burden free up family economic resources for educational investments (e.g., extracurricular activities, books in the home) (Becker & Tomes, 1986). Students’ academic achievement could also benefit from less crowded and higher quality housing, which may reduce stress and provide space for students to complete work without distractions (Conger et al., 1994; Mistry et al., 2002). Overall, living in LIHTC units should be beneficial for students and their educational outcomes. Spillover Effects LIHTC may also produce spillovers for residents who live nearby the housing site. Local residents may benefit from LIHTC investment if it removes disamenities from the community (e.g., blighted buildings, vacant lots) and reduces area crime rates (Diamond & McQuade, 2019; Freedman & Owens, 2011; Amy E. Schwartz et al., 2006). LIHTC investment could also attract more investment to an area by “demonstrating” to other developers that an area is revitalizing (Caplin & Leahy, 1998). Retail may also accompany LIHTC construction because these projects could include mixed-use development with residential and retail space. 27 While academic achievement is not an outcome examined in this study, Di and Murdoch (2013) provide the only evidence connecting LIHTC and school achievement. Using school fixed effects, they find new units in high income areas and newly rehabilitated units in low-income areas are associated with school achievement gains. 63 Depending on the types of businesses introduced by LIHTC investment, community residents could gain access to useful amenities that were not available before the development (e.g., neighborhood corner store). Finally, LIHTC investment may also change the composition of residents by drawing new residents into the area. Who is drawn into areas with LIHTC development may depend on the neighborhood; one study suggests that LIHTC development draws higher-income home buyers into low-income areas and lower-income home buyers into high-income areas (Diamond & McQuade, 2019; although Brummet & Bartalotti (2016) find no evidence of composition changes). Students living near LIHTC investments could benefit from these spillover effects. For instance, reductions in residential and school mobility could ensue if residents feel less inclined to leave an area that is experiencing improvements in neighborhood amenities and saftey. Families that are homeless or on the brink of homelessness could also benefit from additional neighborhood amenities and services spurred by LIHTC (e.g., nonprofits, social services). Reductions in homelessness and mobility would improve academic achievement and attendance (Miller, 2011; Welsh, 2017). Students’ attendance and achievement could also be impacted by reductions in crime (e.g., Burdick-Will, Stein, & Grigg, 2019; Sharkey, Schwartz, Ellen, & Lacoe, 2014) and improvements in neighborhood amenities that complement learning (e.g., Leventhal & Brooks-Gunn, 2000). Changes in neighborhood composition could have varied impacts for students. If higher-income residents move to areas with LIHTC development, then aggregate school and residential mobility and homelessness in the area could decrease and attendance could increase as a function of who is living (and attending school) in the area. The reverse could also be the case if lower-income residents are drawn into areas with LIHTC investment. 64 However, spillover effects might also negatively impact student outcomes. For example, LIHTC construction increases property values in the surrounding area (Diamond & McQuade, 2019; Amy E. Schwartz et al., 2006), which may lead to rent increases in surrounding properties and, consequently, higher rent burden. This process might also lead to residents moving out of the community because they are “priced out”. There is also some evidence that LIHTC increases poverty concentration and neighborhood inequality (Baum- Snow & Marion, 2009; Funderburg & MacDonald, 2010), although the evidence is mixed (Brummet & Bartalotti, 2016; Ellen et al., 2009; Freedman & McGovak, 2015). Such increases may limit the presence of public goods and resources in the community, which could in turn have troubling effects for homelessness, residential and school mobility, and educational outcomes. While the conceptual model provides a general overview of how LIHTC impacts important student outcomes, there are at least two types of students that could be more acutely impacted by LIHTC investment: those who have been identified as underserved by education programs (i.e., Title I, MEP), and those who live in communities with high rent burden or residential mobility. Educationally underserved students (and their families) may be the most sensitive to affordable housing development because they include the most economically disadvantaged students and, thus, are the target of affordable housing. These students should be the most likely to experience housing effects because their incomes are low-enough to qualify for subsidized rentals. The same can be said for students who live in areas with the highest levels of rent burden. These areas have the highest need for affordable housing development so they may be the most sensitive to its introduction. As a result, these students should experience many of the benefits from housing effects described above . 65 We may also find that educationally underserved students who also live in LIHTC units are the most sensitive to educational improvement because they are being jointly targeted by education and housing programs. Students receiving a more comprehensive set of supports may translate to better school outcomes (e.g., higher attendace). These students may also be the most sensitive to spillover effects because they have the fewest resources to begin with so any resources introduced to (or taken away from) the community may especially affect them. Depending on the types of spillover effects that result from LIHTC investment, these students may have the most to gain (or lose) from the construction of affordable housing. Data In this study, I examine the impact of LIHTC development from 2004-2012 induced by a neighborhood’s QCT status on three outcomes measured in 2013-14 for students living in the area. To do so, I bring together student-level data from the state of Michigan with data related to the LIHTC program and census tract characteristics. Sources To examine the effects of LIHTC on students, I observe 1.52 million unique students enrolled in K-12 public schools in the state of Michigan for the 2013-14 school year. 28 I examine three outcomes: student mobility, homelessness, and attendance. First, to measure mobility I use school enrollment information observed in the administrative data set to generate a “new to school” indicator. A student is considered new to the school if they attend a different school in year t compared to t-1 and do not do so because of structural mobility (completing all available grade levels at their previous school site, such as elementary to middle school moves). I also observe students’ geocoded addresses, which is used to generate a “new to census tract” indicator. Students 28 Data from the 2012-13 school year is used to generate the new to school and new to census tract indicators described in greater detail below. 66 observed living in a different census tract in year t compared to t-1 are considered to be new to their census tract. 29 The administrative dataset also includes a homeless indicator, which describes if a student was identified as homeless in the school year. Under the McKinney-Vento Act, school districts in Michigan are required to identify students that “lack a fixed, regular, and adequate nighttime residence” as homeless, including the vast majority of students who live “doubled up” or temporarily with others due to housing loss or economic hardship (McKinney-Vento Homeless Assistance Act, 2015). Other agencies have more limited charters and exclude “doubled up” families and their children in homelessness counts (e.g., HUD). The Michigan Department of Education monitors compliance with the McKinney-Vento Act by requiring districts to document their efforts in identifying and supporting homeless students. While generally homeless populations are challenging to identify because they are transient and stigmatized, the state’s monitoring provides some assurance that districts are using a common set of definitions and procedures to consistently identify homeless students (Cowen, 2017). Finally, I observe the share of days a student attended school and I observe whether the student (or their school) was enrolled in a special program for low-income or highly mobile populations, such as Title I funding and technical assistance and MEP. I add data related to QCT status and LIHTC investments to the student-level data. QCT status was essentially time invariant from 2003 (when it was released using 2000 decennial census 29 Because of timing differences between the end of the LIHTC treatment period and the outcome year (discussed below), the “new to school” and “new to census tract” indicators should not detect transition effects of students who are initially moving into LIHTC units. Meaning, these mobility indicators should not capture the mobility that occurs when students initially move into newly developed units (i.e., when the units are first put on the market). Instead, these indicators detect subsequent mobility (or lack thereof) after students have had an opportunity to move into LIHTC units. 67 data) to 2012. 30 I use HUD provided data on each census tract’s poverty rate and income share (i.e., the share of households earning 60% AMGI criteria) that was used to identify QCTs. To determine LIHTC activity from 2004 to 2012, I rely on a publicly available database of LIHTC- funded projects developed by HUD and, to ensure data coverage, I supplement this with data from the Michigan State Housing Development Authority for projects in these same years. The database includes the date the project was “placed in service” (i.e., ready for tenant use), the address of the project, the census block and tract the project is located in, the number of units, and the number of units made available at below-market rents. 31 To allow sufficient lead time following the 2003 QCT designation, I limit the analysis to projects placed in service beginning in 2004 because these projects were likely planned after the 2003 QCT designations were released. Following Dizon- Ross (2020), the treatment period extends to 2012, the last year before QCT status was redesignated using the 2010 census. As such, the treatment period (2004 to 2012) allows adequate time to observe effects from the housing, spillover, and composition mechanisms. 32 33 Last, I incorporate baseline demographic and housing variables observed at the census tract-level from the 2000 decennial census. Variables include fraction White, Black, and Hispanic residents, respectively, residents less than 18 years old, residents greater than 65 years old, occupied housing units, own-occupied housing units, total housing units, and median household income. 30 For 14 census tracts, QCT does change between 2002 and 2012 due to updates made to census tract housing income data. I exclude these tracts from the analysis to ensure QCT status is functionally time invariant. 31 While HUD provides information on whether the project was new construction or rehabilitated unit and whether the project targeted special populations (e.g., homeless, seniors, families), the same data is not available from the Michigan State Housing Development Authority. Thus, I am unable to disaggregate projects in this way. 32 Other published papers using this approach use data from 1994 to 1999 because they use decennial census data for their observed outcomes. Because I am not limited by census data, I use the full range of years that the observed QCT status was in effect. 33 I do not have access to student-level data before the 2009-10 school year, which limits my ability to examine temporal dynamics in LIHTC effects (e.g., examining different treatment window lengths). 68 Sample Table 3.1 shows how QCTs and non-QCTs differ along the key variables examined in this study. Because QCT status is based on poverty and income, these census tracts should appear more disadvantaged than non-QCTs. Statistics on neighborhood demographics confirms this is the case: QCTs have much smaller shares of White residents (26% vs. 86%) and larger shares of Black (62% vs. 8%) and Hispanic residents (7% vs. 3%) than non-QCTs. Additionally, QCTs tend to have lower home ownership rates (43% vs. 81%) and much lower median household incomes ($23k vs. $51k). Consistent with the “qualified” assignment criteria, QCTs have a higher poverty rate (34% vs. 8%) and a larger share of households with incomes that are <60% AMGI (61% vs. 26%). On average, QCTs receive more LIHTC construction or rehabilitation projects (0.52 vs. 0.14) and these projects results in 31 more LIHTC-funded rental units in QCTs. Essentially all of these units in QCTs are offered at below-market rates for low-income tenants. The descriptive statistics also show that QCTs have more students that have changed schools (28% vs. 14%) and neighborhoods (24% vs. 13%) in the past year. Students in QCTs are homeless (4% vs. 2%) at higher rates and they have lower attendance rates (88% vs. 94%). These descriptive results, however, do not show whether LIHTC investment causes changes in student mobility, homelessness, or attendance. I explain how I generate causal estimates next. Methods To generate causal estimates of LIHTC investment, I exploit a discontinuity in the LIHTC funding formula that generates exogenous variation in affordable housing due to QCT status. Other than the concentration of LIHTC units, we would expect census tracts just above and below the threshold to be similar in other respects (Baum-Snow & Marion, 2009; Brummet & Bartalotti, 69 2016; O. Jackson & Kawano, 2015). In what follows, I describe how I leverage the funding formula (and the exogenous variation in affordable housing it creates) in a fuzzy regression discontinuity design (RDD) to generate causal estimates, but first I begin with an explanation of how I address the multiple running variables that determine QCTs. Attending to Multiple Running Variables Typically, RDDs utilize a single running variable that dictates treatment status. QCT status, however, is assigned using multiple eligibility criteria (i.e., poverty rate and share of households with incomes that are 60% of AMGI) so there are multiple running variables to choose from. Figure 3.2 show the probability of being a QCT based on poverty rate (see panel A) and income share (see panel B). The dashed line shows the eligibility threshold, and the gray line shows the predicted probabilities of being a QCT at each value of the running variable. The figure shows a clear discontinuity in the probability of being a QCT based on either eligibility criteria. Yet, when we look at values to the left of the threshold —values where the census tract would be considered ineligible for QCT status based on the respective running variable examined—there are non-zero probabilities that show some of these tracts are QCTs. These are census tracts where QCT status “binds” only on one running variable, or census tracts that are considered qualified based on either their value on the poverty rate or income share running variable but not both. As a result, these census tracts do not appear to comply with the eligibility cutoff when the non-binding running variable is examined. The fact that neither running variable alone perfectly determines QCT status adds another layer of “fuzziness” to the fuzzy RDD approach. To account for multiple running variables, following Reardon and Robinson (2012) and Wong et al. (2013), I generate a centered binding score that perfectly 70 determines QCT status 34 : 1 ( =max (3<Z-[\]^_\- ' ,!+a<=-b ℎ_[- ) ) (1) Where 1 ( is the maximum of the two running variables that have been centered at their respective cutoffs, such that QCT=1 if 1 ( ≥0 and QCT=0 if 1 ( <0. 1 ( is a single centered binding score that dictates QCT status perfectly, as shown in Figure 3.1C where the probability of being a QCT is 1 at and after the threshold and 0 before the threshold. Fuzzy RDD Fuzzy RDD is appropriate when there is a discontinuity in the probability of treatment. In the case of LIHTC, because QCT status generates exogenous variation in affordable housing investment, there is a discontinuity in the likelihood of receiving LIHTC investment around the QCT threshold. I use QCT status as an instrument for the number of units funded by LIHTC investment and examine the effects of LIHTC investment induced by additional funds (i.e., QCT status) on student outcomes. I estimate a fuzzy RDD using local quadratic regression with triangular kernel weights and bias-adjusted optimal bandwidths (Bartalotti & Brummet, 2017; Calonico, Cattaneo, Farrell, & Titiunik, 2017; Calonico, Cattaneo, & Titiunik, 2014; Cattaneo, Idrobo, & Titiunik, 2017). 35 In the first stage, I establish the relationship between QCT status and LIHTC investments conditional on the centered binding score. First stage 34 An alternative approach (referred to as the univariate approach) is to select only one running variable and drop the qualified census tracts that do not bind on that running variable. Results from the univariate approach are qualitatively similar (although first stage results are only significant for the poverty rate eligibility variable) and are available upon request. I choose to use the binding score approach because it allows me to retain all census tracts. Furthermore, the univariate approach requires the use of adhoc bandwidths, not data-driven ones, due to a lack of variability in observations to the left of threshold. 35 The robustness check section includes checks on the local regression specification and bandwidth selection. All regression models are implemented using the rdrobust command in Stata 16, which incorporates recent advancements in non-parametric RDD estimation (Calonico et al., 2017, 2014). 71 e!2fg ( =h(g-+i)+,)+jba<[ - ( )+6 ' kgf ( +l / + 7 ( (2) I estimate LIHTC development (e!2fg ( ) between 2004 and 2012 defined as three outcomes: 1) the number of LIHTC units, 2) the number of LIHTC units at below-market rates for low- income tenants, and 3) the number of LIHTC projects. I regress these outcomes on kgf ( , an indicator variable for whether student i lives in a QCT. Equation 2 is estimated while controlling for a vector of baseline census tract characteristics from the 2000 decennial census related to demographics and housing, l / , including the fraction of Black residents, fraction of Hispanic residents, log total housing units and the median household income. 36 Standard errors are clustered at the census tract level. Second stage ; ( =h(g-+i)+,)+jba<[ - ( )+/ ' e!2fg m ( +l / + 7 ( (3) In the second stage equation, ; ( represents four respective outcomes: 1) the likelihood that student i was new to their census tract, 2) the likelihood that student i was new to their school, 3) the likelihood that student i was identified as homeless, and 4) the share of school days attended by student i. I leverage the values of e!2fg m ( , the number of LIHTC-funded units induced by additional funds given to projects in QCTs, from the first stage equation. / ' , the coefficient on e!2fg m ( , provides the causal impact of LIHTC units on these outcomes. This coefficient captures the difference in the likelihood a student is mobile or homeless, and the difference in their attendance rate, respectively, that can be attributed to LIHTC investment induced by additional 36 Following common practice (Calonico & Cattaneo, 2014), I use controls to increase the precision of the regression estimates. Results with and without the use of controls are presented. 72 funds for neighborhoods around the QCT threshold. I incorporate the same vector of baseline census tract characteristics, l / as equation 2 and cluster standard errors to the census tract level. Both first and second stage equations are estimated using the same bandwidths, which are calculated using a cluster-robust mean squared error (MSE) bandwidth selection procedure (Bartalotti & Brummet, 2017; Calonico et al., 2014). Assumptions for Internal Validity Use of the fuzzy RDD relies on two key assumptions. First, QCT status (i.e., the instrument) should strongly predict the number of LIHTC units (i.e., the endogenous regressor). Second, conditional on the running variable, no other relevant factors should change discontinuously at the QCT threshold. Put another way, after controlling for the running variable, a tract’s location on either side of the QCT threshold should be as good as random. The results from the first stage equation, discussed in the results section below, provide evidence to verify if the first assumption is met. To investigate the second assumption, I look for evidence of systematic manipulation of the running variable around the QCT threshold. Systematic manipulation would suggest that residents manipulated their 2000 decennial census data to qualify as a QCT, which is unlikely to begin with. In Figure 3.3, I show histograms of the distribution of census tracts along the poverty rate (A), income share (B), and centered binding score (C). None of these histograms suggest there is any heaping, which would be indicated by a substantially greater distribution of census tracts just after the eligibility threshold than before it. I further examine evidence of heaping using a formal test developed by Cattaneo, Jansson, and Ma (2018) that assesses if there is a discontinuity in the density of units around the cutoff. I find no statistically significant evidence that there is a discontinuity in the density of variables around the poverty rate (n =0.57), income share (n =0.94), or centered binding score (n =0.75). 73 Results The Effect of QCT Status on LIHTC Investment In order for the fuzzy RDD to be appropriate for this study, I first must show there is a discontinuity in LIHTC investment around the QCT eligibility cutoff. Figure 3.4 displays the graphical first stage results, where the black line represents a fourth order global polynomial fit and the gray dots are sample averages for binned student observations within the bandwidth examined. There is a large discontinuity in LIHTC investment around the cutoff when using the centered binding score, including in the number of units (panel A), number of units offered at below-market rents (panel B), and the number of projects receiving LIHTC investment (panel C). There appears to be substantially more units or projects just to the right of the cutoff than just to the left of the cutoff. Table 3.2 reports first stage regression results estimated using equation 2. 37 Consistent with the graphical results, the coefficients show there are substantially more units or projects at the threshold. In particular, there are 51 to 72 more units receiving LIHTC investment, of these units 45 to 61 of them are offered at below market rates for low-income tenants. These estimates are significant with (p<.01) and without controls (p<.05). There are also more LIHTC projects, ranging from 0.3 to 0.6, as the census tract crosses the QCT threshold, although estimates are only statistically significant at the 0.05 level with the inclusion of controls. The Effect of LIHTC Investment on Student Mobility, Homelessness, and Attendance I leverage the discontinuity in LIHTC investment around the QCT cutoff to generate causal estimates of LIHTC investment induced by additional funds on student mobility, homelessness, and attendance. Table 3.3 shows results for equation 3, where I use the number of units (in 100s) induced by additional LIHTC funds to estimate the likelihood a student is new to their school and 37 I use a nonparametric regression approach because, as seen in Figure 3.4, the functional form between the running variable and the outcome is not readily apparent. 74 new to their tract, respectively, and that the student is homeless. 38 An additional 100 units (equivalent to over two times the average number of LIHTC units observed in a QCT) could make a sizeable contribution to addressing the affordable housing shortage for low-income renters (Getsinger, Posey, MacDonald, Leopold, & Abazajian, 2017). Across student outcomes, I find no statistically significant effects of a 100-unit LIHTC investment on student mobility, homelessness, or attendance in neighborhoods around the QCT threshold. 39 Subsample Analysis of Students in Underserved Education Programs & Communities Recent research on the impacts of LIHTC shows that the policy can have heterogeneous effects based on who the development impacts and where it occurs (e.g., Diamond & McQuade, 2019). I examine whether LIHTC has different effects for the students and communities where additional investment may matter the most: underserved students already targeted by education programs (Title I and the Migrant Education Program [MEP]) and census tracts with high baseline rent burden (i.e., the share of residents dedicating 30% or more of monthly income to rent greater than the state median using the 2000 decennial census). To do so, I re-estimate equations 2 and 3 on a subsample of 1) students who quality for Title I 40 , which provides federal funds that are used to improve education for low-income and low performing students, or MEP, which disburses federal funds to assist students whose families have migrated for work in the past three years such as farm workers (referred to as underserved students), and 2) on a subsample of students who do not qualify for Title I and do not quality for MEP (referred to as relatively advantaged students). 38 Results are qualitatively similar when using the number of LIHTC units at below-market rates, see Appendix Table A3.1. 39 Results from a statistical power test suggest that the sample size would be large enough to detect effects equivalent to -0.5 to 0.5 at a power of 80. 40 This includes students who attend schools that operate school-wide Title 1 programs (e.g., greater than 40% of students qualify for the free and reduced-price lunch program). This includes the vast majority of FRL-eligible students. I do not limit the sample using FRL-eligibility because all students who are homeless are also FRL- eligible; thus, I am not able to estimate the effects on student homelessness when I limit the sample to students who are not FRL eligible. 75 Table 3.4 shows the first stage results for underserved students and relatively advantaged students. While I observe significant first stage effects for the neighborhoods in which both of these subsamples live (i.e., underserved and relatively advantaged students, respectively), when conditioning on relatively advantaged students, QCT status induces more LIHTC units and projects in the neighborhoods around the cutoff. For example, when examining underserved students, crossing the QCT threshold is associated with 52 units (panel A, column 2) compared with 128 units when examining students not eligible for these programs (panel B, column 2). Similar differences occur when examining below-market units (columns 3 and 4) and LIHTC projects (columns 5 and 6). These differences suggest that developers induced by QCT status build even more units in areas where more advantaged students live compared to areas where underserved students live. Indeed, Baum-Snow & Marion (2009) use a similar estimation approach and find developers differentially select QCTs that are gentrifying. The fact that there are larger first stage results when examining relatively advantaged students does not invalidate the RDD identification strategy. To some extent, there should be sorting on unobservables on groups of tracts on either side of the cutoff. However, as long as the processes by which the sorting occurs is the same on either side then the sorting is differenced out by the RDD estimator (Baum-Snow & Marion, 2009). Table 3.5 shows the second stage results for underserved students and relatively advantaged students. Like the main results, after limiting the sample to underserved students (panel A), I still do not find statistically significant evidence that LIHTC investment affects student outcomes in neighborhoods around the QCT threshold. Conditioning on relatively advantaged students, I find that a 100-unit LIHTC results in -0.06 to -0.07 percentage point reduction in the likelihood of being new to the census tract (columns 1 and 2) and -0.05 percentage point reduction 76 in the likelihood of being new to the school (columns 3 and 4) in neighborhoods around the QCT threshold. In other words, in neighborhoods around the QCT threshold, relatively more advantaged students experience reductions in mobility due to LIHTC investment when compared to other similarly advantaged students, but the same is not true for underserved students. These reductions are quite small, however, when considering both the size of the LIHTC investment and the average neighborhood and school mobility rates. For context, a 100-unit investment reduces neighborhood mobility by 0.17 to 0.20 SD and school mobility by 0.14 SD. I explore what may be driving these results in the next section. Table 3.6 shows the first and second stage results for high rent burden census tracts. While the first stage results are still statistically, similar to the main results, I find non-significant second stage results. In other words, in census tracts with high rent burdens and thus where there is demand for affordable housing, changes in LIHTC do not have a discernible effect on student outcomes. Examining Housing and Spillover Effects Relatively more advantaged students may experience reduced school and neighborhood mobility because of the three mechanisms identified in the conceptual framework. To identify which of these mechanisms may explain these results, I look for the presence of housing, spillover, and composition effects for students that are not eligible for Title I or MEP. 41 Housing effects could be driving the subsample results if relatively more advantaged students experience greater stability because they move into LIHTC housing. I do not have access to students’ exact addresses so I cannot identify which students live in LIHTC housing to generate a true measure of housing effects. Instead, I look for evidence of housing effects by limiting the subsample of students who live in QCTs to those who also live in a census block that receives LIHTC investment. 42 Students 41 Results for the entire sample are similar to the results I describe in this section. 42 Census blocks are roughly the size of a city block in urban areas. 77 living in these census blocks are the most likely to live in LIHTC-funded housing and, consequently, would be the most likely to experience housing effects. Results from this subsample analysis are displayed in Table 3.7 and are largely not significant because of the smaller sample, particularly the number of students to the right of the QCT threshold. However, even with this smaller sample, students living in the same block as LIHTC investment experience reductions in neighborhood mobility and the magnitude is similar to the previous results, although only significant at conventional levels with the inclusion of controls (see columns 1 and 2). These results suggest that relatively more advantaged students may be capturing LIHTC housing and, as a result, experience reduced neighborhood mobility. There are not statistically significant effects for school mobility, homelessness, and attendance. Spillover effects of LIHTC investment could also reduce mobility for students who live close to a LIHTC project. I explore the presence of spillover effects by limiting the sample of students who live in QCTs to students who live close to LIHTC investment but do not live in a block with LIHTC investment. Students living close to LIHTC investment are those that live in census blocks without LIHTC investment but where there is LIHTC investment within one kilometer from their block’s center. Other studies use a similar ring approach (and one kilometer ring radius) to identify spillovers (Baum-Snow & Marion, 2009; Diamond & McQuade, 2019; Amy E. Schwartz et al., 2006). Table 3.8 shows that students living in these blocks experience reductions in school mobility and in the likelihood of being homeless (see columns 3-6). While the results for neighborhood mobility are no longer statistically significant, they are a similar magnitude to the main results. These results suggest that relatively more advantaged students benefit from positive spillover effects of LIHTC investment. It is unusual that school mobility is reduced for students living near LIHTC investment, but neighborhood mobility is not. Typically, 78 school and neighborhood mobility happen together (e.g., changing homes, changing schools). One potential confounding factor could be the presence of charter schools. If other documented spillovers of LIHTC development (e.g., increased property values for low-income areas and lower property values for high-income area) deters charter school growth when compared to similar areas without LIHTC development, then perhaps reductions in school mobility are a function of fewer school options (less school choice) in areas with LIHTC development. Another explanation for these results could be composition effects. That is, relatively more advantaged students may appear to have reduced mobility because relatively more stable students are drawn into communities by LIHTC investment. I check for composition effects by examining whether the composition of census tract residents changes between 2000 and 2013 using data from the decennial census and the American Community Survey. 43 I re-estimate the second stage equation (equation 3) using the difference in the following census tract characteristics between 2013 and 2000: share of residents with at least a bachelor’s degree and log median household income. In Table 3.9, the composition changes detected do not necessarily point to more stable students moving into LIHTC communities. For instance, there is evidence of statistically significant and small reductions in the share of residents with a college degree (see columns 1 and 2). Such changes, however, are the opposite of what would be expected if more stable residents are entering communities with LIHTC investment. The reported results for relatively more advantaged students do not seem to be driven by composition effects. 44 Robustness & Sensitivity Checks 43 I draw on 2000 data from the decennial census data and I use 2013 data from the 5-year American Community Survey (ACS) estimates. The ACS 5-year estimates cover a range of years, so I use the 2010-15 ACS because its midpoint is 2013. The 2010-15 ACS census tracts estimates are generated using the 2010 tract geographies. To ensure comparability with the 2000 tract geographies used by the 2000 decennial census, I interpolate the ACS data to the 2000 tract geographies using tract weights from the Longitudinal Tract Database (Logan, Stults, & Xu, 2016). 44 I also look for evidence of housing, spillover, and composition effects for students who are considered underserved by education programs and do not detect statistically significant evidence of either mechanism. 79 I check the robustness and sensitivity of my estimation strategy in several ways. As previously mentioned, one of the key assumptions of RDD is, conditional on the running variable, no other relevant factors should change discontinuously at the QCT threshold. I further assess this assumption by examining whether QCT status and the number of LIHTC units during the treatment period, respectively, predicts census tracts’ baseline characteristics (collected in the 2000 decennial census). I examine the baseline fraction of Black residents, fraction of Hispanic residents, log housing units and log median income. If there is evidence of a discontinuity at the threshold in these baseline characteristics, then this would suggest that there were systematic differences between communities around the threshold even before the release of QCT status and subsequent LIHTC investment. Appendix Table A3.2 shows non-significant effects of QCT status (panel A) and non-significant effects of LIHTC units (panel B) on several baseline characteristics. Another way to think of the key RDD assumption is that, in the absence of QCT status, the regression function should be continuous for the number of LIHTC units (Cattaneo et al., 2017). If there were not additional funds available for projects in QCTs, LIHTC development should be continuous across the running variable (i.e., no abrupt changes). I investigate this assumption by examining whether there is a discontinuity in LIHTC development at points other than the QCT threshold. I re-estimate the first stage for the number of LIHTC units (equation 2) using placebo thresholds for control (students in non-QCTs) and treatment (students in QCTs) observations separately. The thresholds were selected to include 0.5 SD, 1 SD, and 1.5 SD of the running variable. Significant results at the placebo thresholds would cast doubt that LIHTC development is continuous in the absence of the discontinuity in the funding formula. Appendix Table A3.3 shows the placebo cutoff results, which do not show statistically significant results. 80 Additionally, I assess how sensitive results are to the removal or addition of observations by re-estimating results using different bandwidths. I re-estimate equations 2 (number of LIHTC units only) and 3 using a different data-driven bandwidth selector (coverage error probability [CER]- optimal selector) and I also check whether results are sensitive to multiplying the MSE bandwidths by a constant. Appendix Table A3.4 shows results from equation 2 and Table A3.5 shows results from equation 3. Results are qualitatively similar to the main specification across these bandwidths, with significant first stage effects and non-significant second stage effects. Lastly, I check the sensitivity of results to the local regression specification by re-estimating equations 2 and 3 using a local linear regression (Appendix A3.6). I find that the magnitudes of the first stage results are smaller, although these results are still significant. Limitations There are a few important limitations that may preclude my ability to detect effects of LIHTC on the outcomes I examine. First, due to data constraints, I cannot examine the near-term impacts of LIHTC development. LIHTC development may have the largest impacts in the year residents move in to the unit—when these developments are initially completed. The impacts of LIHTC development may fade overtime. Due to data constraints, I cannot observe key student outcomes in the year of or the year after LIHTC development for the far majority of years development occurred (2004 to 2009). As a result, in some cases, I am examining the effects of LIHTC development that occurred 2-8 years from the school year in which outcomes were measured (2013-2014). I may be unable to detect effects of LIHTC development because any initial impacts that occurred when the development was newly established have faded with time. Second, I do not have access to detailed student address data that would allow me to identify which students live in LIHTC development. This means I cannot directly estimate housing effects for students who live in 81 LIHTC properties. Discussion As educational leaders and policymakers continue to prioritize improving educational outcomes for underserved students, there is a firm understanding that programs outside of education may be necessary. In this study, I consider the joint impacts of housing and education programs to understand whether LIHTC, the federal government’s largest affordable housing development program, produces the conditions (i.e., reducing homelessness and mobility, and increasing attendance) that promote students’ academic success. The results show that a discontinuity in the LIHTC funding formula generates exogenous variation in LIHTC investments: students living in census tracts just above the QCT threshold experience greater LIHTC investment than do those living in census tracts just below the threshold. Compared to other published papers, QCT status in Michigan between 2004 and 2012 generated larger increases in both the number of units and projects funded. These larger estimates are likely due to differences in the types of projects examined. For example, I examine all LIHTC-funded projects (new construction and rehabilitation) and find that QCTs receive 0.3 to 0.6 additional projects. Meanwhile, others examine only new construction projects and find that QCT status induced between 0.05 to 0.12 additional projects (Baum-Snow & Marion, 2009; Brummet & Bartalotti, 2016). Differences may also result from the time period and the state context, both which have not been previously studied. LIHTC investment induced by additional funds largely does not result in statistically significant effects on student mobility, homelessness, and attendance, however. This is true both when examining students and neighborhoods that may be the most sensitive to affordable housing development—underserved students who are already identified as such by educational programs and students who live in neighborhoods with high mobility and rent burden. Instead, I observe that 82 LIHTC investment nominally reduces school and neighborhood mobility for students who are relatively advantaged (i.e., students who do not attend Title I schools or qualify for Title I assistance, and students who are not identified by the MEP). Relatively more advantaged students seem to be benefitting from housing and/or positive spillover effects of living in or near LIHTC investment. Given this unexpected finding, the question remains: why would these students benefit? First, even though these students are not identified as underserved by education programs, they may still be low income enough to qualify for LIHTC’s subsidized units. LIHTC is designed to house families that fall along an income gradient of low income, very low income, or extremely low income (O’Regan & Horn, 2013). Relatively advantaged families that qualify as low income may move into LIHTC housing because they face fewer constraints than families with extremely low incomes, who are likely identified as underserved by educational programs. LIHTC does not house residents with extremely low incomes at similar rates as other HUD housing programs, such as public housing and housing vouchers, because the HUD-defined rents may still be too expensive for these families (McClure, 2019; O’Regan & Horn, 2013; Williamson, 2011). Housing effects could only accrue to those families that have the funds to pay rent. Real-estate developer preferences could also shape who moves into these units. For instances, developers may prefer to house relatively advantaged families at the higher end of the HUD-specified income range. While there is less evidence of this in the context of LIHTC, developers and property managers generally screen out applicants based on credit scores, criminal conviction histories, and landlord-tenant disputes and work history (Dunn & Grabchuk, 2010). As a result, families and students that are considered underserved by education programs may not be able to secure LIHTC housing, while relatively more advantaged students may be able to gain 83 access to LIHTC housing and capture the benefits (i.e., reduced school and neighborhood mobility). Relatively more advantaged students may experience positive spillover effects because developers prefer to propose projects closer to where these students live. We know developers are more likely to build in gentrifying areas (Baum-Snow & Marion, 2009) and perhaps relatively more advantaged students live near these areas. Heterogeneous effects of LIHTC have been identified by others. Diamond and McQuade (2019) find heterogeneity in LIHTC spillover effects on housing prices where higher-income areas experience negative spillovers (price reductions) and lower-income areas experience positive spillovers (price increases). The presence of spillover effects for only relatively more advantaged students adds to the evidence that LIHTC has different implications based on who and where is impacted. The fact that this study does not detect effects of the largest affordable housing development program for underserved K-12 students is noteworthy and has important policy implications. One would expect that outcomes closely related to housing factors, such as homelessness and mobility, would be impacted by affordable housing development. The lack of significant effects for the most disadvantaged families may also portend poorly for LIHTC’s potential to improve educational outcomes for underserved students. As such, there remains an opportunity to better align LIHTC with educational and social welfare goals. Because LIHTC is largely administered by states, one way to do so is by tinkering with how SFHA administer the program. State QAPs could incentivize developers to propose projects that could be beneficial for underserved K-12 students by providing set-asides for permanent supportive housing and additional points given for projects in high-opportunity neighborhoods that provide access to better quality schools and neighborhood amenities could benefit underserved students and promote 84 education goals (Chetty et al., 2016). SFHA could work alongside state education and social service agencies in designing the evaluation criteria for rating developer proposals. Indeed, existing research has found that developers are responsive to changes in QAPs including incentives to develop in high-opportunity areas (Ellen & Horn, 2018). To reach families with extremely low-incomes, more fundamental changes may be needed to LIHTC and housing policy. The HUD-defined subsidized LIHTC rental rates are still too high for households with extremely low incomes. To reach these households, rental rates would need to be further subsidized through an expansion of housing vouchers or by reducing HUD determined rental limits and offering developers more tax credits in exchange. Either one of these changes would likely require greater federal investment in the program. Tenants with extremely low-incomes may also require more supports to find and secure housing, including help locating vacant LIHTC development and help defraying move in costs like security deposits. Real estate developers could be forced to provide these types of supports in exchange for receiving tax credits if states or the federal government mandates this. Changes in how the program is administered may result in LIHTC creating conditions that support academic success for the most educationally underserved students. The results from this study offers insight into whether LIHTC creates conditions that support student academic success and for whom. While LIHTC does not seem to support students who are identified as underserved by educational programs, it does seem to create increased stability in the lives of relatively more advantaged students. Issues related to improving educational outcomes and improving access to affordable housing will likely continue to weigh heavily on the minds of policymakers, advocates, and citizens. Modifying the LIHTC program may offer a way to improve educational outcomes and access to affordable housing. 85 References Acock, A. C. (2013). Discovering structural equation modeling using Stata. College Station, TX: Stata Press. https://doi.org/10.1080/10538712.2011.607753 Alexander, K., & Eckland, B. K. (1975). Contextual effects in the high school attainment process. American Sociological Review, 40(3), 402–416. Anyon, J. (2014). Radical possibilities: Public policy, urban education, and a new social movement. New York, NY: Routledge. Bartalotti, O., & Brummet, Q. (2017). Regression discontinuity designs with clustered data. Advances in Econometrics, 38, 383–420. https://doi.org/10.1108/S0731- 905320170000038017 Baum-Snow, N., & Marion, J. (2009). The effects of low income housing tax credit developments on neighborhoods. Journal of Public Economics, 93(5–6), 654–666. https://doi.org/10.1016/j.jpubeco.2009.01.001 Becker, G. S., & Tomes, N. (1986). Human capital and the rise and fall of families. Journal of Labor Economics, 4(3 Pt. 2), 1–47. https://doi.org/10.1086/298118 Bourdieu, P. (1973). Cultural reproduction and social reproduction. In R. Brown (Ed.), Knowledge, education, and cultural change: Papers in the sociology of education (pp. 71– 112). London: Tavistock Publications. Brummet, Q., & Bartalotti, O. (2016). The effect of low-income housing on neighborhood mobility: Evidence from linked micro-data (Economics Working Papers No. 16004). Buckner, J. C., Mezzacappa, E., & Beardslee, W. R. (2003). Characteristics of resilient youths living in poverty: The role of self-regulatory processes. Development and Psychopathology, 15(1), 139–162. https://doi.org/10.1017/S0954579403000087 86 Buckner, J. C., Mezzacappa, E., & Beardslee, W. R. (2009). Self-regulation and its relations to adaptive functioning in low income youths. American Journal of Orthopsychiatry, 79(1), 19–30. https://doi.org/10.1037/a0014796 Burdick-Will, J., Ludwig, J., Raudenbush, S. W., Sampson, R. J., Sanbonmatsu, L., & Sharkey, P. (2011). Converging evidence for neighborhood effects on children’s test scores: An experimental, quasi-experimental, and observational comparison. In G. J. Duncan & R. J. Murnane (Eds.), Whither opportunity? Rising inequality, schools, and children’s life chances. New York, NY: Russell Sage Foundation. Burdick-Will, J., Stein, M. L., & Grigg, J. (2019). Danger on the way to school: Exposure to violent crime, public transportation, and absenteeism. Sociological Science, 6, 118–142. https://doi.org/10.15195/V6.A5 Buron, L., Nolden, S., Heintzi, K., & Stewart, J. (2000). Assessment of the economic and social characteristics of LIHTC residents and neighborhoods. Cambridge, MA. Calonico, S., & Cattaneo, M. D. (2014). Regression discontinuity designs using covariates. Review of Economics and Statistics, 96(4), 710–728. https://doi.org/10.1162/REST Calonico, S., Cattaneo, M. D., Farrell, M. H., & Titiunik, R. (2017). Rdrobust: Software for regression-discontinuity designs. Stata Journal, 17(2), 372–404. https://doi.org/10.1177/1536867x1701700208 Calonico, S., Cattaneo, M. D., & Titiunik, R. (2014). Robust data-driven inference in the regression-discontinuity design. Stata Journal, 14(4), 909–946. https://doi.org/10.1177/1536867x1401400413 Caplin, A., & Leahy, J. (1998). Miracle on Sixth Avenue: Information externalities and search. The Economic Journal, 108(446), 60–74. https://doi.org/10.1111/1468-0297.00273 87 Carter, G. R. (2011). From exclusion to destitution: Race, affordable housing, and homelessness. Cityscape: A Journal of Policy Development and Research, 13(1), 33–70. https://doi.org/10.2139/ssrn.1808950 Cattaneo, M. D., Idrobo, N., & Titiunik, R. (2017). A practical introduction to regression discontinuity designs. Cambridge, UK. https://doi.org/10.1017/9781108684606 CDE and LACOE. (2018). Access to higher education: Students experiencing homelessness. Los Angeles, CA. https://doi.org/10.4324/9781351024662 Chetty, R., Friedman, J. N., Hilger, N., Saez, E., Schanzenbach, D. W., & Yagan, D. (2010). How does your kindergarten classroom affect your earnings? Evidence from Project STAR (No. 16381). Cambridge, MA. https://doi.org/10.3386/w16381 Chetty, R., Friedman, J. N., & Rockoff, J. E. (2014a). Measuring the impacts of teachers I: Evaluating bias in teacher value-added estimates. The American Economic Review, (1971). Chetty, R., Friedman, J. N., & Rockoff, J. E. (2014b). Measuring the impacts of teachers II: Teacher value-added and student outcomes in adulthood. American Economic Review, 104(9), 2633–2679. https://doi.org/10.1017/CBO9781107415324.004 Chetty, R., Hendren, N., & Katz, L. F. (2016). The effects of exposure to better neighborhoods on children: New evidence from the Moving to Opportunity experiment. American Economic Review, 106(4), 855–902. Childs, J., & Lofton, R. (2021). Masking attendance: How education policy distracts from the wicked problem(s) of chronic absenteeism. Educational Policy, 35(2), 213–234. https://doi.org/10.1177/0895904820986771 Clotfelter, C. T., Ladd, H. F., & Vigdor, J. L. (2007). Teacher credentials and student achievement: Longitudinal analysis with student fixed effects. Economics of Education 88 Review, 26(6), 673–682. https://doi.org/10.1016/j.econedurev.2007.10.002 Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of Sociology, 94, 95–120. Coleman, J. S., Campbell, E. Q., Hobson, C. J., McPartland, J., Mood, A. M., Weinfeld, F. D., & York, R. L. (1966). Equality of educational opportunity. Washington D.C. Retrieved from http://www.eric.ed.gov/PDFS/ED012275.pdf Collinson, R., & Winter, B. (2010). U.S. rental housing characteristics: Supply, vacancy, and affordability (No. 10–01). Washington D.C. Conger, R. D., Ge, X., Elder, G. H., Lorenz, F. O., & Ronald, L. (1994). Economic stress, coercive family process, and developmental problems of adolescents. Child Development, 65(2), 541–561. Cordes, S. A., Schwartz, A. E., & Stiefel, L. (2019). The effect of residential mobility on student performance: Evidence from New York City. American Educational Research Journal, 56(4), 1380–1411. https://doi.org/10.3102/0002831218822828 Cordes, S. A., Schwartz, A. E., Stiefel, L., & Zabel, J. (2016). Is neighbourhood destiny? Exploring the link between neighbourhood mobility and student outcomes. Urban Studies, 53(2), 400–417. https://doi.org/10.1177/0042098014563469 Cowen, J. M. (2017). Who are the homeless? Student mobility and achievement in Michigan 2010–2013. Educational Researcher, 46(1). https://doi.org/10.3102/0013189X17694165 Culhane, D. P., Lee, C.-M., & Wachter, S. M. (1996). Where the homeless come from: A study of the prior address distribution of families admitted to public shelters in New York City and Philadelphia. Housing Policy Debate, 7(2), 327–365. https://doi.org/10.1080/10511482.1996.9521224 89 Cutuli, J. J., & Herbers, J. E. (2014). Promoting resilience for children who experience family homelessness: Opportunities to encourage developmental competence. Cityscape, 16(1), 113–140. Retrieved from http://www.jstor.org/stable/26326860 De Gregorio, S., Dhaliwal, T. K., Owens, A., & Painter, G. (2020). Growing up homeless: Student homelessness and educational outcomes in Los Angeles (EdWorkingPaper No. 20– 334). de la Torre, M., & Gwynne, J. (2009). Changing Schools: A look at student mobility trends in Chicago Public Schools since 1995. Deng, L. (2007). Comparing the effects of housing vouchers and low-income housing tax credits on neighborhood integration and school quality. Journal of Planning Education and Research, 27(1), 20–35. https://doi.org/10.1177/0739456X07301467 Dhaliwal, T. K., De Gregorio, S., Owens, A., & Painter, G. (2021). Putting homelessness in context: The schools and neighborhoods of students experiencing homelessness. The ANNALS of the American Academy of Political and Social Science, 693(1), 158–176. Di, W., & Murdoch, J. C. (2013). The impact of the low income housing tax credit program on local schools. Journal of Housing Economics, 22(4), 308–320. https://doi.org/10.1016/j.jhe.2013.10.002 Diamond, R., & McQuade, T. (2019). Who wants affordable housing in their backyard? An equilibrium analysis of low-income property development. Journal of Political Economy, 127(3), 1063–1117. https://doi.org/10.1086/701354 DiStefano, C., Zhu, M., & Mîndrilǎ, D. (2009). Understanding and using factor scores: Considerations for the applied researcher. Practical Assessment, Research and Evaluation, 14(20). 90 Dizon-Ross, E. (2020). Neighborhood development and school diversity: Evidence from federal low-income housing tax credits. Dunn, E., & Grabchuk, M. (2010). Background checks and social effects: Contemporary residential tenant-screening problems in Washington state. Seattle Journal for Social Justice, 9(1), 319–399. Early, D. W., & Olsen, E. O. (2002). Subsidized housing, emergency shelters, and homelessness: An empirical investigation using data from the 1990 Census. Advances in Economic Analysis & Policy, 2(1). https://doi.org/10.2202/1538-0637.1011 Edwards, E. J. (2019). Hidden success: Learning from the counternarratives of high school graduates impacted by student homelessness. Urban Education. https://doi.org/10.1177/0042085919877928 Edwards, E. J. (2020). Young, Black, successful, and homeless: examining the unique academic challenges of Black students who experienced homelessness. Journal of Children and Poverty, 1–25. https://doi.org/10.1080/10796126.2020.1776688 Ellen, I. G., & Horn, K. M. (2018). Points for place: Can state governments shape siting patterns of Low-Income Housing Tax Credit developments? Housing Policy Debate, 28(5), 727– 745. https://doi.org/10.1080/10511482.2018.1443487 Ellen, I. G., Horn, K. M., & O’Regan, K. M. (2016). Poverty concentration and the Low Income Housing Tax Credit: Effects of siting and tenant composition. Journal of Housing Economics, 34, 49–59. https://doi.org/10.1016/j.jhe.2016.08.001 Ellen, I. G., O’Regan, K. M., & Voicu, I. (2009). Siting, spillovers, and segregation: A re- examination of the Low Income Housing Tax Credit program. In E. L. Glaeser & J. M. Quigley (Eds.), Housing markets and the economy: Risk, regulation, and policy. Lincoln 91 Institute of Land Policy. https://doi.org/10.4135/9781452218380.n126 Eriksen, M. D., & Rosenthal, S. S. (2010). Crowd out effects of place-based subsidized rental housing: New evidence from the LIHTC program. Journal of Public Economics, 94(11–12), 953–966. https://doi.org/10.1016/j.jpubeco.2010.07.002 Fantuzzo, J. W., LeBoeuf, W. A., Chen, C.-C., Rouse, H. L., & Culhane, D. P. (2012). The unique and combined effects of homelessness and school mobility on the educational outcomes of young children. Educational Researcher, 41(9), 393–402. https://doi.org/10.3102/0013189X12468210 Fargo, J. D., Munley, E. A., Byrne, T. H., Montgomery, A. E., & Culhane, D. P. (2013). Community-level characteristics associated with variation in rates of homelessness among families and single adults. American Journal of Public Health, 103(52). https://doi.org/10.2105/AJPH.2013.301619 Federal register. (2011) (Vol. 76). Washington D.C. Federal register. (2014) (Vol. 79). Washington D. C. Freedman, M., & McGovak, T. (2015). Low-Income housing development, poverty concentration, and neighborhood inequality. Journal of Policy Analysis and Management, 29(3), 451–478. https://doi.org/10.1002/pam Freedman, M., & Owens, E. G. (2011). Low-income housing development and crime. Journal of Urban Economics, 70(2–3), 115–131. https://doi.org/10.1016/j.jue.2011.04.001 Funderburg, R., & MacDonald, H. (2010). Neighbourhood valuation effects from new construction of low-income housing tax credit projects in Iowa: A natural experiment. Urban Studies, 47(8), 1745–1771. https://doi.org/10.1177/0042098009356122 Getsinger, L., Posey, L., MacDonald, G., Leopold, J., & Abazajian, K. (2017). The housing 92 affordability gap for extremely low-income renters in 2014. Washington D.C. Retrieved from https://www.urban.org/sites/default/files/publication/54106/2000260-The-Housing- Affordability-Gap-for-Extremely-Low-Income-Renters- 2013.pdf%0Ahttp://www.urban.org/sites/default/files/publication/54106/2000260-The- Housing-Affordability-Gap-for-Extremely-Lo Gonzalez, R. (2016). Educational rights and guidelines for youth in foster care, experiencing homelessness and/or involved in the juvenile justice system. Grossman, P., Cohen, J., Ronfeldt, M., & Brown, L. (2014). The test matters: The relationship between classroom observation scores and teacher value added on multiple types of assessment. Educational Researcher, 43(6). Gubtis, D., Shinn, M., Wood, M., Brown, S. R., Dastrup, S., & Bell, S. H. (2010). What interventions work best for families who experience homelessness? Impact estimates from the Family Options Study. Journal of Policy Analysis and Management, 29(3), 451–478. https://doi.org/10.1002/pam Hair, J. F., Anderson, R. E., Tatham, R. L., & Black, W. C. (1995). Multivariate data analysis. New York, NY: Macmillan. Hanratty, M. (2017). Do local economic conditions affect homelessness? Impact of area housing market factors, unemployment, and poverty on community homeless rates. Housing Policy Debate, 27(4), 640–655. https://doi.org/10.1080/10511482.2017.1282885 Hanushek, E. A., Kain, J. F., Markman, J. M., & Rivkin, S. G. (2003). Does peer ability affect student achievement? Journal of Applied Econometrics, 18(5), 527–544. https://doi.org/10.1002/jae.741 Hanushek, E. A., Kain, J. F., & Rivkin, S. G. (2004). Disruption versus Tiebout improvement: 93 The costs and benefits of switching schools. Journal of Public Economics, 88(9–10), 1721– 1746. https://doi.org/10.1016/S0047-2727(03)00063-X Hanushek, E., Kain, J., O’Brien, D., & Rivkin, S. (2005). The market for teacher quality (No. NBER Working Paper No. 11154). National Bureau of Economic Research. Cambridge, MA. Herbers, J. E., Cutuli, J. J., Supkoff, L. M., Heistad, D., Chan, C.-K., Hinz, E., & Masten, A. S. (2012). Early reading skills and academic achievement trajectories of students facing poverty, homelessness, and high residential mobility. Educational Researcher, 41(9), 366– 374. https://doi.org/10.3102/0013189X12445320 Hollar, M. (2014). Understanding whom the LIHTC program serves: Tenants in LIHTC units as of December 31, 2012. Washington, D.C. https://doi.org/10.2139/ssrn.2563151 Hollar, M., & Usowski, K. (2007). Low-Income housing tax credit qualified census tracts. Cityscape: A Journal of Policy Development and Research, 9(3), 153–160. Holmes, S. A. (2001, June 28). Bureau won’t distribute census data on homeless. New York Times. Horn, K. M., Ellen, I. G., & Schwartz, A. E. (2014). Reprint of “Do housing choice voucher holders live near good schools?” Journal of Housing Economics, 24, 109–121. https://doi.org/10.1016/j.jhe.2014.04.004 Jackson, C. K. (2012). Non-cognitive ability, test scores, and teacher quality: Evidence from 9th grade teachers in North Carolina (No. Working Paper 18624). NBER Working Paper Series. Jackson, C. K., Johnson, R. C., & Persico, C. (2015). The effects of school spending on educational and economic outcomes: Evidence from school finance reforms (No. 20487). 94 National Bureau of Economic Research. Cambridge. https://doi.org/10.1093/qje/qjv036.Advance Jackson, C. K., Rockoff, J. E., & Staiger, D. O. (2014). Teacher effects and teacher-related policies. Annual Review of Economics, 6(1), 801–825. https://doi.org/10.1146/annurev- economics-080213-040845 Jackson, O., & Kawano, L. (2015). Do increases in subsidized housing reduce the incidence of homelessness? Evidence from the Low-Income Housing Tax Credit (No. 15–11). Boston, MA. https://doi.org/10.2139/ssrn.2205490 Jacob, B. A. (2004). Public housing, housing vouchers, and student achievement: Evidence from public housing demolitions in Chicago. The American Economic Review, 94(1), 233–258. Jacob, B. A. (2013). Public housing, housing vouchers and student achievement: Evidence from public housing demolitions in Chicago. Journal of Chemical Information and Modeling, 53(9), 1689–1699. https://doi.org/10.1017/CBO9781107415324.004 Jargowsky, P. A., & El Komi, M. (2009). Before or after the bell? School context and neighborhood effects on student achievement. In S. Wachter, E. L. Birch, & H. Newberger (Eds.), How place matters. Philadelphia, PA: University of Pennsylvania Press. https://doi.org/10.1177/1078087416633533 Jencks, C., & Mayer, S. E. (1990). The social consequences of growing up in a poor neighborhood. In Inner-city poverty in the United States (pp. 111–186). Washington, D.C.: National Academies Press. Kerbow, D. (1996). Patterns of urban student mobility and local school reform. Washington, D. C. https://doi.org/10.1207/s15327671espr0102_5 Khadduri, J., Climaco, C., Burnett, K., Gould, L., & Elving, L. (2012). What happens to low- 95 income housing tax credit properties at year 15 and beyond? Washington D. C. Retrieved from https://www.huduser.gov/publications/pdf/what_happens_lihtc_v2.pdf Kraft, M. A., Papay, J. P., & Chi, O. L. (2020). Teacher skill development: Evidence from performance ratings by principals. Journal of Policy Analysis and Management, 39(2), 315– 347. https://doi.org/10.1002/pam.22193 Ladd, H. F., & Sorensen, L. C. (2015). Returns to teacher experience: Student achievement and motivation in middle school (National Center for Analysis of Longitudinal Data in Education Research No. Working Paper 112). Lee, B. A., & Price-Spratlen, T. (2004). The geography of homelessness in American communities: Concentration or dispersion? City and Community, 3(1), 3–27. https://doi.org/10.1111/j.1535-6841.2004.00064.x Leventhal, T., & Brooks-Gunn, J. (2000). The neighborhoods they live in: The effects of neighborhood residence on child and adolescent outcomes. Psychological Bulletin, 126(2), 309–337. https://doi.org/10.1037/0033-2909.126.2.309 Leventhal, T., & Newman, S. (2010). Housing and child development. Children and Youth Services Review, 32(9), 1165–1174. https://doi.org/10.1016/j.childyouth.2010.03.008 Logan, J. R., Stults, B. J., & Xu, Z. (2016). Validating population estimates for harmonized census tract data, 2000–2010. Annals of the American Association of Geographers, 106(5), 1013–1029. https://doi.org/10.1080/24694452.2016.1187060 Massey, D. S., & Denton, N. A. (1988). The dimensions of residential segregation. Social Forces, 67(2), 281–315. https://doi.org/10.1093/sf/67.2.281 Masten, A. S. (2001). Ordinary magic: Resilience processes in development. American Psychologist, 56(3), 227–238. https://doi.org/10.1037/0003-066X.56.3.227 96 Masten, A. S., Cutuli, J. J., Herbers, J. E., & Reed, M. G. J. (2009). Resilience in development. In S. J. Lopez & C. R. Snyder (Eds.), The Oxford Handbook of Positive Psychology, (2 Ed.) (pp. 1–28). Oxford, UK: Oxford University Press. https://doi.org/10.1093/oxfordhb/9780195187243.013.0012 Masten, A. S., Herbers, J. E., Desjardins, C. D., Cutuli, J. J., McCormick, C. M., Sapienza, J. K., … Zelazo, P. D. (2012). Executive function skills and school success in young children experiencing homelessness. Educational Researcher, 41(9), 375–384. https://doi.org/10.3102/0013189X12459883 Masten, A. S., Miliotis, D., Graham-Bermann, S. A., Ramirez, M., & Neemann, J. (1993). Children in homeless families: Risks to mental health and development. Journal of Counseling and Clinical Psychology, 61(2), 335–343. https://doi.org/10.1037/0022- 006X.61.2.335 Masten, A. S., Sesma, A., Si-Asar, R., Lawrence, C., Miliotis, D., & Dionne, J. A. (1997). Educational risks for children experiencing homelessness. Journal of School Psychology, 35(1), 27–46. https://doi.org/10.1016/S0022-4405(96)00032-5 Mawhinney-Rhoads, L., & Stahler, G. (2006). Educational policy and reform for homeless students: An overview. Education and Urban Society, 38(3), 288–306. https://doi.org/10.1177/0013124506286943 McClure, K. (2019). What should be the future of the Low-Income Housing Tax Credit program? Housing Policy Debate, 29(1), 65–81. https://doi.org/10.1080/10511482.2018.1469526 McKinney-Vento Homeless Assistance Act, Pub. L. No. § 11431-11435. (2015). Title IX, Part A of the Every Student Succeeds Act. 42. 97 Miliotis, D., Sesma, A., & Masten, A. S. (1999). Parenting as a protective process for school success in children from homeless families. Early Education and Development, 10(2), 111– 133. https://doi.org/10.1207/s15566935eed1002_2 Miller, P. M. (2011). A critical analysis of the research on student homelessness. Review of Educational Research, 81(3), 308–337. https://doi.org/10.3102/0034654311415120 Mistry, R. S., Vandewater, E. A., Huston, A. C., & McLoyd, V. C. (2002). Economic well-being and children’s social adjustment: The role of family process in an ethnically diverse low- income sample. Child Development, 73(3), 935–951. https://doi.org/10.1111/1467- 8624.00448 National Association for the Education of Homeless Children and Youth. (2010). A critical moment: Child & youth homelessness in our nation’s schools. Minneapolis, MN. National Center for Homeless Education. (2019). Federal data summary school years 2014-2015 to 2016-17: Education for homeless children and youth. Greensboro, NC. Retrieved from https://nche.ed.gov/wp-content/uploads/2019/02/Federal-Data-Summary-SY-14.15-to- 16.17-Final-Published-2.12.19.pdf Newman, S. J., & Holupka, C. S. (2015). Housing Affordability and Child Well-Being. Housing Policy Debate, 25(1), 116–151. https://doi.org/10.1080/10511482.2014.899261 O’Regan, K. M., & Horn, K. M. (2013). What can we learn about the Low-Income Housing Tax Credit Program by looking at the tenants ? Housing Policy Debate, 23(3), 597–613. https://doi.org/10.1080/10511482.2013.772909 Obradović, J. (2010). Effortful control and adaptive functioning of homeless children: Variable- focused and person-focused analyses. Journal of Applied Developmental Psychology, 31(2), 109–117. https://doi.org/10.1016/j.appdev.2009.09.004 98 Obradović, J., Long, J. D., Cutuli, J. J., Chan, C. K., Hinz, E., Heistad, D., & Masten, A. S. (2009). Academic achievement of homeless and highly mobile children in an urban school district: Longitudinal evidence on risk, growth, and resilience. Development and Psychopathology, 21(2), 493–518. https://doi.org/10.1017/S0954579409000273 Owens, A. (2010). Neighborhoods and schools as competing and reinforcing contexts for educational attainment. Sociology of Education, 83(4), 287–311. https://doi.org/10.1177/0038040710383519 Owens, A., & Candipan, J. (2020). Social and spatial inequalities of educational opportunity: A portrait of schools serving high- and low-income neighbourhoods in US metropolitan areas. Urban Studies, 56(15), 3178–3197. https://doi.org/10.1177/0042098018815049 Pavlakis, A. E. (2014). Living and learning at the intersection: Student homelessness and complex policy environments. Urban Review, 46(3), 445–475. https://doi.org/10.1007/s11256-014-0287-4 Quigley, J. M., Raphael, S., & Smolensky, E. (1999). Homeless in America, homeless in California. Journal of Urban History, 25(2), 258–270. https://doi.org/10.1177/009614429902500204 Rafferty, Y., & Shinn, M. (1991). The impact of homelessness on children. American Psychologist, 51(6), 721–736. https://doi.org/10.1177/0002764207311984 Reardon, S. F., & Robinson, J. P. (2012). Regression discontinuity designs with multiple rating- score variables. Journal of Research on Educational Effectiveness, 5, 217–244. Rendón, M. G. (2014). Drop out and “disconnected” young adults: Examining the impact of neighborhood and school contexts. Urban Review, 46(2), 169–196. https://doi.org/10.1007/s11256-013-0251-8 99 Rothstein, R. (2017). The color of law: A forgotten history of how our government segregated America. New York: Liveright Publishing Corporation. Rucinski, C. L., Brown, J. L., & Downer, J. T. (2017). Teacher-child relationships, classroom climate, and children’s social-emotional and academic development. Journal of Educational Psychology, 110(7), 992–1004. Rumberger, R. W. (2003). The causes and consequences of student mobility. The Journal of Negro Education, 72(1), 6–21. https://doi.org/10.2307/3211287 Rumberger, R. W., Larson, K. A., Ream, R. K., & Palardy, G. J. (1999). The educational consequences of mobility for California students and schools. PACE Policy Brief. Berkeley, CA. Rumberger, R. W., & Palardy, G. J. (2005). Does segregation still matter? The impact of student composition on academic achievement in high school. Teachers College Record, 107(9), 1999–2045. https://doi.org/10.1111/j.1467-9620.2005.00583.x Sacerdote, B. (2011). Peer effects in education: How might they work, how big are they and how much do we know thus far? In E. A. Hanushek, S. J. Machin, & L. Woessmann (Eds.), Handbook of the Economics of Education (Vol. 3, pp. 249–277). North Holland. https://doi.org/10.1016/B978-0-444-53429-3.00004-1 Sampson, R. J. (1997). Collective regulation of adolescent misbehavior: Validation results from eighty Chicago Neighborhoods. Journal of Adolescent Research, 12(2), 227–244. Sampson, R. J. (2012). Great American City: Chicago and the Enduring Neighborhood Effect. Chicago: University of Chicago Press. Sampson, R. J., Sharkey, P., & Raudenbush, S. W. (2008). Durable effects of concentrated disadvantage on verbal ability among African-American children. Proceedings of the 100 National Academy of Sciences of the United States of America, 105(3), 845–852. https://doi.org/10.1073/pnas.0710189104 Scally, C. P., Gold, A., Herman, C., Gerken, M., & DuBois, N. (2018). The Low-Income Housing Tax Credit: Past achievements, future challenges. Washington D.C. Retrieved from https://www.urban.org/sites/default/files/publication/98761/lithc_past_achievements_future _challenges_final_0.pdf Schwartz, Amy E., Ellen, I. G., Voicu, I., & Schill, M. H. (2006). The external effects of place- based subsidized housing. Regional Science and Urban Economics, 36(6), 679–707. https://doi.org/10.1016/j.regsciurbeco.2006.04.002 Schwartz, Amy Ellen, Stiefel, L., & Cordes, S. A. (2017). Moving matters: The causal effect of moving schools on student performance. Education Finance and Policy, (November), 87. https://doi.org/10.1162/EDFP Schwartz, H. (2010). Housing policy is school policy: Economically integrative housing promotes academic success in Montgomery County, Maryland. The Century Foundation. New York, NY. Sharkey, P., & Faber, J. W. (2014). Where, when, why, and for whom do residential contexts matter ? Moving away from the dichotomous understanding of neighborhood effects. Annual Review of Sociology, 40. https://doi.org/10.1146/annurev-soc-071913-043350 Sharkey, P., Schwartz, A. E., Ellen, I. G., & Lacoe, J. (2014). High stakes in the classroom, high stakes on the street: The effects of community violence on student’s standardized test performance. Sociological Science, 1(14), 199–220. https://doi.org/10.15195/v1.a14 Shinn, M., & Gillespie, C. (1994). The roles of housing and poverty in the origins of 101 homelessness. American Behavioral Scientist, 37(4), 505–521. https://doi.org/10.1177/07399863870092005 Swanson, C. B., & Schneider, B. (1999). Students on the move: Residential and educational mobility in America’s schools. Sociology of Education, 72(1), 54–67. U.S. Census Bureau. (2017). Declining mover rate driven by renters, Census Bureau reports. Retrieved from https://www.census.gov/newsroom/press-releases/2017/mover-rates.html U.S. Department of Education. (2020). EDFacts Data Files. Warne, R. T. (2017). Research on the academic benefits of the advanced placement program: Taking stock and looking forward. SAGE Open, 7(1). https://doi.org/10.1177/2158244016682996 Welsh, R. O. (2017). School hopscotch: A comprehensive review of K–12 student mobility in the United States. Review of Educational Research, 87(3), 475–511. https://doi.org/10.3102/0034654316672068 Williamson, A. R. (2011). Can they afford the rent? resident cost burden in low income housing Tax Credit developments. Urban Affairs Review, 47(6), 775–799. Wilson, W. J. (1996). When Work Dissappears: The Word of the New Urban Poor. Vintage. Wong, V. C., Steiner, P. M., & Cook, T. D. (2013). Analyzing regression-discontinuity designs with multiple assignment variables: A comparative study of four estimation methods. Journal of Educational and Behavioral Statistics, 38(2), 107–141. Xu, Z., Hannaway, J., & D’Souza, S. (2009). Student transience in North Carolina: The effect of school mobility on student outcomes using longitudinal data (No. 22). National Center for Analysis of Longitudinal Data in Eduation Research. Washington, D.C. 102 Tables Table 1.1. School & Neighborhood Characteristics of Homeless & Nonhomeless Students Variable (1) Homeless (2) Doubled-Up (3) Nonhomeless Diff (1)-(3) Diff (2)-(3) Panel A. School characteristics % underrepresented students of color 92.96% 93.29% 86.33% 6.62% 6.96% % FRL 86.48% 87.13% 78.95% 7.53% 8.18% % EL 33.56% 33.47% 29.43% 4.13% 4.05% % homeless 4.69% 4.93% 2.14% 2.55% 2.79% % SPED eligible 11.95% 11.97% 11.62% 0.33% 0.35% % born outside US 10.02% 9.41% 10.26% -0.25% -0.86% Avg suspension rate 2.21% 2.03% 2.05% 0.16% -0.03% Avg attendance rate 94.65% 94.85% 95.02% -0.37% -0.17% % change schools 16.43% 16.58% 14.68% 1.75% 1.90% Avg ELA Z score -0.55 -0.54 -0.30 -0.25 -0.25 Avg math Z score -0.50 -0.47 -0.30 -0.20 -0.16 Traditional public school 84.56% 85.07% 81.18% 3.38% 3.89% Magnet school 1.93% 1.75% 4.15% -2.22% -2.40% SPED school 0.27% 0.22% 0.58% -0.31% -0.36% Alternative school 3.60% 2.94% 4.64% -1.04% -1.70% Avg school disadvantage index 0.31 0.33 0.001 0.31 0.33 Panel B. Neighborhood characteristics % female-headed households 22.34% 22.75% 20.43% 1.91% 2.32% % employed 64.17% 64.24% 65.17% -1.00% -0.94% % below the poverty line 24.57% 24.28% 21.15% 3.42% 3.13% % Black residents 11.81% 12.01% 9.12% 2.69% 2.89% % high school graduates 23.61% 23.67% 22.32% 1.29% 1.35% % college graduates 10.61% 10.35% 13.46% -2.85% -3.11% Avg median income $39244.43 $39691.05 $44186.92 -$4942.49 -$4495.87 % professional occupations 11.49% 11.35% 11.89% -0.40% -0.55% % severe rent burden 34.92% 34.89% 33.36% 1.56% 1.53% % severe overcrowding 10.42% 10.15% 9.28% 1.14% 0.87% Avg neighborhood disadvantage index 0.73 0.73 0.441 0.29 0.29 Notes. Significance tests show statistical significance at p<.001 for all variables. Underrepresented students of color consist of Native American, Black, Filipino, Latinx, and Pacific Islander. FRL = Free/Reduced-Price Lunch eligible. EL = English Learner. SPED = Special Education. ELA = English/Language Arts 103 Table 1.2. Prevalence and Direction of Mobility Before Homelessness During Homelessness After Homelessness Panel A. School mobility % students that change schools 28.7% 59.5% 26.6% Average school disadvantage 0.38 0.32 0.22 % significant upward moves 38.4% 35.8% 37.4% % moves to schools in same decile 17.0% 17.7% 18.6% % significant downward moves 15.6% 17.7% 15.4% Panel B. Neighborhood mobility (doubled-up students only) % students that change addresses 36.0% 58.6% 15.8% Average neighborhood disadvantage 0.75 0.73 0.69 % significant upward moves 9.1% 12.1% 11.0% % moves to neighborhoods in same decile 49.9% 44.8% 46.7% % significant downward moves 9.1% 11.4% 12.3% Notes. Significant upwards moves are defined as moves in year t to neighborhoods or schools that are 2 or more deciles less disadvantaged than neighborhoods or schools in year t-1 (based on the neighborhood or school disadvantaged indices). Significant downwards moves are defined as moves in year t to neighborhoods or schools that are 2 or more deciles more disadvantaged than neighborhoods or schools in year t-1 (based on the neighborhood or school disadvantaged indices). Percentages do not add to 100 because modest moves up or down (1 decile) are excluded. We limit the analysis of neighborhood mobility to doubled-up students. 104 Table 2.1. Factor Loadings for School & Neighborhood Indices Variables Factor loadings Panel A. Peer advantage index Share of peers not FRL-eligible 0.79 Share of peers not homeless 0.42 Share of peers stable in school 0.66 Share of peers stable in neighborhood 0.88 Panel B. Teacher qualifications and experience index Average years of experience 0.81 Share new to the profession (1-2 years of experience) -0.79 Share with masters or doctoral degree 0.61 Share rated highly effective on educator evaluator system 0.21 Panel C. Neighborhood advantage index Share of female-headed households -0.92 Share of Black residents -0.89 Share of White residents 0.89 Share employed 0.76 Share children not living in poverty 0.80 Share not living in rent-burdened conditions 0.43 Share not living in overcrowded conditions 0.69 Panel D. Neighborhood educational and occupational attainment index Share college graduates 0.89 Share high school dropouts -0.68 Share working in professional occupations 0.40 Median household income 0.86 Notes. FRL= free- and reduced-price lunch. 105 Table 2.2. Homeless and Non-homeless Student Characteristics Variable Ever Homeless Mean Never Homeless Mean Diff White 56.3% 68.2% -11.9% Asian 0.5% 3.3% -2.8% Black 26.4% 17.7% 8.7% Hispanic 9.9% 7.0% 2.9% Other race 7.0% 3.8% 3.2% FRL 91.6% 47.1% 44.5% EL 3.9% 6.1% -2.2% SPED eligible 21.5% 13.1% 8.4% Suspensions 5.7% 2.2% 3.6% Attendance rate 88.8% 93.8% -5.0% School Mobility 30.9% 14.1% 16.9% Standardized ELA achievement -0.49 0.03 -0.52 Standardized Math achievement -0.53 0.04 -0.57 Notes. FRL= free- and reduced-price lunch, EL= English learner, SPED=special education, ELA=English language arts. Ever homeless= students who are identified as homeless at least once in the data panel, never homeless= students never identified as homeless in the data panel. All differences are statistically significant at p<.001. 106 Table 2.3. Schools and Neighborhoods of Students Experiencing Homelessness Panel A. Ever and never homeless students Variable Ever Homeless Mean Never Homeless Mean Diff Peer advantage index -0.58 0.04 -0.62 Teacher quality & qualifications index -0.15 0.01 -0.16 Neighborhood advantage index -0.26 0.02 -0.28 Neighborhood educational & occupational attainment index -0.45 0.03 -0.49 Panel B. Students when and when not homeless Variable Homeless Mean Not Homeless Mean Diff Peer advantage index -0.62 -0.56 -0.06 Teacher quality & qualifications index -0.18 -0.14 -0.04 Neighborhood advantage index -0.20 -0.30 0.10 Neighborhood educational & occupational attainment index -0.44 -0.46 0.02 Notes. The sample for Panel B is limited to ever homeless students. Ever homeless= students who are identified as homeless at least once in the data panel, never homeless= students never identified as homeless in the data panel. All differences are statistically significant at p<.001. 107 Table 2.4. Relationship between Homelessness and Outcomes (1) (2) (3) Attendance rate Math Z Score ELA Z Score Homeless -0.012*** -0.016*** -0.012*** (0.000) (0.002) (0.003) Constant 0.749*** 1.008*** 1.070*** (0.003) (0.021) (0.024) Observations 8,779,203 4,468,667 4,455,871 R-squared 0.515 0.849 0.813 Notes. All models include a vector of time-varying student characteristics, student fixed effects, and year fixed effects. Robust standard errors in parentheses. *** p<0.001, ** p<0.01, * p<0.05. Table 2.5. Relationship between School Resource Indices, Attendance, and Academic Achievement (1) (2) (3) (4) (5) (6) Attendance Rate Attendance Rate Math Z Score Math Z Score ELA Z Score ELA Z Score Homeless -0.012*** -0.009*** -0.017*** -0.019*** -0.012*** -0.016*** (0.000) (0.000) (0.002) (0.003) (0.003) (0.003) Peer advantage 0.011*** 0.011*** 0.001 0.001* -0.001 -0.001 (0.000) (0.000) (0.001) (0.001) (0.001) (0.001) Teacher qualifications & experience -0.000* -0.000 0.017*** 0.017*** 0.011*** 0.011*** (0.000) (0.000) (0.000) (0.000) (0.001) (0.001) Homeless X Peer advantage 0.004*** -0.004 -0.007* (0.000) (0.003) (0.003) Homeless X Teacher qualifications & experience -0.001 -0.002 -0.001 (0.000) (0.002) (0.003) Constant 0.760*** 0.761*** 1.026*** 1.026*** 1.084*** 1.084*** (0.003) (0.003) (0.021) (0.021) (0.024) (0.024) Observations 8,733,619 8,733,619 4,465,229 4,465,229 4,452,419 4,452,419 R-squared 0.518 0.518 0.849 0.849 0.813 0.813 Notes. All models include a vector of time-varying student characteristics, student fixed effects, and year fixed effects. Robust standard errors in parentheses. *** p<0.001, ** p<0.01, * p<0.05. 108 Table 2.6. Relationship between Neighborhood Resource Indices, Attendance, and Academic Achievement (1) (2) (3) (4) (5) (6) Attendance Rate Attendance Rate Math Z Score Math Z Score ELA Z Score ELA Z Score Homeless -0.012*** -0.012*** -0.016*** -0.017*** -0.011*** -0.014*** (0.000) (0.000) (0.003) (0.003) (0.003) (0.003) Neighborhood advantage 0.007*** 0.007*** 0.005*** 0.005*** -0.001 -0.001 (0.000) (0.000) (0.001) (0.001) (0.001) (0.001) Neighborhood educational & occupational attainment 0.001*** 0.001*** 0.002 0.002 0.005*** 0.006*** (0.000) (0.000) (0.001) (0.001) (0.001) (0.001) Homeless X Neighborhood advantage 0.001** 0.004 0.004 (0.001) (0.003) (0.003) Homeless X Neighborhood educational & occupational attainment 0.001 -0.005 -0.008 (0.001) (0.004) (0.005) Constant 0.749*** 0.749*** 1.009*** 1.009*** 1.067*** 1.067*** (0.003) (0.003) (0.021) (0.021) (0.024) (0.024) Observations 8,656,636 8,656,636 4,407,013 4,407,013 4,394,380 4,394,380 R-squared 0.517 0.517 0.849 0.849 0.813 0.813 Notes. All models include a vector of time-varying student characteristics, student fixed effects, and year fixed effects. Robust standard errors in parentheses. *** p<0.001, ** p<0.01, * p<0.05. 109 110 Table 3.1. Descriptive Statistics for Non-QCTs & QCTs Non-QCTs QCTs Mean SD Mean SD Baseline Tract Characteristics (2000) Fraction Black 0.080 0.196 0.619 0.360 Fraction Hispanic 0.026 0.033 0.065 0.130 Fraction White 0.856 0.204 0.264 0.289 Fraction <18 Years Old 0.277 0.050 0.335 0.090 Fraction >65 Years Old 0.132 0.057 0.104 0.066 Log Total Households 7.185 0.586 6.816 0.684 Fraction Occupied Housing Units 0.908 0.125 0.871 0.079 Fraction Owner-Occupied Housing Units 0.812 0.154 0.426 0.173 Median Household Income 50701.660 18804.174 23187.348 6760.289 LIHTC-related Characteristics (2004-2012) Share of Households Below 60% AMGI 0.261 0.099 0.613 0.105 Poverty Rate 0.076 0.051 0.343 0.107 Centered Binding Score -0.168 0.056 0.137 0.101 Number of LIHTC Funded Projects 0.139 0.486 0.521 1.071 Number of Units Funded 9.842 43.180 40.923 103.795 Number of Below Market Rate Units for Low Income Tenants Funded 8.995 39.491 39.556 100.755 K-12 Characteristics (2013-14 school year) K-12 Students Enrolled 599.360 356.217 370.131 290.316 Share of Students New to Census Tract 0.134 0.072 0.244 0.107 Share of Students New to School 0.143 0.074 0.275 0.106 Share of Homeless Students 0.023 0.022 0.038 0.063 Attendance Rate 0.944 0.024 0.884 0.040 Observations 2279 405 Notes. The unit of observation is the census tract. Census tracts that were not linked to K-12 students enrolled in public MI schools in the 2013-14 school year were excluded. 111 Table 3.2. The Effect of QCT Status on LIHTC Investment (1) (2) (3) (4) (5) (6) Number of units Number of units Number of below-market units Number of below- market units Number of projects Number of projects QCT status 51.22* 72.37** 44.84* 61.18** 0.321 0.635* (21.60) (23.22) (19.89) (20.97) (0.217) (0.258) Covariates N Y N Y N Y Obs BW L 41712 33267 42835 37047 106114 60279 Obs in BW R 76645 60064 75830 60064 73674 53173 Bandwidth L 0.0335 0.0287 0.0349 0.0308 0.0787 0.0503 Bandwidth R 0.0971 0.0751 0.0969 0.0752 0.0935 0.0678 Notes. ***p<.001, **p<.01, * p<.05. Obs= observations, BW=bandwidth, L=left, R=right. The unit of observation is the student. Results are estimated using a local quadratic regression with triangular kernel weights. 112 Table 3.3. The Effect of LIHTC Investment on Student Homelessness, Mobility, and Attendance (1) (2) (3) (4) (5) (6) (7) (8) New to tract New to tract New to school New to school Homeless indicator Homeless indicator Attendance Attendance LIHTC units (100s) 0.0047 0.00695 -0.0511 -0.0427 -0.0444 -0.0311 -0.0167 -0.0231 (0.0608) (0.0392) (0.0639) (0.0458) (0.0405) (0.0200) (0.0346) (0.0208) Covariates N Y N Y N Y N Y Obs BW cutoff left 29327 29327 30637 30637 33267 33267 33037 33037 Obs in BW cutoff right 52922 52922 55248 55248 60064 60064 59718 59718 Bandwidth L 0.0287 0.0287 0.0287 0.0287 0.0287 0.0287 0.0287 0.0287 Bandwidth R 0.0751 0.0751 0.0751 0.0751 0.0751 0.0751 0.0751 0.0751 Notes. ***p<.001, **p<.01, * p<.05. Obs= observations, BW=bandwidth, L=left, R=right. The unit of observation is the student. Results are estimated using a local quadratic regression with triangular kernel weights. 113 Table 3.4. Title I & MEP Eligibility: The Effect of QCT Status on LIHTC Investment Panel A. Eligible for Title I or MEP (1) (2) (3) (4) (5) (6) Number of units Number of units Number of below-market units Number of below- market units Number of projects Number of projects QCT status 38.96 51.94** 34.98 46.91** 0.194 0.281 (20.98) (19.11) (19.47) (17.97) (0.209) (0.209) Covariates N Y N Y N Y Obs BW L 50298 47246 52048 47246 141021 100571 Obs in BW R 60722 60722 60722 60385 61977 52174 Bandwidth L 0.0470 0.0434 0.0475 0.0434 0.0996 0.0801 Bandwidth R 0.0817 0.0804 0.0819 0.0783 0.0824 0.0683 Panel B. Not Eligible for Title I and MEP (1) (2) (3) (4) (5) (6) Number of units Number of units Number of below-market units Number of below- market units Number of projects Number of projects QCT status 136.0** 127.9** 122.0** 120.2** 0.897 1.151* (46.55) (46.40) (42.48) (41.68) (0.537) (0.473) Covariates N Y N Y N Y Obs BW L 5203 4962 5219 4976 14293 9027 Obs in BW R 2648 2648 2648 2647 2689 1936 Bandwidth L 0.0470 0.0430 0.0475 0.0434 0.0996 0.0801 Bandwidth R 0.0817 0.0800 0.0819 0.0783 0.0824 0.0683 Notes. ***p<.001, **p<.01, * p<.05. Obs= observations, BW=bandwidth, L=left, R=right. The unit of observation is the student. Results are estimated using a local quadratic regression with triangular kernel weights. 114 Table 3.5. Title I & MEP Eligibility: The Effect of LIHTC Investment on Student Homelessness, Mobility, and Attendance Panel A. Eligible for Title I or MEP (1) (2) (3) (4) (5) (6) (7) (8) New to tract New to tract New to school New to school Homeless indicator Homeless indicator Attendance Attendance LIHTC units (100s) 0.000788 -0.0149 -0.0467 -0.0532 -0.0591 -0.038 -0.0235 -0.0211 (0.0763) (0.0494) (0.0848) (0.0594) (0.0587) (0.0261) (0.0479) (0.025) Covariates N Y N Y N Y N Y Obs BW cutoff left 41240 41240 42988 42988 46736 46736 46404 46404 Obs in BW cutoff right 53433 53433 55767 55767 60722 60722 60337 60337 Bandwidth L 0.0430 0.0430 0.0430 0.0430 0.0430 0.0430 0.0430 0.0430 Bandwidth R 0.0800 0.0800 0.0800 0.0800 0.0800 0.0800 0.0800 0.0800 Panel B. Not eligible for Title I or MEP (1) (2) (3) (4) (5) (6) (7) (8) New to tract New to tract New to school New to school Homeless indicator Homeless indicator Attendance Attendance LIHTC units (100s) -0.0562* -0.0723** -0.0487*** -0.0483** -0.00924 -0.00935 0.0128 0.0114 (0.0324) (0.0356) (0.0189) (0.0204) (0.00831) (0.00866) (0.0106) (0.0108) Covariates N Y N Y N Y N Y Obs BW cutoff left 4460 4460 4673 4673 4962 4962 4949 4949 Obs in BW cutoff right 2389 2389 2506 2506 2648 2648 2639 2639 Bandwidth L 0.0430 0.0430 0.0430 0.0430 0.0430 0.0430 0.0430 0.0430 Bandwidth R 0.0800 0.0800 0.0800 0.0800 0.0800 0.0800 0.0800 0.0800 Notes. ***p<.001, **p<.01, * p<.05. Obs= observations, BW=bandwidth, L=left, R=right. The unit of observation is the student. Results are estimated using a local quadratic regression with triangular kernel weights. 115 Table 3.6. High Rent Burden Census Tracts (Above State Median) Panel A. The effect of QCT status on LIHTC investment (1) (2) (3) (4) (5) (6) Number of units Number of units Number of below-market units Number of below- market units Number of projects Number of projects QCT status 41.92* 45.75** 42.35* 48.61** 0.371 0.474* (20.10) (16.79) (18.74) (15.68) (0.227) (0.219) Covariates N Y N Y N Y Obs BW L 69204 68809 66757 61542 86782 68809 Obs in BW R 84099 100497 81868 99888 73002 80530 Bandwidth L 0.0667 0.0657 0.0637 0.0570 0.0785 0.0651 Bandwidth R 0.111 0.133 0.108 0.131 0.0970 0.105 Panel B. The effect of LIHTC investment on student outcomes (1) (2) (3) (4) (5) (6) (7) (8) New to tract New to tract New to school New to school Homeless indicator Homeless indicator Attendance Attendance LIHTC units (100s) -0.00884 -0.00296 -0.00222 -0.00432 -0.0505 -0.0372 0.000218 -0.00116 (0.0641) (0.0461) (0.071) (0.0507) (0.0425) (0.0244) (0.0339) (0.0216) Covariates N Y N Y N Y N Y Obs BW cutoff left 60892 60892 63505 63505 68809 68809 68375 68375 Obs in BW cutoff right 87955 87955 92000 92000 100112 100112 99377 99377 Bandwidth L 0.0657 0.0657 0.0657 0.0657 0.0657 0.0657 0.0657 0.0657 Bandwidth R 0.133 0.133 0.133 0.133 0.133 0.133 0.133 0.133 Notes. ***p<.001, **p<.01, * p<.05. Obs= observations, BW=bandwidth, L=left, R=right. The unit of observation is the student. Results are estimated using a local quadratic regression with triangular kernel weights. 116 Table 3.7. Housing Effects for Non-Title I & MEP Eligible Students: The Effect of LIHTC Investment on Student Homelessness, Mobility, and Attendance Panel A. The effect of QCT status on LIHTC investment (1) (2) (3) (4) (5) (6) Number of units Number of units Number of below-market units Number of below- market units Number of projects Number of projects QCT status 197.7*** 183.6*** 179.8*** 172.1*** 2.533*** 2.282*** (10.57) (9.371) (8.926) (8.048) (0.236) (0.209) Covariates N Y N Y N Y Obs BW L 4782 4782 4782 4782 4782 4782 Obs in BW R 83 83 83 83 83 83 Bandwidth L 0.0430 0.0430 0.0430 0.0430 0.0430 0.0430 Bandwidth R 0.0800 0.0800 0.0800 0.0800 0.0800 0.0800 Panel B. The effect of LIHTC investment on student outcomes (1) (2) (3) (4) (5) (6) (7) (8) New to tract New to tract New to school New to school Homeless indicator Homeless indicator Attendance Attendance LIHTC units (100s) -0.0741 -0.0788* -0.000868 0.000738 -0.0377 -0.0383 -0.00166 -0.00181 (0.0382) (0.0389) (0.0152) (0.0177) (0.0195) (0.0205) (0.00642) (0.00638) Covariates N Y N Y N Y N Y Obs BW cutoff left 4290 4290 4496 4496 4782 4782 4769 4769 Obs in BW cutoff right 69 69 75 75 83 83 83 83 Bandwidth L 0.0430 0.0430 0.0430 0.0430 0.0430 0.0430 0.0430 0.0430 Bandwidth R 0.0800 0.0800 0.0800 0.0800 0.0800 0.0800 0.0800 0.0800 Notes. ***p<.001, **p<.01, * p<.05. Obs= observations, BW=bandwidth, L=left, R=right. The unit of observation is the student. Results are estimated using a local quadratic regression with triangular kernel weights. 117 Table 3.8. Spillover Effects for Non-Title I & MEP Eligible students: The Effect of LIHTC Investment on Student Homelessness, Mobility, and Attendance Panel A. The effect of QCT status on LIHTC investment (1) (2) (3) (4) (5) (6) Number of units Number of units Number of below-market units Number of below- market units Number of projects Number of projects QCT status 172.0*** 152.6*** 152.8*** 141.3*** 2.226*** 2.059*** (29.06) (26.04) (26.39) (23.69) (0.220) (0.194) Covariates N Y N Y N Y Obs BW L 3945 3945 3945 3945 3945 3945 Obs in BW R 574 574 574 574 574 574 Bandwidth L 0.0430 0.0430 0.0430 0.0430 0.0430 0.0430 Bandwidth R 0.0800 0.0800 0.0800 0.0800 0.0800 0.0800 Panel B. The effect of LIHTC investment on student outcomes (1) (2) (3) (4) (5) (6) (7) (8) New to tract New to tract New to school New to school Homeless indicator Homeless indicator Attendance Attendance LIHTC units (100s) -0.06 -0.0744 -0.0776** -0.0717*** -0.017* -0.0152 0.00924 0.00835 (0.0343) (0.0402) (0.0244) (0.0202) (0.00864) (0.0103) (0.0088) (0.00899) Covariates N Y N Y N Y N Y Obs BW cutoff left 3551 3551 3717 3717 3945 3945 3934 3934 Obs in BW cutoff right 443 443 467 467 491 491 491 491 Bandwidth L 0.0430 0.0430 0.0430 0.0430 0.0430 0.0430 0.0430 0.0430 Bandwidth R 0.0800 0.0800 0.0800 0.0800 0.0800 0.0800 0.0800 0.0800 Notes. ***p<.001, **p<.01, * p<.05. Obs= observations, BW=bandwidth, L=left, R=right. The unit of observation is the student. Results are estimated using a local quadratic regression with triangular kernel weights. 118 Table 3.9. Composition Effects for Non-Title I & MEP Eligible Students: The Effect of LIHTC Investment on Student Homelessness, Mobility, and Attendance Panel A. The effect of QCT status on LIHTC investment (1) (2) (3) (4) (5) (6) Number of units Number of units Number of below-market units Number of below- market units Number of projects Number of projects QCT status 136.0** 127.9** 122.0** 120.2** 0.897 1.151* (46.55) (46.40) (42.48) (41.68) (0.537) (0.473) Covariates N Y N Y N Y Obs BW L 5203 4962 5219 4976 14293 9027 Obs in BW R 2648 2648 2648 2647 2689 1936 Bandwidth L 0.0470 0.0430 0.0475 0.0434 0.0996 0.0801 Bandwidth R 0.0817 0.0800 0.0819 0.0783 0.0824 0.0683 Panel B. The effect of LIHTC investment on student outcomes (1) (2) (3) (4) Diff % Bach. Degree Diff % Bach. Degree Log Median HH Income Log Median HH Income LIHTC units (100s) -0.0554** -0.0665* 0.341 0.3 (0.0201) (0.0306) (0.185) (0.183) Covariates N Y N Y Obs BW cutoff left 4962 4962 4284 4284 Obs in BW cutoff right 2648 2648 2456 2456 Bandwidth L 0.0430 0.0430 0.0430 0.0430 Bandwidth R 0.0800 0.0800 0.0800 0.0800 Notes. ***p<.001, **p<.01, * p<.05. Obs= observations, BW=bandwidth, L=left, R=right. The unit of observation is the student. Results are estimated using a local quadratic regression with triangular kernel weights. Diff= difference, Bach=bachelors, HH=household.. 119 Figures Figure 1.1. Distribution of Homeless Students in LAUSD Schools and Neighborhoods, 2016- 2017 Neighborhood Disadvantage Index A B C D 120 Notes. *** p<0.001, ** p<.01. Significance tests are conducted between the mobile and nonmobile means within each time period. Figure 1.2. Comparing Average School and Neighborhood Disadvantage When Mobile and Not Mobile 0.24 0.26 0.28 0.3 0.32 0.34 0.36 Before Homeless Homeless After Homeless Average School Disadvantage Index Nonmobile Mobile 0.68 0.7 0.72 0.74 0.76 0.78 0.8 Before Homeless Homeless After Homeless Average Neighborhood Disadvantage Index Nonmobile Mobile *** *** *** ** 121 Figure 2.1. Factors Related to Resilience Positive Effect Negative Effect All Children Promotive factors Risk factors Children Experiencing Adversity Protective factors Vulnerability factors 122 Figure 2.2. Summary of Promotive, Risk, Protective and Vulnerability Factors Across Outcomes Attendance Math Achievement ELA Achievement Promotive Peer advantage Neighborhood advantage Teacher qualifications & experience Neighborhood advantage Teacher qualifications & experience Neighborhood educational & occupational attainment Risk Teacher qualifications & experience Protective Peer advantage Neighborhood advantage Vulnerability Peer advantage 123 Figure 3.1. Conceptual Model of How LIHTC Investments Impacts Student Outcomes LIHTC investment constructing or rehabilitating units Treatment Housing effects Mechanisms Spillover effects Provides community amenities/ removes disamenities Changes in composition of peers & resources Reduces financial strain & stress Outcomes Residential mobility School mobility Homelessness Attendance 124 (A) Poverty rate running variable (B) Income share running variable (C) Centered binding score running variable Figure 3.2. Probability of QCT Status on Running Variables Notes. Dashed line represents the threshold for QCT status. Gray lines represent local linear regressions estimated separately for each side of the cutoff. The unit of observation is the student. 125 (A) Poverty rate running variable (B) Income share running variable (C) Centered binding score running variable Figure 3.3. Distribution of Running Variables Notes. Black line signifies the QCT eligibility cutoff. Census tract observations are divided into 100 bins. 126 (A) Number of units (B) Number of below-market rate units (C) Number of projects Figure 3.4. LIHTC Investment by Centered Binding Score Notes. Each graph includes a bandwidth of observations estimated using the cluster-robust MSE bandwidth. Bandwidths: panel A, left= 0.03 and right=0.08 ;panel B, left= 0.03 and right=0.08; panel C, left= 0.05 and right=0.07. The unit of observation is the student. Dots represent averages within evenly spaced bins. Black lines represent a global polynomial using a 4 th -degree polynomial. 127 Appendix Table A1.1. Characteristics of Homeless and Non-Homeless Students (1) (2) (3) (4) Variable All Homeless Doubled- Up Other Homeless Non- Homeless Difference (1)–(4) Difference (2)–(3) Asian 1.22% 1.09% 1.38% 4.32% -3.10% -0.29% Black 18.67% 17.79% 19.65% 8.98% 9.69% -1.86% Latinx 75.05% 76.52% 73.43% 74.58% 0.47% 3.09% White 3.32% 2.75% 3.96% 9.40% -6.08% -1.21% Native American 0.50% 0.44% 0.57% 0.30% 0.20% -0.13% Filipino 0.88% 1.03% 0.72% 2.06% -1.17% 0.31% Pacific Islander 0.35% 0.39% 0.30% 0.37% -0.02% 0.10% FRL 94.12% 94.37% 93.85% 78.78% 15.34% 0.52% EL 32.74% 33.73% 31.65% 29.36% 3.38% 2.07% SPED eligible 13.12% 12.17% 14.16% 11.59% 1.52% -1.99% Suspensions 3.46% 3.03% 3.94% 2.03% 1.43% -0.90% Attendance rate 92.50% 92.86% 92.10% 95.16% -2.66% 0.76% Born outside of US 13.91% 15.13% 12.55% 10.17% 3.74% 2.58% School mobility 38.16% 38.16% 38.16% 27.72% 10.44% -0.01% Residential mobility 45.69% 46.15% 45.18% 21.44% 24.26% 0.97% Exiting district 17.44% 16.18% 18.79% 14.70% 2.74% -2.61% Std ELA achievement -0.32 -0.34 -0.30 0.01 -0.33 -0.04 Std Math achievement -0.32 -0.33 -0.31 0.01 -0.33 -0.02 Notes. Other homeless includes all non-doubled-up students (e.g., those in shelters, cars, hotels/motels, unsheltered). Significance tests show statistically significant differences between all homeless and non-homeless students at p<.001 for all variables except for Pacific Islander; and statistically significant differences between doubled-up and other homeless students at p<.001 for all variables except Pacific Islander, school and residential mobility, and standardized math achievement. FRL = Free/Reduced-Price Lunch eligible. EL = English Learner. SPED = Special Education. ELA = English/Language Arts 128 Table 1.2. Rotated Factor Loadings for School and Neighborhood Concentrated Disadvantage Indices, 2016-17 Variable Factor Loadings School Concentrated Disadvantage Index % Underrepresented minority students 0.92 % FRL eligible 0.93 % EL 0.51 % SPED 0.16 Neighborhood Concentrated Disadvantage Index % Female-headed households 0.77 % Employed -0.35 % Families below the poverty line 0.81 % Black residents 0.23 % High school graduates 0.71 % College graduates -0.87 Median log household income -0.82 % Professional/scientific/managerial occupations -0.54 % Severe rent burden 0.43 % Overcrowded 0.55 Notes. Underrepresented Minority Students includes Native American, Black, Filipino, Latinx, and Pacific Islander. FRL = Free/Reduced-Price Lunch eligible. EL = English Learner. SPED = Special Education. Factor loadings from 2016-17 school year; loadings are similar in all years. 129 Table A1.3. Schools by Homeless Student Enrollment Rate, 2016-17 School Enrollment < 2% homeless >2% and <3% homeless >3% and <5% homeless >5% and <9% homeless >9% homeless Percent of district schools 34 19 25 17 5 130 Table A1.4. Comparing Homeless Students Who Exit the District to Those who Remain Panel A. Homeless student sample for school mobility analyses Variable Exit while homeless No exit Difference Asian 1.24% 1.02% 0.22% Black 20.97% 16.26% 4.71% Latinx 72.53% 78.13% -5.60% White 3.68% 2.90% 0.78% Native American 0.55% 0.44% 0.12% Filipino 0.69% 0.92% -0.23% Pacific Islander 0.34% 0.35% -0.01% FRL 89.55% 93.32% -3.77% EL 29.01% 34.36% -5.35% SPED eligible 11.45% 14.49% -3.05% Suspensions 5.66% 2.60% 3.06% Attendance rate 91.10% 93.63% -2.54% Born outside of US 15.55% 11.25% 4.30% School mobility 40.58% 37.64% 2.93% Std ELA achievement -0.30 -0.33 0.03 Std Math achievement -0.30 -0.30 0.00 Change addresses 49.89% 38.42% 11.47% School disadvantage index 0.23 0.34 -0.11 Neighborhood disadvantage index 0.71 0.73 -0.02 Panel B. Doubled-up student sample for neighborhood mobility analyses Variable Exit while doubled up No exit Difference Asian 1.11% 0.88% 0.23% Black 20.53% 16.17% 4.35% Latinx 73.70% 78.47% -4.77% White 3.06% 2.59% 0.47% Native American 0.50% 0.40% 0.11% Filipino 0.80% 1.04% -0.24% Pacific Islander 0.30% 0.45% -0.16% FRL 89.98% 93.12% -3.14% EL 31.09% 33.39% -2.30% SPED eligible 10.81% 13.38% -2.57% Suspensions 5.27% 2.29% 2.97% Attendance rate 91.39% 93.73% -2.33% Born outside of US 16.13% 11.60% 4.52% School mobility 40.49% 37.61% 2.88% Std ELA achievement -0.34 -0.33 -0.01 Std Math achievement -0.32 -0.31 -0.02 Change addresses 51% 40% 11.26% 131 School disadvantage index 0.28 0.34 -0.07 Neighborhood disadvantage index 0.74 0.73 0.01 Notes. Significance tests show statistical significance at p<.001 for all variables except for Pacific Islander. FRL = Free/Reduced-Price Lunch eligible. EL = English Learner. SPED = Special Education. ELA = English/Language Arts. We limit the analysis of neighborhood mobility to doubled-up students. 132 Table A1.5. Disadvantage Deciles for Homeless Students, Year Before Mobility Disadvantage Decile Freq. Percent Panel A. School Disadvantage Index 1 1295 2.0% 2 3559 5.6% 3 3707 5.8% 4 4140 6.5% 5 6697 10.6% 6 8222 13.0% 7 8926 14.1% 8 8850 14.0% 9 9343 14.7% 10 8677 13.7% Panel B. Neighborhood Disadvantage Index (doubled-up students only) 1 279 1.0% 2 433 1.5% 3 1011 3.5% 4 1523 5.2% 5 2023 7.0% 6 2362 8.1% 7 3298 11.3% 8 4449 15.3% 9 5705 19.6% 10 7982 27.5% Notes. This table describes the disadvantage deciles of homeless students’ schools and neighborhoods in the year before they change schools or residences. We limit panel B to doubled-up students (the same sample used for the neighborhood mobility analyses). 133 Table A1.6. Fixed Effects Regression Results (1) (2) School Concentrated Disadvantage Neighborhood Concentrated Disadvantage Mobile -0.033*** 0.000 (0.003) (0.006) Years Homeless 0.020*** -0.004 (0.003) (0.006) Years After Exiting Homelessness 0.006 -0.009 (0.005) (0.008) Mobile* Years Homeless -0.003 0.001 (0.004) (0.008) Mobile* Years After -0.000 0.019* (0.005) (0.009) Constant 0.529*** 0.740*** (0.004) (0.011) Observations 190,325 83,303 R-squared 0.763 0.814 Notes. *** p<0.001, ** p<0.01, * p<0.05. Standard errors are clustered at the student level and are shown in parentheses. Student and grade-level fixed effects are included in the model. We limit the analysis of neighborhood mobility to doubled-up students. Table A3.1. The Effect of LIHTC Investment on Student Homelessness and Mobility (using below-market rate units) (1) (2) (3) (4) (5) (6) (7) (8) New to tract New to tract New to school New to school Homeless indicator Homeless indicator Attendance Attendance Below-market rate LIHTC units (100s) -0.000863 0.00806 -0.0641 -0.0492 -0.0513 -0.0361 -0.0188 -0.0268 (0.0702) (0.0454) (0.0729) (0.0524) (0.047) (0.0233) (0.0405) (0.0247) Covariates N Y N Y N Y N Y Obs BW cutoff L 32713 32713 34146 34146 37047 37047 36801 33037 Obs in BW cutoff R 52922 52922 55248 55248 60064 60064 59718 59718 Bandwidth L 0.0308 0.0308 0.0308 0.0308 0.0308 0.0308 0.0308 0.0287 Bandwidth R 0.0752 0.0752 0.0752 0.0752 0.0752 0.0752 0.0752 0.0752 Notes. ***p<.001, **p<.01, * p<.05. Obs= observations, BW=bandwidth, L=left, R=right. The unit of observation is the student. Results are estimated using a local quadratic regression with triangular kernel weights. 134 135 Table A3.2. The Effect of QCT Status and LIHTC Investment on Pre-determined Census Tract Characteristics (using below-market rate units) Panel A. Reduced form estimates (1) (2) (3) (4) Fraction Black Residents Fraction Hispanic Residents Log Housing Units Log Median Income QCT status 0.0657 0.0639 -0.0755 1,064 (0.106) (0.0750) (0.119) (1,566) Covariates N Y N Y Obs BW cutoff L 270173 143617 284141 123393 Obs in BW cutoff R 104612 97124 81019 97124 Bandwidth L 0.131 0.0956 0.134 0.0860 Bandwidth R 0.134 0.123 0.102 0.123 Panel B. IV estimates (1) (2) (3) (4) Fraction Black Residents Fraction Hispanic Residents Log Housing Units Log Median Income Below-market rate LIHTC units (100s) -0.159 0.291 -0.5 2328 (0.338) (0.236) (0.499) (4782) Covariates N Y N Y Obs BW cutoff L 33267 33267 33267 33267 Obs in BW cutoff R 60064 60064 60064 60064 Bandwidth L 0.0287 0.0287 0.0287 0.0287 Bandwidth R 0.0751 0.0751 0.0751 0.0751 Notes. ***p<.001, **p<.01, * p<.05. Obs= observations, BW=bandwidth, L=left, R=right. The unit of observation is the student. Results are estimated using a local quadratic regression with triangular kernel weights. 136 Table A3.3. The Effect of QCT Status on the Number of LIHTC Units Using Placebo Thresholds (1) (2) (3) (4) (5) (6) Treatment Treatment Treatment Control Control Control Placebo=.10 Placebo=.15 Placebo=.20 Placebo=-.10 Placebo=-.15 Placebo=-.20 QCT status -40.20 -13.79 -181.4 -40.73 9.150 -10.97 (59.50) (25.09) (116.1) (31.51) (13.61) (5.917) Covariates Y Y Y Y Y Y Obs BW L 18989 26976 10158 214181 199670 218714 Obs in BW R 44232 16019 7937 41972 128952 239614 Bandwidth L 0.0241 0.0458 0.0296 0.0509 0.0259 0.0131 Bandwidth R 0.0752 0.0390 0.0432 0.0173 0.0260 0.0222 Notes. ***p<.001, **p<.01, * p<.05. Obs= observations, BW=bandwidth, L=left, R=right. The unit of observation is the student. Results are estimated using a local quadratic regression with triangular kernel weights. 137 Table A3.4. The Effect of QCT Status on the Number of LIHTC Units using Alternative Bandwidths (1) (2) (3) (4) (5) (6) CER-Optimal BWs MSE BWs * 1.5 MSE BWs * 2 QCT status 71.30* 94.78** 49.62** 58.71*** 35.89* 43.23** (28.63) (29.86) (19.07) (17.71) (16.42) (15.47) Covariates N Y N Y N Y Obs BW L 26902 24896 52222 52222 69543 69543 Obs in BW R 43744 32398 88264 88264 110934 110934 Bandwidth L 0.0213 0.0183 0.0430 0.0430 0.0574 0.0574 Bandwidth R 0.0619 0.0478 0.113 0.113 0.150 0.150 Notes. ***p<.001, **p<.01, * p<.05. Obs= observations, BW=bandwidth, L=left, R=right. The unit of observation is the student. Results are estimated using a local quadratic regression with triangular kernel weights. 138 Table A3.5. The Effect of LIHTC Investment on Student Homelessness and Mobility Panel A. CER-Optimal BWs (1) (2) (3) (4) (5) (6) (7) (8) New to tract New to tract New to school New to school Homeless indicator Homeless indicator Attendance Attendance LIHTC units (100s) 0.0267 0.0185 -0.071 -0.0605 -0.0415 -0.016 -0.039 -0.0283 (0.0696) (0.0405) (0.0745) (0.0541) (0.0449) (0.0197) (0.0537) (0.0275) Covariates N Y N Y N Y N Y Obs BW cutoff L 21970 21970 22947 22947 24896 24896 24728 24728 Obs in BW cutoff R 28482 28482 29794 29794 32398 32398 32223 32223 Bandwidth L 0.0183 0.0183 0.0183 0.0183 0.0183 0.0183 0.0183 0.0183 Bandwidth R 0.0478 0.0478 0.0478 0.0478 0.0478 0.0478 0.0478 0.0478 Panel B. MSE BWs * 1.5 (9) (10) (11) (12) (13) (14) (15) (16) New to tract New to tract New to school New to school Homeless indicator Homeless indicator Attendance Attendance LIHTC units (100s) -0.0106 -0.0185 -0.0243 -0.0311 -0.0499 -0.0373* -0.0231 -0.0288 (0.0556) (0.0401) (0.059) (0.0439) (0.0387) (0.0215) (0.0457) (0.0259) Covariates N Y N Y N Y N Y Obs BW cutoff L 46165 46165 48142 48142 52222 52222 51877 51877 Obs in BW cutoff R 77709 77709 81182 81182 88264 88264 87667 87667 Bandwidth L 0.0430 0.0430 0.0430 0.0430 0.0430 0.0430 0.0430 0.0430 Bandwidth R 0.113 0.113 0.113 0.113 0.113 0.113 0.113 0.113 Panel C. MSE BWs * 2 (17) (18) (19) (20) (21) (22) (23) (24) New to tract New to tract New to school New to school Homeless indicator Homeless indicator Attendance Attendance LIHTC units (100s) -0.012 -0.0253 0.00284 -0.0153 -0.0538 -0.0358 -0.012 -0.0253 (0.0674) (0.048) (0.0756) (0.052) (0.04620 (0.0256) (0.0674) (0.048) Covariates N Y N Y N Y N Y Obs BW cutoff L 61523 61523 64170 64170 69543 69543 69090 69090 Obs in BW cutoff R 97441 97441 101962 101962 110934 110934 110152 110152 Bandwidth L 0.0574 0.0574 0.0574 0.0574 0.0574 0.0574 0.0574 0.0574 Bandwidth R 0.150 0.150 0.150 0.150 0.150 0.150 0.150 0.150 Notes. ***p<.001, **p<.01, * p<.05. CER=coverage error probability, MSE= mean squared error, Obs= observations, BW=bandwidth, L=left, R=right. The unit of observation is the student. Results are estimated using a local quadratic regression with triangular kernel weights. Table A3.6 First and Second Stage Results using Local Linear specifications Panel A. The effect of QCT status on LIHTC investment (1) (2) (3) (4) (5) (6) Number of units Number of units Number of below- market units Number of below- market units Number of projects Number of projects QCT status 29.82 44.48* 28.74 43.12** 0.234 0.438* (17.20) (17.52) (15.81) (16.15) (0.186) (0.203) Covariates N Y N Y N Y Obs BW L 47075 44279 46053 42835 64533 57267 Obs in BW R 46235 34929 46235 34929 43136 24949 Bandwidth L 0.0383 0.0368 0.0378 0.0345 0.0529 0.0475 Bandwidth R 0.0629 0.0500 0.0626 0.0495 0.0582 0.0403 Panel B. The effect of LIHTC investment on student homelessness and mobility (1) (2) (3) (4) (5) (6) (7) (8) New to tract New to tract New to school New to school Homeless indicator Homeless indicator Attendance Attendance LIHTC units (100s) -0.00611 -0.0193 -0.0374 -0.0506 -0.0365 -0.0229 -0.0106 -0.00792 (0.0691 (0.0474) (0.0713) (0.0491) (0.04) (0.0206) (0.0368) (0.0203) Covariates N Y N Y N Y Obs BW cutoff L 39112 39112 40865 40865 44279 44279 44016 44016 Obs in BW cutoff R 30712 30712 32105 32105 34929 34929 34734 34734 Bandwidth L 0.0368 0.0368 0.0368 0.0368 0.0368 0.0368 0.0368 0.0368 Bandwidth R 0.0500 0.0500 0.0500 0.0500 0.0500 0.0500 0.0500 0.0500 Notes. ***p<.001, **p<.01, * p<.05. Obs= observations, BW=bandwidth, L=left, R=right. K-12 students enrolled in MI schools in the 2013-14 school year linked to census tracts. The unit of observation is the student. Results are estimated using a local linear regression with triangular kernel weights. 139

Abstract (if available)

Abstract Researchers have acknowledged for some time that housing matters a great deal for students’ performance in schools. I present three dissertation papers that examine how issues related to housing affordability impact students and whether current policy solutions create conditions that promote student learning. In the first paper, my co-authors and I examine the school and neighborhood contexts of homeless students in the Los Angeles Unified School District (LAUSD). To do so, we use a combination of descriptive analyses and geospatial techniques. In the second study, I examine whether schools and neighborhoods moderate the relationship between homelessness and academic and behavioral outcomes using a restricted state administrative dataset from Michigan and a fixed-effects model. In the third study, I examine the effects of the federal government’s largest affordable housing program, the Low-Income Housing Tax Credit (LIHTC) program, on student homelessness, mobility, and attendance. I leverage a discontinuity in the LIHTC funding formula and administrative educational data from Michigan, along with publicly available data, to estimate the causal effects of this program. Jointly, these studies elevate our understanding of how housing affordability impacts students by focusing on symptoms of housing affordability, student homelessness, and a housing solution, LIHTC.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

English language development materials in Texas: a study of effectiveness and selection

PDF

Levels of interest: the effects of teachers' unions on Congress, statehouses, and schools

PDF

Teacher evaluation reform in situ: three essays on teacher evaluation policy in practice

PDF

The role of the timing of school changes and school quality in the impact of student mobility: evidence from Clark County, Nevada

PDF

Building networks for change: how ed-tech coaches broker information to lead instructional reform

PDF

Student mobility in policy and poverty context: two essays from Washington

PDF

Choosing wisely: a three paper dissertation exploring how parents evaluate and choose schools

PDF

Preparing teachers for social emotional learning driven instruction and practice

PDF

The interactions between housing and business

PDF

School resource allocation in times of economic boom and bust

PDF

Three essays on aging, wealth, and housing tenure transitions

PDF

Affordable south Los Angeles: survival, support, and different futures

PDF

Teacher curriculum supplementation as phenomenon and process

PDF

Fever dreams: the promise and limitations of Black and Latinx parent organizing during the first year of the COVID-19 pandemic

PDF

Care and social-emotional well-being: organizational conditions in policy and practice

PDF

A multi-perspective examination of developmental education: student progression, institutional assessment and placement policies, and statewide regulations

PDF

Math and the making of college opportunity: persistent problems and possibilities for reform

PDF

What do you notice? What do you wonder? A mixed-methods investigation into community science data talks

PDF

To what extent does being a former high school English learner predict success in college mathematics? Evidence of Latinx students’ duality as math achievers

PDF

Titrating the solution: the diffusion and institutionalization of the logic of continuous improvement

Asset Metadata

Creator Dhaliwal, Tasminda Kaur (author)

Core Title No place like home: a three paper dissertation on K-12 student homelessness & housing affordability

School Rossier School of Education

Degree Doctor of Philosophy

Degree Program Urban Education Policy

Degree Conferral Date 2021-08

Publication Date 07/31/2021

Defense Date 05/12/2021

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

Tag education policy,Homelessness,Housing,OAI-PMH Harvest

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Marsh, Julie (committee chair), Strunk, Katharine (committee chair), Owens, Ann (committee member), Painter, Gary (committee member)

Creator Email dhaliw20@msu.edu,tkdhaliwal@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC15672111

Unique identifier UC15672111

Legacy Identifier etd-DhaliwalTa-9958

Document Type Dissertation

Format application/pdf (imt)

Rights Dhaliwal, Tasminda Kaur

Type texts

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

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu