University of Southern California Dissertations and Theses

Designing an early warning system for Hawaii: identifying indicators of positive high school outcomes

(USC Thesis Other)

Designing an early warning system for Hawaii: identifying indicators of positive high school outcomes

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

DESIGNING AN EARLY WARNING SYSTEM FOR HAWAII:
IDENTIFYING INDICATORS OF POSITIVE HIGH SCHOOL OUTCOMES

by

Tammi J. Oyadomari-Chun

____________________________________________________________

A Dissertation Presented to the
FACULTY OF THE USC ROSSIER SCHOOL OF EDUCATION
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF EDUCATION

August 2010

Copyright 2010 Tammi J. Oyadomari-Chun

ii
ACKNOWLEDGEMENTS
Much credit for this completed dissertation goes to my husband, David, and
my son, Joshua, who believed that I could become a ―doctor of education‖ and were
more confident than I that this would be done. Josh attended the first Ed.D. class as
a one-year old, and as a four-year old, he played and took naps in my office while I
wrote. Now that mommy is done with her dissertation, Josh is going to Disneyland.
I owe a debt of gratitude to Mom and Dad who always made education a
priority for our family. Their sacrifices provided wonderful opportunities to stretch,
grow and serve.
I have been blessed with many wonderful mentors and role models in the
journey to this dissertation including David Menefee-Libey at Pomona College,
Mary Chambers at LEARN, Peg Goertz and Didi Massell at CPRE, Gail Zellman,
Cathy Stasz, and Brian Stecher at RAND, Kati Haycock at Eduation Trust, and
Linda Johnsrud at the University of Hawaii.
Dominic Brewer at USC was an amazing dissertation chair in guiding and
cajoling me to focus and finish. Ron Heck at the University of Hawaii was a
remarkable external committee member and provided tremendous help with the
quantitative analyses as well as incredibly constructive and timely feedback. Larry
Picus at USC was also a valuable committee member.
Team Dom provided advice, critical feedback, references, encouragement
and lots of peer pressure which were critical to completing the dissertation.

iii
TABLE OF CONTENTS
Acknowledgements ii
List of Tables iv
List of Figures vi
Abstract vii
Chapter One: Introduction 1
Chapter Two: Literature Review 19
Chapter Three: Methodology 39
Chapter Four: Analysis 59
Chapter Five: Discussion and Conclusion 89
References 109
Appendix: Descriptions of Models Analyzed 125

iv
LIST OF TABLES
Table 3.1: Description of Data and Data Sources 43
Table 3.2: Summary of Regression Models 52
Table 4.1: Students‘ Background Characteristics, Categorical Variables 60
(N=9,955)
Table 4.2: Students‘ Prior Academic Achievement, Hawai‗i State 61
Assessment, 2002
Table 4.3: Students‘ Ninth Grade Educational Experiences, Categorical 62
Variables (N=9,955)
Table 4.4: Students‘ Ninth Grade Academic Performance, Scaled Variables 63
Table 4.5: School Climate Variables (N=44) 65
Table 4.6: High School Outcomes for Student Sub-Groups Based on 69
Students‘ Background Characteristics (N=9,955)
Table 4.7: Correlation Matrix of Independent Variables for Analysis 71
Table 4.8: Summary of Models‘ Significant Odds Ratios 74
Table 4.9: Summary of Variables Associated with Increased Probability 76
for Positive High School Outcomes
Table A1: Summary of Regression Models 125
Table A2: Binary Logistic Models Predicting On-Time High School 127
Completion, Models 1 and 2
Table A3: High School Outcomes and Students‘ Background 132
Characteristics, Model 3
Table A4: Multinomial Logistic Regression of Students‘ Background 136
Characteristics and Ninth Grade Academic Performance on
High School Outcomes, Model 4

v
Table A5: HLM Logistic Model Predicting On-Time High School 140
Completion, Model 5

Table A6: HLM Multinomial Logistic Model Predicting High School 142
Outcomes, Model 6

vi
LIST OF FIGURES
Figure 1.1: Educational Pipeline, 2006 8
Figure 2.1: Factors Identified in the Literature as Related to High School 30
Outcomes
Figure 3.1: Sample Selection Process and Results, Class of 2006 Cohort 41
Figure 3.2: High School Outcomes of Sample Cohort (N=9,955) 45
Figure 3.3: Statistical Analysis Plan for Modeling High School Outcomes 51
Figure 4.1: High School Outcomes for Students, by Ethnic Group 67
Figure 4.2: High School Outcomes for Students, Based on Prior Academic 68
Achievement

vii
ABSTRACT

Many studies examine the impact of students‘ characteristics and behaviors
on high school outcomes: high school completion, college enrollment or college
completion. This study uses regression analyses to explore the association of
students‘ characteristics and behaviors and students‘ positive high school outcomes:
graduating on-time, enrolling in a two-year college or enrolling in a four-year
college. The study analyzes the longitudinal student records for a Hawaii cohort of
public high school graduates.
Academic factors are powerful predictors of student outcomes. Students‘
successful academic transition to ninth grade are critical to positive high school
outcomes and are better predictors of outcomes than students‘ socioeconomic status,
ethnicity or gender. Higher grade point average and passing core subjects during
freshman year are critical on-track indicators and success factors. Conversely, poor
grades and course failures indicate that students are off-track for positive high school
outcomes and may serve as early warning indicators to identify students needing
additional support to improve their probability of graduating on-time and enrolling in
college, particularly four-year colleges. The schools that students attend have an
impact on outcomes, and students who attend schools with a more positive learning
environment are more likely to graduate high school on-time.

1
CHAPTER ONE
INTRODUCTION
Technology and globalization have changed business and workforce needs in
the United States (Friedman, 2005; National Governors Association, Council of
Chief State School Officers and Achieve, 2008; Wagner, 2008). Automation and
outsourcing have changed dramatically the employment opportunities for Americans
over the last three decades. In 1973, one-third of jobs required less than a high
school diploma, but that proportion shrunk to only nine percent of jobs by 2001;
whereas the proportion of jobs requiring a college education doubled during the
same period, increasing to 60% of the labor market in 2001 (Carnevale and
Desrochers, 2003). In 2006, approximately 80% of all U.S. jobs required at least
some postsecondary education or training with about one-third of jobs requiring a
bachelors degree or higher, and the trend is projected to continue through 2014
(Holzer and Lerman, 2009). In Hawaii, one-third of projected categories of job
openings from 2006-2016 requires postsecondary education or training, and jobs that
require postsecondary education and training pay an average salary twice as much as
those that do not require postsecondary education: $50,000 vs. $25,000 (State of
Hawaii Department of Labor and Industrial Relations Research and Statistics Office,
2008).
America’s Educational Standing
While global economic changes require higher levels skill and knowledge
represented by the credentials of the domestic workforce, American educational

2
attainment is declining relative to other countries. Among 27 Organisation for
Economic Cooperation and Development (OECD) countries and partner countries,
the U.S. ranked 21
st
in high school completion rates and slipped from 2nd to 15th in
college graduation rates between 1995 and 2008 (Organisation for Economic
Cooperation and Development, 2008). Stagnant or declining high school graduation
and college going rates hamper the U.S.‘s standing though the other countries‘ rapid
gains in educational attainment is the primary reason for America‘s loss in rank of
educational attainment among its peers internationally (National Governors
Association et al., 2008).
American students also trail their international counterparts on measures of
academic skills and knowledge. On the OECD‘s Programme for International
Student Assessment (PISA), U.S. 15-year olds ranked 25
th
in mathematics and 21
st
in
science achievement among 30 participating countries (National Governors
Association et., al. 2008). U.S. students fared better on the 2003 Trends in
International Mathematics and Science Study (TIMSS), scoring about average
among a larger group of countries; however, when compared with industrialized
nations, U.S. students‘ performance is ―below the 12-country average at both low-
and high-skill levels and low and high-levels of item difficulty‖ in mathematics and
science (Ginsburg, Cooke, Leinwand, Noell, and Pollack, 2005, pp. iv-v).
Hawaii’s Educational Standing
Historically, Hawaii‘s residents have been well educated with educational
attainment rates above the national average (U.S. Census Bureau, 2009). However,

3
analyzing current educational attainment among different age cohorts reveals that
attainment in Hawaii is declining among younger residents; fewer of Hawai‗i‘s
younger adults (ages 25 to 34) are earning college degrees than prior cohorts
(Johnsrud, 2006). Hawaii‘s educational achievement levels have also been low with
the state‘s K-12 students scoring in the bottom quartile among states on the National
Assessment of Educational Progress (National Council of Educational Statistics,
2009) and below the national average on other measures such as the College Board
SAT test (College Board, 2008).
Benefits of Educational Attainment
The private benefits of educational attainment for individuals and public
benefits for society are significant. Higher educational credentials are associated
consistently with higher wages. In 2007, median salary was $26,894 for high school
diploma recipients vs. $32,874 for those with some college experience or an
associate‘s degree vs. $46,805 for those with a four-year college degree; over a
lifetime, the difference is $1.1 million (U.S. Census Bureau, 2009a; Institute for
Higher Education, 2005). Furthermore, higher levels of education are associated
other positive benefits including individuals‘ health and well-being, civic
engagement (i.e., voting, charitable giving), and avoidance of public assistance and
unemployment (Institute for Higher Education, 2005).
Public benefits of educational attainment are also significant. A direct result
of increased earnings related to higher educational attainment is higher tax revenue
at the national and state levels. Furthermore, Hanushek and Woessman (2008) found

4
that increases in a country‘s levels of educational attainment would lead to
―substantial gains‖ in national economic productivity. At the state level, the
educational level of a region‘s population is the single most important factor in
determining a community‘s economic growth. (Cohen, 2000; Federal Reserve Bank
of Cleveland, 2005; Gottlieb and Fogarty, 2003). Business and government leaders
rate an educated workforce sixth among 21 factors that affect businesses‘ decisions
about where to locate their companies (Montague, 1987). Additionally, Hanushek
and Woessmann (2008) found that higher educational attainment leads to a more
equitable distribution of income within a country.
Concept of Educational Pipeline
In the mid-1990s, states initiated projects to improve educational attainment,
often focusing on students‘ transition from high school to college, ―a policy agenda
usually termed K-16‖ (Ewell, Jones and Kelly, 2003). The K-16 agenda referred to
the coordination of policies and transition of students between the K-12 and higher
education systems as a means to increase educational attainment. The discussion
introduced the concept of an ―educational pipeline.‖ The educational pipeline is
analogous to a business supply chain where the output of each producer or
educational system is dependent on the quality, quantity and timeliness of its input
which is also the output of the previous educational system. The educational
pipeline brings attention to of a ―series of successive transitions‖ which can be
managed and coordinated to improve outcomes (Ewell et al., 2003, p. 2). Reports
frequently identify four key transition points for students: high school graduation,

5
college enrollment, continuing enrollment higher education and receipt of a college
degree ―within 150% of time,‖ three years for a two year degree and six years for a
four year degree (Ewell et al., 2003). The pipeline concept assumes that students
progress through the sequence of transitions to earn a degree, and thus increasing
attainment of college degrees requires that high schools send more graduates onto
college and that high school graduates be better prepared to succeed.
The concept of an educational pipeline represents a shift from focusing on K-
12 and higher education as independent to inter-dependent systems. As an example,
Education Week, a national weekly education publication, has issued an annual
report about ―state level efforts to improve public education‖ since 1997 (Education
Week, 2004). In 2007, Education Week titled its annual report, ―From Cradle to the
Career, Connecting American Education from Birth through Adulthood‖ and
explained that the report represented a ―transitional document… (in the) move from
an exclusive focus on K-12 education to a broader perspective on the connections
between K-12 education and other systems with which it intersects‖ (Education
Week, January 2007, p. 1).
Rather than consider the educational production functions of K-12 or higher
education exclusively, the educational pipeline considers input at an earlier stage,
such as high school or even elementary school, and considers high school graduation
and college going rates to be throughputs toward college completion which is the
pipeline‘s output. The National Center for Higher Education Management Systems
(NCHEMS) publishes a biennial report on the ―flows‖ through the education

6
pipeline, highlighting data at ―four key transition points‖: (1) high school graduation
within four years of entering high school; (2) enrollment in college the fall semester
after receiving a high school diploma; (3) return for the second year of college; and
(4) completion of an associate‘s degree within three years or a bachelor‘s degree
within six years of enrolling in college. (NCHEMS, 2009). The report presents the
data as a morbidity or ―survival rate‖ of 100 ninth graders which makes the statistics
compelling.
Pipeline analyses of educational flows and outputs provides a broader scope
than the traditional education production function focused on a single educational
system, either K-12 or higher education. Comparing states‘ pipelines reveal
variation in college completion rates and transition points along pipeline identifying
need for different policy interventions in different states. Nationally, of 100 ninth
graders, 69 complete high school on-time, 42 enter college directly after high school,
and 20 complete within 150% of expected program completion time (NCHEMS,
2009). South Dakota, the highest ranked state in 2006, produced nearly three times
as many graduates out of this traditional education pipeline compared with the
nation, with 30 of 100 ninth graders completing college degrees within 150% of time
to degree; whereas Alaska, the lowest ranked state, only produced 7 college
graduates of 100 ninth graders. New Jersey, the state with the highest high school
graduation rate in NCHEM‘s 2009 report, ranks 17
th
in producing college graduates
with 23 of 100 ninth graders earning college degrees within ten years of starting high
school. Such pipeline data illustrate the interdependent nature of educational

7
attainment. Achieve‘s annual report, Closing the Expectations Gap, examines the
intersection between the educational systems and reports on various aspects of
alignment between secondary and postsecondary education in the 50 states (2009).
Statement of the Problem
The analysis of the ―educational pipeline‖ reveals that the status of Hawaii‘s
pipeline is poor. In the most recent report of 2006 data, Hawaii ranks 49
th
among
states and the District of Columbia in the output of college graduates per 100 ninth
graders (NCHEMS, 2009). According to the NCHEMS analysis, of 100 Hawaii‘s
ninth graders, 68 graduate high school on-time, 41 enter college directly, 24 continue
college onto their second year consecutively, and 12 complete college degrees,
working their work directly through the educational pipeline (see Figure 1.1).
Hawaii‘s educational pipeline is leaky.
At each transition in the pipeline, Hawaii‘s rates are below the nation‘s and
well below the best performing states. Increasing Hawaii‘s educational attainment
will require identifying and intervening in weak areas of the pipeline to improve
policies and practices to strengthen the pipeline and student supports to improve the
outputs.

8
Figure 1.1: Educational Pipeline, 2006

Source: NCHEMS (2009)

Purpose of the Study
This study focused on exploring the first two transition points of Hawaii‘s
educational pipeline: graduating high school and entering college. This study used
longitudinal student data to examine throughput of Hawaii‘s educational pipeline.
NCHEMS (2009) used cross-sectional data to calculate the throughput and
outputs of the educational pipeline: data in a given year about enrollment at each
point in the pipeline for its calculation. Using cross-sectional data to report data can
be problematic. For example, the NCHEMS analysis used a cross-sectional data to
calculate a high school graduation rate, which compared the proportion of graduates
and ninth graders in a given year and reported Hawaii‘s public schools‘ graduation
Of 100 9th graders

9
rate as 68% (NCHEMS, 2009). However, the State of Hawaii Department of
Education (2009) used a longitudinal method to track students who began ninth
grade in Hawaii and graduated within four years and reported a four-year completion
rate of 80% for the same period. Longitudinal data, rather than cross-sectional data,
can provide a more accurate report of the productivity of the educational system
because it can track individual participants throughout the pipeline.
To meet the Hawai‗i State Department of Education‘s goal of a 90% on-time
graduation rate (within four years), Hawai‗i‘s schools need to improve the
graduation rate, irrespective of the calculation method (State of Hawai‗i Department
of Education, 2007). Understanding the relationship between student characteristics
and high school outcomes--high school graduation and college going--is important to
improving throughputs of the pipeline, ultimately increasing the state‘s educational
attainment levels.
This study used longitudinal student data to track educational participation,
progress and outcomes of recent Hawaii public high school graduates from eighth
grade through the year after high school graduation. The study identified indicators
of whether students are ―on track‖ for high school success—as measured by high
school graduation and subsequent college enrollment—with a focus on the ninth
grade year in which students transition to high school. The study focused on ninth
grade since recent longitudinal studies have found ninth grade to be a pivotal year in
students‘ high school success or failure (Allensworth and Easton, 2005; Neild,
Stoner-Eby and Furstenburg, 2001; Silver, Saunders, and Zarate, 2008). This study

10
analyzed one cohort of Hawaii public high school graduates in the high school
graduating class of 2006.
The cohort study is the first phase of two phases that Jerald (2006)
recommended as the basis of building ―an effective and efficient early warning data
system‖ (p. 19). National educational advocacy groups, such as Alliance for
Excellent Education (Pinkus, 2008), American Youth Policy Forum (2009), National
High School Center (Heppen and Therriault, 2008) and Data Quality Campaign
(n.d.) have promoted identification of on-track indicators related to high school
graduation and promoted the development of ―early warning systems‖ to identify
students who are at risk for dropping out and not graduating high school.
This analysis identified freshman year characteristics of students who were
on-track for high school graduation and college matriculation. Such on-track
indicators for positive high school outcomes should inform the development of high
school-related policies and practices by improving the throughput of the educational
pipeline. By definition, students who were not on-track for positive high school
outcomes are ―off-track‖ and at risk of dropping out of high school. Thus, the
analysis also identified risk factors for not graduating high school.
Research Questions
The overall research question was: Which of students‘ high school
experiences in the ninth grade year indicate that they are on-track for positive high
school outcomes? The following sub-questions were examined:

11
1. Which of students‘ background characteristics are risk or success factors?
2. Do indicators of risk or success vary by type of high school outcome
(graduating on-time and college enrollment)?
3. What is the relationship of students‘ academic experiences in the high
school freshman year to their high school outcomes? And to what extent
do risk or success factors vary by students‘ ninth grade academic
performance?
4. To what extent does the school attended relate to outcomes?
Significance of the Study
This study identified indicators of whether students are on-track for
graduating high school on-time and enrolling in college, using currently available
administrative data about students. These indicators identify key opportunities to
intervene when students become off-track. The interventions are opportunities to
leverage change in students‘ trajectory for high school success, improving Hawaii
students‘ high school outcomes. The study has implications for schools and
institutions, state policymakers and researchers.
Schools and institutions. The study may inform school practice by
identifying opportunities for interventions. High schools could identify students who
exhibit characteristics or behaviors identified as off-track as candidates for
intervention. Schools could develop intervention plans for individual students or
groups of students who are off-track and allocate resources to interventions most

12
likely to have a positive return on investment. School-level indicators identify areas
for schoolwide investigation and improvement.
Furthermore, factors associated with college going can inform the K-12 and
higher education institutions. Since this analysis considered ethnicities, it can inform
efforts to improve achievement for underrepresented groups. The University of
Hawai‗i has a strategic goal of increasing college enrollment and completion among
ethnic groups underrepresented at the University of Hawai‗i (University of Hawai‗i
Office of Vice President of Academic Planning and Policy, 2009), and the State of
Hawai‗i Department of Education (DOE) has a goal of closing gaps in student
achievement and college enrollment particularly for Native Hawaiians and low-
income students (State of Hawai‗i DOE, 2010). Findings about factors associated
with college going can inform higher education system and institutions‘ enrollment
management plans. This analysis can inform the UH and DOE strategies to increase
college access for the underrepresented groups and focus institutional investment in
interventions that increase the likelihood of high school students‘ positive high
school outcomes.
State policymakers. The study could inform state-level policies related to
Hawaii‘s public high schools and higher education system. The identified risk
factors may serve as the basis for an early warning system, a data tool to track
students‘ participation, progress and performance in schools. The early warning
system could produce reports on whether students are on-track for positive high
school outcomes and flag students who exhibit risk factors. Additionally, the

13
analyses may identify schools with more successful student outcomes.
Understanding school effects can inform school improvement interventions for high
schools, including support for school reform and design of high school
accountability measures. college enrollment or target schools that need more
assistance.
Researchers. Two aspects of this study present an opportunity to inform the
research field. First, analysis of Hawaii‘s unique ethnic composition of its student
population presents an opportunity to inform research. Hawaii‘s student population
is diverse with 65% identified as Asian or Pacific Islander (State of Hawaii
Department of Education, 2009a). Nationally, Asian and Pacific Islanders are often
represented as a single group in statistics and analyses (e.g., National Council for
Education Statistics, 2009a). In Hawaii, DOE‘s student records include one self-
identified ethnic group, and there are seven categories of Asian and Pacific Island
ethnicities recorded by the DOE; for the cohort studied, Hawaiians were the largest
Asian and Pacific Islander group representing 35% of Asian and Pacific Islanders.
Due to the data available and variation among Asian and Pacific Islander
populations, analyses can disaggregate among Asian and Pacific Island ethnic groups
and may reveal different on-track indicators or risk factors among ethnic groups.
Second, most research on high school outcomes focus on a single outcome,
such as high school dropout (e.g., Gleason and Dynarski, 2002; Roderick and
Camburn, 1999), high school graduation (e.g., Allensworth and Easton, 2007),
college enrollment (e.g., Roderick, Nagaoka, Coca, Moeller with Roddie, Gilliam

14
and Patton, 2008) or college completion (e.g., Adelman, 2006; Allensworth, 2006).
However, this analysis focused on factors related to four categories of outcomes: not
graduated on-time, graduated high school on-time, enrolled in a two-year college and
enrolled in a four-year college. This research may inform the field by expanding
scope of early warning systems from high school graduation to include college
enrollment.
Limitations, Delimitations, Assumptions
This study is limited in geography, time and scope. First, the study focused
on Hawaii public high school students expected to graduate in 2006 who were also
enrolled in the public education system as eighth graders. The generalizability is
limited by factors that affect students‘ educational experiences, such as rigor of the
graduation requirements applicable to the cohorts studied. Also, generalizability to
other localities is limited since Hawaii public school students‘ records were
analyzed, and Hawaii has a higher proportion of students attending private schools
than nationally: 17% versus 10% (Hoffman and Sable, 2006; Broughman and
Swaim, 2006).
Second, the analysis focused on students‘ educational experiences during the
ninth grade year. However, research has linked a wide range of students‘
educational experiences to high school outcomes. For example, researchers found
that participation in Head Start early childhood programs was associated with a
higher likelihood of graduating high school among African American males and with
higher likelihood of graduating high school and enrolling in college among Whites

15
(Garces, Thomas, and Currie, 2002). Other research has linked college enrollment to
participation in college preparation programs (Gandara and Bial, 2001; Domina,
2009). Additionally, the scope of the study was limited to data available in DOE data
sets. Since the purpose of an early warning system is to identify students based on
data available currently from administrative data sets, only data available in
administrative data sets were considered, excluding potentially relevant but
unavailable data such as student attendance. Furthermore, this analysis was also
constrained by the quality of data available from the DOE.
Finally, the analysis identified relationships between student characteristics
and high school outcomes. Early warning systems are developed from indicators
drawn from analyses of location-specific data with statistical methods being used to
account for differences in students‘ background; in such analyses, results describe
associations but not causal relationships between student characteristics and
outcomes (McEwan and McEwan, 2003). Thus, the early warning indicators are not
definitive. Also the findings from other sites may not generalize to Hawai‗i. This
analysis used historical DOE administrative data to identify patterns of relationships
of Hawai‗i‘s students and schools with high school outcomes.
Definitions
For this study, the following is a definition of terms:
Academic press: ―Extent to which school members… place a strong
emphasis on academic success and conformity to specific standards of achievement‖
(Lee, Smith, Perry and Smylie, 1999, p. 2).

16
Binary logistic regression: Statistical method that predicts the probability of
an event occurring on the basis of continuous and/or categorical independent
variables when there is a dichotomous dependent variable (Hair, Anderson, Tatham
and Black, 1998).
Cohort analysis: ―Study in which… any group of individuals who are linked
in some way… (which) measures some characteristic… at two or more points in
time‖ (www.socialresearch methods.net, 2009).
College going rate: Proportion of high school graduate who enrolled in
college as first-time freshmen (NCHEMS, 2009).
Core subjects: English, mathematics, social studies/history, and science.
Early warning system: System that uses ―indicators based on readily
accessible data can predict whether students are on the right path toward eventual
graduation‖ (Heppen and Therriault, 2008, p.1).
Educational pipeline: Continuous progression of students from pre-K, to
elementary school, to secondary, to college, and into the workforce (Ewell et al.,
2003).
Educational attainment: ―Highest diploma or degree, or level of work
towards a diploma or degree an individual has completed (U.S. Bureau of Labor
Statistics, 2009).
Free/Reduced Price Lunch: Subsidy offered to eligible low-income students
by the U.S. Department of Agriculture‘s National School Lunch Program. Based on
family income and family size, students may qualify for either free or a reduced price

17
meals. Researchers often use students‘ eligibility for Free/Reduced Price Lunch as a
proxy for being low-income.
Grade Point Average (GPA): Mean grade where letter grades (A, B, C, D or
F) are assigned numbers (4, 3, 2 1 and 0, respectively). The grade values are
weighted based on the number of credits assigned to a course. Often used as an
indicator of students‘ overall academic progress.
Graduation rate: Proportion of students who complete high school in a
given period.
Hawaii State Assessment: Annual criterion-referenced test given to Hawaii
public school students in grades 3-8 and 10 in reading, mathematics, and science and
used for federal accountability under the No Child Left Behind law.
Hierarchal Linear Modeling: Statistical technique used to analyze data and
uses HLM software developed by Raudenbush, Bryk and Congdon (2004). HLM is
appropriate for analyzing situations in which ―individual subjects of the study may
be classified or arranged in groups which themselves have qualities that influence the
study‖ (Scientific Software International, 2010).
Longitudinal data: Data in which the same units are observed over multiple
time periods over many years. (U.S. Bureau of Labor Statistics, 2009)
Multinomial logistic regression: Statistical technique used to model the odds
of discrete choices as a ―function of the covariates and to express the results in terms
of odds ratios for choice of different plans‖ (Hosmer and Lemeshow, 2000 p. 260).

18
Odds ratio: ―Measure of association… (which) approximates how much
more likely or unlikely it is for the outcome to be present… (and) approximates a
quantity called the relative risk‖ (Hosmer and Lemeshow, 2000, p. 50).
Overage: Students beyond expected age for their grade level.
Retention: When students fail to promote to the next grade level with their
peers and subsequently repeat a grade level.
Student engagement: When "students make a psychological investment in
learning‖ (Newmann, 1992, p. 2).

19
CHAPTER TWO
LITERATURE REVIEW
This study examined the relationships between Hawaii students‘ high school
experiences, particularly related to the ninth grade year, and high school outcomes:
graduating high school on-time and/or enrolling in postsecondary education
immediately after high school. The study‘s purpose was to identify indicators related
to high school graduation and college going to inform development of an early
warning system which identifies students who are off-track for interventions
designed to increase their likelihood of positive high school outcomes.
The literature review consists of the following sections:
1. Review of high school dropout studies
2. Review of college access studies
3. Current early warning systems
The first two sections represent the two areas of study related to high school
outcomes. First, high school dropout studies identify factors associated with students
leaving high school or failing to complete high school within four years. Second,
college access studies identify factors associated with students attending
postsecondary education. The third section reviews current early warning systems
being implemented. The review concludes by identifying gaps in the literature
which informed this study‘s design.

20
High School Dropout Studies
Research on high school outcomes began in the 1980s, following A Nation at
Risk report (National Commission on Excellence in Education, 1983). In response to
the report‘s identification of ―a rising tide of mediocrity,‖ states increased their high
school graduation requirements prompting concerns about higher requirements
increasing the number of students dropping out of high school (National Commission
on Excellence in Education, 1983). Subsequent research, until ten years ago,
focused primarily on comparing demographic and family background characteristics
of high school dropouts and graduates and the consequences of dropping out
(McNeal, 1997). On the basis of ―risk factors‖ associated with dropping out--
typically being a minority ethnic group or low-income or having prior poor academic
performance--students were identified as at-risk for dropping out and provided
intensive services. However, a few key studies pointed out the importance of
identifying risk factors rather than relying simply comparing descriptive statistics of
dropouts and high school graduates.
Gleason and Dynarski (1998) evaluated a federally funded dropout program,
the U.S. Department of Education‘s School Dropout Demonstration Assistance
Program, and found that dropout programs often failed to target the right students.
Of 2,615 high school students identified as being at risk for dropping out, only 15%
actually dropped out. The authors found single risk factors based on descriptive
statistics to be inefficient in identifying dropouts. Too often, dropout programs
failed to identify and serve large number of students who dropped out and served

21
many students who were unlikely to drop out. Gleason and Dynarski called for
predictive validity of identified risk factors and recommended using longitudinal
data to identify risk factors since longitudinal data allows for tracking the outcomes
of students who exhibit different risk factors. They also speculated that dropping out
of high school was the culmination of a student‘s school experiences over time and
would best be predicted on basis of multiple risk factors over time.
Roderick (1993) studied Fall River, Massachusetts, following a cohort of
students from fourth grade to high school graduation. She found two distinct groups
of dropouts: those who dropped out by ninth grade and those who dropped out
between tenth and twelfth grade. She found that the groups had different
characteristics. Roderick‘s study highlighted the importance of examining student
patterns of academic performance and engagement longitudinally rather than simply
comparing groups of dropouts and graduates.
Other studies confirm the importance of longitudinal data in understanding
the dropout phenomenon. Catterall (1998) used National Education Longitudinal
Study of 1988 data to examine students‘ progress from eighth grade to high school
graduation. Catterall critiqued many studies on student success as focused on group-
level probabilities of failure rather than actual school performance or behavior over
time. In a study of Baltimore children which followed students from first grade
through high school graduation, Alexander, Entwisle and Kabbani (2001) found that
most students dropped out occur in high school years, due in part to compulsory
school age requirements, but that dropout was rarely an event but a process of

22
disengagement beginning as early as the first grade. The Baltimore study reinforced
the importance of tracking students over time to identify risk factors associated with
high school dropouts.
A recent literature review of high school dropout studies found significant
variation in the study designs (Hammond, Linton, Smink, and Drew, 2007). The
review identified 25 significant individual and family-related factors that at least two
studies identified as significant and found four factors that affected dropout behavior
at all level of schooling: low socioeconomic status, ―overage‖ for grade, low
achievement, and poor attendance. These categories of factors represent students‘
background characteristics, students‘ high school academic experiences, students‘
and engagement in high school. Furthermore, several studies in the review identified
the influence of the high school students‘ attendance on dropping out.
Students’ background characteristics. Background characteristics refers to
demographic descriptors of students such as ethnicity, gender, age and
socioeconomic status. High school graduation rates vary significantly by such
demographics. For example, in the U.S. in 2005-06, 90% of Asian Americans
graduated high school ―on-time (in four years)‖ versus 59% of Blacks (Stillwell and
Hoffman, 2008). Studies in this review considered the association of multiple
background characteristics and high school dropout or completion.
Certain student demographic characteristics, such being male, older, from
ethnic minority groups or families with lower socioeconomic status or with single
parent or stepparents are associated with being more likely to drop out (Bryk and

23
Thum, 1989; Rumberger, 1995). However, the consideration of multiple background
characteristics affects the outcomes since the interaction of different background
characteristics are related to dropping out. Catterall (1998) found that low
socioeconomic status affected high school completion negatively, except among
Hispanic and African American cohorts when controlling for ethnicity.
Studies have found that students‘ age relative to their peers affects high
school outcomes. Being overage upon high school entry increased odds of students
dropping out (Roderick, 1994). In studying high school dropouts in Fall River,
Massachusetts, Roderick found that being overage for among peers was a better
predictor of high school dropout more than being retained in grade prior to high
school. The reasons for students being overage include being retained in elementary
school or being placed at lower grade than their age for entering school as an English
Language Learner. Thus, Roderick found the association between age and high
school graduation to be more important than considering retention in prior grades.
Another factor considered in dropout studies is students‘ mobility. The
relationship of students‘ mobility to dropping out or graduating high school is
complex. Changing high schools is often associated with dropping out (Rumberger
and Larson, 1998 in Lee and Burkam, 2003; Schneider, Swanson and Riegle-Crumb,
1998). However, for some students, changing high schools is beneficial. Since high
schools tend to ―track‖ students into a particular curricular sequence of courses,
moving schools can change students‘ opportunities to take higher level courses
leading to positive school outcomes since the transfer requires the student to be

24
assigned to another school‘s explicit or de facto curricular pathway (Schneider et al.,
1998). Thus, considering mobility is important but the direction of the association is
not clearly identified in the literature.
Students’ high school academic experiences. Many studies document the
relationship between students‘ academic experiences and high school outcomes.
Considering high school academic experiences in statistical models of high school
graduation or dropout often increases the explanatory power of the model
significantly. A recent analysis of Los Angeles student outcomes found that
considering high school academic experiences in addition to students‘ background
characteristics, increased explanatory power of the model from 17 to 29% (Silver et
al., 2008). Chicago students‘ background characteristics (e.g., demographics, test
scores) explained only 12% of the variance in graduation rates while freshman year
grades explained 39% of the variance in graduation rates (Allensworth and Easton,
2007). In Chicago, ninth grade course failures and grade point averages predict 80%
of students who graduate (Allensworth and Easton, 2007).
Studies of Chicago high school students found that high school grade point
average (GPA) was related to positive outcomes; Chicago freshman with a B or
better average had a 95% chance of graduating high school but those with less than a
C average were more likely to drop out than graduate (Allensworth and Easton,
2007). Similarly, Neild, Stoner-Eby, and Furstenberg (2008) found that Philadelphia
students with larger percentages of eighth grade Ds and Fs were more likely to drop

25
out of high school and that for every percent increase in ninth grade courses failed,
the odds of dropping out of high school increased by 2.4%.
Studies have also found that grades in specific courses to be strong signals of
whether a student will graduate high school or dropout. In studying Chicago high
school students, Allensworth (2006) found that a failing grade in any core subject –
English, mathematics, social studies or science – in any semester of the freshman
year was highly predictive of dropping out. In contrast, passing Algebra I by the
ninth grade in Los Angeles was associated with a 75% increase in likelihood of
graduating on-time (Silver et al., 2008). Success in ninth grade coursework is
critical because many students fail freshman year courses. The Los Angeles study
found that 63% of the study cohort received a D or F in Algebra I (Silver et al.,
2008), and a Chicago study found that 53% of freshmen fail at least one major
subject in the first semester (Allensworth and Easton, 2007).
Being retained in grade is usually the result of failing courses and falling
short of credits needed to promote to the next grade. In Philadelphia, most students
dropped out as ninth graders but only five percent of students dropped out during
their first year of high school; instead most students dropout while repeating ninth
grade for a second or third year (Neild et al., 2001). The Los Angeles study found
that students repeating ninth grade were four times less likely to graduate than their
peers who promoted on-time after their first year in high school (Silver et al., 2008).
Student engagement in high school. Fredricks, Blumenfeld and Paris (2004)
reviewed literature about school engagement. They identified three types of

26
engagement: behavioral, emotional and cognitive engagement and recommended
studying engagement as a multi-faceted construct. Behavioral engagement refers to
involvement in academic, social or extracurricular activities. Emotional engagement
refers to interactions with teachers, classmates and schools and students‘ association
with the institution and willingness to do the work. Cognitive engagement refers to
mental effort students exert related to learning. Behavioral engagement is most often
measured and reported because it is included school administrative data: attendance
and behavioral problems (or absence thereof). Thus, student behaviors are often
considered in studies of high school outcomes.
Attendance is an important predictor of high school completion, serving as an
indicator of students‘ disengagement from school and a precursor to course failure
(Bryk and Thum, 1989). In a national study, students with poor attendance – who
cut class at least once a week or were tardy ten times or more a month—were six
times more likely to dropout than their peers with better attendance (Pinkus, 2009).
Research on Philadelphia students found that ninth graders who did not enter high
school with risk factors but attended less than 70% of time had a probability of
dropping out of at least 75% (Neild and Balfanz, 2006).
However, high school attendance can be difficult to measure. High schools
vary in their definition of attendance, such as whether student attendance is recorded
for each class period for once for the day at a designated marking period. Thus,
absenteeism is necessary to consider but often insufficient in explaining outcomes.

27
Students with behavioral problems often struggle to complete high school. A
survey of African American dropouts found that 80% had been suspended (Barnes,
1992 in Golschmidt and Wang, 1999). Neild, Balfanz and Herzog (2007) found that
a final unsatisfactory behavior mark received as a sixth grader to be as powerful a
predictor of high school dropout as receiving a failing grade in mathematics or
English.
School climate. In addition to student characteristics and experiences, studies
have found the schools, as organizations, influence outcomes. Typically, studies use
multi-level analyses of student data consider school and student level effects on high
school outcomes. For example, Roderick and Camburn (1999) found that Chicago
students‘ ability to ―recover‖ from ninth grade course failure varies by school even
when individual student risk factors are controlled. Chicago students‘ background
characteristics explained only about one-third of across-school variance in course
failures and school effects mattered. Allensworth and Easton (2007) also found
significant school effects on student outcomes and attributed 4.4 days per semester of
absences among Chicago freshmen to school differences.
Measuring school climate is difficult, but research, particularly studies from
the Chicago Consortium for School Research identify school climate regarding
―academic press‖ to be associated with students‘ high school graduation or dropout.
Academic press is the extent to which the school culture is focused on academic
success. Lee et al. (1999) defined academic press as teachers‘ expectations for
student learning, when teachers press students to work hard and achieve. Roderick

28
and Engel (2001) studied low achieving sixth and eighth graders in Chicago and
found that when teachers created an environment of social and academic support,
student motivation increased and resulted higher than average test score gains and
promotion rates. Allensworth and Easton (2007) defined ―schoolwide academic
press for the future‖ as ―students‘ views of the academic norms of academic
expectations at their schools,‖ as measured by annual surveys. At Chicago schools
with strong academic press, freshmen exhibited fewer risk factors related to absences
and grades.
Academic press is also reflected in schools‘ curricular offerings for their
students. In schools that offer mainly academic courses and few nonacademic
courses—which Lee and Burkam (2003) described as having ―constrained academic
curriculum‖--student are less likely to drop out. Students in schools that offer
calculus exhibited 56% lower odds of dropping out between 10
th
and 12
th
grades.
Offering fewer courses below Algebra I reduced odds of dropping out by 28% in an
analysis of High School Effectiveness Supplement of NELS:88 using hierarchal
linear modeling (Lee and Burkam, 2003). Students in schools that offered Advanced
Placement, college level, courses had a 41% greater odds of graduating (Heck and
Mahoe, 2006).
Research also finds that school climate is also a function of the school‘s
organization. A key factor is school size. Rumberger and Thomas (2000) found
higher dropout rates in schools with larger enrollment. However, Lee and others
posit that school size is an important influence on other organizational aspects of

29
schools and may not have a direct relationship with student achievement (Lee, 1999;
Lee and Loeb, 2000). Heck and Mahoe (2006) found that students in schools
working aggressively to improve student learning by implementing new practices or
policies were more likely to graduate; a one standard deviation (1 SD) increase in
school improvement activity increased students‘ likelihood of receiving a diploma by
42%.
College Access Studies
College access studies address a variety of factors related to students‘
enrollment in postsecondary education. Unlike high school dropout studies focused
on negative risk factors, college access studies typically focus on positive factors
associated with enrolling in college after high school graduation. College enrollment
is usually studied separately from high school completion though many factors are
associated with both outcomes (see Figure 2.1). Since the focus of this study is on
using longitudinal data to identify factors associated with high school outcomes, the
literature reviewed for this study is limited to studies analyzing longitudinal data
which included students‘ school records from at least ninth grade and used
multivariate analyses.
Students’ background characteristics. Student demographic characteristics
are highly correlated with college going rates. Studies find that low-income students
are less likely to enroll in postsecondary education even when statistical analyses
control for factors in addition to background characteristics (Hanson, 1994).

30
Figure 2.1: Factors Identified in the Literature as Related to High School Outcomes

High School Dropout College Access

Risk Factors

Success/Enabling Factors

Students‘
background
characteristics
Students‘ high
school experiences
Students‘ high
school engagement
High school climate
Gender
Ethnicity
Age
Socioeconomic
status of student‘s
family
Prior academic
achievement
9
th
grade course
failures
9
th
grade GPA
9
th
grade retention
Rigorous math
courses taken
Combination of
rigorous courses
Mobility
Behavioral
offenses
Attendance
Academic press
School size
(enrollment)
Socioeconomic
status of school
community

However, college going patterns are complex. Schneider et al. (1998) found
differences in the interaction among student demographic characteristics and types of
colleges enrolled. Females and Asians were more likely to attend postsecondary
education overall but Black students were more likely to not attend college or attend
a two year college than to attend a four year college.
Students’ high school academic experiences. The literature on college going
is rife with studies on the relationship between high school courses that students have
taken and the likelihood of attending college. Many studies identify academic
preparation as the most effective means to increase the likelihood that students will
graduate high school ready for college and complete college (e.g., Adelman, 2006;
Bedworth, Colby and Doctor, 2006).

31
Some studies identify specific high school courses as being associated with
college enrollment and completion. A frequent finding is that taking higher level
mathematics courses increases the likelihood of college enrollment (ACT, 2006;
Friedkin and Thomas, 1997; Horn and Kojaku, 2001; Pallas and Anderson, 1983;
Rose and Betts, 2001). Adelman (2006) reported that completing high school
mathematics beyond Algebra 2 doubled students‘ odds of completing a bachelor‘s
degree (p. 31). He also found that earning six or more college credits while in high
school (equivalent of taking two college courses) increased students‘ likelihood of
completing a bachelor‘s degree. Also, several studies identified laboratory science
courses, especially physics, as contributing to improved outcomes (ACT, 2006;
Adelman, 2006; Horn and Kojaku, 2001; Meyer, 1999; Rose and Betts, 2001).
Taking a college preparatory curriculum—not just a specific course—is also
associated with college enrollment. Attewell and Domina (2008) estimated that a
one-quintile increase in the rigor of high school courses, as categorized by the
National Education Longitudinal Study of 1988, is associated with about a ten
percent increase in the likelihood of attending a four-year college. Roderick et al.
(2006) found that Chicago students taking more advanced coursework, such as
honors and Advanced Placement, were more likely to enroll in more selective
colleges.
Additionally, Chicago researchers found a strong association between high
school grades and students‘ postsecondary success. They identified high school
grade point average ―perhaps the most important determinant of students‘ access to

32
college and likelihood of college graduation‖ (Roderick, Nagaoka and Allensworth
with Coca, Correa and Stoker, 2006, p. 87).
School climate. Several studies identify that schools which emphasize
college-going have positive impact on students. The Consortium on Chicago School
Research‘s Postsecondary Transition Project studies found school climate to affect
students‘ high school outcomes significantly (Allensworth, 2006; Allensworth and
Easton, 2007; Roderick et al, 2008). Consortium researchers measured school
climate using student and teachers‘ survey responses to annual surveys, and their
analyses found measures of academic press and student-teacher relationships to be
reliable in distinguishing between school and individual responses (Consortium on
Chicago School Research, 2009).
Easton, Ponisciak, and Luppescu (2008) defined college-going culture as
teachers having high expectations for students‘ postsecondary education. They
found that students in schools with a strong college-going culture were 12 percentage
points more likely to plan to attend a four-year college and 17 percentage points
more likely attend a college that matched their qualifications than equally qualified
students in schools with weak college going cultures. Building college-going culture
increased the likelihood of students graduating high school prepared for college and
completing college degrees (Schramm and Sagawa, 2008).
College access studies often focus on factors that mediate college
matriculation. Longitudinal studies of Chicago Public Schools‘ graduates have
found mismatches between students and their college choices, even among

33
academically strong students (Roderick, Nagaoka, Coca and Moeller, 2009).
Students‘ college enrollment and their choices of postsecondary institutions are
mediated by a number of factors including students‘ high school academic
experiences, students‘ background characteristics, and high school attended
(Roderick et al., 2008) as well as their social capital as described by parent-student
discussion about college preparation and guidance received at high school (Plank and
Jordan, 2001).
Additionally, a large literature documents many other factors affecting
college enrollment which were not included in longitudinal studies reviewed. For
example, students and families overestimate the college costs; researchers associate
this misinformation with reducing college enrollment, especially for low-income
students, even among academically qualified students (Horn, Chen and Chapman,
2003; Long, 2004). Due to lack of social networks with knowledge about college
admission or poor high school counseling, students are not always well informed
about college admission processes (Orfield and Paul, 1993). Also, faculty and
administrators of K-12 and postsecondary institutions rarely coordinate their policies
so even high school graduates may not be prepared to meet academic requirements
for college admission or placement into college-level courses (Kirst and Bracco,
2004).
Current Early Warning Systems
As educational advocacy groups have promoted early warning systems, states
and districts have begun to identify indicators, develop data tools to identify students

34
as they fall off-track, and implement interventions to get students back on-track for
high school graduation. However research basis for most systems‘ indicators is not
as well documented as Chicago and Philadelphia. Also, early warning systems are at
different stages of development with some sites, such as Portland and Chicago
currently implementing a system, and other sites such as Philadelphia, having
identified its indicators and proposed a system but not yet implemented a data tool or
interventions.
Chicago. The Consortium on Chicago School Research is conducting an
extensive research project, ―Chicago Postsecondary Transition Project,‖ dissecting
the pathway to high school achievement, graduation, college enrollment and college
completion. The project has produced a series of reports titled, From High School to
the Future. As part of the project, the Consortium identified indicators of whether
high school freshmen are on-track for graduation and factors that predict high school
graduation. Allensworth and Easton (2007) identified freshman factors associated
with graduation as grade point average of B or better, passing all classes and fewer
than one week of absences in a semester. In Fall 2008, Chicago Public Schools
launched an early warning tool which provide schools with a ―watchlist‖ of students
who are falling off-track and with intervention strategies (2008).
Philadelphia. Based on research by Neild and colleagues, Philadelphia
identified risk factors associated with students dropping out. In 2007, Neild and
Herzog reported early warning indicators at sixth, eighth and ninth grades. Sixth
grade indicators are attending school less than 80% of the time, receiving a poor final

35
behavior mark, failing sixth grade mathematics and failing sixth grade English.
Eighth grade indicators are failing eighth grade math, failing eighth grade English
and attendance of less than 80%. Ninth grade indicators are earning fewer than two
credits, not being promoted to tenth grade on-time and attending school less than
70% of the time. The Philadelphia School District‘s strategic plan for 2014 includes
development of an early warning indicator system to plan interventions for schools,
student groups and individual students (School District of Philadelphia, 2009).
Indiana. The Indiana State Department created a composite index of age,
retention patterns, middle school test scores, mobility and attendance. The model
only considered data collected at the statewide level so it did not include transcript
information (e.g., course taken or grades) or demographics such as ethnicity or
socioeconomic status. Based on the index, students are identified as at high, medium
or low risk for dropping out, serving as the basis of an early warning system (Kline,
personal communication, March 23, 2009).
North Carolina. North Carolina uses three indicators to signal an early
warning for students‘ risk for high school dropout: having failed one or more grades
before ninth grade, not completing Algebra I in the eighth grade, and being from a
low-wealth family (Coble, personal communication, March 23, 2009).
Portland, Oregon. Analysis Portland Public Schools‘ data for the Class of
2004 revealed that only 54% received a high school diploma (Stid, O‘Neill and
Colby, 2009). The analysis found virtually all dropouts experienced at least one of
five risk factors. Portland administrators selected two freshman year indicators to

36
identify students likely to dropout and to target for intervention: failing grades in
two core courses and excessive absences (more than 20 days). Prior to 2007-08
school year, Portland school administrators identified incoming ninth graders who
had exhibited a risk factor as an eighth grader then implemented interventions to ease
students‘ transition to high school. During that year, Portland saw a modest decline
in the freshman year indicators of students likely to dropout: 57 to 50% of students
failing more than two courses and 40 to 35% of students with more than 20 absences.
Louisiana. Louisiana State Department of Education developed a Louisiana
Dropout Early Warning Systems (DEWS) is based on historical data which identified
four early warning indicators: student attendance, student course achievement,
student behavior, student age (Pinkus, 2008). The system allows participating
schools and districts to receive information identifying about at-risk students twice a
month. Schools are expected to use to identify ninth grade students needing
interventions.
Gaps in Literature
There are two major gaps in the literature. First, the literature is bifurcated
between studies of high school dropouts and college access. Second, the literature
lacks studies of Hawai‗i and student populations relevant to Hawai‗i.
Diversity of high school outcomes. Studies typically focus on a single high
school outcome: high school dropouts compared with high school graduates
(Allensworth and Easton, 2007; Gleason and Dynarski, 1998; Roderick and
Camburn, 1999) or four year college enrollment (Adelman, 2006; Roderick et al.,

37
2006). Studies do not address the sequence of graduating high school on-time and
going onto a two-year or four-year college. In fact, studies reviewed did not
compare factors associated with students‘ selection of two-year or four-year colleges.
The national discussion, stimulated in part by the American Recovery and
Reinvestment Act (2009), is moving from a focus on high school completion to
college and career readiness and from college access to college completion.
Therefore, understanding the range of high school outcomes is increasing in
importance. It is necessary to bridge the gap between research about high school
dropouts and college access to identify factors that will enable high school and
college persistence and completion. This study will address both high school
completion and college enrollment.
Hawaii’s student population. Patterns of students‘ dropout or success vary
across subgroups, regions and locations (Jerald, 2006), and research identifies these
differences by examining interactions between student characteristics and location
(Hammond, et al., 2007).
Only one study reviewed analyzed longitudinal data about Hawaii high
schools. Heck, Price and Thomas (2004) studied a ninth grade cohort from one
Hawaii high school to identify patterns of students‘ coursetaking. Within one high
school they found seven different patterns of coursetaking among student groups.
The study revealed defacto tracking of students into different curriculua and found
the group membership to be associated with different ethnic groups among Hawaii‘s
diverse student population. However, while the study considered students‘ academic

38
record through senior year and post-high school plans, it excluded dropouts from the
sample and did not include high school graduation or college enrollment as variables
for analysis.
This exploratory study analyzed Hawaii data which determine whether there
are location-specific associations between student and school characteristics and high
school outcomes. Also, the literature does not address the specific ethnic groups of
Hawai‗i‘s diverse population. In Hawai‗i, the Asian and Pacific Islander population
makes up 65% of K-12 student enrollment, and Hawaii which identifies ethnic
groups using more specific categories than used by the federal government in the
Census or required for No Child Left Behind reporting (State of Hawaii DOE,
2009a). High school outcomes for Asian/Pacific Islander ethnic groups varies, and
this study allowed for disaggregation of the Asian and Pacific Islander groups to
explore differences in the relationships between students‘ characteristics and high
school outcomes among very diverse groups usually categorized homogenously.

39
CHAPTER THREE
METHODOLOGY
This study examined the relationships between student characteristics and
positive high school outcomes for Hawaii public school students: graduating high
school on-time and matriculating in postsecondary education immediately after high
school. The relationships identified indicators of whether students were on-track for
high school graduation and for entering college. The indicators may serve as the
basis for an early warning system to identify students who are falling off-track in
order to target interventions to support positive high school outcomes. This study
aimed to identify risk and success factors for high school outcomes, as advised by
Jerald (2006) to ―uncover patterns in how those risk factors play out for students
over time and assessing how to measure the impact of schools‖ (p. 15). This study
focused on factors related temporally to students‘ transition year into high school.
This chapter describes the methodology used in this quantitative study by identifying
the research questions, and describing the sample, data sources and data analysis.
Research Question
This study addressed the question: Which of Hawai‗i students‘ high school
experiences in the ninth grade year indicate whether they are on-track for positive
high school outcomes? The following sub-questions were examined:
1. Which of students‘ background characteristics are risk or success factors?
2. Do indicators of risk or success vary by type of high school outcome
(graduating on-time and college enrollment)?

40
3. What is the relationship of students‘ freshman year academic experiences
to their high school outcomes? And to what extent do risk or success
factors vary by students‘ ninth grade academic performance?
4. To what extent does the school attended relate to outcomes?
The research questions are grounded in the literature about factors affecting students‘
high school completion as well as college enrollment. Indicators in existing state
and district-developed early warning systems and available data also informed the
research design.
Research Design
This study was a secondary analysis of administrative data collected by the
State of Hawaii DOE. The data are longitudinal and track individual students over
time. The study focused on identifying patterns and pathways of student progress,
success and failure, improving upon earlier attempts to simply correlate risk factors
with high school outcomes (Gleason and Dynarski, 2002). This exploratory study
used quantitative analyses of longitudinal data to understand the relationship
between student characteristics and their high school outcomes.
Sample and Population
This study analyzed a cohort of students in the DOE. The cohort consisted of
11,915 students who were eighth graders in June 2002 and also attended ninth grade
the following year in a DOE school. The cohort participants who completed high
school on-time (within four years) graduated in Spring 2006. The sample consisted
of the entire cohort of students with the exception of 1,132 students who transferred

41
legitimately out of DOE schools (e.g., private school, out of state or country, home
school) and were excluded. The sample also excluded students who transferred into
DOE schools in grades 9-12. Finally, the study also excluded 828 students who
transcript records were incomplete; many of the ineligible students attended small
schools which did not participate in the electronic student information system
(including most charter schools) and one comprehensive high school. The final
sample consisted of 9,955 students (see Figure 3.1).

Figure 3.1: Sample Selection Process and Results, Class of 2006 Cohort

Did not enroll in college
n=3,721 (31%)

Graduated HS on-time
n=8,230 (69%)

Enrolled in 2-yr college
n=2,730 (23%)

Enrolled in 4-yr college
n=1,779 (15%)

Graduated late
n=115 (1%)

Cohort (8th --> 9th grade)
N=11,915

Did not graduate on-time
n=1,725 (14%)

Still enrolled in 2007-08
n=9 (0%)

Dropped out
n=1,601 (13%)

Transferred (legitimate
exit) n=986 (8%)

Exited study cohort
n=1,960 (16%)
Transcript ineligible
n=828 (7%)

Missing post-high data
n=146 (1%)
Note: Proportion of cohort in parentheses

42
Data Collection
This study examined the relationship of students‘ high school outcomes with
their background characteristics, educational experiences and school attended.
Student data for this study were drawn from various DOE administrative data sets
(see Table 3.1). Data about school climate were obtained from the Department of
Education‘s School Quality Survey (SQS) which is administered biennially. Data
about students‘ postsecondary matriculation came from datasets purchased annually
from the National Student Clearinghouse.
The DOE provided relevant administrative data, and the researcher developed
a dataset for analysis by linking administrative data through the unique student
identifier (using pseudo-identification number provided by the DOE) and school
organizational identification code. The researcher examined the dataset to check
completeness and consistency of the data. The researcher also reviewed minimum
and maximum values for each data field ensure that the values were reasonable.
Also, the researcher also checked the dataset for missing data.

43
Table 3.1: Description of Data and Data Sources

Data source Variables Years of data
Student enrollment
records
Enrollment/withdrawal status
Demographics (ethnicity, socioeconomic
status, gender, birthdate)
Grade in school
School attended
2001-02 through
2007-08
Student test scores
(Hawai‗i State
Assessment)
Scaled scores, reading and math
Proficiency levels, reading and math
Spring 2002 (8
th

grade test)
Student transcripts Courses taken (e.g., rigor of math course)
Grades
2002-03
Graduation awards Diploma type earned
Date of diploma awarded
School graduated from
2005-09
College enrollment
(National Student
Clearinghouse)
Date of enrollment (by term)
Type of institution (2 year vs. 4 year)
2006-09
School characteristics
(Hawai‗i Department
of Education Trend
Report)
Student enrollment
Community socioeconomic status (Percent
of students eligible for Free/Reduced Price
Lunch)
2003
School climate
(Hawai‗i Department
of Education School
Quality Survey)
Percent of stakeholders responding
positively on dimensions of school quality
Average score by stakeholder group on
dimensions of school quality
2003

Data Analysis
The study examined longitudinal data for the cohort, tracking students‘
enrollment and educational experiences focused on the ninth grade transition into
high school as well as enrollment in institutions of higher education after high

44
school. The study used descriptive and inferential statistics to examine relationship
between students and their high school outcomes. SPSS 17 was used as the
statistical analysis tool for data analysis.
Dependent Variables
The dependent variable was students‘ high school outcome. Based on high
school graduation data from 2005-2008 and National Student Clearinghouse higher
education enrollment data in the fall following high school graduation, students were
identified as not graduating on-time, graduating but not attending college, graduating
and enrolling in two-year college, and graduating and enrolling in four-year college
(see Figure 3.2). Students did not graduate on-time if they dropped out of school,
continued to be enrolled in high school beyond four years, or graduated after four
years. Students graduated high school on-time if they completed high school with a
diploma or certificate within four years (by 2006). Students‘ college enrollment was
based on the type of institution in which they first enrolled—either a two- or four-
year college, as reported by the National Student Clearinghouse—and whether they
enrolled by June 2007, one year after graduating high school. Students who
graduated high school on-time but enrolled in college after one year were considered
as having graduated high school but not enrolled in college.

45
Figure 3.2: High School Outcomes of Sample Cohort (N=9,955)

Note: Proportion of sample in parentheses

Independent Variables
Three groups of independent variables were explored: students‘ background
characteristics, students‘ academic experiences and school climate.
Students’ background characteristics. The focus of the analysis was to
identify indicators associated with student success and failure, particularly during the
period of students‘ transition to high school. Therefore, background characteristics
consisted of information about students unrelated to their high school experience:
gender, ethnicity, birth date (to calculate age) and socioeconomic status. These data
were collected by schools and reported to the DOE. For the purpose of these

46
analyses, the students‘ information was based on their eighth grade records from
which the sample was drawn. DOE allowed students to self-identify as associating
with one of 13 ethnic groups; for this analysis ethnicity categories were collapsed
from 13 to 6 groups based on similarity of students‘ outcomes. Students were
considered ―overage‖ for the cohort if they were older than 14.99 years old in
December 2002; given Hawai‗i‘s kindergarten entry policy which considers a
student eligible for kindergarten if they turn five years old by December 31 in the
year they enter kindergarten (Kauerz, 2005), ―expected‖ age of students in the cohort
was 14 years old by December 31, 2002. Students were considered low-income if
they were eligible for Free or Reduced Price Lunch as eighth graders. Finally,
students‘ prior achievement as measured by test scores on the eighth grade Hawaii
State Assessment in reading and mathematics in 2002 was included as a background
characteristic; the analysis used the Hawai‗i State Assessment scaled scores which
range from 100 to 500 points for each subject.
Students’ educational experiences. Many studies report the importance of
students‘ academic experiences as predictors of students‘ post high school outcomes
(Adelman, 2006; Attewell and Domina, 2008; Roderick et. al. 2006). This analysis
considered students‘ academic experiences in high school, drawing from the
literature to identify likely predictors. Freshman year grade point average (GPA) and
the number of core subjects which students failed were examined, as Allensworth
and Easton (2007) identified those factors as on-track indicators for graduation in
Chicago Public Schools. Freshman year GPA was defined as the average of

47
students‘ final grades, weighted for the number of credits associated with each grade.
An ―A‖ grade was given four points, a ―B‖ grade was given three points, a ―C‖ grade
was given two points, a ―D‖ grade was given three points, and failing a course
resulted in no points. Core classes were defined as English, mathematics, science
and social studies courses. In Hawaii, these courses are all required for high school
graduation. Students in the cohort were required to complete the equivalent of four
credits or years of English and of social studies courses and the equivalent of three
credits or years of mathematics and of science courses to receive a high school
diploma (State of Hawai‗i Board of Education, 2010). This study measured ―failures
in core subjects‖ by counting the number of subjects in which students received a
final failing grade; for example, if a student took two, one semester mathematics
courses during the year (e.g., Algebra IA and Algebra IB) and failed the second
course, then their Algebra IB failure counted as a failure in mathematics. The level
of mathematics taken in freshman year and high school was also be considered as
many studies report the relationship of the rigor of high school mathematics course
taken to college going (e.g., Greene and Forester, 2003). These data were drawn
from students‘ high school transcripts which are reported in the DOE‘s student
information system. Students‘ on-time promotion from ninth grade was considered;
this information was derived from enrollment records indicating students‘ grade
level.
School climate. Students‘ educational experiences are mediated by the
school in which they are enrolled. Studies have identified the school climate as

48
affecting student outcomes. In this study, as with other studies, school climate was
measured at the school-level, rather than at the student-level, due to data availability.
Therefore, this study‘s measure of climate does not reflect the individual student‘s
perspective and experience, but the school environment as measured by the
collective perceptions of different school stakeholders. Results of DOE‘s School
Quality Survey were used for measures of school climate. School-level data from the
2003 administration of School Quality Survey were associated with students based
on the school they attended in their freshman year of high school.
The School Quality Survey, administered biennially to teachers, students
(grades 5, 8 and 11) and parents (randomly selected) asks students and teachers
questions for their opinions about their schools. SQS measures dimensions of
standards-based learning, quality student support, professionalism and capacity of
system, coordinated team work, responsiveness of the system, focused and sustained
action. The subscale ―Dimension A: Standards-Based Learning‖ was used as a
measure of academic press. The subscale represents the respondents‘ perception of
the emphasis the school placed on standards and implementation, rated on a five-
point, Likert-type scale. Heck and Hallinger (2009) found this subscale to have an
alpha of .91, indicating that the subscale is reliable (p. 670). This study used the
percentage of respondents who provided a positive response, 4 or 5 on the five point
scale, as a measure of positive response to the school‘s emphasis on standards-based
education. The intensity of positive response from parents, teachers, and students
was considered.

49
In addition to Hawai‗i‘s School Quality Survey results, this study considered
schools‘ academic offerings as a measure of academic press. The percentage of
ninth graders in the cohort taking Algebra or higher was included as a measure of
academic press as Lee and Burkam (2003) identified the dispersion of rigorous
coursetaking, typically measured by students‘ enrollment in mathematics courses, as
a measure of ―constrained curriculum.‖
Additionally, other school-level variables were used as controls: proportion
of students eligible for Free/Reduced Price lunch as an indicator of relative poverty
of the school community and school enrollment.
Statistical Analysis
Descriptive statistics. First, summary statistics about the cohort are presented
to describe the cohort on the dimensions of interest: background characteristics,
educational experiences, school climate and high school outcomes. The summary
statistics include frequencies, means and distributions of students‘ characteristics and
educational experiences. Cross-tabulations are presented on dimensions of interest,
such as ethnicity and high school outcomes, to describe the educational pipeline and
high school outcomes for different student groups.
Logistic regressions. Second, results of logistic regressions are presented.
Logistic regressions were used to analyze the relationship between students and their
high school outcomes. High school outcomes were defined as discrete outcomes:
failure to graduate high school on-time, graduate high school on-time but not enroll
in college immediately after high school, graduate on-time and enroll in a two-year

50
college, and graduate on-time and enroll in a four-year college. While the outcomes
represent a continuum of outcomes, the outcomes are ―non-metric‖ and categorical.
Thus, logistic regression was appropriate for analyzing the relationship of the
categorical dependent variable and independent variables (Schwab, n.d.). The
logistic regression assumes an underlying relationship between the independent and
dependent variables which can be described as a probability function (Cabrera,
1994). Binomial logistic regression was used to analyze the relationship between
independent variables and whether students‘ graduated high school on-time.
Multinomial logistic regression was used to analyze the relationship between
independent variables and the continuum of high school outcomes.
The SPSS (2009) output for the regressions included an odds ratio for each
independent variable, and the odds ratio represents the change in the odds of an
outcome for one unit change in the predictor and is calculated by exp
b
, where b is the
logistic regression coefficient. Thus, statistically significant odds ratios identified
independent variables related to outcomes.
The analyses were organized into three stages using different regression
techniques (see Figure 3.3) and consisted of six models described in Table 3.2. The
first stage of the logistic regression analyses focused on the outcome of high school
graduation. The first stage of the analysis compared students in the cohort who
graduated high school within four years and students who did not graduate within
four years. The second stage of the analysis included consideration of post-
graduation outcomes and considered factors related to college enrollment.

51
Figure 3.3: Statistical Analysis Plan for Modeling High School Outcomes
Categories Independent Variables Dependent Variables

Stage 1: Binary
logistic regression
Stage 2:
Multinomial
logistic regression

Students‘
background
characteristics
Gender
Ethnicity
Age
Socioeconomic
status
Prior academic
achievement
Stage 3: Hierarchal Linear Models (HLM)
Did not graduate on-
time
Did not graduate
on-time

Graduated high
school on-time
Graduated on-time,
did not enroll in
college
immediately

Students‘ high
school
experience
9
th
grade course
failures
9
th
grade GPA
9
th
grade retention
Rigor of 9
th
grade
math courses

Graduated on-time
and enrolled in 2-yr
college

Graduated on-time
and enrolled in 4-yr
college

Add as Level 2. School-level variables
Did not graduate on-
time
Did not graduate
on-time

High school
climate
Academic press:
positive standards-
based learning
environment,
proportion of 9
th

graders enrolled in
Algebra or higher
School size
(enrollment)
Community‘s
socioeconomic
status

Graduated high
school on-time
Graduated on-time,
did not enroll in
college
immediately

Graduated on-time
and enrolled in 2-yr
college

Graduated on-time
and enrolled in 4-yr
college

52

Table 3.2: Summary of Regression Models

Stage Model
Regression
Type Dependent Variable Independent Variables

On-time
high
school
graduation
High
school
outcomes
including
college
enrollment
Students‘
background
characteristics
9
th
grade
academic
experiences
School
climate
1 1
Binomial
logistic
X X
a

2
Binomial
logistic
X X
a
X
2 3
Multinomial
logistic
X X
a

4
Multinomial
logistic
X X
a
X
3 5
HLM
Binomial
logistic
X Level 1 Level 1
Level
2
6
HLM
multinomial
logistic
X Level 1 Level 1
Level
2
a
Included ethnicity

The first stage consisted of developing a model estimating the likelihood of
students graduating high school within four years or not. The model considered the
association of multiple independent variables (background characteristics, students‘
academic experiences, and school climate) with the dependent variable of graduating
high school on-time or not. Since logistic regression predicts the odds of an outcome
based as each unit changes in the dependent variable, it was appropriate to use a
binary logistic regression to analyze the relationships (Tabachnick and Fidel, 1996).

53
In the second stage, a model estimating the likelihood of range of students‘
high school outcomes including college enrollment was developed. The model‘s
dependent variable consists of four outcomes: not graduate high school on-time,
graduate on-time but not enroll in college within the first year, graduate and enroll in
a two-year college, and graduate and enroll in a four-year college. Since the high
school outcomes were categorical and there were more than two dependent variables
with this stage of the analysis considering college enrollment in addition to high
school graduation (without going to college) and not completing high school on-time
multinomial logistic regression was appropriate (SPSS, 2009). The multinomial
logistic regression results indicated the association of each unit change of
independent variables to each possible outcome, as compared with a reference group.
For these analyses, the referent group was the group graduating high school on-time
but not attending college.
Finally, in a third stage, multi-level modeling was used to also consider
school-level effects on high school outcomes. Students are nested within schools, so
Hierarchal Linear Modeling (HLM) considered the associations of both school- and
student-level independent variables on high school outcomes. This analysis
consisted of two-level models using HLM 6 software (Raudenbush et al., 2004).
This approach expanded upon the prior logistic regression models (Models 1-4) by
considering school effects.
The Level 1 model consisted of student-level characteristics identified in the
preferred logistic regression models, including students‘ background characteristics

54
of gender, socioeconomic status and age as well as students‘ ninth grade academic
performance as measured by year-end grade point average and failures in core
subject areas. The Level 2 model accounts for the relationship of school
characteristics to high school outcomes: school size, ―constrained curriculum‖
(percentage of sample enrolled in Algebra courses or higher in the ninth grade),
school community‘s socioeconomic status, and School Quality Survey results on
standards-based learning dimension. The HLM analyses were run for both
dependent variables: graduating high school on-time and high school outcomes
including college going. Odds ratios are reported for significant variables at the
student and school levels.
The results of the different analyses are models that identify factors
associated with different high school outcomes. These factors are potential
indicators for an early warning system since they identify opportunities for
interventions to support students exhibiting characteristics or behaviors found to be
off-track for positive high school outcomes in the studied cohort.
Limitations
This study is limited by the data and analytical tools used in the analysis:
quality of data collected and reported, difficulty in measuring certain constructs, data
available, and the limitations of analytical tools affect the strength of the validity of
the study. Also, the study examined variables that are highly correlated to one
another which made interpretation of the results more difficult.

55
Administrative data about students were limited by the nature of the data
collected. Sometimes data were limited because data sources were incomplete. For
example, the National Student Clearinghouse provides data about DOE graduates‘
enrollment in institutions of higher education. Clearinghouse includes 92% of higher
education enrollment nationwide (National Student Clearinghouse, 2009).
Clearinghouse includes University of Hawaii System enrollment where two-thirds of
Hawaii public school graduates enroll; however, one in-state private college, Hawaii
Pacific University does participate in the Clearinghouse so students enrolling at HPU
are mis-categorized as not enrolling in college (Education Trust, 2009); however,
only small numbers of Hawaii high school graduates attend HPU immediately after
high school. Also, individual student‘s college enrollment information was not
available if the student requested that the information not be disclosed, in accordance
with their rights under the Federal Educational Rights and Privacy Act. However,
other researchers, such as Roderick et al. (2006) at the Consortium on Chicago
School Research have relied on the Clearinghouse as the source of college
enrollment and completion data for college going studies of Chicago Public School
graduates in spite of the data source‘s limitations. The limitations affect the results
so that when graduates‘ postsecondary matriculation is not reported by
Clearinghouse, the student is mis-classified as graduated but not enrolled in college.
Data quality of the Hawaii‘s administrative data was an issue. For example,
school registrars sometimes enter data in the incorrect field or use invalid values
(Patton, 2008). Approximately 212 students were excluded from the sample because

56
their ninth grade transcript records contained an invalid number of credits; some
students‘ records, for example, contained more than eight credits because when they
transferred during the freshman year, their records were duplicated in the file.
Additionally, data about cohort students‘ special needs (e.g., special education,
English as a Second Language) did not reflect the states‘ official reports of special
needs incidence and DOE officials cautioned about the data quality for the specific
variable. Acknowledging the need for data quality, State of Hawaii DOE (2008)
identified data quality improvement as a priority in its successful application for the
U.S. Department of Education Statewide Longitudinal Data Systems Grant Program.
However, the improvements resulting from standardized policies and procedures to
increase the accuracy of information will not benefit the quality of data used for
these analyses.
Certain constructs are difficult to measure. Many studies identify factors
related to student success identify student engagement as significant. However,
Fredricks et al. (2004) found several measurement problems such as a common
definition of engagement that represents the multi-dimensional nature of engagement
and the practical matter of measuring engagement to reflect a multi-dimensional
concept. Also, quality of the high school experience is difficult to measure.
Measures of school climate, such as academic press or students‘ perception of the
relevance of the curriculum, are only proxies for the quality of students‘ high school
experience.

57
Furthermore, relevant data are sometimes not collected at the unit of interest.
The State of Hawaii DOE School Quality Survey measures campus climate at the
school-level. Thus, individual student‘s perceptions of their school are not linked to
their academic record. Thus, measures of school climate are applied to all students
in the school, irrespective of the individual student‘s experience. However, in
analyzing similar survey data aggregated at the school-level in Chicago, Allensworth
and Easton (2007) found that the school-level ―learning climate‖ measures to be
sufficiently robust to explain differences in student performance among schools.
Also, system-level DOE records only contain aggregated, school-level attendance
data but not student-level data on attendance.
Another limitation of the models is that multinomial logistic regression is
limited in explaining effect sizes. In a linear regression, the R-squared represents the
portion of variance explained by the regression model. However, logistic regression
does not have an equivalent measure (Cabrera, 1994). SPSS (2009) recommends
analyzing pseudo R-squared among the models to understand the relative
contribution of different models to explaining the outcome but not to express the
variance in the outcome explained by the model.
Finally, different student characteristics associated with high school dropout
or college attendance are often correlated with one another. For example, poor
school attendance contributes to course failure contributes to being retained in grade
because of a shortfall of credits needed to be promoted. Thus, it is difficult to
interpret the ―array of interconnected factors that shape student performance‖

58
(Roderick and Camburn, 1999, p. 335) and factors that predict the outcome may be
signaling underlying causal factors of weak academic skills, lack of motivation or
lack parental support for education (Neild et al., 2007).

59
CHAPTER FOUR
ANALYSIS
In this chapter, descriptive data are reported and analyzed, organized by the
research questions presented in Chapters One and Three. Next, the results of models
are summarized, for models exploring the outcome of graduating high school on-
time as well as models exploring the outcome of college enrollment choices. Finally,
the findings are presented, organized by research question.
Descriptive Statistics
Students’ Background Characteristics
The sample cohort of students was diverse. The largest ethnic group, for the
sample constructed for this analysis, was Pacific Islander (see Table 4.1). The
Pacific Islander group included Native Hawaiians and Samoans and represented
nearly 30% of the sample. Filipinos represented 24% of the sample. Together, the
white, black and Hispanic groups made up only 14% of the cohort. The group was
51% male, and 42% of students were low income as defined by eligibility for Free or
Reduced Price Lunch. The average age of the cohort in December of their eighth
grade year was 13.58 (SD 0.43); students were considered ―overage‖ if they were 14
years or older as of December 31, 2001, their eighth grade year, or 15 years or older
as ninth graders.
Students‘ prior academic achievement was measured using their eighth grade
test scores as a proxy. About 42% of students demonstrated proficiency in reading

60
and 20% of students demonstrated proficient in mathematics as eighth graders on the
2002 inaugural administration of the Hawai‗i State Assessment (see Table 4.2).

Table 4.1: Students‘ Background Characteristics, Categorical Variables (N=9,955)

Characteristic n % of Sample
Students‘ Background
Characteristics

Gender
Male 5,098 51.2
Female 4,857 48.8
Low-Income (Free/Reduced
Lunch eligible)
4,179 42.0
Ethnicity
―Asian‖
Chinese 357 3.6
Japanese 1,371 13.8
Korean 147 1.5
―Black and Hispanic‖
Black 109 1.1
Hispanic 168 1.7
Filipino 2,399 24.1
―Pacific Islander‖
Hawaiian 2,568 25.8
Samoan 325 3.3
―White‖
Portuguese 220 2.2
White 988 9.9
―Other‖
American Indian 21 .2
Southeast Asian 88 .9
Other 1,194 12.0
Overage 1,314 13.2

61

Table 4.2: Students‘ Prior Academic Achievement, Hawai‗i State Assessment, 2002

Scaled Score n Mean
Standard
Deviation Minimum Maximum
Reading 9,416 283.08 61.71 100 500
Mathematics 9,470 244.67 63.94 100 500

Students’ Academic Experiences
This study included analyses of students‘ freshman year transcripts that
included information about courses taken and grades received. Students in the
sample attempted between five and eight credits during their freshman year. This is
typical for Hawai‗i students since some schools have a ―traditional‖ schedule
consisting of six credits per year (representing six year-long courses, 12 semester
courses or a combination thereof) and other schools use alternate schedules, such as
a ―block‖ schedule, including up to eight course opportunities per year. As reported
in Table 4.3, about one-half of students took Algebra I or higher in their freshman
year with about 20% of students in Geometry or higher, indicating that they
completed Algebra I by the eighth grade which is regarded in the literature as a
strong predictor of future college enrollment (e.g., Smith, 1996).

62
Table 4.3: Students‘ Ninth Grade Educational Experiences, Categorical Variables
(N=9,955)

Characteristic n % of Sample
Students‘ Grade 9 Academic Rigor
Registered for Rigorous Mathematics Course
Algebra or Higher 5,126 51.5
Geometry or Higher 1,935 19.4
Students‘ Grade 9 Summative Outcomes
Retained in Ninth Grade 1,253 12.6

On average, students had a grade point average of 2.41, reflecting grades of
Bs and Cs (see Table 4.4). However, many students struggled academically during
their freshman year. On average, students failed one class (mean of 0.95, standard
deviation of 1.98). About 30% of students failed at least one class in a core subject
(i.e., English, mathematics, social studies, science). Thirteen percent of cohort was
retained and repeated ninth grade in 2003-04 school year (see Table 4.3). In the
2002-03 school year as ―true freshmen,‖ retained students received failing final
course grades in 2.74 of four core subjects on average (SD 1.06), compared with .29
average failures among students who promoted on-time to tenth grade (SD 0.67, not
tabled).

63
Table 4.4: Students‘ Ninth Grade Academic Performance, Scaled Variables

Characteristic n Mean
Standard
Deviation Minimum Maximum
Grade Point Average, Year End 9,955 2.41 1.00 0 4
Final Grades in Core Classes
English 9,911 2.23 1.22 0 4
Social Studies 9,842 2.37 1.18 0 4
Science 9,781 2.24 1.27 0 4
Mathematics 9,955 2.15 1.25 0 4
Final Failing Grades in Core
Classes

English 9,911 .17 .51 0 7
Social Studies 9,842 .20 .53 0 5
Science 9,781 .12 .35 0 4
Mathematics 9,781 .09 .29 0 1
Total Number of Fs (of 4) 9,955 .77 1.57 0 16

Graduation outcomes were better for students who passed more freshman
year classes. Students who did not graduate on-time failed an average of 1.78 core
subjects (not tabled). Descriptive data also showed that 93% of students who passed
all core subjects in ninth grade graduated on-time compared with only 77% of
students who failed a core subject. Fewer than one in four students graduated on-
time if they failed all core subjects in freshman year (not tabled).
School Climate
School climate was measured on the basis of two academic press criteria and
two school characteristics. Stakeholders‘ responses on the 2003 School Quality
Survey Standards-Based Learning Dimension provide data on different points of

64
view about the extent to which schools had a standards-based learning environment
(see Table 4.5). This study used the measure of stakeholders‘ positive response that
their school had a standards-based learning environment for analysis. For the cohort
studied, Hawai‗i high school students were the least generous stakeholder group
among teachers and parents in rating schools‘ learning environment on the School
Quality Survey, and where more students rated their schools positively on statements
such as ―My schoolwork is challenging,‖ and ―I feel my classes are preparing me for
future education and work,‖ high school graduation outcomes were better (State of
Hawai‗i DOE, 2010). Overall, teachers were the most optimistic with 80% of
teachers reporting that their school had a standards-based learning environment.
Parents were next most optimistic with 56% perceiving their child‘s school to be
implementing standards-based learning, and students were the most critical with less
than half reporting a standards-based learning environment. Additionally, a measure
of rigorous coursetaking was used to measure academic press. The percent of the
sample enrolled in Algebra or higher as a ninth grader was used as an indicator of
whether a school had a constrained curriculum (Lee and Burkam, 2003). In the
sample, the school average was 55% of students were enrolled in Algebra or higher,
but the range of Algebra enrollment among schools was wide with as few as 22%
enrolled in ninth grade to as many as 78% enrolled.

65
Table 4.5: School Climate Variables (N=44)

Variable Mean
Standard
Deviation Minimum Maximum
School Quality Survey: Standards-
Based Learning (Dimension A), 2003

Students‘ response (percent
positive)
46.26 5.24 36.70 59.95
Teachers‘ response (percent
positive)
79.50 6.77 46.43 90.26
Parents‘ response (percent positive) 56.36 8.87 37.34 77.78
Constrained curriculum (percent of
cohort enrolled in Algebra or higher as
ninth graders, 2002-03)
54.77 24.63 21.56 100.00
School size (enrollment, 2003) 1255 595 29 2421
School community‘s socioeconomic
status (percent eligible for Free or
Reduced Price Lunch, 2003)
41.41 21.01 8.50 96.55

Other school characteristics ranged as well. School size varied with school
enrollment ranging from 29 to 2,421 students. School enrollment averaged 1,255
with a large standard deviation of 595 indicating a variety of school sizes. Likewise,
schools‘ socioeconomic status varied widely. The average concentration of
Free/Reduced Lunch students was 41% with a large standard deviation of 21%.
Some schools had as few as nine percent of students on Free/Reduced Lunch while
other schools had 97% of students on Free/Reduced Lunch.
Student Outcomes
Most students in the cohort graduated high school (see Figure 3.2). Eighty-
three percent graduated high school on-time, earning a diploma or certificate by June

66
2006. Seventeen percent of students did not graduate on-time. Of the 1,725 students
who did not graduate on-time, 1601 (93%) dropped out, 115 graduated late (7%) and
9 students were still enrolled two years after their cohort graduated (see Figure 3.1).
Among students who graduated on-time, 4,509 (54%) enrolled in college within a
year of high school. One-third of graduates enrolled initially in two-year colleges
and 21% enrolled in four-year colleges.
The descriptive data provides information about Hawai‗i‘s educational
pipeline, at least for the cohort of ninth graders graduating high school in 2006. The
NCHEMS (2009) data comparing states‘ educational pipeline productivity
(presented in Figure 1.1) used cross-sectional data to report graduation and college
going rates, but using longitudinal data, this study allowed for tracking individual
students through the educational pipeline from ninth grade to high school graduation
and college enrollment. The analysis found that 83% of the sample graduated on-
time; this is more similar to the DOE‘s report that 80% of students graduate on-time
(2009) than NCHEMS‘ (2009) cross-sectional data report of 65% of ninth graders
graduating on-time.
Students‘ high school outcomes differed by their characteristics. Students
who demonstrated proficiency on the 2002 Hawai‗i State Assessment in eighth grade
reading or mathematics as well as Asian students had the highest graduation rates
with 90% of students in each group completing high school on-time, and enrolling in
a two or four year college immediately after high school (see Figure 4.1). Students
who were overage for their freshman year or who scored at the level of ―well below

67
proficient‖ on the 2002 Hawai‗i State Assessment had the lowest rates of graduating
on-time with 25% or more of students not graduating on-time (see Figure 4.2). The
lowest college going rates are among students scoring well below proficient on the
state assessment in either math (23%) or reading (18%). Chi-square tests for gender,
ethnic group, income, age, and prior academic performance were all significant (see
Table 4.6).

Figure 4.1: High School Outcomes for Students, by Ethnic Group

Percent of students in each sub-group

68

Figure 4.2: High School Outcomes for Students, Based on Prior Academic
Achievement

69
Table 4.6: High School Outcomes for Student Sub-Groups Based on Students‘
Background Characteristics (N=9,955)

Graduated high school on-time
Student sub-groups
Did not
graduate high
school on-time
Did not
enroll in
college
Enrolled in
two-year
college
Enrolled in
four-year
college
Ethnic Groups****
Asian 9% 18% 33% 40%
Black/Hispanic 18% 43% 27% 12%
Filipino 13% 39% 33% 14%
Other 21% 36% 24% 19%
Pacific Islander 24% 49% 22% 6%
White 19% 36% 25% 20%
Gender****
Female 16% 34% 28% 22%
Male 19% 41% 27% 14%
Socioeconomic Status****
Low-Income 24% 46% 21% 8%
Not Low-Income 12% 31% 32% 25%
Age****
Expected age 15% 38% 28% 19%
Overage 32% 36% 22% 10%
Prior Academic Achievement, 8
th
grade Hawai‗i State Assessment in Mathematics****
Well below proficiency 27% 50% 21% 2%
Approaching
proficiency
15% 38% 33% 14%
Proficient 6% 20% 24% 50%
Exceeds proficiency 1% 16% 6% 77%
Prior Academic Achievement, 8
th
grade Hawai‗i State Assessment in Reading****
Well below proficiency 27% 55% 17% 1%
Approaching
proficiency
20% 43% 30% 7%
Proficient 9% 27% 29% 35%
Exceeds proficiency 3% 18% 10% 70%
Note: Chi-square tests significant at *p<.10, **p<.05, ***p<.01, ****p<.001

70
There are also significant differences among schools in their students‘ outcomes.
On-time high school completion rates by schools ranged from 45-97% with an
average of 84% (not tabled). The college going rate (two and four year college
enrollment combined) ranged from 18-71% with an average of 45% (not tabled).
Table 4.7 presents the inter-correlations between independent variables.
While most independent variables have a statistically significant correlation with one
another, the correlations are strong in only a few cases. There was a strong
correlation between students‘ Hawaii State Assessment scores in reading and
mathematics (.73). Likewise, there was a strong correlation between ninth grade
GPA and core courses failed in ninth grade (-.76). Weaker though significant
correlations existed between ninth grade performance and prior academic
achievement as measured by the Hawaii State Assessment (e.g., .47 correlation
between ninth grade GPA and Hawaii State Assessment reading score). However,
most associations were less than .20. Fewer than one-half of the associations had a
correlation of greater than .10.

71
Table 4.7: Correlation Matrix of Independent Variables for Analysis

Low-
income Male Overage
Prior
achievement in
reading (scaled
score)
Prior
achievement in
math (scaled
score)
Grade point
average (9
th

grade)
Failures in
core
subjects
Students‘ positive
perception of school‘s
standards-based
learning environment
Low-income Pearson
Correlation
1 .008 .078
**
-.250
**
-.272
**
-.207
**
.160
**
.112
**

Sig. (2-tailed) .442 .000 .000 .000 .000 .000 .000
N 9955 9955 9955 9416 9470 9955 9955 9950
Male Pearson
Correlation
.008 1 .072
**
-.231
**
-.062
**
-.162
**
.066
**
-.021
*

Sig. (2-tailed) .442 .000 .000 .000 .000 .000 .036
N 9955 9955 9955 9416 9470 9955 9955 9950
Overage Pearson
Correlation
.078
**
.072
**
1 -.091
**
-.093
**
-.096
**
.101
**
.043
**

Sig. (2-tailed) .000 .000 .000 .000 .000 .000 .000
N 9955 9955 9955 9416 9470 9955 9955 9950
Prior
achievement in
reading (scaled
score)
Pearson
Correlation
-.250
**
-.231
**
-.091
**
1 .731
**
.466
**
-.257
**
-.004
Sig. (2-tailed) .000 .000 .000 .000 .000 .000 .725
N 9416 9416 9416 9416 9164 9416 9416 9414
Prior
achievement in
math (scaled
score)
Pearson
Correlation
-.272
**
-.062
**
-.093
**
.731
**
1 .492
**
-.274
**
-.031
**

Sig. (2-tailed) .000 .000 .000 .000 .000 .000 .002
N 9470 9470 9470 9164 9470 9470 9470 9467

72
Table 4.7, Continued

Low-
income Male Overage
Prior
achievement
in reading
(scaled score)
Prior
achievement in
math (scaled
score)
Grade
point
average (9
th

grade)
Failures in
core
subjects
Students‘ positive
perception of
school‘s standards-
based learning
environment
Grade point average
(9
th
grade)
Pearson
Correlation
-.207
**
-.162
**
-.096
**
.466
**
.492
**
1 -.759
**
.028
**

Sig. (2-tailed) .000 .000 .000 .000 .000 .000 .006
N 9955 9955 9955 9416 9470 9955 9955 9950
Failures in core
subjects
Pearson
Correlation
.160
**
.066
**
.101
**
-.257
**
-.274
**
-.759
**
1 .015
Sig. (2-tailed) .000 .000 .000 .000 .000 .000 .124
N 9955 9955 9955 9416 9470 9955 9955 9950
Students‘ positive
perception of
school‘s standards-
based learning
environment
Pearson
Correlation
.112
**
-.021
*
.043
**
-.004 -.031
**
.028
**
.015 1
Sig. (2-tailed) .000 .036 .000 .725 .002 .006 .124
N 9950 9950 9950 9414 9467 9950 9950 9950
Note: Correlation significant at the *.05 level, ** .01 level.

73
Summary of Models
Statistical analyses were conducted to consider the relationship of multiple
factors associated with students‘ high school outcomes. Analyses consisted of a
series of logistic regressions on the longitudinal student data to identify risk and
success factors associated with positive high school outcomes. These regression
analyses addressed the shortcomings, pointed out by Gleason and Dynarski (1998),
of relying solely on associating single risk factors with outcomes using descriptive
statistics.
Six models were developed to describe the relationship of various student and
school characteristics and high school outcomes (see Table 3.2 for descriptions of
models). Models 1, 2 and 5 focused on the outcome of on-time high school
graduation, and Models 3, 4, and 6 also considered enrollment in two- and four-year
colleges as high school outcomes. Models 5 and 6 used HLM to consider school-
level characteristics in addition to student-level variables. The models are further
described in the Appendix.
The odds ratios of statistically significant independent variables are
summarized on Table 4.8. Variables associated with increased probability of at least
10% more positive outcomes are presented in Table 4.9. Then the statistically
significant findings of the models are organized by research question.

74
Table 4.8: Summary of Models‘ Significant Odds Ratios

Model 1 2 3
a
4
a
5 6
a

Grad on-
time
Grad on-
time
Not grad
on-time
Enroll in
2-year
college
Enroll in
4-year
college
Not grad
on-time
Enroll in
2-year
college
Enroll in
4-year
college
Grad on-
time
Not grad
on-time
Enroll in
2-year
college
Enroll in
4-year
college
Student
Characteristics

Background
Characteristics

Low-Income .63 (1.60) .66 (1.52) 1.26 .61 (1.64) .54 (1.85) 1.24 .61 (1.64) .53 (1.88) .72 (1.40) .59 (1.69) .45 (2.22)
Male .88 (1.14) .79 (1.23) .60 (1.68) .83 (1.21) .82 (1.22) .86 (1.12) .84 (1.18) .85 (1.17)
Overage .43 (2.34) .42 (2.39) 2.17 .73 (1.38) 2.23 .87 (1.16) .63 (1.59) .42 (2.35) 2.30 .60 (1.66)
Prior Achievement
(for 10 point score
increase)

Reading, 2002
(Scaled Score)
1.02 1.02 1.13 1.02 1.02 1.08 1.02 1.01 1.09
Mathematics,
2002 (Scaled
Score)
1.06 1.03 1.03 1.16 1.02 1.09 1.02 1.11
Ethnicity
Asian 1.68 1.72 2.28 2.89 2.28 2.88
Black/Hispanic 1.46 1.85 .70 (1.42) .56 (1.78)
Filipino 1.92 1.99 .57 (1.76) 1.35 .55 (1.83) 1.34
Other 1.24 1.39 1.34
Pacific Islander 1.49 .83 (1.20) .81 (1.24) .50 (2.02) .63 (1.59) .82 (1.21) .60 (1.67)
White
b

Note. Bold indicates statistical significance at 99% level or higher; (reciprocal odds ratio for significant negative coefficients)
a
Referent group was students who graduated on-time but did not enroll in college immediately
b
Referent group for ethnic groups
Schools
Ninth graders in sample cohort

75
Table 4.8, Continued

Model 1 2 3
a
4
a
5 6
a

Grad
on-
time
Grad on-
time
Not
grad
on-
time
Enroll in
2-year
college
Enroll
in 4-
year
college
Not grad
on-time
Enroll in
2-year
college
Enroll in
4-year
college
Grad on-
time
Not grad
on-time
Enroll in
2-year
college
Enroll in
4-year
college
Ninth Grade Academic
Performance

Grade Point Average
(Year-End)
3.69 .335 (2.99) 1.14 4.83 3.95 .31 (3.27) 1.27 5.58
Core Subjects Failed .81 (1.23) 1.25 .90 (1.15) .85 (1.17) 1.22 -
School Characteristics
Academic Press
Students‘ perception of
schools‘ standards-based
learning environment (percent
positive, z-score)
1.33 .71 (1.40) .85 (1.17)
Constrained curriculum
(percent of cohort enrolled in
Algebra or higher)

School size (for 100 student
increase in enrollment)
1.04 1.02
School community‘s
socioeconomic status (for 10
percentage point increase in
proportion of students who are
low-income)
.91 (1.10) .92 (1.08)
Pseudo R
2
(Nagelkerke) .070 .220 .343 .507
Reliability estimate (Level 1) .793 .780 .676 .739
Note. Bold indicates statistical significance at 99% level or higher; (reciprocal odds ratio for significant negative coefficients)
a
Referent group was students who graduated on-time but did not enroll in college immediately
b
Referent group for ethnic groups

76
Table 4.9: Summary of Variables Associated with Increased Probability for Positive
High School Outcomes

Increased
probability
of positive
outcome
Graduate high
school on-time
(compared with not
graduating on-time)
a

Not graduate
on-time
b

Enroll in 2-year
college
Enroll in 4-year
college
(compared with graduating on-time)
c

10-24% ―Other‖ ethnic group
(vs. White)
Passed one more
core subject
Lower concentration
of poverty in school
community (10
percentage point
decrease)
Failed one
more core
subject
Female
White (vs.
Pacific Islander)
Fewer students
report positive
standards-based
learning
environment (1
SD)
Female
Higher 8th
grade
mathematics
score (10
points)
25-49% Pacific Islander (vs.
White)
More students report
positive standards-
based learning
environment (1 SD)
Fewer students
report positive
standards-
based learning
environment (1
SD)
Filipino (vs.
White)
One point
higher GPA

50-74% Higher income (not
Free/Reduced
Lunch)
Asian (vs. White)
White (vs.
Pacific
Islander)
Higher family
income (not
Free/Reduced
Lunch)
Expected age
Other ethnic
group (vs.
White)
White (vs.
Pacific Islander)
Expected age
Black/Hispanic (vs.
White)
Filipino (vs. White)
One point higher
GPA
Overage
White (vs.
Black/Hispanic,
Filipino)
One point
lower GPA
Asian (vs.
White)
Higher income
Asian (vs.
White)
One point
higher GPA
Note: Early warning student-level indicators for students in bold and school level indicators
in italics.
a
Based on results of Models 2 and 5
b
These variables are associated negatively with high school outcomes since they represent
significant factors associated with not graduating high school on-time (compared with
graduating on-time).
c
Based on results of Models 3 and 6

77
Research Question 1: Which of students’ background characteristics are risk or
success factors?
Many of students‘ background characteristics analyzed were associated with
their high school outcomes. Being the expected age or higher income was
significantly increased the odds of positive outcomes: graduating on-time or
graduating and enrolling in college. Students from higher income families were
more likely to have successful outcomes irrespective of whether ninth grade
academic performance was considered. In Model 4 which considered ninth grade
academic performance, higher income students were 1.24 times more likely to
graduate on-time and had an increased likelihood of attending college by 1.64 times
for two-year colleges and 1.88 times for four-year colleges. Gender mattered though
the significance of gender varied across the models and outcomes; males often had
an advantage for graduating high school on-time but female graduates had a slight
advantage over male graduates for enrolling in college.
Ethnicity. The regressions identified a complex relationship between ethnicity
and the outcomes. Being white was not an advantage in this Hawai‗i cohort. For
example, Filipinos had a higher likelihood of graduating on-time by 1.76 times in
Model 3 (see Table 4.7). Being Filipino was more of an advantage to graduating on-
time than having higher socioeconomic status. Students in Black/Hispanic and
Pacific Islander groups were both more likely to graduate high school on-time than
whites: 1.42 and 1.20 times, respectively.

78
When ninth grade academic performance was considered in Model 4, the
disadvantage of being white for graduating on-time increased. Compared with
Model 3, the statistical significance and the likelihood of graduating on-time
increased for students of color compared with whites. For Black/Hispanic students
compared with white students, when freshman grades were considered, the
likelihood of graduating on-time increased from 1.42 times to 1.78 times. The
likelihood of graduation for Filipinos increased from 1.76 to 1.83 times and for
Pacific Islander from 1.20 to 1.59 times. However, when ninth grade academic
performance was considered, the relationship of ethnicity with college enrollment
outcomes did not change notably.
Whether ninth grade academic performance was considered or not, ethnicity
had a strong relationship to college going outcomes for some ethnic groups. Asians
had higher likelihood of enrolling in higher education: 2.28 times more than whites
for two-year colleges and 2.88 times more than whites for four-year colleges (see
Table 4.8). While Pacific Islanders had an advantage over whites for graduating
high school on-time, whites were more likely than Pacific Islanders to attend college:
1.21 times more likely to attend two-year colleges and 1.67 times more likely to
attend four-year colleges. Black/Hispanic and white students did not appear to differ
in their college enrollment patterns once other background characteristics and
academic performance were considered.
Prior academic achievement. Prior academic achievement, as measured by
students‘ eighth grade Hawai‗i State Assessment scores (2002), only mattered for

79
certain high school outcomes. Prior academic achievement correlated highly with
positive high school outcomes but once background characteristics were considered,
did not have much impact on whether students graduated high school on-time. The
models showed only small increases in likelihood of graduating on-time. The
highest coefficient for prior academic achievement was in Model 1; a ten point
higher mathematics score on the eighth grade test (of 500 possible points; s.d. of 64
points) increased the student‘s probability of graduating on-time by six percent. In
both binary and multinomial logistic models, when ninth grade academic
performance is considered, the effect of prior achievement was reduced or became
insignificant in explaining whether students graduated high school on-time.
However, higher prior academic achievement had a more meaningful
relationship with college enrollment. Higher eighth grade test scores were associated
positively with college enrollment, particularly for four-year colleges. However,
ninth grade academic performance mediated the relationship of prior achievement
with college enrollment, especially at four-year colleges, as evidenced by smaller
odds ratios for prior achievement in Model 4 than Model 3.
Research Question 2: Do indicators of risk or success vary by type of high school
outcome?
Different relationships between students‘ characteristics, their ninth grade
performance, and their high school outcomes were revealed between the binary
logistic models (Models 1 and 2) which focused on graduating high school on-time
and the multinomial logistic models (Models 3 and 4) which considered college

80
enrollment outcomes. The multinomial models which considered variation among
on-time graduates‘ postsecondary enrollment choices were stronger models; they had
higher pseudo R-squareds than the binary logistic models indicating that the
multinomial models explained more of the variance in the data (see Table 4.8). Also,
differences between comparing binary and multinomial logistic models results in the
odds ratios and statistical significance of the independent variables indicated that
there were different structures of the equations when considering enrollment in
different types of colleges compared with just examining whether students graduated
high school on-time.
Considering college enrollment variables in Models 3 and 4 revealed a
complex relationship of ethnicity and outcomes. In Models 1 and 2, all ethnic groups
of color had an advantage over whites in completing high school on-time, and these
relationships increased in significance and magnitude when background
characteristics and ninth grade academic performance were considered (see
Appendix, Table A1). Similarly, when comparing graduating high school on-time
and not graduating on-time in Models 3 and 4, accounting for freshman year
academic performance increased the significance and magnitude of the odds of
graduating high school for select ethnic groups: Black/Hispanic, Filipino and Pacific
Islander. However, the relationship of ethnicity and college enrollment outcomes
varied. Being Asian increased the odds of attending a two or four year college
significantly, by 2.28 and 2.88 times respectively compared with whites (see
Appendix, Tables A2 and A3). Filipinos were 1.35 times more likely to attend a two-

81
year college than whites with the similar background characteristics but did not have
different odds than whites for enrolling in a four-year college. Conversely, students
classified in the ―other‖ ethnic group had the same odds as whites of attending a two-
year college but were 1.39 times more likely to attend a four-year college. Pacific
Islanders, students who identified themselves as Hawaiian or Samoan in the DOE
system, were less likely than whites to attend college. Being white instead of Pacific
Islander increased students‘ likelihood of going to a two-year college by 1.24 times
and four year college by 2.02 times. When other background characteristics and
academic performance were equal, Pacific Islanders were the only ethnic group at
risk for worse college going outcomes compared with whites.
Research Question 3: What is the relationship of students’ academic experiences in
the high school freshman year to their high school outcomes? And to what extent do
risk or success factors vary by students’ ninth grade academic performance?
Students‘ ninth grade academic performance contributed to explaining their
high school outcomes. For models considering either set of outcomes--graduating
on-time and/or college enrollment--the explanatory power of the model increased, as
measured by pseudo R-squared, when ninth grade academic performance was added
to the model of students‘ background characteristics. In general, ninth grade
performance variables were statistically significant and meaningful.
In these analyses, students‘ ninth grade performance was measured by the
average of their grades across all courses (GPA) and number of core subjects that
they failed. Many ninth grade performance measures, including course grades in core

82
subjects, number of failed courses, proportion of failed courses, first semester grades,
and the level of rigor of mathematics courses, had considered and tested for the
analyses. However, models which incorporated the average of final course grades
(GPA) and number of failed core subjects had the highest pseudo-R squares, and the
variables were significant when included in the model.
Freshman year grade point average. Grades were a robust predictor of
positive high school outcomes. Grade point average (GPA) was a statistically
significant at p < .001 across Models 2, 4 and 6. In Model 2, a one point increase in
GPA (on a scale of 4 in which 4 represents a ―A‖ and 0 represents an ―F‖) increased
the likelihood of graduating high school on-time by 3.69 times. GPA was a much
more powerful of a predictor than any other variable; the odds ratio of GPA was 3.69
compared with the next highest magnitude odds ratio of 2.39 for students being
overage for ninth grade. In Models 4 and 6 which also examined college enrollment
outcomes, GPA was significant for each of the outcomes though the predictive
power of GPA varied across the outcomes. Having a one point higher GPA
increased the likelihood of graduating on-time vs. not graduating on-time by 2.99
times in Model 4 and 3.27 times in Model 6. Compared with graduating on-time but
not going to college, having a higher GPA increased the likelihood of enrolling in a
two-year college by 1.14 times in Model 4 and 1.27 times in Model 6. Grade point
average had an especially strong association with students‘ enrollment in a four-year
college. An increase of one point in GPA increased the likelihood of enrolling in
four-year colleges by 4.83 times in Model 4 and 5.58 times in Model 6.

83
Among all of the independent variables in the models, an incremental
(though non-trivial) increase in GPA of one point (one standard deviation) had the
largest impact on outcomes. For example, students who did not graduate on-time
had an average GPA of 1.78 (not tabled), representing grades of mostly Cs and some
Ds; an improvement in grades of one standard deviation (1.50 points) would result in
a GPA of 3.28, representing grades of mostly Bs and a couple of As, and nearly six
times increased likelihood of graduating on-time. For on-time graduates, a one
standard deviation increase in GPA of .84 points to 3.13 points would have resulted
in increased likelihood of college enrollment: 1.27 times for two-year college and
5.59 times for four-year college. The impact of GPA was much greater on
graduating on-time or enrolling in four year colleges than enrolling in two-year
colleges.
Grades were significant predictors of students‘ high school outcomes. In
particular, grades differentiated students who graduated on-time vs. those who did
not and also differentiated among on-time graduates who chose four-year colleges
from those who did not go to college. A GPA that was one point higher increased
the probability of the aforementioned better outcomes by more than 75% (see Table
4.9).
Failures in core subjects. Failing courses in the core subjects was related to
high school outcomes, particularly for the graduating high school on-time. In Model
1, each additional core subject failure increased the likelihood of not graduating on-
time by 1.23 times. Even when college going outcomes were considered in Model 4

84
and school-level effects were considered in Models 5 and 6, the relationship of
failing core subjects and the likelihood of students graduating on-time remained
relatively stable, ranging from 1.17 to 1.25 times. Failing a core subject increased
the likelihood of not graduating on-time by 1.23 times (Model 6) and failing two
core subjects increased the likelihood of not graduating by 1.50 times (calculated,
not tabled). Conversely, passing all four core subjects nearly doubled the likelihood
of graduating on-time (calculated, not tabled). However, association between
failures in core subjects and outcomes diminished as college enrollment was
considered in Models 4 and 6. Compared with graduating on-time, the significance
and magnitude of the odds for failing core subjects were reduced for enrolling a two-
year college, and were statistically insignificant for enrolling in a four-year college.
Change in risk factors. Ninth grade academic performance was related to
high school outcomes. When ninth grade performance was considered in the
relationship between students‘ background characteristics and on-time high school
graduation, ninth grade performance variables were significant.
When accounting for college enrollment outcomes, however, the impact of
academic performance on students‘ other background characteristics were mixed.
Ninth grade performance explained some of the college going advantage of girls; for
example, the odds ratio for females enrolling in four-year college decreased from
1.68 to 1.16 between Models 3 and 4. Likewise, considering freshman year
performance decreased the odds associated with eighth grade test scores for students
who enrolled in four-year colleges.

85
All ethnic groups of color were more likely than whites to complete high
school on-time, and these advantages increased in significance and magnitude when
ninth grade academic performance was considered in addition to background
characteristics (see Table 4.8, comparing Models 1 and 2, and Models 3 and 4).
However, when college enrollment was also considered in Models 3 and 4, the
relationship of ethnicity and outcomes became more complex. Controlling for
freshman year academic performance in Model 4 increased the significance and
magnitude of the odds of graduating high school for select ethnic groups—
Black/Hispanic, Filipino and Pacific Islander—compared with Model 3. However,
the effect of freshman year grades and college enrollment varied among ethnic
groups. The magnitude of the odds changed for some groups. While the odds of
college enrollment did not change significantly for Asians or Filipinos, considering
freshman year grades reduced the odds of other significant relationships. Students in
the ―other‖ ethnic group were 1.34 times more likely to enroll in a four year college
compared with whites, and white students were 1.21 times as likely as Pacific
Islanders to enroll in a two-year college and 1.67 times as likely as Pacific Islanders
to enroll in a four-year college.
Research Question 4: To what extent does the school attended relate to outcomes?
Multi-level Models 5 and 6 found significant school and student effects
related to high school outcomes.
School effects. The academic press variable was based on students‘ responses
on the 2003 School Quality Survey and concentration of ninth grade enrollment in

86
rigorous math courses. Analyses suggested a relationship between stronger
perceived school academic orientation of a school and increased likelihood of
students graduating on-time. In Models 5 and 6, more positive student responses
about their schools‘ standards-based learning environment (one standard deviation
higher) increased the likelihood of graduating high school on-time by 1.33 and 1.40
times, respectively. However, the proportion of students in the sample cohort taking
Algebra as ninth graders did not have any significant association with high school
outcomes.
Other school effects were identified. Smaller school size increased the
likelihood of students graduating on-time. Students attending schools with 100
fewer students had a slightly higher probability of graduating on-time: 4% when
considering on-time graduation only and 2% when also considering college
enrollment. Schools‘ socioeconomic status had some effect on on-time graduation.
Attending a school with ten percentage point less in proportion of students eligible
for Free and Reduced Lunch was associated with a 10% higher probability of
graduating on-time.
The relationship of college enrollment outcomes and school effects was more
complex. None of the school-level effects was significant in comparing students
who enrolled in four-year college with students who did not enroll. However, there
were school effects when comparing students who enrolled in two-year colleges with
students who graduated but did not choose to enroll in higher education. For
individuals with the same background characteristics and ninth grade performance,

87
the high school attended affected the probability of positive outcomes; graduates of
schools with fewer students reporting positive standards-based learning environment
or lower concentration of poverty were more likely to choose two year colleges than
not attend college.
Student effects. When school effects were considered, some of the
relationships of students‘ characteristics or academic performance with students‘
high school outcomes changed. In Models 5 and 6, accounting for school
characteristics reduced the negative impact of individual students‘ low-income status
on graduating high school on-time or choosing a two-year college; in fact, when
school variables are considered in Model 6, students‘ Free/Reduced Lunch status and
the poverty level of the school they attend were not associated significantly with
students graduating on-time. Conversely, considering school effects intensified the
impact of individual students‘ ninth grade performance, particularly their grade point
average. Higher GPA (e.g., having Bs instead of Cs) increased the likelihood of
students graduating on-time (vs. not graduating on-time) by three times and the
likelihood of enrolling college (vs. not enrolling) by 1.27 times for two-year colleges
and 5.59 times for four-year colleges.
Summary
The analyses identified a number of independent variables associated with
positive high school outcomes. The regression analyses revealed effects of both
individual and school-level characteristics on outcomes. Students‘ prior academic
performance on eighth grade state assessments correlated strongly with high school

88
outcomes but the association decreased significantly when considering students‘
ninth grade academic performance. Students‘ freshman year academic performance
had the strongest association with positive outcomes, particularly graduating on-time
and enrolling in a four-year college.
The relationship of ethnicity and outcomes was complex. In general, whites
were not particularly advantaged once controlling for other background factors and
freshman year performance. However, being Asian more than doubled the
likelihood of students going to college, either a two- or four-year college. While
Pacific Islanders were more likely than whites to graduate high school on-time, the
Pacific Island graduates were less likely to enroll in college.
Regression models yielded different results when comparing on-time
graduation and college enrollment outcomes. Risk and success factors for college
enrollment differed among graduates depending on their post-high school outcomes.
Multi-level models further identified some relationships between outcomes and
school factors.

89
CHAPTER FIVE
DISCUSSION AND CONCLUSION
This section begins with a brief summary of the study followed by a
summary and discussion of the results. Next, implications for practice and research
will be discussed. Finally, conclusions will be presented.
Summary of the Study
A number of initiatives seek to increase the educational levels of Hawai‗i‘s
citizens. Increasing the productivity of the ―educational pipeline‖ is a strategy for
increasing the educational attainment in Hawai‗i. These strategies are represented by
the University of Hawai‗i President‘s Hawai‗i Graduation Initiative (2010), Hawai‗i
P-20 Partnerships for Education‘s goal that 55% of working age adults have a two-
or four-year degree (Hawai‗i P-20, 2010), and a Memorandum of Agreement
between Hawai‗i‘s Governor, Superintendent of Schools, and UH President (Hawai‗i
Economic Stimulus Oversight Commission, 2010). Based on cross-sectional data,
NCHEMS (2009) estimated that in Hawai‗i, of 100 ninth graders, only 68 will
graduate high school on-time, and only 41 will college immediately after high
school.
This purpose of this study was to examine the ninth grade to college
enrollment components of Hawai‗i‘s educational pipeline and to identify indicators
of whether students are on-track for positive high school outcomes, both graduating
on-time and enrolling in college. These indicators are intended to inform

90
development of an ―early warning system‖ which flags students falling off-track for
positive outcomes for intervention.
This study focused indicators of whether students are on-track during their
freshman year of high school. The study analyzed one cohort of students, tracking
them longitudinally from ninth grade entry for six years. The study examined the
relationship of students‘ high school outcomes with their background characteristics,
ninth grade academic performance, and the climate of the school they attended as
freshmen. The study focused on the question: Which of students‘ high school
experiences in the ninth grade year indicate whether they are ―on-track‖ for positive
high school outcomes? The following sub-questions were examined:
1. Which of students‘ background characteristics are risk or success factors?
2. Do indicators of risk or success vary by type of high school outcome
(graduating on-time and college enrollment)?
3. What is the relationship of students‘ academic experiences in the high
school freshman year to their high school outcomes? And to what extent
do risk or success factors vary by students‘ ninth grade academic
performance?
4. To what extent is the school attended related to outcomes?
The study employed quantitative methods to answer the questions. A dataset
of longitudinal student and school data for a cohort of students was created. The
dataset included students‘ background characteristics, ninth grade academic
performance, information about the schools that students attended as ninth graders,

91
and students‘ high school outcomes. Logistic regression was conducted on two
dependent variables: (1) on-time high school graduation, and (2) high school
outcomes including college going. Six models were created, three for each
dependent variable. The set of three models for each outcome variable consisted of
models that (1) only considered students‘ background characteristics, (2) included
ninth grade academic performance, and (3) also included school characteristics,
particularly those focused on school climate.
Discussion of Results
The study revealed a number of variables with significant relationships to
high school outcomes. An early warning system for the ninth grade would
incorporate variables which are predictors of positive high school outcomes. Based
on indicators identified in this study, an early warning system would identify ninth
grade academic performance as key indicators. Additionally, student characteristics
associated with less positive outcomes can be used to identify particular student
populations for intervention and further research.
On-Track Indicators
The overall research question focused on identifying students‘ ninth grade
experiences that are indicators of being on-track for positive high school outcomes.
Given the available administrative data, the study examined students‘ ninth grade
transcripts and found that students‘ ninth grade academic performance was
significantly related to their outcomes. This finding was confirmed by reviewing the
strength of the models; the models‘ explanatory power increased when academic

92
performance was included in the model. Specifically, the number of failed core
subjects was an important indicator of whether students would graduate high school
on-time. Additionally, students‘ freshman year GPA was a robust predictor of high
school outcomes across all of the models which considered academic performance.
Higher GPA also contributed strongly to the likelihood of students graduating on-
time and enrolling in college, especially four-year colleges. A one point increase in
GPA increased the probability of a student graduating on-time or enrolling in a four-
year college by three times or more.
These findings about the importance of ninth grade academic performance
affirm prior research. A study of Los Angeles students by Silver et al. (2008) found
the explanatory power of the model of student outcomes increased from 17% to 29%
when adding high school academic experiences to students‘ background
characteristics. Allensworth and Easton (2007) found that freshman year grades
explained 39% of variance in Chicago‘s graduation rates. Likewise, Allensworth
(2006) found that a failing grade in any core subject in any semester of the freshman
year to be highly predictive of dropping out, and Neild et al. (2008) found that the
odds of dropping out of high school in Philadelphia to be directly related to ninth
grade course failures. Roderick et al. (2006) identified high school grade point
average ―perhaps the most important determinant of students‘ access to college and
likelihood of college graduation‖ (p. 87) in an extensive series of Chicago studies.
Thus, freshman year grades are important indicators of whether students are
on-track for positive high school outcomes. While failures in core subjects and GPA

93
are related to one another, both are statistically significant in most of the models
suggesting that failures in core subject courses have an especially strong impact on
outcomes. As with Chicago‘s off-track indicators (Allensworth, 2006), this study
suggests that a failing grade in any core subject during freshman year in Hawai‗i is
an early warning indicator for on-time graduation. Additionally, in this study, GPA
was an important differentiator of whether students graduated on-time and if they
enrolled in a four-year college. Thus, passing all courses in core subjects and having
a GPA of 2.00 or higher were strong indicators that students were on-track for
positive high school outcomes.
Students’ Background Characteristics
Students‘ background characteristics, such as being male, overage, lower
socioeconomic status and ethnicity, were often related to students‘ outcomes. Across
the analytical models, background characteristics of low-income and overage
emerged as factors associated with poorer outcomes consistently. Low-income
students were less likely to graduate on-time, and those who graduated were less
likely to enroll in college. Likewise, other researchers also found low-income
students less likely to graduate or enroll in postsecondary education (Bryk and
Thum, 1989; Rumberger, 1995; Hanson, 1994). Being overage emerged as a
consistent risk factor for graduating on-time or graduates enrolling in four-year
colleges compared to not attending college; being overage doubled the likelihood of
students not graduating on-time in each of the six models. This finding affirmed
Roderick‘s Fall River, Massachusetts study (1994) which identified being overage

94
upon high school entry as highly predictive of high school dropout, irrespective of
the reason that the student was overage (e.g., whether due to retention in prior
grades, late entry in school, or English Language Learners placed in a lower grade
than students their age upon school entry).
A number of other background characteristics were explored but not
identified as early warning indicators. Positive high school outcomes did not
consistently favor girls or boys, and prior academic achievement, as measured by
eighth grade test scores, had positive though small impact on high school outcomes.
The relationship of ethnicity and high school graduation was straightforward.
All ethnic groups were at least as likely as whites to graduate high school on-time.
However, when examining the relationship of ethnicity to college enrollment, a more
complex pattern emerged. Across the four categories of high school outcomes, no
ethnic group had a consistent relative position to whites as either advantaged or
disadvantaged. Asians had a greater likelihood of enrolling in college than whites,
and Pacific Islanders were less likely than whites to enroll. Likewise, Schneider et al.
(1998) found complex differences in college going patterns among different student
demographics. Heck and Mahoe (2006) posited that different patterns of
race/ethnicity and likelihood to graduate may be attributed to school organizational
factors, such as tracking of tracking and distribution of teacher quality among certain
student groups, rather than to ethnicity directly.
While high schools cannot affect students‘ family income and age are upon
entry, early warning systems sometimes incorporate these risk factors. For example,

95
Indiana‘s early warning system is based entirely on background characteristics: age,
retention patterns, middle school test scores, mobility and attendance (Kline,
personal communication, March 23, 2009). North Carolina includes low-income
status (Coble, personal communication March 23, 2009), and Louisiana includes age
(Pinkus, 2008). However, further research would be needed to understand that
interaction between background characteristics and other outcomes to identify early
warning indicators which may be supported by interventions.
Variance in Risk/Success Factors by High School Outcome
Research on high school dropout, graduation, and college access converge on
factors about students‘ background, academic experiences, and school as related to
high school outcomes (see Figure 2.1). However, most studies focus on their
dropout behavior, graduation or college enrollment while this study included
consideration of the continuum of outcomes. Models 1, 2 and 5 focused on the
outcome of on-time high school graduation, while Models 3, 4 and 6 also considered
enrollment in two- and four-year colleges as high school outcomes. Comparing
statistically significant variables across models as well as the variables‘ odds ratios
(see Table 4.8) shows that inclusion of college going outcomes changes the
equations associated with graduating on-time (vs. not graduating on-time). For
example, prior academic achievement was not significantly associated with
graduation in Models 2 and 5; however, when college enrollment was included in the
outcome variable in Models 4 and 6, both reading and mathematics tests scores were
positively associated with graduates‘ enrollment in college, either two or four year,

96
while reading scores were associated with a marginal increase in not graduating on-
time in the same models. Likewise, considering college enrollment in Models 3 and
4 changed the identified association between ethnic groups and outcomes. Whereas
in Model 2, all ethnic groups of color had a statistically significant advantage over
whites for graduating on-time, Models 3 and 4 showed a more nuanced set of
relationships with Asians much more likely than whites to enroll in college and
Pacific Islanders as the only group less likely than whites to enroll in college.
Furthermore, the equations for four-year college enrollment compared with
graduation but not attending college were different. Prior achievement and grade
point average had much larger odds ratios in equations for four-year outcomes than
equations for other outcomes. In Model 6, the four-year college equation was the
only one in which school-level variables were insignificant. These findings suggest
that identifying risk or success factors for college going requires more in-depth
analysis; factors identified in high school dropout studies or early warning systems
focused on high school graduation cannot be assumed to apply to college enrollment
behavior.
Freshman Year Academic Experiences
As described earlier, ninth grade academic performance, as measured by
GPA and failures in core classes, emerged as strong predictors of high school
outcomes. Many studies identified student attendance as a key predictor of
graduating on-time (NCES, 1988; Neild and Balfanz, 2006), and most school
systems‘ early warning systems including Chicago, Philadelphia, Portland and

97
Louisiana include attendance as an indicator. However, reliable attendance data for
students in the sample were not available systematically from the DOE. Thus,
academic performance as measured by GPA and performance in core subjects
reflected important freshman year academic experience from data readily accessible.
Future studies, perhaps at the school-level, should examine attendance to strengthen
models as attendance poses potential omitted variable bias.
School Climate
High schools mattered. The size of the school as well as the school climate
were related to outcomes, particularly whether students graduated on-time or
enrolled in a two-year college. Attending a smaller school increased the likelihood
of students graduating or attending two-year colleges affirming the Rumberger and
Thomas study (2000) which found higher dropout rates in schools with larger
enrollments. As with Lee‘s prior studies (Lee, 1999; Lee and Loeb, 2000),
Rumberger and Thomas‘ study did not explain whether school size itself is a factor
in the outcomes or if other factors associated with smaller schools such as school
organization, quality of teacher-student interactions, or ninth grade transition
programs were responsible for the benefit to students.
Students‘ perception of whether their school had a standards-based learning
environment was related to graduation outcomes. Students at schools with higher
positive student response on the School Quality Survey were more likely to graduate
on-time and to enroll in a two-year college. This finding about schools‘ academic
press affirms other research about the important mediating role of the high school in

98
affecting outcomes. Easton, Ponisciak and Luppescu (2008) found that the quality of
the high school, as measured by its academic culture is a major factor in determining
student achievement, beyond prior student achievement: ―There are many students
who start in the same place but end up very different from each other. It is students‘
school experiences that play such a strong role in determining academic
achievement‖ (p. 13).
Implications
The study identified important factors related to students‘ positive high
school outcomes and lead to recommendations for policy and research. The findings
point to on-track indicators to inform development of an early warning system to
identify freshmen for intervention to improve their likelihood of graduating and
enrolling in college. The study also revealed that the relationship of student
variables and outcomes was different when considering college enrollment. Also,
study demonstrated the power of longitudinal data in further explaining the
relationships between students‘ background characteristics, their academic
performance and the schools they attend.
Policy Implications
Many policy advocacy groups, such as the American Youth Policy Forum
(2008), Achieve (Jerald, 2006) and the National High School Center (Kennelly and
Monrad, 2007) have promoted development of early warning systems. This study
identified Hawaii-specific evidence-based on-track indicators which can be the basis
of an early warning system. Based on available administrative data analyzed for this

99
study, ninth grade GPA of at least 2.0 and passing all core subjects during freshman
year were identified as on-track indicators. Higher GPA increased the likelihood of
graduating on-time and enrolling in a four-year college significantly, and passing all
four core subjects ninth grade was critical to graduate on-time.
These findings suggest that schools need to track students‘ academic progress
closely during the transition into high school, particularly since research has shown
that it is difficult for students to recover from ninth grade failure (Roderick and
Camburn, 1999). Failing even one core subject during the transition year to high
school reduced a student‘s probability of graduating on-time by almost one-quarter.
DOE needs to develop information technology systems and tools that track students‘
progress, especially their grades, in real-time and that include tools which flag
students who fall off-track from positive high school outcomes. Once students are
identified as off-track, they need to be provided support to improve their
performance; schools need to develop and evaluate interventions which may include
academic tutoring, developing students‘ study skills, or counseling.
Different factors were significant or varied in magnitude in the relationships
between students and high school outcomes. Being on-track for four-year college
enrollment had different indicators than high school graduation. Ninth grade
performance mattered as well, but GPA was a significant differentiator between
students who graduated but did not choose college and students who enrolled in a
four-year college. A higher GPA increased the likelihood of a student enrolling in a
four-year college dramatically. Especially for students who aspire to four-year

100
colleges or who are interested in career paths which may require four-year college
degrees, school personnel need to be aware of students‘ ambitions early in order to
provide appropriate support. School personnel could identify and document such
student interests in the process of guiding students in writing their post-high school
plans as freshman, and school personnel should inform students of the importance of
their grades in increasing the likelihood of their on-time graduation and college
enrollment. Even for students who do not express interest in a four-year college,
ensuring that students remain on-track provides them with the most post-high school
options; overlooking students without college aspirations early in high school may
indeed close college options for students. School strategies and supports focused
solely on on-time high school graduation as the goal are insufficient to support
students‘ preparation for colleges.
Students‘ outcomes were not simply the result of their individual
characteristics and performance. At schools where more students reported a focus on
quality learning, students were more likely to graduate and to enroll in a two-year
college. Students were more likely to have graduated on-time if they were in schools
with characteristics associated with smaller schools, though the school‘s emphasis on
rigorous curriculum, at least as measured by the proportion of ninth graders enrolled
in algebra or higher, was not significant.
None of the school characteristics were significantly associated with students
who enrolled in four-year colleges. This suggests that four-year college going
behavior was driven by individual student characteristics, such as prior academic

101
achievement as well as other latent or unmeasured variables such as motivation.
This finding differs from Roderick et al. (2008) who found that the most consistent
predictor of college enrollment was a strong college climate. One explanation is that
individual characteristics and behavior may be driving the four-year college behavior
in the absence of school culture and supports promoting four-year college
attendance. Since school size and socioeconomic status were also not related to
four-year college matriculation, the findings suggest that students may be
individually selecting four-year colleges in spite of their schools. Another
explanation is that the measures used for these analyses are insufficient proxies for
school climate.
This study also identified student background characteristics related to high
school outcomes. Low-income status and being overage were consistently risk
factors for graduation and college enrollment. Some early warning systems
incorporate these demographics as risk factors. However, since these background
factors are not mutable by school officials; thus schools should develop or promote
programs which support students who have high risk profiles based on their
demographic characteristics. For example, low-income students are less likely to
achieve positive high school outcomes; therefore interventions such as federal
Upward Bound and GEAR UP programs may provide critical support for low-
income students‘ college preparation through facilitation of college planning
processes such as selecting colleges and applying for financial aid (Roderick et al.,
2008). Schools should incorporate these programs as part of a school and district

102
strategy to provide critical supports for students with high risk profiles based on
background characteristics rather than supplemental services for interested
individuals who receive individual benefits.
Among ethnic groups, Pacific Islander was the only group to have a lower
likelihood to enroll in college. Even when ninth grade performance was considered,
compared to equally qualified white students, Pacific Islanders were less likely to
attend college. Thus, academic supports alone are probably insufficient to achieve
ambitious Native Hawaiian college going targets of the Hawai‗i DOE (2010a) and
the University of Hawai‗i (UH Office of Vice President of Academic Planning and
Policy, 2009). More research is needed to understand why Pacific Islanders are at a
disadvantage for college enrollment even controlling for student background,
academic performance and school factors.
By providing more than descriptive statistics correlating one dimension of
student characteristics to high school outcomes, this study explored the relationship
between multiple characteristics of students—their demographics, their prior
achievement, their freshman year academic performance, and the school they
attended—and their outcomes. Since this study is correlational, not causal, it
identified variables with predictive power about the outcomes and can serve as the
basis for targeting students for interventions more efficiently, as recommended by
Gleason and Dynarski (2002).
However, the results may suffer from omitted variable bias since a limited set
of available administrative data was considered and the analysis was not an

103
experimental design. Omitted variable bias occurs when a variable not considered in
the analysis is related to both the outcome and explanatory variables. In this
analysis, for example, student attendance was not included since the reliable data
were not available from DOE‘s administrative datasets. Other locales have found
student attendance to be a predictor of high school graduation and student attendance
is likely to be correlated with students‘ grades which were found to be significant
predictors in this analysis (e.g., Neild and Balfanz, 2006; Allensworth and Easton,
2007). Thus, further data collection and analyses are necessary to identify causal
relationships between students‘ high school outcomes and their academic
experiences, background characteristics, and school attended. An initial step would
be to analyze the data at the school-level, rather than the state-level, since schools
collect student attendance data and attendance data are likely to be reliable at the
school-level.
The next step is for DOE staff to create a reporting tool through its
information technology system to provide schools and teachers with lists of current
high school underclassmen exhibiting early warning indicators. Then, schools need
to identify and apply interventions to students exhibiting risk factors to get them
back on-track for graduation and college. Also, schools need to build supports so
that students remain on-track. These supports include development of a college
going culture, monitoring of students‘ academic performance, and tutorial support to
avoid course failures.

104
Further development of a longitudinal student data system is needed in
Hawai‗i. The dataset for this analysis was created by DOE staff creating ad hoc data
reports from approximately 20 datasets. DOE staff developed the datasets for
analyses including pseudo-student identification numbers, and the researcher
consolidated the multiple datasets. A longitudinal data system which houses student
records and school data would facilitate further research on similar matters.
Furthermore, the dataset was limited by the focus on freshman year grades
and also by available data. Thus, important information such as students‘ special
needs status or student attendance was not analyzed. For example, further analyses
should examine students‘ entire high school record, including academic performance
in grades 10-12 to determine whether sequences or clusters of coursetaking lead to
different high school outcomes. A longitudinal data system which tracks students
over time will facilitate identification of more complete patterns of student progress
and outcomes. Such analyses will refine policy implications and programs.
Research Implications
As U.S. President Barack Obama further defines his education agenda to
focus on ―college- and career-ready students‖ and rates of college enrollment directly
from high school are publicized as part of the American Recovery and Reinvestment
Act (ARRA) State Fiscal Stabilization Funds reporting, more attention will shift
from high school completion to college matriculation (U.S. Department of
Education, 2010). Research on the differences in past graduates‘ patterns of high
school completion and college enrollment will inform development of strategic

105
supports for students to meet their personal educational goals; to create tools, for
schools and educators to use to monitor students‘ progress, to identify students for
interventions, and to motivate and build capacity of educators, schools and school
systems to support improved student outcomes.
First, further research is needed to better understand relationship between
students and a range of high school outcomes. Prior research analyzed dichotomous
outcomes, such as graduating high school vs. dropping out (e.g., Allensworth and
Easton, 2007). This study compared on-time graduation with graduation and college
going and found differences in relationship of student characteristics and behavior
with the range of positive high school outcomes. This study points to the need for
further analyses to develop a more complex understanding of high school outcomes.
Second, more complete data about students‘ K-12 records should be
analyzed. Most early warning systems include student attendance data; student
attendance is often the leading indicator of a student falling off-track (Neild and
Herzog, 2007; Allensworth and Easton, 2007). DOE historical records did not
include student-level attendance data. In the short term, further analyses may be
done at the school-level where historical attendance data are available, and in the
medium term, attendance data should be collected statewide. Also, more years of
students‘ longitudinal records should be analyzed. Some studies point to student
data at much earlier grades as predictive of later outcomes. These studies have
identified grades in the fourth grade (Roderick, 1994) or middle school course

106
failures, attendance and behavior (Neild, Balfanz and Herzog, 2007; Parthenon
Group, 2007) as predictors of high school success.
Third, more complete data about students‘ post-high school outcomes should
be analyzed. Just as the K-12 policy discussion is moving to ―college- and career-
ready,‖ the higher education discussion is moving to a focus on higher education
persistence and completion. For example, Obama set a goal that America will have
the highest proportion of college graduates in the world and proposed the American
Graduation Initiative focused on increasing community college graduation rates
(White House, 2009). However, this study was limited to students‘ enrollment in
college within one year of high school graduation. Further quantitative studies
should extend the research to include outcome data about students‘ progress in
higher education (e.g., credits earned) and college completion.
Fourth, multiple cohorts of Hawai‗i data should be analyzed. Replicating the
analyses using data for 2007, 2008 and/or 2009 graduating cohorts will provide a
sensitivity analysis of the models. The repeated analyses will indicate whether the
models are reliable; stability of models across the years which would increase
confidence about applying analyses of prior cohort data to future cohorts, especially
since the findings will be applied to classes at least five years after the study sample
entered high school. Also, sensitivity analyses of multiple cohorts of data allow for
assessment of validity of other locales‘ early warning indicators to Hawai‗i.
Fifth, qualitative research needs to be done to understand students‘
experiences in the process of selecting a college. Students‘ college matriculation is

107
mediated by the process of selecting, applying and enrolling in higher education. As
pointed out by Roderick et al. (2008), many factors contribute to the college
enrollment decision including applying for financial aid and advising by teachers and
counselors. Similarly, Hawai‗i would benefit from interviewing students about their
motivation, college aspirations, plans and outcomes as well as reviewing students‘
transcripts and tests scores; such a study could inform development of the student
supports at high schools and by higher education outreach programs to prepare and
recruit students for higher education.
Sixth, further research is needed regarding the high school outcome and
college going patterns of different ethnic groups in Hawai‗i. In these analyses of
Hawai‗i‘s diverse ethnic groups, whites were at a disadvantage for graduating high
school on-time compared with the other ethnic groups of color. Further analyses
would provide more information about the heterogeneity of student experiences
within designated ethnic groups.
Conclusion
Longitudinal data analyses of students‘ high school to college records
provides important information about students‘ patterns of success and failure. As
with other studies, this study found that academic factors are powerful predictors of
student outcomes. Students‘ successful academic transition to the ninth grade was
found to be critical to success and was often a better predictor than students‘
socioeconomic status, ethnicity or gender of whether a student graduated on-time or
enrolled in a four-year college. Also, the analyses found that students are much

108
more likely to graduate from schools where more students report a positive
standards-based learning environment.
Achieving local and national policy goals for educational attainment requires
that students not only graduate high school on-time, enroll and complete higher
education degree programs. This study analyzed the educational pipeline from ninth
grade through college for a cohort of Hawai‗i students and identified indicators
which predicted student success or risk. The study also identified different
predictors for two- and four-year college going behavior which indicate that focusing
on high school graduation alone will not achieve human capital policy goals. This
study provides the foundation for developing an early warning system for Hawai‗i.
Improving the productivity of the Hawai‗i‘s educational pipeline depends on the
success of using the output of the early warning system and implementing supports
to ensure that more students transition successfully to high school and onto positive
high school outcomes.

109
REFERENCES
Abelmann, C., Elmore, R., Even, J., Kenyon, S. and Marshall, J. (1999). When
accountability knocks, will anyone answer? (CPRE Research Report Series
RR-42). Philadelphia, PA: Consortium for Policy Research in Education.

Achieve (2009). Closing the expectations gap. Washington D.C.: Author.

ACT (2006). Ready for college and ready for work: Same or different? Iowa:
Author.

ACT (2008). The forgotten middle: Ensuring that all students are on target for
college and career readiness before high school. Iowa City, IA: Author.

Adelman, C. (2006). The toolbox revisited: Paths to degree completion from high
school through college. Washington, D.C.: U.S. Department of Education.

Alexander, K., Entwisle, D. and Kabbani, N. (2001). The dropout process in life
course perspective: Early risk factors at home and school. Teachers College
Record 103(5), 760-822.

Allensworth, E. (2006). Update to From high school to the future: A first look at
Chicago Public School graduates‘ college enrollment, college preparation
and graduation from four-year colleges. Chicago: Consortium on Chicago
School Research.

Allensworth, E. and Easton, J. (2005). The on-track indicator as a predictor of high
school graduation. Chicago: Consortium on Chicago School Research.

Allensworth, E. and Easton, J. (2007). What matters for staying on track and
graduating in Chicago Public High Schools: A close look at course grades,
failures and attendance in the freshman year. Chicago: Consortium on
Chicago School Research.

American Youth Policy Forum (2008). Improving the transition from middle grades
to high schools: The role of early warning indicators. (Forum brief).
Retrieved April 11, 2009 from
http://www.aypf.org/forumbriefs/2008.fb021508.htm.

Attewell, P. and Domina, T. (2008). Raising the bar: curricular intensity and
academic performance. Educational Evaluation and Policy Analysis 30(1),
pp. 51-71.

110
Bedworth, W., Colby, S. and Doctor, J. (2006). Reclaiming the American dream.
Boston, MA: Bridgespan Group.

Broughman, S.P., and Swaim, N.L. (2006). Characteristics of private schools in the
United States: Results from the 2003–2004 Private School Universe Survey
(NCES 2006-319). Washington, DC: U.S. Department of Education National
Center for Education Statistics.

Bryk, A.S, Raudenbush, S.W., & Congdon, R. (1996). HLM 4 for Windows
[Computer software]. Chicago, IL: Scientific Software International, Inc.

Bryk, A. S., & Thum, Y. M. (1989). The effects of high school organization on
dropping out: An exploratory investigation. American Educational Research
Journal, 26(3), 353-383.

Cabrera, A. (1994). Logistic regression analysis in higher education: an applied
perspective. In J. Smart (Ed.) Higher education: Handbook of theory and
research Volume 10 (pp. 225-256). New York: Agathon.

Carnevale, A. and Desrochers, D. (2002). The missing middle: Aligning education
and the knowledge economy. Princeton, NJ: Educational Testing Service.

Carnevale, Anthony P. and Desrochers, D. (2003) Standards for what? The economic
roots of K–16 reform. Princeton, NJ: Educational Testing Service.

Catterall, J. (1998). Risk and resilience in student transitions to high school.
American Journal of Education 106(2), 302-333.

Chicago Public Schools (September 6, 2008). CPS launches early warning tool to
support freshmen. Press release. Retrieved June 6, 2009 from
http://www.cps.edu/news/press-releases/pages/pressreleases.apx.

Cohen, N. (2000). Business location decision making and the cities: bringing
companies back. Washington, D.C.: The Brookings Institution Center on
Urban and Metropolitan Policy. Retrieved May 23, 2009 from
http://www.brookings.edu/~/media/Files/rc/reports/2000/05
metropolitanpolicy_cohen/cohen.pdf.

Cohort studies (2009). Retrieved June 4, 2009 from
http://www.socialresearchmethods.net/tutorial/Cho2/cohort.html.

111
College Board (2008). College bound seniors 2008. Retrieved May 23, 2009 from
http://professionals.collegeboard.com/data-reports-research/sat/cb-seniors-
2008.

Consortium on Chicago School Research (2009). 2007 Consortium survey
measures. Retrieved June 6, 2009 from
http://ccsr.uchicago.edu/surveymeasures2007/.

Data Quality Campaign (n.d.) A policymaker’s guide: Leveraging longitudinal
student data to develop college and career ready high school graduates.
Retrieved June 4, 2009 from
http://www.dataqualitycampaign.org/files/Policymakers_Guide.pdf.

DesJardins, S., McClall, B., Ahlburg, D. and Moye, M. (2002). Adding a timing
light to the ―Tool Box.‖ Research in Higher Education 43(1), 83-114.

Domina, T. (2009). What works in college outreach: Assessing targeted and
schoolwide interventions for disadvantaged students. Educational
Evaluation and Policy Analysis 31(2), 127-152.

Easton, J., Ponsciak, S. and Luppescu, S. (2008). From high school to the future:
Pathway to 20. Chicago: Consortium on Chicago School Research.

Education Trust (2009). College results on-line. Hawaii Pacific University
Retrieved May 3, 2009 from
http://www.collegeresults.org/search1a.aspx?institutionid=141644.

Education Week (2004). Quality Counts archive. Retrieved May 29, 2009 from
http://www.edweek.org/rc/articles/2004/10/15/qc-archive.html.

Education Week (January 2007). From cradle to career. Connecting American
education from birth through adulthood. Education Week. Retrieved 15
February 2009 from
http://www.edweek.org/media/ew/qc/2007/17shr.hi.h26.pdf.

Ewell, P., Jones, D., and Kelly, P. (2003). Conceptualizing and researching the
educational pipeline. Retrieved February 15, 2009 from
http://www.higheredinfo.org/analyses/Pipeline percent20Article.pdf.

Federal Reserve Bank of Cleveland (2005). Altered states: A perspective on 75 years
of state income growth. Cleveland, OH: Author.

112
Fredricks, J. Blumenfeld, P. and Paris, A. (2004). School engagement: potential of
the concept, state of the evidence. Review of Educational Research 74(1),
59-109.

Friedkin, N. and Thomas, S. (1997). Social positions in schooling. Sociology of
Education 70(4), 239-255.

Friedman, T. (2005). The world is flat: A brief history of the 21
st
century. New York:
Farrar, Straus, and Giroux.

Gandara, P. and Bial, D. (2001). Paving the way to postsecondary education: K-12
intervention programs for underrepresented youth (NCES 2001-205).
Washington, D.C.: U.S. Department of Education, National Center for
Education Statistics.

Garces, E., Thomas, D. and Currie, J. (2002). Long term effects of Head Start.
American Economic Review, 92(4), 999-1012.

Garson, G. (2010). Logistic regression. Statnotes: Topics in multivariate analysis.
Retrieved April 13, 2010 from
http://faculty.chass.ncsu.edu/garson/pa765/statnote.htm.

Ginsburg, A, Cooke, G., Leinwand, S., Noell, J . and Pollack, E. (2005).
Reassessing U.S. international mathematics performance: New findings from
the 2003 TIMSS and PISA. Washington, D.C.: American Institutes for
Research.

Gleason, P. and Dynarski, M. (1998). Do we know whom to serve? Issue in using
risk factors to identify dropouts. Princeton, NJ: Mathematica Policy
Research.

Gleason, P. and Dynarski, M. (2002). Do we know whom to serve? Issue in using
risk factors to identify dropouts. Journal of Education for Students Placed at
Risk, 7, 25-41.

Goldschmidt, P. and Wang, J. (1999). When can schools affect dropout behavior? A
longitudinal multilevel analysis. American Educational Research Journal
36(4), 715-738.

Gottlieb, P. D.and Fogarty, M. (2003). Educational attainment and metropolitan
growth. Economic Development Quarterly, 17(4), 325-336.

113
Greene, J. and Forster, G. (2003). Public high school graduation and college
readiness rates in the United States. New York: Manhattan Institute for
Policy Research. Retrieved June 4, 2009 from www.manhattan-
institute.org/html/ewp_03.htm.

Greene, J.P., & Forster, G. (2003). Public high school graduation and college
readiness rates in the United States. Education Working Paper, No. 3,
September 2003. Center for Civic Innovation at the Manhattan Institute.

Guthrie, J. and Wigfield, A. (2002). Engagement and motivation in reading, pp. 403-
422 in M. Kamil, M. and P. Mosenthal, (Eds). Handbook of Reading
Research Volume 3. Mahwah, NJ: Lawrence Earlbum.

Hammond, C., Linton, D., Smink, J. and Drew, S. (2007). Dropout risk factors and
exemplary programs: A technical report. Clemson, SC: National Dropout
Center Prevention Network.

Hanson, S. (1994). Lost talent: Unrealized educational aspirations and expectations
among U.S. youths. Sociology of Education 67(3), 159-183.

Hanushek, E. and Woessmann, L. (2008). The role of cognitive skills in economic
development. Journal of Economic Literature 46(3), 607-668.

Hawaii Economic Stimulus Oversight Commission (2009). Memorandum of
Agreement between the Governor of the State of Hawai‗i and the Department
of Education and the University of Hawai‗i to significantly improve student
achievement in Hawai‗i using American Recovery and Reinvestment Act
(―ARRA‖) resources to advance education reform. Retrieved April 22, 2010
from
http://capitol.hawaii.gov/session2010/otherreports/GOVResponses(Suppleme
ntal).pdf.

Hawaii P-20 (2010). Hawai‗i P-20 Partnerships for Education (Fact sheet).
Retrieved April 22, 2010 from
http://www.p20hawaii.org/sites/default/files/P-
20%20Fact%20Sheet%20draft%207.pdf.

Hawaii Workforce Informer (November 2007). Hawai‘i’s hot 50 jobs. Honolulu: HI:
State of Hawai‗i Department of Labor & Industrial Relations.

He, S., Yuan, S., Illukpitiya, P. and Yuen, S. (2007). Economic well-being in
Hawaii: Family and individual self-sufficiency – AUW Report. Honolulu, HI:
University of Hawaii at Manoa Center on the Family.

114
Heck, R. and Hallinger, P. (2009). Assessing the contribution of distributed
leadership to school improvement and growth in math achievement.
American Educational Research Journal 46(3), 659-689.

Heck, R. and Mahoe, R. (2006). Student transition to high school and persistence:
Highlighting the influences of social divisions and school contingencies.
American Journal of Education 112, 209-443.

Heck, R., Price, C. and Thomas, S. (2004). Tracks as emergent structures; A
network analysis of student differentiation in a high school. American
Journal of Education 110, 321 -353.

Heppen, J. and Therriault, S. (2008). Developing early warning systems to identify
potential high school dropouts. Washington, D.C.: National High School
Center. Retrieved June 4, 2009 from
http://www.betterhighschools.org/pubs/ews_guide.pdf.

Hoffman, L., and Sable, J. (2006). Public elementary and secondary students, staff,
schools, and school districts: School year 2003–04 (NCES 2006-307).
Washington, DC: U.S. Department of Education National Center for
Education Statistics. Retrieved June 5, 2010 from
http://www.ecs.org/html/offsite.asp?
document=http%3A%2F%2Fnces%2Eed%2Egov%2Fpubs2006%2F2006307
%2Epdf.

Holzer, H. and Lerman, R. (2009). The future of middle-skill jobs. Washington,
D.C.: Brookings Institution. Retrieved May 23, 2009 from
http://www.brookings.edu/~/media/Files/rc/papers/2009/02_middle_skill_job
s_holzer/02_middle_skill_jobs_holzer.pdf

Horn, L., Chen, X. and Chapman, C. (2003). Getting ready to pay for college: What
students and their parents know about the cost of college tuition and what
they are doing to find out. (NCES 2003-030). Washington, D.C.: U.S.
Department of Education National Center for Education Statistics.

Horn, L. and Kojaku, L. (2001). High school academic curriculum and the
persistence path through college (NCES 2001-163). Washington, D.C.: U.S.
Department of Education National Center for Education Statistics.

Hosmer, D. and Lemeshow, S. (2000). Applied logistic regression. 2
nd
edition.
USA: John Wiley & Sons, Inc.

115
Institute for Higher Education (2005). The investment payoff: A 50 state analysis of
public and private benefits of higher education. Washington, D.C.: Author.
Retrieved June 6, 2009 from http://www.ihep.org/Publications/publications-
detail.cfm?id=43.

Jerald, C. (2006). Identifying potential dropouts: Key lessons for building an early
warning data system. Washington, D.C.: Achieve.

Jerald, C. (2008). Stacking the deck: using early warning data to improve high
school outcomes. Presentation to the American Youth Policy Forum.

Johnsrud, L. (2006). University of Hawaii Second Decade project. Powerpoint
presentation. Retrieved June 3, 2009 from http://hawaii.edu/offices/app/.

Kauerz, K. (2005). Full day kindergarten: A study of state policies in the United
States. Denver, CO: Education Commission of the States.

Kennelly, L. and Monrad, M. (2007). Approaches to dropout prevention: Heeding
early warning signs with appropriate interventions. Washington, D.C.:
National High School Center.

Kirst, M. and Bracco, K. (2004). Bridging the great divide: How the K-12 and
postsecondary split hurts students and what can be done about it. In M. Kirst
and A. Venezia (Eds). From high school to college: Improving opportunities
for success in postsecondary education (pp. 1-30). San Francisco: Jossey-
Bass.

Kirst, M. and Venezia, A. (2004). From high school to college: Improving
opportunities for success in postsecondary education. San Francisco: Jossey-
Bass.

Lee, V. (1999). School size and organization of secondary schools, In Hallinan, M.
(Ed). Handbook of the sociology of education (pp. 327-344). New York:
Kluwer Academic/Plenum Publishers.

Lee, V. and Burkam, D. (2003). Dropping out of high school: The role of school
organization and structure. American Educational Research Journal 40(2),
353-393.

Lee, V. and Loeb, S. (2000). School size in Chicago elementary schools: Effects on
teachers‘ attitudes and students‘ achievement. American Educational
Research Journal 37(1), 3-31.

116
Lee, V. and Smith, J. (1996). Collective responsibility for learning and its effects on
gains in achievement for early secondary school students. American Journal
of Education 104(2), 103-147.

Lee, V., Smith, J., Perry, T. and Smylie, M. (1999). Social support, academic press
and student achievement: A view from the middle grades in Chicago.
Chicago: Consortium on Chicago School Research.

Long, B. (2004). The role of perceptions and information in college access: An
exploratory review of the literature and possible data sources. Boston, MA:
The Education Resources Institute. Retrieved June 6, 2009 from
http://teri.org/pdf/research-studies/Research Report_Long.pdf.

McEwan, E. and McEwan, P. (2003). Making sense of research: What’s good,
what’s not, and how to tell the difference. Thousand Oaks, CA: Corwin Press.

McNeal, R. (1997). Are students being pulled out of high school? The effect of
adolescent employment on dropping out. Sociology of Education 70(3), 206-
220.

Montague, W. (September 30, 1987). School rated a key factor in business-site
decisions. Education Week, 1, 14.

National Center for Education Statistics (2008). Table 2. Public school number of
graduates and averaged freshman graduation rate, by race/ethnicity and state
or jurisdiction: School year 2005-06. Common Core of Data. Retrieved May
16, 2009 from http://nces.ed.gov/ccd/tables/2008353_02.asp.

National Center for Education Statistics (2009). Integrated Postsecondary
Education Data System. Retrieved June 4, 2009 from
http://nces.ed.gov/IPEDS/about/.

National Center for Public Policy and Higher Education (2004). The educational
pipeline: Big investment, big return. Retrieved February 16, 2009 from
http://www.highereducation.org/reports/pipeline/

National Center for Public Policy and Higher Education (2006). Educational
pipeline FAQ. Retrieved May 23, 2009 from
http://www.highereducation.org/reports/pipeline/faq.shtml

117
National Commission on Excellence in Education (1983). A nation at risk:
Imperative for educational reform. Washington, D.C.: U.S. Department of
Education. Retrieved May 16, 2009 from
http://www.ed.gov/pubs/NatAtRisk/title.html.

National Council on Educational Statistics (2009). National Assessment of
Educational Progress state comparisons. Washington, D.C.: U.S.
Department of Education. Retrieved May 23, 2009 from
http://nces.ed.gov/nationsreportcard/nde/statecomp/sorting.asp

National Council on Educational Statistics (2009a). The Condition of education
2010. Washington, D.C.: U.S. Department of Education. Retrieved June 12,
2010 from http://nces.ed.gov/programs/coe/2010/section2/table-mgp-3.asp

National Council on Higher Education Management (2009). Student pipeline
transition and completion rates from ninth grade to college. Retrieved May
23, 2009 from
http://www.higheredinfo.org/dbrowser/index.php?submeasure=119&year=20
06&level=nation&mode=data&state=0.

National Governors Association, Council of Chief State School Officers and Achieve
(2008). Benchmarking for success: ensuring U.S. students receive a world
class education. Washington, D.C.: National Governors Association.

National Student Clearinghouse (2009). Measure the college success of your high
school graduates. Retrieved May 3, 2009 from
http://www.studentclearinghouse.org/highschools/default.htm

Neild, R. and Balfanz, R. (2001). An extreme of difficulty: The demographics of
ninth grade in non-selective high schools in Philadelphia. Paper presented at
the Annual Meetings of the American Sociological Association, Anaheim,
CA.

Neild, R. and Balfanz, R. (2006). Unfulfilled promise: The dimensions and
characteristics of Philadelphia’s dropout crisis, 2000-05. Philadelphia:
Philadelphia Youth Transitions Collaborative.

Neild, R. and Herzog, L. (2007). Research and interventions. Presentation at 4
th

annual Middle Grades Matter conference. Retrieved June 4, 2009 from
http://www.philaedfund.org/pels/pels_pubs.html.

Neild, R., Balfanz, R. and Herzog, L. (2007). An early warning system.
Educational Leadership 65(2), 28-33.

118
Neild, R., Stoner-Eby, S., and Furstenberg, F. (2008). Connecting entrance and
departure: The transition to ninth grade and high school dropout. Education
and Urban Society 40(5), 543-569.

Newmann, F. (1992) Student engagement and achievement in American secondary
schools. New York: Teachers College Press.

Orfield, G. and Paul, F. (1993). High hopes and long odds: A report on Hoosier
teens and the American dream. Indianapolis, IN: Indiana Youth Institute.

Organisation for Economic Co-operation and Development (2006). Programme for
International Student Assessment (PISA) 2006: Science competencies for
tomorrow’s world, Vol I: Analysis. Paris: Author.

Organisation for Economic Cooperation and Development (2007). PISA 2006:
Science competencies for tomorrow’s world: OECD briefing note for the
United States. Paris, France: Author.

Organisation for Economic Co-operation and Development. (2008). Education at a
glance 2008. Paris: Author. Retrieved May 23, 2009 from
http://www.oecd.org/dataoecd/23/46/41284038.pdf

Organisation for Economic Co-operation and Development. (2009). Glossary of
statistical terms: Educational attainment. Retrieved May 31, 2009 from
http://stats.oecd.org/glossary/detail.asp?ID=742.

Pallas, A. and Anderson, K. (1983). Sex differences in quantitative SAT
performance: New evidence on the differential coursework hypothesis.
American Educational Research Journal 20(2), 164-182.

Parthenon Group (2007). Strategic planning to serve off-track youth: Data review
and strategic implications. Boston: Author.

Patton, M. (2008). HI-PASS report on data quality. (Memo).

Perna, T. and Titus, M. (2002). The role of parental involvement as social capital in
promoting the college enrollment of African Americans and Hispanics:
Implications for college preparation programs. Paper presented at the
Annual Meeting of the Association for the Study of Higher Education,
Sacramento, CA.

119
Pinkus, L. (2008). Using early-warning data to improve graduation rates: Closing
cracks in the education system. Washington, D.C.: Alliance for Excellent
Education.

Pinkus, L. (2009). Moving beyond AYP: High school performance indicators.
Washington, D.C.: Alliance for Excellent Education.

Plank, S. and Jordan, W. (2001). Effects of information, guidance and actions on
postsecondary destinations: A study of talent loss. American Educational
Research Journal 38(4), 947-979.

Powers, D. and Zie, Y. (2008). Statistical methods for categorical data analysis,
second edition. Bingley, UK: Emerald Group Publishing Limited.

Railsback, J. (2004). Increasing student attendance: Strategies from research and
practice. Portland, OR: Northwest Regional Educational Laboratory.

Roderick, M. (1993). The path to dropping out: evidence for intervention. Westport,
CT: Auburn House.

Roderick, M. (1994). Grade retention and school dropout: Investigating the
association. American Educational Research Journal 31(4), 729-759.

Roderick, M. and Camburn, E. (1999). Risk and recovery from course failure in the
early years of high school. American Educational Research Journal. 36(2),
303-343.

Roderick, M. and Engel, M. (2001). The grasshopper and the ant: Motivational
responses of low performing students to high stakes testing. Education
Evaluation and Policy Analysis 23(3), 197-227.

Roderick, M., Nagaoka, J. and Allensworth, E. with Coca, V., Correa, M. and
Stoker, G. (2006) From high school to the future: A first look at Chicago
Public School graduates’ college enrollment, college preparation and
graduation from four-year colleges. Chicago: Consortium for Chicago
School Research.

Roderick, M., Nagaoka, J., Coca, V. and Moeller, E. (2009). From high school to
the future: Making hard work pay off. Chicago: Consortium on Chicago
School Research.

120
Roderick, M., Nagaoka, J., Coca, V. and Moeller, E. with Roddie, K., Gilliam, J. and
Patton, D. (2008). From high school to the future: Potholes on the road to
college. Chicago: Consortium on Chicago School Research.

Rose, H. and Betts, J. (2001) Math matters: Links between high school curriculum,
college graduation and earnings. San Francisco, CA: Public Policy Institute
of California.

Rumberger, R. (1995). Dropping out of middle school: A multilevel analysis of
students and schools. American Educational Research Journal 32(3), 583-
625.

Rumberger, R., and Larson, K. (1998). Student mobility and the increased risk of
high school dropout. American Journal of Education, 107(1), 1-35.

Rumberger, R., and Thomas, S. (2000). The distribution of dropout and turnover
rates among urban and suburban high schools. Sociology of Education, 73(1),
39-67.

Schneider, B. Swanson C. and Riegle-Crumb, C. (1998). Opportunities for learning:
Course sequences and positional advantages. Sociology of Education 2, 25-
53.

School District of Philadelphia (2009). Imagine 2014: Building a system of great
schools. Philadelphia: Author. Retrieved June 6, 2009 from
http://webgui.phila.k12.pa.us/offices/s/strategic-planning.

Schwab (n.d.). Multinomial logistic regression basic relationships. [Powerpoint
presentation for University of Texas SW388R7 Data Analysis and Computers
II]. Retreived May 3, 2009 from
www.utexas.edu/courses/schwab/sw388r7/SolvingProblems/
MultinomialLogisticRegression_BasicRelationships.pp.

Scientific Software International (2010). HLM – hierarchal linear and nonlinear
modeling (HLM). Accessed April 17, 2010 from http://ssicentral.com/hlm/.

Silver, D., Saunders, M., and Zarate, E. (2008). What factors predict high school
graduation in Los Angeles Unified School District. Santa Barbara, CA:
California Dropout Research Project.

Smith, J. (1996). Does an extra year make any difference? The impact of early
access to algebra on long-term gains in mathematics achievement.
Educational Evaluation and Policy Analysis 18(2), 141-153.

121
SPSS (2009). Multinomial logistic regression. SPSS tutorial. Retrieved May 3,
2009 from
file:///Applications/SPSSInc/Statistics17/SPSSStatistics17.0.app/Contents/bin
/lang/en/tutorial/case_studies/casestudies_intro.htm

State of Hawaii Board of Education (2010). Board policy 4540, Graduation and
related policy. Retrieved April 17, 2010 from
http://lilinote.k12.hi.us/STATE/BOE/POL1.NSF/85255a0a0010ae828525553
40060479d/91ae48edb6f3e0140a2566a3006f0dec?OpenDocument.

State of Hawaii Department of Education (2001). School Quality Survey dimension
descriptions. Honolulu, HI: State of Hawaii Department of Education
Planning, Budget and Resource Development Office. Retrieved May 24,
2009 from http://arch.k12.hi.us/PDFs/sqs/DimDesc.pdf.

State of Hawaii Department of Education (2007). Department of Education Strategic
Plan, July 1, 2008-June 30, 2011. Honolulu, HI: State of Hawaii Department
of Education Office of the Superintendent. Retrieved April 13, 2010 from
http://doe.k12.hi.us/curriculum/strategicplan.pdf.

State of Hawaii Department of Education (2008). Enabling effective longitudinal
data usage in Hawaii. Application to U.S. Department of Education
Statewide Longitudinal Data Systems Grant Program CFDA 84.732.

State of Hawaii Department of Education (2009). Department of Education trend
report: Educational and fiscal accountability. Honolulu, HI: Author.
Retrieved May 24, 2009 from
http://arch.k12.hi.us/PDFs/trends/2008/Act51_State.pdf.

State of Hawaii Department of Education (2009a). Introduction, enrollment.
Honolulu, HI: Author. Retrieved June 6, 2009 from
http://doe.k12.hi.us/about/intro_enrollment.htm.

State of Hawaii Department of Education (2010). School Quality Survey item
correspondence tables, 2001-07. Honolulu, HI: Author. Retrieved April 21,
2010 from
http://arch.k12.hi.us/PDFs/sqs/SQS%20Item%20Correspondence%20Tables2
001-07.PDF.

State of Hawai‗i Department of Education (2010a). Hawai‘i’s Race to the Top
summary, January 2010 application. Honolulu, HI: Author. Retrieved April
21, 2010 from http://doe.k12.hi.us/arra/RaceToTheTop.pdf.

122
State of Hawaii Department of Labor and Industrial Relations Research and Statistics
Office (2008). Hawaii hot 50. Honolulu, HI: Author. Retrieved May 23,
2009 from
http://www.hiwi.org/admin/uploadedPublications/1486_Hot_50_4pgs_2008_
Revised.pdf

Stevenson, D., Schiller K. and Schneider, B. (1994). Sequences of opportunities for
learning. Sociology of Education 67, 184-198.

Stid, D., O‘Neill, K. and Colby, S. (2009). Portland Public Schools: From data to
decisions to implementation and results on dropout prevention. Boston:
Bridgespan Group.

Stillwell, R. and Hoffman, L. (2008). Public school graduates and dropouts from the
Common Core of Data: School year 2005–06 (NCES 2008-353rev).
Washington, D.C.: National Center for Education Statistics, Institute of
Education Sciences, U.S. Department of Education. Retrieved April 27, 2010
from http://nces.ed.gov/pubsearch/pubsinfo.asp?pubid=2008353rev.

Swanson, C. (2003). NCLB implementation report: State approaches for calculating
high school graduation rates. Washington, D.C.: The Urban Institute.

Swanson, C. (2004). Who graduates? Who doesn’t? A statistical portrait of public
high school graduation, Class of 2001. Washington, D.C.: Urban Institute.

Tabachnick, B. and Fidel, L. (1996). Using multivariate statistics Third Edition.
New York: Harper Collins College Publishers.

U.S. Bureau of Labor Statistics (2009). Bureau of Labor Statistics Information
Glossary. Retrieved May 31, 2009 from
http://stats.bls.gov/bls/glossary.htm#L

U.S. Census (1990). American Fact Finder Detailed Tables. Retrieved May 3, 2009
from http://factfinder.census.gov/servlet/DTTable?_bm=y&-context=dt&-
ds_name=DEC_1990_STF1_&-mt_name=DEC_1990_STF1_P006&-
mt_name=DEC_1990_STF1_P007&-mt_name=DEC_1990_STF1_P008&-
mt_name=DEC_1990_STF1_P009&-mt_name=DEC_1990_STF1_P010&-
CONTEXT=dt&-tree_id=100&-all_geo_types=N&-geo_id=04000US15&-
search_results=01000US&-format=&-_lang=en

123
U.S. Census Bureau (2000). Profile of general demographic characteristics: 2000,
Geographic area: Hawaii (DP-1). Retrieved May 3, 2009 from
http://factfinder.census.gov/servlet/QTTable?_bm=y&-geo_id=04000US1r&-
qr_name=DEC_2000_SF1_U_DP1&-ds_name=DEC_2000_SF1_U

U.S. Census Bureau (2009). A half-century of learning: historical statistics on
educational attainment in the United States, 1940-2000. (PHC-T-41).
Accessed May 23, 2009 from
http://www.census.gov/population/www/socdemo/education/phct41.html.

U.S. Census Bureau (2009a). Educational attainment in the United States 2007.
(P20-560). Retrieved May 23, 2009 from
http://www.census.gov/prod/2009pubs/p20-560.pdf.

U.S. Department of Education (2010). Blueprint for reform. The reauthorization of
the Elementary and Secondary Education Act. Washington, D.C.: Author.
Retrieved April 21, 2010 from
http://www2.ed.gov/policy/elsec/leg/blueprint/blueprint.pdf.

University of California Los Angeles Academic Technology Services Statistical
Consulting Group (2010). FAQ: What are pseudo R-squareds? Retrieved
March 27, 2010 from
http://www.ats.ucla.edu/stat/mult_pkg/faq/genearl_Pseudo_RSquareds.htm.

University of Hawaii Office of the President (2010). The University of Hawai‘i for
the 21
st
century: Putting the power of higher education to work for the State
of Hawai‘i. [Transcript of speech to Legislature] Retrieved April 22, 2010
from http://hawaii.edu/offices/op/2010/0203-address.html.

University of Hawaii Office of Vice President of Academic Planning and Policy
(2009). University of Hawaii System Strategic outcomes and performance
measures, 2008-2015. Retrieved June 3, 2009 from
http://www.hawaii.edu/ovppp/uhplan/.

Venezia, A., Kirst, M. and Antonio, A. (2003). Betraying the college dream: how
disconnected K-12 and postsecondary education systems undermine student
aspirations. Palo Alto: Stanford University Bridge Project.

Wagner, T. (2008). The global achievement gap: Why even our best schools don’t
teach the new survival skills our children need and what we can do about it.
New York: Basic Books.

124
Watt, K., Powell, C., Mendiola, I. (2004). Implications of one comprehensive school
reform model for secondary school students underrepresented in higher
education. Journal of education for students placed at risk 9(3), 241-259.

White House (July 14, 2009). American Graduation Initiative fact sheet. [Press
release]. Retrieved April 22, 2010 from
http://www.whitehouse.gov/the_press_office/Excerpts-of-the-Presidents-
remarks-in-Warren-Michigan-and-fact-sheet-on-the-American-Graduation-
Initiative/.

125
APPENDIX
DESCRIPTIONS OF MODELS ANALYZED
As described in Chapter 3, logistic regressions were used to analyze the data
to identify the relationship between students‘ high school outcomes and independent
variables of students‘ characteristics and experiences. As summarized in Table A1,
the analyses consisted of six models exploring data to identify leading indicators
during students‘ high school freshman year of graduating on-time and enrolling in
college. This appendix presents the results of the models.

Table A1: Summary of Regression Models

Stage Model
Regression
Type Dependent Variable Independent Variables

On-time
high
school
graduation
High school
outcomes
including
college
enrollment
Students‘
background
characteristics
9
th
grade
academic
experiences
School
climate
1 1
Binomial
logistic
X X
a

2
Binomial
logistic
X X
a
X
2 3
Multinomial
logistic
X X
a

4
Multinomial
logistic
X X
a
X
3 5
HLM
Binomial
logistic
X Level 1 Level 1
Level
2
6
HLM
multinomial
logistic
X Level 1 Level 1
Level
2
a
Included ethnicity

126
The results of the logistic regressions include coefficients which are
interpreted more easily as as odds ratios. Odds ratios are calculated using the
formula e
b
where e=2.71728 where b is the log-odds coefficient. Odds ratios are
interpreted as increased odds of the dependent variable occurring for each unit
increase in the independent variable. Positive log-odds coefficients result in odds
ratios greater than one and suggest an increased likelihood of the positive outcome,
on-time graduation in this case. Negative log-odds coefficients, on the other hand,
result in odds ratios less than one and suggest an increased likelihood of the negative
outcome, not completing high school on-time. Negative log-odds coefficients are
most simply interpreted as the reciprocal of the odds ratio associated with the
negative outcome (Powers and Xie, 2008). For example, one unit increase in an
independent variable with a negative coefficient leads to an increase in the negative
outcome by the reciprocal value of the odds ratio.
Stage 1: On-Time High School Graduation Models
The following models use binary logistic regression to measure the
association of students‘ background characteristics and academic performance with
whether students graduated high school on-time or not (see Table A2). Table 4.8
reports coefficients, significance and odds ratios, observations and pseudo R-squares
for these models.

127
Table A2: Binary Logistic Models Predicting On-Time High School Completion,
Models 1 and 2

Model 1 Model 2
Characteristic Estimate Odds Ratio Estimate Odds Ratio
Background Characteristics
Low-Income -.466**** .627 (1.595) -.420**** .657 (1.5221)
Male -.133** .876 (1.142) .109 1.115
Overage -.849**** .428 (2.336) -.872**** .418 (2.392)
Prior Achievement
Reading, 2002
(Scaled Score)
.002*** 1.002 -.001 .989
Mathematics, 2002
(Scaled Score)
.006*** 1.006 .001 1.001
Ethnic Group
Asian .499**** 1.678 .541*** 1.718
Black/Hispanic .375* 1.455 .614*** 1.847
Filipino .652**** 1.919 .686**** 1.987
Other .075 1.078 .218* 1.244
Pacific Islander .065 1.068 .399**** 1.490
White --
a
--
a

Ninth Grade Academic
Performance

Grade Point Average
(Year-End)
1.307**** 3.694
Core Subjects Failed -.208**** .812 (1.232)
Constant -.158 .854 -.874** .417
Number of Observations 9164 9164
Pseudo R
2
(Nagelkerke) .070 .220
Note: Reciprocal odds ratios reported in parentheses for significant, negative coefficients.
a
Referent

group is whites
*p<.10 **p<.05 ***p<.01, ****p<.001

128
Model 1: Students’ Background Characteristics
A model examining the relationship between students‘ background
characteristics and graduating high school on-time revealed significant associations.
Being low-income, male, and overage for freshman year increased the odds of not
graduating high school on-time. Conversely, higher prior student achievement as
measured by eighth grade test scores and not being white (ethic group) increased the
odds of students graduating on-time.
In the models examining the relationship between students‘ background
characteristics and graduating high school on-time, identified risk factors for not
graduating high school which were consistent with the literature. Fewer students
who were low-income, male, or older than the cohort‘s expected age graduated on-
time (see Table A2). The greatest effect came from being overage. Overage students
were more than two times as likely to not graduate on time. Being low-income
increased odds of not graduating by 1.60 times, and being male increased odds of not
graduating by 1.14 times.
On the other hand, students who were not white (ethnic group) were more
likely to graduate on-time. Every group had a positive log-odds coefficient although
the coefficient was significant for only 3 of 5 groups. Asians and Filipinos were
more likely to graduate by 1.68 and 1.92 times, respectively; these results were very
significant (p<.0001). Though the significance was not as strong for Blacks and
Hispanics, these traditionally lower performing students were 1.46 times more likely
to graduate than students who identified ethnically as white (including Portuguese).

129
Students‘ prior academic achievement was also associated with graduating
on-time. For each point increase in students‘ Hawai‗i State Assessment scores, the
likelihood of graduating increased slightly: 1.002 times for each point in reading and
1.006 for each point in mathematics. Since the standard deviation of the reading
score is 61 points, being one standard deviation above the mean score increased the
likelihood of graduating by 1.15 times. The standard deviation of the mathematics
score is 64 points, and students who score one standard deviation above the mean
score are 1.47 times more likely to graduate on-time.
Model 2: Students’ Background Characteristics and Ninth Grade Academic
Performance
In Model 2 (see Table A2), students‘ ninth grade academic performance was
considered in addition to their background characteristics. In this model, background
characteristics of being low-income and overage were associated significantly with
increased likelihood of not graduating on-time. Overage students were 2.39 times
more likely to not graduate on-time, and low-income students were 1.52 times more
likely not to graduate on time. All ethnic groups of color, in comparison with
whites, were more likely to graduate on-time. While the significance and likelihood
levels varied, not identifying as white increased students‘ likelihood of graduating at
a statistically significant level. The association of being of an ethnic group of color
and graduating on-time ranged from increased odds of 1.24 times for ―other‖
students to 1.85 times for Filipino students when other background characteristics

130
and ninth grade academic performance are considered. The log-odds coefficients for
gender and prior academic achievement were not significant.
Different combinations of students‘ freshman year academic performance,
described in Tables 4.3 and 4.4, were considered in the models. Considering year-
end performance resulted in higher pseudo R-squared as well as higher odds ratios
than considering first semester performance, so year-end grades and course failures
were used to represent ninth grade academic performance. Also, analyzing
individual course grades or failures in core subjects identified some significant
variables in either group but no more than two of four variables were significant.
Ninth grade retention was entered as a variable as an alternate measure of academic
performance; while students‘ status as repeating or promoted ninth grade in the
2003-04 school year was associated significantly with high school outcomes, the
pseudo R-squared of grades were higher than retention status, and furthermore, since
retention represented the consequence of multiple negative performance indicators in
the freshman year, students‘ grades rather than retention status provided more
practical opportunities for intervention.
This model showed that students‘ ninth grade academic performance was
associated significantly with graduating on-time. Each unit increase in grade point
average increased the likelihood of graduating on-time by 3.69 times. For each core
subject (i.e., English, mathematics, social studies, science) failed, students increased
their likelihood of not graduating on-time by 1.23 times. Once ninth grade

131
performance was considered, prior academic achievement was not significantly
related to graduating on-time.
Stage 2: High School Outcomes Models
The following models measure the association of students‘ background
characteristics and academic performance with students‘ high school outcomes:
completion of high school on-time and post-high college enrollment outcomes.
Multinomial logistic regression was used since there are four categories for the
dependent variable, high school outcomes: did not graduate on-time, graduated on-
time but did not enroll in college, graduated on-time and enrolled in two-year
college, and graduated on-time and enrolled in four-year college.
Model 3: Students’ Background Characteristics
The relationship of students‘ background characteristics with outcomes
varied depending on the outcome. In Model 3 which considered only students‘
background characteristics, the reference group was students who graduated but did
not enroll in college in the year after high school (see Table A3).

132

Table A3: High School Outcomes and Students‘ Background Characteristics,
Model 3

Graduated High School On-Time
Did Not Graduate High
School On-Time
Enrolled in Two Year
College
Enrolled in Four Year
College
Estimate Odds Ratio Estimate Odds Ratio Estimate Odds Ratio
Background
Characteristics

Low-Income .230
***
1.259 -.495
****
.610 (1.639) -.613
****
.542 (1.845)
Male -.051 .985 (1.015) -.241
****
.786 (1.272) -.520
****
.595 (1.681)
Overage .776
****
2.173 -.122 .885 (1.130) -.322
**
.725 (1.379)
Prior Achievement
Reading, 2002
(Scaled Score)
.000 1.000 .002
***
1.002 .012
****
1.012
Mathematics,
2002 (Scaled
Score)
-.003
****
.997 (1.003) .003
****
1.003 .015
****
1.015
Ethnicity
Asian .097 1.102 .822
****
2.276 1.060
****
2.886
Black/ Hispanic -.356
*
.701 (1.423) .079 1.079 -.129 .879
Filipino -.565
****
.568 (1.761) .301
***
1.352 .011 1.011
Other .003 1.003 .098 1.103 .326
**
1.386
Pacific Islander -.183
*
.833 (1.200) -.215
**
.807 (1.239) -.703
****
.495 (2.020)
White --
a
--
a
--
a

Constant .844
****

-
2.152
****

-
9.315
****

Number of
Observations

Pseudo R
2

(Nagelkerke)
.343
Note: Reciprocal odds ratios reported in parentheses for significant, negative coefficients.
a
Referent

group is whites
*p<.10 **p<.05 ***p<.01, ****p<.001

133
Did not graduate high school on-time. The largest impact on graduating high
school on-time was students‘ age; students who are overage were 2.17 times more
likely to not graduate on-time. However income also mattered. Not being low -
income increased students likelihood of graduating on-time by 1.26 times. Prior
achievement has a small effect with each additional point in a student‘s eighth grade
mathematics test score increasing students‘ odds of graduating on-time by 1.003; one
standard deviation increase in mathematics test scores increases students‘ odds of
graduating by 1.21 times. Likelihood of graduating on-time differed for some ethnic
groups. Black/Hispanic, Filipino and Pacific Island students were more likely than
whites to graduate high school on-time by 1.42, 1.76 and 1.20 times, respectively.
Graduated high school on-time and enrolled in two-year college. Students
who graduated on-time and enrolled in two year colleges had different background
characteristics from students who graduated on-time but did not enroll in college.
Being female and higher income (not-low income) increased students‘ likelihood of
enrolling in a two-year college by 1.27 and 1.64 times, respectively. Higher prior
academic achievement increases the likelihood of attending a two-year college
marginally; scoring one standard deviation higher on the eighth grade state
assessment increased the likelihood of attending a two-year college by 1.15 times for
reading and 1.21 times for mathematics. Asians were twice as likely as whites to
enroll in two- year colleges, and Filipinos were 1.35 times more likely to enroll in a
two-year college than whites. However, white were 1.24 times more likely than
Pacific Islanders to enroll in a two-year college enrollment.

134
Graduated high school on-time and enrolled in four-year college. Most
background characteristics had a significant relationship with students enrolling in a
four-year college vs. not enrolling after graduating on-time. Being female, expected
age for high school freshmen, and higher income increased students‘ odds of
enrolling in a four-year college by 1.68, 1.38 and 1.85 times, respectively. Higher
prior academic achievement is more strongly associated with four-year college
enrollment; students scoring one standard deviation above the mean on the eighth
grade tests increased the likelihood of enrolling in a four-year college by 2.25 for
reading and 2.61 for mathematics compared with graduating but not choosing
college immediately. The ethnic groups of Asians and Others were more likely than
whites to enroll in a four-year college, especially Asians who were 2.89 times more
likely to attend. However, being a Pacific Islander made a student twice as likely to
not attend college compared with whites.
Model 4: Students’ Background Characteristics and Ninth Grade Academic
Performance
Students‘ ninth grade academic performance was associated with their high
school outcomes. Most background characteristics identified as significant in Model
3, which only considered high school graduation, continued to be significant, and
with a few exceptions, the magnitude of the odds ratios were similar (see Table A4).
Did not graduate high school on-time. Not being low-income and being the
expected age of high school freshmen increased the likelihood of graduating on-time
by 1.24 and 2.23 times, respectively. Higher prior mathematics achievement, as

135
measured by the eighth grade mathematics test, marginally improved students‘
likelihood of graduating on-time. Being from any ethnic group of color increased
the likelihood of graduating on-time vs. not graduating when compared with whites.
Black/Hispanics, Filipinos, and Pacific Islanders had a higher likelihood of
graduating on time: 1.59 times for Pacific Islanders and 1.83 times for Filipinos.
Ninth grade academic performance affected graduation outcomes. Each unit
increase in grade point average increased the likelihood of graduating on-time by
almost three times, while each core subject failed increased likelihood of not
graduating on-time by 1.25 times.

136
Table A4: Multinomial Logistic Regression of Students‘ Background Characteristics
and Ninth Grade Academic Performance on High School Outcomes, Model 4

Graduated High School On-Time
Did Not Graduate High
School On-Time
Enrolled in Two
Year College
Enrolled in Four
Year College
Estimate Odds Ratio Estimate Odds Ratio Estimate Odds Ratio
Background
Characteristics

Low-Income .217
***
1.243 -.495
****
.610 (1.639) -.632
****
.532 (1.880)
Male -.192
**
.825 (1.212) -.197
***
.821 (1.218) -.147
*
.864 (1.157)
Overage .800
****
2.225 -.147
*
.864 (1.157) -.464
***
.629 (1.590)
Prior Achievement
Reading, 2002
(Scaled Score)
.002
*
1.002 .002
**
1.002 .008
****
1.008
Mathematics, 2002
(Scaled Score)
.001 1.001 .002
***
1.002 .009
****
1.009
Ethnicity
Asian -.060 .942 .825
****
2.282 1.057
****
2.878
Black/Hispanic -.576
**
.562 (1.779) .088 1.092 .003 1.003
Filipino -.602
****
.548 (1.825) .294
***
1.341 -.145 .865
Other -.161 .852 .110 1.117 .293
**
1.340
Pacific Islander -.460
****
.631 (1.585) -.193
**
.824 (1.214) -.514
****
.598 (1.672)
White --
a
--
a
--
a

Ninth Grade Academic
Performance

Grade Point Average
(Year-End)
-1.095
****
.335 (2.985) .130
***
1.138 1.574
****
4.828
Core Subjects Failed .221
****
1.248 -.109
**
.896 (1.151) .834 1.035
Constant 1.405
****
-2.176
****
-11.300
Number of
Observations
9115
Pseudo R
2
(Nagelkerke) .507
Note: Reciprocal odds ratios reported in parentheses for significant, negative coefficients.
a
Referent

group is whites
*p<.10 **p<.05 ***p<.01, ****p<.001

137
Graduated on-time and enrolled in a two-year college. Being female, the
expected age for high school freshmen and not low-income increased the likelihood
of enrolling in a two-year college, compared with graduating and not going to
college. Prior achievement in reading and mathematics had small though positive
impact on likelihood of enrolling in a two-year college. Asians were more than
twice as likely as whites to enroll in a two-year college, and Filipinos were 1.34
times as likely to enroll as whites. However, whites were 1.21 times more likely
than Pacific Islanders to enroll in a two-year college. Better academic performance
in the ninth grade was associated with attending a two-year college. A one point
improvement in grade point average, such as moving from grades of Cs to Bs on
average, increased the likelihood of attending a two-year college by 1.14 times.
Each failure in a core subject increased the likelihood of not enrolling in college by
1.15 times.
Graduated on-time and enrolled in a four-year college. Students enrolling in
a four-year college had different characteristics than those who graduated high
school but did not proceed to postsecondary education immediately. Being female,
the expected age for high school freshmen and not low-income increased the
likelihood of enrolling in a four-year college compared with graduating and not
going to college. Higher prior achievement was associated with increased likelihood
of enrolling in a four-year college. Students scoring one standard deviation higher
on the eighth grade Hawai‗i State Assessment in reading were 1.15 times more likely
to enroll in a four-year college and in mathematics, 1.14 times more likely. Ethnicity

138
also had an impact. Asians were much more likely than whites to enroll in a four-
year college: 2.89 times more likely, and students classified as Other were 1.34 times
more likely than whites to enroll in baccalaureate programs. However, whites were
1.67 times more likely to enroll in a four-year college than Pacific Islanders. Failing
courses in core subject areas did not have a significant relationship with four-year
college going, but grade point average had a large impact with one point increase in
grade point average increasing the odds of four year-college enrollment by 4.83
times.
Stage 3: Multilevel Models
Multilevel models permitted analysis of school effects. The Hierarchal
Linear Model (HLM) considered the relationship of school-level characteristics to
student outcomes. HLM analyses consisted of two-level models with level-one
representing student-level characteristics and level-two representing school-level
characteristics. Independent variables in the level-one models were based on
characteristics identified as significant in Models 1 through 4. Level-two models
consisted of school climate variables representing the academic press of the school
as measured by the percentage of students reporting that their school had a standards-
based learning environment and the proportion of the sample enrolled in Algebra or
higher as ninth graders, school size, and the school community‘s relative poverty.
HLM was used to identify school-level effects for both dependent variables: on-time
high school completion (Model 5) and high school outcomes including college
enrollment (Model 6).

139
Model 5: Relationship of on-time high school completion with student and school
characteristics
School effects. Model 5 identified school effects on students‘ on-time high
school completion (see Table A5). The academic press of the school had an effect on
students‘ on-time graduation. Students‘ positive perception of whether their school
had a standards-based learning environment was associated significantly with
graduating high school; for a one standard deviation increase in students reporting
positively that their school had a standards-based environment, the likelihood of
students‘ graduating on-time increased by 1.32 times. However, the proportion of
students enrolled in Algebra or higher was not significantly associated with
graduating on-time. School characteristics of size and socioeconomic status were
significant. Larger school size had a negative effect on completing high school on-
time. More specifically, if school size increased by one standard deviation, the odds
of a student not graduating on-time increased by 1.24 times (not tabled). Increases in
school communities‘ socioeconomic status had more impact on students‘ outcomes;
an increase by one standard deviation in poverty (21% of students eligible for
Free/Reduced Price lunch) would result in an increase in likelihood of a student not
graduating on-time by 1.56 times (not tabled).

140
Table A5: HLM Logistic Model Predicting On-Time High School Completion,
Model 5

Characteristic Estimate Odds Ratio
Level One: Student Characteristics
Background Characteristics
Low-Income -.333*** .717 (1.396)
Male .0928 1.097
Overage -.856**** .425 (2.353)
Prior Achievement
Reading, 2002 (Scaled Score) -.001 .999
Mathematics, 2002 (Scaled Score) .001 1.001
Ninth Grade Academic Performance
Grade Point Average (Year-End) 1.374**** 3.950
Core Subjects Failed -.158**** .854 (1.171)
Level Two: School Characteristics
Academic Press
Students‘ perception of schools‘ standards-based
learning environment (percent positive, z-score)
.285** 1.330
Constrained curriculum (percent of cohort enrolled
in Algebra or higher)
-.500 .606
School size (total enrollment) -.0004* .9996 (1.0004)
School community‘s socioeconomic status (percent of
students who are low-income)
-.981* .3749 (2.667)
Constant 2.559
Reliability estimate (Level 1) .793
Note: Reciprocal odds ratios reported in parentheses for significant, negative coefficients.
a
Referent

group is whites
*p<.10 **p<.05 ***p<.01, ****p<.001

141
Student effects. Adding school effects to the model did not significantly
change the findings from Model 2 which examined the same outcome. The same
individual characteristics were associated significantly with graduating high school
on-time: higher income, expected age for high school freshmen, grade point average
and number of core subjects failed. The magnitude of the associations was similar to
Model 2 with students‘ status as low-income, overage, and failing courses being
associated with not completing high school on-time, and students with higher grade
point average with much higher likelihood of graduating on-time.
Model 6: Relationship of high school outcomes with student and school
characteristics
Model 6 was a two-level HLM using multinomial logistical regression.
Level-one variables consisted of student characteristics also examined in Model 4,
and level-two variables consisted of school characteristics also explored in Model 5.
School climate variables had some association with outcomes, with positive school
climate variables related to students graduating on-time. As with Model 5, the
proportion of cohort enrolled in algebra or higher as a proxy for constrained
curriculum was not significantly associated with graduation outcomes. The results
of Model 6 are presented in Table A6.

142
Table A6: HLM Multinomial Logistic Model Predicting High School Outcomes,
Model 6

Graduated High School On-Time
Did Not Graduate High
School On-Time
Enrolled in Two
Year College
Enrolled in Four
Year College

Estimate
Odds
Ratio Estimate
Odds
Ratio Estimate Odds Ratio
Level One: Student
Characteristics

Background Characteristics
Low-Income .135 1.144 -.524
****

.592
(1.689)
-.796
****

.451
(2.217)
Male -.169
**

.844
(1.184)
-.158
***

.854
(1.171)
-.083
.920
(1.087)
Overage .834
****
2.304 -.123
.884
(1.131)
-.506
****

.603
(1.659)
Prior Achievement
Reading, 2002 (Scaled
Score)
.0019
*
1.0019 .001
*
1.001 .008
****
1.008
Mathematics, 2002
(Scaled Score)
.0009 1.0009 .002
***
1.002 .010
****
1.010
Ninth Grade Academic
Performance

Grade Point Average
(Year-End)
-1.184
****

.306
(3.267)
.240
****
1.271 1.72
****
5.586
Core Subjects Failed .203
****
1.225 -.089
*

.915
(1.093)
.060 1.062
Level Two: School
Characteristics

Academic Press
Students‘ perception of
schools‘ standards-based
learning environment
(percent positive, z-score)
-.339
**

.712
(1.404)
-.159
**

.853
(1.168)
-.101
.904
(1.106)
Constrained curriculum
(percent of cohort enrolled
in Algebra or higher)
.375 1.45 -.122
.885
(1.130)
.165 1.179
School size (total enrollment) .0004
*
1.0004 .0002 1.0002 .00003 1.00003
School community‘s
socioeconomic status (percent
of students who are low-
income)
.717 2.049 -.800
**

.450
(2.226)
-.466
.628
(1.594)
Constant -1.804
****

.165
(6.074)
-.085
.919
(1.089)
-1.656
****

.191
(5.238)
Reliability estimate (Level 1) .780 .676 .739
Note: Reciprocal odds ratios reported in parentheses for significant, negative coefficients.
a
Referent

group is whites
*p<.10 **p<.05 ***p<.01, ****p<.001

143
Did not graduate high school on-time. Students‘ likelihood of on-time
graduation increased when more students‘ positively reported that their school had a
standards-based learning environment. Students in schools with a standard deviation
higher School Quality Survey report on ―Dimension A‖ were 1.33 times more likely
to graduate on-time. Students attending smaller schools had a higher likelihood of
graduating on-time; as school enrollment increased, the likelihood of not graduating
on-time increased, though the effect was small (1.0004) for each additional student
(i.e., for 100 students added, the effect would be 1.04, or a four percent increase in
likelihood to graduate on-time). Socioeconomic status, either the school
community‘s relative poverty or the individual student‘s eligibility for Free/Reduced
Price Lunch, did not have a significant relationship with the likelihood of graduating
on-time when school characteristics were taken into account. However, students‘
ninth grade academic performance had a significant relationship with their likelihood
of graduating on-time. Students with higher grades were more likely to graduate;
each one point increase in grades on the four point scale increased the likelihood of
graduating by 3.30 times. Failing core subjects had a negative impact on graduating
with failure increasing the likelihood of not graduating on-time by 1.17 times for
each of four core subjects failed. Males were more likely to graduate on-time than
females, and being overage was a significant disadvantage since overage students
were 2.35 times more likely not to graduate on-time.
Graduated on-time and enrolled in a two-year college. Students attending
schools where more students reported a stronger standards-based learning

144
environment were more likely to graduate on-time but not go to college than students
who enrolled in two-year colleges. Income of the community and of the student‘s
family were both associated significantly with graduates not enrolling in college
compared with enrollment in a two-year college. Low-income and male students
were more likely to not enroll in a two-year college (1.69 and 1.17 times,
respectively). Grades remained an important factor in explaining two-year college
enrollment but, compared with other outcomes and models, the GPA and passing
core subjects did not have a strong impact on two-year college enrollment compared
with graduating but not choosing college as increasing GPA by one point increased
the likelihood of enrolling in a two-year college by 1.27 times and failing a core
subject reduced the likelihood of enrolling in a two-year college by 1.09 times.
Graduated on-time and enrolled in a four-year college. In comparing on-
time graduates who enrolled in a four-year college with those who did not enroll in
college, none of the school process variables was significant. The model suggests
that students‘, not schools‘, characteristics and performance were related to four-year
college going behavior. Low-income graduates were 2.22 times more likely to not
enroll in college, and overage graduates were 1.66 times more likely to also not
enroll in college. Higher prior academic achievement increased the likelihood of
going to a four-year college; a student had an increased likelihood of enrolling in a
four-year college of 1.77 times if they had a one standard deviation increase in eighth
grade reading scores and 1.99 times in eighth grade mathematics scores. Ninth grade
GPA was very significant in explaining graduates‘ four year college enrollment;

145
graduates‘ likelihood of enrolling in a four-year college increased by 5.58 times for
each point increase in GPA.
Strength of Models
An important consideration of regression models is the extent to which the
models explain the variation in the data. Typically, the explanatory power of a
model is measured by the R-squared. However, R-squared cannot be calculated for
logistic regressions which have categorical outcomes. Pseudo R-squared is
associated with logistic regressions. While pseudo R-squared cannot be interpreted
as explaining the proportion of variance in the sample data explained by the model,
different models‘ pseudo R-squared are ―valid and useful in evaluating multiple
models predicting the same outcome on the same dataset‖ (UCLA Academic
Technology Services, 2010). Thus, higher pseudo R-squared indicates the relative
strength of a model to explain a dataset. The Nagelkerke Pseudo R-squared was
selected among the pseudo R-squared statistics reported by SPSS for each logistic
regression because Nagelkerke‘s R-squared is the ―most reported of the pseudo R-
squared estimates‖ (Garson, 2010). When a model perfectly predicts the outcome,
the value of the Nagelkerke R-Squared is one (UCLA Academic Technology
Services, 2010).
Pseudo R-squared statistics of models presented in this analysis indicate that
including ninth grade academic performance strengthens the model. In both the
binary and the multinomial logistic regressions, the pseudo R-squared increased
when ninth grade academic performance was considered in addition to students‘

146
background characteristics. Furthermore, Models 3 and 4 which considered college
enrollment in addition to graduating high school on-time had much higher pseudo R-
squared than Models 1 and 2.
The multi-level models also demonstrated high reliability, ranging from .67
to .79, for level one models in Models 5 and 6.

Abstract (if available)

Abstract Many studies examine the impact of students’ characteristics and behaviors on high school outcomes: high school completion, college enrollment or college completion. This study uses regression analyses to explore the association of students’ characteristics and behaviors and students’ positive high school outcomes: graduating on-time, enrolling in a two-year college or enrolling in a four-year college. The study analyzes the longitudinal student records for a Hawaii cohort of public high school graduates.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

The influence of counselors and high school organization on the selection of participants for a dual credit program

PDF

Perceptions of Hawai`i TRIO Talent Search staff and target high school administrators of college access factors and project effectiveness

PDF

Hawaii Board of Education's middle grade promotion policy: a policy implementation study

PDF

Who do we tell? School violence and reporting behavior among native Hawaiian high school students

PDF

The estimated cost of public preschool in Hawaii

PDF

School funding and the evidence based model: an examination of high school budget allocation in Hawaii

PDF

Factors inhibiting application for financial aid by low-income students at the University of Hawaii at Manoa

PDF

Changing course, creating opportunity: a study on the implementation of an early college model for African American ninth grade students

PDF

Predictors of success in meeting the high school mathematics and English requirements for college entrance: a longitudinal study

PDF

An examination of resource allocation strategies that promote student achievement: case studies of rural elementary schools in Hawaii

PDF

Native Hawaiians and college success: Does culture matter?

PDF

Personnel resource allocations: a case study of one Hawaii complex

PDF

Accreditation and accountability processes in California high schools: a case study

PDF

High school curriculum: Self-regulation for the secondary student

PDF

Resource allocation strategies and educational adequacy: an examination of an academic & financial plan used to allocate resources to strategies that promote student achievement in Hawaii

PDF

Factors related to college graduation among private and public secondary school students

PDF

The perceptions of faculty and administrators of remedial mathematics education in two higher education institutions

PDF

The open enrollment of advanced placement classes as a means for increasing student achievement at the high school level

PDF

Personnel resource allocation in a Hawaii school complex

PDF

A case study of student engagement in a high performing urban continuation high school

Asset Metadata

Creator Oyadomari-Chun, Tammi J. (author)

Core Title Designing an early warning system for Hawaii: identifying indicators of positive high school outcomes

School Rossier School of Education

Degree Doctor of Education

Degree Program Education

Publication Date 08/06/2010

Defense Date 05/07/2010

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

Tag college enrollment,early warning,high school drop out,high school graduation,longitudinal student data,ninth grade transition,OAI-PMH Harvest

Place Name Hawaii (states)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Brewer, Dominic J. (committee chair), Heck, Ronald (committee member), Picus, Lawrence (committee member)

Creator Email tammi@hawaii.edu,tammichun@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m3322

Unique identifier UC1190224

Identifier etd-OyadomariChun-3881 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-371247 (legacy record id),usctheses-m3322 (legacy record id)

Legacy Identifier etd-OyadomariChun-3881.pdf

Dmrecord 371247

Document Type Dissertation

Rights Oyadomari-Chun, Tammi J.

Type texts

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

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu