Essay Grade

Volume 19 ✤ Number 1 ✤ Fall 2007 ✤ pp. 32–64
More Than One Gap:
Dropout Rate Gaps Between and Among
Black, Hispanic, and White Students
Dick M. Carpenter II & Al Ramirez
University of Colorado, Colorado Springs
t
This study is the second in a series of investigations designed to
explore issues surrounding the achievement gap, or, as concluded
in our prior work, achievement gaps. The first study (Carpenter,
Ramirez, & Severn, 2006) used data from the National Education
Longitudinal Study of 1988 (NELS: 88) to examine nuances of
academic achievement gaps among Black, White, and Hispanic
students, with a particular focus on not only gaps between groups
but also within groups. Findings from the analysis showed
unique patterns and multiple achievement gaps, both between
and within groups. In fact, results indicated within-group gaps
were often more significant than gaps between groups.
This research extends that effort by examining variables
associated with dropout behavior as a measure of achievement
gaps. As in the first study, comparisons were made among Black,
White, and Hispanic students, paying particular attention to
gaps in dropout rates both between and within groups. The
research progressed in two phases. Phase I used the same variables from the prior investigation to determine if the patterns
among independent variables would prove consistent with a different dependent variable (dropout status rather than academic
achievement on tests). Phase II added another index of variables
more conceptually aligned with dropout behavior.
Results from Phase I showed little consistency with findings
from our first investigation and the resulting logistic hierarchi32
Copyright © 2007 Prufrock Press, P.O. Box 8813, Waco, TX 76714
The achievement gap, traditionally measured by test scores, also can
be documented by dropout behavior. Examining dropout behavior
among Black, White, and Hispanic students, with a particular focus
on gaps within groups and not just between Whites and minorities,
shows a clearer picture of the achievement gap. The results of our study
show multiple achievement gaps both between and within groups, ultimately concluding that within-group gaps were often more significant
than gaps between groups. Through hierarchical linear modeling, we
and number of suspensions. Hispanic and White students showed three
additional predictors in common—time spent on homework, gender,
and family composition. White and Black students shared only one
common predictor beyond suspensions and being held back: parental
involvement. Black and Hispanic students shared no additional common predictors. Finally, race/ethnicity generally proved not to be a
significant predictor of dropping out. Gaps within groups may be more
significant than those between groups. Such differences further reinforce our concern about the practice of establishing policy initiatives
that conflate all minority group students into a monolithic whole. Our
research suggests that policy makers and school leaders should craft
dropout prevention policies and programs with sufficient flexibility to
allow school-level personnel to individualize said policies and practices
based on local conditions.
Carpenter, D. M., II, & Ramirez, A. (2007). More than one gap: Dropout rate gaps
between and among Black, Hispanic, and White students. Journal of Advanced
Academics, 19, 32–64.
summary
found two common predictors for all three groups—being held back
Dropout Rates
cal generalized linear models (HGLM) explained little variance.
HGLM results from Phase II indicated some notable patterns
when comparing models between Black, White, and Hispanic
students. Specifically, significant predictors for White and
Hispanic students showed some commonality between groups,
but significant predictors for Black students showed less overlap
with the other two student groups. Finally, Phase II models also
explained little overall variance.
Literature Review
In the large and growing literature on closing the achievement gap, a common theme is a singular definition of the term.
Yet, as we demonstrated in a prior study, this singular definition
fails to describe the actual multilayered definition of differences
in achievement, whereby there is not one gap but many gaps
(Carpenter et al., 2006). Moreover, while the singular definition
typically describes achievement differences between White and
minority students, our results indicated within-group gaps can
be significantly greater than differences between groups. Indeed,
when examining significant predictors among Black, White, and
Hispanic students, results indicated substantive overlap between
variables in best fit models for each group. By substantive overlap,
we mean all three models included socioeconomic status (SES),
inclusion in an ESL program, and parental involvement, and
coefficient directions across groups were identical. Increases in
SES and parental involvement resulted in higher math achievement. Moreover, the Black and White models shared homework
as a significant predictor, and Hispanic and White models shared
number of units in Algebra I.
A second common theme in the achievement gap literature,
including our prior work (Carpenter et al., 2006), is the use of academic achievement as the dominant dependent variable, as often
measured by test scores and other related indicators. The use of
dropout status to measure gaps in achievement is less common.
This is an unfortunate dynamic given the general acknowledg34
Journal of Advanced Academics
Carpenter & Ramirez
ment that dropout rates are comparably greater among minority
students (Darling-Hammond, 2006, 2007). Indeed, authors such
as Roscigno (1999) note that minority students, particularly Black
students, drop out at higher levels than their White counterparts.
However, much of this research is based on qualitative studies that
are informative, but often lack the depth of investigation and the
precision needed to guide viable policy initiatives.
For example, among the most widely circulated reports specifically on the Hispanic dropout problem is Secada et al. (1998).
This effort sought to collect expert opinions from researchers
and practitioners about the causes and solutions to the problem. Public hearings and commissioned papers were part of the
methodology as well. In another example, Neumann (1996)
used observations, surveys, and interviews with students, teachers, and school administrators to understand the factors that
explained the low dropout rate for Mexican American students
in one California high school. He identified a myriad of programs and policies as the reason for the low number of dropouts.
In an earlier study of dropouts in California, Pulido (1991) also
used interviews and observations to collect data from 18 high
schools with high Hispanic enrollments and low dropout rates.
His purpose was to identify factors that contributed to the high
retention and correlate these findings with the literature about
effective schools.
Researchers who take a quantitative approach to the examination of dropping out commonly draw on large national datasets,
such as those produced by the National Center for Education
Statistics (NCES) or the U.S. Census Bureau. For example,
using the High School and Beyond database, Melnick and Sabo
(1992) investigated the influence of interscholastic sports on
dropping out. They found this aspect of school offered a limited deterrence for a small number of students. Perrira, Harris,
and Lee (2006) examined data from the National Longitudinal
Study of Adolescent Health and found human, cultural, community, and family capital explained why second generation
children of immigrants were at higher risk of dropping out than
first generation children. The researchers looked at data from
Volume 19 ✤ Number 1 ✤ Fall 2007
35
Dropout Rates
the in-school survey of students, which included more than
12,000 participants. The study found differences in the influence
of the selected variables both between and within ethnic and
racial groups. Generational differences also were reported within
immigrant groups.
The National Educational Longitudinal Study of 1988
(NELS: 88) also has seen particularly frequent use in dropout
research (Warren & Lee, 2003; Yin & Moore, 2004). For example, Lan and Lanthier (2003) used NELS: 88 to measure the
relationship between dropping out and academic performance,
relationships with teachers, relationships with peers, perceptions
of school, participation in school activities, motivation for school
work, effort expended in school work, self-esteem, and locus of
control. Results showed a developmental pattern of the personal attributes of dropout students and identified the transition
to high school as a particularly critical time for interventions.
Croninger and Lee (2001) examined the relationship between
social capital and dropping out using the NELS: 88 database and
discovered that teachers are an important source of social capital,
which can reduce the probability of dropping out by as much as
half. Students with past academic difficulties find guidance and
assistance from teachers especially helpful. Teachman, Paasch,
and Carver (1996) likewise studied the relationship between
dropping out and social capital using the NELS: 88 database
and found changing schools is particularly detrimental.
Some authors, such as Lee and Burkam (2003) and
Goldschmidt and Wang (1999), examined school factors and
the likelihood of dropping out also using NELS: 88. Lee and
Burkam applied multilevel methods to explore the influence
of school factors, such as curriculum, size, and social relations,
taking into account students’ academic and social background,
including race/ethnicity. In schools that offer mainly academic
courses, students are less likely to drop out. Similarly, students in
smaller schools more often stay in school, as do those where relationships between teachers and students are positive. For their
part, Goldschmidt and Wang used NELS: 88 data to study early
dropouts, those who leave in middle school, and late dropouts,
36
Journal of Advanced Academics
Carpenter & Ramirez
those who leave in high school, paying particular attention to
differences between the groups. Results showed a general difference in significant predictors between the groups, although
being held back is the strongest predictor of dropping out for
both early and late leavers.
As revealing as these studies are, however, they do not necessarily focus exclusively on differences between groups based
on race/ethnicity. The authors do include race/ethnicity in the
research, but it is more often used as a covariate. Moreover, a
robust collection of empirical studies that examine dropout status as a measure of achievement gap is missing. This study seeks
to contribute to that collection by focusing on both dropout status as the measure of achievement gaps and differences within
groups as well as between.
Predictors of Dropout Behavior
To do so, we examined within-group differences in the likelihood of dropping out for Black, Hispanic, and White students
separately by running hierarchical generalized linear models for
each group. We also combined all three groups into one sample
to examine if there were significant differences in the likelihood
of dropping out based on race/ethnicity. As described in detail
below, we proceeded in two phases. Phase I included variables
from our first achievement gaps study to determine if those same
variables proved significant with dropping out as the dependent
variable rather than academic achievement. Because Phase I
results proved inconclusive, Phase II introduced an additional set
of predictor variables more closely aligned with dropping out.
The Phase I variables included: time spent on homework during the week outside of school, SES, number of units of Algebra
1, participation in an ESL program, language other than English
regularly spoken at home, family composition, parental involvement, student race/ethnicity, teacher certification, enrollment,
percentage of White students in the school, school type, and
urbanicity. To varying degrees, each of these variables has been
shown to influence the likelihood of dropping out. Beginning
Volume 19 ✤ Number 1 ✤ Fall 2007
37
Dropout Rates
with variables at the student/family level, Alexander, Entwisle,
and Kabbani (2001) considered at-risk factors for dropping out
among children in the Baltimore public schools and identified
socioeconomic status of the family as a key predictor. Cairns,
Cairns, and Neckerman (1989) also found socioeconomic status
of the family along with aggressive behavior and poor grades as
variables associated with dropping out.
Language issues also have been examined as associated with
dropping out, such as Theobald’s (2003) study of dropout statistics relative to the kind of English language acquisition program
to which Hispanic students were assigned. Additionally, family
characteristics, such as family composition, influence decisions to
complete high school (Astone & McLanahan, 1994). Rumberger,
Ghatak, Poulos, Ritter, and Dornbusch (1990) underscored the
value of parent involvement in education, while pointing out that
students who are on their own regarding schooling decisions are
at higher risk of leaving school before graduation.
How students use their time in and out of school, such as
time spent on homework at home or courses they take at school,
has been analyzed in connection with dropout statistics. Natriello,
McDill, and Pallas (1985) showed a curvilinear relationship
between time spent on homework and likelihood of dropping out,
and Fratt (2006) reported on the positive relationship between
success in algebra courses and completing high school.
Among the school-level variables, several authors have examined the relationships between dropping out and school size and
composition of the student body. Merritt (1983) and Alspaugh
(1998) concluded students in larger schools drop out more often,
as do students in schools with a greater percentage of minorities (Rumberger & Palardy, 2005). School type and urbanicity
likewise appear to predict dropping out. Specifically, Hendrie
(2004) reported students in private schools drop out less than
those in public schools, and rural school students drop out more
often than those in other settings (Roscigno & Crowley, 2001).
Finally, numerous authors posit a relationship between teacher
quality and student outcomes, concluding lower teacher qual-
38
Journal of Advanced Academics
Carpenter & Ramirez
ity contributes to a greater likelihood of dropping out (DarlingHammond, 2006; Davis & Dupper, 2004; Heck, 2007).
The Phase II variables included: if the student was ever held
back, number of suspensions, inclusion in a dropout program,
country of birth, gender, hours per day watching TV, hours per
week spent working, hours per week in extracurricular activities, how often the student uses a computer per week, number
of siblings who dropped out, 8th-grade reading test score, 8thgrade math test score, 10th-grade reading test score, 10th-grade
math test score, percent of 10th graders who drop out before
graduation, percent of students in a dropout program, if a test
is required for graduation, if the school district allows choice in
enrollment, the level of gang problems in the school, and how
much influence gangs have in compelling others to dropout.
The decision to include some of these variables is rather selfevident, such as inclusion in a dropout program, number of siblings
who have dropped out, percent of students in a dropout program,
or the amount of influence gangs play in compelling others to
dropout. For other variables, the conceptual alignment is not as
self-evident but still conceptually rational. For example, poor academic achievement has been shown to be related to dropping out
(Natriello et al., 1985; Reyes & Jason, 1991), as has an increase
in the number of hours of paid employment (Warren & Cataldi,
2006; Warren & Lee, 2003), retention in a grade (Frey, 2005;
Shepard & Smith, 1990), and exclusion from school for disciplinary reasons (Kokko, Tremblay, Lacourse, Nagin, & Vitaro, 2006).
Still other variables may not appear closely aligned to dropout behavior, but have been shown to be important nonetheless.
For example, student gender interacts with dropout behavior
when teenage girls are pregnant (Turner, 1995; Warren & Lee,
2003) or begin families without marrying (Cairns et al., 1989).
Both circumstances serve as predictors of dropping out of high
school (Manlove, 1998). Another includes how students use
their time outside of school, such as in extracurricular activities.
McNeal (1995) proffered that not all student participation in
extracurricular activities has the same effect of deterring students
from leaving school. For example, sports and fine arts do appear
Volume 19 ✤ Number 1 ✤ Fall 2007
39
Dropout Rates
to reduce dropout rates of participants, but academic and vocational clubs do not. He went on to point out that the strength of
these dynamics supersedes race, economic status, and gender.
Another predictor with somewhat mixed results includes a
student’s country of birth. Specifically, children of first generation immigrant parents with greater social and cultural capital
drop out less often than those with less capital, but that dynamic
wanes among children of parents in second generations and
beyond (Perrira et al., 2006). Finally, some have examined the
relationship between school reform efforts and their effect on
dropout activities (Natriello et al., 1985). High-stakes testing
(Shriberg & Shriberg, 2006) and exit examinations from high
school (Viadero, 2005) are two factors cited as contributing to
higher dropout rates.
Methods
Using the aforementioned variables, this study was implemented in two phases. The first phase applied models resulting
from our 2006 research (called Study 1 hereafter) using dropout
status as the dependent variable, rather than academic achievement. In so doing, we sought to determine if (a) the same predictors from Study 1 and (b) the pattern of within-group and
between-group gaps proved consistent with a different dependent variable. Because the models from Study 1 did not produce similar results with the dropout status dependent variable,
we proceeded to a second phase wherein we modeled dropout
status among Black, White, and Hispanic students using the
aforementioned additional index of variables more conceptually
aligned to dropping out. Specific procedures in each phase are
included below.
Data and Sample
Data for this study came from NELS: 88. Conducted by the
National Center for Education Statistics (NCES), NELS: 88
40
Journal of Advanced Academics
Carpenter & Ramirez
represents the third in a series of longitudinal studies of cohorts
of American students. NELS: 88 began collecting data on students during their eighth-grade year and continued into high
school, postsecondary education, and into the labor force. The
design for NELS: 88 included a questionnaire and a cognitive
test for each student in the sample. Questionnaires also were
administered to each student’s parents, school principal, and two
of his or her teachers.
Student questionnaires asked for information about selected
background characteristics, including English language proficiency, attitudes, career and college plans, school experience, and
extracurricular activities. Principals and headmasters answered
specific questions about the school. Parents reported on family resources, parent involvement in school, educational opportunities supported outside of school, and financial planning for
college. Two teachers of each sampled student completed a questionnaire designed to collect data and evaluations concerning
the educational progress and motivation of the student, the academic difficulty of the class in which the student was enrolled,
the school itself, and the teachers’ prior educational experiences.
NELS: 88 employed a two-stage, stratified random sample
design. To ensure a balanced sample, schools were first stratified
by region, urbanicity, and percentage of minority students prior
to sampling. The school sample was restricted to regular public
and private schools (including independent, Catholic, and other
types of religious schools) that enrolled eighth graders. The second stage of the sampling process selected the students within
the schools.
Successive follow-ups, or waves, occurred in 1990 (F1—10th
grade), 1992 (F2—12th grade), 1994 (F3—2 years after high
school), and 2000 (F4—8 years after high school). F1 and F2
included school administrator, teacher, and student questionnaires and student cognitive tests. F2 also included a parent survey and high school transcripts. F3 included only a student survey,
and F4 utilized a student questionnaire and college transcripts.
Volume 19 ✤ Number 1 ✤ Fall 2007
41
Dropout Rates
Sample
The total sample in this study includes 17,613 participants
measured at F2; 2,010 were Black (10.4%), 2,445 were Hispanic
(12.6%), and 13,158 were White (67.9%). Excluded from the
sample are students in other racial groups (i.e., Asian, Pacific
Islander, American Indian, Alaska Native) or any participants
not present in the baseline year (BY), F1, and F2. Those not
present in all three waves does not mean dropouts were excluded,
as dropping out did not mean elimination from NELS data
collection. Data were gathered in all three waves on students
regardless of enrollment status. Those not present in all three
waves included participants lost through common mortality
(e.g., death, choosing to leave the study) or those added after
the base year to create freshened samples. Because of missing
data, cases also were excluded at the analysis stage. More specific details about sample sizes are included below in discussions
about the phases of the study.
Phase I Procedures
Phase I began with a three-level HGLM logistic model using
all predictors from Study 1 (Carpenter et al., 2006) and race/
ethnicity to examine if significant differences exist in the probability of dropping out based on race/ethnicity. Table 1 lists all of
the Study 1 predictors and briefly describes the coding of each.
Three-level HGLM modeling was used because the data structure for the first model of this phase includes students nested
within teachers nested within schools. Specifically, seven of these
variables (time spent on homework during the week outside of
school, SES, number of units of Algebra 1, participation in an
ESL program, language other than English regularly spoken at
home, family composition, parental involvement, and race/ethnicity, dummy coded) are measured at the student/family level.
One variable (teacher certification) is measured at the teacher/
classroom level. Four variables are measured at the school level
(enrollment, percentage of White students in the school, school
42
Journal of Advanced Academics
Carpenter & Ramirez
type, and urbanicity, which is dummy coded with suburban as
the reference). Therefore, the three level model is:
Level 1:
Level 2:
Level 3:
η = π0 + π1(homework) + π2(SES) + π3(Alg) + π4(ESL)
+ π5(Eng) + π6(Family) + π7(Par inv) + π8(Black) +
π9(Hispanic)
π0 = β00 + β01 (Teacher cert) + r0
β00 = γ000 + γ001(Enroll) + γ002(Per white) + γ003(School
type) + γ004(Urban) + γ005(Rural) + u00
where η represents the log odds of dropping out of school.
The sample sizes for this part of Phase 1 included 6,940 at
level one; 2,364 at level two; and 654 at level three for all three
groups combined.
Among researchers, practitioners, and policy makers, what
constitutes dropping out remains contested (Warren & HalpernManners, 2007). For example, NCES proffers no less than four
perspectives on dropping out: the event dropout, the status dropout, the status completion rate, and the average freshmen graduation rate (Laird, DeBell, & Chapman, 2006). Greene (2001)
attempted to present a clear picture of high school completion
by arguing that government generated dropout and graduation
rates that misreport and mask the true extent of the problem.
His research calculated a graduation rate as the number of regular diplomas issued compared to eighth-grade enrollment 4 years
earlier.
NELS: 88 also measures dropout status in different ways.
Thus, the variable used in this study for the dependent variable
was F2RWTST, which is the participant’s enrollment status
at F2, similar to the status dropout listed above. In its original
form, this variable includes three categories—in school/in grade,
in school/out of grade, and dropout. This was transformed into
a dichotomous (1 = yes/0 = no) variable where “yes” included all
of those who dropped out, and “no” included those who were
enrolled, despite in or out of grade status.
Results from this model showed, among other things, that
teacher certification was not a significant predictor, β = -.218
Volume 19 ✤ Number 1 ✤ Fall 2007
43
44
†
Journal of Advanced Academics
Nominal, 1 = yes/0 = no
Nominal, 1 = yes/0 = no
Ordinal, 0 = none, 1 = 1 hour or less, 2 = 2–3 hours, 3 = 4–6 hours, 4 = 7–9 hours, 5 =
10–12 hours, 6 = 13–15 hours, 7 = more than 15 hours
Ordinal, 1 = 1–399, 2 = 400–599, 3 = 600–799, 4 = 800–999, 5 = 1000–1199, 6 =
1200–1599, 7 = 1600–1999, 8 = 2000–2499, 9 = 2500+
Nominal, 1 = public/0 = private
Nominal, urban, suburban, rural†
Ordinal, 0 = 91–100, 1 = 76–90, 2 = 51–75, 3 = 26–50, 4 = 0–25
Nominal, 1 = standard certification/0 = less than standard certification
Ordinal, 0 = less than one year, 1 = one year, 2 = more than one year
Nominal, 1 = two parents or guardians in the home/0 = one parent or guardian in the
home
Ordinal, 0 = not involved, 1 = somewhat involved, 2 = very involved
Nominal, Black, Hispanic, White‡
Student ever in ESL program
Language other than English regularly spoken at home
Time spent on homework out of school
School enrollment
School type
Urbanicity
Percentage of White students in the school
Teacher certification
Years of Algebra I completed
Family composition
Parental involvement
Race/Ethnicity
Dummy coded, suburban as reference ‡ Dummy coded
Continuous (a composite variable including mother’s education, father’s education,
mother’s occupation, father’s occupation, and family income)
Scale of Measurement
SES
Variable
Independent Variables in Phase I
Table 1
Dropout Rates
Carpenter & Ramirez
(.32), odds ratio = .80, p = .498. This proved to be true not only
with the entire sample but also when separate models were run
for each racial/ethnic group: for Black students, teacher certification β = -.323(.78), odds ratio = .72, p = .682; for Hispanic students teacher certification β = 1.04(1.15), odds ratio = 2.85, p =
.364; and for White students teacher certification β = .134(.35),
odds ratio = 1.03, p = .708. For models run for each racial/ethnic
group, sample sizes were as follows: for Black students, 591 at
level one, 375 at level two, and 198 at level three; for Hispanic
students, 540 at level one, 354 at level two, and 193 at level three;
and for White students, 5,056 at level one, 1,793 at level two,
and 577 at level three.
In an effort to make the modeling more parsimonious,
we dropped teacher certification and collapsed the three-level
modeling into two levels, because teacher certification was the
only predictor in level two. The two-level HGLM models then
included students nested within schools using the same list of
variables described above. Therefore, the models are:
Level 1:
Level 2:
η = β0 + β1(homework) + β2(SES) + β3(Alg) + β4(ESL)
+ β5(Eng) + β6(Family) + β7(Par inv) + β8(Black) +
β9(Hispanic)
β00 = γ00 + γ01(Enroll) + γ02(Per white) + γ03(School
type) + γ04(Urban) + γ05(Rural) + u0
Using two level models, we first used the entire sample to examine whether there were significant differences in probability of
dropping out based on race/ethnicity. The sample sizes for this
model included 11,228 at level one and 762 at level two.
Finally, Phase I ended by running separate models for each
racial/ethnic group to facilitate a comparison of models between
groups. In so doing, we sought to create the most parsimonious model for each group containing only significant predictors, which then enabled us to determine (a) how much overlap
would be present among the resulting models and (b) how well
those models corresponded to the ones ascertained in our first
study. For this part of Phase I and all of Phase II (described
Volume 19 ✤ Number 1 ✤ Fall 2007
45
Dropout Rates
below), sample sizes for Black students included 1,142 students
at level one and 303 students at level two. For Hispanic students,
level one had 1,326 students and level two had 328. For White
students, level one had 8,010 students and level two had 700.
Phase II Procedures
As discussed below, Phase I results did not produce models of great consistency with Study 1. Therefore, we introduced
into the two-level HGLM modeling an additional set of predictor or independent variables, as indicated in Table 2. We did
so by retaining the significant predictors from Phase I for each
racial/ethnic group and adding the new index of variables to
each group’s modeling. The additional variables were chosen due
to their conceptual tie to dropping out. Of these variables, 15
were student/family-level variables and entered at level one (ever
held back, number of suspensions, ever in a dropout program,
country of birth, gender, hours per day watching TV, hours per
week spent working, hours per week in extracurricular activities,
how often uses a computer per week, number of siblings who
dropped out, 8th-grade reading test score, 8th-grade math test
score, 10th-grade reading test score, and 10th-grade math test
score). The remaining six were school-level variables and entered
at level two (percent of 10th graders who drop out before graduation, percent of students in a dropout program, test required for
graduation, school district allows choice in enrollment, the level
of gang problems in the school, and how much influence gangs
have in compelling others to dropout).
As in Phase I, multiple models were run for each racial group
separately to create parsimonious models, which, in turn, facilitated
a comparison of significant predictors between groups. Finally, all
of the Phase I and Phase II independent variables were introduced
into a full model with race/ethnicity as an additional predictor to
measure, again, if differences in probability of dropping out were
significant based on race/ethnicity. As in Phase I when using the
entire sample for two-level modeling, the sample sizes for this
model included 11,228 at level one and 762 at level two.
46
Journal of Advanced Academics
Scales of Measurement
Nominal, 1 = yes/0 = no
Ordinal, 0 = never, 1 = 1–2 times, 2 = 3–6 times, 3 = 7–9 times, 4 = more than 10 times
Nominal, 1 = yes/0 = no
Nominal, 1 = USA/0 = elsewhere
Nominal, 1 = male/0 = female
Nominal, 1 = yes/0 = no
Ordinal, 0 = none, 1 = less than 1 hour, 2 = 1–2 hours, 3 = 2–3 hours, 4 = 3–4 hours, 5 = 4–5
hours, 6 = more than 5 hours
Hours spent working during week
Ordinal, 0 = 0–10, 1 = 11–20, 2 = 21–30, 3 = 31–40, 4 = more than 40
Hours per week in extracurricular activities
Ordinal, 0 = none, 1 = less than 1 hour, 2 = 1–4 hours, 3 = 5–9 hours, 4 = 10–19 hours, 5 =
20 hours or more
How often uses computer at home
Ordinal, 0 = none, 1 = less once a week, 2 = once or twice a week, 3 = every day
8th-grade reading test score
Continuous, estimated number right using IRT
8th-grade math test score
Continuous, estimated number right using IRT
10th-grade reading test score
Continuous, estimated number right using IRT
10th-grade math test score
Continuous, estimated number right using IRT
Percent of 10th graders who drop out before graduation Continuous
Percent of students in dropout program
Ordinal, 0 = 0–10, 1 = 11–24, 2 = 25–49, 3 = 50–74, 4 = 75–100
Gang activity a problem at school
Ordinal, 0 = no problem, 1 = minor problem, 2 = moderate problem, 3 = serious problem
Gangs influence others to dropout
Ordinal, 0 = no influence, 1 = small, 2 = some, 3 = moderate, 4 = major
School allows some element of enrollment choice
Nominal, 1 = yes/0 = no
Students must pass test to receive diploma
Nominal, 1 = yes/0 = no
Variable
10th grader ever held back
How many times suspended from school
Student ever in dropout program
10th grader’s birthplace
Student gender
If student had siblings who dropped out
Hours spent watching TV weekdays
Phase II Independent Variables
Table 2
Carpenter & Ramirez
Volume 19 ✤ Number 1 ✤ Fall 2007
47
Dropout Rates
Limitations
In readi