Effects of High School Course-Taking on Test Scores, High ...
Transcript of Effects of High School Course-Taking on Test Scores, High ...
Draft version. Please do not cite without authors' permission.
Effects of High School Course-Taking on Test Scores, High School Graduation, and Postsecondary Entry*
Mark C. Long University of Washington
Daniel J. Evans School of Public Affairs Box 353055
Seattle, WA 98195-3055 206-543-3787 [email protected]
Dylan Conger
The George Washington University Trachtenberg School of Public Policy and Public Administration
805 21st Street NW, MPA 601G Washington, DC 20052
202-994-1456 [email protected]
Patrice Iatarola Florida State University
Department of Educational Leadership & Policy 113G Stone Building
Tallahassee, FL 32306-4452 850-644-8162 [email protected]
April 2, 2009
Running head: Effects of High School Course-Taking ______________________________________________________________________________*Prepared for the City University of New York Office of Policy Research's Higher Education Policy Seminar Series (April 9). Funding for this research was provided by the Florida Department of Education and the U.S. Department of Education, Institute of Education Sciences, Grant R305B070131. We are especially grateful to the Florida Department of Education for maintaining such a comprehensive and organized data system, and for making the data available to us for analysis. We also thank Danielle Fumia for superb research assistance. All authors contributed equally.
1
Abstract
Many states have increased the number and rigor of courses that are required for graduation, yet the
literature on the effects of course-taking has not kept pace with these policy developments. Using
panel data from a census of public school students in the state of Florida, we examine the effects of
course-taking in various subjects on 10th grade test scores, high school graduation, and entry into
postsecondary institutions. We control for course-availability, 8th grade test scores, and other
student, school, and neighborhood characteristics. We use ordinary least squares, propensity score
matching, and instrumental variables in our effort to identify causal effects.
2
1. Introduction
U.S. students fare poorly on international exams relative to students in other industrialized
nations and their high school graduation rates remain surprisingly low (NCES, 2005; Heckman &
LaFontaine, 2008). Many educators place blame on poor quality public school investments and
undemanding high school graduation requirements. In response, states have increased the number
and rigor of courses students are required to take in order to graduate. Yet, the research literature on
the effect of course-taking on achievement patterns in high school and college preparedness has not
kept pace with these policy developments. Much of the existing literature focuses strictly on math
and science courses, fails to control for other influences on achievement and course-taking, and
relies on now-dated samples of high school students. In addition, few studies closely examine the
relative performance returns to courses in different subjects and even fewer examine the
determinants of the variation in course-taking effects across subgroups of students and across high
schools with different characteristics. As a result, we know very little about which courses across the
high school curriculum improve student learning most, whether all students benefit equally from
their coursework, and the characteristics of the high schools that appear to produce the largest
course-taking effects.
Using panel data from several recent cohorts of 8th grade public school students in the state
of Florida, we examine the effects of course-taking in 9th and 10th grade on 10th grade test scores
and high school graduation. We further evaluate the link between high school course-taking and
entry into 2-year and 4-year postsecondary institutions. In estimating these effects, we control for
8th grade test scores, high school course-availability, and student, school, and neighborhood
characteristics. We also improve our estimates by using propensity score matching and instrumental
variables.
3
2. Prior Research
The prior empirical work on course-taking effects includes studies that focus on the effects
of tracking students into different sections of the same subject according to their prior preparation,
and others that focus on the effect of the number, level (e.g. Advanced Placement), and subject (e.g.
Algebra I) of courses in a given area, irrespective of the assigned track. For simplicity, we combine
the literatures since both address the underlying effects of course-taking. Among those studies that
have controlled for prior achievement, the relationship between the number and level of courses a
student takes in a given subject or the overall intensity of the curriculum and her proficiency on tests
of achievement in that subject have been relatively strong (Gamoran, 1987; Byrk, Lee, & Smith,
1990; Cool & Keith, 1991; Stevenson et al., 1994; Rock & Pollack, 1995; NCES, 1997; Schneider et
al., 1998; Attewell & Domina, 2008). Most of this work has concentrated on mathematics and
science course-taking. For instance, using the National Educational Longitudinal Survey (NELS), a
panel study of 8th graders in 1988, NCES (1997) shows a strong correlation between the number of
science courses taken and gains in science proficiency from the 8th to the 12th grade. Using the High
School and Beyond (HSB) cohort of 10th graders in 1980, Gamoran (1987) examines the effect of
the number of courses and number of advanced courses on 10th to 12th grade gains on tests in math,
science, reading, vocabulary, writing, and civics. Gamoran finds relatively strong effects of the
number of courses, and particularly the number of advanced courses, on gains in all subject areas,
controlling for other student and school inputs. The influence of math courses is particularly strong
and is the only effect that holds when he examines the relationship between course-taking and
achievement separately for students who scored below average in math on the prior exam.
Schneider et al. (1998) are among the few to explore the link between course-taking and high
school graduation. Using the NELS, Schneider and her colleagues find that students who take more
advanced mathematics courses in their last two years of high school are more likely to complete high
4
school. Other research has found a positive correlation between course-taking and achievement on
college admissions examinations, such as the Scholastic Achievement Test (SAT) or the ACT
(Sebring, 1987; Schneider, 2003; ACT, 2005; Attewell & Domina, 2008). And a few others find that
students who take more AP courses in high school—often, courses that may be used for college
credit—are more likely to attend college (e.g. Schneider et al., 1998; NCES, 2000). Indeed, taking
AP courses and receiving high grades in the course along with a pass on the AP exam is becoming
something of a prerequisite to gain admissions into several selective universities (Geiser & Santelices,
2004). In two of our previous papers using the Florida data, we also find that students who take
higher level math courses are less likely to require remediation in college (Long, Iatarola, & Conger,
2009) and that students who take Advanced Placement courses earn higher grades in college
(Conger & Long, 2009).
Collectively, the prior research suggests there may be a strong effect of course-taking on
high school achievement and college entry that varies by subject matter and by student, and that
appears to hold under controls for prior achievement. Yet we are uncertain as to the strength of
these relationships and the effects of course-taking on other educational outcomes because of the
following limitations of the existing work. First, the overwhelming majority of the research on
course-taking relies on two surveys of students conducted more than 17 years ago: the NELS and
the HSB. These surveys are limited as they are restricted to a single cohort of high school students
and have modest numbers of minority and low-income students. Since our data is a census of
Florida students, which includes a high number of minority and low-income students, we will be
able to answer more nuanced questions about these groups than is possible with the HSB and NELS
datasets. In addition, significant changes in high school curriculum and graduation requirements in
the past 20 years may have affected the relationship between high school courses and educational
outcomes. Given that most of the work has relied on national samples, little is known about course-
5
taking patterns and effects within a single state, where minimum graduation requirements, course
classification systems, and state-level education policies are the same for all schools and students.
Second, most of the prior research has focused narrowly on the determinants and effects of math
and science course-taking, to the neglect of other potentially important subjects, such as English,
foreign-languages, and history as well as the tradeoffs between the various subjects. Third, most
research focuses on achievement gains among students who graduate from high school and, further,
on the effects of the last two years of high school coursework. Less attention has been paid to the
first two years of coursework, a potentially crucial time, or to the relationship between courses in
these early years and high school dropout or post secondary enrollment. Fourth, the prior literature
has not sufficiently examined how course-taking effects vary across high schools with different
characteristics and, thus, we know little about the schools that produce the largest benefits. Finally,
most prior studies rely on severely under-specified models that fail to address biases on estimated
coefficients due to unobservables or to non-linear effects of observables.
3. Model
Our basic model is the following:
(1) Yim = β0 + β 1Ci + β 2Ti + β 3Ai + β 4Di + β 5Ei + νim + εi,
where Yim is the outcome of interest, Ci is a vector of credits earned in specific courses (or in a single
course), Ti is the total number of credits earned, Ai is a vector of 8th grade test scores, Di is a vector
of demographic characteristics, Ei is a vector of educational need characteristics, and νim is 8th grade
campus fixed effects. This specification allows us to test whether the hypothesis that taking specific
courses raises or lowers academic achievement/attainment, controlling for total credits earned. That
is, we explicitly test whether a shift in the composition of courses affects outcomes.
6
One should be concerned with the possibility of a selection effect: that is, the possibility that
those students who choose to take particular courses would have higher values of Y even if they had
not taken the course. The variables A, D, E, and ν are included in the specification in an attempt to
minimize the possibility of omitted variable bias. Student's ability, demographics, educational needs,
and 8th grade campus (and it's associated neighborhood) are likely to both influence the student's
course-taking and the student's educational outcomes. Even with these controls, there may be two
remaining sources of bias..
The first is the possibility that the linear specification may not adequately control for these
observable characteristics, and that we may be using students with low probability of taking
particular courses to identify the effects for students who did take the courses. We address this
possibility using propensity score matching. We first estimate Equation (2) using a probit
specification for our binary course-taking measures:
(2) Probability(Ci=1) = Φ(γ0 + γ 1Ti + γ 2Ai + γ 3Di + γ 4Ei + εi).
We estimate this specification for each 8th grade campus then predict the propensity to take the
course for each student based on the results for that student's 8th grade campus.1 We then
construct a sample by matching each "treated" student (i.e., a student who took the course) to an
"untreated" student who had a similar propensity to take the course but who did not take the
course.2 We impose a common support restriction by dropping treated students whose propensity
to take the course was greater than any untreated student, and dropping untreated students whose
propensity to take the course was less than any treated student. The average effect of the treatment
on the treated is given by Equation (3) and is based on a difference of mean outcomes for the
treated and untreated matched samples: 1 If everyone in the 8th grade campus took the course, we set the predicted propensity to one, and if no one took the course, we set the predicted propensity to zero. 2 In subsequent drafts of this paper, we will replace nearest neighbor matching with kernel matching. This nearest neighbor matching is done with replacement.
7
(3) Estimated treatment effect = mean(YTreated Students) - mean(YMatched Untreated Students).
Standard errors for this estimated treatment effect are estimated by computing the standard error
across 50 estimates derived from 50 bootstrapped samples.3
We estimate the treatment effect using students from the entire state of Florida, and then
repeat the estimation for each high school in the state. By estimating the effects for each high
school, we can examine the variation in effects across the high schools, and determine the school-
level predictors of the variation with the following regression:
(4) Estimated treatment effecth = δ0 + δ1Sh + εh,
where Sh is a vector of characteristics of high school h. Since the dependent variable in Equation (4)
is an estimated value with a non-constant level of variance, each high school level observation must
be weighted appropriately when estimating Equation (4). The technique for producing these
weights is described in Hanushek (1974) and Lewis and Linzer (2005).4
Returning to the first-stage individual-level regression, our second possible source of bias is
due to unobserved characteristics that influence a student's course-taking and academic outcomes,
and thereby bias the estimation of the treatment effect (for both the regression and propensity score
matching techniques). Since experiments are not available to estimate causal course-taking effects,
we need an instrumental variable to identify the effects non-experimentally. Of course, as in most
settings, finding a suitable instrument is challenging. We use the share of a high school's teachers
that are certified in various subjects to instrument for the likelihood that a student takes a course in a
3 In subsequent drafts of this paper, we will increase the number of bootstrapped samples to more accurately estimated the standard errors. 4 The authors would like to thank Drew Linzer for sharing their code for producing these weights. An alternative approach to the one described above would be to combine Equations 1 and 4 using a multilevel model (e.g. random effects, hierarchical linear modeling) to estimate the coefficients. The downside of using a multilevel model would be that we would need to sacrifice the propensity score estimation technique in the first-stage estimation. Since there may be substantial gains in using propensity score estimation over a linear first-stage model, and since our primary interest is the first-stage results, we prefer the two-stage process that we are using. We have also sufficiently corrected the standard errors in the second stage model so the results should be similar to what we would have gotten had we combined to the two levels.
8
particular subject. For this specification, we include high school fixed effects. Thus, the
instrumental variable identifies how changes in the composition of a high school's teaching staff
changes their students' course-taking patterns. For this IV strategy to work, requires two
assumptions: 1) that the composition of the teaching staff does not directly affect student outcomes
except through the effect on students' course-taking, and 2) that parents do not change their selected
high school for their child in response to changes in the composition of the teachers. Neither of
these assumptions is likely to hold perfectly. Thus, the utility of this modeling strategy rests on the
extent to which there is residual correlation in teacher composition and student outcomes (net of the
effect via course-taking) and the extent of omitted variable bias in the estimates given in Equations
(1) and (3).
The IV approach is as follows. First, estimate Equation (5):
(5) Ci = ρ0 + ρ1Vh + ρ2Ti + ρ3Ai + ρ4Di + ρ5Ei + ηih + τi + εi,
where Vh is a vector denoting the share of a high school's teachers credentialed in math, English,
science, social studies, and foreign language, ηim is high school fixed effects, and τi is year fixed
effects. Second predict the likelihood of the student taking each course; modify Equation (1) by
replacing Ci with a vector of predicted course-taking variables, replacing the middle school fixed
effects with high school fixed effects, and adding time fixed effects; and then re-estimate Equation
(1).5
4. Description of Data
Our data have been provided by the Florida Department of Education (FDOE). For this
draft version of the paper, we rely on one cohort of students (later versions will include later
5 While we have described this as a two-stage procedure, we actually perform the estimation in one step (using "ivregress" in Stata) so as to produce the correct standard errors.
9
cohorts). To construct our analytic cohort, we begin with all students who were enrolled in the 8th
grade in 1998-99, along with students who entered the cohort in subsequent years and grades and
who would be on track to graduate in 2003-04 assuming normal progression (those students first
observed in the 9th grade in 1999-2000, 10th grade in 2000-01, and so on). For our analyses of the
effects of course-taking through the 10th grade, we use data on the 128,611 students who were
observed in the 9th and 10th grades. For our analyses of completed high school course-taking, we use
data on the 106,709 students who were observed in at least three high school grades and who
received a high school diploma or general equivalency degree (GED) within four years of entering
school. Restricting the sample to students who are observed in at least three grades minimizes the
measurement error that would be attributed to making assumptions about the high school course-
taking patterns of students that we do not observe for their entire high school careers. It also
prevents us from confounding the effects of high school dropout and limited high school course-
taking.
The Florida data contains detailed records of students’ high school course-taking. For each
student, the data record the course code and name as well as the term in which the course was taken
and the number of credits the student earned in the course. All schools in the state adhere to a
common course code with the FDOE maintaining a course code directory of authorized courses
along with course descriptions that allow us to determine the subject of each course. The FDOE
also provides a classification scheme that identifies whether a course is at level one, two, or three,
where the higher number indicates greater difficulty. For instance, in the subject of Algebra, "Pre-
Algebra" is a Level 1 course, "Algebra 1" is a Level 2 course, and "Honors Algebra 1" is a Level 3
course. Only very basic or remedial courses are identified as Level 1; for instance "Consumer
Mathematics" and "English Skills 1" are Level 1 courses. For our models, we use various measures
of course-taking, including total credits earned, the number of credits earned by subject, the highest
10
math course taken (1= no math course to 13 = calculus), whether the student completed Algebra 1,
whether the student completed a Level 3 math course, and whether the student took an Advanced
Placement (AP) or International Baccalaureate (IB) course prior to graduation.
Table 1 provides descriptive statistics on our two samples of students. Consistent with at least
some estimates (see Laird, Lew, DeBell, and Chapman, 2006), only 73% of the students in our
cohort (who were observed in both 9th and 10th grade) graduate from high school on time. A fairly
large percentage (59% of all students observed in both 9th and 10th grade and 71% of on-time high
school graduates) pursue a higher degree, with a slightly higher share going on to a 4-year versus a 2-
year institution. The 10th grade achievement scores are standardized with a mean of zero and a
standard deviation of one.6 The middle of the table provides the distribution on our course-taking
measures with the average student earning approximately 10 credits by the 10th grade and 26 credits
by graduation. By the 10th grade, the mean student has completed a Level 1 course or a Level 2
course other than Algebra 1 (the value of 5 on our highest math course measure) but not yet Algebra
1 (the value of 6). By the 12th grade, the mean graduate has completed Geometry (the value of 7)
but not Algebra 2 (the value of 8). Forty-three percent of the graduates complete a Level 3 course
upon graduation. In both samples, students average slightly higher credits in English than math and
science, reflecting Florida's high school graduation requirements (4 credits required in English, 3
credits in math, and 3 credits in Science). The majority of the credits are earned in high school level
courses- Level 2 and 3—with few students requiring remediation at the high school level.
At the bottom of the panel, we provide descriptive statistics on the model covariates, which
include 8th grade Florida Comprehensive Assessment Test (FCAT) scores in reading and math along
6 Since these scores are standardized among all test-takers, the means for all test-takers equal 0.00. However, the mean 10th grade test score is 0.09 for both reading and math among our sample of students observed in both 9th and 10th grades. This suggests that our sample has somewhat higher achievement than the full sample of 10th graders. This result is not surprising as our sample consists of students with more stability than the full sample, which includes students who move into and out of Florida's public schools.
11
with student socio-demographic and educational characteristics. Approximately 31% of the 9th and
10th grade sample and 26% of the graduate sample is missing data on one or more right-hand side
variables.7 To improve efficiency and to avoid sampling bias, we impute the missing values using
multiple imputation by chained equations (Rubin, 1987; Royston, 2004) creating five multiply-
imputed datasets.8 The estimated effects resulting from the five imputed datasets are averaged and
standard errors are adjusted to account for uncertainty caused by imputation.
Finally, the 8th grade campus fixed effects and the student's main high school were
determined using only "regular" campuses, excluding juvenile justice facilities and other alternative
institutions. If the student attended multiple "regular" campuses, we select the campus with the
most terms of enrollment (or select randomly amongst the campuses tied for most terms of
enrollment). Students who were never enrolled in a "regular" 8th grade campus were assigned into
one group for the purpose of computing the 8th grade fixed effect.
5. Results
5a. Effects of 9th and 10th grade course-taking
OLS Results
Table 2 presents the OLS and Probit results for the effects of 9th and 10th grade course-
taking. Controlling for total credits earned through the 10th grade, increasing the highest math
course taken by one level raises the likelihood of graduation from high school by 3.3 percentage
points, raises the likelihood of attending a four-year college by 6.0 percentage points, and raises 10th
grade math and reading FCAT scores by 0.10 and 0.06 standard deviations. Since it is somewhat
challenging to interpret the meaning of "the highest math course taken by one level", in the second
7 8th grade test scores are missing for roughly one-fifth of the students in both samples. 8 For a relatively small amount of missing information in the data there is little efficiency gains to be made with more than five imputations of the data (Rubin, 1987).
12
panel we use a discrete measure: "Completed Algebra 1 or higher by 10th grade". These results are
comparable; completing Algebra 1 or higher raises the likelihood of graduation from high school by
10.0 percentage points, raises the likelihood of attending a four-year college by 9.7 percentage points,
and raises 10th grade math and reading FCAT scores by 0.20 and 0.11 standard deviations. Note
that Florida requires students to complete Algebra 1 or an equivalent course in order to earn a diploma.
Thus, the fact that completion of Algebra 1 is not perfectly correlated with graduation suggests that
these alternative pathways to graduation are taken by many students.
In the next panel, we consider the relative effects of credits in various subjects. Since we are
controlling for total credits earned we can interpret the effects on each subject as the effect of
replacing a credit in a "non-listed" subject (e.g., art, physical education, home economics, etc.) with a
credit in that particular subject. For example, if one were to replace a "non-listed" credit with a
credit in math earned by the end of the 10th grade, the student's likelihood of earning an on-time
high school diploma would increase by 3.6 percentage points. For the most part, the listed subjects
have positive effects on these outcomes but there are some notable differences in the magnitudes of
these effects.
First, the largest effects are found for credits in foreign languages, which raise the likelihood
of graduation from high school by 8.3 percentage points, raises the likelihood of attending a four-
year college by 10.6 percentage points, and raises 10th grade math and reading FCAT scores by 0.10
and 0.08 standard deviations.
Second, credits earned in English and social studies are negatively related to receiving a high
school diploma and credits earned in English are negatively related to attending a four-year college
relative to credits earned in non-listed subjects. On the other hand, relative to credits in non-listed
subjects, credits earned in English and social studies raise the likelihood of attending a two-year
college and raise 10th grade FCAT math and reading scores. In fact, more credits in English are
13
positively correlated with 10th grade math scores, yet more credits in math is uncorrelated with 10th
grade reading scores. Thus, except for 10th grade reading scores, math credits may lead to greater
returns than English credits.
Third, controlling for credits in the five listed subjects, a student who earns more credits in
the non-listed subjects (which tend to be in non-academic subjects9) are slightly more likely to
graduate high school, but less likely to attend a postsecondary institution and earn slightly lower 10th
grade FCAT scores. That is, amassing credits in non-academic subjects may help the student attain
the credits needed to complete high school, but these non-academic subjects have negative effects
on subsequent academic outcomes.
The bottom-panel of Table 2 repeats the analysis in the third-panel, but focuses on credits in
Level-2 or Level-3 courses only (thus not including credits in remedial courses). The results are
largely the same as those in the third-panel.
Propensity Score Matching Results: Discrete Treatment
Table 3 repeats the analysis for the effect of completing Algebra 1 or higher by 10th grade
using propensity score matching (PSM). Essentially, this method creates a subsample consisting of
all students who completed Algebra 1 or higher, and matches these students to "non-treated"
students who had similar probabilities of completing Algebra 1 or higher, based on their 8th grade
test scores, 8th grade campuses, and other individual characteristics. We then compute the
difference in the means of the outcomes for the "treated" and "untreated" students.
The PSM results show strikingly larger effects. Completing Algebra 1 or higher by 10th
grade is associated with a 27.3 percentage point increase in the likelihood of receiving an on-time
9 The definition of what should be counted as an "academic" subject varies. " The No Child Left Behind Act of 2001 defines "core academic subjects" as: English, reading or language arts, mathematics, science, foreign languages, civics and government, economics, arts, history, geography." (http://ritter.tea.state.tx.us/nclb/hottopics/def.core.acad.subj.html). With the exception of "arts", these subjects are covered in our five listed subjects.
14
high school diploma, a 28.0 percentage point increase in the likelihood of attending a four-year
college, and raises 10th grade math and reading FCAT scores by 1.05 and 0.97 standard deviations.
There are two possible explanations for the higher magnitudes of the estimates. First, is the
possibility that the linear specification was inadequate as it included students with low probabilities
of taking higher-level math courses in the counterfactual. Thus, the semi-parametric PSM estimate
with the "common support" requirement produces more accurate estimates of the treatment effect.
The other possibility is that the propensity score matching process has not successfully matched the
treated and untreated students on their covariates. We have not yet tested whether our first-stage
probit predicting the likelihood of completing Algebra 1 or higher satisfies the "balancing property"
(i.e., that the means of the covariates in the treated and untreated samples are roughly equivalent).
We will test the balancing of the samples in the next draft of the paper.
Propensity Score Matching Results: Continuous Treatment
Since the discrete propensity score matching results produced such interesting results, we are
now exploring using a propensity score matching technique to evaluate the effects of a one unit
increase in credits in each subject, controlling for credits taken in other subjects. The goal is
produce propensity score matching results that correspond to the results shown in the third panel of
Table 2. This section of the paper describes very preliminary results using a 5% sample of all
Florida students which does not yet incorporate 8th grade campus fixed effects.
Hirano and Imbens (2004) introduce the methodology and theory behind propensity score
matching with continuous (as opposed to binary) treatments. To estimate the "dose-response" and
"treatment effect" functions, we first estimate the conditional distribution of the treatment variable
(the number of credits taken in a particular subject) given the covariates. The treatment ("dose") is
the number of credits earned in one of the following subjects: math, english, social studies, science,
or foreign language. Here the covariates included standardized math and reading test scores in 8th
15
grade, demographic variables (race, free or reduced price lunch recipient, sex, age), educational
demographic variables (lep and exceptional), and total credits earned by 10th grade. Additionally, we
also included credits earned in the other named subjects (math, English, social studies, science, and
foreign language, with the subject of interest omitted).
Bai and Mattei (2007) wrote a procedure for Stata (doseresponse.ado) which follows the
work of Hirano and Imbens (2004). We employe this program to estimate the dose-response and
treatment effect functions. Following Hirano and Imbens (2004) and Bai and Mattei (2007), we
assume the treatment to be distributed normally given the covariates and estimate the generalized
propensity score (GPS). The procedure tests the normality of the distribution of the treatment
given covariates and the balancing property, which is done by dividing the treatment into quartiles.
We find that the balancing property is satisfied below the 0.01 level, while the normality assumption
is not statistically satisfied at the 0.05 level. Appendix Figures 1a-1e plot the distribution of the
treatment variables along with the normal distribution. These provide a visual indication of the
distribution of the treatment, although these distributions are not conditional on the covariates.
While the distributions are not normally distributed, they are not grossly non-normal (with the
exception of foreign languages).
We then estimated the dose-response function and the treatment effect function by
estimating the expected outcome given the GPS, the treatment, and the interaction of these variables.
The outcomes of interest were the standardized math and reading FCAT scores for 10th grade,
whether an individual graduated high school within 4 years, and whether an individual attended a (1)
4-year college within 5 years, (2) 2-year college within 5 years, and (3) any college within 5 years.
The average expected outcome is then estimated for each level of the treatment.
To calculate the treatment effect, we estimate the expected outcome at each level if the
number of credits was increased by one (1 additional year of credits) and subtract the expected
16
outcome with the shift (i.e. the dose-response). Figures 1a-1e plot the dose-response and treatment
effect functions for each treatment (i.e. subject) and the outcome of graduating high school within
4-years ("hsd4") . The plot on the left-side of Figure 1a gives the predicted dose response function,
and can be interpreted as the expected probability of on-time graduation for an individual in a given
treatment level—i.e. with a given number of math credits taken by 10th grade. For example, an
individual with about 2 credits in math by 10th grade would have approximately 0.78 probability of
graduating high school in four years. That probability would increase to about 0.87 with 7.75 credits
in math. The plot on the right-side of Figure 1a gives the predicted treatment effect function, and
provides an estimate of what would happen if an individual in a given treatment level increased his
or her number of credits in math by 10th grade by 1. Thus, the treatment effect gives the change in
the expected outcome given a change in the treatment of 1 (roughly speaking, this is a plot of the
slope of the curve shown in the left-panels). For example, an individual with 1 credit in math by 10th
grade would see an approximately 7 percentage point increase in his or her expected probability of
graduating within four years if they increase the number of credits in math to 2 credits by 10th grade,
but increasing from 2 to 3 credits would lower the expected probability of graduating within four
years by nearly the same amount. In other words, if we look at the dose-response function on the
left, we can see that the probability of graduating would drop from approximately 0.78 to about 0.72
with an additional year of math before 10th grade for those at 2 credit treatment level. An individual
with 6 credits of math, however, would see about 3 percentage point increase in his or her
probability of graduating within 4 years if they took an additional year of math by 10th grade
increasing from a probability of about 0.81 to 0.84 again referring to the dose response function on
the left. The other figures can be interpreted similarly.
While these results are very preliminary, they suggest some interesting patterns. From Figure
1a, we see the marginal effect of math credits on high school graduation is mostly positive except for
17
students with 2 math credits. At this point, we can only speculate on why the marginal effect is
negative at 2 credits of math. First, it could be the case that the sample of individuals with three
credits of math by 10th grade contain a disproportionate share of students who were retained in 9th
or 10th grade (and thus had three years to take courses in the 9th and 10th grade). Students who
were retained are likely to be more prone to not graduate high school within four years of high
school entry. A second hypthosesis is that moving from the second to third credit in math leads to
an imbalance in the credits earned that makes it more difficult to graduate. For example, an extra
math course could displace an English credit needed for high school graduation. (However, one
would expect the displacement effect to be greater still with an increase from three to four credits).
In Figure 1b, we see that the first two credits in English raise the likelihood of high school
graduation, while the third and fourth credits in English lower the likelihood of graduation. The
dose response function shown in Figure 1c for science credits mirrors Figure 1a for math credits;
additional credits in science raise the likelihood of high school graduation, except for the third credit.
Likewise, the dose response function shown in Figure 1d for social studies and Figure 1e for foreign
language mirrors Figure 1b for English credits; the first 1.5 credits in social studies and foreign
language raise the likelihood of high school graduation, while subsequent credits lower the likelihood
of high school graduation.
The next set of figures (2a-2e) repeat the analysis with 10th grade math FCAT scores as the
dependent variable. Unlike high school graduation, this outcome doesn't mechanically depend on
the composition of credits earned. Nonetheless, we still see the same patterns (in general), with
inflections in the effect at two credits. Thus, it appears that taking a balanced set of classes has effects
on both math acheivement and high school graduation. (In the next draft, we will attempt to
disentangle the issue of student retention on these outcomes).
18
Finally, note that Figures 2a-2e show positive effects of the first 1.5 credits in all subjects on
10th grade math FCAT scores. These results again appear to confirm that 10th grade math scores
do not simply depend on math course-taking.
Instrumental Variable Results
IV estimates are not yet computed, but will be included in the next draft of the paper.
5b. Effects of all high school course taking
OLS Results
Table 4 shows the effects of complete high school course-taking (not just courses through the
10th grade) on post-secondary entry for the sample of students who were observed in at least three
high school grades and who completed a high school diploma or GED within four years of high
school entry.10
There is a strong association of earning credits in advanced math and AP/IB classes and
attending any college or a four-year college, while credits earned in advanced math and AP/IB
classes are negatively associated with attending a two-year college. We view these results as more
likely to be reflecting selection on the part of the student than a causal effect of course-taking on
postsecondary entry. That is, students who take advanced math or AP/IB courses are likely doing
so with the understanding that these courses will improve their postsecondary enrollment prospects.
The fourth panel of Table 4 is potentially more interesting. There is a strong correlation
between taking science credits in high school and attending a 4-year institution. Among the five
listed subjects, additional credits in English have the lowest association with attending a four-year
institution (although the association is still positive). Each additional credit in the non-listed (mostly
non-academic) subjects lowers the likelihood of attending a four-year institution by 1.5 percentage
10 98.5% of this sample earned a high school diploma.
19
points. Thus, again, amassing credits in these non-listed subjects may increase the likelihood of the
student graduating high school, but appears to lessen the chance of the student attending a four-year
college.
5c. Variation in the Effects Across Students, by Race, Gender, and Poverty
We now explore whether the relationships between course-taking and outcomes vary across
demographic subgroups (white, Hispanic, black, poor, non-poor, female, and male). We repeat the
specification shown in the third panel for Table 2, but restrict the sample to one demographic
group at a time. We then compute the marginal effects and evaluate whether the marginal effects are
larger or smaller for "disadvantaged" groups (black, Hispanic, poor, and male) relative to their
reference group (whites, non-poor, or female students).11
The results are shown in Tables 5a and 5b. For the most part, the patterns of effects by
course type are maintained for various student subgroups. Where there are significant differences
between the effects by subgroup, the effects are slightly more favorable for disadvantaged students.
Ignoring the 4-year and 2-year college results, there are 29 significant positive differences, 17
significant negative differences, and 58 insignificant differences.
There are some interesting patterns in these results that deserve attention. First, additional
credits in English have a more positive effect on 10th grade math and reading scores for blacks,
Hispanics, poor students, and males. But, additional credits in English significantly lowers the
chances of black, Hispanic, and poor students from receiving a high school diploma, while having
insignificant effects on the likelihood of enrollment for whites and non-poor students. This result
11 For the dichotomous dependent variables a difference in the marginal effect between two demographic groups could come as a result of either a difference in the effect of the course on the outcome or from the demographic groups having different base likelihoods. For example, the marginal effect of taking an additional math course on a high school diploma is dHSD/dMath = φ(X'β)*βMath. Thus, if the demographic groups have different values of φ(X'β), they would have different marginal effects even with the same value of βMath.
20
suggests that there could be a tension between meeting the Adequate Year Progress requirements of
No Child Left Behind and promoting graduation for these subgroups.
Other results are noteworthy. Additional credits in social studies tend to have positive
effects on disadvantaged students, with significant positive differences for blacks on 10th grade
reading, Hispanics on HS diploma, poor students on 10th grade reading and math and
PSE entry, and males on 10th grade reading. Additional credits in science have larger positive
effects on 10th grade math scores for females than for males. Finally, and not surprisingly,
additional credits in foreign languages have significantly smaller effects for Hispanics than for whites.
5d. Variation in the Effects Across High Schools, by School Characteristics
To get a sense of how these effects of course-taking vary across high schools, we
recomputed the propensity score estimates for the effect of completing Algebra 1 or higher that are
shown in Table 3 for each of 333 high schools in the state of Florida. These estimates differ from
those in Table 3 in that the nearest neighbor matching is computed within high schools, rather than
within the entire state of Florida. (The first-stage propensity score estimates are computed within
each 8th grade campus, as was done previously).
Figure 3 plots the distribution of the effect size estimates across the 333 high schools. The
peaks of the density estimates roughly correspond to the state-wide estimates shown in Table 3 (i.e.,
roughly 0.25 for high school diploma and attending a four-year college, and 1.1 for 1.0 for math and
reading FCAT scores). Each of the distribution is uni-peaked -- yet there is a large degree of
variation in the effects. At the high school-level, the effects of competing Algebra 1 or higher are
generally positive for each of the outcomes. However, the effects of competing Algebra 1 or higher
are negative for attending a two-year college for a sizable share of high schools (and positive and
large for another group of high schools). This variation in the magnitude of the effects motivates
21
our investigation of the relationship between school characteristics and the school-level effect
magnitudes.
Before turning to regression estimates, Figure 4 shows the bivariate relationships between
school characteristics and the effect of completing Algebra 1 or higher by 10th grade on attending a
four-year college. The effects tend to be higher in schools with more enrollment, fewer poor
students, yet lower resources (i.e., lower expenditures per pupil, fewer teachers per pupil, lower share
of teachers with advanced degrees, and non-magnet schools). In the regressions that follow, we
then test whether these bivariate associations are maintained when controlling for the other school
characteristics.
Table 6 shows the regression results. In general (but depending on the outcome evaluated),
the bivariate results are maintained. Most interestingly, the positive effects of taking more advanced
math on attending a four-year college are muted in schools with more poor students (i.e., schools
with a higher share of their students receiving free or reduced-price lunch). Whereas, the effect of
taking more advanced math on attending a two-year college (which is negative for some schools)
tends to be more positive if the school has more poor students. These differential effects could be
due to differences in the quality of the advanced courses in high- and low-poverty schools or to
other resource differentials proxied by the poverty of the students in the school, such as the college
guidance counselors. Another interpretation is that the financial limitations of students influences
the direction of the effects of advanced course-taking; in other words, the share of poor students in
the school proxies for traits of the students themselves that we have not sufficiently held constant
with student poverty level.
Similar interpretations may apply to the somewhat counterintuitive effects of school
resources. The effects of advanced course-taking on most of these outcomes tend to be lower in
schools with high resources. For instance, the effect of advanced courses tends to be lower in
22
schools with higher teacher-pupil ratios. There may be ceiling effects such that students in these
schools are likely to have parents with higher incomes and are already quite likely to attend college.
Thus, the effects of course-taking on these advantaged students is lower.
We previously found strong effects of magnet school attendance on the likelihood that the
student would take Level-3 courses and AP/IB courses (Conger, Long, & Iatarola, 2009). It is
possible that these magnet schools are pushing students into courses where they are not fully
prepared, and this lessens the effect of the course on their subsequent outcomes. On the other hand,
in Long, Iatarola, and Conger (2009) we found sizable effects of highest math course taken on
reducing the need for college math remediation even for students with lower math scores.
Nonetheless, as shown in Table 6, we find that schools with more prepared students (i.e., with
higher 8th grade math FCAT scores) show larger effects of advanced math course-taking on 10th
grade FCAT scores and the likelihood of going to a four-year college. Thus, the cumulative
evidence seems to show that the course-taking effects are larger when students are more prepared,
and in schools that are less likely to push students with less preparation into the advanced courses.
6. Conclusions
This paper aims to fully investigate the effects of high school course-taking on student
achievement and attainment. This first draft provides an overview of the scope of the analysis: we
use several measures of course-taking, examine outcomes at the secondary and post-secondary level,
explore variation in the effects of course-taking across subgroups of students and the schools they
attend, and address specification errors common to most prior studies.
The preliminary analyses presented here offer several findings that will be investigated
further. First, and perhaps unsurprisingly, obtaining more credits and higher-level credits in high
school associates with higher levels of achievement and attainment. Second, and more interestingly,
23
the type of credits earned matters: a) foreign-language courses are more positively correlated with
outcomes than any other type of course; b) except for 10th grade reading scores, math credits may
lead to greater educational performance returns than English credits; c) amassing credits in non-
academic subjects may help the student attain the credits needed to complete high school (in fact,
helps even more than taking English credits), but these non-academic subjects may have negative
effects on subsequent academic outcomes; and d) there is a strong positive association between
taking science courses and attending a 4-year post-secondary institution. Third, most of the
relationships between course-taking and outcomes hold across demographic groups, and where
there are significant differences the differences tend to favor disadvantaged students (blacks,
Hispanics, males, and poor students). Finally, the effects of courses matter differently according to
the characteristics of the high school students attend. In general, we find that the effects of
advanced courses are smaller in schools with somewhat greater resources, including fewer poor
students, fewer students overall, and higher teacher-pupil ratios.
We caution the reader that these analyses are very preliminary. However, they suggest some
important areas to pursue further including the tradeoff between credits in different subjects and the
differential effects of these tradeoffs for different kinds of schools. Stay tuned.
24
References
Altonji, J.G. (1995). The effects of high school curriculum on education and labor market outcomes. The Journal of Human Resources, 30(3), 409-438.
Argys, L. M., Rees, D. I., & Brewer, D. J. (1996). Detracking America’s schools: Equity at zero cost? Journal of Policy Analysis and Management, 15(4), 623-645.
Attewell, P. & Domina, T. (2008). Raising the bar: Curricular intensity and academic performance. Educational Evaluation and Policy Analysis, 30(1), 51-71.
Betts, J. R., & Shkolnik, J. L. (2000). The effects of ability grouping on student achievement and resource allocation in secondary school. Economics of Education Review, 19, 1-15.
Bia, M., & Mattei, A. (2008). A Stata package for the estimation of the dose–response function through adjustment for the generalized propensity score. The Stata Journal, 8(3), 354–373.
Bryk, A., Lee, V., & Smith, J. (1990). High school organization and its effects on teachers and students: an interpretive summary of the research. Choice and Control in American Education, 1, 135- 227.
Catsambis, S. (1994). The path to math: Gender and racial-ethnic differences in mathematics participation from middle school to high school. Sociology of Education, 67(3), 199-215.
Cool, V.A. & Keith, T.Z. (1991). Testing a model of school learning: Direct and indirect effects on academic achievement. Contemporary Educational Psychology, 16(1), 28-44.
Conger, D., Long, M.C., & Iatarola, P. (2009). Explaining Race, Poverty, and Gender Disparities in Advanced Course-taking. Working Paper.
Conger, D. & Long, M.C. (2009). Why are Men Falling Behind? Gender Gaps in College Performance and Persistence. Working Paper.
Dougherty, C., Mellor, L., & Jian, S. (2006). Orange juice or orange drink: Ensuring that “advanced courses” live up to their labels. National Center for Educational Accountability Policy Brief No. 1.
Figlio, D. N., & Page, M. E. (2002). School choice and the distributional effects of ability tracking: Does separation increase inequality? Journal of Urban Economics, 51(3), 497-514.Hanushek, Eric A. (1974). "Efficient Estimators for Regressing Regression Coefficients," The American Statistician, 28(2), pp. 66-67.
Gamoran, A. (1987). The stratification of high school learning opportunities. Sociology of Education, 60(3), 135-155.
Gamoran, A., & Mare, R. D. (1989). Secondary school tracking and educational inequality: Compensation, reinforcement, or neutrality? The American Journal of Sociology, 94(5), 1146-1183.
Geiser, S. & Santelices, V. (2004). The role of advanced placement and honors courses in college admissions. Center for Studies in Higher Education, University of California at Berkeley. Research & Occasional paper Series CSHE.4.04.
Heckman, J.J, & LaFontaine, P.A. (2008). The American high school graduation rate: Trends and levels. Institute for the Study of Labor Discussion Paper No. 3216.
Hirano, Keisuke and Imbens, Guido W. (2004), "The Propensity Score with Continuous Treatments." In Andrew Gelman and Xiao-Li Meng (eds.), Applied Bayesian Modeling and Causal Inference from Incomplete-Data Perspectives. West Sussex: John Wiley and Sons, 73-84.
Hoffer, T.B., Rasinski, K.A., & Moore, W. (1995). Social background differences in high school mathematics and science course-taking and achievement. (NCES 95-206). Washington DC: US Department of Education, National Center for Education Statistics.
Laird, J., Lew, S., DeBell, M., and Chapman, C. (2006). Dropout Rates in the United States: 2002 and 2003 (NCES 2006-062). U.S. Department of Education. Washington, DC: National Center for Education Statistics.
25
Lewis, Jeffrey B., and Linzer, Drew A. (2005). "Estimating Regression Models in Which the Dependent Variable Is Based on Estimates," Political Analysis, 13, pp. 345–364
Long, M.C., Iatarola, P., & Conger, D. (2009). Explaining gaps in readiness for college-level math: The role of high school courses. Education Finance and Policy, 4(1), 1-33.
National Center for Education Statistics. (2005). The condition of education. Washington DC: U.S. Government Printing Office (NCES 2005-094).
Planty, M. Provasnik, S., & Daniel, B. (2007). High School Coursetakings: Findings from the Condition of Education 2007 (NCES 2007-065). U.S. Department of Education. Washington, DC: National Center for Education Statistics.
Rock, D.A., & Pollack, J.M. (1995). The relationship between gains in achievement in mathematics and selected course taking behaviors. Washington DC: US Department of Education, National Center for Education Statistics.
Schneider, B. (2003). Strategies for success: High school and Beyond. In Brookings Papers on Education Policy, ed D. Ravitch, Washington DC: Brookings Institution Press, p. 55-94.
Schneider, B., Swanson, C. B., & Reigle-Crumb, C. (1998). Opportunities for learning: Course sequences and positional advantages. Social Psychology of Education, 2, 25-53.
Sebring, P.A. (1987). Consequences of differential amounts of high school course work: Will the new graduation requirements help? Educational Evaluation and Policy Analysis, 9, 258-273.
Slavin, R. E. (1990). Achievement effects of ability grouping in secondary schools: A best-evidence synthesis. Review of Educational Research, 60, 471-499.
Stevenson, D. L., Schiller, K. S., & Schneider, B. (1994). Sequences of opportunities for learning. Sociology of Education, 67(3), 184-198.
Royston P. (2004). Multiple imputation of missing values. Stata Journal, 4(3), 227-241. Rubin, D.B. (1987). Multiple Imputation for Nonresponse in Surveys. New York: John Wiley and Sons.
26
Table 1: Descriptive Statistics
Mean SD Mean SD
Outcome Variables Receive a High School Diploma Within Four Years of HS Entry 0.73Attended Any 4‐Year College Within Five Years of HS Entry 0.31 0.39Attended Any 2‐Year College (But No 4‐Year College) Within Five Years of HS Entry 0.29 0.32Attended Any College Within Five Years of HS Entry 0.59 0.7110th Grade Math FCAT Score 0.09 1.0410th Grade Reading FCAT Score 0.09 1.02
Course‐taking Variables Total Credits Earned 9.82 4.58 26.23 4.89Highest Math Level (1‐13, with 13 = Calculus) 5.38 2.05 7.80 2.42Completed Algebra 1 0.72Completed Level 3 Math Course 0.43Credits Earned in Math 1.51 0.79 3.75 1.08Credits Earned in English 1.54 0.87 4.41 1.40Credits Earned in Science 1.27 0.63 3.39 1.03Credits Earned in Social Studies 0.95 0.71 3.65 1.19Credits Earned in Foreign Language 0.53 0.68 1.71 1.15Level 2 or 3 Credits Earned in Math 1.27 0.74 3.40 1.16Level 2 or 3 Credits Earned in English 1.39 0.76 4.10 1.29Level 2 or 3 Credits Earned in Science 1.22 0.65 3.31 1.11Level 2 or 3 Credits Earned in Social Studies 0.95 0.70 3.56 1.16Level 2 or 3 Credits Earned in Foreign Language 0.54 0.69 1.70 1.15Credit Earned in Any Advanced Placement or International Baccalaureate Course 0.28
Control Variables 8th Grade Math FCAT Score 0.15 0.99 0.35 0.868th Grade Reading FCAT Score 0.15 0.99 0.35 0.87Student is White 0.58 0.60Student is Black 0.22 0.19Student is Hispanic 0.17 0.17Student is Asian American 0.02 0.02Student is Male 0.50 0.47Student's Age (as of July 1st, 2003) 18.50 0.53 18.41 0.47Student Received Free‐ or Reduced‐Price Lunch in 8th Grade or High School 0.44 0.38Student Classified as Limited English Proficient in 8th Grade or High School 0.14 0.13Student Classified as Having a Disability in 8th Grade or High School 0.15 0.11
Number of Observations 128,611 106,709
(Measured through the 10th grade for the first two columns)
Students observed in both 9th and 10th
Grade
Students observed in at least three of the four HS grades and who received a HS
diploma or GED within 4 years of HS entry
27
Table 2: Effects of 10th Grade Course Taking, as Estimated by OLS or Probit Regressions
Highest Math Level by 10th Grade 0.033 (0.001) 0.060 (0.001) ‐0.005 (0.001) 0.059 (0.001) 0.103 (0.001) 0.062 (0.002)
Total Credits Earned by 10th Grade 0.012 (0.000) ‐0.005 (0.000) ‐0.002 (0.000) ‐0.003 (0.000) ‐0.011 (0.001) ‐0.007 (0.001)
Completed Algebra 1 or Higher by 10th Grade 0.102 (0.004) 0.097 (0.004) 0.067 (0.004) 0.138 (0.004) 0.197 (0.005) 0.113 (0.007)
Total Credits Earned by 10th Grade 0.012 (0.000) 0.000 (0.000) ns ‐0.004 (0.000) 0.000 (0.000) ns ‐0.007 (0.001) ‐0.004 (0.001)
Credits Earned in Math by 10th Grade 0.036 (0.002) 0.007 (0.003) 0.011 (0.002) 0.018 (0.003) 0.054 (0.004) ‐0.004 (0.004) ns
Credits Earned in English by 10th Grade ‐0.035 (0.002) ‐0.014 (0.002) 0.025 (0.002) ‐0.009 (0.002) 0.048 (0.003) 0.048 (0.003)
Credits Earned in Science by 10th Grade 0.040 (0.003) 0.027 (0.003) 0.024 (0.003) 0.034 (0.004) 0.022 (0.004) 0.035 (0.004)
Credits Earned in Social Studies by 10th Grade ‐0.015 (0.003) 0.020 (0.003) 0.008 (0.002) 0.013 (0.003) 0.048 (0.004) 0.070 (0.004)
Credits Earned in Foreign Language by 10th Grade 0.083 (0.003) 0.106 (0.002) ‐0.023 (0.002) 0.125 (0.003) 0.100 (0.003) 0.077 (0.003)
Total Credits Earned by 10th Grade 0.008 (0.000) ‐0.008 (0.001) ‐0.008 (0.000) ‐0.008 (0.001) ‐0.021 (0.001) ‐0.015 (0.001)
Level 2 or 3 Credits Earned in Math by 10th Grade 0.036 (0.003) 0.028 (0.003) 0.036 (0.002) 0.041 (0.003) 0.095 (0.003) 0.037 (0.005)
Level 2 or 3 Credits Earned in English by 10th Grade ‐0.033 (0.003) ‐0.012 (0.003) 0.031 (0.002) ‐0.010 (0.003) 0.008 (0.003) 0.030 (0.004)
Level 2 or 3 Credits Earned in Science by 10th Grade 0.017 (0.003) 0.024 (0.003) 0.029 (0.003) 0.026 (0.004) 0.076 (0.005) 0.074 (0.005)
Level 2 or 3 Credits Earned in Social Studies by 10th Grade ‐0.019 (0.003) 0.020 (0.003) ‐0.006 (0.003) 0.003 (0.003) ns 0.026 (0.004) 0.051 (0.004)
Level 2 or 3 Credits Earned in Foreign Language by 10th Grade 0.084 (0.003) 0.105 (0.002) ‐0.028 (0.002) 0.123 (0.003) 0.086 (0.003) 0.068 (0.003)
Total Credits Earned by 10th Grade 0.012 (0.000) ‐0.010 (0.001) ‐0.009 (0.000) ‐0.006 (0.001) ‐0.019 (0.001) ‐0.016 (0.001)
Notes: 1) Each panel shows the results from a separate regression. Each regression controls for 8th grade test scores, demographic characteristics, educational needs, and 8th grade fixed effects. 2) Results for the indicator variables show the marginal effect computed at the mean of the independet variables. 3) Robust standard errors are shown in parentheses. 4) All results are significant, except where noted by "ns" for "not significant" or "ms" for "marginally significant" (i.e., 10% level).
Attended Any 2‐Year College (But No 4‐Year College)
Within Five Years of HS Entry
Attended Any 4‐Year College Within Five Years of HS Entry
Receive a High School Diploma Within Four Years of
HS Entry
Attended Any College Within Five Years of HS Entry
10th Grade Math FCAT Score 10th Grade Reading FCAT Score
28
Table 3: Effects of Completing Algebra 1 or Higher by 10th Grade, as Estimated by Propensity Score Matching
Effect 0.273 (0.037) 0.280 (0.028) 0.056 (0.035) 0.336 (0.038) 1.053 (0.076) 0.969 (0.081)
Notes: 1) Each panel shows the effect size from a separate propensity score estimation. Each student "treated" student (i.e. who took Algebra 1 by 10th grade) is matched with their nearest "untreated" neighbor (i.e. the student who has the most similar propensity to take Algebra 1, but who did not do so). Propensity scores are based on a first‐stage probit regression run for each 8th grade campus, which controlled for 8th grade test scores, demographic characteristics, and educational needs. 2) Standard errors are computed by the standard error across 100 random subsamples of the state of Florida. 3) All effects shown are significant, except where noted by "ns" for "not significant" or "ms" for "marginally significant" (i.e., 10% level).
Receive a High School Diploma
Within Four Years of HS Entry
Attended Any 4‐Year College Within Five Years of HS Entry
Attended Any 2‐Year College (But No 4‐Year College) Within Five Years of HS Entry
Attended Any College Within Five Years of
HS Entry
10th Grade Math FCAT Score
10th Grade Reading FCAT Score
29
Table 4: Effects of High School Course Taking, as Estimated by OLS or Probit Regressions
Highest Math Level 0.114 (0.001) ‐0.018 (0.001) 0.080 (0.001)Total Credits Earned 0.010 (0.000) 0.003 (0.000) 0.008 (0.000)
Completed Level 3 Math 0.426 (0.004) ‐0.120 (0.003) 0.275 (0.004)Total Credits Earned 0.014 (0.000) 0.003 (0.000) 0.013 (0.000)
Credit Earned in AP or IB Course 0.457 (0.004) ‐0.189 (0.004) 0.247 (0.004)Total Credits Earned 0.017 (0.000) 0.003 (0.000) 0.015 (0.000)
Credits Earned in Math 0.053 (0.002) ‐0.020 (0.002) 0.027 (0.002)Credits Earned in English 0.016 (0.002) 0.023 (0.001) 0.016 (0.001)Credits Earned in Science 0.089 (0.002) ‐0.028 (0.002) 0.048 (0.002)Credits Earned in Social Studies 0.065 (0.002) 0.011 (0.002) 0.054 (0.002)Credits Earned in Foreign Language 0.130 (0.002) ‐0.006 (0.001) 0.108 (0.002)Total Credits Earned ‐0.015 (0.001) 0.002 (0.000) ‐0.008 (0.001)
Level 2 or 3 Credits Earned in Math 0.059 (0.002) 0.016 (0.002) 0.050 (0.002)Level 2 or 3 Credits Earned in English 0.012 (0.002) 0.021 (0.001) 0.012 (0.002)Level 2 or 3 Credits Earned in Science 0.072 (0.002) ‐0.019 (0.002) 0.039 (0.002)Level 2 or 3 Credits Earned in Social Studies 0.023 (0.002) 0.009 (0.002) 0.016 (0.002)Level 2 or 3 Credits Earned in Foreign Language 0.128 (0.002) ‐0.015 (0.001) 0.103 (0.002)Total Credits Earned ‐0.008 (0.001) ‐0.002 (0.000) ‐0.004 (0.000)
Notes: 1) Each panel shows the results from a separate regression. Each regression controls for 8th grade test scores, demographic characteristics, educational needs, and 8th grade fixed effects. 2) Results show the marginal effect computed at the mean of the independet variables. 3) Robust standard errors are shown in parentheses. 4) All results are significant, except where noted by "ns" for "not significant" or "ms" for "marginally significant" (i.e., 10% level).
Attended Any 4‐Year College Within Five Years of HS Entry
Attended Any 2‐Year College (But No 4‐Year College) Within Five Years of HS Entry
Attended Any College Within Five Years of
HS Entry
30
Table 5a: Effects of 10th Grade Course Taking, by Demographic Subgroup
Student Subgroup Credits Earned by 10th Grade
(Non‐Hispanic) White In Math 0.034 (0.003) 0.016 (0.004) 0.006 (0.003) 0.015 (0.004) 0.060 (0.005) 0.002 (0.005)In English ‐0.013 (0.002) ‐0.006 (0.003) 0.025 (0.003) ‐0.007 (0.003) 0.029 (0.004) 0.033 (0.004)In Science 0.033 (0.004) 0.024 (0.005) 0.022 (0.004) 0.019 (0.005) 0.024 (0.006) 0.031 (0.007)In Social Studies ‐0.017 (0.003) 0.015 (0.004) 0.012 (0.003) 0.009 (0.004) 0.040 (0.006) 0.064 (0.006)In Foreign Languages 0.087 (0.003) 0.111 (0.003) ‐0.029 (0.003) 0.135 (0.004) 0.097 (0.003) 0.076 (0.004)Total Credits Earned 0.007 (0.001) ‐0.009 (0.001) ‐0.010 (0.001) ‐0.006 (0.001) ‐0.019 (0.001) ‐0.013 (0.002)
(Non‐Hispanic) Black In Math 0.045 (0.005) 0.002 (0.004) ‐ 0.017 (0.004) + 0.022 (0.006) 0.039 (0.006) ‐ ‐0.018 (0.007) ‐In English ‐0.070 (0.004) ‐ ‐0.022 (0.004) ‐ 0.025 (0.003) ‐0.010 (0.005) 0.078 (0.006) + 0.071 (0.006) +In Science 0.053 (0.007) + 0.026 (0.005) 0.025 (0.005) 0.053 (0.007) + 0.007 (0.008) 0.032 (0.008)In Social Studies ‐0.029 (0.006) 0.016 (0.004) 0.004 (0.005) 0.015 (0.006) 0.063 (0.007) + 0.082 (0.007)In Foreign Languages 0.120 (0.008) + 0.077 (0.004) ‐ ‐0.014 (0.005) + 0.123 (0.007) 0.096 (0.008) 0.079 (0.007)Total Credits Earned 0.012 (0.001) + ‐0.004 (0.001) + ‐0.007 (0.001) + ‐0.009 (0.001) ‐0.023 (0.001) ‐0.017 (0.002)
Hispanic In Math 0.038 (0.006) ‐0.003 (0.007) ‐ 0.011 (0.005) 0.004 (0.005) 0.053 (0.008) ‐0.002 (0.009)In English ‐0.055 (0.005) ‐ ‐0.023 (0.006) ‐ 0.016 (0.003) ‐0.020 (0.004) ‐ 0.050 (0.008) + 0.051 (0.008)In Science 0.046 (0.007) 0.031 (0.008) 0.022 (0.005) 0.030 (0.006) 0.047 (0.011) 0.058 (0.011) +In Social Studies 0.012 (0.007) + 0.039 (0.007) + 0.003 (0.005) 0.019 (0.006) 0.050 (0.009) 0.068 (0.009)In Foreign Languages 0.036 (0.006) ‐ 0.065 (0.006) ‐ 0.002 (0.004) + 0.057 (0.005) ‐ 0.074 (0.007) ‐ 0.047 (0.008) ‐Total Credits Earned 0.007 (0.001) ‐0.008 (0.002) ‐0.006 (0.001) + ‐0.003 (0.001) + ‐0.023 (0.002) ‐0.016 (0.002)
Notes: 1) Each panel shows the results from a separate regression applied only to members of that demographic subgroup. Each regression controls for 8th grade test scores, demographic characteristics, educational needs, and 8th grade fixed effects. 2) Results for the indicator variables show the marginal effect computed at the mean of the independet variables. 3) Robust standard errors are shown in parentheses. 4) + / ‐ indicates significantly larger/smaller marginal effect (at the 5% level) than the reference group (i.e., white, non‐poor, or female).
Receive a High School Diploma
Within Four Years of HS Entry
Attended Any 4‐Year College Within Five Years of HS Entry
Attended Any 2‐Year College (But No 4‐Year College) Within Five Years of HS
Entry
Attended Any College Within Five Years of HS Entry
10th Grade Math FCAT Score
10th Grade Reading FCAT Score
31
Table 5b: Effects of 10th Grade Course Taking, by Demographic Subgroup
Student Subgroup Credits Earned by 10th Grade
Non‐Poor In Math 0.029 (0.003) 0.012 (0.005) 0.008 (0.004) 0.011 (0.004) 0.067 (0.005) 0.007 (0.004)In English ‐0.015 (0.002) ‐0.008 (0.004) 0.024 (0.003) ‐0.005 (0.003) 0.019 (0.004) 0.026 (0.005)In Science 0.027 (0.004) 0.029 (0.006) 0.026 (0.004) 0.027 (0.005) 0.037 (0.006) 0.046 (0.006)In Social Studies ‐0.012 (0.003) 0.019 (0.004) 0.008 (0.004) 0.007 (0.004) 0.039 (0.006) 0.060 (0.006)In Foreign Languages 0.068 (0.003) 0.135 (0.003) ‐0.036 (0.003) 0.108 (0.003) 0.092 (0.004) 0.070 (0.004)Total Credits Earned 0.005 (0.001) ‐0.010 (0.002) ‐0.009 (0.001) ‐0.005 (0.001) ‐0.019 (0.002) ‐0.014 (0.001)
Poor In Math 0.045 (0.004) + 0.004 (0.003) 0.014 (0.003) 0.022 (0.004) + 0.039 (0.005) ‐ ‐0.016 (0.006) ‐In English ‐0.057 (0.003) ‐ ‐0.016 (0.002) 0.019 (0.002) ‐0.012 (0.003) 0.075 (0.004) + 0.067 (0.004) +In Science 0.053 (0.005) + 0.022 (0.003) 0.018 (0.004) 0.037 (0.005) 0.012 (0.005) ‐ 0.028 (0.007) ‐In Social Studies ‐0.014 (0.004) 0.021 (0.003) 0.003 (0.003) 0.021 (0.004) + 0.056 (0.005) + 0.080 (0.007) +In Foreign Languages 0.093 (0.005) + 0.057 (0.003) ‐ 0.015 (0.003) + 0.114 (0.005) 0.088 (0.004) 0.074 (0.005)Total Credits Earned 0.011 (0.001) + ‐0.005 (0.001) + ‐0.007 (0.001) ‐0.009 (0.001) ‐ ‐0.022 (0.001) ‐0.016 (0.002)
Female In Math 0.033 (0.003) 0.001 (0.004) 0.012 (0.004) 0.014 (0.004) 0.058 (0.005) 0.000 (0.005)In English ‐0.035 (0.002) ‐0.017 (0.004) 0.024 (0.003) ‐0.012 (0.003) 0.036 (0.004) 0.036 (0.004)In Science 0.031 (0.004) 0.025 (0.005) 0.023 (0.004) 0.028 (0.005) 0.033 (0.006) 0.040 (0.007)In Social Studies ‐0.013 (0.003) 0.022 (0.004) 0.005 (0.004) 0.008 (0.004) 0.046 (0.005) 0.061 (0.005)In Foreign Languages 0.066 (0.003) 0.110 (0.003) ‐0.022 (0.003) 0.106 (0.004) 0.087 (0.004) 0.071 (0.004)Total Credits Earned 0.008 (0.001) ‐0.008 (0.002) ‐0.008 (0.001) ‐0.005 (0.001) ‐0.021 (0.001) ‐0.015 (0.001)
Male In Math 0.040 (0.003) 0.012 (0.003) 0.004 (0.003) 0.021 (0.004) 0.051 (0.005) ‐0.008 (0.006)In English ‐0.035 (0.003) ‐0.011 (0.003) 0.003 (0.003) ‐ ‐0.006 (0.003) 0.060 (0.004) + 0.058 (0.005) +In Science 0.050 (0.004) + 0.027 (0.004) 0.004 (0.004) ‐ 0.040 (0.005) 0.013 (0.005) ‐ 0.032 (0.006)In Social Studies ‐0.017 (0.004) 0.018 (0.003) 0.004 (0.003) 0.018 (0.004) 0.048 (0.007) 0.078 (0.007) +In Foreign Languages 0.101 (0.004) + 0.098 (0.003) ‐ 0.003 (0.003) + 0.142 (0.004) + 0.113 (0.004) + 0.083 (0.004) +Total Credits Earned 0.008 (0.001) ‐0.008 (0.001) 0.001 (0.001) + ‐0.010 (0.001) ‐ ‐0.021 (0.001) ‐0.015 (0.002)
Notes: 1) Each panel shows the results from a separate regression applied only to members of that demographic subgroup. Each regression controls for 8th grade test scores, demographic characteristics, educational needs, and 8th grade fixed effects. 2) Results for the indicator variables show the marginal effect computed at the mean of the independet variables. 3) Robust standard errors are shown in parentheses. 4) + / ‐ indicates significantly larger/smaller marginal effect (at the 5% level) than the reference group (i.e., white, non‐poor, or female).
Receive a High School Diploma
Within Four Years of HS Entry
Attended Any 4‐Year College Within Five Years of HS Entry
Attended Any 2‐Year College (But No 4‐Year College) Within Five Years of HS
Entry
Attended Any College Within Five Years of HS Entry
10th Grade Math FCAT Score
10th Grade Reading FCAT Score
32
Table 6: Relationship Between School Characteristics and the High School's Estimated Effect of Completing Algebra 1 or Higher by 10th Grade, as Estimated by Propensity Score Matching
Total Enrollment (000s) ‐0.004 (0.000) 0.051 (0.000) *** ‐0.061 (0.000) *** ‐0.023 (0.000) ** ‐0.020 (0.000) ‐0.042 (0.000) **Teacher‐Pupil Ratio ‐0.326 (0.162) ** ‐0.249 (0.146) * ‐0.069 (0.203) ‐0.409 (0.229) * ‐0.856 (0.431) ** ‐0.677 (0.426)Expenditures Per Pupil ($000s) ‐0.022 (0.009) ** 0.014 (0.008) * ‐0.025 (0.009) *** ‐0.020 (0.010) ** ‐0.016 (0.022) ‐0.040 (0.021) *Percentage of Teachers with Advanced Degrees 0.001 (0.001) 0.000 (0.001) ‐0.002 (0.001) *** ‐0.002 (0.001) ** ‐0.003 (0.002) ‐0.003 (0.002)Magnet School ‐0.185 (0.047) *** ‐0.182 (0.044) *** 0.004 (0.041) ‐0.228 (0.047) *** ‐0.280 (0.114) ** ‐0.240 (0.112) **Mean 8th Grade Math FCAT Standardized Score 0.019 (0.040) 0.105 (0.033) *** 0.008 (0.034) 0.137 (0.038) *** 0.332 (0.086) *** 0.321 (0.085) ***Percentage of FRPL Students 0.032 (0.074) ‐0.137 (0.061) ** 0.116 (0.065) * 0.039 (0.070) 0.111 (0.159) 0.264 (0.157) *Constant 0.358 (0.057) *** 0.165 (0.050) *** 0.364 (0.052) *** 0.568 (0.057) *** 1.288 (0.133) *** 1.275 (0.128) ***
Observations 333 333 333 333 333 333Adjusted R‐Squared 0.100 0.351 0.221 0.194 0.122 0.099
Receive a High School Diploma Within Four Years of HS Entry
Attended Any 4‐Year College Within Five Years of HS Entry
Attended Any 2‐Year College (But No 4‐Year College)
Within Five Years of HS Entry
Attended Any College Within Five Years of HS Entry
10th Grade Math FCAT Score 10th Grade Reading FCAT Score
33
Figure 1a: Effect of Additional Credits in Math by the 10th Grade on the Likelihood of Graduating from High School in 4 Years (Estimated by Generalized Propensity Score Matching)
Figure 1b: Effect of Additional Credits in English by the 10th Grade on the Likelihood of Graduating from High School in 4 Years (Estimated by Generalized Propensity Score Matching)
.4.6
.81
E[h
d4(t)
]
0 2 4 6 8Treatment level
Dose Response Low bound
Upper bound
Confidence Bounds at .95 % levelDose r esponse function = Probability of a positive outcomeRegression command = logit
Dose Response Function
-.10
.1.2
.3
E[hd
4(t+
1)]-E
[hd4
(t)]
0 2 4 6 8Treatment level
Treatment Effect Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Probability of a positive outcomeRegression command = l ogit
Treatment Effect Function.4
.5.6
.7.8
E[h
d4(t)
]
0 2 4 6 8Treatment level
Dose Response Low bound
Upper bound
Confidence Bounds at .95 % levelDose r esponse function = Probability of a positive outcomeRegression command = logit
Dose Response Function
-.2-.1
0.1
.2
E[hd
4(t+
1)]-E
[hd4
(t)]
0 2 4 6 8Treatment level
Treatment Effect Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Probability of a positive outcomeRegression command = l ogit
Treatment Effect Function
34
Figure 1c: Effect of Additional Credits in Science by the 10th Grade on the Likelihood of Graduating from High School in 4 Years (Estimated by Generalized Propensity Score Matching)
Figure 1d: Effect of Additional Credits in Social Studies by the 10th Grade on the Likelihood of Graduating from High School in 4 Years (Estimated by Generalized Propensity Score Matching)
.4.5
.6.7
.8.9
E[h
d4(t)
]
0 2 4 6Treatment level
Dose Response Low bound
Upper bound
Confidence Bounds at .95 % levelDose r esponse function = Probability of a positive outcomeRegression command = logit
Dose Response Function
-.2-.1
0.1
.2.3
E[hd
4(t+
1)]-E
[hd4
(t)]
0 2 4 6Treatment level
Treatment Effect Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Probability of a positive outcomeRegression command = l ogit
Treatment Effect Function.2
.4.6
.8
E[h
d4(t)
]
0 2 4 6Treatment level
Dose Response Low bound
Upper bound
Confidence Bounds at .95 % levelDose r esponse function = Probability of a positive outcomeRegression command = logit
Dose Response Function
-.2-.1
0.1
.2
E[hd
4(t+
1)]-E
[hd4
(t)]
0 2 4 6Treatment level
Treatment Effect Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Probability of a positive outcomeRegression command = l ogit
Treatment Effect Function
35
Figure 1e: Effect of Additional Credits in Foreign Language by the 10th Grade on the Likelihood of Graduating from High School in 4 Years (Estimated by Generalized Propensity Score Matching)
Figure 2a: Effect of Additional Credits in Math by the 10th Grade on the Likelihood of Graduating from High School in 4 Years (Estimated by Generalized Propensity Score Matching)
.2.4
.6.8
1
E[h
d4(t)
]
0 2 4 6Treatment level
Dose Response Low bound
Upper bound
Confidence Bounds at .95 % levelDose r esponse function = Probability of a positive outcomeRegression command = logit
Dose Response Function
-.2-.1
0.1
.2
E[hd
4(t+
1)]-E
[hd4
(t)]
0 2 4 6Treatment level
Treatment Effect Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Probability of a positive outcomeRegression command = l ogit
Treatment Effect Function-.5
0.5
E[m
10(t)
]
0 2 4 6 8Treatment level
Dose Response Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Linear prediction
Dose Response Function
-.4-.2
0.2
.4
E[m
10(t+
1)]-E
[m10
(t)]
0 2 4 6 8Treatment level
Treatment Effect Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Linear prediction
Treatment Effect Function
36
Figure 2b: Effect of Additional Credits in English by the 10th Grade on the Likelihood of Graduating from High School in 4 Years (Estimated by Generalized Propensity Score Matching)
Figure 2c: Effect of Additional Credits in Science by the 10th Grade on the Likelihood of Graduating from High School in 4 Years (Estimated by Generalized Propensity Score Matching)
-1-.5
0.5
11.
5
E[m
10(t)
]
0 2 4 6 8Treatment level
Dose Response Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Linear prediction
Dose Response Function
-.50
.51
E[m
10(t+
1)]-E
[m10
(t)]
0 2 4 6 8Treatment level
Treatment Effect Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Linear prediction
Treatment Effect Function-1
-.50
.5
E[m
10(t)
]
0 2 4 6Treatment level
Dose Response Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Linear prediction
Dose Response Function
-.50
.5
E[m
10(t+
1)]-E
[m10
(t)]
0 2 4 6Treatment level
Treatment Effect Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Linear prediction
Treatment Effect Function
37
Figure 2d: Effect of Additional Credits in Social Studies by the 10th Grade on the Likelihood of Graduating from High School in 4 Years (Estimated by Generalized Propensity Score Matching)
Figure 2e: Effect of Additional Credits in Foreign Language by the 10th Grade on the Likelihood of Graduating from High School in 4 Years (Estimated by Generalized Propensity Score Matching)
-.6-.4
-.20
.2.4
E[m
10(t)
]
0 2 4 6Treatment level
Dose Response Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Linear prediction
Dose Response Function
-.50
.5
E[m
10(t+
1)]-E
[m10
(t)]
0 2 4 6Treatment level
Treatment Effect Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Linear prediction
Treatment Effect Function-1
.5-1
-.50
.51
E[m
10(t)
]
0 2 4 6Treatment level
Dose Response Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Linear prediction
Dose Response Function
-1-.5
0.5
1
E[m
10(t+
1)]-E
[m10
(t)]
0 2 4 6Treatment level
Treatment Effect Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Linear prediction
Treatment Effect Function
38
Figure 3: Distribution of High School-Level Effects of Completing Algebra 1 (or Higher) by 10th Grade
01
23
Den
sity
-.5 0 .5 1Effect
kernel = epanechnikov, bandwidth = .04
Distribution of the Effectsof Algebra 1 (or Higher) by 10th Grade
on HS DiplomaAcross High Schools
0.5
11.
5D
ensi
ty
0 .5 1 1.5 2Effect
kernel = epanechnikov, bandwidth = .08
Distribution of the Effectsof Algebra 1 (or Higher) by 10th Grade
on 10th Grade Math FCAT ScoreAcross High Schools
0.5
11.
5D
ensi
ty
-.5 0 .5 1 1.5 2Effect
kernel = epanechnikov, bandwidth = .08
Distribution of the Effectsof Algebra 1 (or Higher) by 10th Gradeon 10th Grade Reading FCAT Score
Across High Schools
01
23
Den
sity
-.5 0 .5 1Effect
kernel = epanechnikov, bandwidth = .04
Distribution of the Effectsof Algebra 1 (or Higher) by 10th Grade
on 4-Year College EntryAcross High Schools
01
23
Den
sity
-.5 0 .5 1Effect
kernel = epanechnikov, bandwidth = .04
Distribution of the Effectsof Algebra 1 (or Higher) by 10th Grade
on 2-Year College EntryAcross High Schools
01
23
Den
sity
-.5 0 .5 1Effect
kernel = epanechnikov, bandwidth = .04
Distribution of the Effectsof Algebra 1 (or Higher) by 10th Grade
on Any College EntryAcross High Schools
39
Figure 4: Relationship Between School Characteristics and Estimated High School-Level Effects of Completing Algebra 1 (or Higher) by 10th Grade on Attending a 4-Year College
01
Effe
ct
0 5000Enrollment
01
Effe
ct
.05 .1Teacher/Pupil
01
Effe
ct
30008000Expend./Pupil
01
Effe
ct
0 100Adv. Degrees
01
Effe
ct
0 1Magnet School
01
Effe
ct
-1 18th Math Score
01
Effe
ct
0 1Pct. FRPL
Relationship Between High School Characteristicsand Effects of Algebra 1 (or Higher) by 10th Grade
on 4-Year College Entry
40
Appendix Figure 1a: Distribution of total credits taken in math by 10th Grade:
Appendix Figure 1b: Distribution of total credits taken in English by 10th Grade:
0.2
.4.6
Den
sity
0 2 4 6 8 10(sum) emat_tot
Kernel density estimateNormal density
kernel = epanechnikov, bandwidth = .3
Kernel density estimate0
.2.4
.6D
ensi
ty
0 2 4 6 8(sum) eeng_tot
Kernel density estimateNormal density
kernel = epanechnikov, bandwidth = .3
Kernel density estimate
Appendix Figure 1c: Distribution of total credits taken in science by 10th Grade:
Appendix Figure 1d: Distribution of total credits taken in social studies by 10th Grade:
Appendix Figure 1e: Distribution of total credits taken in foreign language by 10th Grade
0.2
.4.6
.8D
ensi
ty
0 2 4 6 8(sum) esci_tot
Kernel density estimateNormal density
kernel = epanechnikov, bandwidth = .15
Kernel density estimate0
.2.4
.6D
ensit
y
0 2 4 6 8(sum) esoc_tot
Kernel density estimateNormal density
kernel = epanechnikov, bandwidth = .29
Kernel density estimate
0.2
.4.6
.8D
ensit
y
0 1 2 3 4(sum) efor_tot
Kernel density estimateNormal density
kernel = epanechnikov, bandwidth = .28
Kernel density estimate