Doubling Up: Intensive Math Instruction and Educational ... · PDF fileIntensive Math...
Transcript of Doubling Up: Intensive Math Instruction and Educational ... · PDF fileIntensive Math...
1
Doubling Up: Intensive Math Instruction and Educational Attainment
Kalena E. Cortes*
The Bush School of Government and Public Service Texas A&M University
Joshua Goodman John F. Kennedy School of Government
Harvard University [email protected]
Takako Nomi
Consortium on Chicago School Research University of Chicago
This version: September 14, 2012
Abstract Success or failure in freshman math has long been thought to have a strong impact on subsequent high school outcomes. We study an intensive math instruction policy in which students scoring below average on an 8th grade exam were assigned in 9th grade to an algebra course that doubled instructional time and emphasized problem solving skills. Using a regression discontinuity design, we confirm prior work showing relatively little short-run impact on algebra passing rates and math scores. We show, however, positive and substantial long-run impacts of double-dose algebra on college entrance exam scores, high school graduation rates and college enrollment rates. The majority of this impact comes from students with average math skills but below average reading skills, perhaps because the intervention focused on verbal exposition of mathematical concepts. These facts point to the importance both of evaluating interventions beyond the short-run and of targeting interventions toward appropriately skilled students. This is the first evidence we know of to demonstrate long-run impacts of such intensive math instruction. JEL Classifications: I20, I21, I24, J10, J13, J15, J48 Keywords: Double-dose intervention, Intensive Math Instruction, Student Outcomes, Postsecondary Enrollment
* We are indebted to Chicago Public Schools for sharing their data with us and to Sue Sporte, Director of Research Operations, Consortium on Chicago School Research (CCSR), University of Chicago, for facilitating this sharing. Special thanks for helpful comments from Richard Murnane, Bridget Terry Long, Jeffrey D. Kubik, Lori Taylor, Jacob Vigdor, Caroline Hoxby, Martin West, Kevin Stange, and Nora Gordon, as well as seminar and conference participants at Harvard University’s Program on Education Policy and Governance, State of Texas Education Research Center at Texas A&M University, the Association for Education Finance and Policy, and the National Bureau of Economic Research Economics of Education Program. Colin Sullivan, Heather Sarsons and Shelby Lin provided outstanding research assistance. The research reported here was supported by the Institute of Education Sciences, U.S. Department of Education, through Grant #R305A120466 to Harvard University. The opinions expressed are those of the authors and do not represent views of the Institute or the U.S. Department of Education. Institutional support from Texas A&M University and Harvard University are also gratefully acknowledged. Research results, conclusions, and all errors are our own.
2
1. Introduction
The high school graduation rate for American students has declined since the 1970s to about
75 percent, with black and Hispanic graduation rates hovering around 65 percent (Heckman and
LaFontaine 2010). Poor academic preparation of students entering high school is often cited as a
major source of such high dropout rates. Results from the 2011 National Assessment of Educational
Progress suggest that only 35 percent of students enter high school with math skills considered
proficient. Black and Hispanic students’ proficiency rates are an even lower 13 and 20 percent,
respectively.1 These low academic skills may explain observed high failure rates in 9th grade
coursework, particularly in algebra (Herlihy 2007, Horwitz and Snipes 2008).
Such high failure rates are particularly worrying because of their close association with
dropout rates in later grades. Early course failures prevent students from progressing to more
advanced coursework and from earning the credits needed to graduate (Allensworth and Easton
2007). In the Chicago Public Schools (CPS), the focus of this study, roughly half of high school
freshmen fail at least one course, with the highest failure rates in math courses (Allensworth and
Easton 2005). Concern about this fact and the apparent failure of remedial math courses to correct it
led CPS to implement a double-dose algebra policy starting with students entering high school in the
fall of 2003. Under this policy, students scoring below the national median on an 8th grade math test
were subsequently assigned to two periods of freshman algebra rather than the usual one period.
CPS hoped that this doubling of instructional time, along with an increased emphasis on problem
solving skills, would improve algebra passing rates in the short-run and high school graduation rates
in the longer-run.
1 Similarly large skill gaps by income are also apparent. Students poor enough to qualify for free lunch under the National School Lunch Program have a proficiency rate of 17 percent, compared to a 47 percent proficiency rate among students who do not qualify for such subsidies. See “The Nations Report Card: Mathematics 2011” published by the National Center for Education Statistics.
3
To analyze the effect of the double-dose policy, we employ a regression discontinuity design
comparing students just above and below the threshold for assignment to additional instructional
time. Using longitudinal data that tracks students from 8th grade through college, we confirm prior
work showing little short-run impact on algebra passing rates and math scores. We show, however,
positive and substantial longer-run impacts of double-dose algebra on college entrance exam scores,
high school graduation rates and college enrollment rates. Interestingly, the majority of this policy
impact comes from students with average math skills but below average reading skills, perhaps
because the intervention focused on verbal exposition of mathematical concepts. These facts point
to the importance both of evaluating interventions beyond the short-run and of targeting
interventions toward appropriately skilled students. Our study is the first evidence we know of to
demonstrate long-run impacts of such intensive math instruction.
Our work contributes to three strands of the research literature. First, given that the
intervention studied here doubled the amount of time students were exposed to freshman algebra,
our study adds to the literature on the importance of instructional time to student achievement.
Some education reformers have pushed U.S. schools to lengthen school days and years, noting that
students in many academically successful nations, particularly in Asia, spend substantially more time
in school than do American students. Proponents of this view point to evidence on summer learning
loss (Cooper et al. 1996), the impact of snow days (Marcotte and Hemelt 2008), the association
between charter school effectiveness and instructional time (Dobbie et al. 2011, Hoxby and Murarka
2009), and other such patterns linking student achievement to hours spent learning (Lavy 2010,
Fitzpatrick et al. 2011). Another set of studies suggests this evidence is weaker than it first appears,
with Fryer Jr. and Levitt (2004) observing little differential summer learning loss, Goodman (2012a)
showing little impact of snow days on achievement, Angrist et al. (2011) showing little relation
between instructional time and charter school effectiveness, and Associates (2011) showing little
4
effect of an intervention that substantially increased schools’ instructional times. The emerging
consensus from this literature is that increasing instructional time is no guarantee of better student
outcomes if such time is not well spent. Our results are consistent with heterogeneous impacts of
increased instructional time by math and reading skills.
Second, our work adds to the literature concerning the short-run impact of curricular
interventions, particularly for students struggling in mathematics. Recent years have seen three main
curriculum approaches tried by American schools. Remediation, which diverts students into basic
courses prior to taking regular courses, has generally had little discernible impact on student
achievement, particularly at the college level where it has most often been studied (Jacob and
Lefgren 2004, Lavy and Schlosser 2005, Calcagno and Long 2008, Bettinger and Long 2009,
Martorell and McFarlin Jr 2011, Boatman and Long 2010, Scott-Clayton and Rodriguez 2012).
Algebra “for all”, which pushes students to take algebra courses in earlier grades than they otherwise
would, actually harms student achievement by forcing students into subjects for which they are not
sufficiently prepared (Clotfelter et al. 2012, Nomi 2012). Double-dosing, which places students in
regular courses but supplements those courses with additional instructional time, has generated
modest short-run gains in some settings and no gains in others (Nomi and Allensworth 2009, Nomi
and Allensworth 2010, Roland G. Fryer 2011, Taylor 2012, Dougherty 2012). Perhaps because of
perceived effectiveness at raising short-run achievement levels, the double-dose strategy has become
increasingly common, with half of large urban districts reporting it as their most common form of
support for struggling students.2
Third, and perhaps most important, we contribute to the literature on the long-run impacts
of curriculum on student outcomes. Nearly all such research point to a close association between
coursework completed in high school and later outcomes such as college enrollment and labor
2 See “Urban Indicator: High School Reform Survey, School Year 2006-2007”, by the Council of Great City Schools, 2009.
5
market earnings (Altonji 1995, Levine and Zimmerman 1995, Rose and Betts 2004, Attewell and
Domina 2008, Long et al. 2009, Long et al. 2012). Most such papers attempt to deal with the bias
generated by selection into coursework by controlling for a rich set of covariates, either through
OLS or propensity score matching. However, such methods leave open the possibility that the
remaining unobservables are still important factors. The few papers that use quasi-experimental
methods to convincingly eliminate such selection bias also, however, find strong associations
between completed coursework and long-run outcomes, suggesting that such selection bias is not
generating the central findings (Joensen and Nielsen 2009, Goodman 2012b). This paper is one of
the better identified links between high school coursework and educational attainment.
The structure of our paper is as follows. In section 2, we describe in detail the double-dose
algebra policy. In section 3, we describe the data and offer descriptive statistics about students in our
sample. In section 4, we explain the regression discontinuity underlying our identification strategy. In
sections 5 and 6, we describe the impact of double-dosing on students’ educational experiences,
grades, test scores and educational attainment. In section 7, we discuss robustness, heterogeneity and
spillovers onto other subjects. In section 8, we conclude. We now turn to a description of the
double-dose algebra policy itself.
2. Implementing Double-Dose Algebra
Since the late 1990s, Chicago Public Schools (CPS) has been at the forefront of curriculum
reform designed to increase the rigor of student coursework and prepare students for college
entrance. Starting with students entering high school in the fall of 1997, CPS raised its graduation
requirements to align with the New Basics Curriculum.3 CPS eliminated lower-level and remedial
3 The new basics curriculum was a minimum curriculum recommended by the National Commission of Excellence in Education in 1983, which consists of four years of English, three years of each mathematics, science, and social studies, and one-half year of computer science. The CPS requirements are actually slightly higher than the New Basics
6
courses so that all first-time freshmen would enroll in algebra in 9th grade, geometry in 10th grade
and algebra II or trigonometry in 11th grade. Soon after these reforms, CPS officials realized that
students were unable to master the new college-prep curriculum. Passing rates in 9th grade algebra
were quite low, largely because students entered high school with such poor math skills (Roderick
and Camburn 1999).
In response to these low passing rates in 9th grade algebra, CPS launched the double-dose
algebra policy for students entering high school in the fall of 2003. Instead of reinstating the
traditional remedial courses from previous years, CPS required enrollment in two periods of algebra
coursework for all first-time 9th graders testing below the national median on the math portion of
the 8th grade Iowa Tests of Basic Skills (ITBS).4 Such students enrolled for two math credits, a full-
year regular algebra class plus a full-year algebra support class.5 Three student cohorts, those
entering high school in the fall of 2003, 2004 and 2005, were subject to the double-dose policy. Our
analysis focuses on the first two cohorts because the test score-based assignment rule was not
followed closely for the final cohort. We will refer to these as the 2003 and 2004 cohorts.
Prior to the double-dose policy, algebra curricula had varied considerably across CPS high
schools, due to the fairly decentralized nature of the district. Conversely, CPS offered teachers of
double-dose algebra two specific curricula called Agile Mind and Cognitive Tutor, stand-alone lesson
plans they could use, and thrice annual professional development workshops where teachers were
Curriculum, which includes two years of a foreign language and specific courses in mathematics (i.e., algebra, geometry, advanced algebra, and trigonometry). 4 All CPS high schools were subject to the double-dose algebra policy, including 60 neighborhood schools, 11 magnet schools, and 6 vocational schools (Nomi and Allensworth 2009). 5 Double-dose algebra students received 90 minutes of math class time every day for a full academic year. The first math course, regular algebra, consisted mostly of class lectures. The second math course, algebra with support, focused on building math skills that students lacked and covered materials in a different than the textbook. Double-dose teachers used various instructional activities, such as working in small groups, asking probing and open-ended questions, and using board work (Wenzel et al. 2005, Starkel and Price 2006).
7
given suggestions about how to use the extra instructional time.6 Though it is difficult to know
precisely what occurred in these extra classes, Nomi and Allensworth (2010) surveyed students to
learn more about the classroom learning environment. They found that students assigned to double-
dose algebra reported much more frequently: writing sentences to explain how they solved a math
problem; explaining how they solved a problem to the class; writing math problems for other
students to solve; discussing possible solutions with other students; and applying math to situations
in life outside of school. The additional time thus focused on building verbal and analytical skills
may have conferred benefits in subjects other than math.
CPS also strongly advised schools to schedule their algebra support courses in three specific
ways. First, double-dose algebra students should have the same teacher for their two periods of
algebra. Second, the two algebra periods should be offered consecutively. Third, double-dose
students should take their algebra support class with the same students who are in their regular
algebra class. Most CPS schools followed these recommendations in the initial year (Nomi and
Allensworth 2009). For the 2003 cohort, 80 percent of double-dose students had the same teacher
for both courses, 72 percent took the two courses consecutively, and rates of overlap between the
two classes’ rosters exceeded 90 percent. By 2004, schools began to object to the scheduling
difficulties of assigning the same teacher to both periods so CPS removed that recommendation.
For the 2004 cohort, only 54 percent of double-dose students had the same teacher for both courses
and only 48 percent took the two courses consecutively. Overlap between the rosters remained,
6 The district made the new double-dose curricula and professional development available only to teachers teaching double-dose algebra courses, but there was a possibility of spillover effects for teachers in regular algebra. However, the professional development was geared towards helping teachers structure two periods of algebra instruction. Moreover, based on CPS officials and staff members’ observations of double-dose classrooms, they found that even teachers who taught both single-period and double-dose algebra tended to differentiate their instruction between the two types of classes. Specifically, teachers tended to use new practices with the double-period class, but continued to use traditional methods with the single-period class. Teachers told them that they did not feel they needed to change methods with the advanced students (i.e., non double-dose students), and that they were hesitant to try new practices that may be more time-consuming with just a single period. The double period of algebra allowed these teachers to feel like they had the time to try new practices (e.g., cooperative groups).
8
however, close to 90%. In our analysis, we also explore whether the program’s impacts vary by
cohort in part because of this variation in implementation.
The treatment under consideration here thus had many components. Assignment to double-
dose algebra doubled the amount of instructional time and exposed students to the curricula and
activities discussed above. As we will show, the recommendation that students take the two classes
with the same set of peers caused tracking by skill to increase, thus reducing classroom
heterogeneity. All of these factors were likely to, if anything, improve student outcomes (Duflo et al.
2011). We will also show, however, that the increase tracking by skill placed double-dosed students
among substantially lower skilled peers than non-double dosed students. This factor is likely to, if
anything, hurt student outcomes. Our estimates will, therefore, capture the net impact of all of these
components.
3. Data and Descriptive Statistics
We use longitudinal data from CPS that tracks students from 8th grade through college
enrollment. These data include demographic information, detailed high school transcripts, numerous
standardized test scores, and graduation and college enrollment information. Our sample consists of
all students entering 9th grade for the first time in the fall of 2003 and 2004. We include only
students who have valid 8th grade math scores and who enroll in freshman algebra. We include only
high schools in which at least one classroom of students was assigned to double-dose algebra. For
binary outcomes, students who leave the CPS school system for any reason are coded as zeroes. CPS
attempts to track students’ reasons for leaving. In our sample, students who leave CPS are about
evenly divided between those who are known dropouts, those who leave for other schools (private
schools or public schools outside of Chicago), and those whose reasons for leaving are unknown.
9
The summary statistics of the analytic sample are shown in Table 1. Column (1) includes the
entire sample and column (2) includes only students within 10 percentiles of the double-dose
threshold, our main analytic sample. Columns (3) and (4) separate that sample by cohort. As seen in
panel (A), about 90 percent of CPS students are black or Hispanic, with 20 percent in special
education. Though not shown here, over 90 percent of CPS students are low income as indicated by
participation in the federal subsidized lunch program. We will therefore use as controls more
informative socioeconomic and poverty measures constructed for each student’s residential block
group from the 2000 Census and standardized within the full sample.7 We also include as a control
each student’s 8th grade reading percentile.
The first row of panel (B) shows our instrument, each student’s 8th grade score on the math
portion of the Iowa Test of Basic Skills (ITBS), which all CPS 8th graders are required to take. The
mean CPS 8th grade student scores are between the 45th and 46th percentiles on this nationally
normed exam. As shown in column (1), 55 percent of CPS students score below the 50th percentile
and thus should be assigned to double-dose algebra, though the transcript data reveal that only 42%
actually are assigned to this class, suggesting imperfect compliance with the rule. As a result, the
average CPS freshman in our sample takes 1.4 math courses freshman year.
The transcript data also allow for detailed exploration of the treatment itself. We construct
variables, shown in panel (B), showing the extent to which schools were complying with CPS’
guidelines for implementing double-dose algebra. The average student attended a school in which 63
percent of double-dosed students had their two algebra courses during consecutive periods, in which
71 percent of double-dosed students had the same teacher for both courses, and in which 90 percent 7 Our data are linked to the 2000 U.S. Census at the block level corresponding with each student’s home address. Indicators of students’ socioeconomic status and concentration of poverty were derived from the census data about the economic conditions in students’ residential block groups. More specifically, socioeconomic status variable is based on percentage of employed persons 16 years (or older) who are managers and executives and the mean level of education among people over 18. The concentration of poverty variable is based on the percent of males over 18 who are employed one or more weeks during the year and the percent of families above the poverty line.
10
of double-dosed students’ regular algebra classmates were themselves double-dosed. Consistent with
schools’ complaints that scheduling double-dose algebra according to these guidelines was quite
challenging, compliance was substantially lower in 2004 than in 2003, as can be seen in columns (3)
and (4).
We focus on two primary sets of outcomes. First, in panel (C), we explore whether double-
dosing helps student’s academic achievement by constructing a variety of variables measuring
grades, coursework and standardized test scores. The grades and coursework variables reveal that
only 62 percent of the full sample pass algebra (i.e., receiving a D or higher), while even fewer pass
higher level courses such as geometry and trigonometry. Given that grades are subjective measures,
we construct a variety of test scores standardized by cohort, including the PLAN exam, which all
CPS students take in September of both their second and third years in high school, and the ACT
exam, which all CPS students take in April of their third year and is commonly used in the Midwest
for college applications.
Second, in panel (D), we explore whether double-dosing improves educational attainment by
constructing measures of high school graduation and college enrollment. Students are coded as high
school graduates if they received a regular CPS diploma within four or five years of starting high
school. About 50 percent of CPS students in our sample graduate high school within four years,
with another 5 percent graduating in their fifth year. CPS has matched its data on high school
graduates with the National Student Clearinghouse (NSC) data on college enrollment, allowing us to
observe initial college enrollment for any CPS student with a high school diploma.8 We construct
indicators for enrollment in college by October 1 of the fifth year after starting high school. Only 29
percent of the full sample both graduate from a CPS high school and enroll in college within this
8 The NCS collects information on students’ postsecondary education, which includes semester-by-semester college enrollment and attainment. NCS covers 91 percent of colleges (more than 2,800 postsecondary institutions) in the United States. CPS graduates mostly enroll in local colleges that participate in NCS. We therefore have excellent college enrollment records for about 95 percent of CPS graduates.
11
time frame, more than half of whom enroll in two-year colleges. We also construct the number of
semesters of college a student has enrolled within eight years of starting high school.9 The average
student in this sample has enrolled for 1.8 semesters of college, nearly half of which occurs at two-
year colleges. We cannot explore college completion rates because students in our sample have not
yet had sufficient time to graduate from four-year colleges and because NSC has little graduation
data from two-year colleges in this sample.
4. Empirical Strategy
Comparison of the outcomes of students who are and are not assigned to double-dosed
algebra would likely yield biased estimates of the policy’s impacts given potentially large differences
in unobserved characteristics between the two groups of students. To eliminate this potential bias,
we exploit the fact that students scoring below the 50th percentile on the 8th grade ITBS math test
were supposed to enroll in double-dose algebra. This rule allows us to identify the impact of double-
dose algebra using a regression discontinuity design applied to the two treated cohorts. We use the
assignment rule as an exogenous source of variation in the probability that a given student will be
double-dosed.
We implement the regression discontinuity approach using the regressions below:
itititititit mathlowscoremathlowscoreY 88 3210 (1)
itititititit mathlowscoremathlowscoreDoubleDose 88 3210 (2)
itititititit mathlowscoremathDoubleDoseY 88 3210 (3)
where for student i in cohort t, lowscore indicates an 8th grade math score below the 50th percentile,
math8 is each student’s 8th grade math score re-centered around the 50th percentile cutoff,
DoubleDose is an indicator for assignment to the extra algebra period, and Y represents an outcome
9 This outcome is based on weighting half-time enrollment as 0.5 semesters and less than half-time enrollment as 0.25 semesters. Note that NSC data contains no information on credits completed, just the number of semesters enrolled.
12
of interest. The lowscore coefficient ( 1 ) from equation (1) estimates the discontinuity of interest by
comparing the outcomes of students just below and just above the double-dose threshold. This
reduced form equation produces an intent-to-treat estimate because of imperfect compliance with
the assignment rule. The lowscore coefficient ( 1 ) from the first stage equation (2) measures the
difference in double-dose rates between students just below and just above the threshold.
Our ultimate estimate of interest is thus the DoubleDose coefficient ( 1 ) from equation (3), in
which DoubleDose has been instrumented with lowscore using that first-stage. This approach estimates
a local average treatment effect (LATE) of double-dose algebra for students near the 50th percentile
of 8th grade math skill. Here, students just above the threshold serve as a control group for students
just below the threshold, so that these estimates will be unbiased under the assumption that no other
factors change discontinuously around the threshold itself. The density of 8th grade math scores is
quite smooth around the cutoff, suggesting little scope for manipulation of such scores by students
or teachers and little impact of the threshold on the probability of appearing in our sample.10
Our preferred specification will fit straight lines on either side of the threshold using a
bandwidth of 10 percentiles, and will also control for gender, race, special education status,
socioeconomic and poverty measures, and 8th grade reading scores. We will show that our central
results are robust to alternative choices of controls and bandwidths. In all specifications,
heteroskedasticity robust standard errors will be clustered by each student’s initial high school to
account for within high school correlations in the error terms.
5. The Treatment
We first explore the treatment itself to learn more about how the double-dose algebra policy
was changing students’ freshman year experiences. Before turning to regression results, we look at
10 See Figure A.1.
13
visual evidence. Figure 1 plots the proportion of students double-dosed for each 8th grade math
percentile. If compliance with the double-dose rule were perfect, we would expect to see 100%
remediation rates to the left of the 50th percentile threshold. Instead, panel (A) shows imperfect
compliance, with assignment rates reaching a maximum of about 80 percent for students in the 20-
40th percentiles. Students in the lowest percentiles have lower double-dose rates because they are
more likely to be supported through special education programs (who may be taking low-level
courses that do not technically qualify as double-dose algebra). Also, some students above the
threshold are double-dosed, perhaps because schools cannot perfectly divide students into
appropriately sized classes by the assignment rule. Panel (B) reveals that compliance for students just
below the threshold is substantially lower in the 2004 cohort than the 2003 cohort, providing further
motivation to analyze program impacts separately by cohort.
Table 2 shows the first-stage results using a low 8th grade math score indicator as an
instrument for assignment to double-dose algebra. Panel (A) pools the cohorts while panel (B)
allows the estimate to vary by cohort. Column (1) implements equation (2), including only cohort
indicators as controls. Column (2) adds demographic and test score controls as described in the
table. Column (3) includes those controls and high school fixed effects. Column (4) expands the
bandwidth from 10 to 20 percentiles and fits quadratic functions on either side of the threshold
instead of linear functions.
The first-stage discontinuity estimates suggest that students just below the threshold were 40
percentage points more likely to be double-dosed than students just above the threshold. Consistent
with Figure 1, this discontinuity was much larger in 2003 (48 percentage points) than in 2004 (32
percentage points), further motivating our decision throughout the paper to explore differential
impacts by cohort. These estimates are highly robust to inclusion of controls, inclusion of high
school fixed effects, and to alternative bandwidth and functional form assumptions. Table 3
14
explores heterogeneity by reading skill, by gender and by race/ethnicity. Each column replicates the
second column from Table 2 with the sample limited to the identified sub-group. There is little
indication of differential compliance by reading skill, gender or race/ethnicity, with the exception of
lower compliance rates among the small number of white students in our sample. These patterns
hold true for both the 2003 and 2004 cohorts.
The double-dose treatment involved multiple changes to the freshman algebra experience.
Figures 2, 3 and 4 show visual evidence of the channels through which double-dose may have
affected student outcomes. Figure 2 shows the most obvious impact of double-dose algebra, namely
the substantial increase in the number of class periods devoted to algebra. Doubling of instructional
time was not, however, the only impact of the intervention. Because double-dose students were also
supposed to be placed into classes with similarly double-dosed students, tracking in freshman math
courses increased. The next two figures show that the CPS guideline that double-dose students
should be in regular algebra classes with their double-dose classmates increased tracking by academic
skill. Figure 3 shows that double-dosed students’ regular algebra classes were more homogeneous, as
measured by the standard deviation of 8th grade math scores in those classes. Lastly, figure 4 shows
that double-dosed students regular algebra peers had substantially lower mean math skills than did
students not assigned to double-dose.
Table 4 explores the impact of assignment to double-dose algebra on the freshman academic
experience. All of the coefficients come from regressions in which assignment to double-dose
algebra has been instrumented by eligibility, as in Table 2. As such, these are treatment-on-the-
treated (TOT) estimates of the impact of double-dosing on those actually double-dosed. In panel
(A), column (1) shows that being double-dosed increased the number of freshman math courses
taken by one, as would be expected from the double-dose strategy. The policy thus doubled
instructional time. Columns (2)-(4) show that students fit this additional math course into their
15
schedule not by replacing core academic courses in English, social studies and science, but by
replacing other courses such as fine arts and foreign languages. The result was a positive but small
change in the total number of courses taken freshman year.
Columns (5)-(8) highlight channels other than instructional time and measure various
characteristics of students’ regular (i.e., not double-dose) algebra courses. The increased skill tracking
implied by CPS’ guidelines meant that double-dosed students took algebra classes with peers whose
8th grade math scores were substantially lower than the peers of non-double-dosed students. The
estimates in column (5) imply that double-dosing lowered the mean peer skill of double-dosed
students near the threshold by over 19 percentiles. Column (6) suggests that double-dosed students
near the threshold were in more homogeneous classrooms than their non-double-dosed peers, with
the standard deviation of math skill roughly 3 percentiles lower. Column (7) implies that double-
dosing increased the distance of students to their peers’ mean skill by about 3 percentiles, which
could have negative repercussions if teachers focus their energies on the mean student. Column (8)
suggests that, near the threshold, double-dosed students were in regular algebra classes 2.4 students
larger than non-double-dosed students. Thus, the double-dose policy thus doubled instructional
time in math by replacing other coursework and increased homogeneity of algebra classrooms, but
lowered peer skill, increased distance from the class mean and increased class size. Panel (B) suggests
that none of these effects varied substantially by cohort. We now turn to analysis of the overall
impact of these various changes on coursework, test scores and educational attainment.
6. Grades, Test Scores and Educational Attainment
A visual preview of our results is available in Figures 5, 6 and 7. Using a bandwidth of 10
percentiles on either side of the threshold, each of these figures plots the mean of the given outcome
16
by 8th grade math score, as well as fitted straight lines that come from the regressions we will run.11
Figure 5 suggests little clear impact of double-dosing on algebra passing rates but a noticeable
impact on the fraction of students receiving at least a B. Figure 6 suggests little impact of double-
dosing on math scores the fall after freshman year ended, but a clear impact on such scores a full
year after that. Figure 7 suggest impacts on both high school graduation and college enrollment
rates. We now turn to regression analysis to measure more precisely the magnitude and statistical
significance of these discontinuities.
Table 5 explores the impact of the double-dose policy on math coursework and grades.
Double-dosing increased the proportion of students earning at least a B in freshman algebra by 9.4
percentage points, a more than 65 percent increase from a base of 13.8 percentage points. Though
passing rates for freshman algebra increased by 4.7 percentage points, the increase is statistically
insignificant. Double-dosed students were also no more likely to pass geometry. They were,
however, substantially more likely to pass trigonometry, a course typically taken in the third year of
high school. Mean GPA across all math courses taken after freshman year increased by a marginally
significant 0.14 grade points on a 4.0 scale. As a whole, these results imply that the double-dose
policy greatly improved freshman algebra grades for the upper end of the double-dosed distribution,
but had relatively little impact on passing rates. This latter fact is one of the primary reasons that
CPS has since moved away from this strategy. There is, however, some evidence of improved
passing rates and GPA in later math courses, suggesting the possibility of longer-run benefits
beyond freshman year. Though coursework and grades matter for students’ academic trajectories,
the subjective nature of course grading motivates us to turn to standardized achievement measures
as a potentially better measure of the impact of double-dosing on math skill.
11 Our preferred specification will include student-level demographic controls but these lines come from OLS regressions with no controls other than 8th grade math scores.
17
Table 6 explores the impact of double-dosing on mathematics test scores as measured by the
PLAN exam taken in September of a student’s second year, the PLAN exam taken in September of
a student’s third year, and the ACT exam taken in April of each student’s third year. The PLAN
exams contain a pre-algebra/algebra section and a geometry section, which we analyze separately
given that double-dose classes focused on algebra skills. Column (1) suggests impacts on the first
PLAN exam of 0.09 standard deviations. Though statistically insignificant, it is worth noting that the
positive coefficient comes almost entirely from an improvement in algebra, while the point estimate
for geometry is almost exactly zero. Double-dosing does, however, improve scores on later exams.
Double-dosing increases algebra scores by a statistically significant 0.15 standard deviations and has
a slightly smaller though statistically insignificant impact on geometry. Double-dosing thus raises
overall math scores in the fall of students’ third years by 0.16 standard deviations. Perhaps more
importantly, a nearly identical effect is seen on the math portion of the ACT, with double-dose
algebra raising such scores by a statistically significant 0.15 standard deviations on an exam used by
many colleges as part of the admissions process. These effects are nearly identical across the two
cohorts. The insignificant coefficient in column (7), which uses as an outcome an indicator for
taking the ACT, shows that these results are not driven by selection into exam-taking. Results for
the PLAN exams are similarly insignificant. Table 6 thus suggests that double-dosed students
experienced medium-run achievement gains that persisted at least two years after the end of double-
dose classes.
Table 7 explores the impact of double-dosing on educational attainment. In panel (A),
columns (1) and (2) show that double-dosing increases 4- and 5-year graduation rates by 8.7 and 7.9
percentage points, respectively. This represents a 17 percent improvement over the 51 percent of
non-double-dosed students at the threshold who graduate within 4 years. Panel (B) shows that these
improvements in high school graduation rates were larger in magnitude for the 2003 cohort, but not
18
in any statistically significant way. Columns (3)-(7) show that double-dosing also dramatically
improve college enrollment outcomes. Double-dosed students are 8.6 percentage points more likely
to ever enroll in college within five years of starting high school, a 30 percent increase over the base
college enrollment rate of 29 percent. As columns (4) and (5) show, nearly all of this college
enrollment increase comes from two-year colleges, with the majority of that coming from part-time
enrollment in such colleges. Given the relatively low academic skills and high poverty rates of CPS
students at the double-dose threshold, it is unsurprising that double-dosing improved college
enrollment rates at relatively inexpensive and non-selective two-year postsecondary institutions.
Columns (6) and (7) show that, within eight years of starting high school and thus within four years
of expected high school graduation, double-dosed students have enrolled in 0.6 additional semesters
of college, 0.4 of which are in two-year colleges.
7. Robustness, Heterogeneity and Spillovers
Our primary results suggest double-dose algebra improved students’ math achievement, high
school graduation rates and college enrollment rates. We turn now to questions of the robustness of
these results, heterogeneity of the policy’s impacts, and spillovers into subjects other than math.
Table 8 shows robustness checks for the central results of the previous tables. Panel (A) fits straight
lines on either side of the threshold, using a bandwidth of 10 percentiles. The top row includes no
controls other than cohort indicators, the second row (our preferred specification) adds
demographic controls as described in previous tables, and the third line adds high school fixed
effects. All three specifications yield very similar results. Panel (B) replicates the second row of panel
(A) using successively smaller bandwidths. The magnitudes of the coefficients remain quite similar
or, if anything, increase as the bandwidth is reduced, though their statistical significance diminishes
in some cases due to the reduced sample size. Panel (C) includes three placebo tests, each of which
19
replicates the reduced form version of the second row of panel (A). The first and second placebo
tests use the 40th and 60th percentiles as the double-dose threshold, while the third uses the
untreated 2001 and 2002 cohorts. Only two of the 24 coefficients are marginally significant, as
would be expected by chance. There is no sign of spurious discontinuities. This table implies that
our central results are robust to alternative specifications and that such discontinuities appear only at
thresholds and for cohorts where we expect them.
Table 9 explores whether the impacts of double-dosing varied by the academic skill of the
double-dosed student. Our primary regression discontinuity results, which we repeat in panel (A),
estimate a LATE for students near the given threshold. In our case, our estimates apply to students
near the 50th percentile of math skill. Such students vary, however, in their reading skills as
measured by their 8th grade ITBS reading scores. We exploit this fact in panel (B), where we divide
students into those who scored above (“good readers”) and below (“poor readers”) the 45th
percentile on the 8th grade ITBS reading test, which represents roughly the median reading skill of
students at the double-dose threshold. We then interact double-dosing (and its instruments) with
indicators for those two categories. The results are striking. For all of the outcomes shown, double-
dosing had larger positive effects on poor readers than on good readers, differences which are often
statistically significant. For example, double-dosing raised below median readers’ ACT scores by 0.22
standard deviations but raised above median readers’ ACT scores by only 0.09 standard deviations.
The impact of double-dosing on high school graduation and college enrollment is driven almost
entirely by students in the lower part of the reading distribution. Similar hetereogeneity analysis by
gender and race yields little evidence of differential impacts along these dimensions.
To explore whether double-dosing’s impact varied by math skill, we implement in panel (C)
a difference-in-difference specification using all students in the untreated 2001 and 2002 cohorts as a
control for all students in the treated 2003 and 2004 cohorts. By controlling for differences between
20
low- and high-scoring students in the pre-treatment cohorts and for overall differences between
cohorts, we can thus estimate how the difference in outcomes between low- and high-scoring
students changed at the time double-dose algebra was introduced. This approach estimates an
average treatment effect (ATE) of double-dose algebra for all students double-dosed because of the
policy. Such students are, on average, lower skilled than students near the threshold itself. These
results suggest that, across all double-dosed students, double-dosing did improve passing rates and
short-run algebra scores, but had no discernible impact on later test scores or high school graduation
rates. Data limitations prevent exploration of college outcomes for the earlier cohorts. Together,
panels (B) and (C) suggest that double-dose algebra had little long-run impact on the average
double-dosed student but had substantial positive impacts on double-dosed students with relatively
high math skills but relatively low reading skills. That the majority of the positive long-run impact of
double-dosing came through its effect on low skilled readers may be due to the intervention’s focus
on reading and writing skills in the context of learning algebra.
Finally, the increased focus on algebra at the cost of other coursework may potentially have
affected achievement in other academic subjects. Table 10 looks at outcomes in other subjects. We
find strong evidence that, rather than harming other achievement, double-dosing had positive
spillovers in reading and science. Double-dosed students scored nearly 0.2 standard deviations
higher on the verbal portion of their ACTs, were substantially more likely to pass chemistry classes
usually taken in 10th or 11th grade, and had marginally high GPAs across all of their non-math
classes in years after 9th grade. If anything, the skills gained in double-dose algebra seem to have
helped, not hindered, students in other subjects and subsequent years.
21
8. Conclusion
Double-dose strategy has become an increasingly popular way to aid students struggling in
mathematics. Today, nearly half of large urban districts report doubled math instruction as the most
common form of support for students with lower skills (Council of Great City Schools 2009). The
central concern of urban school districts is that algebra may be a gateway for later academic success,
so that early high school failure in math may have large effects on subsequent academic achievement
and graduation rates. As the current policy environment calls for “algebra for all” in 9th grade or
earlier grades, providing an effective and proactive intervention is particularly critical for those who
lack foundational mathematical skills. A successful early intervention may have the greatest chance
of having longer-term effects on students’ academic outcomes.
We provide evidence of positive and substantial long-run impacts of one particular form of
intensive math instruction on college entrance exam scores, high school graduation rates and college
enrollment rates. We show that this intensive math instruction was most successful for students with
average math skills but relatively low reading skills, and not successful for the average treated
student. This highlights the importance of carefully targeting such interventions to students most
likely to benefit from them. Also, like other recent studies, we find that the test score impacts of this
policy dramatically understate its long-run benefits as measured by educational attainment (Deming,
2009; Chetty et al., 2011). In our sample, OLS suggests that a 0.15 standard deviation increase in
ACT math scores is associated with a 2 percentage point increase in college enrollment rates. We
observe college enrollment effects four times that size, highlighting the fact that long-run analyses of
such interventions may yield very different conclusion than short-run analyses.
One final piece of evidence suggests that CPS’s double-dose policy impacts could be
replicated in other urban school districts across the U.S. Recall that the district strongly suggested
that double-dose students should have the same teacher in regular algebra and in their double-dose
22
class, that the two algebra courses should be offered sequentially, and that double-dose students
should take their double-dose class with the same set of students that were in their regular algebra
class. Table 11 explores whether the impact of double-dosing varied by the extent to which each
school complied with CPS’ implementation guidelines. For each school we construct the fraction of
double-dosed students whose schedule adhered to each of these guidelines. We average all three
measures of compliance in panel (A) and break them out separately in panel (B). In neither panel is
there any evidence that compliance with these guidelines is related to the impact of double-dosing.
This is consistent with the fact that the impact of double-dosing seems not to have varied by the
2003 and 2004 cohorts, even though schools were less likely to comply with the guidelines for the
second cohort. Districts looking to adopt the double-dose strategy could likely reap its benefits
without needing to radically restructure their school days.
23
References
Allensworth, E. and J. Easton (2005). The on-track indicator as a predictor of high school graduation. Consortium on Chicago School Research, University of Chicago. Allensworth, E. and J. Easton (2007). What matters for staying on track and graduating in Chicago Public High Schools. Chicago, IL: Consortium on Chicago school research. 17. Altonji, J. (1995). The effects of high school curriculum on education and labor market outcomes. The Journal of Human Resources 30(3), 409–438. Angrist, J. D., P. A. Pathak, and C. R. Walters (2011). Explaining charter school effectiveness. Working Paper 17332, National Bureau of Economic Research. Associates, A. (2011). Evaluation of the expanded learning time initiative. Annual report. Attewell, P. and T. Domina (2008). Raising the bar: Curricular intensity and academic performance. Educational Evaluation and Policy Analysis 30(1), 51–71. Bettinger, E. and B. Long (2009). Addressing the needs of underprepared students in higher education. The Journal of Human Resources 44(3), 736–771. Boatman, A. and B. Long (2010). Does remediation work for all students? How the effects of postsecondary remedial and developmental courses vary by level of academic preparation. Calcagno, J. C. and B. T. Long (2008, July). The impact of postsecondary remediation using a regression discontinuity approach: Addressing endogenous sorting and noncompliance. Working Paper 14194, National Bureau of Economic Research. Clotfelter, C. T., H. F. Ladd, and J. L. Vigdor (2012, June). The aftermath of accelerating algebra: Evidence from a district policy initiative. Working Paper 18161, National Bureau of Economic Research. Cooper, H., B. Nye, K. Charlton, J. Lindsay, and S. Greathouse (1996). The effects of summer vacation on achievement test scores: A narrative and meta-analytic review. Review of Educational Research 66(3), 227–268. Dobbie, W., R. Fryer, and G. Fryer Jr (2011). Are high-quality schools enough to increase achievement among the poor? Evidence from the harlem children’s zone. American Economic Journal: Applied Economics 3(3), 158–187. Dougherty, S. (2012). Bridging the discontinuity in adolescent literacy: Evidence of effectiveness from one district. Unpublished manuscript. Duflo, E., P. Dupas, and M. Kremera (2011). Peer effects, teacher incentives, and the impact of tracking: Evidence from a randomized evaluation in kenya. The American Economic Review 101(5), 1739–1774.
24
Fitzpatrick, M., D. Grissmer, and S. Hastedt (2011). What a difference a day makes: Estimating daily learning gains during kindergarten and first grade using a natural experiment. Economics of Education Review 30(2), 269–279. Fryer Jr, R. and S. Levitt (2004). Understanding the black-white test score gap in the first two years of school. The Review of Economics and Statistics 86(2), 447–464. Goodman, J. (2012a). Flaking out: Snowfall, disruptions of instructional time, and student achievement. Working Paper Series. Goodman, J. (2012b). The labor of division: Returns to compulsory math coursework. Working Paper Series. Heckman, J. and P. LaFontaine (2010). The american high school graduation rate: Trends and levels. The Review of Economics and Statistics 92(2), 244–262. Herlihy, C. (2007). State and district-level support for successful transitions into high school. National High School Center, 14. Horwitz, A. and J. Snipes (2008). Supporting successful transitions to high school. Council of Great City Schools 6. Hoxby, C. M. and S. Murarka (2009). Charter schools in new york city: Who enrolls and how they affect their students’ achievement. Working Paper 14852, National Bureau of Economi Research. Jacob, B. and L. Lefgren (2004). Remedial education and student achievement: A regression discontinuity analysis. The Review of economics and statistics 86(1), 226–244. Joensen, J. and H. Nielsen (2009). Is there a causal effect of high school math on labor market outcomes? The Journal of Human Resources 44(1), 171–198. Lavy, V. (2010). Do differences in schools instruction time explain international achievement gaps in math, science, and reading? Evidence from developed and developing countries. Working Paper 16227, National Bureau of Economic Research. Lavy, V. and A. Schlosser (2005). Targeted remedial education for underperforming teenagers: Costs and benefits. Journal of Labor Economics 23(4), pp. 839–874. Levine, P. and D. Zimmerman (1995). The benefit of additional high-school math and science classes for young men and women. Journal of Business & Economic Statistics, 137–149. Long, M., P. Iatarola, and D. Conger (2009). Explaining gaps in readiness for college-level math: The role of high school courses. Education Finance and Policy 4(1), 1–33. Long, M. C., D. Conger, and P. Iatarola (2012). Effects of high school course-taking on secondary and postsecondary success. American Educational Research Journal 49(2), 285–322.
25
Marcotte, D. and S. Hemelt (2008). Unscheduled school closings and student performance. Education Finance and Policy 3(3), 316–338. Martorell, P. and I. McFarlin Jr (2011). Help or hindrance? The effects of college remediation on academic and labor market outcomes. The Review of Economics and Statistics 93(2), 436–454. Nomi, T. (2012). The unintended consequences of an algebra-for-all policy on high-skill students: Effects on instructional organization and students’ academic outcomes. Educational Evaluation and Policy Analysis. Nomi, T. and E. Allensworth (2009). “double-dose” algebra as an alternative strategy to remediation: Effects on students’ academic outcomes. Journal of Research on Educational Effectiveness 2(2), 111–148. Nomi, T. and E. Allensworth (2010). The effects of tracking with supports on instructional climate and student outcomes in high school algebra. Working paper, University of Chicago Consortium on Chicago School Research. Roderick, M. and E. Camburn (1999). Risk and recovery from course failure in the early years of high school. American Educational Research Journal 36(2), 303–343. Roland G. Fryer, J. (2011, October). Injecting successful charter school strategies into traditional public schools: Early results from an experiment in Houston. Working Paper 17494, National Bureau of Economic Research. Rose, H. and J. Betts (2004). The effect of high school courses on earnings. The Review of Economics and Statistics 86(2), 497–513. Scott-Clayton, J. and O. Rodriguez (2012). Development, diversion, or discouragement? a new framework and new evidence on the effects of college remediation. Stacy A. Wenzel, Kareem Lawal, B. C. C. R. F. S. R. S. (2005). Data brief: Algebra problem solving teachers talk about their experiences, December 2004. Brief report, UIC CMSI Evaluation Project. Starkel, R., M. J. and K. Price (2006). Two-period algebra in the 05-06 school year: Implementation report. Technical report. Taylor, E. (2012). Allocating more of the school day to math: Regression-discontinuity estimates of returns and costs. Unpublished manuscript.
Figure 1: Double-Dosing Rates
0.2
.4.6
.8F
ract
ion
do
ub
le−
do
sed
0 10 20 30 40 50 60 70 80 90 1008th grade math percentile
(A) Cohorts pooled
0.2
.4.6
.8F
ract
ion
do
ub
le−
do
sed
0 10 20 30 40 50 60 70 80 90 1008th grade math percentile
2003 cohort
2004 cohort
(B) Cohorts separated
21
Figure 2: Freshman Algebra Periods
11.
21.
41.
61.
8F
resh
man
alg
ebra
per
iod
s
0 10 20 30 40 50 60 70 80 90 1008th grade math percentile
(A) Cohorts pooled
11.
21.
41.
61.
8F
resh
man
alg
ebra
per
iod
s
0 10 20 30 40 50 60 70 80 90 1008th grade math percentile
2003 cohort
2004 cohort
(B) Cohorts separated
22
Figure 3: Standard Deviation of Peer Math Skill
911
1315
17S
D o
f p
eer
mat
h s
kil
l
0 10 20 30 40 50 60 70 80 90 1008th grade math percentile
(A) Cohorts pooled
911
1315
17S
D o
f p
eer
mat
h s
kil
l
0 10 20 30 40 50 60 70 80 90 1008th grade math percentile
2003 cohort
2004 cohort
(B) Cohorts separated
23
Figure 4: Mean Peer Math Skill
1020
3040
5060
7080
90M
ean
pee
r m
ath
sk
ill
0 10 20 30 40 50 60 70 80 90 1008th grade math percentile
(A) Cohorts pooled
1020
3040
5060
7080
90M
ean
pee
r m
ath
sk
ill
0 10 20 30 40 50 60 70 80 90 1008th grade math percentile
2003 cohort
2004 cohort
(B) Cohorts separated
24
Figure 5: Freshman Algebra Grades
.58
.6.6
2.6
4.6
6.6
8P
asse
d a
lgeb
ra
40 50 608th grade math percentile
(A) Passed algebra
.14
.16
.18
.2.2
2A
or
B i
n a
lgeb
ra
40 50 608th grade math percentile
(B) A or B in algebra
25
Figure 6: Math Achievement
−.2
−.1
0.1
.2F
all
10 m
ath
sco
re
40 50 608th grade math percentile
(A) Fall 10 math score
−.2
−.1
0.1
.2F
all
11 m
ath
sco
re
40 50 608th grade math percentile
(B) Fall 11 math score
26
Figure 7: Educational Attainment
.5.5
5.6
.65
Gra
du
ated
HS
in
5 y
ears
40 50 608th grade math percentile
(A) Graduated high school in 5 years
.2.2
5.3
.35
.4A
ny
co
lleg
e
40 50 608th grade math percentile
(B) Enrolled in college
27
Table 1: Summary Statistics
(1) (2) (3) (4)Both cohorts, Both cohorts 2003 cohort, 2004 cohort,
full sample near threshold near threshold near threshold
(A) Controls
Female 0.50 0.54 0.54 0.54Black 0.57 0.56 0.58 0.55Hispanic 0.33 0.36 0.34 0.37Special education 0.20 0.08 0.08 0.08Census block poverty measure -0.00 -0.02 -0.01 -0.03Census block SES measure -0.00 -0.01 -0.02 -0.018th grade reading percentile 43.33 46.60 46.48 46.71
(B) Double-dose
8th grade math percentile 45.69 49.92 50.01 49.82Eligible for double-dose 0.55 0.48 0.47 0.49Double-dosed 0.42 0.41 0.41 0.40Freshman math courses 1.40 1.39 1.39 1.39Consecutive periods 0.63 0.65 0.77 0.54Same teacher 0.71 0.73 0.85 0.60Extent of tracking 0.90 0.90 0.94 0.86
(C) Achievement
Passed algebra 0.62 0.63 0.63 0.64Passed geometry 0.57 0.59 0.58 0.59Passed trigonometry 0.51 0.52 0.51 0.53Fall 10 math z-score 0.00 -0.03 -0.04 -0.02Fall 11 math z-score 0.00 -0.05 -0.05 -0.04ACT math z-score 0.00 -0.21 -0.21 -0.22
(D) Attainment
Graduated HS in 4 years 0.49 0.51 0.50 0.53Graduated HS in 5 years 0.54 0.56 0.55 0.57Enrolled in any college 0.29 0.31 0.29 0.33Enrolled in 2-year college 0.16 0.17 0.16 0.19Semesters of college 1.86 1.93 1.89 1.96Semesters of 2-year college 0.82 0.93 0.91 0.94
N 41,122 11,507 5,734 5,773
Notes: Mean values of each variable are shown by sample. Column (1) is the full sample of students from the 2003and 2004 cohorts. Column (2) limits the sample to students within 10 percentiles of the double-dose threshold.Columns (3) and (4) separate column (2) by cohort.
28
Table 2: Eligibility as an Instrument for Double-Dose Algebra
(1) (2) (3) (4)Y = double-dosed No Demographic High school Quadratic,
controls controls fixed effects bandwidth=20
(A) Cohorts pooled
Eligible for double-dose 0.400∗∗∗ 0.399∗∗∗ 0.406∗∗∗ 0.387∗∗∗
(0.038) (0.039) (0.037) (0.039)F (Z) 108.4 106.4 120.3 97.6
(B) Cohorts separated
Eligible * 2003 0.478∗∗∗ 0.478∗∗∗ 0.486∗∗∗ 0.442∗∗∗
(0.039) (0.039) (0.038) (0.038)Eligible * 2004 0.321∗∗∗ 0.319∗∗∗ 0.325∗∗∗ 0.331∗∗∗
(0.047) (0.048) (0.046) (0.047)p (β2003
= β2004) 0.000 0.000 0.000 0.002
N 11,507 11,507 11,507 21,539
Notes: Heteroskedasticity robust standard errors clustered by initial high school are in parentheses (* p<.10 ** p<.05*** p<.01). Each column in each panel represents the first-stage regression of semester of double-dose algebra oneligibility as determined by eighth grade math score. Panel (A) pools the two treatment cohorts. Below panel (A) isthe value from an F-test of the instrument. Panel (B) interacts the instrument with cohort indicators. Below panel(B) is the p-value from an F-test of the equality of the two listed coefficients. Column (1) fits straight lines on bothsides of the threshold using a bandwidth of 10 percentiles. Cohort indicators are included but not shown. Column(2) adds to column (1) controls for gender, race, census block poverty and socioeconomic status, and eighth grademath and reading scores. Column (3) adds to column (2) high school fixed effects. Column (4) adds to column (3)quadratic terms on either side of the threshold and expands the bandwidth to 20 percentiles.
29
Tab
le3:
Fir
st-S
tag
eH
eter
og
enei
ty
(1)
(2)
(3)
(4)
(5)
(6)
(7)
Bel
ow
Ab
ov
eY
=d
ou
ble
-do
sed
med
ian
med
ian
Mal
eF
emal
eB
lack
His
pan
icW
hit
ere
ader
read
er
(A)
Co
ho
rts
po
ole
d
Eli
gib
lefo
rd
ou
ble
-do
se0.
405∗
∗∗
0.38
4∗∗∗
0.38
0∗∗∗
0.41
4∗∗∗
0.41
8∗∗∗
0.40
2∗∗∗
0.23
7∗∗
(0.0
41)
(0.0
46)
(0.0
41)
(0.0
42)
(0.0
38)
(0.0
65)
(0.1
04)
F(Z
)95
.869
.285
.198
.611
9.8
38.6
5.2
(B)
Co
ho
rts
sep
arat
ed
Eli
gib
le*
2003
0.47
5∗∗∗
0.47
9∗∗∗
0.46
2∗∗∗
0.49
1∗∗∗
0.49
8∗∗∗
0.47
7∗∗∗
0.32
2∗∗∗
(0.0
43)
(0.0
46)
(0.0
40)
(0.0
44)
(0.0
40)
(0.0
70)
(0.0
92)
Eli
gib
le*
2004
0.33
5∗∗∗
0.28
8∗∗∗
0.30
4∗∗∗
0.33
1∗∗∗
0.33
6∗∗∗
0.33
1∗∗∗
0.14
8(0
.050
)(0
.056
)(0
.051
)(0
.051
)(0
.052
)(0
.071
)(0
.121
)p
(β2003=
β2004)
0.00
20.
000
0.00
00.
001
0.00
40.
010
0.02
9
N5,
612
5,89
55,
291
6,21
66,
494
4,11
889
5
No
tes:
Het
ero
sked
asti
city
rob
ust
stan
dar
der
rors
clu
ster
edb
yin
itia
lh
igh
sch
oo
lar
ein
par
enth
eses
(*p<
.10
**p<
.05
***
p<
.01)
.E
ach
colu
mn
rep
lica
tes
the
seco
nd
colu
mn
fro
mta
ble
2w
ith
the
sam
ple
lim
ited
toth
eid
enti
fied
sub
-gro
up
.
30
Tab
le4:
Do
ub
le-D
ose
Alg
ebra
,Fre
shm
anC
ou
rsew
ork
and
Pee
rs
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Mat
hA
cad
emic
Oth
erT
ota
lM
ean
pee
rS
Do
fD
ista
nce
toC
lass
cou
rses
cou
rses
cou
rses
cou
rses
skil
lsk
ill
mea
np
eer
size
(A)
Co
ho
rts
po
ole
d
Do
ub
le-d
ose
d0.
994∗
∗∗
-0.1
69∗
-0.6
69∗∗∗
0.15
6∗-1
9.40
1∗∗∗
-3.1
01∗∗∗
3.20
2∗∗∗
2.44
3∗∗∗
(0.0
22)
(0.0
96)
(0.0
88)
(0.0
89)
(1.4
40)
(0.7
61)
(1.2
16)
(0.7
76)
(B)
Co
ho
rts
sep
arat
ed
Do
ub
le-d
ose
d*
2003
0.97
9∗∗∗
-0.1
71∗
-0.6
75∗∗∗
0.13
3-1
9.44
3∗∗∗
-3.3
45∗∗∗
3.32
7∗∗∗
2.34
9∗∗∗
(0.0
22)
(0.0
89)
(0.0
87)
(0.0
82)
(1.4
11)
(0.8
28)
(1.0
47)
(0.8
48)
Do
ub
le-d
ose
d*
2004
1.02
0∗∗∗
-0.1
66-0
.657
∗∗∗
0.19
7∗-1
9.32
7∗∗∗
-2.6
71∗∗∗
2.97
9∗2.
609∗
∗∗
(0.0
24)
(0.1
23)
(0.1
03)
(0.1
17)
(2.1
16)
(0.8
64)
(1.7
32)
(0.8
83)
p(β
2003=
β2004)
0.00
70.
948
0.77
80.
388
0.95
10.
354
0.76
40.
737
Yat
thre
sho
ld1.
212
3.51
32.
255
6.98
051
.609
15.7
9310
.705
20.8
13N
11,5
0711
,507
11,5
0711
,507
11,5
0711
,501
11,5
0711
,507
No
tes:
Het
ero
sked
asti
city
rob
ust
stan
dar
der
rors
clu
ster
edb
yin
itia
lh
igh
sch
oo
lar
ein
par
enth
eses
(*p<
.10
**p<
.05
***
p<
.01)
.E
ach
colu
mn
inea
chp
anel
rep
rese
nts
the
inst
rum
enta
lv
aria
ble
sre
gre
ssio
no
fth
eli
sted
ou
tco
me
on
sem
este
rso
fd
ou
ble
-do
seal
geb
ra,
wh
ere
do
ub
le-d
ose
isin
stru
men
ted
by
elig
ibil
ity
acco
rdin
gto
the
firs
t-st
age
giv
enin
tab
le2.
Pan
el(A
)p
oo
lsth
etw
otr
eatm
ent
coh
ort
s.P
anel
(B)
inte
ract
sth
ein
stru
men
tan
den
do
gen
ou
sre
gre
sso
rw
ith
coh
ort
ind
icat
ors
.B
elo
wp
anel
(B)
isth
ep
-val
ue
fro
man
F-t
est
of
the
equ
alit
yo
fth
etw
oli
sted
coef
fici
ents
.A
lso
list
edis
the
mea
nv
alu
eo
fea
cho
utc
om
eat
the
elig
ibil
ity
thre
sho
ld.
Co
lum
n(2
)in
clu
des
all
cou
rses
insc
ien
ce,
En
gli
shan
dso
cial
stu
die
s.C
olu
mn
(3)
incl
ud
esal
lco
urs
eso
ther
than
mat
h,s
cien
ce,E
ng
lish
and
soci
alst
ud
ies.
31
Table 5: The Impact of Double-Dose Algebra on Math Coursework
(1) (2) (3) (4) (5)A or B in Passed Passed Passed Math GPAalgebra algebra geom. trig. 10-12
(A) Cohorts pooled
Double-dosed 0.094∗∗ 0.047 -0.022 0.084∗∗ 0.143∗
(0.042) (0.046) (0.049) (0.041) (0.082)
(B) Cohorts separated
Double-dosed * 2003 0.095∗∗ 0.035 -0.010 0.075∗ 0.149∗
(0.038) (0.045) (0.043) (0.039) (0.079)Double-dosed * 2004 0.093∗ 0.067 -0.044 0.101∗ 0.132
(0.053) (0.057) (0.067) (0.055) (0.107)p (β2003
= β2004) 0.930 0.417 0.432 0.512 0.832
Y at threshold 0.138 0.624 0.576 0.544 1.398N 11,507 11,507 11,507 11,507 10,439
Notes: Heteroskedasticity robust standard errors clustered by initial high school are in parentheses (* p<.10 **p<.05 *** p<.01). Each column in each panel represents the instrumental variables regression of the listed outcomeon semesters of double-dose algebra, where double-dose is instrumented by eligibility according to the first-stagegiven in table 2. Panel (A) pools the two treatment cohorts. Panel (B) interacts the instrument and endogenousregressor with cohort indicators. Below panel (B) is the p-value from an F-test of the equality of the two listedcoefficients. Also listed is the mean value of each outcome at the eligibility threshold.
32
Tab
le6:
Th
eIm
pac
to
fD
ou
ble
-Do
seA
lgeb
rao
nM
ath
Ach
iev
emen
t
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Fal
l10
Fal
l10
Fal
l10
Fal
l11
Fal
l11
Fal
l11
Sp
rin
g11
To
ok
mat
hal
geb
rag
eom
etry
mat
hal
geb
rag
eom
etry
AC
Tm
ath
AC
T
(A)
Co
ho
rts
po
ole
d
Do
ub
le-d
ose
d0.
086
0.08
50.
008
0.15
9∗∗
0.14
9∗∗
0.10
10.
153∗
∗-0
.017
(0.0
67)
(0.0
70)
(0.0
82)
(0.0
64)
(0.0
59)
(0.0
85)
(0.0
71)
(0.0
39)
(B)
Co
ho
rts
sep
arat
ed
Do
ub
le-d
ose
d*
2003
0.07
70.
079
0.02
30.
145∗
∗0.
140∗
∗0.
053
0.14
0∗∗
0.00
4(0
.059
)(0
.064
)(0
.073
)(0
.059
)(0
.056
)(0
.077
)(0
.064
)(0
.036
)D
ou
ble
-do
sed
*20
040.
100
0.09
7-0
.017
0.18
4∗∗
0.16
5∗∗
0.18
2∗0.
173∗
∗-0
.055
(0.0
90)
(0.0
90)
(0.1
07)
(0.0
80)
(0.0
78)
(0.1
10)
(0.0
88)
(0.0
54)
p(β
2003=
β2004)
0.67
00.
726
0.50
40.
375
0.66
30.
043
0.45
50.
146
Yat
thre
sho
ld-0
.023
-0.1
07-0
.101
-0.0
76-0
.119
-0.1
31-0
.211
0.57
6N
8,21
08,
210
8,21
07,
428
7,42
87,
428
6,70
011
,507
No
tes:
Het
ero
sked
asti
city
rob
ust
stan
dar
der
rors
clu
ster
edb
yin
itia
lh
igh
sch
oo
lar
ein
par
enth
eses
(*p<
.10
**p<
.05
***
p<
.01)
.E
ach
colu
mn
inea
chp
anel
rep
rese
nts
the
inst
rum
enta
lv
aria
ble
sre
gre
ssio
no
fth
eli
sted
ou
tco
me
on
sem
este
rso
fd
ou
ble
-do
seal
geb
ra,
wh
ere
do
ub
le-d
ose
isin
stru
men
ted
by
elig
ibil
ity
acco
rdin
gto
the
firs
t-st
age
giv
enin
tab
le2.
Pan
el(A
)p
oo
lsth
etw
otr
eatm
ent
coh
ort
s.P
anel
(B)
inte
ract
sth
ein
stru
men
tan
den
do
gen
ou
sre
gre
sso
rw
ith
coh
ort
ind
icat
ors
.B
elo
wp
anel
(B)
isth
ep
-val
ue
fro
man
F-t
est
of
the
equ
alit
yo
fth
etw
oli
sted
coef
fici
ents
.A
lso
list
edis
the
mea
nv
alu
eo
fea
cho
utc
om
eat
the
elig
ibil
ity
thre
sho
ld.
33
Tab
le7:
Th
eIm
pac
to
fD
ou
ble
-Do
seA
lgeb
rao
nE
du
cati
on
alA
ttai
nm
ent
(1)
(2)
(3)
(4)
(5)
(6)
(7)
Gra
du
ated
Gra
du
ated
En
roll
ed,
En
roll
ed,
Par
t-ti
me,
Sem
este
rs,
Sem
este
rs,
hig
hsc
ho
ol
hig
hsc
ho
ol
any
two
-yea
rtw
o-y
ear
any
two
-yea
rin
4y
ears
in5
yea
rsco
lleg
eco
lleg
eco
lleg
eco
lleg
eco
lleg
e
(A)
Co
ho
rts
po
ole
d
Do
ub
le-d
ose
d0.
087∗
0.07
9∗∗
0.08
6∗∗∗
0.07
9∗∗
0.06
6∗∗∗
0.59
9∗∗
0.42
3∗∗
(0.0
47)
(0.0
39)
(0.0
33)
(0.0
36)
(0.0
26)
(0.2
39)
(0.1
67)
(B)
Co
ho
rts
sep
arat
ed
Do
ub
le-d
ose
d*
2003
0.10
5∗∗
0.09
4∗∗∗
0.09
4∗∗∗
0.07
0∗∗
0.05
7∗∗
0.65
6∗∗∗
0.32
4∗∗
(0.0
44)
(0.0
36)
(0.0
31)
(0.0
34)
(0.0
24)
(0.2
36)
(0.1
57)
Do
ub
le-d
ose
d*
2004
0.05
50.
052
0.07
20.
093∗
0.08
3∗∗
0.49
80.
598∗
∗
(0.0
60)
(0.0
52)
(0.0
46)
(0.0
48)
(0.0
37)
(0.3
29)
(0.2
41)
p(β
2003=
β2004)
0.17
30.
238
0.51
20.
514
0.35
80.
559
0.15
7
Yat
thre
sho
ld0.
509
0.56
30.
290
0.15
90.
077
1.79
50.
873
N11
,507
11,5
0711
,507
11,5
0711
,507
11,5
0711
,507
No
tes:
Het
ero
sked
asti
city
rob
ust
stan
dar
der
rors
clu
ster
edb
yin
itia
lh
igh
sch
oo
lar
ein
par
enth
eses
(*p<
.10
**p<
.05
***
p<
.01)
.E
ach
colu
mn
inea
chp
anel
rep
rese
nts
the
inst
rum
enta
lv
aria
ble
sre
gre
ssio
no
fth
eli
sted
ou
tco
me
on
sem
este
rso
fd
ou
ble
-do
seal
geb
ra,
wh
ere
do
ub
le-d
ose
isin
stru
men
ted
by
elig
ibil
ity
acco
rdin
gto
the
firs
t-st
age
giv
enin
tab
le2.
Pan
el(A
)p
oo
lsth
etw
otr
eatm
ent
coh
ort
s.P
anel
(B)
inte
ract
sth
ein
stru
men
tan
den
do
gen
ou
sre
gre
sso
rw
ith
coh
ort
ind
icat
ors
.B
elo
wp
anel
(B)
isth
ep
-val
ue
fro
man
F-t
est
of
the
equ
alit
yo
fth
etw
oli
sted
coef
fici
ents
.A
lso
list
edis
the
mea
nv
alu
eo
fea
cho
utc
om
eat
the
elig
ibil
ity
thre
sho
ld.
34
Tab
le8:
Ro
bu
stn
ess
Ch
eck
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Pas
sed
Fal
l10
Fal
l11
Sp
rin
g11
Gra
du
ated
En
roll
ed,
Sem
este
rs,
Sem
este
rs,
fres
hm
anm
ath
mat
hA
CT
mat
hh
igh
sch
oo
lan
yan
ytw
o-y
ear
alg
ebra
sco
resc
ore
sco
rein
5y
ears
coll
ege
coll
ege
coll
ege
(A)
Ban
dw
idth
=10
No
con
tro
ls0.
032
0.06
90.
137∗
∗0.
141∗
0.06
00.
067∗
0.46
7∗0.
368∗
∗
(0.0
46)
(0.0
71)
(0.0
64)
(0.0
74)
(0.0
42)
(0.0
37)
(0.2
55)
(0.1
74)
+d
emo
gra
ph
icco
ntr
ols
0.04
70.
086
0.15
9∗∗
0.15
3∗∗
0.07
9∗∗
0.08
6∗∗∗
0.59
9∗∗
0.42
3∗∗
(0.0
46)
(0.0
67)
(0.0
64)
(0.0
71)
(0.0
39)
(0.0
33)
(0.2
39)
(0.1
67)
+h
igh
sch
oo
lfi
xed
effe
cts
0.04
50.
044
0.12
9∗∗
0.12
7∗0.
071∗
0.07
3∗∗
0.48
6∗∗
0.36
8∗∗
(0.0
45)
(0.0
63)
(0.0
62)
(0.0
73)
(0.0
39)
(0.0
34)
(0.2
36)
(0.1
62)
(B)
Oth
erb
and
wid
ths
Ban
dw
idth
=8
0.09
9∗0.
135∗
0.21
0∗∗∗
0.17
1∗∗
0.09
9∗∗
0.07
3∗0.
538∗
0.34
6(0
.052
)(0
.073
)(0
.075
)(0
.072
)(0
.048
)(0
.042
)(0
.288
)(0
.221
)B
and
wid
th=
60.
069
0.09
40.
291∗
∗∗
0.16
7∗0.
153∗
∗0.
100∗
∗0.
608∗
0.17
5(0
.071
)(0
.084
)(0
.103
)(0
.095
)(0
.069
)(0
.047
)(0
.334
)(0
.264
)B
and
wid
th=
40.
045
0.06
90.
306∗
∗0.
184
0.14
20.
117∗
0.81
30.
352
(0.0
79)
(0.1
26)
(0.1
37)
(0.1
30)
(0.0
88)
(0.0
66)
(0.5
11)
(0.3
26)
(C)
Pla
ceb
ote
sts
Th
resh
old
at40
thp
erce
nti
le-0
.000
0.03
3-0
.003
0.02
50.
036∗
0.00
5-0
.010
0.04
2(0
.017
)(0
.032
)(0
.030
)(0
.025
)(0
.021
)(0
.016
)(0
.100
)(0
.067
)T
hre
sho
ldat
60th
per
cen
tile
0.03
1∗0.
027
-0.0
11-0
.029
0.02
10.
021
0.14
80.
115
(0.0
18)
(0.0
28)
(0.0
31)
(0.0
30)
(0.0
22)
(0.0
21)
(0.1
59)
(0.0
94)
Ear
lier
coh
ort
s0.
019
0.01
8-0
.016
-0.0
010.
013
-0.0
07-0
.064
0.00
8(0
.014
)(0
.030
)(0
.030
)(0
.029
)(0
.020
)(0
.015
)(0
.116
)(0
.066
)
No
tes:
Het
ero
sked
asti
city
rob
ust
stan
dar
der
rors
clu
ster
edb
yin
itia
lh
igh
sch
oo
lar
ein
par
enth
eses
(*p<
.10
**p<
.05
***
p<
.01)
.P
anel
(A)
fits
stra
igh
tli
nes
on
eith
ersi
de
of
the
thre
sho
ld,
usi
ng
ab
and
wid
tho
f10
per
cen
tile
s.P
anel
(B)
rep
lica
tes
the
seco
nd
row
of
pan
el(A
)b
ut
usi
ng
oth
erb
and
wid
ths.
Pan
el(C
)in
clu
des
thre
ep
lace
bo
test
s,ea
cho
fw
hic
hre
pli
cate
sth
ere
du
ced
form
ver
sio
no
fth
ese
con
dro
wo
fp
anel
(A).
Th
efi
rst
and
seco
nd
pla
ceb
ote
sts
use
the
40th
and
60th
per
cen
tile
sas
the
do
ub
le-d
ose
thre
sho
ld,w
hil
eth
eth
ird
use
sth
eu
ntr
eate
d20
01an
d20
02co
ho
rts.
35
Tab
le9:
Het
ero
gen
eity
by
Aca
dem
icS
kil
l
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Pas
sed
Fal
l10
Fal
l11
Sp
rin
g11
Gra
du
ated
En
roll
ed,
Sem
este
rs,
Sem
este
rs,
fres
hm
anm
ath
mat
hA
CT
mat
hh
igh
sch
oo
lan
yan
ytw
o-y
ear
alg
ebra
sco
resc
ore
sco
rein
5y
ears
coll
ege
coll
ege
coll
ege
(A)
RD
,ov
eral
l
Do
ub
le-d
ose
d0.
047
0.08
60.
159∗
∗0.
153∗
∗0.
079∗
∗0.
086∗
∗∗
0.59
9∗∗
0.42
3∗∗
(0.0
46)
(0.0
67)
(0.0
64)
(0.0
71)
(0.0
39)
(0.0
33)
(0.2
39)
(0.1
67)
Yat
thre
sho
ld0.
633
-0.0
28-0
.067
-0.1
920.
576
0.29
01.
837
0.89
8N
11,5
078,
210
7,42
86,
700
11,5
0711
,507
11,5
0711
,507
(B)
RD
,by
read
ing
skil
l
Do
ub
le-d
ose
d*
po
or
read
er0.
085
0.08
80.
184∗
∗0.
219∗
∗∗
0.12
7∗∗∗
0.13
3∗∗∗
0.82
0∗∗∗
0.54
8∗∗∗
(0.0
52)
(0.0
81)
(0.0
73)
(0.0
75)
(0.0
48)
(0.0
41)
(0.2
82)
(0.1
98)
Do
ub
le-d
ose
d*
go
od
read
er0.
009
0.08
00.
136∗
∗0.
093
0.03
10.
039
0.37
90.
299∗
(0.0
43)
(0.0
63)
(0.0
68)
(0.0
77)
(0.0
40)
(0.0
36)
(0.2
70)
(0.1
73)
p(β
below=
βabove)
0.01
80.
884
0.39
20.
015
0.01
50.
019
0.09
80.
112
N11
,507
8,21
07,
428
6,70
011
,507
11,5
0711
,507
11,5
07
(C)
DD
,ov
eral
l
Do
ub
le-d
ose
d0.
064∗
∗∗
0.07
6∗∗∗
0.02
40.
009
0.01
6(0
.019
)(0
.027
)(0
.029
)(0
.032
)(0
.013
)
Yfo
rb
elo
w*b
efo
re0.
551
-0.5
20-0
.512
-0.5
360.
457
N79
,394
53,3
9749
,653
44,5
8579
,394
No
tes:
Het
ero
sked
asti
city
rob
ust
stan
dar
der
rors
clu
ster
edb
yin
itia
lh
igh
sch
oo
lar
ein
par
enth
eses
(*p<
.10
**p<
.05
***
p<
.01)
.P
anel
(A)
con
tain
sth
ere
gre
ssio
nd
isco
nti
nu
ity
resu
lts
fro
mp
rio
rta
ble
s,b
elo
ww
hic
his
the
mea
nv
alu
eo
fea
cho
utc
om
eat
the
elig
ibil
ity
thre
sho
ld.
Pan
el(B
)se
par
ates
stu
den
tsb
y8t
hg
rad
ere
adin
gsc
ore
,in
tera
ctin
gb
oth
the
inst
rum
ent
and
end
og
eno
us
reg
ress
or
wit
hre
adin
gsk
ill
gro
up
s.B
elo
wea
chco
effi
cien
tin
pan
el(B
)is
the
p-v
alu
efr
om
anF
-tes
to
fth
eeq
ual
ity
of
the
two
coef
fici
ents
.P
anel
(C)
con
tain
sin
stru
men
tal
var
iab
les
reg
ress
ion
su
sin
gth
ed
iffe
ren
ce-i
n-d
iffe
ren
cesp
ecifi
cati
on
des
crib
edin
the
tex
t.B
elo
wp
anel
(C)
isth
em
ean
val
ue
of
each
ou
tco
me
for
stu
den
tsb
elo
wth
eel
igib
ilit
yth
resh
old
inth
e20
01an
d20
02co
ho
rts.
36
Tab
le10
:S
pil
lov
ers
on
toO
ther
Su
bje
cts
(1)
(2)
(3)
(4)
(5)
(6)
(7)
Fal
l10
AC
TF
all
10A
CT
Pas
sed
No
nm
ath
No
nm
ath
ver
bal
ver
bal
scie
nce
scie
nce
chem
istr
yG
PA
,9G
PA
,10-
12
(A)
Co
ho
rts
po
ole
d
Do
ub
le-d
ose
d0.
106∗
0.18
5∗∗∗
-0.0
010.
039
0.09
9∗∗
-0.0
430.
150∗
(0.0
63)
(0.0
65)
(0.0
83)
(0.0
85)
(0.0
44)
(0.0
80)
(0.0
85)
(B)
Co
ho
rts
sep
arat
ed
Do
ub
le-d
ose
d*
2003
0.10
2∗0.
173∗
∗∗
0.05
60.
074
0.09
2∗∗
0.00
60.
152∗
∗
(0.0
57)
(0.0
59)
(0.0
76)
(0.0
79)
(0.0
42)
(0.0
69)
(0.0
75)
Do
ub
le-d
ose
d*
2004
0.11
10.
205∗
∗-0
.101
-0.0
150.
111∗
-0.1
310.
145
(0.0
82)
(0.0
92)
(0.1
11)
(0.1
07)
(0.0
58)
(0.1
13)
(0.1
19)
p(β
2003=
β2004)
0.86
00.
650
0.02
00.
221
0.66
90.
070
0.92
8
Yat
thre
sho
ld-0
.069
-0.1
23-0
.047
-0.1
680.
434
1.92
81.
786
N8,
228
6,70
08,
184
6,69
911
,507
11,5
0710
,509
No
tes:
Het
ero
sked
asti
city
rob
ust
stan
dar
der
rors
clu
ster
edb
yin
itia
lh
igh
sch
oo
lar
ein
par
enth
eses
(*p<
.10
**p<
.05
***
p<
.01)
.E
ach
colu
mn
inea
chp
anel
rep
rese
nts
the
inst
rum
enta
lv
aria
ble
sre
gre
ssio
no
fth
eli
sted
ou
tco
me
on
sem
este
rso
fd
ou
ble
-do
seal
geb
ra,
wh
ere
do
ub
le-d
ose
isin
stru
men
ted
by
elig
ibil
ity
acco
rdin
gto
the
firs
t-st
age
giv
enin
tab
le2.
Pan
el(A
)p
oo
lsth
etw
otr
eatm
ent
coh
ort
s.P
anel
(B)
inte
ract
sth
ein
stru
men
tan
den
do
gen
ou
sre
gre
sso
rw
ith
coh
ort
ind
icat
ors
.B
elo
wp
anel
(B)
isth
ep
-val
ue
fro
man
F-t
est
of
the
equ
alit
yo
fth
etw
oli
sted
coef
fici
ents
.A
lso
list
edis
the
mea
nv
alu
eo
fea
cho
utc
om
eat
the
elig
ibil
ity
thre
sho
ld.
37
Tab
le11
:C
han
nel
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Pas
sed
Fal
l10
Fal
l11
Sp
rin
g11
Gra
du
ated
En
roll
ed,
Sem
este
rs,
Sem
este
rs,
fres
hm
anm
ath
mat
hA
CT
mat
hh
igh
sch
oo
lan
yan
ytw
o-y
ear
alg
ebra
sco
resc
ore
sco
rein
5y
ears
coll
ege
coll
ege
coll
ege
(A)
By
ov
eral
lco
mp
lian
ce
Do
ub
le-d
ose
d0.
052
0.09
10.
160∗
∗0.
151∗
∗0.
078∗
0.09
0∗∗
0.63
8∗∗
0.44
2∗∗
(0.0
46)
(0.0
69)
(0.0
65)
(0.0
72)
(0.0
40)
(0.0
35)
(0.2
53)
(0.1
77)
Do
ub
le-d
ose
d*
Gu
idel
ine
com
pli
ance
-0.1
04-0
.176
∗-0
.023
0.07
20.
020
-0.0
93-0
.823
∗-0
.395
(0.0
80)
(0.1
06)
(0.0
79)
(0.1
03)
(0.0
78)
(0.0
79)
(0.4
99)
(0.3
95)
(B)
By
each
gu
idel
ine
Do
ub
le-d
ose
d0.
052
0.09
30.
156∗
∗0.
153∗
∗0.
080∗
0.08
9∗∗
0.64
4∗∗
0.42
7∗∗
(0.0
47)
(0.0
69)
(0.0
67)
(0.0
72)
(0.0
41)
(0.0
36)
(0.2
56)
(0.1
79)
Do
ub
le-d
ose
d*
Co
nse
cuti
ve
per
iod
s-0
.142
∗0.
070
0.08
30.
155∗
-0.0
690.
009
-0.4
25-0
.572
∗
(0.0
78)
(0.0
80)
(0.0
73)
(0.0
93)
(0.0
75)
(0.0
70)
(0.4
82)
(0.3
11)
Do
ub
le-d
ose
d*
Sam
ete
ach
er0.
013
-0.1
67∗∗
-0.0
66-0
.091
-0.0
29-0
.072
-0.6
250.
260
(0.1
04)
(0.0
81)
(0.1
22)
(0.1
31)
(0.1
11)
(0.0
95)
(0.6
71)
(0.3
62)
Do
ub
le-d
ose
d*
Ex
ten
to
ftr
ack
ing
0.10
5-0
.141
-0.1
07-0
.030
0.25
8-0
.047
0.74
1-0
.062
(0.1
69)
(0.1
92)
(0.1
97)
(0.2
67)
(0.2
05)
(0.1
95)
(1.2
47)
(0.6
64)
N11
,507
8,21
07,
428
6,70
011
,507
11,5
0711
,507
11,5
07
No
tes:
Het
ero
sked
asti
city
rob
ust
stan
dar
der
rors
clu
ster
edb
yin
itia
lh
igh
sch
oo
lar
ein
par
enth
eses
(*p<
.10
**p<
.05
***
p<
.01)
.B
oth
pan
els
rep
lica
teth
em
ain
spec
ifica
tio
ns
fro
mp
rio
rta
ble
sb
ut
inte
ract
bo
thth
ein
stru
men
tan
den
do
gen
ou
sre
gre
sso
rw
ith
the
giv
enm
easu
res
of
gu
idel
ine
com
pli
ance
.T
ho
sem
easu
res
of
gu
idel
ine
com
pli
ance
are
also
con
tro
lled
for
dir
ectl
yb
ut
no
tsh
ow
n.
38