Upload
others
View
8
Download
0
Embed Size (px)
Citation preview
The Multidimensional Impact of Teachers on Students
Nathan Petek and Nolan G. Pope*
January 2021
Abstract
For decades, policymakers and researchers have used value-added models that rely solely onstudent test scores to measure teacher quality. However, since teaching ability is multidimen-sional, test-score value-added measures of teacher quality may not fully capture the impact ofteachers on students. In this paper, we use test-score and non-test-score measures of studentachievement and behavior from over a million students in the Los Angeles Unified SchoolDistrict to estimate multiple dimensions of teacher quality. We find that test-score and non-test-score measures of teacher quality are only weakly correlated, and that both measures ofteacher quality affect students’ performance in high school. A teacher-removal policy simula-tion that uses both dimensions of teacher quality improves most long-term student outcomesby over 50 percent compared to a policy that uses test scores alone. Our results also show thatthe long-term effects of teachers in later grades are larger than in earlier grades and that per-formance in core elementary school subjects matters more for long-term outcomes than othersubjects.
*Petek: Federal Trade Commission, 600 Pennsylvania Ave NW, Washington, DC 20580 ([email protected]).Pope: The University of Maryland, 3114 Tydings Hall 7343 Preinkert Dr, College Park Maryland 20742([email protected]). We would like to thank John Eric Humphries, Steven Levitt, Jens Ludwig, Ofer Malamud,Magne Mogstad, Derek Neal, and Eric Nielsen for helpful comments and discussion. The views expressed in thisarticle are those of the authors. They do not necessarily represent those of the Federal Trade Commission or any of itsCommissioners.
1 Introduction
Teacher quality has garnered the attention of policymakers and researchers for many years. Re-
searchers have primarily measured teacher quality using a test-score value-added framework.1 Al-
though the use of test-score value-added has substantially impacted education research, people
have long recognized that good teachers likely affect a wide range of student outcomes. In fact,
early theoretical formulations of value-added used an education production function that modeled
educational output as a “multidimensional factor” (Hanushek 1971). Consequently, measures of
teacher quality that rely solely on student test scores may not fully capture the impact of teachers
on students.
In this paper, we are interested in whether teachers can noticeably impact measures of student
achievement beyond just test scores, and to what extent the impact on non-test-score measures
is important for the future success of students. We use the value-added framework to construct
separate measures of teacher ability to increase test scores, behavior, and a plausible measure of
noncognitive skills. We use these multiple value-added estimates of teacher ability to explore
the effect of teachers on students’ long-term outcomes and the relative importance of cognitive
and noncognitive skills in the production of human capital. We illustrate the benefits of using
broader measures of teacher ability by investigating the extent to which using multiple measures
of teacher ability increases the efficacy of teacher selection and assignment policies, improves the
measurement of the cumulative return to high quality teaching, and allows the measurement of
teacher quality in untested subjects.
We gather administrative data from the Los Angeles Unified School District (LAUSD) for stu-
dents in grades K-12 from 2003 to 2015. These data link over a million students to teachers, and
track students over time as they progress through the LAUSD system. Our three measures of stu-
dent achievement are constructed from (1) student math and English state test scores, (2) measures
of student behavior, including suspensions, attendance, GPA, and grade retention, and (3) teacher
1An important exception is a paper by Kirabo Jackson (2018) that estimates non-test-score measures of teacherquality that we discuss in more detail below.
2
assessments of student effort and 14 learning skills which are plausibly measures of noncognitive
ability. The learning skills include teacher assessments such as whether a student makes good use
of time, exercises self-control, and resolves conflicts appropriately. We measure the long-term ef-
fects of teachers using student performance in high school, including dropping out of high school,
taking the SAT, SAT scores, high school exit exam scores, GPA, teacher assessments of effort and
cooperation, attendance, suspensions, and grade retention.
We first document that elementary school students with better test scores, behavior, and learn-
ing skills perform better in high school. We then estimate teacher value-added measures of three
dimensions of teacher quality – student test scores (using math and English state tests), student
behavior (using GPA, attendance, suspensions, and grade retention), and student learning skills
(using teacher assessments of effort and 14 learning skills). To avoid bias and potential teacher
manipulation when using teacher-reported non-test-score variables, we modify the standard value-
added framework to use student outcomes from the year after the student was in a teacher’s class,
instead of the contemporaneous year. Using these value-added measures, we show that teachers
affect both test-score and non-test-score dimensions of student achievement, but little evidence that
that learning-skills value-added in particular affects high school outcomes.
We find that having a high test-score value-added teacher in elementary school improves stu-
dents’ high school performance. These long-term effects of test-score value-added are not sub-
stantially reduced by adding teachers’ behavior or learning-skills value-added to the model. This
result suggests that the long-term effects of test-score value-added may not be biased by omitting
non-test-score teaching ability.
We also find that behavior value-added is only weakly correlated with test-score value-added,
and has a similarly large effect on students’ long-term outcomes. Therefore, test-score value-added
misses the dimensions of teacher quality captured by behavior value-added that matter for long-
term outcomes. Consequently, test-score value-added underestimates the total effect of teachers
on students. However, we find no evidence of an interaction effect for teachers who are better or
worse on both dimensions of teacher ability.
3
The low correlation between the two value-added measures also suggests that using behavior
value-added in conjunction with test-score value-added may substantially enhance the accuracy
with which overall teacher quality is measured. We illustrate how behavior value-added enhances
the measurement of teacher quality using a hypothetical policy simulation that replaces teachers
in the bottom 5 percent of the teacher quality distribution with district average teachers. Relative
to relying on test-score value-added alone, a simple rule that equally weights the test-score and
behavior value-added of a teacher increases the efficacy of the policy by at least 50 percent for
the likelihood of dropping out of high school, taking the SAT, high school GPA, suspensions,
absences, and on-time progression. These gains are obtained with little to no decline in student
test scores, are similar to the gains obtained if an optimal weighting scheme is used, and do not
require administering additional tests or using data beyond what schools typically collect.
Finally, we use test-score and non-test-score measures of ability in two applications. First,
we estimate the effect of test-score and behavior value-added for each grade 3 to 12. We find
that middle school and high school teachers have a larger effect on outcomes measured in 11th or
12th grade than elementary school teachers. This result suggests that teachers in later grades may
play a more important role in benefiting the long-term student outcomes than teachers in earlier
grades under the strong assumption that a one standard deviation change in teacher value-added
induces the same amount of learning in each grade. Assuming constant returns to higher quality
teachers, these results imply large cumulative benefits of teacher value-added. For example, giving
students a standard deviation better test-score value-added teacher each year from grades 3 to 12
increases the likelihood of taking the SAT by 8.1 percentage points, and reduces the likelihood
of dropping out of high school by 0.5 percentage points. Giving students a standard deviation
better behavior value-added teacher over the same period increases the likelihood of taking the
SAT by 8.4 percentage points, and reduces the likelihood of dropping out of high school by 5.9
percentage points. The cumulative effects of better teachers are only somewhat reduced by controls
for tracking.
Second, the focus on test scores has limited the study of teacher effects to a few regularly tested
4
subjects (i.e., math and English). We instead use subject-specific GPAs to compute value-added
measures of teacher quality in 10 elementary school subjects. We find that students with higher
value-added teachers in reading and health perform better in high school, whereas having a higher
value-added teacher in speaking has negative effects on high school performance. Hiring teachers
who are relatively better at teaching reading and spending more time on it could potentially benefit
the long-term outcomes of student.
From a policy perspective, there are potentially large benefits from adopting a measure of
teacher quality that includes both test-score and non-test-score dimensions. For example, poli-
cymakers can use non-test-score value-added to measure teacher quality for all teachers, not just
math and English teachers. In addition, since focusing on only one output of the multidimensional
education production function (i.e., test scores) may distort the efficient allocation of teachers’ time
and resources, using a broader measure of teacher quality may help alleviate this distortion. Fi-
nally, using a better measure of overall teacher quality can make school districts’ hiring and tenure
decisions more effective.
Our paper contributes to a literature that estimates the effect of various non-test score value-
added measures on contemporaneous outcomes (Jennings and DiPrete 2010, Ruzek et al. 2014,
Gershenson 2016, Blazar and Kraft 2017, and Kraft 2019), on within-high school outcomes (Jack-
son 2018), and on outcomes of 20-year-olds (Fleche 2017).2 The paper most closely related to ours
is Jackson (2018). Using North Carolina data, he finds that above and beyond test scores, teachers
effect proxies for noncognitive skills (behavior value-added) in 9th grade and subsequently out-
comes in 12th grade such as high school completion, SAT-taking, and intentions to attend college.
He finds that including both test-score and behavioral value-added measures in 9th grade more
than doubles the predictable variability of teacher effects on 12th grade outcomes.
2The non-test-score value-added measures include social and behavioral skills (Jennings and DiPrete 2010); mo-tivation (Ruzek et al. 2014); absences (Gershenson 2016); belief in the ability to do math and happiness in math class(Blazar and Kraft 2017); and grit, growth mindset, effort; answering open-ended questions (Kraft 2019); absences, sus-pensions, grades, and grade progression (Jackson 2018); and internalizing and externalizing behavior (Fleche 2017).Araujo et al. (2016) produce short-term estimates of classroom value-added on measures of executive function. Otherstudies assess multidimensional teacher effects using non-value-added approaches (Mihaly et al. 2013 and Rockoffand Speroni 2010).
5
Our paper contributes to the literature in several ways. First, we show the long term effects of
non-test-score value-added extend back to elementary school. We also show there are no interac-
tions in the educational production function between our measures of teacher ability to increase
cognitive and noncognitive skills. In addition, we quantify the potential gains to long-term student
performance from a policy that assesses teachers using both test-score and non-test score value-
added rather than just test-score value-added.
Second, our estimates of the long-term effect of teacher quality by grade (for both test-score
and non-test-score measures) allow for estimates of the benefits of switching teachers between
grades and the cumulative effect of increasing teacher quality across grades that is not possible
in previous work (Chetty, Friedman, and Rockoff (2014b) estimated test-score value-added for
grades 4 through 8 and Jackson (2018) for test-score and non-test-score value-added in grade 9).
The test-score value added literature also has been limited to measuring teacher performance in
tested subjects like English and math (Jackson, Rockoff, and Staiger 2014). Our extension of the
value-added framework to measure subject-specific GPA value-added provides novel estimates of
the effect of teaching quality on long-term outcomes in subjects outside of English and math.
Finally, we contribute to the broader literature of the role of cognitive and noncognitive skills
in the production of human capital and long-term outcomes (e.g., Cunha, Heckman, Schennach
2010; Heckman, Stixrud, and Urzua 2006) by analyzing how teachers with varying levels of ability
to increase students’ cognitive and noncognitive skills affect their students’ long-term outcomes.
More specifically, we contribute to understanding the role that the development of cognitive and
non-cognitive skills play on the long-term effects of educational interventions (Heckman, Pinto,
and Savelyev 2013) using a different source of variation in cognitive and noncognitive skills.
The rest of the paper will proceed as follows. Section 2 provides background information on
value-added scores. Section 3 describes the LAUSD data that we use and, in particular, describes
the variables used to measure test-score, behavior, and learning-skills value-added. Section 4 out-
lines the empirical method for estimating teacher value-added measures and estimating the effect
of teacher value-added on long-term student outcomes. Section 5 presents the descriptive results of
6
the test score, behavior, and learning-skills value-added of teachers, and then reports the results for
how teachers affect students’ concurrent and long-term outcomes. The gains from teacher-removal
policies that use multiple dimensions of teacher quality are also presented. Section 6 presents the
relative value of higher quality teachers over the students’ educational life cycle and in specific
subjects. Section 7 concludes.
2 Background on Test-Score Value-Added
Since the early 1970s, researchers have used test-score valued-added to measure teacher quality
(Hanushek 1971). This research led states and school districts to use test-score value-added in
teacher evaluations as early as the 1990s (Horn and Sanders 1994). Since then, the use of test-score
value-added has expanded, and 27 states require that teacher evaluations include “growth measures
as a significant criterion” (The National Council of Teacher Quality 2015). This increased use
of test-score value-added has largely been due to a lack of other predictors of teacher quality
(Hanushek and Rivkin 2010). Much of the recent work in the value-added literature focuses on the
validity of value-added models (Bacher-Hicks, Kane, Staiger 2014; Chetty, Friedman, and Rockoff
2014a; Kane and Staiger 2008; Kane et al. 2013; Rockoff 2004; Rothstein 2010; Rothstein 2017;
Chetty, Friedman, and Rockoff 2017), gains from using them in personnel decisions (Goldhaber
and Hansen 2010; Gordon, Kane, and Staiger 2006; Hanushek 2011), and theoretical and empirical
studies of their use in pay-for-performance (Fryer 2013; Goodman and Turner 2013; Neal 2011).
In particular, Chetty, Friedman, and Rockoff (2014b), who find that students with higher test-score
value-added teachers earn significantly more by their late 20s, have fewer births as teenagers, and
are more likely to attend college.
3 Los Angeles Student Data
The LAUSD is the second largest school district in the United States, educating over 600,000
students each year. In 2003, the school district was 71.9 percent Hispanic, 12.1 percent black,
7
and 9.4 percent white.3 We use a panel of student-level administrative data on all public school
students in the LAUSD. The panel links students to teachers over time and includes the 2002-03
to 2014-15 school years, which we reference by year of graduation (e.g., we refer to the 2002-03
school year as 2003). Our analysis focuses on the over 110,000 3rd to 5th grade students studying
in the LAUSD each year.
These data are unique in the level of detail they provide about each student’s academic perfor-
mance. For grades 2 through 11, math and English California state test (CST) scores are available
for each student. The testing regime is relatively consistent over this period, with the only major
change being an essay section added to the 4th- and 7th-grade English test in 2011. For all grades,
these data contain the number of days a student was suspended, the number of days a student was
absent, and whether a student did not progress on time to the next grade (i.e., held back). Both
elementary and high school students received progress reports with their grades by subject and a
number of additional teacher assessments of student performance.
Elementary school progress reports (grades K-5) are given each trimester by the student’s sole
classroom teacher, and contain achievement grades in 10 subjects (e.g., reading, mathematics, art,
etc.), effort grades for the same 10 subjects, grades for five “work and study habits” (e.g., “makes
good use of time,” “organizes materials,” etc.), and grades for nine “learning and social skills”
(e.g., “resolves conflicts appropriately,” “exercises self-control,” etc.). All grades are on a 4-point
scale, with no fractional points given. We compute an annual GPA for each of the four groups
listed above. Figure 1 shows a template of the progress report.
Starting in the 6th grade, middle school and high school students receive progress reports each
semester from multiple classroom teachers, with three categories of grades for each of their classes:
achievement (i.e., academic performance), “work habits,” which we term effort (i.e., “effort,” “re-
sponsibility,” “attendance,” and “evaluation”), and “cooperation” (i.e., “courtesy,” “conduct,” “im-
provement,” and “class relations”). Achievement is graded on a 4-point scale, and effort and
cooperation are graded on a 3-point scale, with no fractional points given. We compute annual
3Statistics can be found at http://dq.cde.ca.gov/dataquest.
8
GPAs for each of these three groups of measures. Appendix Figure A.1 shows additional details
on grading criteria.
Additional data are available for middle and high school students, including whether a student
dropped out of high school (i.e., the student enrolled in the LAUSD in grade 9 and did not graduate
high school in the LAUSD within 5 years), graduated from the LAUSD conditional on enrolling
in the LAUSD in 12th grade, SAT scores, PSAT scores, math and English California High School
Exit Examination (CAHSEE) scores, science CST scores (grades 5, 8, and 10), social science CST
scores (grades 8, 11, and world history), and the number of AP courses taken. All test scores are
normalized to be mean zero and standard deviation one at the grade-year level, except both SAT
and PSAT scores, which we place on a 600-2400 scale (PSAT is normally on a 60-240 scale, and
for some years, the SAT was on a 400-1600 scale). We top code days absent at 180 days per year,
and report log absences as the log of one plus the number of absences.
Described fully in sections 4.1 and 4.2, we compute test-score value-added measures using
math and English CST scores, and behavior value-added using log days absent, achievement GPA,
an indicator for suspensions, and an indicator for being held back. For elementary school teach-
ers, we compute learning-skills value-added using the three additional types of elementary school
GPAs. We reduce the dimensionality of both the inputs to the value-added variables and the value-
added variables themselves by creating equally weighted indices.
Our main outcome variables are measures of high school performance, including an indicator
for dropping out of high school, an indicator for taking the SAT, SAT scores, the three high school
GPA measures averaged from grades 9-12, math and English California high school exit-exam
scores, days suspended in grades 9-12, log absences in grades 9-12, and an indicator for being held
back in grades 9-12. We treat graduation as a supplemental measure because it is conditional on
enrolling in the LAUSD in 12th grade. The unit of observation in the main analyses data set is a
student-academic year, where the outcome is typically a measure of the student’s performance in
high school and the independent variables of interest are the three teacher value-added indices for
a student in a particular academic year. We focus on the pooled effect of teachers in grades 3, 4,
9
and 5, and thus the same student often appears in the main analysis three times.
Summary statistics for these data are shown in Table 1. Panel A shows summary statistics for
all students in grades 3-5 from the 2004 to 2010 school years. Panel B shows high school summary
statistics for the students in Panel A who attended high school in the LAUSD. Appendix Figure
A.2 shows retention rates by grade. While the dropout numbers seem high, they are consistent
with LAUSD’s official graduation rate which was 64% in 2005-2006 to 62% in 2009-2010. The
LAUSD dropout variable overestimates the dropout rate as a measure of people who never graduate
from any school in CA by a factor of about 1.5 because it includes both dropouts and students
who transferred to schools outside of the LAUSD. The actual LAUSD dropout rate, estimated by
the LAUSD, was 37.2 percent in 2010. The graduated variable is an overestimate of the actual
graduation rate because it is conditioned on entering the 12th grade.
4 Empirical Method
4.1 Estimating Teacher Value-Added
Let Si jt be a measure of student i’s test scores, behavior, or learning skills in academic-year t (in
teacher j’s class). For example, Si jt could be a standardized test score, an indicator for whether
the student was suspended, or a teacher’s assessment of a particular learning skill. The goal is to
estimate the effect of a teacher on several measures of students’ test scores, behavior, and learn-
ing skills. Recent research estimating teacher test-score value-added and its affects on long-term
outcomes has used slightly different estimation strategies (Chetty, Friedman, and Rockoff 2014b;
Jackson 2018; Rothstein 2017). Our approach combines elements of each, and we show in the
Appendix that the main results are robust across a range of estimation strategies.
For test-score value-added, we use the following estimation procedure. Although very similar,
a small but important adjustment is made when estimating behavior and learning-skills value-
added, which is discussed below. We construct value-added measures by first residualizing the
achievement measure, Si jt , by regressing it on a vector of controls, Xi jt , for lagged student achieve-
10
ment and the classroom environment using Equation (1). The baseline controls Xi jt include (1)
lags of a cubic polynomial of the student’s math test score, English test score, achievement GPA,
effort GPA, work and study habits GPA, and learning and social skills GPA; (2) lags of log days
absent, an indicator for suspensions, and an indicator for held back; (3) current English language
learner status; (4) a cubic polynomial of both class and grade level averages of lagged math test
score, English test score, achievement GPA, effort GPA work and study habits GPA, and learning
and social skills GPA; (5) class and grade level averages of lagged of log days absent, an indicator
for suspensions, an indicator for being held back, and English language learner status; (6) current
class size; and (7) grade fixed effects and year fixed effects. Each of the controls except current
English language learner status and current class size are interacted with grade fixed effects. In
middle and high school, our controls Xi jt are the same as in elementary school except we exclude
effort GPA, work and study habits GPA, and learning and social skills GPA controls for lack of
data.
Si jt = ΓXi jt + εi jt , (1)
εi jt = µ jt +αc + γit (2)
We assume the error term, εi jt , is an additively separable function of teacher quality (µ jt),
classroom shocks (αc), and student-year shocks (γit) as defined in Equation (2). This specification
of the error term, εi jt , is more flexible than one often used in the value-added literature that requires
a teacher’s quality be the same in each year this approach requires a stationarity assumption to
separately identify µ jt and αc.4
Let νi jt be the residualized student achievement computed as follows
νi jt = Si jt− Γ̂Xi jt . (3)
4See, assumption 1 of Chetty, Friedman, and Rockoff (2014a).
11
The residualization purges Si jt of measures of the prior achievement of each student, each student’s
class, and each student’s grade. For middle and high school students we obtain the residuals using
only students who had one teacher per subject for English or math.
We then take the mean of the residuals, ν̄ jt =1N
∑Ni=1 νi jt , by year for each teacher j. This
provides an estimate of the teacher’s value-added score which is a measure of their ability to af-
fect student achievement in each year t. It is unbaised as long as certain teachers do not tend to
receive students with relatively better or worse than average unobserved achievement, specifically,
if E [αc + γit | j] = E [αc + γit ] (Chetty, Friedman, and Rockoff 2011). Although this is a strong as-
sumption, it is plausible in this context because the value-added model includes extensive controls
for students’ prior achievement and behavior in school that have been shown to account for most
student sorting in test-score value-added models (Bacher-Hicks, Kane, and Staiger 2014; Chetty,
Friedman, and Rockoff 2014a). To help alleviate concerns about student sorting based on unob-
servable components of student achievement, in section 5.5, we check for forecast bias, examine
the effect of teacher value-added on predicted outcomes as a placebo test, and perform a quasi-
experimental analysis that uses teachers switching grades and schools. We also show our estimates
are robust to including controls for tracking in middle and high school.
We then predict teacher quality in year t with the teacher’s estimated value-added scores in the
surrounding years using this equation ν̂ jt = ∑t+as=t−a ψ̂sν̄ js1[s 6= t], where a equals six years for test
scores and five years for behavior and learning skills. This approach measures of teacher quality
with data from the surrounding years to avoid biasing estimates of the long-term effects of teacher
quality on student outcomes (Jacob, Lefgren, and Sims 2010). Including year t in the prediction
would likely bias the long-term estimates because unobservables in year t that affect any dimension
of student performance may also affect both the estimated value-added measure in year t and the
long-term outcomes. We allow the weight placed on the value-added measure, ψ̂s, to vary by the
number of years before or after year t. We estimate the weights by minimizing the mean-squared
error of the difference between ν̄ jt and predictions of ν̄ jt made using the teacher value-added
measures the years before and after t. In particular, by solving the following minimization prob-
12
lem, ψ ={ψt−a,...,ψt+a} ∑Jj(ν̄ jt−∑
t+as=t−a ψν̄ js1[s 6= t]
)2.5 This procedure produces leave-year-out,
jackknife, value-added predictions that allows teacher quality to change over time and shrinks the
value-added predictions to the mean through Bayesian shrinkage (Chetty, Friedman, and Rockoff
2014a).
We modify this procedure when we calculate non-test-score value-added measures by using
the lead of the achievement measure, Si j(t+1), as the outcome variable. This approach contrasts
with most of the non-test-score value-added literature that uses contemporaneous student outcomes
instead of outcomes in the next year, but is closely related to approaches used to calculate the value-
added of professors (Carrell and West 2010; Figlio, Schapiro, and Soter 2015). This approach
requires the main assumption – that teachers do not systematically receive students with relatively
better or worse unobserved achievement – holds for two years instead of just one. This assumption
in identical to that used in the value-added literature but is required for two years instead of one.
We use Si j(t+1) because using Si jt creates the potential for the non-test-score value-added mea-
sures to capture aspects of teacher behavior unrelated to teachers’ ability to affect students’ be-
havior or learning skills. For example, grades are likely affected not only by how much a teacher
helps a student learn and work diligently, but also by how strictly the teacher grades. Similarly,
suspensions are affected both by whether a teacher helps develop student behavior and how harshly
or leniently a teacher chooses to punish a student. These types of measurement error could lead to
biased estimates of the effect of teacher value-added on student outcomes.
A related concern is that teachers could directly affect long-term outcomes without affecting a
student’s behavior or learning skills. For example, if a teacher is more likely than other teachers
to recommend a student be held back, that student may be more likely to drop out of high school
even if the teacher actually has no effect on the student’s behavior or learning skills. This potential
direct effect could bias the effect of teacher value-added on long-term outcomes in the direction of
affecting long-term outcomes.
We remove bias from variation in teacher strictness (or leniency), and the direct effect of teach-
5See, Chetty, Friedman, and Rockoff (2014a) for additional details.
13
ers on long-term student outcomes, by using the lead of the student achievement measure (i.e.,
achievement in year t +1, rather than the measure of student achievement in year t):
Si j(t+1) = ΓXi jt + εi jt . (4)
This approach introduces noise to our estimates because it partially captures the effect of the
teacher in year t +1, but removes systematic bias from teachers evaluating their own students. In
addition, using student achievement in year t +1 makes it more difficult for teachers to manipulate
their behavior or learning-skills value-added. It does not eliminate all measurement problems as
there are a limited number of year t + 1 teachers evaluating each teacher’s students and some
measures have an inherent subjectivity that is not fixed using this approach. However, we show
later in the paper that the GPA and learning skills measures, Si jt , are highly correlated with long-
term student outcomes, GPA and learning skills measures value-added affects outcomes after year
t +1. which suggests these variables measure meaningful skills.
In order for this approach to work, teachers must be able to exert some long-term influence on
their students behavior. One possible channel is directly by helping students develop noncognitive
skills (e.g., Cunha, Heckman, Schennach 2010; Heckman, Pinto, and Savelyev 2013; and Kraft
2019). Teachers may also affect parents behaviors and beliefs, e.g., about the value of attending
school or their child’s performance in school, and induce parents to reduce absences or help their
children with their homework.6
4.2 Estimating the Long-Term Effects of Teacher Value-Added
Once we have leave-year-out estimates of teacher quality, v̂ jt , we ask how having either a higher-
or lower-quality teacher along some dimension of teacher quality affects a student in the long
6For example, in CA 42-56% of elementary school parents were contacted by their school about attendance ina 6th month period (AdCouncil 2015). Nationally, 90% of parents of 3-5 graders report attending parent-teacherconferences (McQuiggani and Megra 2017) and one of the topics LAUSD suggests teachers inform parents about isthe benefit of better attendance (LAUSD New Teacher Resource Guide). There is also evidence from three RCTs thatcommunication from teachers or schools to parents substantially increases attendance (Cook et al 2017; Kraft andRogers 2015, Robinson et al. 2018; Rogers and Feller 2018)
14
term. Let yi be a long-term outcome of interest such as whether a student is a high school dropout,
an indicator for taking the SAT, or a score on a test required for high school graduation. Let k
index the distinct leave-year-out value-added measures of test scores, behavior, or learning skills.
Elementary school students only have one teacher per year so we only have one measure of v̂ jkt
per student-year. Let ni jt be the number of classes the student, i, has with a teacher, j, in academic-
year, t. For a high school or middle school student, i, and subject s∈ {English, Math}, we construct
v̂sikt =
1
∑Jj=1 ni jt
∑Jj=1 ni jv̂s
i jkt , which is the mean value-added score weighted by the number of
classes the student has with the teacher.
We regress outcome, yi, on a number of value-added measures and our controls from equation
(1) and Xi jt which is the vector of baseline controls used in equation (1), which is defined in section
4.1, with the exception of variables that are not consistently available across years including work
and study habits GPA and learning and social skills GPA.7 The estimates of β̂k for each value-added
measure assess how each dimension of teacher quality affects the outcome of interest:
yi =K
∑k=1
βkν̂ jkt +ΓXi jt +ηi jt . (5)
We reduce the dimensionality of the estimates of teacher quality by constructing three indices
of the value-added variables. The first index is computed using teacher math and English test-
score value-added, which we call the test-score value-added, or θ̂ sjt . The second value-added index
is computed using value-added for suspensions, log days absent, GPA, and not progressing to the
next grade on time (i.e., held back), which we call the behavior value-added, or θ̂ bjt . The third
value-added index is computed using the value-added from effort GPA, work and study habits
GPA, and learning and social skills GPA, which we call learning-skills value-added, or θ̂ ljt . We
chose these three groups because they separate test scores from non-test scores, and because the
behavior value-added includes variables that are available for all grades, whereas the learning-skills
7In particular, the controls include lags of a cubic polynomial of the student’s math test score, English test score,GPA, and effort GPA; a cubic polynomials of lagged class and grade level means of each of those variables; and currentEnglish learner status and lagged log days absent, suspensions, and held back for the individual, the class average, andthe grade average. Each of these variables except English learner status is fully interacted with grade fixed effects anda control for class size is included.
15
value-added is available only for elementary school students. Our behavior value-added measures
for middle and high school students include measures for both their math and English teachers.
The indices are computed by summing the standardized value-added variables, recoded so each
has the same expected sign, and then standardizing the resulting index to be mean zero, standard
deviation one. In the Appendix, we show that the main results are robust to grouping GPA with
learning skills, using factor analysis to construct the three indices, and using exploratory factor
analysis to choose the factors and the factor load on each value-added measure.
We estimate the long-term effect of these value-added measures using the following specifica-
tion:
yi = βsθ̂
sjt +β
bθ̂
bjt +β
lθ̂
ljt +ΓXi jt +ηi jt . (6)
Where Xi jt is the vector of baseline controls used in equation (5). We also compare the estimates
from equation (6) with the estimates from a model that omits non-test-score value-added indices.
This comparison allows us to sign the bias from omitting non-test-score measures in papers that es-
timate the effect of teachers’ test-score value-added on long-term outcomes. If we find that β̂ s falls
when we move from a model that excludes θ̂ bjt and θ̂ l
jt to one that includes them, it suggests that
typical estimates of the long-term effects of test-score value-added are biased upward by omitted
measures of behavioral or noncognitive skills. Alternatively, if θ̂ bjt or θ̂ l
jt affect long-term out-
comes, and the estimate of β̂ s is unaffected by adding θ̂ bjt or θ̂ l
jt , the long-term effects of test-score
value-added may be unbiased, but estimates of the total effect of teachers on students is larger
than the effects found when using test-score value-added alone.8 We cluster the standard errors
at the high school level and show in Appendix Table A.1 that the standard errors not substantially
effected by bootstrapping over the estimation of the value-added scores described in Section 4.1
and estimation of the long-term effects using Equation (6).
Let tildes denote residualized student value-added indices, for example, θ̃ sjt = θ̂ s
jt − Γ̂X −
β̂ bθ̂ bjt− β̂ l θ̂ l
jt . To interpret the estimates in equation (6) as causal, we must assume cov(θ̃ sjt ,ηi jt) =
8We use the modal high school that students in each elementary school progress to so all students in a givenelementary school are in the same cluster.
16
cov(θ̃ bjt ,ηi jt) = cov(θ̃ l
jt ,ηi jt) = 0; the residualized leave-year-out predicted teacher value-added
indices and student unobservables that affect the outcome, yi, are uncorrelated. Although a strong
assumption, it does not rule out all sorting on unobservables. For example, suppose yi is an indi-
vidual’s SAT score and ηi jt is a scalar that captures the effect of an SAT preparation course (if any)
on an SAT score that is orthogonal to X . People who have unexplained variation in SAT prepa-
ration courses in high school could disproportionately be assigned to types of elementary school
teachers without biasing the results. It must only be the case that the sorting is unrelated to the ele-
mentary school teachers’ residualized value-added indices (Chetty, Friedman, and Rockoff 2011).
To help alleviate some of the concerns with this assumption, in section 5.5, we examine the effect
of teacher value-added on predicted outcomes as a placebo test and perform a quasi-experiment
analysis that uses teachers switching grades and schools.
In addition, there are reasons to believe that this approach is conservative. First, we find some-
what larger, although much less precisely estimated, effects using a quasi-experimental design that
uses variation in teachers switching between grades and schools. Second, we estimate smaller
effects than if we use the approach taken by Chetty, Friedman, and Rockoff (2014b).
We also extend this analysis in two ways. First, we examine the dynamic effects of a teacher
on student outcomes for years τ ∈ [0,1, ..,7]:
yi(t+τ) = βsθ̂
sjt +β
bθ̂
bjt +β
lθ̂
ljt +ΓXi jt +ηi jt . (7)
The model shows the extent to which the effect of teacher value-added on student outcomes persists
or fades over a number of years. Second, we assess the effects on long-term outcomes by grade to
see in which grades high-quality teachers have the most impact on students.
17
5 Results
5.1 Descriptive Results
5.1.1 Descriptive Relationships in Student Data
To assess whether multiple dimensions of teacher quality might matter for long-term outcomes,
we estimate the relationship between measures of student achievement, both with each other and
with long-term outcomes. Appendix Table A.2 shows bivariate correlations between each of the
measures of student achievement. English and math test scores are highly correlated. The rela-
tionship between test scores and students’ GPA, learning-skills GPA, and effort GPA is weaker
but the correlation still ranges from 0.45 to 0.68. The correlations of each of these variables with
attendance, days suspended, and being held back are substantially weaker, and suggests these vari-
ables largely capture different aspects of student achievement. These correlations suggest that test
scores, behavior, and learning skills are related, but that some room remains for them to have
an independent effect on long-term outcomes. Reducing the dimensionality of these variables by
separately computing test-score, behavior (i.e., attendance, days suspended, being held back, and
GPA), and learning-skills (i.e., learning-skills GPA and effort GPA) indices, as described in section
4.2, yields correlations between 0.46 and 0.55 (Table 2).
Next, we assess whether these measures of student achievement are related to long-term out-
comes, conditional on the same set of controls we use to compute the value-added measures. En-
glish and math test scores, GPA, learning-skills GPA, suspensions, log days absent, and held back
measured in grades 3 through 5 typically (but not always) have a statistically significant (or in one
case marginally significant) relationship with high school outcomes (Appendix Table A.3). After
reducing the dimensionality to the three indices of student achievement, test scores, behavior, and
learning skills measured in grades 3 through 5 nearly always have a statistically significant effect
on high school outcomes (Table 3). For many of the high school outcomes, behavior and learning
skills are as predictive of the outcome as test scores.
These results are consistent with test scores, behavior, and learning skills each independently
18
affecting long-term outcomes. However, despite the fact that we control for a wide range of mea-
sures of student achievement, these estimates may be biased because of unobservables. Conse-
quently, these results may not hold in situations in which there is exogenous variation in students’
test scores, behavior, and learning-skills achievement. To address this concern, we move to a
teacher value-added framework in which omitted variables are less likely to bias the results.
5.1.2 Descriptive Relationships in Teacher Value-Added Data
We compute teacher value-added as described in section 4.1. Appendix Table A.4 shows the re-
lationship between the value-added measures. English and math test-score value-added are highly
correlated. The correlations between test-score value-added and all other variables are much
weaker, but they are positively correlated with GPA, effort GPA, and learning-skills GPA, which
have correlations between 0.14 and 0.20. The GPA, effort GPA, and learning-skills GPA value-
added are highly correlated with each other. The three value-added measures of student behavior
– log absences, days suspended, and held back – are all weakly correlated with each other, test
scores, and GPA measures. These correlations suggest that math and English test-score measures
of teacher quality are closely related, as are GPA-based measures of teacher quality, whereas the
ability to influence student behavior relates less closely. Table 4 shows similar results. The correla-
tion between test-score value-added and behavior value-added is 0.15, the correlation of test-score
value-added with learning-skills value-added is 0.17, and the correlation of behavior value-added
with learning-skills value-added is 0.46.
5.2 Effects of Teacher Quality on Long-Term Outcomes
5.2.1 Single Value-Added Effects
Figures 2 through 4 show the effect of teachers’ test-score, behavior, and learning-skills value-
added individually on each high school outcome, conditional on the set of controls used to compute
value-added measures (also see Appendix Table A.5 for the results in this subsection reported in a
19
regression table). The plotted points show the relationship between the mean residualized outcome
and the mean residualized value-added variables (with the unconditional mean of the outcome and
value-added variables added back in) for 20 equally sized bins of teacher value-added measures.
The coefficients and standard errors reported in the figures are from an OLS regression, using the
micro data, of the outcome variable on the value-added variable, conditional on the same set of
controls.
Figure 2 shows that students with better teachers in grades 3 to 5, as measured by the test-score
value-added, score significantly higher on the SAT, have marginally significantly higher effort
GPAs and significantly higher cooperation GPAs, and score significantly higher on the high school
exit exams. We find no significant effects on dropping out of high school, taking the SAT, GPA,
being held back, log days absent, or being suspended. These results are consistent with the existing
literature that shows benefits in adulthood from higher test-score value-added teachers, although
research that demonstrates positive effects of elementary school teachers on high school outcomes
is rare (Rothstein 2017).
Figure 3 shows the effect of teachers’ behavior value-added on each outcome. We observe
at least marginally statistically significant effects in the expected direction on all of the outcome
variables except SAT score and English exit exam score. This indicates that in the absence of test-
score value-added, having a teacher with a higher behavior valued-added impacts the high school
outcomes of students in a meaningful way. Figure 4 shows the effect of teachers’ learning-skills
value-added on each outcome. We find less evidence of an effect than for the other two value-added
measures. The coefficient on the learning-skills value-added typically has the expected sign, but the
only marginally significant effect is on days suspended. The confidence intervals are sufficiently
small to reject moderate effects, but for several of the outcomes we cannot reject the effects of
he same size as the behavior value-added estimates. These results suggest that elementary school
teachers affect students’ long-term outcomes by increasing student achievement as measured by
both test-score and non-test-score data.
Comparing the magnitudes across the analyses, we tend to find that test-score value-added has
20
a large effect on outcomes that have more cognitive content than the behavior or learning-skills
value-added, whereas the pattern of results is reversed for outcomes that have more noncognitive
content. For example, having a teacher with a standard deviation higher test-score value-added
increases the math high school exit exam score by 0.023 standard deviations, whereas the increase
for behavior value-added is 0.014 standard deviations, and the statistically insignificant increase
for the learning-skills value-added is 0.004 standard deviations. However, the effect of having
a teacher with a standard deviation higher test-score value-added on days suspended is less than
0.001, whereas behavior and learning-skills value-added both reduce days suspended by 0.003
days, a 2 percent decrease.
The test-score value-added estimates appear to have two sets of potential nonlinear effects.
First, the effect of test-score value-added on all three GPA measures is positive until teachers be-
come above average, and then the relationship is, if anything, negative. Second, there is suggestive
evidence that the top vingtile or two of the test-score value-added distribution has a smaller effect
on several outcomes than would be predicted from the rest of the test score value-added distribu-
tion (Figure 2). Chetty, Friedman, and Rockoff (2014b) find a similar anomaly in their 4th-8th
grade test-score value-added results, and drop the top 1 percent of teachers because of evidence
of “test manipulation.” We leave those teachers in, although including them biases the effects of
test-score value-added toward zero if “test manipulation” exists. We find less evidence of non-
monotonicities for both behavior and learning-skills value-added, and outliers in the top vingtile
are less common. One explanation for this finding is that because non-test-score value-added mea-
sures are constructed using student achievement in year t + 1, teachers are unable to manipulate
their non-test-score value-added measure unless they influence the actions of their students’ teach-
ers in the subsequent year. Supporting this explanation, we find that when Figure 2 is recreated
with value-added measures using test scores in t +1 instead of year t (see Appendix Figure A.3),
these nonlinearities no longer exist.
Taken together, these results suggest that multiple components of teacher quality affect long-
term outcomes. Our findings also indicate that in situations in which no test-score data are avail-
21
able, but other administrative data such as grades, attendance, suspensions, and held back are
available, creating estimates of teacher quality that are associated with long-term benefits to stu-
dents is possible. Some evidence also suggests non-test-score value-added measures calculated
using our approach are less prone to manipulation by teachers, although they might begin to be
manipulated if used in high-stakes settings.
5.2.2 Multivariate Value-Added Effects
Now that we have found that two of the dimensions of teacher quality affect high school outcomes,
we can determine whether more than one value-added measure independently affects long-term
outcomes. Significant effects of more than one value-added measure would suggest that teacher
quality is multidimensional in a way that both matters for long-term outcomes and is measurable
using a value-added approach. In addition, this analysis informs the extent to which the long-term
effects of test score value-added measures are driven by teachers’ affect on behavior and learning
skills.
Table 5 shows the effect that each of the three elementary school value-added measures have on
high school outcomes in an OLS regression in which all three value-added measures are included
simultaneously along with the baseline controls (equation 6). Including behavior and learning-
skills value-added only slightly affects the coefficients on the test-score value-added measures.
For example, the coefficient in the SAT-score regression falls from 6.39 to 6.24 SAT points (or is
a constant 0.021 standard deviations), the coefficient in the math high school exit exam regression
falls from 0.023 to 0.022 standard deviations, and the coefficient in the high school GPA regression
falls from 0.004 to 0.002 GPA points. These results indicate that the long-term effects of test-score
value-added are likely not driven by teachers’ effects on students’ behavior and learning skills that
are correlated with test-score value-added.
The effects of behavior value-added on most outcomes is also not affected substantially by
conditioning on the test-score and learning-skills value-added. Behavior value-added picks up a
dimension of teacher quality that is largely unrelated to the other two value-added measures and
22
that matters for long-term outcomes. In addition, Appendix Table A.6 shows that there is no ev-
idence of an interactive effect between an elementary school teacher’s test-score and behavioral
value-added on students’ high school outcomes with coefficient estimates that suggest any interac-
tion effect is very close to zero.
Adding the other value-added measures does not affect the evidence for an independent effect
of teachers on long-term outcomes through learning skills. None of the coefficients on the learning-
skills value-added in Table 5 are statistically significant. For some of these outcomes, conditioning
on test score and behavior value-added results in the learning-skills value-added coefficients no
longer has the expected sign. The confidence intervals often, but do not always exclude effects of
the magnitude of the coefficients on behavior value-added.
This approach assumes teacher quality is not grade specific. We expect that if a competent of
teacher quality is grade specific we will tend to underestimate the long-term effect of teachers on
students because of measurement error in the independent variable of interest. To check this we
estimate grade-specific measures of teacher value-added. We find that while teacher value-added
measured for the same teacher in different grades is closely related, the slope is well below one
(Appendix Figure A.4). We then re-estimate the effect of teachers on long-term outcomes using
grade-specific measures of teacher value-added (Appendix Table A.7). The point estimates test-
score value-added increase by a factor of at least two for many of the outcomes and are newly
significant for drop out and achievement GPA. The point estimates for behavior-value generally
also increase (although not by as much) and are newly significant for SAT score and the English
exit exam, although, the effect on taking the SAT is no longer statistically significant. These result
suggest there is some grade-specific component of teacher ability and by assuming teacher quality
is constant across grades we tend to under estimate the effect of teachers on long-term outcomes.
Figure 5 shows how the value-added measures affect a number of outcomes that can be tracked
over time beginning in elementary school. The effect of test-score value-added on test scores shows
the expected pattern of results. Having a standard deviation higher test-score value-added teacher
has a large effect on math and English test scores in year zero that largely, but not completely, fades
23
out over the next seven years. Behavior value-added and learning-skills value-added show less ev-
idence of fade out, but our approach to constructing these variables should result in measures with
less fade out than test-score value-added. By measuring behavior and learning-skills value-added
using the effect of a teacher this year on student achievement in the next year, we are effectively
removing the first year of fade out from the estimates. In addition, because some of the student
achievement variables are grades, and students may be graded on a curve, seeing little effect of
behavior and learning-skills value-added on GPA measures in year zero would not be surprising.
The size of the long-term results can be interpreted using the cross-sectional relationship be-
tween test scores and earnings. Hanushek and Wossman (2008) find consistent evidence that in
the cross section, a standard deviation increase in test scores at the end of high school increases
earnings by 12 percent (Lazear 2003; Mulligan 1999; Murnane et al. 2000). Approximately the
same relationship holds between elementary and middle school test scores and earnings. Chetty,
Friedman, and Rockoff (2014b) show that direct estimates of a standard deviation increase in test-
score value-added for 4th-8th graders on earnings, and a back of the envelope estimate using the
cross-sectional relationship between primary school test scores and earnings, yield estimates of
the effect of teachers on earnings in the 1.3 to 1.5 percent range. We find effects of approximately
the same size or larger using the effect of teachers on contemporaneous test scores. If we instead
use the effect of having a standard deviation higher test-score value-added teacher in elementary
school on high school test scores, which ranges from 0.016 to 0.022 standard deviations, the esti-
mated increase in earnings would be 0.23 percent. This much smaller effect is likely driven by the
fade-out in the effect of teachers on test scores over time.
Combined, these results indicate that teacher quality is multidimensional in a way that matters
for long-term outcomes. Importantly, this multidimensionality can be measured using a combina-
tion of test scores and other data that schools routinely collect.
24
5.2.3 Multiple Hypothesis Testing
In this section we check whether the statistical significance of the results is robust to adjusting for
multiple hypothesis testing using several approaches. For the first approach, we simply constructed
indices of three families of outcomes: test scores (math and English CAHSEE)9, GPA (GPA, effort
GPA, cooperation GPA), and behaviors (took SAT, days suspended, log absences, held back). This
approach greatly reduces the number of hypotheses tested, makes use of the information in all our
outcomes, and it is much less computationally intensive than other methods. Appendix Table A.8
shows single factor results using the index outcomes. Test-score value-added significantly affects
the test score and GPA indices, behavior value-added significantly affects all three indices, and
learning skills has no significant effect on these outcomes. These results suggest the long-term
effects of teachers are robust to reducing the number of hypothesis tests. We also apply the index
approach to our main specification Table 5 where we include all three measures of teacher value-
added simultaneously, and find similar results except the effect of behavior value-added on the test
score index is not statistically significant (Appendix Table A.9).
Next, we use a number of methods to adjust for multiple hypothesis testing for each of our
independent variables of interest. We recomputed the p-values using the Bonferroni-Holm, Sidak-
Holm, and Romano-Wolf step-down methods along with the Westfall and Young (1993) algo-
rithm.10 We grouped the outcomes into three families of hypothesis tests: test scores, grades,
and all other outcomes. We then adjusted for multiple hypothesis testing within each family and
independent variable of interest. Appendix Table A.10 shows the results for each of these four
approaches (using 10,000 bootstrap draws where applicable). The p-values for test-score value-
added increase, however, the statistical significance of the results is generally robust to allowing
for multiple hypothesis testing. Although, effort GPA is no longer marginally significant under
most adjustments. Similarly, for behavior value-added the statistical significance of the results are
generally robust to these approaches with the exception that the effect on math test scores is now
9We do not include the SAT score because it would either made our sample much small or required us to imputea SAT score for a large fraction of students.
10We use code from Jones et al. (2018) and Clarke et al. (2019).
25
generally only marginally significant and held back is generally statistically insignificant
If instead of adjusting for multiple hypothesis tests by family, we instead adjust across all out-
comes, the statistical significance of test-score value-added is the same as using p-values adjusted
within family (Appendix Table A.11). For behavior value-added, the effect on math test scores
and cooperation GPA are no longer significant or marginally significant. The other outcomes that
were statistically significant remain at least marginally significant. We also re-computed p-values
for the index outcomes, allowing for multiple hypothesis testing across all three outcomes. The
statistical significance is unaffected except the p-values for test-score value-added on GPA are only
marginally significant for some adjustments (Appendix Table A.11).
Broadly these results show that we detect long-term, significant effects of having a higher
test-score and behavior value-added teacher and our results are not driven by multiple hypothesis
testing. However, the significance of some individual outcomes are not robust to these adjustments.
Notably the evidence that behavior-value added has long-term effects on test scores, conditional
on test score value-added, is weaker, although behavior value-added still significantly affects other
outcomes.
5.3 Teacher Selection Policies
Policies that use teachers’ test-score value-added to hire, fire, or incentivize teachers have been
widely criticized because making decisions using only one (potentially gameable) dimension of
teacher quality is considered unfair, or even counterproductive. However, the effect on long-term
outcomes of having higher test-score and behavior value-added teachers implies that policies that
shift the distribution of teacher quality upward in these dimensions benefit students. In comparison
to just using test-score value-added, we show that using multiple dimensions of teacher quality in
teacher-removal policies substantially improves the measurement of teacher quality and students’
long-term outcomes.
Figure 6 shows scatter plots of teacher quality as measured by value-added in a given year. The
dashed lines show the 5th percentile of teachers for a given value-added measure. The first panel
26
plots test-score and behavior value-added, and shows that although both dimensions of teacher
quality are positively correlated, the correlation is relatively weak, and some teachers who perform
poorly as measured by test-score value-added perform well on the behavior value-added dimen-
sion. For example, the majority of teachers who are in the bottom 5 percent of teachers as measured
by the test-score value-added are not in the bottom 5 percent of teachers as measured by the be-
havior value-added. Therefore, a linear combination of a teacher’s value-added measures might be
a better predictor of teacher quality and measure for teacher-removal policies.
One way to assess the value of using multiple measures of teacher quality is to ask to what ex-
tent students’ long-term outcomes could be improved under a policy that replaces a school district’s
bottom 5 percent of teachers with average teachers as measured by only test-score value-added ver-
sus different linear combinations of the three value-added measures. Panel A of Table 6 shows the
effect on a student’s high school outcomes of being assigned an average teacher instead of a teacher
in the bottom 5 percent of teachers as measured by a teacher’s true value-added (realized value-
added ex post, ν̄ jt). This panel shows the upper bound on the effects of the teacher-removal policy.
The simulation uses estimated effects of teacher value-added on high school outcomes (Figures 2
through 4 and Table 5) and the within-teacher correlations between the three teacher value-added
measures (Table 4). Standard errors for the estimated forecasts are shown in parentheses.
Each cell in row 1 shows the effect on students’ high school outcomes if their bottom 5 percent
test-score value-added elementary school teacher was replaced by an average teacher. For example,
the students that the policy would affect (about 5 percent) would see their SAT scores increase by
13 points and their high school GPA by 0.008 GPA points. Row 2 shows the effect on students’
high school outcomes if their bottom 5 percent behavior value-added elementary school teacher
was replaced by an average teacher. The benefits from using behavior value-added are comparable
to using test-score value-added, and in some cases, the benefits are larger.
Row 3 uses the average of teachers’ test-score and behavior value-added. When this combined
measure is used, students affected by the policy would see beneficial effects on all but one of the
high school outcomes. Row 5 shows the percent change in students’ outcomes if the replacement of
27
the bottom 5 percent of teachers is performed using the average of teachers’ test-score and behavior
value-added instead of just teachers’ test-score value-added. There is over a 100 percent increase
in the beneficial effects on students for dropping out of high school, taking the SAT, GPA, effort
GPA, days suspended, log absences, and held back. Importantly, these gains are accompanied by
only small decreases in English exit exam test scores and SAT scores.
Row 4 uses a maximization procedure to choose the optimal weights to be placed on a lin-
ear combination of teachers’ test-score, behavior, and learning-skills value-added to determine the
bottom 5 percent of teachers for the indicated outcome variable. The optimal weights vary depend-
ing on the outcome variable, so simultaneously improving all outcomes by the calculated amount
would not be possible. However, for most of the outcomes, a policy that uses the optimal weights
for a particular outcome only slightly outperforms a simple policy that places equal weights on
test-score and behavior value-added.
Panel B of Table 6 shows analogous results using teachers’ estimated value-added based on the
three previous years of student data. These results reflect the potential student gains if the teacher
removal policy were to be implemented for teachers who had taught for three years. Similar to
Panel A, student gains can be obtained if both test-score and behavior value-added are used to
make the teacher-removal decision. Because the autocorrelation between years for the behavior
measure is smaller than for the test-score measure (Appendix Figure A.5), the percent gain from
using both value-added measures instead of just the test-score value-added is smaller.
These results suggest that the dimensions of teacher quality captured by behavior value-added
are roughly as important for long-term outcomes as test-score value-added, and in combination,
can benefit student. Most of these benefits do not require new tests or assessments – only a new
use for data that schools already collect.
5.4 Which Behaviors Matter for Long-Term Outcomes?
Behavior value-added includes several weakly correlated value-added measures, some of which
may matter more than others for long-term outcomes. A straightforward way to assess which
28
variables matter most is to regress high school outcomes on the full set of value-added measures
that we use to construct the lower dimensional representation of teacher quality plus the usual set
of controls. We regress each outcome on each component of test score, behavior, and learning-
skills value-added on each outcome, conditional on our baseline controls so each cell in the table
is from a separate regression (like in Appendix Table A.5).
The test-score value-added measures have significant effects on SAT taking, the GPA measures
(English test-score value-added), and high school exit exams Appendix Table A.12. GPA VA
has significant effects on SAT taking and the English exit exam, and a similar magnitude but
insignificant effect on the math exit exam. Two other components of behavior value-added, absence
and suspension-value added, have significant effects on outcomes with both cognitive and non-
cognitive content. There is less evidence of an effect of held back value-added, but the held back
confidence interval general cannot reject effects of the magnitude of the absence or suspension-
value added measures. We generally do not see evidence of effect of the various components
of learning skills value-added. The coefficients typically have the expected sign, but the only
significant effect is of effort GPA on SAT taking. These estimates are somewhat less precise than,
e.g., the test score value added outcomes, and we often cannot reject effects of the magnitude of the
test score effects. Although the estimates tend to be slightly more precise than absences and days
suspended VA. If we instead include all value-added measures at once, we find broadly similar
effects except interpreting test score and GPA based measured results is more difficult because the
test score value-added measures are highly correlated, as are the GPA based value-added measures
(Appendix Table A.13). The results suggests that the behavior value-added results are driven
primarily by teachers’ effects on suspensions and absences.
Another way to illustrate this is to move GPA value-added from behavior value-added to
learning-skills value-added, and conduct the main analysis again. The new behavior value-added
constructed only from absences, suspension, and being held back has a significant or marginally
significant effect in the expected direction on six high school outcomes (Appendix Table A.14).
The new GPA-based value-added has a marginally significant effect on only taking the SAT. The
29
point estimates of the GPA-based value-added are often smaller than the significant effects of the
other value-added measures, and although the confidence intervals generally cannot reject effects
of the magnitude of the behavior value-added estimates.
These results suggest multiple dimensions through which teachers affect long-term student
outcomes, one that is closely related to increased performance on tests and others related to reduced
absences and suspensions. The abilities reflected in achievement GPA, effort GPA, and learning-
skills GPA matter for long-term outcomes, but the portion of these abilities that teachers are able
to affect is largely captured by test scores and the ability to keep the students in the classroom.
From a policy perspective, these results suggest that finding ways to keep children in the class-
room may have large benefits, e.g., the effect of reducing absence value-added on SAT scores is
about the same size as the sum of the effects of the two test score value-added measures. There
is also little of evidence that lack-of on-time progression in elementary school results in worse
high school outcomes, although our held-back value-added measure may be less informative as
our other value-added measures because most elementary school students progress on-time to the
next grade.
5.5 Checking for Bias in Long-Term Effects
We conduct four analyses to look for evidence of bias in the estimates of the long-term effects of
teachers. Consistent with Chetty, Friedman, and Rockoff (2014a) results for 4-8th grade test-score
value-added, most tests show no evidence of bias, and the magnitude of the bias in the remaining
tests is sufficiently small such that it does not substantially affect our conclusions.
First, we show that the value-added measures are forecast unbiased. Specifically, all but one
of the leave-year-out value-added variables cause an increase in the corresponding residualized
achievement variable that is statistically indistinguishable from one (Appendix Table A.15). Al-
though this is a weak test, failing this test would be problematic. Only math test scores are statisti-
cally different than one, for which a one unit increase in the math test-score value-added causes a
0.99 increase in math test scores. The confidence intervals are relatively tight for most outcomes,
30
although absences, suspensions, and held back are exceptions.
Second, we show that after conditioning on the main controls, students expected to perform bet-
ter in elementary school based on their twice-lagged values of achievement are largely not sorting
to higher value-added teachers. The estimated forecast bias from selection on student characteris-
tics is between−1.6 to 1.3 percent, which is smaller than Chetty, Friedman, and Rockoff’s (2014a)
point estimate of 2.2 percent for test-score value-added (Rothstein 2007). The forecast bias is only
marginally significant for GPA with a point estimates of 1.3 percent (Appendix Table A.15). An
analogous calculation using predicted high school outcomes from the twice-lagged values of the
control variables shows no evidence of upward bias. The only significant point estimates are for
behavior value-added, but each suggests that better students are sorted to worse teachers (Table 7).
Third, we aggregate these data to the school-grade-year level and estimate long-term effects
using quasi-experimental variation in teacher value-added caused by teachers switching between
grades or schools. The analysis removes variation in teacher value-added caused by students sort-
ing to teachers within a grade. Following Chetty, Friedman, and Rockoff (2014b), we regress
changes in school-grade-year high school outcomes on changes in the mean teacher value-added
weighted by the number of students.11 Appendix Table A.16 shows that the signs on the estimated
coefficients are generally consistent with the main results in Table 5, and the point estimates tend
to be larger. However, the estimates are much less precise. Despite this loss in statistical power, we
observe a significant effect of test-score value-added on math exit exams, and either significant or
marginally significant effects in the expected direction of behavior value-added for the three GPA
outcomes. As in the main table, learning-skills value-added is often wrong-signed and for some
outcomes are statistically significant.
Finally, we address the effects of tracking in middle and high school. During the period we use
to calculate value-added scores, students are tracked in math classes starting in 8th grade and they
have the opportunity to take more advanced English classes starting in 11th grade and students
11The sample is limited to cases in which we have value-added measures for all teachers in a given school-gradeyear in two consecutive years, to the subset of students for which we have both the long-term outcome variable and ateacher value-added measure, and to value-added measures that can be computed leaving out both year t and t−1.
31
may take elective English courses in both middle and high school. To the extent our interest is
estimating the benefit to an individual elementary school student of being shifted into a higher
value-added teacher’s classroom, as a parent might want to know, our estimates are not biased by
future tracking. Second, if a better teacher in elementary school results in a student being prepared
to take rigorous courses in high school or increases the quality of the students peers by directly
increasing the peers performance, we view that as an outcome of interest rather than a source of
bias. This is particularly true in the LAUSD where a large number of students struggle in rigorous
coursework and do not progress from 9th to 10th grade on time. However, some benefits of a
better elementary school teacher are zero sum, so our estimates may be too large for certain policy
experiments.
To help address this concern we re-estimate our main results from Table 5 after adding controls
for tracking including fixed effects for the number future middle and high school English electives,
English helper classes, English AP classes, math classes on each of the three math pathways, math
helper or below grade level classes, and AP math courses. Including these controls likely result in
us over-controlling for tracking. We find that the effect of a better teacher in elementary school on
most high school outcomes falls slightly and the statistical significance of the effects is generally
unaffected (Appendix Table A.17). For example, the effect of test-score value-added on math exit
exams falls from 0.22 to 0.20. Two exceptions to this pattern are the effect of test-score value-added
on LAUSD dropout which becomes marginally significant and the effect of test-score value-added
on effort GPA, which is no longer marginally significant. Additionally, Figure 5 provides some
evidence that tracking is not driving our results. Tracking starts 3-5 years after the current year for
math and 5-8 years after the current year for other occurs. There do not appear to be appreciable
changes in the effects of a better teacher in Figure 5 in the post-tracking years.
5.6 Robustness Checks
We conduct a number of robustness checks in which we use alternative approaches to constructing
the value-added indices, different specifications to estimate the long-term effects, and additional
32
high school outcomes. The results of the robustness checks are qualitatively consistent with the
main results.
Appendix Table A.18 reports the effect of teacher value-added on additional high school out-
comes such as graduating from the LAUSD if enrolled in the LAUSD in 12th grade, took the PSAT,
PSAT score, 11th grade English CST score (the last grade the CST is administered), 11th grade
math CST score, 8th grade science CST score, 10th grade science CST score, 8th grade social
studies CST score, 11 grade social studies CST score, world history CST score, and the number
of AP courses. We see significant or marginally significant effects of test-score value-added on
all outcomes except LAUSD graduation and took the PSAT. Test-score value-added affects test-
score outcomes by between 0.013 and 0.018 standard deviations, and does not vary noticeably by
subject. Having a high test-score value-added teacher in elementary school increases the long-
term performance across a number of subjects, not just English and math. The coefficients on
behavior value-added are typically the expected sign but are only marginally significant for one
outcome, whereas the coefficients on the learning-skills value-added are typically wrong-signed
and are statistically significant for one outcome. There is also little evidence that teacher value-
added in grades 3 through 5 affects whether students leave the school district in subsequent years
(Appendix Figure A.6).
We now show that the results are robust to a number of changes to our approach to computing
the value-added indices and long-term effects. The main results are larger and more often statisti-
cally significant if we follow Chetty, Friedman, and Rockoff (2014b) when computing value-added
measures by residualizing the achievement data using within-teacher variation in the controls (Ap-
pendix Table A.19). The effects are even larger if we use Chetty, Friedman, and Rockoff’s (2014b)
approach to computing long-term effects by residualizing the outcome variables using within-
teacher variation in the controls and then regressing the residualized high school outcomes on
teacher value-added with no controls (Appendix Table A.20). There is also more evidence of an
effect of learning skills value-added on long-term outcomes including significant effects in the
expected direction on dropout, and marginally significant effects on SAT score and English high
33
school exit exam score. The main results are essentially unchanged if we use factor analysis to
construct the three indices for teacher ability (Appendix Tables A.21 and A.22), and weaker for
the non-test score factor if we use exploratory factor analysis to construct two orthogonal factors
(Appendix Tables A.23 and A.24). Finally, we convert the test score value-added into deciles and
test score outcomes into percentiles, which means we use ordinal, rather than cardinal measures of
teacher value-added and test score outcomes. We continue to find significant effects of test score-
value-added on exit exam test scores and significant or marginally significant effects of behavior
value-added on took SAT, math test scores, absences, and held back (Appendix Table A.25).
Appendix Table A.26 shows that if we remove behavior value-added from the main analysis,
learning-skills value-added no longer has one wrong-signed coefficient. This suggests that part of
the reason for the unintuitive results for learning-skills value-added is that behavior value-added
and learning-skills value-added are moderately correlated.
Another concern with our interpretations of the results is that the effect of teacher value-added
on long-term outcomes could be driven by the direct effect of teacher actions like suspensions or
retaining a student on both teacher value-added and the student’s long-term outcomes. While this
is likely happening to some degree, it is unlikely this effect is large enough to influence our main
results because both types of events are rare for elementary school students. In any given, year
only about 0.7% of elementary school students are held back and the average elementary school
student is suspended for 0.05 days (Table 1). To check the magnitude of this effect we replicate our
main table excluding students who were held back or suspended. We find only slight changes in
our results (Appendix Table A.27). The magnitudes are very similar to the baseline results in Table
5, and there are only minor changes in the statistical significance of the coefficients of interest.
This suggests the direct effect of suspension and grade retention plays a small role in our estimated
results.
We also test the sensitivity of our results to using year t + 1 measures of the behavior and
learning skills to estimate behavior and learning skills value-added. First we re-estimate test-score
value-added using year t + 1 test scores, and then we replicate our main results from Table 5
34
using test-score, behavior and learning skills value-added all estimated using year t +1 measures.
Generally, we find larger long-term effects of test-score value-added estimated using t+1 data than
year t data. There are also slight decreases in the effect of behavior value-added on some long-
term outcomes (Appendix Table A.28). We also re-estimate behavior and learning skills value-
added using year t data than year t + 1. The tests score VA measures are essentially unchanged
while the estimates of behavior value-added are generally smaller and the effects on SAT taking,
math test scores, and grade retention are no-longer statistically significant. Although, there are
still significant or marginally significant results on the three GPA measures and log days absent.
Interestingly, the learning skills estimates tend to be slightly larger in absolute value and the results
suggests higher learning skills value added leads to significantly or marginally significantly worse
results for six high school outcomes (Appendix Table A.29).
We also check if our results are sensitive to using GPA measured in z-scores rather than grade
points both when computing value-added scores and as long-term outcomes (Appendix Table
A.30). Converting the GPA measures to z-scores before estimating teacher VA had no effect on the
relationship between the teacher value-added indices and the GPA outcomes in points. The statisti-
cal significance GPA outcomes in SD units is the same as when they are measured in grade points.
The change in magnitudes is what would be expected give the SD of the high school outcomes
reported in Table 1.
6 Applications of Non-Test-Score Measures
We take the general approach from the last section to demonstrate that non-test-score measures
of achievement are useful for answering additional questions related to the effects of teachers on
students. We first examine teacher effects over the educational life cycle, and ask to what extent
there could be long-term gains from moving high value-added teachers between grades and what
are the cumulative benefits from increasing teacher quality. We then take the approach used to
construct non-test-score value-added measures to compute GPA value-added for specific subjects
35
in order to test the long-term value of having a better teacher in different subjects. This analysis
could also be interpreted as suggesting how long-term student performance could be obtained from
reallocating time between subjects.
6.1 Implications for Long-Term Effects of Teacher Quality over the Educa-
tional Life Cycle
The approach in this analysis is to compute the test-score and behavior value-added for teachers in
grades 3 through 12, and ask how having a standard deviation better teacher in each grade affects
long-term outcomes as measured in 11th or 12th grade.12 We do not compute learning-skills value-
added because we do not have learning-skills data for middle and high school students. Previous
work on long-term teacher effects by grade has estimated the effects of test-score value-added
for grades 4 through 8 (Chetty, Friedman, and Rockoff 2014b), and test-score and non-test-score
value-added for grade 9 in North Carolina (Jackson 2018).
Figure 7 reports the results of this analysis for outcome variables measured as late as possible
in a student’s career. In each graph, we also report the sum of the coefficients across all grades.
With some assumptions, particularly no tracking of students and no diminishing returns to having
consecutive high quality teachers (Kinsler 2016), this sum reflects an upper bound on the cumu-
lative effect of having a standard deviation higher test-score or behavior value-added teacher in
each grade from 3rd through 12th grade. If tracking students plays a large role, or if diminishing
returns exist, this sum overestimates the cumulative effect and is an upper bound. However, Chetty,
Friedman, and Rockoff (2014b) find evidence for only a small amount of tracking, conditional on
controls for lagged achievement in grades 4-8, and we include a robustness check that directly
controls for tracking.
We find that having a standard deviation higher test-score teacher in grades 3 through 12 has a
beneficial effect on taking the SAT, SAT scores, and math and English test scores. The cumulative
12We cannot compute teacher value-added in 12th grade, so value-added measures for teachers in 12th grade useestimates of teacher quality in earlier grades.
36
effect for each of these outcomes is quite large. Using the cross-sectional relationship between test
scores and earnings, and the cumulative effects on math and English tests scores, having a standard
deviation higher test-score value-added teacher in each grade increases a student’s adult earnings
by 2.7 to 5.2 percent.
We also find that having a standard deviation higher behavior value-added teacher in grades 3
through 12 has a large beneficial effect on dropping out of high school, graduation, taking the SAT,
the three GPA measures, absences, suspensions, and grade retention. For example, the cumulative
effects suggest that having a standard deviation higher behavior value-added teacher in each grade
decreases the likelihood of dropping out of high school by 9.0 percentage points. Once adjusted for
dropping out of high school being over estimated due to students leaving the LASUD, this effect
is still a 5.9 percentage point decrease.
We also show the magnitude of the cumulative effects of better teachers in Figure 7 is not is
driven by tracking. We re-estimate the value-added scores for middle and high school after adding
controls for contemporaneous tracking. Then we re-estimate the grade-by-grade long-term effects
in the elementary school controlling for future tracking. For middle and high school grades we
use the value-added measures estimated with tracking controls, and when estimating the effect of
teacher-value added on longer term outcomes we include controls for contemporaneous tracking
(Appendix Figure A.7). We control for future tracking using the same approach as in Appendix
Table A.17, and control for contemporaneous tracking with fixed effects for the number of English
elective, English helper classes, English AP classes, classes on each of the three math pathways,
math helper or below grade level classes, and AP math classes in each year. The effects of a
better teacher on each outcome are generally smaller than in Figure 7, but only slightly and our
conclusions are unaffected. For example, cumulative effects on SAT taking fall from 8.1 and
8.4 percentage points for test score and behavior value-added, to 7.8 and 7.7 percentage points,
respectively.
We re-estimate the cumulative effects using an alternative approach where we compute the
average test-score and behavior-value added for each student’s teachers across grades 4-11, then
37
regress the outcomes reported in Figure 7 on the average test-score and behavior-value added and
controls measured in fourth grade.13 The support of the distribution includes plus or minus one
standard deviation in teacher value-added (Appendix Figure A.8). The effect of behavior value-
added on long-term outcomes are systematically smaller than in Figure 7, but they still imply
substantial effects of providing students with better behavior value-added teachers for a number of
outcomes (Appendix Table A.31). The effect of test-score value-added on long-term outcomes is
approximately as large and for some outcomes notably larger than the estimates reported in Figure
7. The outcomes where test-score value-added increases the most tend to be ones where behavior
value-added decreases substantially, which suggests this empirical approach puts more weight on
better test-score value-added teachers than behavior value-added teachers.
Since we have data on all grades 3 to 12, we can also examine whether having a high value-
added elementary school teacher or high school teacher has a larger long-term effect on student
outcomes under the strong assumption that a one standard deviation change in teacher value-added
induces the same amount of learning in each grade. Models of human capital formation in which
past human capital production is complementary to current human capital production suggests that
having a high value-added elementary school teacher is more important (Kinsler 2016). Alterna-
tively, the substantial fade out we observe in the effect of teacher value-added suggests that high
school teachers will have a larger effect on student outcomes.
The results generally show that having a high value-added English and math teacher in middle
school or high school has a bigger impact on long-term outcomes than having a high value-added
elementary school teacher. This pattern of results is especially clear for dropping out of high
school, test scores, the GPA measures, absences, suspensions, and grade retention. For example,
having a one standard deviation higher behavior value-added teacher has little effect on whether
a student drops out of high school in grades 3 to 5, but reduces the likelihood of dropping out by
about one percentage point per year in grades 6 to 12. Exceptions exist, notably for taking the SAT,
but the pattern of results is fairly clear. The strength of the middle school and high school effects
13We start in fourth grade rather than third grade because it substantially increases our sample size. We include allstudents with at least seven observations of teacher value-added in the eight years from grade four to eleven.
38
is somewhat surprising because we only calculate value-added for English and math teachers, with
whom a student spends less than half of her school day, whereas in elementary school, the value-
added measures are calculated using a classroom teacher with whom students spend much more
time. However, these results should be interpreted with an abundance of caution since common
cardinality of the VA measures across grades is difficult to fully defend given that achievement
measures are standardized within each grade-year. It is possible that one standard deviation in the
VA distribution at later grades represents a larger magnitude shift in an underlying distribution of
skill relative to a one standard deviation shift in earlier grades.
6.2 Measuring Teacher Quality in Untested Subjects
A significant shortcoming of the test-score value-added framework is that a test must be adminis-
tered in a subject to measure a teacher’s quality in that subject. Consequently, we cannot evaluate
teachers using value-added measures in subjects or grades that are untested, or compare the im-
portance of high quality teachers across untested subjects. We extend the approach for calculating
non-test-score value-added described in section 4.1 to compute value-added measures for elemen-
tary school teachers by subject using students’ grades in each subject. We measure a teacher’s
quality using the grade each student receives in a subject in year t +1, controlling for the baseline
controls from year t−1.
Appendix Table A.32 shows long-term student outcomes regressed on students’ grades and
the standard set of controls. Better grades in virtually all subjects improve students’ long-term
outcomes. Two exceptions are speaking and PE, which indicate that students who perform better
in speaking or PE perform worse in high school, even conditional on their grades in other subjects
and prior achievement. However, these estimates may not be detecting the true effect of ability in
a particular subject, but unobserved characteristics associated with both elementary school grades
and high school outcomes.
Table 8 reports the results after redoing this analysis using teacher value-added for each sub-
ject. Teachers who excel at teaching reading and health have students who perform better in high
39
school across several measures. The effects for math-GPA value-added are subject specific. The
effect is only positive and marginally significant for math high school exit exam scores. Unex-
pectedly, students with better elementary school math teachers are more likely to be held back in
high school. The effects of reading-GPA value-added are more widespread, with significant or
marginally significant effects on dropout, taking the SAT, SAT score, and both math and English
high school exit exam scores. Writing-GPA value-added significantly affects the probability of
being held back. Health-GPA value-added significantly affects for taking the SAT, cooperation
GPA, and is marginally significant for both math and English high school exit exam scores. Social
studies has positive and marginally significant effects on two of the grade measures.
Speaking-GPA value-added has a positive and significant effect on dropout, and negative and
significant or marginally significant effects on SAT scores, both high school exit exam measures,
and all three GPA measures. Perhaps talking in class is not well rewarded in high school. The
coefficients on PE, Arts, and science are also generally suggest negative effects on students but in
most cases are not statistically significant.
The confidence intervals on the statistically insignificant estimates are such that they often
cannot reject effects of the magnitude of reading or health GPA on the outcomes. Although for
example, the effect of reading GPA on taking the SAT and test scores is outside the confidence
interval for many outcomes as is the effect. Similarly, the negative effects on students of the
magnitude of speaking value-added on e.g., dropout and SAT score, are often rejected by the
confidence intervals of the other value added measures.
These results broadly support the traditional view that reading is a building-block skills that has
long-term benefits. The health results are unexpected, suggesting that health knowledge at a young
age could have long-term benefits, though this explanation should be interpreted with caution.
Besides the negative effect of speaking, we find relatively little evidence of negative effects of
other subjects. These results suggest that elementary schools could potentially create long-term
benefits for students by hiring and retaining strong reading teachers and focusing more of their
time on teaching reading.
40
7 Conclusion
The results demonstrate that teacher quality is multidimensional. We show that teachers’ test-score
value-added has significant effects on long-term outcomes, and that adding controls for behavior
and learning-skills value-added does not influence the estimated effects. This finding indicates the
long-term effects of having a high test-score value-added teacher may not be biased upward by
omitting measures of behavior or learning-skills value-added.
We also find that a teacher value-added measure that combines the teacher valued-added effects
on GPA, absences, suspensions, and grade retention affects many high school outcomes. These ef-
fects are similar in magnitude to test-score value-added. We find little evidence that learning-skills
value-added individually effects high school outcomes. The second dimension of teacher quality
is only weakly correlated with test-score value-added, and allows for a substantial enhancement
in the measurement of teacher quality. For example, a policy that uses both dimensions and three
years of data to identify the bottom 5 percent of teachers and replaces them with average teachers
increases the efficacy of the policy by over 50 percent versus a policy that uses only test-score
value-added for dropout rates, the likelihood of taking the SAT, GPA, effort GPA, absences, and
being held back. Despite substantial gains in many areas, high school test scores experience only
minimal declines.
We then demonstrate that this value-added framework can be extended to analyze effects by
grade and all elementary school subjects. We find that cumulative effect of better elementary,
middle, and high school teachers is large, even after controlling for tracking. We also show that
teachers who are relatively better at teaching reading and health improve their elementary school
students’ high school outcomes, whereas teachers who are better at teaching speaking worsen
them. Teaching reading may have long-term benefits for students, which suggests schools should
focus on increasing teaching quality in that area.
Overall, this paper shows the multifaceted role that teachers play in influencing the outcomes of
their students. In addition to benefiting students through increasing performance on tests, teachers
meaningfully impact students long-term outcomes through additional channels.
41
References
AdCouncil. 2015. “California Attendance Parent Survey Results,” Unpublished Presentation.Araujo, M. Caridad, Pedro Carneiro, Yyannú Cruz-Aguayo, and Norbert Schady. 2016. “Teacher
Quality and Learning Outcomes in Kindergarten.” The Quarterly Journal of Economics,131(3): 1415–1453.
Bacher-Hicks, Andrew, Thomas J. Kane, and Douglas O. Staiger. 2014. “Validating TeacherEffect Estimates Using Changes in Teacher Assignments in Los Angeles.” National Bureauof Economic Research Working Paper No. 20657.
Blazar, David, and Matthew A. Kraft. 2017. “Teacher and Teaching Effects on Students’ Atti-tudes and Behaviors.” Educational Evaluation and Policy Analysis, 39(1), 146-170.
Carrell, Scott E., and James E. West. 2010. "Does Professor Quality Matter? Evidence fromRandom Assignment of Students to Professors." Journal of Political Economy, 118(3):409-432.
Chetty, Raj, John N. Friedman, and Jonah E. Rockoff. 2011. "The Long-Term Impacts ofTeachers: Teacher Value-Added and Student Outcomes in Adulthood." Working Paper.
Chetty, Raj, John N. Friedman, and Jonah E. Rockoff. 2014a. "Measuring the Impacts of Teach-ers I: Evaluating Bias in Teacher Value-Added Estimates." American Economic Review,104(9): 2593-2632.
Chetty, Raj, John N. Friedman, and Jonah E. Rockoff. 2014b. "Measuring the Impacts of Teach-ers II: Teacher Value-Added and Student Outcomes in Adulthood." American EconomicReview, 104(9): 2633-2679.
Chetty, Raj, John N. Friedman, and Jonah E. Rockoff. 2017. "Measuring the Impacts of Teach-ers: Reply." American Economic Review, 107(6): 1685-1717.
Clarke, Damian, Joseph P. Romano, and Michael Wolf. 2019. "The Romano-Wolf MultipleHypothesis Correction in Stata." Working Paper.
Cook, Philip J., Kenneth A. Dodge, Elizabeth J. Gifford, and Amy B. Schulting. 2017. "A newprogram to prevent primary school absenteeism: Results of a pilot study in five schools."Children and Youth Services Review, 82: 262-270.
Cunha, Flavio, James J. Heckman, and Susanne M. Schennach. 2010. “Estimating the Technol-ogy of Cognitive and Noncognitive Skill Formation.” Econometrica, 78(3): 883–931.
Figlio, David N., Morton O. Schapiro, and Kevin B. Soter. 2015. "Are Tenure Track ProfessorsBetter Teachers?" The Review of Economics and Statistics, 97(4): 715-724.
Fleche, Sarah. 2017. “Teacher Quality, Test Scores, and Non-Cognitive Skills: Evidence fromPrimary School Teachers in the UK.” Working Paper.
Fryer, Roland G. 2013. “Teacher Incentives and Student Achievement: Evidence from NewYork City Public Schools.” Journal of Labor Economics, 31(2): 373-407.
Gershenson, Seth. 2016. “Linking Teacher Quality, Student Attendance, and Student Achieve-ment.” Education Finance and Policy, 11(2): 123-149.
Goldhaber, Dan, and Michael Hansen. 2010. “Assessing the Potential of Using Value-AddedEstimates of Teacher Job Performance for Making Tenure Decisions.” National Center forAnalysis of Longitudinal Data in Education Research Working Paper No. 31.
Goodman, Sarena, and Lesley Turner. 2013. "The Design of Teacher Incentive Pay and Edu-cational Outcomes: Evidence from the New York City Bonus Program." Journal of Labor
42
Economics, 31(2): 409-420.Gordon, Robert, Thomas J. Kane, and Douglas O. Staiger. 2006. “Identifying Effective Teachers
using Performance on the Job.” Washington, DC: The Brookings Institution.Hanushek, Eric A. 1971. “Teacher Characteristics and Gains in Student Achievement: Esti-
mation Using Micro Data.” American Economic Review Papers and Proceedings, 61(2):280-288.
Hanushek, Eric A. 2011. “Valuing Teachers: How Much Is a Good Teacher Worth?” EducationNext, 11(3): 41-45.
Hanushek, Eric A., and Ludger Woessmann. 2008. “The Role of Cognitive Skills in EconomicDevelopment.” Journal of Economic Literature, 46(3): 607- 668.
Hanushek, Eric A., and Steven G. Rivkin. 2010. “Generalizations about Using Value-AddedMeasures of Teacher Quality.” American Economic Review Papers and Proceedings, 100(2):267-271.
Heckman, James J., Jora Stixrud, and Sergio Urzua. 2006. “The Effects of Cognitive andNoncognitive Abilities on Labor Market Outcomes and Social Behavior.” Journal of LaborEconomics, 24(3): 411-482.
Heckman James, Rodrigo Pinto, and Peter Savelyev. 2013. “Understanding the MechanismsThrough Which an Influential Early Childhood Program Boosted Adult Outcomes.” Amer-ican Economic Review, 103(6): 2052–2086.
Horn, Sandra P., and William L. Sanders. 1994. “The Tennessee Value-Added AssessmentSystem (TVAAS): Mixed-Model Methodology in Education Assessment.” Journal of Per-sonnel Evaluation in Education, 8(3): 299-311.
Jackson, C. Kirabo. 2018. “What Do Test Scores Miss? The Importance of Teacher Effects onNon-Test Score Outcomes.” Journal of Political Economy, 126(5): 2072-2107.
Jackson, C. Kirabo, Jonah E. Rockoff, and Douglas O. Staiger. 2014. “Teacher Effects andTeacher- Related Policies.” Annual Review of Economics, 6:801–825.
Jacob, Brian A., Lars Lefgren, and David P. Sims. 2010. "The Persistence of Teacher-InducedLearning." Journal of Human Resources, 45(4): 915-943.
Jennings, Jennifer L., and Thomas A. DiPrete. 2010. “Teacher Effects on Social and BehavioralSkills in Early Elementary School.” Sociology of Education, 83(2): 135-159.
Jones, Damon, David Molitor, and Julian Reif. 2019. “What Do Workplace Wellness ProgramsDo? Evidence from the Illinois Workplace Wellness Study.” The Quarterly Journal ofEconomics, 134(4): 1747-1791.
Kane, Thomas J., and Douglas O. Staiger. 2008. “Estimating Teacher Impacts on StudentAchievement: An Experimental Evaluation.” National Bureau of Economic ResearchWorking Paper No. 14607.
Kane, Thomas J., Daniel F. McCaffrey, Trey Miller, and Douglas O. Staiger. 2013. "Have WeIdentified Effective Teachers? Validating Measures of Effective Teaching Using RandomAssignment." In Research Paper. MET Project. Bill & Melinda Gates Foundation.
Kinsler, Josh. 2016. "Teacher Complementarities in Test Score Production: Evidence fromPrimary School." Journal of Labor Economics, 34(1): 29-61.
Kraft, Matthew A. 2019. "Teacher Effects on Complex Cognitive Skills and Social-EmotionalCompetencies." Journal of Human Resources, 54(1): 1-36.
Kraft, Matthew A., and Todd Rogers. 2015. "The underutilized potential of teacher-to-parentcommunication: Evidence from a field experiment." Economics of Education Review, 47:
43
49-63.Lazear, Edward P. 2003. “Teacher Incentives.” Swedish Economic Policy Review, 10(3): 179-
214.McQuiggan, Meghan and Mahi Megra. 2017. “Parent and Family Involvement in Education:
Results from the National Household Education Surveys Program of 2016.” U.S. Depart-ment of Education. Washington, DC: National Center for Education Statistics.
Mihaly, Kata, Daniel F. McCaffrey, Douglas O. Staiger, and J. R. Lockwood. 2013. "A Com-posite Estimator of Effective Teaching." Seattle, WA: Bill & Melinda Gates Foundation.
Mulligan, Casey B. 1999. “Galton versus the Human Capital Approach to Inheritance.” Journalof Political Economy, 107(6): S184-224.
Murnane, Richard J., John B. Willett, Yves Duhaldeborde, and John H. Tyler. 2000. “How Im-portant Are the Cognitive Skills of Teenagers in Predicting Subsequent Earnings.” Journalof Policy Analysis and Management, 19(4): 547-568.
National Council on Teacher Quality. 2015. “2015 State Teacher Policy Yearbook: NationalSummary.” Technical Report.
Neal, Derek. 2011. “The Design of Performance Pay in Education.” In Handbook of the Eco-nomics of Education Volume 4, edited by Eric A. Hanushek, Stephen Machin, and LudgerWoessmann, 495-550. Amsterdam: North-Holland, Elsevier.
Robinson, Carly. D., Monica G. Lee, Eric Dearing, Todd Rogers. 2018. “Reducing StudentAbsenteeism in the Early Grades by Targeting Parental Beliefs.” American EducationalResearch Journal, 55(6), 1163–1192.
Rockoff, Jonah E. 2004. “The Impact of Individual Teachers on Student Achievement: Evidencefrom Panel Data.” American Economic Review Papers and Proceedings, 94(2): 247-252.
Rockoff, Jonah, E., and Cecilia Speroni. 2010. “Subjective and Objective Evaluations ofTeacher Effectiveness.” American Economic Review Papers & Proceedings, 100(2): 261-266.
Rogers, Todd and Avi Feller. 2018. “Reducing student absences at scale by targeting parents’misbeliefs.” Nature Human Behaviour, 2(5): 335-342.
Romano, Joseph P., and Michael Wolf. 2016. "Efficient computation of adjusted p-values forresampling-based stepdown multiple testing." Statistics & Probability Letters, 113: 38-40.
Rothstein, Jesse. 2007. "Does Competition Among Public Schools Benefit Students and Tax-payers? Comment." American Economic Review, 97(5): 2026-2037.
Rothstein, Jesse. 2010. "Teacher Quality in Educational Production: Tracking, Decay, andStudent Achievement." Quarterly Journal of Economics, 125(1): 175-214.
Rothstein, Jesse. 2017. “Measuring the Impacts of Teachers: Comment." American EconomicReview, 107(6): 1656-84.
Ruzek, Erik A., Thurston Domina, AnneMarie M. Conley, Greg J. Duncan, and Stuart A.Karabenick. 2015. “Using Value-Added Models to Measure Teacher Effects on Students’Motivation and Achievement.” Journal of Early Adolescence, 35(5-6): 852-882.
Westfall, Peter H. and S. Stanley Young. 1993. Resampling-based multiple testing: Examplesand methods for p-value adjustment. Vol. 279, John Wiley & Sons.
44
Figure 1: LAUSD Elementary School Progress Report
Note: The figure displays a blank copy of an LAUSD elementary school progress report. Each row labels the academic subject, work and studyhabits, or learning and social skill each student is graded on by their teacher. Columns 1, 2, and 3 correspond to each of the three trimesters studentsreceive a grade. For the academic subjects, the AC column stands for achievement scores and the EF column stands for effort scores. For allacademic subjects, work and study habits, and learning and social skills, students receive a grade ranging from 1 (the poorest performing) to 4 (thebest performing).
45
Figure 2: Effect of Teacher Test-Score Value-Added on High School Outcomes
Note: The sample includes students in grades 3-5 who attended high school in the LAUSD. The unit of observation is a student-academic year.Figure 2 shows binned scatter plots of residualized high school outcome variables and normalized teacher test-score value-added for grades 3, 4, or5. We construct these plots by first residualizing the outcome and teacher value-added variables using the controls shown in equation (6). We thenplot the mean values of both variables in 20 equally sized bins. Lastly, we add back the unconditional mean of both variables. We also plot the bestlinear fit estimated prior to binning the data, and report its slope coefficient and standard error, clustered at the modal high school level. *** p <0.01, ** p < 0.05, and * p < 0.10.
46
Figure 3: Effect of Teacher Behavior Value-Added on High School Outcomes
Note: The sample includes students in grades 3-5 who attended high school in the LAUSD. The unit of observation is a student-academic year.Figure 3 shows binned scatter plots of residualized high school outcome variables and normalized teacher-behavior value-added for grades 3, 4, or5. We construct these plots by first residualizing the outcome and teacher value-added variables using the controls shown in equation (6). We thenplot the mean values of both variables in 20 equally sized bins. Lastly, we add back the unconditional mean of both variables. We also plot the bestlinear fit estimated prior to binning the data, and report its slope coefficient and standard error, clustered at the modal high school level. *** p <0.01, ** p < 0.05, and * p < 0.10.
47
Figure 4: Effect of Teacher Learning-Skills Value-Added on High School Outcomes
Note: The sample includes students in grades 3-5 who attended high school in the LAUSD. The unit of observation is a student-academic year.Figure 4 shows binned scatter plots of residualized high school outcome variables and normalized teacher learning-skills value-added for grades 3,4, or 5. We construct these plots by first residualizing the outcome and teacher value-added variables using the controls shown in equation (6). Wethen plot the mean values of both variables in 20 equally sized bins. Lastly, we add back the unconditional mean of both variables. We also plot thebest linear fit estimated prior to binning the data, and report its slope coefficient and standard error, clustered at the modal high school level. *** p< 0.01, ** p < 0.05, and * p < 0.10.
48
Figure 5: Dynamic Effects of Test-Score, Behavior, and Learning-Skills Teacher Value-Added
Note: Unit of observation is a student-academic year. Each plot shows the effect of test-score, behavior, and learning-skills value-added of teachersin grades 3, 4 or 5 on student outcomes in the concurrent year (the year a student was in a teacher’s classroom) and future years (the years after astudent was in a teacher’s classroom). The estimated effects are obtained by regressing leads of outcome variables on teacher test-score, behavior,and learning-skills value-added and the baseline controls as specified in equation (7). The coefficients on test-score, behavior, and learning-skillsvalue-added are plotted along with 95 percent confidence intervals, with standard errors clustered at the modal high school level.
49
Figure 6: Two-Dimensional Cross Teacher Value-Added Plots
Note: The first scatter plot shows a plot of elementary school teachers’ annual, normalized test-score and behavior value-added within three standarddeviations of the mean. The two dashed lines show the cutoffs for the 5th percentile of the test-score and behavior teacher value-added, respectively.The second and third scatter plots are constructed analogously for test-score value-added versus learning-skills value-added, and behavior value-added versus learning-skills value-added, respectively.
50
Figure 7: Effects of Test-Score and Behavior Teacher Value-Added by Grade
Note: The sample includes students in grades 3-5 who attended high school in the LAUSD. The unit of observation is a student-academic year. Thefigure shows plots of the effect of test-score and behavior teacher value-added on high school outcomes by the grade level of the student. The plottedcoefficients and standard errors (clustered at the modal high school level) are from a regression of the high school outcome variable on test-scoreand behavior teacher value-added, and the vector of controls for high school students specified in section 4.1, estimated separately for each grade.
51
Table 1: Summary Statistics
Variable MeanStandardDeviation
Observations
Panel A: Grades 3 - 5Math CST Score 0.01 1.00 891,643English CST Score 0.00 1.00 891,751GPA 2.88 0.42 861,977Effort GPA 3.14 0.46 861,977Learning Skills GPA 3.10 0.58 614,532Fraction of Days Absent 3.9% 858,308Days Suspended 0.05 0.38 906,193Held Back 0.7% 837,401English Learner 42.0% 906,193Panel B: High School OutcomesLAUSD Dropout 54.6% 333,513Took SAT 50.5% 249,436SAT Score 1,330 298 145,265GPA 2.25 0.96 536,868Effort GPA 2.12 0.52 476,548Cooperation GPA 2.33 0.45 476,548Math CAHSEE Score 0.07 1.01 331,266English CAHSEE Score 0.08 0.98 329,980Days Suspended 0.18 0.83 588,273Fraction of Days Absent 7.8% 542,959Held Back a Grade 29.0% 449,533Graduated if Entered 12th Grade 88.6% 190,278Took PSAT 68.9% 470,703PSAT Score 1,110 248 348,992Math CST Score -0.05 0.99 124,044English CST Score 0.00 0.99 135,769Grade 8 Science CST Score 0.02 1.00 599,880Grade 10 Science CST Score 0.09 1.00 296,069Grade 8 Social Science CST Score 0.03 1.00 548,439Grade 11 Social Science CST Score 0.07 0.99 160,483World History CST Score 0.06 0.98 270,403Number of AP Courses 0.73 1.70 588,273
Note: The sample includes students in grades 3-5 who attended the LAUSD. The unit of observation is a student-academic year. Panel A reportssummary statistics for all LAUSD students in grades 3-5 from 2004 to 2010. Panel B reports high school summary statistics for all LAUSD studentswho were in grades 3-5 from 2004 to 2010 and attend high school in the LAUSD. Elementary school GPA, effort GPA, and learning-skills GPA areon a 4-point scale. GPA in high school is on a 4-point scale, and effort GPA and cooperation GPA in high school are on a 3-point scale. All testscores except the SAT and PSAT are normalized at the grade-year level. Both the SAT score and PSAT score are on a 600-2400 scale. The LAUSDDropout variable is the fraction of students who enrolled in an LAUSD school in 9th grade and did not graduate from the LAUSD within five years.The Graduated if Entered 12th Grade variable shows the fraction of students who enrolled in an LAUSD school in 12th grade and graduated fromthe LAUSD.
52
Table 2: Correlation of Elementary School Student Achievement Measures
Measure Test Scores BehaviorLearning
SkillsTest Scores 1Behavior 0.463 1Learning Skills 0.532 0.550 1
Note: The sample includes students in grades 3-5 who attended the LAUSD. The unit of observation is a student-academic year. Table 2 reportsthe correlations between the three measures of student achievement for grades 3-5. Each of the three measures of student achievement are equallyweighted indices. The test-score index is computed using the students’ normalized math and English test scores. The behavior index is computedusing students’ GPA, suspensions, log days absent, and not progressing to the next grade on time (held back). The learning-skills index is computedusing students’ effort GPA, work and study habit GPA, and learning and social skills GPA.
53
Tabl
e3:
Rel
atio
nshi
pbe
twee
nE
lem
enta
rySc
hool
Ach
ieve
men
tand
Hig
hSc
hool
Out
com
es
Mea
sure
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
es-0
.032
***
0.09
1***
143.
271*
**0.
208*
**0.
103*
**0.
089*
**0.
472*
**0.
398*
**-0
.009
***
-0.1
12**
*-0
.052
***
(0.0
03)
(0.0
04)
(2.6
44)
(0.0
04)
(0.0
02)
(0.0
02)
(0.0
06)
(0.0
06)
(0.0
02)
(0.0
05)
(0.0
03)
Beh
avio
r-0
.020
***
0.02
8***
10.8
26**
*0.
062*
**0.
032*
**0.
030*
**0.
028*
**0.
020*
**-0
.028
***
-0.1
30**
*-0
.017
***
(0.0
02)
(0.0
03)
(1.5
54)
(0.0
04)
(0.0
02)
(0.0
02)
(0.0
03)
(0.0
03)
(0.0
03)
(0.0
06)
(0.0
02)
Lea
rnin
gSk
ills
-0.0
40**
*0.
068*
**-2
.393
*0.
228*
**0.
133*
**0.
121*
**0.
045*
**0.
064*
**-0
.045
***
-0.0
62**
*-0
.061
***
(0.0
02)
(0.0
03)
(1.2
90)
(0.0
03)
(0.0
02)
(0.0
02)
(0.0
03)
(0.0
03)
(0.0
03)
(0.0
04)
(0.0
02)
Obs
erva
tions
134,
356
100,
691
59,5
8232
1,23
625
1,46
025
1,46
016
0,38
515
9,91
134
3,72
030
2,90
224
0,20
4
R-s
quar
ed0.
421
0.17
90.
686
0.32
10.
315
0.32
60.
572
0.57
70.
045
0.28
10.
134
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
pred
ictiv
eef
fect
ofa
stan
dard
devi
atio
nin
crea
sein
each
ofth
eth
ree
mea
sure
sof
stud
enta
chie
vem
ent(
see
Tabl
e2)
ingr
ades
3-5
onhi
ghsc
hool
stud
ento
utco
mes
.Spe
cific
ally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
achi
evem
entm
easu
res
ofst
uden
tsfr
oman
OL
Sre
gres
sion
ofth
ehi
ghsc
hool
stud
ento
utco
me
onth
est
uden
ts’t
hree
achi
evem
entm
easu
res
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.T
heba
selin
eco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fthe
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ner
stat
usan
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,a
ndhe
ldba
ckfo
rth
ein
divi
dual
,the
clas
sav
erag
e,an
dth
egr
ade
aver
age.
Eac
hof
thes
eva
riab
les
exce
ptE
nglis
hle
arne
rsta
tus
isfu
llyin
tera
cted
with
grad
efix
edef
fect
san
da
cont
rolf
orcl
ass
size
isin
clud
ed.
Stan
dard
erro
rscl
uste
red
atth
em
odal
high
scho
olle
vela
rere
port
edin
pare
nthe
ses.
***
p<
0.01
, **
p<
0.05
,and
*p
<0.
10.
54
Table 4: Correlation of Elementary School Teacher Value-Added Measures
Grades 3-5 VATest Score
VABehavior
VALearningSkills VA
Test Score VA 1Behavior VA 0.145 1Learning Skills VA 0.174 0.459 1
Note: The sample includes students in grades 3-5 who attended the LAUSD. The unit of observation is a student-academic year. Table 4 reports thecorrelations between the three measures of teacher value-added for grades 3-5. Each of the three measures are equally weighted indices. The test-score value-added is computed using teachers’ value-added for math and English test scores. The behavior value-added is computed using teachers’value-added for GPA, suspensions, log days absent, and not progressing to the next grade on time (held back). The learning-skills value-added iscomputed using teachers’ value-added for effort GPA, work and study habit GPA, and learning and social skills GPA.
55
Tabl
e5:
Eff
ecto
fEle
men
tary
Scho
olTe
ache
rVal
ue-A
dded
onH
igh
Scho
olO
utco
mes
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
-0.0
02-0
.002
6.23
7***
0.00
20.
003*
0.00
5***
0.02
2***
0.01
6***
0.00
10.
001
0.00
1
(0.0
02)
(0.0
02)
(1.2
62)
(0.0
02)
(0.0
01)
(0.0
01)
(0.0
03)
(0.0
02)
(0.0
01)
(0.0
03)
(0.0
01)
Beh
avio
rVA
-0.0
030.
010*
**1.
955
0.01
3***
0.00
7***
0.00
5**
0.01
3**
0.00
4-0
.003
-0.0
16**
*-0
.006
**
(0.0
02)
(0.0
03)
(2.1
39)
(0.0
04)
(0.0
03)
(0.0
02)
(0.0
06)
(0.0
05)
(0.0
02)
(0.0
05)
(0.0
03)
Lea
rnin
gSk
ills
VA-0
.000
-0.0
01-0
.547
-0.0
02-0
.003
-0.0
03-0
.005
0.00
1-0
.002
0.00
20.
001
(0.0
02)
(0.0
03)
(1.7
56)
(0.0
05)
(0.0
03)
(0.0
03)
(0.0
05)
(0.0
04)
(0.0
02)
(0.0
06)
(0.0
02)
Obs
erva
tions
135,
786
102,
517
60,6
9429
3,02
123
3,07
823
3,07
815
2,34
515
1,82
031
6,11
627
7,33
122
1,75
7
R-s
quar
ed0.
293
0.14
50.
617
0.24
40.
234
0.23
90.
500
0.51
20.
040
0.26
70.
108
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
eth
ree
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
ed(s
eeTa
ble
4)on
stud
ents
’hig
hsc
hool
outc
omes
.Spe
cific
ally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
norm
aliz
edm
easu
res
ofte
ache
rva
lue-
adde
dfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’
high
scho
olou
tcom
eon
the
thre
em
easu
res
ofte
ache
rva
lue-
adde
dfo
rth
est
uden
ts’
teac
hers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.T
heba
selin
eco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fthe
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ners
tatu
san
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,a
ndhe
ldba
ckfo
rthe
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ners
tatu
sis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.Sta
ndar
der
rors
clus
tere
dat
the
mod
alhi
ghsc
hool
leve
lare
repo
rted
inpa
rent
hese
s.**
*p
<0.
01, *
*p
<0.
05,a
nd*
p<
0.10
.
56
Tabl
e6:
Eff
ecto
fRep
laci
ngth
eB
otto
m5
Perc
ento
fEle
men
tary
Scho
olTe
ache
rson
Hig
hSc
hool
Out
com
es
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Pane
lA:U
sing
True
Val
ue-A
dded
Test
Scor
eVA
-0.0
04-0
.002
13.1
76**
*0.
008*
0.00
5*0.
009*
**0.
048*
**0.
035*
**-0
.000
-0.0
030.
001
(0.0
03)
(0.0
04)
(2.2
57)
(0.0
05)
(0.0
03)
(0.0
03)
(0.0
05)
(0.0
05)
(0.0
03)
(0.0
05)
(0.0
02)
Beh
avio
rVA
-0.0
07*
0.01
9***
5.44
60.
026*
**0.
013*
**0.
008*
0.02
8**
0.01
4-0
.007
**-0
.031
***
-0.0
11**
(0.0
04)
(0.0
06)
(4.5
31)
(0.0
08)
(0.0
05)
(0.0
04)
(0.0
11)
(0.0
09)
(0.0
03)
(0.0
10)
(0.0
05)
1 2(T
estS
core
+B
ehav
ior)
-0.0
08*
0.01
2**
12.2
80**
*0.
023*
**0.
012*
**0.
012*
**0.
050*
**0.
033*
**-0
.004
-0.0
22**
-0.0
07
(0.0
04)
(0.0
05)
(3.9
65)
(0.0
07)
(0.0
04)
(0.0
04)
(0.0
09)
(0.0
08)
(0.0
03)
(0.0
09)
(0.0
04)
Opt
imal
Thr
eeVA
-0.0
080.
019*
**13
.733
***
0.02
7***
0.01
4***
0.01
3***
0.05
3***
0.03
6***
-0.0
08-0
.031
***
-0.0
11*
Wei
ghte
dA
vg.
(0.0
06)
(0.0
07)
(3.1
44)
(0.0
09)
(0.0
05)
(0.0
04)
(0.0
08)
(0.0
07)
(0.0
05)
(0.0
11)
(0.0
05)
%G
ain
(Row
1to
Row
3)12
1%20
0%+
-7%
200%
129%
34%
4%-6
%20
0%+
200%
+20
0%+
Pane
lB:U
sing
Prev
ious
Thr
eeY
ears
ofSt
uden
tDat
a
Test
Scor
eVA
-0.0
02-0
.001
7.68
8***
0.00
4*0.
003*
0.00
5***
0.02
8***
0.02
0***
-0.0
00-0
.002
0.00
0
(0.0
02)
(0.0
02)
(1.3
17)
(0.0
03)
(0.0
02)
(0.0
01)
(0.0
03)
(0.0
03)
(0.0
02)
(0.0
03)
(0.0
01)
Beh
avio
rVA
-0.0
01*
0.00
3***
0.96
80.
005*
**0.
002*
**0.
001*
0.00
5**
0.00
3-0
.001
**-0
.006
***
-0.0
02**
(0.0
01)
(0.0
01)
(0.8
05)
(0.0
01)
(0.0
01)
(0.0
01)
(0.0
02)
(0.0
02)
(0.0
01)
(0.0
02)
(0.0
01)
1 2(T
estS
core
+B
ehav
ior)
-0.0
030.
001
6.46
3***
0.00
6**
0.00
4***
0.00
5***
0.02
4***
0.01
7***
-0.0
00-0
.004
-0.0
01
(0.0
02)
(0.0
02)
(1.3
71)
(0.0
03)
(0.0
02)
(0.0
01)
(0.0
03)
(0.0
03)
(0.0
01)
(0.0
03)
(0.0
02)
Opt
imal
Thr
eeVA
-0.0
030.
003*
**7.
688*
**0.
006*
*0.
004*
**0.
005*
**0.
028*
**0.
020*
**-0
.002
-0.0
06**
*-0
.002
**
Wei
ghte
dA
vg.
(0.0
02)
(0.0
01)
(1.3
17)
(0.0
03)
(0.0
02)
(0.0
01)
(0.0
03)
(0.0
03)
(0.0
01)
(0.0
02)
(0.0
01)
%G
ain
(Row
1to
Row
3)29
%20
0%+
-16%
36%
24%
-0%
-15%
-15%
142%
95%
200%
+
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Pan
elA
show
sth
eef
fect
ona
stud
ent’s
high
scho
olou
tcom
esof
bein
gas
sign
edan
aver
age
teac
heri
nste
adof
ate
ache
rin
the
botto
m5
perc
enta
sm
easu
red
byth
ete
ache
rval
ue-a
dded
vari
able
indi
cate
din
each
row
.The
sim
ulat
ion
uses
the
estim
ated
effe
cts
ofte
ache
rva
lue-
adde
dm
easu
res
onhi
ghsc
hool
outc
omes
(fro
mFi
gure
s2-
4an
dTa
ble
5),t
hew
ithin
-tea
cher
corr
elat
ions
betw
een
the
thre
ete
ache
rval
ued-
adde
dm
easu
res
(sho
wn
inTa
ble
4),a
ndas
sum
este
ache
rva
lue-
adde
dm
easu
res
are
norm
ally
dist
ribu
ted.
Row
1in
Pane
lAus
esa
mea
sure
ofte
ache
rs’
test
-sco
reva
lue-
adde
dto
repl
ace
the
botto
m5
perc
ento
fte
ache
rs.
The
refo
re,e
ach
cell
inro
w1
show
sth
eef
fect
ona
stud
ent’s
high
scho
olou
tcom
e(s
how
nin
the
colu
mn)
ifsh
ew
ere
tom
ove
from
ate
ache
rin
the
botto
m5
perc
ento
ftes
t-sc
ore
teac
herv
alue
-add
edto
anav
erag
ete
ache
r.R
ow2
ofPa
nelA
show
sth
eim
prov
emen
tin
outc
omes
for
am
ove
from
ate
ache
rin
the
botto
m5
perc
entt
oan
aver
age
teac
her
asm
easu
red
bybe
havi
orva
lue-
adde
d.R
ow3
ofPa
nelA
uses
the
aver
age
ofte
ache
rs’
test
-sco
reva
lue-
adde
dan
dth
eirb
ehav
iorv
alue
-add
ed.
Row
4of
Pane
lAus
esa
max
imiz
atio
npr
oced
ure
toch
oose
the
optim
alw
eigh
tsto
bepl
ace
onte
ache
rs’t
est-
scor
e,be
havi
or,a
ndle
arni
ng-s
kills
valu
e-ad
ded
tode
term
ine
the
botto
m5
perc
ento
fte
ache
rsfo
rth
ein
dica
ted
outc
ome
vari
able
.R
ow5
ofPa
nelA
show
sth
epe
rcen
tim
prov
emen
tin
stud
ents
’ou
tcom
esif
the
repl
acem
ento
fth
ebo
ttom
5pe
rcen
tof
teac
hers
ispe
rfor
med
byus
ing
the
aver
age
ofte
ache
rs’
test
-sco
rean
dbe
havi
orva
lue-
adde
din
stea
dof
just
teac
hers
’te
st-s
core
valu
e-ad
ded.
Pane
lBsh
ows
anal
ogou
sre
sults
usin
gte
ache
rs’
valu
e-ad
ded
base
don
only
the
thre
epr
evio
usye
ars
ofst
uden
tdat
a.T
hese
valu
ed-a
dded
mea
sure
sar
ees
timat
edus
ing
thre
epr
iory
ears
ofda
taon
each
teac
her,
alon
gw
ithth
eau
toco
rrel
atio
nsin
teac
hers
’val
ue-a
dded
acro
ssye
ars
show
nin
Figu
reA
.5.T
hest
anda
rder
rors
onth
ees
timat
edfo
reca
star
esh
own
inpa
rent
hese
s.**
*p
<0.
01, *
*p
<0.
05,a
nd*
p<
0.10
.
57
Tabl
e7:
Eff
ecto
fEle
men
tary
Scho
olTe
ache
rVal
ue-A
dded
onPr
edic
ted
Hig
hSc
hool
Out
com
es
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
0.00
00.
000
0.42
30.
000
0.00
00.
000
0.00
00.
000
0.00
00.
000
0.00
0
(0.0
00)
(0.0
01)
(0.9
82)
(0.0
01)
(0.0
01)
(0.0
01)
(0.0
02)
(0.0
02)
(0.0
00)
(0.0
01)
(0.0
00)
Beh
avio
rVA
0.00
1*-0
.001
*-2
.024
*-0
.003
**-0
.002
**-0
.001
**-0
.007
**-0
.006
***
0.00
0**
0.00
2*0.
001*
(0.0
00)
(0.0
01)
(1.0
30)
(0.0
01)
(0.0
01)
(0.0
01)
(0.0
03)
(0.0
02)
(0.0
00)
(0.0
01)
(0.0
00)
Lea
rnin
gSk
ills
VA0.
000
0.00
00.
229
0.00
00.
000
0.00
0-0
.001
0.00
10.
000
0.00
10.
000
(0.0
01)
(0.0
01)
(1.1
28)
(0.0
01)
(0.0
01)
(0.0
01)
(0.0
03)
(0.0
03)
(0.0
00)
(0.0
01)
(0.0
00)
Obs
erva
tions
66,3
5450
,137
29,2
6915
8,32
712
3,34
712
3,34
778
,651
78,4
3416
9,92
914
9,19
411
7,76
3
R-s
quar
ed0.
962
0.70
90.
750
0.68
10.
665
0.66
60.
702
0.75
00.
764
0.83
30.
737
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rd-d
evia
tion
incr
ease
inth
eth
ree
mea
sure
sof
elem
enta
ryte
ache
rva
lue-
adde
d(s
eeTa
ble
4)on
stud
ents
’pr
edic
ted
high
scho
olou
tcom
esus
ing
doub
lela
gged
stud
ent
achi
evem
ent.
The
pred
icte
dou
tcom
esar
ecr
eate
dby
estim
atin
gan
OL
Sre
gres
sion
ofth
ehi
ghsc
hool
outc
ome
indi
cate
din
each
colu
mn
onth
ird-
orde
rpol
ynom
ials
ofdo
uble
lagg
ed(y
eart−
2)m
ath
and
Eng
lish
test
scor
es,G
PA,e
ffor
tGPA
,lea
rnin
gsk
ills
GPA
,log
abse
nces
,an
indi
cato
rfo
rsu
spen
sion
,an
indi
cato
rfo
rbe
ing
held
back
,and
anin
dica
tor
ofE
nglis
hle
arne
rst
atus
.T
heco
effic
ient
sob
tain
edfr
omth
isO
LS
regr
essi
onar
eth
enus
edto
pred
ict
stud
ents
’hig
hsc
hool
outc
omes
.Eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
norm
aliz
edm
easu
res
ofte
ache
rval
ue-a
dded
from
anO
LS
regr
essi
onof
the
stud
ents
’pre
dict
edhi
ghsc
hool
outc
ome
onth
eth
ree
mea
sure
sof
teac
herv
alue
-add
edfo
rthe
stud
ents
’tea
cher
sin
grad
es3,
4,or
5,al
ong
with
the
base
line
cont
rols
desc
ribe
din
sect
ion
4.1.
The
base
line
cont
rols
incl
ude
lags
ofa
cubi
cpo
lyno
mia
loft
hest
uden
t’sm
ath
test
scor
e,E
nglis
hte
stsc
ore,
GPA
,and
effo
rtG
PA;a
cubi
cpo
lyno
mia
lsof
lagg
edcl
ass
and
grad
ele
velm
eans
ofea
chof
thos
eva
riab
les;
and
curr
entE
nglis
hle
arne
rst
atus
and
lagg
edlo
gda
ysab
sent
,sus
pens
ions
,and
held
back
fort
hein
divi
dual
,the
clas
sav
erag
e,an
dth
egr
ade
aver
age.
Eac
hof
thes
eva
riab
les
exce
ptE
nglis
hle
arne
rsta
tus
isfu
llyin
tera
cted
with
grad
efix
edef
fect
san
da
cont
rolf
orcl
ass
size
isin
clud
ed.S
tand
ard
erro
rscl
uste
red
atth
em
odal
high
scho
olle
vela
rere
port
edin
pare
nthe
ses.
***
p<
0.01
, **
p<
0.05
,and
*p
<0.
10.
58
Tabl
e8:
Eff
ecto
fEle
men
tary
Scho
olTe
ache
rSub
ject
Em
phas
ison
Hig
hSc
hool
Out
com
es
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Mat
hG
PA0.
004
-0.0
065.
997
-0.0
05-0
.002
0.00
10.
021*
0.00
9-0
.000
0.00
10.
015*
**
(0.0
05)
(0.0
07)
(4.1
39)
(0.0
10)
(0.0
06)
(0.0
05)
(0.0
11)
(0.0
09)
(0.0
04)
(0.0
11)
(0.0
05)
Rea
ding
GPA
-0.0
11**
0.02
9***
9.83
0**
0.00
70.
006
0.00
50.
030*
*0.
028*
**0.
002
0.00
6-0
.005
(0.0
05)
(0.0
08)
(4.3
45)
(0.0
09)
(0.0
05)
(0.0
06)
(0.0
12)
(0.0
10)
(0.0
04)
(0.0
11)
(0.0
05)
Wri
ting
GPA
-0.0
01-0
.012
2.20
20.
016
0.00
80.
007
-0.0
030.
002
0.00
2-0
.005
-0.0
12**
(0.0
05)
(0.0
08)
(4.9
64)
(0.0
10)
(0.0
06)
(0.0
06)
(0.0
14)
(0.0
11)
(0.0
04)
(0.0
12)
(0.0
06)
Lis
teni
ngG
PA-0
.001
-0.0
104.
579
0.00
30.
002
0.00
40.
017
0.01
1-0
.003
-0.0
130.
001
(0.0
06)
(0.0
08)
(4.2
58)
(0.0
11)
(0.0
07)
(0.0
07)
(0.0
15)
(0.0
11)
(0.0
05)
(0.0
10)
(0.0
06)
Spea
king
GPA
0.01
4**
-0.0
03-1
1.32
3**
-0.0
25**
-0.0
13*
-0.0
14*
-0.0
28*
-0.0
25**
0.00
10.
009
0.00
1
(0.0
05)
(0.0
07)
(4.7
00)
(0.0
10)
(0.0
07)
(0.0
07)
(0.0
16)
(0.0
12)
(0.0
04)
(0.0
10)
(0.0
05)
His
tory
/Soc
ialS
cien
ceG
PA-0
.000
-0.0
031.
483
0.00
90.
008*
0.00
8*-0
.002
0.00
1-0
.001
0.00
6-0
.006
(0.0
06)
(0.0
06)
(3.8
24)
(0.0
08)
(0.0
05)
(0.0
05)
(0.0
11)
(0.0
07)
(0.0
03)
(0.0
12)
(0.0
06)
Scie
nce
GPA
-0.0
010.
000
-6.4
70-0
.000
-0.0
05-0
.009
-0.0
24*
-0.0
13-0
.003
-0.0
10-0
.003
(0.0
07)
(0.0
08)
(4.8
24)
(0.0
10)
(0.0
06)
(0.0
06)
(0.0
13)
(0.0
10)
(0.0
04)
(0.0
09)
(0.0
06)
Hea
lthE
duca
tion
GPA
0.00
30.
015*
*5.
572
0.00
70.
008
0.01
2**
0.02
3*0.
019*
-0.0
02-0
.000
0.00
2
(0.0
06)
(0.0
07)
(4.1
98)
(0.0
10)
(0.0
06)
(0.0
06)
(0.0
13)
(0.0
10)
(0.0
03)
(0.0
10)
(0.0
05)
PEG
PA-0
.007
0.01
1-0
.223
0.00
3-0
.005
-0.0
11*
-0.0
080.
000
-0.0
03-0
.012
-0.0
02
(0.0
06)
(0.0
07)
(4.7
95)
(0.0
09)
(0.0
05)
(0.0
06)
(0.0
11)
(0.0
09)
(0.0
04)
(0.0
09)
(0.0
04)
Art
sG
PA-0
.002
0.00
0-4
.331
-0.0
04-0
.005
-0.0
07-0
.009
-0.0
15*
0.00
30.
007
0.00
3
(0.0
06)
(0.0
07)
(5.1
69)
(0.0
10)
(0.0
06)
(0.0
06)
(0.0
12)
(0.0
08)
(0.0
03)
(0.0
11)
(0.0
05)
Obs
erva
tions
136,
125
102,
822
60,8
7529
3,56
923
3,52
923
3,52
915
2,69
315
2,16
231
6,74
027
7,85
822
2,17
4
R-s
quar
ed0.
293
0.14
60.
617
0.24
40.
234
0.24
00.
500
0.51
20.
039
0.26
60.
108
Not
e:T
his
tabl
ere
port
sth
eef
fect
ofa
stan
dard
devi
atio
nin
crea
sein
the
10su
bjec
tele
men
tary
scho
olte
ache
rva
lue-
adde
dm
easu
res
onst
uden
ts’
high
scho
olou
tcom
es.
Spec
ifica
lly,e
ach
colu
mn
ofth
eta
ble
repo
rts
the
coef
ficie
nts
onea
chof
the
10su
bjec
ttea
cher
valu
e-ad
ded
mea
sure
sfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’h
igh
scho
olou
tcom
eon
the
10su
bjec
ttea
cher
valu
e-ad
ded
mea
sure
sfo
rthe
stud
ents
’tea
cher
sin
grad
es3,
4,or
5,al
ong
with
the
base
line
cont
rols
desc
ribe
din
sect
ion
4.1.
The
base
line
cont
rols
incl
ude
lags
ofa
cubi
cpo
lyno
mia
loft
hest
uden
t’sm
ath
test
scor
e,E
nglis
hte
stsc
ore,
GPA
,and
effo
rtG
PA;a
cubi
cpo
lyno
mia
lsof
lagg
edcl
ass
and
grad
ele
velm
eans
ofea
chof
thos
eva
riab
les;
and
curr
entE
nglis
hle
arne
rst
atus
and
lagg
edlo
gda
ysab
sent
,sus
pens
ions
,and
held
back
for
the
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ner
stat
usis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.St
anda
rder
rors
clus
tere
dat
the
mod
alhi
ghsc
hool
leve
lare
repo
rted
inpa
rent
hese
s.**
*p
<0.
01, *
*p
<0.
05,a
nd*
p<
0.10
.
59
Tabl
e9:
Eff
ecto
fEle
men
tary
Scho
olSt
uden
tSki
llsfr
omIn
crea
ses
inTe
ache
r-V
alue
Add
edon
Hig
hSc
hool
Out
com
es
Pool
edG
rade
s3-
5Sk
ills
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eIn
dex
-0.0
310.
015
126.
913*
**0.
117*
**0.
082*
**0.
110*
**0.
402*
**0.
357*
**0.
005
-0.0
94*
-0.0
10
(0.0
22)
(0.0
36)
(17.
773)
(0.0
36)
(0.0
23)
(0.0
20)
(0.0
44)
(0.0
37)
(0.0
19)
(0.0
51)
(0.0
20)
Beh
avio
rInd
ex-0
.031
0.19
0***
49.7
840.
264*
**0.
138*
**0.
102*
**0.
266*
**0.
159*
-0.0
60**
-0.3
03**
*-0
.094
*
(0.0
41)
(0.0
54)
(35.
653)
(0.0
65)
(0.0
36)
(0.0
36)
(0.0
96)
(0.0
94)
(0.0
28)
(0.0
73)
(0.0
49)
Lea
rnin
gSk
ills
Inde
x-0
.013
0.04
3**
30.0
530.
065*
0.02
60.
013
0.06
70.
058*
-0.0
29**
-0.0
75*
-0.0
15
(0.0
16)
(0.0
20)
(20.
226)
(0.0
38)
(0.0
23)
(0.0
23)
(0.0
42)
(0.0
34)
(0.0
12)
(0.0
42)
(0.0
17)
Not
e:T
his
tabl
ere
port
sth
eef
fect
ofa
stan
dard
devi
atio
nin
crea
sein
the
thre
em
easu
res
stud
ent
skill
ses
timat
edby
inst
rum
entin
gfo
rth
est
uden
tsk
illus
ing
the
thre
eva
lue-
adde
dm
easu
res
ofth
eir
elem
enta
rysc
hool
teac
her(
see
Tabl
e4)
onst
uden
ts’h
igh
scho
olou
tcom
es.S
peci
fical
ly,e
ach
colu
mn
ofth
eta
ble
repo
rts
the
coef
ficie
nts
onea
chof
the
thre
eno
rmal
ized
mea
sure
sof
stud
enta
bilit
yfr
oman
2SL
Ses
timat
ion
that
ises
timat
edse
para
tely
for
each
skill
ofth
est
uden
ts’
high
scho
olou
tcom
eon
the
each
mea
sure
ofst
uden
tabi
lity
inye
art+
1in
stru
men
ted
with
the
corr
espo
ndin
gm
easu
res
ofte
ache
rval
ue-a
dded
fort
hest
uden
ts’t
each
ers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.T
heba
selin
eco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fthe
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ner
stat
usan
dla
gged
log
days
abse
nt,
susp
ensi
ons,
and
held
back
fort
hein
divi
dual
,the
clas
sav
erag
e,an
dth
egr
ade
aver
age.
Eac
hof
thes
eva
riab
les
exce
ptE
nglis
hle
arne
rsta
tus
isfu
llyin
tera
cted
with
grad
efix
edef
fect
san
da
cont
rolf
orcl
ass
size
isin
clud
ed.S
tand
ard
erro
rscl
uste
red
atth
em
odal
high
scho
olle
vela
rere
port
edin
pare
nthe
ses.
***
p<
0.01
, **
p<
0.05
,and
*p
<0.
10.
60
Appendix
Figure A.1: Secondary School Criteria for Marks
Note: The figure shows LAUSD’s secondary school teacher guidelines for giving achievement, effort, and cooperation grades.
61
Figure A.2: Share of Students Held Back by Grade0
.05
.1.1
5.2
.25
Sha
re H
eld
Bac
k
2 3 4 5 6 7 8 9 10 11 12Grade
Share Held Back by Grade
Note: The sample includes students in grades 3-5 who attended the LAUSD. The unit of observation is a student-academic year. The figure showsthe share of students held back by grade for all grades in the sample.
62
Figure A.3: Effect of Teacher Test-Score Value-Added Estimated using Year t +1 Test Scores onHigh School Outcomes
Note: The sample includes students in grades 3-5 who attended high school in the LAUSD. The unit of observation is a student-academic year.Figure A.3 shows binned scatter plots of residualized high school outcome variables and normalized teacher test-score value-added for grades 3, 4,or 5 estimated using year t +1 test-scores. We construct these plots by first residualizing the outcome and teacher value-added variables using thecontrols shown in equation (6). We then plot the mean values of both variables in 20 equally sized bins. Lastly, we add back the unconditional meanof both variables. We also plot the best linear fit estimated prior to binning the data, and report its slope coefficient and standard error, clustered atthe modal high school level. *** p < 0.01, ** p < 0.05, and * p < 0.10.
63
Figure A.4: Within Teacher Relationship in Value-Added Across Grades Taught-2
-1.5
-1-.
50
.51
1.5
2G
rade
4 T
est-
Sco
re V
alue
-Add
ed
-2 -1.5 -1 -.5 0 .5 1 1.5 2Grade 3 Test-Score Value-Added
Grade 3 and 4 Test-Score Value-Added
-2-1
.5-1
-.5
0.5
11.
52
Gra
de 4
Beh
avio
r V
alue
-Add
ed
-2 -1.5 -1 -.5 0 .5 1 1.5 2Grade 3 Behavior Value-Added
Grade 3 and 4 Behavior Value-Added
-2-1
.5-1
-.5
0.5
11.
52
Gra
de 4
Lea
rnin
g S
kills
Val
ue-A
dded
-2 -1.5 -1 -.5 0 .5 1 1.5 2Grade 3 Learning Skills Value-Added
Grade 3 and 4 Learning Skills Value-Added
Note: Unit of observation is a teacher-grade. The figure shows binned scatter plots of the within-teacher relationship in their average leave-one-outvalue-added in each grade taught.
64
Figure A.5: Autocorrelations of Teacher Value-Added Across Years
Note: Each graph shows the correlation between mean test-score residuals, using the baseline controls, across classes taught by the same teacher indifferent years. The first graph plots the autocorrections for elementary school and the second graph for secondary school.
65
Figure A.6: Effects of Teacher Value-Added on Student Attrition
Note: The unit of observation is a student-academic year. The figure shows plots of the effect of a teacher’s test-score, behavior, and learning-skillsvalue-added in grades 3, 4, or 5 on leads of student attrition (the first panel) and attrition in future grades (the second panel). In the first panel theplotted coefficients and standard errors (clustered at the modal high school level) are from a regression of the leads of attrition out of the sample ontest-score, behavior, and learning skills teacher value-added and the vector of controls specified in section 4.1, estimated separately for each lead.In the second panel the plotted coefficients and standard errors (clustered at the modal high school level) are from a regression of student attritionin a particular grade on test-score, behavior, and learning skills teacher value-added and the vector of controls specified in section 4.1, estimatedseparately for each grade.
66
Figure A.7: Effects of Test-Score and Behavior Teacher Value-Added by Grade Controlling forTracking
-.02
-.01
5-.
01-.
005
0.0
05C
oeffi
cien
t Est
imat
e
3 4 5 6 7 8 9 10 11 12Grade
Test Score Behavior
Test Score Sum = -0.007Behavior Sum = -0.083
HS Dropout
-.01
5-.
01-.
005
0.0
05.0
1C
oeffi
cien
t Est
imat
e
3 4 5 6 7 8 9 10 11 12Grade
Test Score Behavior
Test Score Sum = 0.016Behavior Sum = 0.025
HS Graduate
-.01
0.0
1.0
2.0
3C
oeffi
cien
t Est
imat
e
3 4 5 6 7 8 9 10 11 12Grade
Test Score Behavior
Test Score Sum = 0.078Behavior Sum = 0.077
Took SAT
-20
-10
010
20C
oeffi
cien
t Est
imat
e
3 4 5 6 7 8 9 10 11 12Grade
Test Score Behavior
Test Score Sum = 36Behavior Sum = -3
SAT Score-.
04-.
020
.02
.04
.06
Coe
ffici
ent E
stim
ate
3 4 5 6 7 8 9 10 11 12Grade
Test Score Behavior
Test Score Sum = -0.041Behavior Sum = 0.184
12th Grade GPA
-.02
0.0
2.0
4C
oeffi
cien
t Est
imat
e
3 4 5 6 7 8 9 10 11 12Grade
Test Score Behavior
Test Score Sum = -0.009Behavior Sum = 0.122
12th Grade Effort GPA
-.02
-.01
0.0
1.0
2.0
3C
oeffi
cien
t Est
imat
e
3 4 5 6 7 8 9 10 11 12Grade
Test Score Behavior
Test Score Sum = 0.006Behavior Sum = 0.079
12th Grade Cooperation GPA
-.05
0.0
5.1
Coe
ffici
ent E
stim
ate
3 4 5 6 7 8 9 10 11Grade
Test Score Behavior
Test Score Sum = 0.382Behavior Sum = 0.036
11th Grade Math Test Scores
-.02
0.0
2.0
4.0
6C
oeffi
cien
t Est
imat
e
3 4 5 6 7 8 9 10 11Grade
Test Score Behavior
Test Score Sum = 0.211Behavior Sum = -0.004
11th Grade English Test Scores
-.15
-.1
-.05
0.0
5C
oeffi
cien
t Est
imat
e
3 4 5 6 7 8 9 10 11 12Grade
Test Score Behavior
Test Score Sum = 0.000Behavior Sum = -0.411
12th Grade Log Days Absent
-.01
-.00
50
.005
Coe
ffici
ent E
stim
ate
3 4 5 6 7 8 9 10 11 12Grade
Test Score Behavior
Test Score Sum = 0.004Behavior Sum = -0.018
12th Grade Days Suspended
-.01
5-.
01-.
005
0.0
05C
oeffi
cien
t Est
imat
e
3 4 5 6 7 8 9 10 11Grade
Test Score Behavior
Test Score Sum = 0.001Behavior Sum = -0.032
11th Grade Held Back
Note: The sample includes students in grades 3-5 who attended high school in the LAUSD. The unit of observation is a student-academic year.The figure shows plots of the effect of test-score and behavior teacher value-added on high school outcomes by the grade level of the student. Theplotted coefficients and standard errors (clustered at the modal high school level) are from a regression of the high school outcome variable ontest-score and behavior teacher value-added, and the vector of controls for high school students specified in section 4.1, controls for future trackingfor elementary school students and current tracking for middle and high school students estimated separately for each grade. The middle and highschool teacher value-added scores are estimated controlling for tracking.
67
Figure A.8: Distribution of the Average Value-Added Score of Students’ Teachers from Grades4-11
0.2
.4.6
.81
Den
sity
-2 -1 0 1 2
Average Test-Score VA Average Behavior VA
Note: The sample is limited to students for whom we observe the value-added scores of their teachers for at least 7 of the 8 years they spend ingrades 4-11. The average value-added scores are computed within a student across their grade 4-11 teachers.
68
Tabl
eA
.1:E
ffec
tofE
lem
enta
rySc
hool
Teac
herV
alue
-Add
edon
Hig
hSc
hool
Out
com
esw
ithB
oots
trap
ped
Stan
dard
Err
ors
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
Poin
tEst
imat
e-0
.001
7-0
.001
66.
2375
0.00
240.
0026
0.00
460.
0222
0.01
650.
0010
0.00
070.
0009
Clu
ster
edSE
(0.0
017)
(0.0
022)
(1.2
618)
(0.0
024)
(0.0
014)
(0.0
013)
(0.0
028)
(0.0
025)
(0.0
013)
(0.0
032)
(0.0
014)
Boo
tstr
appe
dSE
(0.0
025)
(0.0
031)
(1.5
706)
(0.0
034)
(0.0
020)
(0.0
017)
(0.0
039)
(0.0
036)
(0.0
021)
(0.0
046)
(0.0
020)
Beh
avio
rVA
Poin
tEst
imat
e-0
.003
30.
0099
1.95
520.
0132
0.00
710.
0049
0.01
280.
0040
-0.0
027
-0.0
160
-0.0
057
Clu
ster
edSE
(0.0
025)
(0.0
033)
(2.1
391)
(0.0
043)
(0.0
026)
(0.0
024)
(0.0
056)
(0.0
046)
(0.0
020)
(0.0
051)
(0.0
026)
Boo
tstr
appe
dSE
(0.0
034)
(0.0
042)
(2.4
830)
(0.0
050)
(0.0
029)
(0.0
027)
(0.0
060)
(0.0
051)
(0.0
028)
(0.0
058)
(0.0
031)
Lea
rnin
gSk
ills
VA
Poin
tEst
imat
e-0
.000
2-0
.001
4-0
.546
7-0
.002
1-0
.003
1-0
.003
4-0
.005
20.
0007
-0.0
021
0.00
170.
0010
Clu
ster
edSE
(0.0
024)
(0.0
027)
(1.7
562)
(0.0
052)
(0.0
032)
(0.0
032)
(0.0
048)
(0.0
038)
(0.0
023)
(0.0
058)
(0.0
025)
Boo
tstr
appe
dSE
(0.0
031)
(0.0
036)
(2.1
547)
(0.0
057)
(0.0
034)
(0.0
033)
(0.0
057)
(0.0
045)
(0.0
029)
(0.0
068)
(0.0
029)
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
eth
ree
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
ed(s
eeTa
ble
4)on
stud
ents
’hig
hsc
hool
outc
omes
.Spe
cific
ally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
norm
aliz
edm
easu
res
ofte
ache
rva
lue-
adde
dfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’
high
scho
olou
tcom
eon
the
thre
em
easu
res
ofte
ache
rva
lue-
adde
dfo
rth
est
uden
ts’
teac
hers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.T
heba
selin
eco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fth
est
uden
t’sm
ath
test
scor
e,E
nglis
hte
stsc
ore,
GPA
,and
effo
rtG
PA;a
cubi
cpo
lyno
mia
lsof
lagg
edcl
ass
and
grad
ele
velm
eans
ofea
chof
thos
eva
riab
les;
and
curr
entE
nglis
hle
arne
rsta
tus
and
lagg
edlo
gda
ysab
sent
,sus
pens
ions
,and
held
back
fort
hein
divi
dual
,the
clas
sav
erag
e,an
dth
egr
ade
aver
age.
Eac
hof
thes
eva
riab
les
exce
ptE
nglis
hle
arne
rsta
tus
isfu
llyin
tera
cted
with
grad
efix
edef
fect
san
da
cont
rolf
orcl
ass
size
isin
clud
ed.S
tand
ard
erro
rscl
uste
red
atth
em
odal
high
scho
olle
vela
ndbo
otst
rapp
edw
ithin
clas
sroo
man
dbl
ock
boot
stra
pped
byth
em
odal
high
scho
olw
ith1,
000
draw
sar
ere
port
edin
pare
nthe
ses.
69
Table A.2: Correlation of All Student Achievement Measures
MeasuresMath Test
Scores
English
Test ScoresGPA
Learning
Skills GPAEffort GPA
Log
Absences
Days
SuspendedHeld Back
Math Test Scores 1
English Test Scores 0.762 1
GPA 0.636 0.676 1
Learning Skills GPA 0.454 0.454 0.688 1
Effort GPA 0.527 0.539 0.828 0.774 1
Log Absences -0.186 -0.110 -0.170 -0.181 -0.176 1
Days Suspended -0.081 -0.076 -0.098 -0.184 -0.116 0.076 1
Held Back -0.072 -0.077 -0.106 -0.069 -0.082 0.018 0.010 1
Note: The sample includes students in grades 3-5 who attended high school in the LAUSD. The unit of observation is a student-academic year.Table A.2 reports the correlations between each measure of grades 3-5 student achievement.
70
Tabl
eA
.3:R
elat
ions
hip
betw
een
Ele
men
tary
Scho
olA
chie
vem
enta
ndH
igh
Scho
olO
utco
mes
forA
llA
chie
vem
entM
easu
res
Mea
sure
sL
AU
SD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Mat
hTe
stSc
ores
-0.0
23**
*0.
049*
**58
.139
***
0.13
1***
0.06
5***
0.05
4***
0.36
4***
0.07
3***
-0.0
05**
*-0
.087
***
-0.0
36**
*
(0.0
02)
(0.0
03)
(2.2
41)
(0.0
03)
(0.0
02)
(0.0
02)
(0.0
06)
(0.0
04)
(0.0
02)
(0.0
04)
(0.0
02)
Eng
lish
Test
Scor
es-0
.013
***
0.04
9***
94.9
59**
*0.
094*
**0.
050*
**0.
049*
**0.
101*
**0.
371*
**-0
.013
***
-0.0
27**
*-0
.020
***
(0.0
02)
(0.0
04)
(2.0
39)
(0.0
03)
(0.0
02)
(0.0
02)
(0.0
04)
(0.0
06)
(0.0
02)
(0.0
05)
(0.0
02)
GPA
0.00
20.
012*
**25
.921
***
0.00
5-0
.008
**-0
.017
***
0.06
2***
0.08
0***
0.01
0***
0.00
30.
003
(0.0
04)
(0.0
04)
(2.2
95)
(0.0
06)
(0.0
03)
(0.0
03)
(0.0
06)
(0.0
04)
(0.0
03)
(0.0
05)
(0.0
04)
Lea
rnin
gSk
ills
GPA
-0.0
44**
*0.
065*
**-1
2.38
3***
0.23
6***
0.14
4***
0.14
7***
0.02
5***
0.03
3***
-0.0
70**
*-0
.101
***
-0.0
64**
*
(0.0
02)
(0.0
04)
(1.5
71)
(0.0
05)
(0.0
03)
(0.0
03)
(0.0
05)
(0.0
04)
(0.0
04)
(0.0
04)
(0.0
03)
Eff
ortG
PA0.
002
0.00
2-0
.817
-0.0
07-0
.006
*-0
.018
***
-0.0
05-0
.009
*0.
020*
**0.
018*
**0.
000
(0.0
03)
(0.0
04)
(1.6
13)
(0.0
05)
(0.0
04)
(0.0
03)
(0.0
06)
(0.0
05)
(0.0
03)
(0.0
06)
(0.0
03)
Susp
ende
d0.
043*
**-0
.045
***
-17.
708*
*-0
.130
***
-0.0
70**
*-0
.099
***
-0.0
24-0
.054
***
0.23
5***
0.07
5***
0.04
7***
(0.0
09)
(0.0
12)
(8.3
64)
(0.0
15)
(0.0
08)
(0.0
08)
(0.0
17)
(0.0
12)
(0.0
21)
(0.0
16)
(0.0
08)
Log
Abs
ence
s0.
021*
**-0
.033
***
-0.8
06-0
.092
***
-0.0
51**
*-0
.041
***
-0.0
28**
*-0
.002
0.00
4***
0.24
9***
0.02
7***
(0.0
01)
(0.0
02)
(0.9
73)
(0.0
02)
(0.0
01)
(0.0
01)
(0.0
02)
(0.0
02)
(0.0
01)
(0.0
05)
(0.0
01)
Hel
dB
ack
0.24
9***
0.11
7***
36.7
25**
0.06
9**
0.03
7*0.
025
0.28
9***
0.20
7***
-0.1
02**
*-0
.272
***
-0.0
48**
(0.0
24)
(0.0
39)
(15.
603)
(0.0
27)
(0.0
19)
(0.0
18)
(0.0
42)
(0.0
37)
(0.0
15)
(0.0
31)
(0.0
19)
Obs
erva
tions
134,
357
100,
691
59,5
8232
1,23
625
1,46
025
1,46
016
0,38
515
9,91
134
3,72
130
2,90
224
0,20
4
R-s
quar
ed0.
422
0.18
00.
689
0.32
40.
320
0.33
40.
579
0.58
70.
049
0.29
80.
136
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.T
his
tabl
ere
port
sth
epr
edic
tive
effe
ctof
each
ofth
em
easu
res
ofst
uden
tach
ieve
men
t(se
eTa
ble
A.2
)in
grad
es3,
4or
5on
stud
ents
’hi
ghsc
hool
outc
omes
.Sp
ecifi
cally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
em
easu
res
ofst
uden
tac
hiev
emen
tfro
man
OL
Sre
gres
sion
ofth
est
uden
ts’h
igh
scho
olou
tcom
eon
the
stud
ents
’mea
sure
sin
grad
es3-
5,al
ong
with
the
base
line
cont
rols
desc
ribe
din
sect
ion
4.1.
Mat
han
dE
nglis
hte
stsc
ores
,G
PA,l
earn
ing-
skill
sG
PA,a
ndef
fort
GPA
have
each
been
norm
aliz
edat
the
grad
e-ye
arle
vel.
Susp
ende
dan
dH
eld
Bac
kar
ein
dica
tors
forb
eing
susp
ende
dan
dbe
ing
held
back
agr
ade,
resp
ectiv
ely.
Log
Abs
ence
sis
the
log
num
bero
fday
sab
sent
.The
base
line
cont
rols
incl
ude
lags
ofa
cubi
cpo
lyno
mia
loft
hest
uden
t’sm
ath
test
scor
e,E
nglis
hte
stsc
ore,
GPA
,and
effo
rtG
PA;a
cubi
cpo
lyno
mia
lsof
lagg
edcl
ass
and
grad
ele
velm
eans
ofea
chof
thos
eva
riab
les;
and
curr
entE
nglis
hle
arne
rsta
tus
and
lagg
edlo
gda
ysab
sent
,sus
pens
ions
,and
held
back
fort
hein
divi
dual
,the
clas
sav
erag
e,an
dth
egr
ade
aver
age.
Eac
hof
thes
eva
riab
les
exce
ptE
nglis
hle
arne
rsta
tus
isfu
llyin
tera
cted
with
grad
efix
edef
fect
san
da
cont
rolf
orcl
ass
size
isin
clud
ed.S
tand
ard
erro
rscl
uste
red
atth
em
odal
high
scho
olle
vela
rere
port
edin
pare
nthe
ses.
***
p<
0.01
, **
p<
0.05
,and
*p
<0.
10.
71
Table A.4: Correlation of All Elementary School Teacher Value-Added Measures
Grades 3-5 VAMath Test
Scores
English
Test ScoresGPA
Learning
Skills GPAEffort GPA
Log
Absences
Days
SuspendedHeld Back
Math Test Scores 1
English Test Scores 0.761 1
GPA 0.174 0.199 1
Learning Skills GPA 0.145 0.158 0.683 1
Effort GPA 0.142 0.154 0.767 0.760 1
Log Absences -0.064 -0.031 -0.122 -0.110 -0.111 1
Days Suspended -0.028 -0.033 -0.036 -0.078 -0.040 0.062 1
Held Back -0.023 -0.019 -0.016 -0.004 -0.011 0.010 0.006 1
Note: The sample includes students in grades 3-5 who attended high school in the LAUSD. The unit of observation is a student-academic year.Table 4 reports the correlations between measures of grades 3-5 teacher value-added. Each measure of teacher value-added is created as describedin section 4.1.
72
Tabl
eA
.5:E
ffec
tofE
lem
enta
rySc
hool
Teac
herV
alue
-Add
edon
Hig
hSc
hool
Out
com
es
Test
Scor
eV
alue
-Add
ed
Gra
des
3-5
VAL
AU
SD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
-0.0
02-0
.001
6.39
2***
0.00
40.
003*
0.00
4***
0.02
3***
0.01
7***
-0.0
00-0
.002
0.00
0
(0.0
02)
(0.0
02)
(1.0
95)
(0.0
02)
(0.0
01)
(0.0
01)
(0.0
03)
(0.0
02)
(0.0
01)
(0.0
03)
(0.0
01)
Obs
erva
tions
179,
795
136,
152
81,7
0736
6,79
529
6,28
929
6,28
919
7,81
619
7,13
039
6,37
334
7,66
428
1,64
9
R-s
quar
ed0.
293
0.15
20.
631
0.25
50.
245
0.24
90.
512
0.52
60.
040
0.26
50.
112
Beh
avio
rVal
ue-A
dded
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Beh
avio
rVA
-0.0
04*
0.00
9***
2.64
10.
013*
**0.
006*
**0.
004*
0.01
4**
0.00
7-0
.003
**-0
.015
***
-0.0
05**
(0.0
02)
(0.0
03)
(2.1
97)
(0.0
04)
(0.0
02)
(0.0
02)
(0.0
05)
(0.0
04)
(0.0
02)
(0.0
05)
(0.0
02)
Obs
erva
tions
135,
786
102,
517
60,6
9429
3,03
223
3,07
823
3,07
815
2,34
515
1,82
031
6,12
727
7,33
522
1,75
7
R-s
quar
ed0.
293
0.14
50.
616
0.24
40.
234
0.23
90.
500
0.51
20.
040
0.26
70.
108
Lea
rnin
gSk
ills
Val
ue-A
dded
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Lea
rnin
gSk
ills
VA-0
.002
0.00
31.
484
0.00
40.
001
-0.0
000.
004
0.00
5-0
.003
*-0
.005
-0.0
01
(0.0
02)
(0.0
02)
(1.9
31)
(0.0
05)
(0.0
03)
(0.0
03)
(0.0
05)
(0.0
04)
(0.0
02)
(0.0
06)
(0.0
02)
Obs
erva
tions
136,
236
102,
901
60,9
1529
3,78
123
3,69
623
3,69
615
2,80
315
2,27
231
6,97
727
8,05
522
2,33
5
R-s
quar
ed0.
293
0.14
50.
617
0.24
40.
234
0.23
90.
500
0.51
20.
040
0.26
60.
108
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
eth
ree
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
ed(s
eeTa
ble
4)on
stud
ents
’hig
hsc
hool
outc
omes
.Spe
cific
ally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
norm
aliz
edm
easu
res
ofte
ache
rval
ue-a
dded
from
anO
LS
regr
essi
onof
the
stud
ents
’hig
hsc
hool
outc
ome
onon
eof
the
thre
em
easu
res
ofte
ache
rval
ue-a
dded
fort
hest
uden
ts’t
each
ers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.T
heba
selin
eco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fth
est
uden
t’sm
ath
test
scor
e,E
nglis
hte
stsc
ore,
GPA
,and
effo
rtG
PA;a
cubi
cpo
lyno
mia
lsof
lagg
edcl
ass
and
grad
ele
velm
eans
ofea
chof
thos
eva
riab
les;
and
curr
entE
nglis
hle
arne
rsta
tus
and
lagg
edlo
gda
ysab
sent
,sus
pens
ions
,and
held
back
fort
hein
divi
dual
,the
clas
sav
erag
e,an
dth
egr
ade
aver
age.
Eac
hof
thes
eva
riab
les
exce
ptE
nglis
hle
arne
rsta
tus
isfu
llyin
tera
cted
with
grad
efix
edef
fect
san
da
cont
rolf
orcl
ass
size
isin
clud
ed.
Stan
dard
erro
rscl
uste
red
atth
ele
velo
fthe
pred
icte
dhi
ghsc
hool
.are
repo
rted
inpa
rent
hese
s.**
*p
<0.
01, *
*p
<0.
05,a
nd*
p<
0.10
.
73
Tabl
eA
.6:T
heIn
tera
ctio
nE
ffec
tBet
wee
nTe
st-S
core
and
Beh
avio
rVal
ue-A
dded
onH
igh
Scho
olO
utco
mes
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
-0.0
02-0
.002
6.14
0***
0.00
20.
002
0.00
4***
0.02
2***
0.01
6***
0.00
10.
001
0.00
1
(0.0
02)
(0.0
02)
(1.2
35)
(0.0
02)
(0.0
01)
(0.0
01)
(0.0
03)
(0.0
02)
(0.0
01)
(0.0
03)
(0.0
01)
Beh
avio
rVA
-0.0
030.
009*
**1.
716
0.01
2***
0.00
6**
0.00
30.
010*
0.00
4-0
.004
**-0
.015
***
-0.0
05**
(0.0
02)
(0.0
03)
(2.2
40)
(0.0
04)
(0.0
02)
(0.0
02)
(0.0
05)
(0.0
05)
(0.0
02)
(0.0
05)
(0.0
02)
Inte
ract
ion
Term
0.00
10.
001
0.62
50.
001
0.00
10.
000
0.00
00.
001
0.00
0-0
.001
-0.0
00
(0.0
02)
(0.0
02)
(1.2
02)
(0.0
03)
(0.0
01)
(0.0
01)
(0.0
03)
(0.0
03)
(0.0
01)
(0.0
03)
(0.0
01)
Obs
erva
tions
135,
786
102,
517
60,6
9429
3,02
123
3,07
823
3,07
815
2,34
515
1,82
031
6,11
627
7,33
122
1,75
7
R-s
quar
ed0.
293
0.14
50.
617
0.24
40.
234
0.23
90.
500
0.51
20.
040
0.26
70.
108
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
ete
st-s
core
and
beha
vior
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
ed(s
eeTa
ble
4)on
stud
ents
’hig
hsc
hool
outc
omes
alon
gw
ithth
ein
tera
ctio
nbe
twee
nth
etw
om
easu
res.
Spec
ifica
lly,e
ach
colu
mn
ofth
eta
ble
repo
rts
the
coef
ficie
nts
onea
chof
the
two
norm
aliz
edm
easu
res
ofte
ache
rva
lue-
adde
dal
ong
with
the
inte
ract
ion
betw
een
the
two
mea
sure
sfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’
high
scho
olou
tcom
eon
the
two
mea
sure
sof
teac
herv
alue
-add
edan
dth
ere
inte
ract
ion
fort
hest
uden
ts’t
each
ers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.T
heba
selin
eco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fthe
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ners
tatu
san
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,a
ndhe
ldba
ckfo
rthe
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ners
tatu
sis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.Sta
ndar
der
rors
clus
tere
dat
the
mod
alhi
ghsc
hool
leve
lare
repo
rted
inpa
rent
hese
s.**
*p
<0.
01, *
*p
<0.
05,a
nd*
p<
0.10
.
74
Tabl
eA
.7:E
ffec
tofE
lem
enta
rySc
hool
Teac
herV
alue
-Add
edon
Hig
hSc
hool
Out
com
es-T
each
er-G
rade
VA
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
-0.0
09**
*0.
005
19.9
88**
*0.
018*
**0.
013*
**0.
019*
**0.
055*
**0.
054*
**0.
001
-0.0
080.
000
(0.0
03)
(0.0
04)
(2.8
95)
(0.0
05)
(0.0
03)
(0.0
03)
(0.0
08)
(0.0
06)
(0.0
03)
(0.0
07)
(0.0
03)
Beh
avio
rVA
0.00
00.
007
5.15
7**
0.02
2***
0.01
4***
0.01
2***
0.02
0***
0.01
3**
-0.0
01-0
.025
***
-0.0
07**
(0.0
03)
(0.0
04)
(2.5
44)
(0.0
06)
(0.0
04)
(0.0
04)
(0.0
06)
(0.0
05)
(0.0
03)
(0.0
07)
(0.0
03)
Lea
rnin
gSk
ills
VA0.
001
0.00
31.
676
0.00
2-0
.003
-0.0
05-0
.004
-0.0
03-0
.008
**-0
.000
0.00
2
(0.0
04)
(0.0
04)
(2.8
55)
(0.0
09)
(0.0
05)
(0.0
05)
(0.0
07)
(0.0
06)
(0.0
04)
(0.0
08)
(0.0
04)
Obs
erva
tions
92,8
0370
,520
41,6
6120
9,36
416
5,85
816
5,85
810
8,15
710
7,78
022
5,79
319
8,08
715
7,91
3
R-s
quar
ed0.
362
0.14
10.
606
0.23
30.
223
0.22
90.
487
0.49
70.
033
0.26
00.
101
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
eth
ree
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
ed(s
eeTa
ble
4)on
stud
ents
’hig
hsc
hool
outc
omes
.Spe
cific
ally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
norm
aliz
edm
easu
res
ofte
ache
rva
lue-
adde
dfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’
high
scho
olou
tcom
eon
the
thre
em
easu
res
ofte
ache
rva
lue-
adde
dfo
rth
est
uden
ts’
teac
hers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.T
heva
lue-
adde
dm
easu
res
wer
eco
nstr
ucte
dby
teac
her-
grad
era
ther
than
byte
ache
r.T
heba
selin
eco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fth
est
uden
t’sm
ath
test
scor
e,E
nglis
hte
stsc
ore,
GPA
,and
effo
rtG
PA;a
cubi
cpo
lyno
mia
lsof
lagg
edcl
ass
and
grad
ele
velm
eans
ofea
chof
thos
eva
riab
les;
and
curr
entE
nglis
hle
arne
rsta
tus
and
lagg
edlo
gda
ysab
sent
,su
spen
sion
s,an
dhe
ldba
ckfo
rthe
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ners
tatu
sis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.Sta
ndar
der
rors
clus
tere
dat
the
mod
alhi
ghsc
hool
leve
lare
repo
rted
inpa
rent
hese
s.**
*p
<0.
01, *
*p
<0.
05,a
nd*
p<
0.10
.
75
Tabl
eA
.8:E
ffec
tofE
lem
enta
rySc
hool
Teac
herV
alue
-Add
edon
Hig
hSc
hool
Out
com
es-S
ingl
eFa
ctor
Inde
xO
utco
mes
Test
Scor
eV
alue
-Add
ed
Gra
des
3-5
VATe
stSc
ore
Inde
xG
PAIn
dex
Beh
avio
rs
Inde
x
Test
Scor
eVA
0.02
2***
0.00
6**
0.00
2
(0.0
03)
(0.0
03)
(0.0
02)
Obs
erva
tions
194,
271
296,
233
271,
186
R-s
quar
ed0.
577
0.26
30.
250
Beh
avio
rVal
ue-A
dded
Gra
des
3-5
VATe
stSc
ore
Inde
xG
PAIn
dex
Beh
avio
rs
Inde
x
Beh
avio
rVA
0.01
1**
0.01
1**
0.01
4***
(0.0
05)
(0.0
04)
(0.0
04)
Obs
erva
tions
149,
637
233,
037
213,
435
R-s
quar
ed0.
563
0.25
20.
243
Lea
rnin
gSk
ills
Val
ue-A
dded
Gra
des
3-5
VATe
stSc
ore
Inde
xG
PAIn
dex
Beh
avio
rs
Inde
x
Lea
rnin
gSk
ills
VA0.
006
0.00
10.
005
(0.0
04)
(0.0
06)
(0.0
05)
Obs
erva
tions
150,
085
233,
655
213,
992
R-s
quar
ed0.
563
0.25
20.
243
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
eth
ree
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
ed(s
eeTa
ble
4)on
stud
ents
’hig
hsc
hool
outc
omes
.Spe
cific
ally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
norm
aliz
edm
easu
res
ofte
ache
rva
lue-
adde
dfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’
high
scho
olou
tcom
eon
the
thre
em
easu
res
ofte
ache
rva
lue-
adde
dfo
rth
est
uden
ts’
teac
hers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.T
heba
selin
eco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fthe
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ners
tatu
san
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,a
ndhe
ldba
ckfo
rthe
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ners
tatu
sis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.Sta
ndar
der
rors
clus
tere
dat
the
leve
loft
hehi
ghsc
hool
each
stud
ent
isex
pect
edto
atte
ndar
ere
port
edin
pare
nthe
ses.
76
Tabl
eA
.9:
Eff
ect
ofE
lem
enta
rySc
hool
Teac
her
Val
ue-A
dded
onH
igh
Scho
olO
utco
mes
-A
djus
tfo
rM
ultip
leH
ypot
hesi
sTe
stin
gA
cros
sA
llIn
dex
Out
com
es
Pool
edG
rade
s3-
5VA
Out
put
Test
Scor
e
Inde
xG
PAIn
dex
Beh
avio
rs
Inde
x
Test
Scor
eVA
β0.
021
0.00
60.
000
SE(0
.003
)(0
.003
)(0
.003
)
p−
valu
e[0
.000
][0
.031
][0
.895
]
p−
RW[0
.000
][0
.012
][0
.865
]
p−
WY
[0.0
00]
[0.0
60]
[0.8
97]
p−
Bon
f[0
.000
][0
.061
][0
.895
]
p−
Sida
k[0
.000
][0
.060
][0
.895
]
Beh
avio
rVA
β0.
009
0.01
30.
014
SE(0
.005
)(0
.005
)(0
.004
)
p−
valu
e[0
.110
][0
.011
][0
.002
]
p−
RW[0
.062
][0
.013
][0
.004
]
p−
WY
[0.1
48]
[0.0
48]
[0.0
23]
p−
Bon
f[0
.110
][0
.022
][0
.006
]
p−
Sida
k[0
.110
][0
.022
][0
.006
]
Lea
rnin
gSk
ills
VAβ
-0.0
02-0
.006
-0.0
01
SE(0
.004
)(0
.006
)(0
.005
)
p−
valu
e[0
.634
][0
.324
][0
.874
]
p−
RW[0
.729
][0
.343
][0
.831
]
p−
WY
[0.8
31]
[0.5
27]
[0.8
74]
p−
Bon
f[1
.000
][0
.972
][1
.000
]
p−
Sida
k[0
.866
][0
.691
][0
.874
]
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
eth
ree
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
ed(s
eeTa
ble
4)on
stud
ents
’hig
hsc
hool
outc
omes
.Spe
cific
ally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
norm
aliz
edm
easu
res
ofte
ache
rva
lue-
adde
dfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’
high
scho
olou
tcom
eon
the
thre
em
easu
res
ofte
ache
rva
lue-
adde
dfo
rth
est
uden
ts’
teac
hers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.T
heba
selin
eco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fthe
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ners
tatu
san
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,a
ndhe
ldba
ckfo
rthe
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ners
tatu
sis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.Sta
ndar
der
rors
clus
tere
dat
the
leve
loft
hehi
ghsc
hool
each
stud
ent
isex
pect
edto
atte
ndar
ere
port
edin
pare
nthe
ses.
P-va
lues
are
repo
rted
inbr
acke
tsth
atar
eca
lcul
ated
usin
gcl
uste
red
stan
dard
erro
rs,t
heR
oman
oW
olf(
2005
,201
6)co
rrec
tion,
the
Wes
tfall
You
ng(1
993)
algo
rith
m,t
heB
onfe
rron
icor
rect
ion,
and
the
Sida
k-H
olm
corr
ectio
n.
77
Tabl
eA
.10:
Eff
ecto
fEle
men
tary
Scho
olTe
ache
rVal
ue-A
dded
onH
igh
Scho
olO
utco
mes
-Adj
ustf
orM
ultip
leH
ypot
hesi
sTe
stin
gby
Out
com
eV
aria
ble
Fam
ily
Pool
edG
rade
s3-
5VA
Out
put
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
ESA
TSc
ore
GPA
Eff
ortG
PAC
oope
ratio
n
GPA
LA
USD
Dro
pout
Took
SAT
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
β0.
022
0.01
66.
237
0.00
20.
003
0.00
5-0
.002
-0.0
020.
001
0.00
10.
001
SE(0
.003
)(0
.002
)(1
.262
)(0
.002
)(0
.001
)(0
.001
)(0
.002
)(0
.002
)(0
.001
)(0
.003
)(0
.001
)
p−
valu
e[0
.000
][0
.000
][0
.000
][0
.307
][0
.077
][0
.001
][0
.322
][0
.468
][0
.455
][0
.833
][0
.543
]
p−
RW[0
.000
][0
.000
][0
.000
][0
.190
][0
.036
][0
.000
][0
.635
][0
.770
][0
.770
][0
.776
][0
.770
]
p−
WY
[0.0
00]
[0.0
00]
[0.0
00]
[0.3
27]
[0.1
24]
[0.0
04]
[0.8
46]
[0.8
98]
[0.8
98]
[0.8
98]
[0.8
98]
p−
Bon
f[0
.000
][0
.000
][0
.000
][0
.307
][0
.154
][0
.002
][1
.000
][1
.000
][1
.000
][1
.000
][1
.000
]
p−
Sida
k[0
.000
][0
.000
][0
.000
][0
.307
][0
.148
][0
.002
][0
.857
][0
.912
][0
.912
][0
.912
][0
.912
]
Beh
avio
rVA
β0.
013
0.00
41.
955
0.01
30.
007
0.00
5-0
.003
0.01
0-0
.003
-0.0
16-0
.006
SE(0
.006
)(0
.005
)(2
.139
)(0
.004
)(0
.003
)(0
.002
)(0
.002
)(0
.003
)(0
.002
)(0
.005
)(0
.003
)
p−
valu
e[0
.026
][0
.390
][0
.365
][0
.003
][0
.007
][0
.046
][0
.188
][0
.004
][0
.178
][0
.003
][0
.035
]
p−
RW[0
.042
][0
.446
][0
.446
][0
.003
][0
.004
][0
.016
][0
.141
][0
.002
][0
.141
][0
.002
][0
.028
]
p−
WY
[0.1
33]
[0.5
94]
[0.5
94]
[0.0
18]
[0.0
23]
[0.0
64]
[0.3
24]
[0.0
31]
[0.3
24]
[0.0
29]
[0.1
30]
p−
Bon
f[0
.078
][0
.729
][0
.729
][0
.009
][0
.014
][0
.046
][0
.356
][0
.017
][0
.356
][0
.013
][0
.106
]
p−
Sida
k[0
.076
][0
.596
][0
.596
][0
.009
][0
.014
][0
.046
][0
.324
][0
.017
][0
.324
][0
.013
][0
.102
]
Lea
rnin
gSk
ills
VAβ
-0.0
050.
001
-0.5
47-0
.002
-0.0
03-0
.003
0.00
0-0
.001
-0.0
020.
002
0.00
1
SE(0
.005
)(0
.004
)(1
.756
)(0
.005
)(0
.003
)(0
.003
)(0
.002
)(0
.003
)(0
.002
)(0
.006
)(0
.002
)
p−
valu
e[0
.279
][0
.854
][0
.757
][0
.682
][0
.334
][0
.287
][0
.949
][0
.611
][0
.368
][0
.770
][0
.697
]
p−
RW[0
.345
][0
.884
][0
.884
][0
.593
][0
.281
][0
.270
][0
.938
][0
.928
][0
.724
][0
.938
][0
.938
]
p−
WY
[0.5
49]
[0.9
32]
[0.9
32]
[0.6
87]
[0.4
42]
[0.4
42]
[0.9
71]
[0.9
71]
[0.8
84]
[0.9
71]
[0.9
71]
p−
Bon
f[0
.838
][1
.000
][1
.000
][0
.860
][0
.860
][0
.860
][1
.000
][1
.000
][1
.000
][1
.000
][1
.000
]
p−
Sida
k[0
.626
][0
.941
][0
.941
][0
.682
][0
.637
][0
.637
][0
.977
][0
.977
][0
.899
][0
.977
][0
.977
]
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
eth
ree
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
ed(s
eeTa
ble
4)on
stud
ents
’hig
hsc
hool
outc
omes
.Spe
cific
ally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
norm
aliz
edm
easu
res
ofte
ache
rva
lue-
adde
dfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’
high
scho
olou
tcom
eon
the
thre
em
easu
res
ofte
ache
rva
lue-
adde
dfo
rth
est
uden
ts’
teac
hers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.T
heba
selin
eco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fthe
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ners
tatu
san
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,a
ndhe
ldba
ckfo
rthe
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ners
tatu
sis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.Sta
ndar
der
rors
clus
tere
dat
the
leve
loft
hehi
ghsc
hool
each
stud
ent
isex
pect
edto
atte
ndar
ere
port
edin
pare
nthe
ses.
P-va
lues
are
repo
rted
inbr
acke
tsth
atar
eca
lcul
ated
usin
gcl
uste
red
stan
dard
erro
rs,t
heR
oman
oW
olf(
2005
,201
6)co
rrec
tion,
the
Wes
tfall
You
ng(1
993)
algo
rith
m,t
heB
onfe
rron
icor
rect
ion,
and
the
Sida
k-H
olm
corr
ectio
n.T
hefa
mili
esar
eth
eth
ree
test
scor
es,t
heth
ree
GPA
mea
sure
s,an
dal
loth
erm
easu
res.
78
Tabl
eA
.11:
Eff
ect
ofE
lem
enta
rySc
hool
Teac
her
Val
ue-A
dded
onH
igh
Scho
olO
utco
mes
-A
djus
tfo
rM
ultip
leH
ypot
hesi
sTe
stin
gA
cros
sA
llO
utco
mes
Pool
edG
rade
s3-
5VA
Out
put
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
ESA
TSc
ore
GPA
Eff
ortG
PAC
oope
ratio
n
GPA
LA
USD
Dro
pout
Took
SAT
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
β0.
022
0.01
66.
237
0.00
20.
003
0.00
5-0
.002
-0.0
020.
001
0.00
10.
001
SE(0
.003
)(0
.002
)(1
.262
)(0
.002
)(0
.001
)(0
.001
)(0
.002
)(0
.002
)(0
.001
)(0
.003
)(0
.001
)
p−
valu
e[0
.000
][0
.000
][0
.000
][0
.307
][0
.077
][0
.001
][0
.322
][0
.468
][0
.455
][0
.833
][0
.543
]
p−
RW[0
.000
][0
.000
][0
.000
][0
.646
][0
.135
][0
.000
][0
.646
][0
.770
][0
.770
][0
.776
][0
.770
]
p−
WY
[0.0
00]
[0.0
00]
[0.0
01]
[0.8
60]
[0.3
91]
[0.0
11]
[0.8
60]
[0.8
98]
[0.8
98]
[0.8
98]
[0.8
98]
p−
Bon
f[0
.000
][0
.000
][0
.000
][1
.000
][0
.540
][0
.005
][1
.000
][1
.000
][1
.000
][1
.000
][1
.000
]
p−
Sida
k[0
.000
][0
.000
][0
.000
][0
.890
][0
.430
][0
.005
][0
.890
][0
.912
][0
.912
][0
.912
][0
.912
]
Beh
avio
rVA
β0.
013
0.00
41.
955
0.01
30.
007
0.00
5-0
.003
0.01
0-0
.003
-0.0
16-0
.006
SE(0
.006
)(0
.005
)(2
.139
)(0
.004
)(0
.003
)(0
.002
)(0
.002
)(0
.003
)(0
.002
)(0
.005
)(0
.003
)
p−
valu
e[0
.026
][0
.390
][0
.365
][0
.003
][0
.007
][0
.046
][0
.188
][0
.004
][0
.178
][0
.003
][0
.035
]
p−
RW[0
.061
][0
.446
][0
.446
][0
.010
][0
.020
][0
.083
][0
.318
][0
.012
][0
.318
][0
.010
][0
.074
]
p−
WY
[0.2
17]
[0.5
94]
[0.5
94]
[0.0
66]
[0.0
98]
[0.2
68]
[0.5
66]
[0.0
78]
[0.5
66]
[0.0
66]
[0.2
51]
p−
Bon
f[0
.182
][0
.729
][0
.729
][0
.030
][0
.058
][0
.230
][0
.711
][0
.038
][0
.711
][0
.030
][0
.212
]
p−
Sida
k[0
.169
][0
.596
][0
.596
][0
.029
][0
.056
][0
.210
][0
.543
][0
.038
][0
.543
][0
.029
][0
.194
]
Lea
rnin
gSk
ills
VAβ
-0.0
050.
001
-0.5
47-0
.002
-0.0
03-0
.003
0.00
0-0
.001
-0.0
020.
002
0.00
1
SE(0
.005
)(0
.004
)(1
.756
)(0
.005
)(0
.003
)(0
.003
)(0
.002
)(0
.003
)(0
.002
)(0
.006
)(0
.002
)
p−
valu
e[0
.279
][0
.854
][0
.757
][0
.682
][0
.334
][0
.287
][0
.949
][0
.611
][0
.368
][0
.770
][0
.697
]
p−
RW[0
.705
][0
.987
][0
.987
][0
.987
][0
.782
][0
.706
][0
.987
][0
.979
][0
.827
][0
.987
][0
.987
]
p−
WY
[0.8
95]
[0.9
96]
[0.9
96]
[0.9
96]
[0.9
32]
[0.8
95]
[0.9
96]
[0.9
95]
[0.9
48]
[0.9
96]
[0.9
96]
p−
Bon
f[1
.000
][1
.000
][1
.000
][1
.000
][1
.000
][1
.000
][1
.000
][1
.000
][1
.000
][1
.000
][1
.000
]
p−
Sida
k[0
.973
][0
.999
][0
.999
][0
.999
][0
.974
][0
.973
][0
.999
][0
.999
][0
.974
][0
.999
][0
.999
]
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
eth
ree
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
ed(s
eeTa
ble
4)on
stud
ents
’hig
hsc
hool
outc
omes
.Spe
cific
ally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
norm
aliz
edm
easu
res
ofte
ache
rva
lue-
adde
dfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’
high
scho
olou
tcom
eon
the
thre
em
easu
res
ofte
ache
rva
lue-
adde
dfo
rth
est
uden
ts’
teac
hers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.T
heba
selin
eco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fthe
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ners
tatu
san
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,a
ndhe
ldba
ckfo
rthe
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ners
tatu
sis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.Sta
ndar
der
rors
clus
tere
dat
the
leve
loft
hehi
ghsc
hool
each
stud
ent
isex
pect
edto
atte
ndar
ere
port
edin
pare
nthe
ses.
P-va
lues
are
repo
rted
inbr
acke
tsth
atar
eca
lcul
ated
usin
gcl
uste
red
stan
dard
erro
rs,t
heR
oman
oW
olf(
2005
,201
6)co
rrec
tion,
the
Wes
tfall
You
ng(1
993)
algo
rith
m,t
heB
onfe
rron
icor
rect
ion,
and
the
Sida
k-H
olm
corr
ectio
n.
79
Tabl
eA
.12:
Eff
ecto
fAll
Ele
men
tary
Scho
olTe
ache
rVal
ue-A
dded
Mea
sure
son
Hig
hSc
hool
Out
com
esfr
omSe
para
teR
egre
ssio
ns
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Mat
hTe
stSc
ores
VA-0
.001
-0.0
004.
364*
**0.
001
0.00
10.
003*
0.02
5***
0.01
3***
0.00
0-0
.001
0.00
1
(0.0
02)
(0.0
03)
(1.3
08)
(0.0
03)
(0.0
02)
(0.0
02)
(0.0
03)
(0.0
03)
(0.0
02)
(0.0
03)
(0.0
01)
Eng
lish
Test
Scor
esVA
-0.0
03-0
.002
12.3
92**
*0.
009*
**0.
006*
**0.
008*
**0.
034*
**0.
029*
**-0
.000
-0.0
04-0
.000
(0.0
02)
(0.0
03)
(1.7
24)
(0.0
03)
(0.0
02)
(0.0
02)
(0.0
04)
(0.0
04)
(0.0
02)
(0.0
04)
(0.0
02)
GPA
VA-0
.002
0.01
5***
5.52
40.
009
0.00
20.
000
0.01
30.
015*
*-0
.003
-0.0
07-0
.006
*
(0.0
04)
(0.0
04)
(3.7
99)
(0.0
07)
(0.0
05)
(0.0
05)
(0.0
10)
(0.0
07)
(0.0
02)
(0.0
08)
(0.0
03)
5L
earn
ing
Skill
sG
PAVA
-0.0
040.
003
4.42
60.
006
0.00
20.
002
0.00
80.
010
-0.0
03-0
.008
-0.0
02
(0.0
04)
(0.0
04)
(3.6
41)
(0.0
08)
(0.0
05)
(0.0
05)
(0.0
08)
(0.0
07)
(0.0
03)
(0.0
08)
(0.0
03)
9L
earn
ing
Skill
sG
PAVA
-0.0
050.
003
0.25
40.
003
-0.0
02-0
.003
0.00
20.
004
-0.0
05*
-0.0
11-0
.000
(0.0
03)
(0.0
04)
(3.2
04)
(0.0
08)
(0.0
04)
(0.0
04)
(0.0
07)
(0.0
05)
(0.0
03)
(0.0
10)
(0.0
03)
Eff
ortG
PAVA
-0.0
020.
010*
*2.
965
0.01
00.
002
-0.0
020.
011
0.01
0-0
.005
*-0
.005
-0.0
04
(0.0
04)
(0.0
04)
(3.6
79)
(0.0
07)
(0.0
05)
(0.0
05)
(0.0
09)
(0.0
07)
(0.0
03)
(0.0
09)
(0.0
03)
Log
Abs
ence
sVA
0.00
7-0
.001
-11.
244*
*-0
.025
***
-0.0
20**
*-0
.017
**-0
.031
**-0
.012
0.00
20.
017*
0.01
3
(0.0
08)
(0.0
10)
(4.3
16)
(0.0
09)
(0.0
07)
(0.0
08)
(0.0
15)
(0.0
13)
(0.0
05)
(0.0
10)
(0.0
08)
Day
sSu
spen
ded
VA0.
012*
-0.0
24**
-0.6
61-0
.019
-0.0
08-0
.003
-0.0
32*
-0.0
060.
008*
*0.
053*
**-0
.000
(0.0
07)
(0.0
09)
(7.0
91)
(0.0
12)
(0.0
06)
(0.0
05)
(0.0
16)
(0.0
13)
(0.0
04)
(0.0
14)
(0.0
05)
Hel
dB
ack
VA0.
002
-0.0
0612
.345
*-0
.037
***
-0.0
10-0
.005
0.00
90.
018
0.00
60.
002
0.01
0
(0.0
13)
(0.0
14)
(7.1
53)
(0.0
12)
(0.0
09)
(0.0
08)
(0.0
18)
(0.0
17)
(0.0
06)
(0.0
15)
(0.0
12)
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
em
easu
res
ofel
emen
tary
scho
olte
ache
rva
lue-
adde
d(s
eeTa
ble
A.4
)on
stud
ents
’hi
ghsc
hool
outc
omes
.Sp
ecifi
cally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eno
rmal
ized
mea
sure
sof
teac
herv
alue
-add
edfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’h
igh
scho
olou
tcom
eon
each
ofth
em
easu
res
ofte
ache
rval
ue-a
dded
fort
hest
uden
ts’t
each
ers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.T
heba
selin
eco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fthe
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ners
tatu
san
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,a
ndhe
ldba
ckfo
rthe
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ners
tatu
sis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.Sta
ndar
der
rors
clus
tere
dat
the
mod
alhi
ghsc
hool
leve
lare
repo
rted
inpa
rent
hese
s.**
*p
<0.
01, *
*p
<0.
05,a
nd*
p<
0.10
.
80
Tabl
eA
.13:
Eff
ecto
fAll
Ele
men
tary
Scho
olTe
ache
rVal
ue-A
dded
Mea
sure
son
Hig
hSc
hool
Out
com
es
Gra
des
3-5
VAL
AU
SD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Mat
hTe
stSc
ores
VA0.
003
0.00
3-8
.234
***
-0.0
14**
*-0
.008
***
-0.0
06**
-0.0
00-0
.018
***
0.00
20.
011*
*0.
005
(0.0
04)
(0.0
04)
(2.2
90)
(0.0
05)
(0.0
03)
(0.0
03)
(0.0
06)
(0.0
05)
(0.0
03)
(0.0
05)
(0.0
03)
Eng
lish
Test
Scor
esVA
-0.0
06-0
.007
18.2
00**
*0.
019*
**0.
013*
**0.
013*
**0.
033*
**0.
043*
**-0
.001
-0.0
12-0
.004
(0.0
04)
(0.0
04)
(2.8
37)
(0.0
05)
(0.0
03)
(0.0
03)
(0.0
07)
(0.0
06)
(0.0
03)
(0.0
07)
(0.0
04)
GPA
VA0.
001
0.02
4***
7.71
60.
003
0.00
10.
002
0.00
80.
016*
0.00
2-0
.002
-0.0
11**
(0.0
05)
(0.0
07)
(5.4
23)
(0.0
09)
(0.0
06)
(0.0
06)
(0.0
11)
(0.0
09)
(0.0
04)
(0.0
12)
(0.0
05)
Lea
rnin
gSk
ills
GPA
VA-0
.007
-0.0
18**
*-5
.377
-0.0
12-0
.007
-0.0
03-0
.018
*-0
.011
0.00
0-0
.011
0.00
8*
(0.0
06)
(0.0
06)
(4.8
13)
(0.0
08)
(0.0
05)
(0.0
05)
(0.0
10)
(0.0
09)
(0.0
03)
(0.0
11)
(0.0
04)
Eff
ortG
PAVA
0.00
30.
002
-1.6
020.
014*
0.00
5-0
.003
0.00
90.
001
-0.0
070.
009
-0.0
01
(0.0
05)
(0.0
08)
(5.8
99)
(0.0
08)
(0.0
05)
(0.0
05)
(0.0
12)
(0.0
10)
(0.0
05)
(0.0
10)
(0.0
05)
Log
Abs
ence
sVA
0.00
50.
000
-10.
348*
*-0
.024
**-0
.020
**-0
.017
**-0
.027
*-0
.010
0.00
10.
011
0.01
4*
(0.0
08)
(0.0
10)
(4.4
00)
(0.0
10)
(0.0
08)
(0.0
08)
(0.0
15)
(0.0
12)
(0.0
05)
(0.0
10)
(0.0
08)
Day
sSu
spen
ded
VA0.
011
-0.0
21**
1.99
0-0
.017
-0.0
07-0
.002
-0.0
26*
-0.0
010.
008*
*0.
053*
**-0
.002
(0.0
07)
(0.0
09)
(6.3
49)
(0.0
11)
(0.0
06)
(0.0
05)
(0.0
15)
(0.0
12)
(0.0
04)
(0.0
13)
(0.0
05)
Hel
dB
ack
VA0.
002
-0.0
0513
.027
*-0
.035
***
-0.0
09-0
.004
0.01
30.
021
0.00
5-0
.001
0.00
9
(0.0
13)
(0.0
14)
(7.2
67)
(0.0
12)
(0.0
08)
(0.0
07)
(0.0
17)
(0.0
17)
(0.0
05)
(0.0
15)
(0.0
12)
Obs
erva
tions
135,
786
102,
517
60,6
9429
3,02
123
3,07
823
3,07
815
2,34
515
1,82
031
6,11
627
7,33
122
1,75
7
R-s
quar
ed0.
293
0.14
50.
617
0.24
40.
234
0.23
90.
500
0.51
30.
040
0.26
70.
108
Not
e:T
his
tabl
ere
port
sth
eef
fect
ofa
stan
dard
devi
atio
nin
crea
sein
the
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
ed(s
eeTa
ble
A.4
)on
stud
ents
’hig
hsc
hool
outc
omes
.Spe
cific
ally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eno
rmal
ized
mea
sure
sof
teac
herv
alue
-add
edfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’h
igh
scho
olou
tcom
eon
the
mea
sure
sof
teac
herv
alue
-add
edfo
rthe
stud
ents
’tea
cher
sin
grad
es3,
4,or
5,al
ong
with
the
base
line
cont
rols
desc
ribe
din
sect
ion
4.1.
The
base
line
cont
rols
incl
ude
lags
ofa
cubi
cpo
lyno
mia
loft
hest
uden
t’sm
ath
test
scor
e,E
nglis
hte
stsc
ore,
GPA
,and
effo
rtG
PA;a
cubi
cpo
lyno
mia
lsof
lagg
edcl
ass
and
grad
ele
velm
eans
ofea
chof
thos
eva
riab
les;
and
curr
entE
nglis
hle
arne
rst
atus
and
lagg
edlo
gda
ysab
sent
,sus
pens
ions
,and
held
back
for
the
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ner
stat
usis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.St
anda
rder
rors
clus
tere
dat
the
mod
alhi
ghsc
hool
leve
lare
repo
rted
inpa
rent
hese
s.**
*p
<0.
01, *
*p
<0.
05,a
nd*
p<
0.10
.
81
Tabl
eA
.14:
Eff
ecto
fEle
men
tary
Scho
olTe
ache
rVal
ue-A
dded
onH
igh
Scho
olO
utco
mes
Usi
ngA
ltern
ativ
eG
roup
ing
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
-0.0
02-0
.002
6.23
8***
0.00
30.
003*
0.00
5***
0.02
2***
0.01
6***
0.00
10.
000
0.00
1
(0.0
02)
(0.0
02)
(1.2
65)
(0.0
02)
(0.0
01)
(0.0
01)
(0.0
03)
(0.0
03)
(0.0
01)
(0.0
03)
(0.0
01)
Beh
avio
rVA
-0.0
030.
005*
0.61
90.
011*
**0.
006*
**0.
004*
*0.
010*
*0.
000
-0.0
02-0
.015
***
-0.0
03
(0.0
02)
(0.0
03)
(1.5
01)
(0.0
03)
(0.0
02)
(0.0
02)
(0.0
04)
(0.0
04)
(0.0
02)
(0.0
04)
(0.0
02)
All
GPA
VA-0
.001
0.00
4*0.
849
0.00
3-0
.000
-0.0
020.
000
0.00
4-0
.003
-0.0
04-0
.002
(0.0
02)
(0.0
02)
(2.0
44)
(0.0
05)
(0.0
03)
(0.0
03)
(0.0
05)
(0.0
04)
(0.0
02)
(0.0
06)
(0.0
02)
Obs
erva
tions
135,
786
102,
517
60,6
9429
3,02
123
3,07
823
3,07
815
2,34
515
1,82
031
6,11
627
7,33
122
1,75
7
R-s
quar
ed0.
293
0.14
50.
617
0.24
40.
234
0.23
90.
500
0.51
20.
040
0.26
70.
108
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
eth
ree
alte
rnat
ive
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
edon
stud
ents
’hig
hsc
hool
outc
omes
.The
test
-sco
reva
lue-
adde
dis
com
pute
dus
ing
the
teac
hers
’val
ue-a
dded
form
ath
and
Eng
lish
test
scor
es.
The
beha
vior
valu
e-ad
ded
isco
mpu
ted
usin
gte
ache
rs’
valu
e-ad
ded
for
susp
ensi
ons,
log
days
abse
nt,a
ndno
tpro
gres
sing
toth
ene
xtgr
ade
ontim
e(h
eld
back
).T
heA
llG
PAva
lue-
adde
dis
com
pute
dus
ing
teac
hers
’val
ue-a
dded
forG
PA,e
ffor
tGPA
,wor
kan
dst
udy
habi
tGPA
,and
lear
ning
and
soci
alsk
ills
GPA
.Eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
alte
rnat
ive
mea
sure
sof
teac
herv
alue
-add
edfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’h
igh
scho
olou
tcom
eon
the
thre
eal
tern
ativ
em
easu
res
ofte
ache
rval
ue-a
dded
fort
hest
uden
ts’t
each
ers
ingr
ades
3,4,
or5
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.T
heba
selin
eco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fthe
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ners
tatu
san
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,a
ndhe
ldba
ckfo
rthe
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ners
tatu
sis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.Sta
ndar
der
rors
clus
tere
dat
the
mod
alhi
ghsc
hool
leve
lare
repo
rted
inpa
rent
hese
s.**
*p
<0.
01, *
*p
<0.
05,a
nd*
p<
0.10
.
82
Tabl
eA
.15:
Est
imat
esof
Fore
cast
Bia
s
Pool
edG
rade
s3-
5VA
Mat
hTe
st
Scor
e
EL
ATe
st
Scor
eG
PAH
abit
GPA
Eff
ortG
PAL
og
Abs
ence
sSu
spen
ded
Hel
dba
ck
Pane
lA:U
sing
Act
ualR
esid
ualiz
edC
orre
spon
ding
Out
com
eV
aria
ble
Cor
resp
ondi
ngTe
ache
r0.
989*
*0.
987
1.02
70.
997
1.02
01.
057
1.09
01.
195
Val
ue-A
dded
Mea
sure
(0.0
05)
(0.0
10)
(0.0
32)
(0.0
30)
(0.0
30)
(0.0
92)
(0.1
88)
(0.3
65)
Obs
erva
tions
668,
480
668,
712
407,
248
402,
882
402,
882
468,
446
476,
790
369,
890
Pane
lB:U
sing
Pred
icte
dC
orre
spon
ding
Out
com
eV
aria
ble
Cor
resp
ondi
ngTe
ache
r0.
004
0.00
10.
013*
-0.0
010.
004
-0.0
160.
007
-0.0
01
Val
ue-A
dded
Mea
sure
(0.0
03)
(0.0
05)
(0.0
07)
(0.0
05)
(0.0
04)
(0.0
23)
(0.0
11)
(0.0
03)
Obs
erva
tions
413,
826
413,
826
323,
910
323,
568
323,
568
344,
370
345,
914
322,
806
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Eac
hce
llin
Pane
lAre
port
sth
eco
effic
ient
onth
ein
dica
ted
elem
enta
rysc
hool
teac
herv
alue
-add
edm
easu
refr
oman
OL
Sre
gres
sion
ofst
uden
ts’r
esid
ualiz
ed(u
sing
the
base
line
cont
rols
)tes
tsco
re,G
PA,o
rbeh
avio
ralm
easu
reon
the
teac
herv
alue
-add
edm
easu
reca
lcul
ated
usin
gth
atou
tcom
e.Te
ache
rval
ued-
adde
dis
estim
ated
usin
gth
ele
ave-
outy
eara
ppro
ach
deta
iled
inse
ctio
n4.
1,an
dsc
aled
sobo
thth
eou
tcom
ean
dva
lue-
adde
dm
easu
rear
ein
the
sam
eun
its.
Eac
hce
llin
Pane
lBre
port
sth
eco
effic
ient
onth
ein
dica
ted
teac
her
valu
e-ad
ded
mea
sure
from
anO
LS
regr
essi
onof
stud
ents
’pr
edic
ted
outc
ome
(usi
ngdo
uble
-lag
ged
stud
enta
chie
vem
ent)
onte
ache
rs’
valu
e-ad
ded
mea
sure
afte
ral
lva
riab
les
are
resi
dual
ized
usin
gth
eba
selin
eco
ntro
ls.
The
pred
icte
dou
tcom
esar
ecr
eate
dby
estim
atin
gan
OL
Sre
gres
sion
ofth
eou
tcom
ein
dica
ted
inea
chco
lum
non
thir
d-or
der
poly
nom
ials
ofdo
uble
lagg
ed(y
eart−
2)m
ath
and
Eng
lish
test
scor
es,G
PA,e
ffor
tGPA
,lea
rnin
g-sk
ills
GPA
,log
abse
nces
,an
indi
cato
rfo
rsu
spen
sion
,an
indi
cato
rfo
rbe
ing
held
back
,and
anin
dica
toro
fEng
lish
lear
ners
tatu
s.T
heco
effic
ient
sob
tain
edfr
omth
isO
LS
regr
essi
onar
eth
enus
edto
pred
icts
tude
nts’
outc
omes
.Sta
ndar
der
rors
clus
tere
dat
the
mod
alhi
ghsc
hool
leve
lare
repo
rted
inpa
rent
hese
s.
83
Tabl
eA
.16:
Eff
ecto
fEle
men
tary
Scho
olTe
ache
rVal
ue-A
dded
onH
igh
Scho
olO
utco
mes
Usi
ngSc
hool
-Gra
deV
aria
tion
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
-0.0
15-0
.008
3.08
2-0
.001
-0.0
040.
000
0.03
3**
0.01
8-0
.004
-0.0
04-0
.002
(0.0
13)
(0.0
09)
(5.8
44)
(0.0
10)
(0.0
06)
(0.0
05)
(0.0
16)
(0.0
15)
(0.0
06)
(0.0
11)
(0.0
06)
Beh
avio
rVA
-0.0
390.
029
2.10
10.
043*
*0.
026*
*0.
021*
*0.
053
0.03
6-0
.009
-0.0
30-0
.013
(0.0
27)
(0.0
18)
(12.
527)
(0.0
20)
(0.0
12)
(0.0
10)
(0.0
34)
(0.0
31)
(0.0
12)
(0.0
23)
(0.0
12)
Lea
rnin
gSk
ills
VA0.
003
-0.0
22-4
.208
-0.0
49**
*-0
.020
**-0
.018
**-0
.047
*-0
.039
0.01
7*0.
024
0.00
8
(0.0
24)
(0.0
16)
(9.6
83)
(0.0
16)
(0.0
09)
(0.0
08)
(0.0
27)
(0.0
26)
(0.0
10)
(0.0
19)
(0.0
10)
Obs
erva
tions
3,10
63,
094
2,87
86,
250
5,17
55,
175
4,04
44,
043
6,29
06,
233
5,14
1
R-s
quar
ed0.
352
0.02
10.
001
0.00
60.
006
0.00
40.
021
0.01
00.
016
0.32
40.
037
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
scho
ol-g
rade
-yea
r.T
his
tabl
ere
port
sth
eef
fect
ofa
stan
dard
devi
atio
nin
crea
sein
the
thre
em
easu
res
ofel
emen
tary
scho
olte
ache
rva
lue-
adde
d(s
eeTa
ble
4)on
stud
ents
’hi
ghsc
hool
outc
omes
usin
gsc
hool
-gra
deva
riat
ion
inte
ache
rs’
valu
e-ad
ded.
Spec
ifica
lly,e
ach
colu
mn
ofth
eta
ble
repo
rts
the
coef
ficie
nts
onth
efir
stdi
ffer
ence
inth
esc
hool
-gra
dele
velm
ean
ofea
chof
the
thre
em
easu
res
ofte
ache
rval
ue-a
dded
from
ast
uden
twei
ghte
dO
LS
regr
essi
onof
the
first
diff
eren
cein
the
scho
ol-g
rade
leve
lmea
nof
the
outc
ome
vari
able
onth
efir
stdi
ffer
ence
inth
esc
hool
-gra
dele
velm
ean
ofea
chof
the
thre
em
easu
res
ofte
ache
rva
lue-
adde
dfo
rgr
ades
3-5
,alo
ngw
ithye
arfix
edef
fect
s.St
anda
rder
rors
clus
tere
dat
the
mod
alhi
ghsc
hool
leve
lare
repo
rted
inpa
rent
hese
s.**
*p
<0.
01, *
*p
<0.
05,a
nd*
p<
0.10
.
84
Tabl
eA
.17:
Eff
ecto
fEle
men
tary
Scho
olTe
ache
rVal
ue-A
dded
onH
igh
Scho
olO
utco
mes
-Con
trol
ling
forF
utur
eTr
acki
ng
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
-0.0
03*
-0.0
014.
924*
**-0
.001
0.00
10.
003*
**0.
020*
**0.
015*
**0.
001
0.00
40.
001
(0.0
02)
(0.0
02)
(1.0
52)
(0.0
02)
(0.0
01)
(0.0
01)
(0.0
02)
(0.0
02)
(0.0
01)
(0.0
03)
(0.0
01)
Beh
avio
rVA
-0.0
010.
008*
**1.
765
0.01
0***
0.00
5***
0.00
4*0.
010*
*0.
001
-0.0
02-0
.014
***
-0.0
05**
(0.0
02)
(0.0
03)
(1.2
08)
(0.0
03)
(0.0
02)
(0.0
02)
(0.0
04)
(0.0
03)
(0.0
02)
(0.0
04)
(0.0
02)
Lea
rnin
gSk
ills
VA-0
.000
-0.0
03-1
.498
-0.0
04-0
.003
-0.0
04-0
.008
*-0
.000
-0.0
020.
002
0.00
0
(0.0
02)
(0.0
02)
(1.4
86)
(0.0
04)
(0.0
02)
(0.0
03)
(0.0
04)
(0.0
03)
(0.0
02)
(0.0
05)
(0.0
02)
Obs
erva
tions
135,
786
102,
517
60,6
9429
3,02
123
3,07
823
3,07
815
2,34
515
1,82
031
6,11
627
7,33
122
1,75
7
R-s
quar
ed0.
401
0.29
80.
708
0.39
20.
418
0.40
60.
609
0.57
80.
051
0.30
90.
201
Not
e:T
his
tabl
ere
port
sth
eef
fect
ofa
stan
dard
devi
atio
nin
crea
sein
the
thre
em
easu
res
ofel
emen
tary
scho
olte
ache
rva
lue-
adde
d(s
eeTa
ble
4)on
stud
ents
’hi
ghsc
hool
outc
omes
.Sp
ecifi
cally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
norm
aliz
edm
easu
res
ofte
ache
rval
ue-a
dded
from
anO
LS
regr
essi
onof
the
stud
ents
’hig
hsc
hool
outc
ome
onth
eth
ree
mea
sure
sof
teac
her
valu
e-ad
ded
fort
hest
uden
ts’t
each
ers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.T
heba
selin
eco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fthe
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ners
tatu
san
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,
and
held
back
for
the
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ner
stat
usis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.A
dditi
onal
cont
rols
incl
ude
fixed
effe
cts
for
the
num
ber
ofco
urse
sea
chst
uden
ttak
esin
high
scho
olin
four
mat
htr
acks
,num
ber
ofm
ath
AP
clas
ses,
num
ber
ofE
nglis
hA
Pcl
asse
s,nu
mbe
rof
Eng
lish
elec
tives
and
num
bero
f“he
lper
”E
nglis
hC
lass
es.S
tand
ard
erro
rscl
uste
red
atth
em
odal
high
scho
olle
vela
rere
port
edin
pare
nthe
ses.
***
p<
0.01
, **
p<
0.05
,and
*p
<0.
10.
85
Tabl
eA
.18:
Eff
ecto
fEle
men
tary
Scho
olTe
ache
rVal
ue-A
dded
Mea
sure
son
Add
ition
alH
igh
Scho
olO
utco
mes
Pool
edG
rade
s3-
5VA
LA
USD
Gra
duat
eTo
okPS
AT
PSA
TSc
ore
Gra
de11
Mat
hC
ST
Gra
de11
Eng
lish
CST
Gra
de8
Scie
nce
CST
Gra
de10
Scie
nce
CST
Gra
de8
Soci
al
Scie
nce
CST
Gra
de11
Soci
al
Scie
nce
CST
Wor
ld
His
tory
CST
Num
bero
f
AP
Cla
sses
Test
Scor
eVA
0.00
00.
002
4.97
6***
0.01
6***
0.01
5***
0.01
7***
0.01
7***
0.01
8***
0.01
3***
0.01
4***
0.00
7*
(0.0
01)
(0.0
01)
(0.6
42)
(0.0
05)
(0.0
03)
(0.0
03)
(0.0
03)
(0.0
03)
(0.0
04)
(0.0
03)
(0.0
04)
Beh
avio
rVA
0.00
1-0
.002
1.53
00.
005
0.00
30.
008*
0.00
50.
000
0.00
60.
006
0.01
4
(0.0
02)
(0.0
03)
(1.3
12)
(0.0
10)
(0.0
06)
(0.0
04)
(0.0
06)
(0.0
04)
(0.0
06)
(0.0
05)
(0.0
09)
Lea
rnin
gSk
ills
VA-0
.001
0.00
0-0
.329
-0.0
08-0
.006
0.00
1-0
.008
0.00
3-0
.013
**-0
.006
-0.0
06
(0.0
02)
(0.0
03)
(1.0
16)
(0.0
10)
(0.0
05)
(0.0
05)
(0.0
05)
(0.0
05)
(0.0
05)
(0.0
04)
(0.0
08)
Obs
erva
tions
64,0
3922
8,10
318
4,89
448
,330
53,4
4833
1,19
113
6,92
227
3,64
154
,880
113,
571
316,
116
R-s
quar
ed0.
048
0.05
30.
591
0.36
70.
479
0.47
60.
420
0.48
30.
357
0.36
60.
234
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
eth
ree
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
ed(s
eeTa
ble
4)on
stud
ents
’hig
hsc
hool
outc
omes
.Spe
cific
ally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
norm
aliz
edm
easu
res
ofte
ache
rva
lue-
adde
dfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’
high
scho
olou
tcom
eon
the
thre
em
easu
res
ofte
ache
rva
lue-
adde
dfo
rth
est
uden
ts’
teac
hers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.T
heba
selin
eco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fthe
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ners
tatu
san
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,a
ndhe
ldba
ckfo
rthe
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ners
tatu
sis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.Sta
ndar
der
rors
clus
tere
dat
the
mod
alhi
ghsc
hool
leve
lare
repo
rted
inpa
rent
hese
s.**
*p
<0.
01, *
*p
<0.
05,a
nd*
p<
0.10
.
86
Tabl
eA
.19:
Eff
ecto
fEle
men
tary
Scho
olTe
ache
rVal
ue-A
dded
Est
imat
edw
ithTe
ache
rFix
edE
ffec
tson
Hig
hSc
hool
Out
com
es
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
-0.0
03-0
.002
9.27
3***
0.00
6**
0.00
5***
0.00
8***
0.03
0***
0.02
3***
0.00
1-0
.001
0.00
1
(0.0
02)
(0.0
03)
(1.5
04)
(0.0
03)
(0.0
02)
(0.0
01)
(0.0
03)
(0.0
03)
(0.0
02)
(0.0
04)
(0.0
02)
Beh
avio
rVA
-0.0
05**
0.01
0***
2.26
90.
018*
**0.
010*
**0.
007*
**0.
016*
**0.
005
-0.0
03-0
.020
***
-0.0
06**
(0.0
02)
(0.0
04)
(2.1
26)
(0.0
04)
(0.0
03)
(0.0
02)
(0.0
05)
(0.0
05)
(0.0
02)
(0.0
05)
(0.0
03)
Lea
rnin
gSk
ills
VA-0
.001
-0.0
00-0
.395
-0.0
02-0
.004
-0.0
04-0
.006
0.00
1-0
.003
0.00
40.
002
(0.0
02)
(0.0
03)
(1.9
48)
(0.0
06)
(0.0
03)
(0.0
03)
(0.0
05)
(0.0
04)
(0.0
02)
(0.0
07)
(0.0
03)
Obs
erva
tions
135,
786
102,
517
60,6
9429
3,02
123
3,07
823
3,07
815
2,34
515
1,82
031
6,11
627
7,33
122
1,75
7
R-s
quar
ed0.
293
0.14
50.
617
0.24
40.
234
0.23
90.
501
0.51
20.
040
0.26
70.
108
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
eth
ree
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
ed(s
eeTa
ble
4)on
stud
ents
’hig
hsc
hool
outc
omes
.The
valu
e-ad
ded
estim
ates
used
inth
eth
ree
indi
ces
are
estim
ated
usin
gC
hetty
,Fri
edm
an,
and
Roc
koff
’s(2
014a
,201
4b)a
ppro
ach,
whi
ches
timat
esth
ere
latio
nshi
pbe
twee
nco
ntro
lsan
dst
uden
tach
ieve
men
tusi
ngw
ithin
-tea
cher
vari
atio
n.Sp
ecifi
cally
,equ
atio
n(1
)is
mod
ified
toin
clud
ea
teac
her
fixed
effe
ct,b
uteq
uatio
n(3
)is
unch
ange
d.E
ach
colu
mn
ofth
eta
ble
repo
rts
the
coef
ficie
nts
onea
chof
the
thre
eno
rmal
ized
mea
sure
sof
teac
herv
alue
-add
edfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’h
igh
scho
olou
tcom
eon
the
thre
em
easu
res
ofte
ache
rval
ue-a
dded
fort
hest
uden
ts’t
each
ers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.T
heba
selin
eco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fthe
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ner
stat
usan
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,a
ndhe
ldba
ckfo
rthe
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ners
tatu
sis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.Sta
ndar
der
rors
clus
tere
dat
the
mod
alhi
ghsc
hool
leve
lare
repo
rted
inpa
rent
hese
s.**
*p
<0.
01, *
*p
<0.
05,a
nd*
p<
0.10
.
87
Tabl
eA
.20:
Eff
ecto
fEle
men
tary
Scho
olTe
ache
rVal
ue-A
dded
onH
igh
Scho
olO
utco
mes
Est
imat
edU
sing
With
in-T
each
erV
aria
tion
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
-0.0
17**
*0.
002
27.7
88**
*0.
026*
**0.
019*
**0.
024*
**0.
052*
**0.
050*
**-0
.002
-0.0
14**
*-0
.001
(0.0
04)
(0.0
03)
(4.0
85)
(0.0
04)
(0.0
03)
(0.0
04)
(0.0
07)
(0.0
05)
(0.0
02)
(0.0
05)
(0.0
02)
Beh
avio
rVA
-0.0
010.
013*
**3.
267
0.02
2***
0.01
2***
0.00
9**
0.02
6***
0.00
9-0
.005
**-0
.017
***
-0.0
07**
(0.0
04)
(0.0
04)
(3.3
82)
(0.0
05)
(0.0
04)
(0.0
04)
(0.0
08)
(0.0
06)
(0.0
02)
(0.0
06)
(0.0
03)
Lea
rnin
gSk
ills
VA-0
.012
**0.
000
7.42
9*0.
005
0.00
0-0
.001
-0.0
000.
012*
-0.0
04-0
.002
0.00
1
(0.0
05)
(0.0
04)
(4.2
48)
(0.0
07)
(0.0
05)
(0.0
05)
(0.0
07)
(0.0
07)
(0.0
02)
(0.0
09)
(0.0
03)
Obs
erva
tions
135,
786
102,
517
60,6
9429
3,02
123
3,07
823
3,07
815
2,34
515
1,82
031
6,11
627
7,33
122
1,75
7
R-s
quar
ed0.
003
0.00
10.
029
0.00
20.
003
0.00
50.
009
0.00
80.
000
0.00
10.
000
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
eth
ree
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
ed(s
eeTa
ble
4)on
stud
ents
’hig
hsc
hool
outc
omes
.The
valu
e-ad
ded
estim
ates
used
inth
eth
ree
indi
ces
are
estim
ated
usin
gth
eC
hetty
,Fri
edm
an,
and
Roc
koff
(201
4a,2
014b
)app
roac
h,w
hich
estim
ates
the
rela
tions
hip
betw
een
cont
rols
and
stud
enta
chie
vem
entu
sing
with
in-t
each
erva
riat
ion.
Spec
ifica
lly,e
quat
ion
(1)i
sm
odifi
edto
incl
ude
ate
ache
rfix
edef
fect
,but
equa
tion
(3)i
sun
chan
ged.
The
outc
ome
vari
able
sar
ere
sidu
aliz
edus
ing
the
sam
eap
proa
ch.E
ach
colu
mn
ofth
eta
ble
repo
rts
the
coef
ficie
nts
onea
chof
the
thre
eno
rmal
ized
mea
sure
sof
teac
herv
alue
-add
edfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’r
esid
ualiz
edhi
ghsc
hool
outc
ome
onth
eth
ree
mea
sure
sof
teac
herv
alue
-add
edfo
rthe
stud
ents
’tea
cher
sin
grad
es3,
4,or
5(a
ndno
cont
rols
).St
anda
rder
rors
clus
tere
dat
the
mod
alhi
ghsc
hool
leve
lare
repo
rted
inpa
rent
hese
s.**
*p
<0.
01, *
*p
<0.
05,a
nd*
p<
0.10
.
88
Tabl
eA
.21:
Lon
g-Te
rmE
ffec
toft
heTe
ache
rVal
ue-A
dded
Mea
sure
sE
stim
ated
Usi
nga
Fact
orM
odel
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
-0.0
02-0
.002
6.18
7***
0.00
20.
003*
0.00
5***
0.02
2***
0.01
6***
0.00
10.
001
0.00
1
(0.0
02)
(0.0
02)
(1.2
74)
(0.0
02)
(0.0
01)
(0.0
01)
(0.0
03)
(0.0
03)
(0.0
01)
(0.0
03)
(0.0
01)
Beh
avio
rVA
-0.0
040.
013*
**2.
336
0.01
3**
0.00
7**
0.00
40.
015*
*0.
006
-0.0
03-0
.021
***
-0.0
05*
(0.0
03)
(0.0
04)
(2.7
34)
(0.0
05)
(0.0
03)
(0.0
03)
(0.0
07)
(0.0
06)
(0.0
02)
(0.0
06)
(0.0
03)
Lea
rnin
gSk
ills
VA0.
000
-0.0
04-0
.790
-0.0
03-0
.003
-0.0
03-0
.008
-0.0
01-0
.001
0.00
50.
001
(0.0
03)
(0.0
03)
(1.8
93)
(0.0
05)
(0.0
03)
(0.0
03)
(0.0
05)
(0.0
04)
(0.0
02)
(0.0
06)
(0.0
03)
Obs
erva
tions
135,
786
102,
517
60,6
9429
3,02
123
3,07
823
3,07
815
2,34
515
1,82
031
6,11
627
7,33
122
1,75
7
R-s
quar
ed0.
293
0.14
50.
617
0.24
40.
234
0.23
90.
500
0.51
20.
040
0.26
70.
108
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
eth
ree
elem
enta
rysc
hool
teac
her
valu
e-ad
ded
mea
sure
s(c
reat
edus
ing
afa
ctor
mod
el)
onst
uden
ts’
high
scho
olou
tcom
es.
Eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
norm
aliz
edm
easu
res
ofte
ache
rval
ue-a
dded
from
anO
LS
regr
essi
onof
the
stud
ents
’hig
hsc
hool
outc
ome
onth
eth
ree
mea
sure
sof
teac
herv
alue
-add
edfo
rthe
stud
ents
’tea
cher
sin
grad
es3,
4,or
5,al
ong
with
the
base
line
cont
rols
desc
ribe
din
sect
ion
4.1.
The
base
line
cont
rols
incl
ude
lags
ofa
cubi
cpo
lyno
mia
loft
hest
uden
t’sm
ath
test
scor
e,E
nglis
hte
stsc
ore,
GPA
,and
effo
rtG
PA;a
cubi
cpo
lyno
mia
lsof
lagg
edcl
ass
and
grad
ele
velm
eans
ofea
chof
thos
eva
riab
les;
and
curr
entE
nglis
hle
arne
rsta
tus
and
lagg
edlo
gda
ysab
sent
,sus
pens
ions
,and
held
back
fort
hein
divi
dual
,the
clas
sav
erag
e,an
dth
egr
ade
aver
age.
Eac
hof
thes
eva
riab
les
exce
ptE
nglis
hle
arne
rsta
tus
isfu
llyin
tera
cted
with
grad
efix
edef
fect
san
da
cont
rolf
orcl
ass
size
isin
clud
ed.S
tand
ard
erro
rscl
uste
red
atth
em
odal
high
scho
olle
vela
rere
port
edin
pare
nthe
ses.
The
fact
orlo
ads
are
repo
rted
inTa
ble
A.2
2.**
*p
<0.
01, *
*p
<0.
05,a
nd*
p<
0.10
.
89
Table A.22: Factor Loads for Elementary School Value-Added
Grades 3-5 VA Test Score BehaviorLearning
Skills
Math Test Scores 0.804
English Test Scores 0.804
GPA 0.253
Log Absences -0.271
Suspended -0.160
Held Back -0.049
Effort GPA 0.798
Work and Study GPA 0.917
Learning and Social GPA 0.886
Note: This table reports the factor loads for the test-score, behavior, and learning-skills teacher valued-added measures in elementary school (grades3-5). Each of the three measures of teacher value-added is created using factor analysis. The test-score value-added is computed using the teachers’value-added for math and English test scores, and the factor loads are shown in column 1. The behavior value-added is computed using teachers’value-added for GPA, suspensions, log days absent, and not progressing to the next grade on time (held back), and the factor loads are shown incolumn 2. The learning-skills value-added is computed using teachers’ value-added for effort GPA, work and study habit GPA, and learning andsocial skills GPA, and the factor loads are shown in column 3.
90
Tabl
eA
.23:
Eff
ecto
fEle
men
tary
Scho
olTe
ache
rVal
ue-A
dded
onH
igh
Scho
olO
utco
mes
Usi
ngE
xplo
rato
ryFa
ctor
Ana
lysi
s
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Fact
or1
(Loa
dsPr
imar
ily-0
.002
-0.0
016.
507*
**0.
004
0.00
3**
0.00
5***
0.02
3***
0.01
7***
0.00
1-0
.001
0.00
0
onTe
stSc
ores
)(0
.002
)(0
.002
)(1
.246
)(0
.002
)(0
.001
)(0
.001
)(0
.003
)(0
.002
)(0
.001
)(0
.003
)(0
.001
)
Fact
or2
(Loa
dsPr
imar
ily-0
.002
0.00
4*1.
394
0.00
40.
001
-0.0
000.
003
0.00
5-0
.003
*-0
.006
-0.0
02
onN
on-t
estS
core
)(0
.002
)(0
.002
)(2
.039
)(0
.005
)(0
.003
)(0
.003
)(0
.005
)(0
.004
)(0
.002
)(0
.006
)(0
.002
)
Obs
erva
tions
135,
786
102,
517
60,6
9429
3,02
123
3,07
823
3,07
815
2,34
515
1,82
031
6,11
627
7,33
122
1,75
7
R-s
quar
ed0.
293
0.14
50.
617
0.24
40.
234
0.23
90.
500
0.51
20.
040
0.26
60.
108
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
etw
oal
tern
ativ
em
easu
res
ofel
emen
tary
scho
olte
ache
rval
ue-a
dded
onst
uden
ts’h
igh
scho
olou
tcom
es.T
hetw
ofa
ctor
sar
eco
mpu
ted
usin
ga
mod
elw
ithal
lofo
urm
easu
res
ofte
ache
rval
ue-a
dded
incl
uded
sim
ulta
neou
sly
rath
erth
anby
impo
sing
that
only
cert
ain
valu
e-ad
ded
mea
sure
sco
ntri
bute
toce
rtai
nfa
ctor
s.T
heva
lue-
adde
dm
easu
res
used
toco
mpu
teth
efa
ctor
sin
clud
em
ath
and
Eng
lish
test
scor
es,s
uspe
nsio
ns,l
ogda
ysab
sent
,not
prog
ress
ing
toth
ene
xtgr
ade
ontim
e(h
eld
back
),G
PA,e
ffor
tGPA
,wor
kan
dst
udy
habi
tsG
PA,a
ndle
arni
ngan
dso
cial
skill
sG
PA.T
hese
cond
fact
orpr
imar
ilylo
ads
onth
etw
ote
st-s
core
valu
e-ad
ded
mea
sure
san
dto
ale
sser
exte
ntG
PA,w
here
asth
efir
stfa
ctor
load
spr
imar
ilyon
GPA
and
the
othe
rno
n-te
st-s
core
vari
able
s.E
ach
colu
mn
ofth
eta
ble
repo
rts
the
coef
ficie
nts
onea
chof
the
two
alte
rnat
ive
mea
sure
sof
teac
her
valu
e-ad
ded
from
anO
LS
regr
essi
onof
the
stud
ents
’hi
ghsc
hool
outc
ome
onth
etw
oal
tern
ativ
em
easu
res
ofte
ache
rva
lue-
adde
dfo
rth
est
uden
ts’t
each
ers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.T
heba
selin
eco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fthe
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ner
stat
usan
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,a
ndhe
ldba
ckfo
rth
ein
divi
dual
,the
clas
sav
erag
e,an
dth
egr
ade
aver
age.
Eac
hof
thes
eva
riab
les
exce
ptE
nglis
hle
arne
rst
atus
isfu
llyin
tera
cted
with
grad
efix
edef
fect
san
da
cont
rolf
orcl
ass
size
isin
clud
ed.
Stan
dard
erro
rscl
uste
red
atth
em
odal
high
scho
olle
vela
rere
port
edin
pare
nthe
ses.
***
p<
0.01
, **
p<
0.05
,and
*p
<0.
10.
91
Table A.24: Factor Loads for the Exploratory Factor Analysis
Grades 3-5 VA Factor 1 Factor 2
Math Test Scores 0.102 0.807
English Test Scores 0.118 0.808
GPA 0.781 0.126
Log Absences 0.121 0.044
Suspended 0.065 0.029
Held Back 0.008 0.024
Effort GPA 0.845 0.068
Work and Study GPA 0.910 0.088
Learning and Social GPA 0.866 0.050
Note: This table reports the factor loads for the two factors used in Table A.23. The two factors of teacher value-added measures are created usingfactor analysis. All nine of the listed value-added measures are variables included in the factor analysis. Factors with an eigenvalue grater than oneare included.
92
Tabl
eA
.25:
Lon
g-Te
rmE
ffec
toft
heTe
ache
rVal
ue-A
dded
Mea
sure
sE
stim
ated
inPe
rcen
tiles
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
e
Perc
entil
e
GPA
Perc
entil
e
Eff
ortG
PA
Perc
entil
e
Coo
pera
tion
GPA
Perc
entil
e
Mat
h
CA
HSE
E
Perc
entil
e
Eng
lish
CA
HSE
E
Perc
entil
e
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
Dec
ile-0
.012
6-0
.002
20.
0018
-0.0
003
0.00
00-0
.000
00.
0199
***
0.00
97**
*-0
.020
7-0
.037
30.
0064
(0.0
178)
(0.0
183)
(0.0
012)
(0.0
003)
(0.0
002)
(0.0
002)
(0.0
039)
(0.0
021)
(0.0
168)
(0.0
312)
(0.0
138)
Beh
avio
rVA
Dec
ile-0
.030
50.
1065
***
0.00
340.
0003
0.00
020.
0004
0.01
78*
0.00
57-0
.023
9-0
.137
0*-0
.111
8***
(0.0
418)
(0.0
389)
(0.0
025)
(0.0
009)
(0.0
014)
(0.0
014)
(0.0
101)
(0.0
057)
(0.0
333)
(0.0
690)
(0.0
353)
Lea
rnin
gSk
ills
VAD
ecile
-0.0
079
0.00
74-0
.000
1-0
.000
10.
0032
0.00
310.
0036
-0.0
001
-0.0
027
-0.0
187*
-0.0
119
(0.0
179)
(0.0
110)
(0.0
013)
(0.0
001)
(0.0
023)
(0.0
022)
(0.0
022)
(0.0
010)
(0.0
039)
(0.0
107)
(0.0
087)
Obs
erva
tions
135,
786
102,
517
60,6
9429
3,02
123
3,07
823
3,07
815
2,34
515
1,82
031
6,11
627
7,33
122
1,75
7
R-s
quar
ed0.
293
0.14
50.
225
0.17
80.
978
0.97
80.
158
0.14
20.
039
0.26
70.
108
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.T
his
tabl
esh
ows
the
effe
ctof
ade
cile
incr
ease
inth
ree
elem
enta
rysc
hool
teac
her
valu
e-ad
ded
indi
ces
onhi
ghsc
hool
outc
omes
whe
rete
ache
rva
lue-
adde
d,G
PA,a
ndte
stsc
ores
are
mea
sure
din
ordi
nal
units
,dec
iles
inth
eca
seof
teac
her
valu
e-ad
ded
and
perc
entil
esin
the
case
ofte
stsc
ores
and
GPA
.Eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
deci
les
incr
ease
inte
ache
rval
ue-a
dded
from
anO
LS
regr
essi
onof
the
stud
ents
’hig
hsc
hool
outc
ome
onth
eth
ree
mea
sure
sof
teac
herv
alue
-add
edfo
rthe
stud
ents
’tea
cher
sin
grad
es3,
4,or
5,al
ong
with
the
base
line
cont
rols
desc
ribe
din
sect
ion
4.1.
The
base
line
cont
rols
incl
ude
lags
ofa
cubi
cpo
lyno
mia
lof
the
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ners
tatu
san
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,a
ndhe
ldba
ckfo
rthe
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ners
tatu
sis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.Sta
ndar
der
rors
clus
tere
dat
the
mod
alhi
ghsc
hool
leve
lare
repo
rted
inpa
rent
hese
s.T
hefa
ctor
load
sar
ere
port
edin
Tabl
eA
.22.
***
p<
0.01
, **
p<
0.05
,and
*p
<0.
10.
93
Tabl
eA
.26:
Eff
ecto
fthe
Ele
men
tary
Scho
olTe
ache
rVal
ue-A
dded
Mea
sure
son
Hig
hSc
hool
Out
com
esw
ithou
tBeh
avio
rVal
ue-A
dded
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
-0.0
02-0
.001
6.36
4***
0.00
30.
003*
*0.
005*
**0.
023*
**0.
017*
**0.
001
-0.0
000.
001
(0.0
02)
(0.0
02)
(1.2
51)
(0.0
02)
(0.0
01)
(0.0
01)
(0.0
03)
(0.0
03)
(0.0
01)
(0.0
03)
(0.0
01)
Lea
rnin
gSk
ills
VA-0
.002
0.00
30.
391
0.00
40.
000
-0.0
010.
000
0.00
2-0
.003
*-0
.005
-0.0
02
(0.0
02)
(0.0
02)
(1.9
66)
(0.0
05)
(0.0
03)
(0.0
03)
(0.0
05)
(0.0
04)
(0.0
02)
(0.0
06)
(0.0
02)
Obs
erva
tions
136,
236
102,
901
60,9
1529
3,77
023
3,69
623
3,69
615
2,80
315
2,27
231
6,96
627
8,05
122
2,33
5
R-s
quar
ed0.
293
0.14
50.
617
0.24
40.
234
0.23
90.
500
0.51
20.
040
0.26
60.
108
Not
e:T
his
tabl
ere
port
sth
eef
fect
ofa
stan
dard
devi
atio
nin
crea
sein
test
-sco
rean
dle
arni
ng-s
kills
elem
enta
rysc
hool
teac
her
valu
e-ad
ded
(see
Tabl
e4)
onst
uden
ts’
high
scho
olou
tcom
es.
Spec
ifica
lly,
each
colu
mn
ofth
eta
ble
repo
rts
the
coef
ficie
nts
onth
ete
st-s
core
and
lear
ning
-ski
llsno
rmal
ized
mea
sure
sof
teac
herv
alue
-add
edfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’h
igh
scho
olou
tcom
eon
the
two
mea
sure
sof
teac
her
valu
e-ad
ded
for
the
stud
ents
’te
ache
rsin
grad
es3,
4,or
5,al
ong
with
the
base
line
cont
rols
desc
ribe
din
sect
ion
4.1.
The
base
line
cont
rols
incl
ude
lags
ofa
cubi
cpo
lyno
mia
lof
the
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ners
tatu
san
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,a
ndhe
ldba
ckfo
rth
ein
divi
dual
,the
clas
sav
erag
e,an
dth
egr
ade
aver
age.
Eac
hof
thes
eva
riab
les
exce
ptE
nglis
hle
arne
rst
atus
isfu
llyin
tera
cted
with
grad
efix
edef
fect
san
da
cont
rolf
orcl
ass
size
isin
clud
ed.S
tand
ard
erro
rscl
uste
red
atth
em
odal
high
scho
olle
vela
rere
port
edin
pare
nthe
ses.
***
p<
0.01
, **
p<
0.05
,and
*p
<0.
10.
94
Tabl
eA
.27:
Eff
ect
ofE
lem
enta
rySc
hool
Teac
her
Val
ue-A
dded
onH
igh
Scho
olO
utco
mes
-E
xclu
ding
Susp
ende
dan
dH
eld
Bac
kE
lem
enta
rySc
hool
Stud
ents
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
-0.0
01-0
.002
6.20
6***
0.00
10.
002
0.00
4***
0.02
2***
0.01
6***
0.00
10.
001
0.00
1
(0.0
02)
(0.0
02)
(1.2
97)
(0.0
02)
(0.0
01)
(0.0
01)
(0.0
03)
(0.0
03)
(0.0
01)
(0.0
03)
(0.0
01)
Beh
avio
rVA
-0.0
030.
010*
**1.
984
0.01
3***
0.00
7***
0.00
5*0.
013*
*0.
004
-0.0
03-0
.016
***
-0.0
06**
(0.0
02)
(0.0
03)
(2.1
55)
(0.0
04)
(0.0
03)
(0.0
02)
(0.0
06)
(0.0
05)
(0.0
02)
(0.0
05)
(0.0
03)
Lea
rnin
gSk
ills
VA-0
.000
-0.0
01-0
.483
-0.0
02-0
.003
-0.0
03-0
.005
0.00
1-0
.002
0.00
00.
001
(0.0
02)
(0.0
03)
(1.8
20)
(0.0
05)
(0.0
03)
(0.0
03)
(0.0
05)
(0.0
04)
(0.0
02)
(0.0
06)
(0.0
03)
Obs
erva
tions
129,
193
98,0
9958
,529
278,
453
221,
690
221,
690
145,
703
145,
228
299,
648
263,
003
211,
435
R-s
quar
ed0.
300
0.14
40.
617
0.24
10.
230
0.23
40.
500
0.51
20.
035
0.26
90.
107
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
eth
ree
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
ed(s
eeTa
ble
4)on
stud
ents
’hig
hsc
hool
outc
omes
.Spe
cific
ally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
norm
aliz
edm
easu
res
ofte
ache
rva
lue-
adde
dfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’
high
scho
olou
tcom
eon
the
thre
em
easu
res
ofte
ache
rva
lue-
adde
dfo
rth
est
uden
ts’
teac
hers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SDan
dex
clud
eshe
ldba
ckan
dsu
spen
ded
elem
enta
rysc
hool
stud
ents
.T
heba
selin
eco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fth
est
uden
t’sm
ath
test
scor
e,E
nglis
hte
stsc
ore,
GPA
,and
effo
rtG
PA;a
cubi
cpo
lyno
mia
lsof
lagg
edcl
ass
and
grad
ele
velm
eans
ofea
chof
thos
eva
riab
les;
and
curr
entE
nglis
hle
arne
rst
atus
and
lagg
edlo
gda
ysab
sent
,sus
pens
ions
,and
held
back
for
the
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ner
stat
usis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.St
anda
rder
rors
clus
tere
dat
the
mod
alhi
ghsc
hool
leve
lare
repo
rted
inpa
rent
hese
s.**
*p
<0.
01, *
*p
<0.
05,a
nd*
p<
0.10
.
95
Tabl
eA
.28:
Eff
ecto
fEle
men
tary
Scho
olTe
ache
rVal
ue-A
dded
onH
igh
Scho
olO
utco
mes
-Tes
tSco
reVA
Est
imat
edus
ing
Nex
tYea
r’s
Test
Scor
e
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
(t+
1)-0
.005
***
0.00
111
.123
***
0.01
3***
0.00
9***
0.01
1***
0.03
8***
0.02
8***
-0.0
01-0
.003
-0.0
02
(0.0
02)
(0.0
03)
(1.9
06)
(0.0
04)
(0.0
03)
(0.0
03)
(0.0
05)
(0.0
04)
(0.0
02)
(0.0
05)
(0.0
02)
Beh
avio
rVA
-0.0
020.
010*
**0.
862
0.01
2***
0.00
6**
0.00
40.
009*
0.00
1-0
.002
-0.0
16**
*-0
.005
**
(0.0
02)
(0.0
03)
(1.9
43)
(0.0
04)
(0.0
02)
(0.0
02)
(0.0
05)
(0.0
04)
(0.0
02)
(0.0
05)
(0.0
03)
Lea
rnin
gSk
ills
VA0.
000
-0.0
02-1
.143
-0.0
03-0
.004
-0.0
04-0
.008
*-0
.001
-0.0
020.
002
0.00
1
(0.0
02)
(0.0
03)
(1.6
57)
(0.0
05)
(0.0
03)
(0.0
03)
(0.0
04)
(0.0
03)
(0.0
02)
(0.0
06)
(0.0
03)
Obs
erva
tions
131,
999
99,6
2259
,004
285,
915
227,
272
227,
272
148,
368
147,
863
308,
344
270,
602
216,
288
R-s
quar
ed0.
295
0.14
40.
617
0.24
30.
234
0.23
90.
500
0.51
20.
039
0.26
60.
108
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
eth
ree
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
ed(s
eeTa
ble
4)on
stud
ents
’hig
hsc
hool
outc
omes
.Spe
cific
ally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
norm
aliz
edm
easu
res
ofte
ache
rva
lue-
adde
dfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’
high
scho
olou
tcom
eon
the
thre
em
easu
res
ofte
ache
rva
lue-
adde
dfo
rth
est
uden
ts’
teac
hers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.Te
st-s
core
valu
e-ad
ded
ises
timat
edus
ing
test
scor
esin
year
t+1.
The
base
line
cont
rols
incl
ude
lags
ofa
cubi
cpo
lyno
mia
lof
the
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ners
tatu
san
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,
and
held
back
for
the
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ner
stat
usis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.Eac
hof
thes
eva
riab
les
exce
ptE
nglis
hle
arne
rsta
tus
isfu
llyin
tera
cted
with
grad
efix
edef
fect
san
da
cont
rolf
orcl
ass
size
isin
clud
ed.S
tand
ard
erro
rscl
uste
red
atth
em
odal
high
scho
olle
vela
rere
port
edin
pare
nthe
ses.
***
p<
0.01
, **
p<
0.05
,and
*p
<0.
10.
96
Tabl
eA
.29:
Eff
ecto
fEle
men
tary
Scho
olTe
ache
rVal
ue-A
dded
onH
igh
Scho
olO
utco
mes
-Con
tem
pora
neou
sVA
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
-0.0
02-0
.001
6.51
5***
0.00
30.
003*
0.00
4***
0.02
3***
0.01
7***
-0.0
00-0
.001
0.00
1
(0.0
02)
(0.0
02)
(1.0
70)
(0.0
02)
(0.0
01)
(0.0
01)
(0.0
03)
(0.0
02)
(0.0
01)
(0.0
02)
(0.0
01)
Beh
avio
rVA
(t)
-0.0
030.
003
0.29
20.
008*
*0.
004*
*0.
003*
0.00
50.
001
0.00
2-0
.011
***
-0.0
03
(0.0
02)
(0.0
03)
(1.4
63)
(0.0
03)
(0.0
02)
(0.0
02)
(0.0
04)
(0.0
03)
(0.0
02)
(0.0
04)
(0.0
02)
Lea
rnin
gSk
ills
VA(t)
0.00
1-0
.000
-2.0
25-0
.006
*-0
.005
***
-0.0
04**
-0.0
07*
-0.0
06**
-0.0
010.
003
0.00
2*
(0.0
02)
(0.0
02)
(1.3
07)
(0.0
03)
(0.0
02)
(0.0
02)
(0.0
04)
(0.0
03)
(0.0
02)
(0.0
03)
(0.0
01)
Obs
erva
tions
179,
761
136,
115
81,6
8436
6,71
929
6,22
329
6,22
319
7,75
919
7,07
639
6,27
334
7,59
028
1,58
8
R-s
quar
ed0.
293
0.15
20.
631
0.25
50.
245
0.24
90.
512
0.52
60.
040
0.26
50.
112
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
eth
ree
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
ed(s
eeTa
ble
4)on
stud
ents
’hig
hsc
hool
outc
omes
.Spe
cific
ally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
norm
aliz
edm
easu
res
ofte
ache
rva
lue-
adde
dfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’
high
scho
olou
tcom
eon
the
thre
em
easu
res
ofte
ache
rva
lue-
adde
dfo
rth
est
uden
ts’
teac
hers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.B
ehav
iora
ndle
arni
ngsk
ills
valu
e-ad
ded
wer
eco
nstr
ucte
dus
ing
year
tra
ther
than
year
t+1
outc
omes
.The
base
line
cont
rols
incl
ude
lags
ofa
cubi
cpo
lyno
mia
lof
the
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ners
tatu
san
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,a
ndhe
ldba
ckfo
rthe
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ners
tatu
sis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.Sta
ndar
der
rors
clus
tere
dat
the
mod
alhi
ghsc
hool
leve
lare
repo
rted
inpa
rent
hese
s.**
*p
<0.
01, *
*p
<0.
05,a
nd*
p<
0.10
.
97
Tabl
eA
.30:
Eff
ecto
fEle
men
tary
Scho
olTe
ache
rVal
ue-A
dded
onH
igh
Scho
olO
utco
mes
-Z-S
core
GPA
VAan
dO
utco
mes
Gra
des
3-5
VAG
PAz-
scor
e
GPA
Eff
ortG
PAz-
scor
e
Eff
ortG
PA
Coo
pera
tion
GPA
z-sc
ore
Coo
pera
tion
GPA
Test
Scor
eVA
0.00
20.
003
0.00
3*0.
005*
0.00
5***
0.01
0***
(0.0
02)
(0.0
02)
(0.0
01)
(0.0
03)
(0.0
01)
(0.0
03)
Beh
avio
rVA
(Z-G
PA)
0.01
3***
0.01
4***
0.00
7***
0.01
4***
0.00
5**
0.01
1**
(0.0
04)
(0.0
04)
(0.0
03)
(0.0
05)
(0.0
02)
(0.0
05)
Lea
rnin
gSk
ills
VA(Z
-GPA
)-0
.002
-0.0
02-0
.003
-0.0
06-0
.003
-0.0
08
(0.0
05)
(0.0
05)
(0.0
03)
(0.0
06)
(0.0
03)
(0.0
07)
Obs
erva
tions
293,
021
293,
021
233,
078
233,
078
233,
078
233,
078
R-s
quar
ed0.
244
0.24
40.
234
0.23
40.
239
0.23
9
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
eth
ree
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
ed(s
eeTa
ble
4)on
stud
ents
’hig
hsc
hool
outc
omes
.Spe
cific
ally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
norm
aliz
edm
easu
res
ofte
ache
rva
lue-
adde
dfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’
high
scho
olou
tcom
eon
the
thre
em
easu
res
ofte
ache
rva
lue-
adde
dfo
rth
est
uden
ts’
teac
hers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.T
heG
PAva
riab
les
used
toco
nstr
uctt
heva
lue-
adde
dm
easu
res
wer
eco
nver
ted
toz-
scor
esbe
fore
estim
atin
gte
ache
rval
ue-a
dded
.T
heba
selin
eco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fth
est
uden
t’sm
ath
test
scor
e,E
nglis
hte
stsc
ore,
GPA
,and
effo
rtG
PA;a
cubi
cpo
lyno
mia
lsof
lagg
edcl
ass
and
grad
ele
velm
eans
ofea
chof
thos
eva
riab
les;
and
curr
ent
Eng
lish
lear
ner
stat
usan
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,a
ndhe
ldba
ckfo
rth
ein
divi
dual
,the
clas
sav
erag
e,an
dth
egr
ade
aver
age.
Eac
hof
thes
eva
riab
les
exce
ptE
nglis
hle
arne
rst
atus
isfu
llyin
tera
cted
with
grad
efix
edef
fect
san
da
cont
rolf
orcl
ass
size
isin
clud
ed.S
tand
ard
erro
rscl
uste
red
atth
em
odal
high
scho
olle
vela
rere
port
edin
pare
nthe
ses.
***
p<
0.01
, **
p<
0.05
,and
*p
<0.
10.
98
Tabl
eA
.31:
The
Cum
ulat
ive
Eff
ecto
fGra
des
4-11
Teac
herV
alue
-Add
edM
easu
res
onH
igh
Scho
olO
utco
mes
Ave
rage
Gra
de
4-11
VA
LA
USD
Dro
pout
HS
Gra
duat
eTo
okSA
TSA
TSc
ore
12G
rade
GPA
12G
rade
Eff
ortG
PA
12G
rade
Coo
pera
tion
GPA
11G
rade
Mat
hTe
st
Scor
es
11G
rade
Eng
lish
Test
Scor
es
12G
rade
Day
s
Susp
ende
d
12G
rade
Log
Abs
ence
s
Hel
dB
ack
11G
rade
Avg
Test
Scor
eVA
-0.0
15*
0.01
3***
0.06
9***
57.4
71**
*0.
041*
0.03
9***
0.04
8***
0.48
9***
0.24
8***
0.00
0-0
.109
***
-0.0
09**
(0.0
07)
(0.0
05)
(0.0
16)
(5.6
28)
(0.0
23)
(0.0
11)
(0.0
11)
(0.0
29)
(0.0
21)
(0.0
02)
(0.0
34)
(0.0
04)
Avg
Beh
avio
rVA
-0.0
38**
*0.
012*
*0.
031*
**-3
0.03
3***
0.04
6**
0.04
6***
0.02
3*-0
.046
-0.0
41**
-0.0
02**
*-0
.202
***
-0.0
08*
(0.0
09)
(0.0
06)
(0.0
10)
(6.8
27)
(0.0
22)
(0.0
14)
(0.0
12)
(0.0
32)
(0.0
17)
(0.0
01)
(0.0
27)
(0.0
04)
Obs
erva
tions
36,3
7520
,318
30,8
8817
,377
28,4
4118
,001
18,0
0117
,001
18,3
7830
,888
28,3
0029
,572
R-s
quar
ed0.
537
0.02
70.
151
0.62
20.
117
0.11
20.
126
0.40
30.
479
0.00
30.
085
0.02
7
Not
e:T
his
tabl
ere
port
sth
eef
fect
ofa
stan
dard
devi
atio
nin
crea
sein
test
-sco
rean
dle
arni
ng-s
kills
elem
enta
rysc
hool
teac
her
valu
e-ad
ded
(see
Tabl
e4)
onst
uden
ts’
high
scho
olou
tcom
es.
Spec
ifica
lly,
each
colu
mn
ofth
eta
ble
repo
rts
the
coef
ficie
nts
onth
ete
st-s
core
and
lear
ning
-ski
llsno
rmal
ized
mea
sure
sof
teac
herv
alue
-add
edfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’h
igh
scho
olou
tcom
eon
the
two
mea
sure
sof
teac
herv
alue
-add
edfo
rave
rage
dfo
reac
hst
uden
ts’t
each
ers
from
grad
es4-
11,a
long
with
the
cont
rols
desc
ribe
din
sect
ion
4.1
mea
sure
din
grad
es3
and
4.T
heco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fthe
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ners
tatu
san
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,a
ndhe
ldba
ckfo
rthe
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.A
cont
rolf
orcl
ass
size
isin
clud
ed.S
tude
nts
are
incl
uded
ifw
eob
serv
eth
eva
lue-
adde
dsc
ores
ofth
eirt
each
ers
fora
tlea
st7
ofth
e8
year
sth
eysp
end
ingr
ades
4-11
.Sta
ndar
der
rors
clus
tere
dat
the
mod
alhi
ghsc
hool
leve
lare
repo
rted
inpa
rent
hese
s.**
*p
<0.
01, *
*p
<0.
05,a
nd*
p<
0.10
.
99
Tabl
eA
.32:
Rel
atio
nshi
pbe
twee
nE
lem
enta
rySu
bjec
tGPA
and
Hig
hSc
hool
Out
com
es
Pool
edG
rade
s3-
5G
PAL
AU
SD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Mat
hG
PA-0
.038
***
0.05
4***
56.9
02**
*0.
156*
**0.
075*
**0.
060*
**0.
337*
**0.
096*
**-0
.010
***
-0.1
05**
*-0
.046
***
(0.0
03)
(0.0
04)
(2.9
48)
(0.0
06)
(0.0
04)
(0.0
03)
(0.0
07)
(0.0
05)
(0.0
03)
(0.0
06)
(0.0
04)
Rea
ding
GPA
0.00
40.
011*
*53
.384
***
-0.0
17**
*-0
.015
***
-0.0
09**
*0.
019*
**0.
169*
**0.
012*
**-0
.010
0.01
0***
(0.0
03)
(0.0
05)
(2.6
35)
(0.0
05)
(0.0
04)
(0.0
03)
(0.0
07)
(0.0
07)
(0.0
03)
(0.0
07)
(0.0
03)
Wri
ting
GPA
-0.0
29**
*0.
057*
**-0
.449
0.15
1***
0.08
2***
0.05
9***
0.02
2***
0.10
7***
-0.0
19**
*-0
.016
**-0
.038
***
(0.0
04)
(0.0
05)
(2.1
74)
(0.0
06)
(0.0
03)
(0.0
03)
(0.0
06)
(0.0
06)
(0.0
04)
(0.0
07)
(0.0
03)
Lis
teni
ngG
PA-0
.057
***
0.06
4***
-11.
538*
**0.
210*
**0.
131*
**0.
134*
**0.
048*
**0.
025*
**-0
.070
***
-0.0
69**
*-0
.050
***
(0.0
03)
(0.0
04)
(2.2
65)
(0.0
06)
(0.0
03)
(0.0
03)
(0.0
06)
(0.0
05)
(0.0
05)
(0.0
07)
(0.0
04)
Spea
king
GPA
0.02
4***
-0.0
37**
*0.
697
-0.1
30**
*-0
.081
***
-0.0
88**
*-0
.046
***
-0.0
22**
*0.
040*
**0.
063*
**0.
026*
**
(0.0
03)
(0.0
05)
(2.4
76)
(0.0
08)
(0.0
05)
(0.0
04)
(0.0
07)
(0.0
06)
(0.0
04)
(0.0
07)
(0.0
04)
His
tory
/Soc
ialG
PA-0
.016
***
0.01
8***
-3.1
330.
070*
**0.
040*
**0.
031*
**-0
.019
**0.
001
-0.0
13**
*-0
.029
***
-0.0
22**
*
(0.0
04)
(0.0
05)
(2.2
94)
(0.0
07)
(0.0
04)
(0.0
04)
(0.0
07)
(0.0
06)
(0.0
04)
(0.0
07)
(0.0
04)
Scie
nce
GPA
-0.0
010.
007
21.4
26**
*0.
019*
**0.
008*
0.01
0***
0.02
0***
0.03
8***
-0.0
09**
-0.0
30**
*-0
.008
**
(0.0
04)
(0.0
05)
(2.6
53)
(0.0
06)
(0.0
04)
(0.0
04)
(0.0
07)
(0.0
05)
(0.0
04)
(0.0
05)
(0.0
03)
Hea
lthG
PA0.
008*
*0.
001
1.45
1-0
.001
0.00
30.
011*
**0.
006
0.00
3-0
.004
0.00
70.
004
(0.0
04)
(0.0
05)
(3.2
88)
(0.0
07)
(0.0
04)
(0.0
04)
(0.0
09)
(0.0
06)
(0.0
05)
(0.0
07)
(0.0
04)
PEG
PA0.
021*
**-0
.032
***
-30.
433*
**-0
.099
***
-0.0
65**
*-0
.071
***
-0.0
46**
*-0
.115
***
0.04
2***
-0.0
060.
024*
**
(0.0
03)
(0.0
04)
(2.8
03)
(0.0
06)
(0.0
03)
(0.0
03)
(0.0
06)
(0.0
05)
(0.0
05)
(0.0
07)
(0.0
05)
Art
sG
PA-0
.016
***
0.01
2*-0
.000
0.06
3***
0.03
7***
0.03
7***
-0.0
090.
046*
**-0
.032
***
0.00
4-0
.011
***
(0.0
04)
(0.0
07)
(2.7
27)
(0.0
06)
(0.0
04)
(0.0
04)
(0.0
08)
(0.0
06)
(0.0
04)
(0.0
08)
(0.0
04)
Obs
erva
tions
206,
598
156,
359
93,4
1740
0,53
232
8,03
532
8,03
522
2,91
622
2,15
843
3,59
638
0,61
231
1,58
5
R-s
quar
ed0.
288
0.16
70.
645
0.28
60.
276
0.28
20.
537
0.54
60.
043
0.26
50.
121
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
pred
ictiv
eef
fect
ofa
one-
poin
tinc
reas
ein
ast
uden
ts’G
PAin
each
ofth
e10
elem
enta
rysc
hool
subj
ects
onst
uden
ts’h
igh
scho
olou
tcom
es.S
peci
fical
ly,e
ach
colu
mn
ofth
eta
ble
repo
rts
the
coef
ficie
nts
onea
chof
the
stud
ents
’10
subj
ectG
PAs
from
anO
LS
regr
essi
onof
the
stud
ents
’hi
ghsc
hool
outc
ome
onst
uden
ts’
10su
bjec
tGPA
sin
grad
es3,
4or
5.
The
base
line
cont
rols
incl
ude
lags
ofa
cubi
cpo
lyno
mia
lof
the
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ners
tatu
san
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,
and
held
back
for
the
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ner
stat
usis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.Sta
ndar
der
rors
clus
tere
dat
the
mod
alhi
ghsc
hool
leve
lare
repo
rted
inpa
rent
hese
s.**
*p
<0.
01, *
*p
<0.
05,a
nd*
p<
0.10
.
100
Tabl
eA
.33:
Eff
ecto
fEle
men
tary
Scho
olTe
ache
rVal
ue-A
dded
onH
igh
Scho
olO
utco
mes
-Clu
ster
ing
atE
lem
enta
rySc
hool
Lev
el
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
-0.0
02-0
.002
6.23
7***
0.00
20.
003*
0.00
5***
0.02
2***
0.01
6***
0.00
10.
001
0.00
1
(0.0
02)
(0.0
02)
(1.1
52)
(0.0
02)
(0.0
01)
(0.0
01)
(0.0
03)
(0.0
03)
(0.0
01)
(0.0
03)
(0.0
01)
Beh
avio
rVA
-0.0
030.
010*
**1.
955
0.01
3***
0.00
7***
0.00
5**
0.01
3**
0.00
4-0
.003
*-0
.016
***
-0.0
06**
(0.0
02)
(0.0
03)
(2.1
48)
(0.0
04)
(0.0
02)
(0.0
02)
(0.0
05)
(0.0
04)
(0.0
02)
(0.0
05)
(0.0
02)
Lea
rnin
gSk
ills
VA-0
.000
-0.0
01-0
.547
-0.0
02-0
.003
-0.0
03-0
.005
0.00
1-0
.002
0.00
20.
001
(0.0
02)
(0.0
03)
(1.4
85)
(0.0
04)
(0.0
02)
(0.0
02)
(0.0
04)
(0.0
04)
(0.0
02)
(0.0
05)
(0.0
02)
Obs
erva
tions
135,
786
102,
517
60,6
9429
3,02
823
3,07
823
3,07
815
2,34
515
1,82
031
6,12
327
7,33
322
1,75
7
R-s
quar
ed0.
293
0.14
50.
617
0.24
40.
234
0.23
90.
500
0.51
20.
040
0.26
70.
108
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
eth
ree
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
ed(s
eeTa
ble
4)on
stud
ents
’hig
hsc
hool
outc
omes
.Spe
cific
ally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
norm
aliz
edm
easu
res
ofte
ache
rva
lue-
adde
dfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’
high
scho
olou
tcom
eon
the
thre
em
easu
res
ofte
ache
rva
lue-
adde
dfo
rth
est
uden
ts’
teac
hers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.T
heba
selin
eco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fthe
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ners
tatu
san
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,a
ndhe
ldba
ckfo
rthe
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ners
tatu
sis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.Sta
ndar
der
rors
clus
tere
dat
the
elem
enta
rysc
hool
leve
lare
repo
rted
inpa
rent
hese
s.**
*p
<0.
01, *
*p
<0.
05,a
nd*
p<
0.10
.
101
Tabl
eA
.34:
Eff
ecto
fEle
men
tary
Scho
olTe
ache
rVal
ue-A
dded
onH
igh
Scho
olO
utco
mes
-Clu
ster
ing
atth
eE
lem
enta
rySc
hool
Coh
ort
Lev
el
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
-0.0
02-0
.002
6.23
7***
0.00
20.
003*
*0.
005*
**0.
022*
**0.
016*
**0.
001
0.00
10.
001
(0.0
02)
(0.0
02)
(1.0
23)
(0.0
02)
(0.0
01)
(0.0
01)
(0.0
02)
(0.0
02)
(0.0
01)
(0.0
03)
(0.0
01)
Beh
avio
rVA
-0.0
030.
010*
**1.
955
0.01
3***
0.00
7***
0.00
5***
0.01
3***
0.00
4-0
.003
*-0
.016
***
-0.0
06**
*
(0.0
03)
(0.0
03)
(1.4
94)
(0.0
03)
(0.0
02)
(0.0
01)
(0.0
04)
(0.0
03)
(0.0
01)
(0.0
03)
(0.0
02)
Lea
rnin
gSk
ills
VA-0
.000
-0.0
01-0
.547
-0.0
02-0
.003
*-0
.003
**-0
.005
*0.
001
-0.0
020.
002
0.00
1
(0.0
03)
(0.0
02)
(1.1
73)
(0.0
03)
(0.0
02)
(0.0
01)
(0.0
03)
(0.0
03)
(0.0
01)
(0.0
03)
(0.0
02)
Obs
erva
tions
135,
786
102,
517
60,6
9429
3,02
823
3,07
823
3,07
815
2,34
515
1,82
031
6,12
327
7,33
322
1,75
7
R-s
quar
ed0.
293
0.14
50.
617
0.24
40.
234
0.23
90.
500
0.51
20.
040
0.26
70.
108
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
eth
ree
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
ed(s
eeTa
ble
4)on
stud
ents
’hig
hsc
hool
outc
omes
.Spe
cific
ally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
norm
aliz
edm
easu
res
ofte
ache
rva
lue-
adde
dfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’
high
scho
olou
tcom
eon
the
thre
em
easu
res
ofte
ache
rva
lue-
adde
dfo
rth
est
uden
ts’
teac
hers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1.T
heba
selin
eco
ntro
lsin
clud
ela
gsof
acu
bic
poly
nom
ialo
fthe
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,a
ndef
fort
GPA
;acu
bic
poly
nom
ials
ofla
gged
clas
san
dgr
ade
leve
lmea
nsof
each
ofth
ose
vari
able
s;an
dcu
rren
tEng
lish
lear
ners
tatu
san
dla
gged
log
days
abse
nt,s
uspe
nsio
ns,a
ndhe
ldba
ckfo
rthe
indi
vidu
al,t
hecl
ass
aver
age,
and
the
grad
eav
erag
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ners
tatu
sis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.Sta
ndar
der
rors
clus
tere
dat
the
leve
loft
heel
emen
tary
scho
ol-c
ohor
t.ar
ere
port
edin
pare
nthe
ses.
***
p<
0.01
, **
p<
0.05
,and
*p
<0.
10.
102
Tabl
eA
.35:
Eff
ecto
fEle
men
tary
Scho
olTe
ache
rVal
ue-A
dded
onH
igh
Scho
olO
utco
mes
-Mon
oton
icC
ontr
ols
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
-0.0
02-0
.002
5.73
0***
0.00
10.
002
0.00
5***
0.02
1***
0.01
6***
0.00
10.
001
0.00
1
(0.0
02)
(0.0
02)
(1.3
65)
(0.0
03)
(0.0
01)
(0.0
01)
(0.0
03)
(0.0
03)
(0.0
01)
(0.0
03)
(0.0
01)
Beh
avio
rVA
-0.0
030.
010*
**2.
346
0.01
4***
0.00
7***
0.00
5*0.
013*
*0.
004
-0.0
03-0
.016
***
-0.0
06**
(0.0
03)
(0.0
03)
(2.1
25)
(0.0
04)
(0.0
03)
(0.0
02)
(0.0
06)
(0.0
05)
(0.0
02)
(0.0
05)
(0.0
03)
Lea
rnin
gSk
ills
VA-0
.000
-0.0
02-1
.753
-0.0
04-0
.004
-0.0
03-0
.006
-0.0
00-0
.002
0.00
20.
002
(0.0
02)
(0.0
03)
(1.9
90)
(0.0
05)
(0.0
03)
(0.0
03)
(0.0
05)
(0.0
04)
(0.0
02)
(0.0
06)
(0.0
03)
Obs
erva
tions
135,
786
102,
517
60,6
9429
3,02
123
3,07
823
3,07
815
2,34
515
1,82
031
6,11
627
7,33
122
1,75
7
R-s
quar
ed0.
290
0.14
00.
599
0.23
70.
227
0.23
40.
489
0.50
30.
038
0.26
60.
104
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
eth
ree
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
ed(s
eeTa
ble
4)on
stud
ents
’hig
hsc
hool
outc
omes
.Spe
cific
ally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
norm
aliz
edm
easu
res
ofte
ache
rva
lue-
adde
dfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’
high
scho
olou
tcom
eon
the
thre
em
easu
res
ofte
ache
rva
lue-
adde
dfo
rth
est
uden
ts’
teac
hers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1ex
cept
allc
ontr
ols
ente
rth
efu
nctio
nlin
earl
y(n
one
are
cubi
cpo
lyno
mia
ls).
The
base
line
cont
rols
incl
ude
lags
ofa
stud
ent’s
mat
hte
stsc
ore,
Eng
lish
test
scor
e,G
PA,l
ogda
ysab
sent
,an
indi
cato
rfo
rsu
spen
sion
s,an
indi
cato
rfo
rbe
ing
held
back
,and
effo
rtG
PA;
lagg
edcl
ass
and
grad
ele
vel
mea
nsof
each
ofth
ose
vari
able
s;an
dE
nglis
hle
arne
rst
atus
for
the
indi
vidu
al,c
lass
,and
grad
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ner
stat
usis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
aco
ntro
lfor
clas
ssi
zeis
incl
uded
.St
anda
rder
rors
clus
tere
dat
the
mod
alhi
ghsc
hool
leve
lare
repo
rted
inpa
rent
hese
s.**
*p
<0.
01, *
*p
<0.
05,a
nd*
p<
0.10
.
103
Tabl
eA
.36:
Eff
ecto
fEle
men
tary
Scho
olTe
ache
rVal
ue-A
dded
onH
igh
Scho
olO
utco
mes
-Non
mon
oton
icC
ontr
ols
Pool
edG
rade
s3-
5VA
LA
USD
Dro
pout
Took
SAT
SAT
Scor
eG
PAE
ffor
tGPA
Coo
pera
tion
GPA
Mat
h
CA
HSE
E
Eng
lish
CA
HSE
E
Day
s
Susp
ende
d
Log
Abs
ence
sH
eld
Bac
k
Test
Scor
eVA
-0.0
02-0
.001
6.08
6***
0.00
30.
003*
0.00
5***
0.02
2***
0.01
7***
0.00
10.
000
0.00
1
(0.0
02)
(0.0
02)
(1.2
49)
(0.0
02)
(0.0
01)
(0.0
01)
(0.0
03)
(0.0
02)
(0.0
01)
(0.0
03)
(0.0
01)
Beh
avio
rVA
-0.0
020.
009*
**1.
611
0.01
3***
0.00
7***
0.00
5**
0.01
2**
0.00
3-0
.002
-0.0
16**
*-0
.006
**
(0.0
02)
(0.0
03)
(1.9
61)
(0.0
04)
(0.0
03)
(0.0
02)
(0.0
05)
(0.0
05)
(0.0
02)
(0.0
05)
(0.0
03)
Lea
rnin
gSk
ills
VA-0
.000
-0.0
02-0
.549
-0.0
02-0
.003
-0.0
03-0
.005
0.00
1-0
.002
0.00
10.
001
(0.0
02)
(0.0
03)
(1.7
63)
(0.0
05)
(0.0
03)
(0.0
03)
(0.0
05)
(0.0
04)
(0.0
02)
(0.0
06)
(0.0
02)
Obs
erva
tions
135,
786
102,
517
60,6
9429
3,02
123
3,07
823
3,07
815
2,34
515
1,82
031
6,11
627
7,33
122
1,75
7
R-s
quar
ed0.
295
0.14
60.
617
0.24
50.
235
0.24
00.
501
0.51
20.
040
0.26
80.
109
Not
e:T
hesa
mpl
ein
clud
esst
uden
tsin
grad
es3-
5w
hoat
tend
edhi
ghsc
hool
inth
eL
AU
SD.T
heun
itof
obse
rvat
ion
isa
stud
ent-
acad
emic
year
.Thi
sta
ble
repo
rts
the
effe
ctof
ast
anda
rdde
viat
ion
incr
ease
inth
eth
ree
mea
sure
sof
elem
enta
rysc
hool
teac
herv
alue
-add
ed(s
eeTa
ble
4)on
stud
ents
’hig
hsc
hool
outc
omes
.Spe
cific
ally
,eac
hco
lum
nof
the
tabl
ere
port
sth
eco
effic
ient
son
each
ofth
eth
ree
norm
aliz
edm
easu
res
ofte
ache
rva
lue-
adde
dfr
oman
OL
Sre
gres
sion
ofth
est
uden
ts’
high
scho
olou
tcom
eon
the
thre
em
easu
res
ofte
ache
rva
lue-
adde
dfo
rth
est
uden
ts’
teac
hers
ingr
ades
3,4,
or5,
alon
gw
ithth
eba
selin
eco
ntro
lsde
scri
bed
inse
ctio
n4.
1ex
cept
allc
ontr
olva
riab
les
(exc
eptd
umm
yva
riab
les)
ente
rthe
func
tion
ascu
bic
poly
nom
ials
.The
base
line
cont
rols
incl
ude
lags
ofa
thir
d-de
gree
poly
nom
ialo
fth
est
uden
t’sm
ath
test
scor
e,E
nglis
hte
stsc
ore,
GPA
,log
days
abse
nt,a
nin
dica
tor
for
susp
ensi
ons,
anin
dica
tor
for
bein
ghe
ldba
ck,a
ndef
fort
GPA
;thi
rd-d
egre
epo
lyno
mia
lsof
lagg
edcl
ass
and
grad
ele
velm
eans
ofea
chof
thos
eva
riab
les;
and
Eng
lish
lear
ners
tatu
sfo
rthe
indi
vidu
al,c
lass
,and
grad
e.E
ach
ofth
ese
vari
able
sex
cept
Eng
lish
lear
ners
tatu
sis
fully
inte
ract
edw
ithgr
ade
fixed
effe
cts
and
ath
ird-
degr
eepo
lyno
mia
lcla
sssi
zeis
incl
uded
.Sta
ndar
der
rors
clus
tere
dat
the
mod
alhi
ghsc
hool
leve
lare
repo
rted
inpa
rent
hese
s.**
*p
<0.
01, *
*p
<0.
05,a
nd*
p<
0.10
.
104