Easy A’s, Less Pay: The Long-Term Effects of Grade Inflation

Jeffrey T. Denning∗ Rachel Nesbit† Nolan Pope‡ Merrill Warnick§

October 15, 2025

Abstract

Average grades continue to rise in the United States, prompting discussion about the pos-
sibility of grade inflation. However, we know very little about the consequences of grading
standards for students. We study how grade inflating teachers affect students. First, we extend,
develop, and validate teacher-level measures of grade inflation. We construct two measures of
grade inflation, one that measures average grade inflation and another that measures whether
teachers are more likely to give a passing grade. These measures pass forecast bias tests com-
mon in the teacher effects literature. We show that grade inflation is distinct from teacher
value-added, with grade inflating teachers having moderately lower cognitive value-added and
slightly higher noncognitive value-added. Next, we consider the effects of grade inflation on
future outcomes. Mean grade inflation reduces future test scores, reduces the likelihood of
graduating from high school, reduces college enrollment, and ultimately reduces earnings.
However, passing grade inflation reduces the likelihood of being held back, increases high
school graduation, and increases initial enrollment in two-year colleges.

∗
University of Texas at Austin, NBER, IZA, and CESifo, Email: jeffdenning@[Link]
†
RAND, Email: rnesbit@[Link]
‡
University of Maryland, NBER Email: npope@[Link]
§
University of Georgia, Email: [Link]@[Link]

1
1 Introduction

Average grades have risen substantially in recent years (Gershenson, 2018). Although this might,

in principle, reflect real improvements in student learning, standardized test scores have not in-

creased at the same rate (Sanchez and Moore, 2022). This divergence suggests that the rise in

average grades likely reflects a shift toward lower grading standards. Understanding how grading

standards impact students – both in their academic performance and in their subsequent educa-

tional and professional outcomes – constitutes a central question for researchers and policymakers.

Indeed, school districts throughout the country are in the throes of contentious debates about

whether and how to change grading standards, with many teachers reporting an increase in prac-

tices that lower grading standards and raise grade inflation (Alex, 2022; Randazzo, 2023; Graham,

2017; Las Vegas Review-Journal Editorial, 2023; Griffith and Tyner, 2025).

Grading standards can arise from a variety of sources, but teachers interact most closely with

students and generally have discretion over assigning grades. This paper therefore focuses on

teachers’ individual grading standards. Effective teachers benefit students across numerous di-

mensions in both the short- and long-run, including test scores, suspensions, absences, effort, and

earnings in adulthood (Koedel and Rockoff, 2015; Petek and Pope, 2023; Jackson, 2018; Chetty et al.,

2014a,b). However, the practices that make some teachers more effective than others are poorly

understood, which prevents school leaders and policymakers from making optimal personnel,

training, and policy decisions (Staiger and Rockoff, 2010). As a notable exception to this critique,

a small literature shows that teachers’ grading practices, biases, and expectations affect student

outcomes (Carlana, 2019; Figlio and Lucas, 2004; Gershenson et al., 2022; Papageorge et al., 2020).

Those findings suggest that teachers, schools, and districts could actively change or adopt specific

grading policies to benefit students.

Debates on grading hinge on the theoretical ambiguity about whether high standards boost

performance by eliciting effort from students or hamper performance by discouraging students.

One camp suggests that grade inflation1 is harmful to students, leaving them unprepared for future

educational or vocational endeavors and decreasing their overall effort (Wright, 2019). Others
1
Much of the prior literature has referred to this concept as “lower grading standards.” We use the terms grade
inflation, grading leniency, and reduced/heightened grading standards interchangeably; we do not take a stand on the
correct name for this phenomenon.

2
suggest that grade inflation does not negatively affect students and that high grading standards

might even discourage students (Kohn, 2002). The arguments put forth by these two camps are

not necessarily mutually exclusive. Minimum grades are required to pass classes and continue

in high school, so grade inflation may help students by allowing them to continue to accumulate

human capital and to graduate when they otherwise would drop out. However, high standards

for awarding high grades may provide an incentive for students to study and learn the material. If

this incentive is reduced, students may study less and learn less. These two effects are in tension:

inflated grades can reduce early failure and dropping out, but blunt incentives to study and learn

material.2 This is a key insight in Costrell’s model of grade inflation; changing the standards for

graduation may increase effort and learning for some students, while reducing effort and learning

for others (Costrell, 1994).3

The effects of grade inflation also likely differ in importance for different students. Students on

the margin of failing a class are more likely to benefit from a small increase in their assigned grade.

Students who are significantly above the passing margin may be harmed by the reduced incentive

to study. Given the theoretical ambiguity and clear policy importance, empirical evidence can help

settle the question of what grading practices are best for students.

In this paper, we study the effects of grade inflation on high school students. To study the

effects of different types of grading standards, we construct two teacher-specific measures of grade

inflation. The first is “mean grade inflation,” which measures how much higher on average grades

are than expected, given student standardized test scores and other characteristics. This concept

is related to the one measured in Figlio and Lucas (2004), Mozenter (2019), and Gershenson et al.

(2022). The second is a novel measure, which we call “passing grade inflation,” and which mea-

sures inflation at the margin of passing a class – that is, inflating the student’s grade from an F to a

D or above. We find that these two measures of grading leniency are correlated, but distinct, and

highlight the tension between higher standards either improving student performance by eliciting

more effort or hampering performance by discouraging students. Consistent with the literature,

mean grade inflation decreases contemporaneous test scores. Additionally, we show new evidence
2
Landaud et al. (2024) show that randomly higher GPAs have long-term benefits for students. This is likely one
source of the incentive effects of grades.
3
Costrell (1994) nicely illustrates this key tension but makes simplifying assumptions such as a single credential and
a pooling equilibrium for wages for all students with the same credential level.

3
that mean grade inflation has long-term negative effects on students. In contrast, we find that lower

grading standards may help students at the margin: passing grade inflation increases five-year high

school graduation rates and enrollment in postsecondary education following high school.

To construct our measures of grade inflation in high school, we rely on several administrative

data sets of high school students. We use administrative data on students and teachers from

the nation’s second-largest school district, Los Angeles Unified School District (LAUSD). We also

use administrative data on students in the universe of public high schools in Maryland linked to

administrative college and earnings records. These data offer different strengths. More frequent

testing in LAUSD allows us to evaluate the impacts of grading standards on future test scores and

include more grades of high school. The Maryland data allow us to explore additional longer-term

outcomes such as college enrollment, graduation, and earnings. Despite the disparate settings, the

effects of grade inflation on student outcomes are remarkably similar across datasets.

Using these administrative data sets, we confirm that grades are increasing over time and

that these grade increases represent grade inflation and not increases in human capital. Figure

1 shows that average GPAs increased over the 2004-2014 school cohorts. Over these 10 years, the

mean GPA increased from 2.21 to 2.49, roughly a quarter of a letter grade. Over those same 10

years, for students who score within one-tenth of a standard deviation of the average on the state

standardized test, the average GPA rose by 0.41. Students scoring over 3 standard deviations above

the mean saw their average GPA increase by 0.17.4 Average GPA increases are also occurring within

subjects, as shown in Appendix Figure A.1. These trends in grade inflation are corroborated by the

literature. Several studies that have shown that grading has become more lenient over time, both

in high school and in college (Zhang and Sanchez, 2013; Gershenson, 2018; Hurwitz and Lee, 2018;

Denning et al., 2022; Sanchez and Moore, 2022).

This paper makes several contributions to our understanding of grade inflation and grading

practice. Our first is the conceptual contribution of distinguishing mean grade inflation, which is

similar to measures constructed by others in the literature, from passing grade inflation, which is

novel. These measures allow us to understand the trade offs of greater grading leniency along the

dimensions of effort elicitation and discouragement or persistence. These two measures of grading
4
The achievement of these top students is unlikely to have increased substantially over time, so these trends suggest
that the change in grades may reflect grade inflation rather than increases in human capital.

4
leniency are correlated (correlation coefficient of .85 in LAUSD and .3 in Maryland) but distinct.5

Hence we conclude that mean grade inflation and passing grade inflation represent related but

sometimes different grading practices of teachers. Because mean grade inflation and passing grade

inflation empirically often have opposite effects on student outcomes, the distinction is important.

We validate these measures of grading leniency in several ways. We use two tests to show that

our measures are forecast unbiased, and perform as well or better than cognitive value-added in

these tests. Additionally, we construct grade inflation measures with extra demographic controls

available in Maryland and find very similar results. Finally, we estimate the effect of grade inflation

on student outcomes separately for math and English courses, and again find very similar results.

Second, we ask how grade inflation relates to other teacher characteristics. One advantage of

our data is that we can consider several value-added measures, including those related to student

motivation and learning as in Petek and Pope (2023) as well as test score value-added. We docu-

ment that grading leniency is somewhat correlated with other well-established teacher character-

istics, such as cognitive value-added (correlation -.40) and noncognitive value-added (correlation

0.30). If these correlations are causal, they suggest that teachers may face tradeoffs in classroom

practices. This motivates our next series of exercises, which test whether grade inflation is on net

good or bad for students.6

Third, we evaluate the effect of both measures of grade inflation on student academic outcomes

such as high school graduation, future test scores, and SAT test taking. Previous work has docu-

mented that more lenient teachers reduce performance on tests in subsequent grades (Betts and

Grogger, 2003; Figlio and Lucas, 2004; Mozenter, 2019; Gershenson et al., 2022).7 Consistent with

prior evidence, we show that having higher mean grade inflating teachers reduces performance
5
Failure is uncommon in Maryland, which means there is less variation in passing grade inflation.
6
While we report correlations for measures of teacher value-added and grade inflation, conceptually they are different
in important ways. Teacher value-added can be thought of as a black box, and measures the effect of all teacher attributes
and behaviors on student success. Teacher value-added is hard to manipulate via policy, because it is less clear what
contributes to high teacher value-added. In contrast, grading practices are a policy choice of teachers (or other school
administrators) on how to map student performance into grades.
7
Our paper is related to work that uses a change in grading policy in North Carolina and finds that inflating grades
reduced student attendance and reduced ACT scores for some students (Bowden et al., 2023). This paper is very useful
for looking at statewide policy, but does not focus on teachers’ role in affecting grades. Our paper is also related to work
that uses random assignment of college instructors at the US Naval academy and find that teachers that give higher
grades tend to harm performance in follow on courses (Insler et al., 2021). We extend this study by focusing on high
shool students, considering longer term outcomes, and developing and validating measures that can be used in settings
without random assignment.

5
on future tests.8 Having math and English teachers who are on average one standard deviation

higher in mean grade inflation reduces test scores in the next year by approximately 0.03 standard

deviations. We build on previous work by documenting the persistent effects of grading leniency.

We find that having a one standard deviation higher mean grade inflating teacher decreases the

likelihood of graduating high school by 0.8 percentage points and of taking the SAT by 0.8 percent-

age points.

We further contribute to the existing literature by documenting important heterogeneity in

the effects of different types of grade inflation. While higher mean grade inflation is detrimental

to students’ future academic achievement, having a teacher with a higher passing grade inflation

measure is beneficial for students. We show that having higher passing grade inflating teachers

improves performance on future tests and increases the likelihood of graduating from high school

and taking the SAT. Having math and English teachers who are on average one standard devi-

ation higher in passing grade inflation increases students’ next year test scores by 0.01 standard

deviations and increases the likelihood of graduating high school and taking the SAT by 0.6 and

0.7 percentage points, respectively. In addition, having a higher passing grade inflating teacher

decreases the likelihood of being held back in the next year by 1.1 percentage points. Hence,

the nature of the grade inflation is critical for understanding the longer-term effects on student

outcomes, yet has previously been unexplored.

Fourth, we present novel evidence that grade inflation has longer-term impacts on students,

influencing their postsecondary enrollment, graduation, and labor market outcomes. We find

that exposure to higher mean-grade-inflating teachers reduces enrollment in postsecondary ed-

ucation programs, particularly Associate’s degrees, and also reduces graduation from Associate’s

degree programs. Furthermore, mean grade inflation reduces the likelihood that a student will

be employed up to six years after expected high school graduation and their earnings up to seven

years after expected high school graduation. On the other hand, passing grade inflation increases

enrollment in Associate’s programs after high school graduation, but reduces graduation from

Bachelor’s programs. We also find suggestions that passing grade inflation may have a small

positive effect on students’ labor market outcomes via increased earnings and employment, though
8
Mozenter (2019) finds no effect on longer-term outcomes, but the author only considers the effects of one of our
grade inflation measures. He also focuses on middle school students while we focus on high school students. Betts and
Grogger (2003) consider the consequences of grade inflation at the school level.

6
those results are sensitive to the definition of employment and the time period measured.

Fifth, we show that these two measures of grade inflation have heterogeneous effects among

different groups of students, especially along the academic ability distribution. Having a higher

mean-grade-inflating teacher has similar negative effects on students’ future test scores for students

of different academic abilities, but reduces grade retention more (in Maryland and Los Angeles)

and reduces SAT-test-taking less (in Los Angeles) among students in the bottom of the 8th grade

GPA distribution. In addition, the positive effects of passing grade inflation are concentrated

among lower-performing students: having a teacher who engages in passing grade inflation in-

creases graduation rates more and reduces grade retention more among students in the bottom of

the 8th grade GPA distribution. These results suggest that both mean and passing grade inflation

are slightly more beneficial for low-achieving students than for high-achieving students.

The paper proceeds as follows. Section 2 discusses the data and describes grades in the LAUSD.

Section 3 discusses our construction and estimation of the two different types of grading leniency.

Section 4 discusses the results. Section 5 concludes.

2 Data

We study grading practices in two settings, the Los Angeles Unified School District (LAUSD) and

the state of Maryland, using two datasets with complementary strengths. In Los Angeles, students

took end-of-year standardized tests through 11th grade. This allows us to create measures of grade

inflation and value-added, which rely on current and lagged test scores, and evaluate the impact

of those measures on future test scores through 11th grade. In addition, LAUSD teachers graded

their students on noncognitive dimensions (effort and cooperation) along with the traditional GPA,

allowing us to explore how grade inflation relates to both cognitive and noncognitive measures of

value-added. Finally, students in the LAUSD frequently receive failing grades for classes, which

means that the passing margin is meaningful for a large part of this dataset. However, the LAUSD

data lacks information on student demographic characteristics such as race, gender, and measures

of socioeconomic status.

We augment this analysis with linked K-12, college, and employment data from Maryland,

which allows us to explore longer-term outcomes in a sample that covers more recent years with

7
more demographic characteristics. Because in Maryland only approximately two years of math

and English exams are administered in high school,9 our analysis in Maryland is based on only

a subset of high school students that skew toward grades 9 and 10. In addition, our measure of

passing grade inflation in Maryland has less variation than the measure in Los Angeles, as few

students in Maryland fail their classes in high school. We therefore view the two data sources

as complements which together tell a more complete story about the effects of grade inflation.

Wherever possible, we report estimates from both Los Angeles and Maryland in order to provide

additional validation of the results. We describe both data sources in detail below.

Los Angeles Unified School District Data

The LAUSD data contain student-year observations from 2004 to 2013 for high school students.10

In 2004, the school district was 72.5 percent Hispanic, 11.8 percent Black, and 9.1 percent white.11 In

addition, over 70 percent of students received free and reduced price lunch in 2010.12 Historically,

the LAUSD’s academic performance has been lower than the national average. In the early 2000s,

the LAUSD had graduation rates that were below 50 percent but rose to 70 percent by 2014.

During this time period, students in grades 8 through 11 took the end of the year California

state test (CST) in math and English. The CST is a high-stakes, multiple-choice test administered to

all California students each spring. The English and math portions each consist of two 90-minute

parts. We standardize test scores at the grade-year level. In addition to yearly test scores, the data

include information on students’ grades in each course and their overall GPA. Students are given a

grade of A, B, C, D, or F for each class and GPA is measured on a 0 to 4 scale (e.g., A = 4.0, B = 3.0).

These two variables, GPA and end of year test score, are the main components of our measures

of grade inflation. The data also provide information about student behavior which we use to

construct measures of teacher value-added, as controls in our empirical analysis, and as next-year

outcomes. In particular, we use information about the number of days a student was suspended,

the number of days a student was absent, whether a student did not progress on time to the next
9
Specifically, Algebra I, Algebra II, Geometry, and either 9th or 10th grade English are the standardized tests that
might be administered to high school students in Maryland.
10
The data actually begin in 2003, but our main analysis sample starts in 2004 so that we have lagged test scores and
behavioral measures for use as controls.
11
These statistics were constructed using the downloadable files at [Link]
12
These statistics can be found at [Link]

8
grade (i.e., held back), and whether the student is an English Language Learner (ELL).

We evaluate the impact of grade inflation on a variety of outcome variables. These include

whether a student took the SAT and their score conditional on taking it, whether a student took

the PSAT and their score conditional on taking it, and indicators for high school graduation within

four or five years of starting ninth grade. We also look at next-year outcomes including test scores,

absences, suspensions, and whether the student is held back a grade.

Table 1 presents student-level summary statistics for the 985,020 high school students enrolled

between 2004-2013 for which we have information about test scores, grades, and behavior. In this

table we average over years, so that each observation represents one student.13 Several character-

istics are notable in the LAUSD. First, measures of student success are low. Only 51 percent of

ninth graders graduate from the LAUSD within 4 years and 62 percent graduate within 5 years. 12

percent of students are held back at some point, which we define as their administrative grade level

being the same as their grade level in the previous year. The average student is absent for 7 percent

of school days. While the LA data has very few demographic characteristics, 20 percent of students

are English language learners. In the LAUSD, our best measure of college intention is taking the

SAT, which is taken by 36 percent of students in our data. Among those who take the SAT, the

average score is a 1334 on a 2400 scale, which is approximately the 30th percentile of test takers.

Notably, grades are low, with an average GPA of 2.29. Course specific GPAs are even lower for math

and English, with averages of 1.76 and 2.19, respectively. During this time period in the LA school

district, teachers gave students grades based on their effort and their cooperation in addition to the

usual academic course grades. These grades facilitate our estimation of noncognitive teacher value-

added measures. Students perform similarly in these noncognitive dimensions, with an average

effort GPA of 2.16 and an average cooperation GPA of 2.39.

13
We require that students have the following information to be included in this sample: school and district identifiers,
grade level, math or English grade inflation measures, math or English teacher value-added measures, noncognitive
teacher value-added measures (GPA total value-added, fraction days absent value-added, suspension value-added,
and held back value-added), lagged math and English test scores, lagged total GPA, lagged fraction days absent, lagged
suspended, lagged held back, and an indicator for being ELL. This is the same restriction we require in our empirical
analysis; we make the restriction in this table to aid in interpreting our regression results.

9
Maryland Data

The Maryland data contain student-year observations from 2013 to 2023 for high school students.

They are provided via the Maryland Longitudinal Data System.14 These include educational records

from the Maryland State Department of Education linked to Maryland Higher Education Commis-

sion records and Maryland Department of Labor records. For high school students, we observe

student test scores in Algebra, Geometry, and either 9th or 10th grade English, as well as student

grades, demographic characteristics, and teacher identifiers. Notably, Maryland has information

about gender, race, ethnicity, and a proxy for low family income (Free and Reduced Price Meals),

which the LA data does not provide. We construct test scores and grades in a similar fashion to

those in the LA sample. We standardize test scores at the grade-year level, and grades in each

course are measured on a 0 to 4 scale. In addition to only administering end of grade standardized

tests for a subset of courses, Maryland’s exams also changed several times throughout the sam-

ple period. Originally the standardized tests were the High School Assessments (HSA), which

switched to the Partnership for the Assessment of Readiness for College and Careers (PARCC)

around 2014, which then switched to the Maryland Comprehensive Assessment Program (MCAP)

around 2022.15

Most of the higher education records are sourced from the National Student Clearinghouse,

which allows us to track students’ enrollment and degrees achieved at any school in the United

States, with additional information for those students who stay in Maryland. Our main higher

education outcomes of interest are enrollment in any higher education institution one and two

years after high school and graduating from any higher education institution four and six years

after high school. We are also able to explore enrollment and graduation separately for different

degree types as defined by the Maryland Higher Education Commission (Bachelor’s, Associate’s,

Master’s, Doctorate, and Professional or Certificate).

The workforce records from the Department of Labor are based primarily on unemployment

insurance and provide earnings at the quarterly level. They are therefore subject to the usual caveat

that we are only able to see employment with employers who have mandatory reporting to the state

unemployment insurance system; students who are self-employed, for example, are not captured in
14
See [Link]
15
The switch to the MCAP was originally intended for the 2019-2020 school year by was delayed by the pandemic.

10
our employment and earnings data. Our main workforce outcomes of interest are employment one

and six years after high school and annual earnings one and six years after high school. Our main

measure of employment is an indicator for having nonzero earnings in any quarter of the year.16

Our main measure of earnings is winsorized at the 99th percentile. For both higher education and

workforce outcomes, we define the time relative to their expected high school graduation – that

is, based on when the student was first in 9th grade – to avoid basing the timing on a potentially

endogenous outcome such as high school graduation. For example, “enrollment one year after

high school” is measured as whether the student was enrolled in a higher education institution

five years after 9th grade.

Table 2 presents summary statistics for the Maryland sample of high school students enrolled

between 2013 to 2023 for which we have information about test scores, grades, and behavior. As

in LA, in this table we average over years so that each observation represents one student. The

average math and English test scores are skewed slightly positive, consistent with a small positive

selection into the sample given that we require students to have lagged values of test scores and

course grades. The average GPA is 2.59 in math and 2.76 in English and 66 percent of students took

the SAT. Graduation rates are slightly higher than national averages with 92 percent of 9th graders

graduating within 5 years. Thus the students in Maryland appear to have better educational

outcomes than those in LAUSD, where many fewer students took the SAT and graduated within

5 years. The Maryland sample is 41 percent white, 34 percent Black, 14 percent Hispanic, and 7

percent Asian. In terms of longer-term outcomes, 65 percent of students enrolled in postsecondary

education a year after high school and 27 percent graduated within 4 years. Lastly, 62 percent of

students are employed in a job covered by unemployment insurance six years after their expected

high school graduation.17

3 Estimation

In this section, we construct two measures of teacher-specific grade inflation, mean and passing

grade inflation, and show that these measures do not suffer from forecast bias. We also construct
16
We also explore a more restrictive definition that requires earnings in more than two quarters, which yields
qualitatively similar results.
17
This is slightly lower than is seen for public high school graduates in Texas which have an employment rate of 76
percent 9-11 years after high school graduation Black et al. (2023).

11
measures of teacher value-added, and describe the estimation of correlation between different

grade inflation and value-added measures. Finally, we outline how we estimate the effects of grade

inflation on future outcomes.

3.1 Constructing Measures of Grade Inflation

In practice, teachers have a lot of discretion over how they assign grades. For a group of students

with the same underlying performance, a teacher can map that performance into different grades.

Those different mappings of performance to grades can change the number of students who receive

a grade of A, the average GPA of a class, or the number of students who fail. Different mappings of

performance to grades are likely to have different effects. For instance, a teacher who does not fail

students very often may reduce the probability of a student repeating a grade or failing to graduate

because they did not pass a required class. Alternatively, a teacher who gives many students A

grades may reduce the incentive for top students to study, which could hurt their performance

in future classes. From these two examples, it is clear that grade inflation theoretically could

improve or damage a students’ future academic performance. Whether grade inflation helps or

hurts depends critically on both the type of grade inflation in which the teacher engages and the

characteristics of the student.

With this motivation in mind, we explore two types of grade inflation. First, we are interested

in characterizing “mean grade inflation” which measures how much the average GPA is inflated.

Second, we are interested in characterizing “passing grade inflation” which measures how likely

a teacher is to pass a student.

To construct our first measure of mean grading leniency, we follow a method somewhat re-

lated to Figlio and Lucas (2004), Gershenson et al. (2022), and Mozenter (2019).18 We model the
18
Our method differs in one important way from Figlio and Lucas (2004) and Gershenson et al. (2022): we predict
grade as a function of test score and teacher, whereas they predict test score as a function of grade and teacher. Our
approach follows from a structural model in which a student’s measured performance in a course (their grade) is a
function of their underlying academic performance (their test score and prior performance in that subject) combined
with whatever discretion the teacher has in assigning the grade (we call that discretion “grade inflation” or “grading
leniency”).

12
student’s grade as in Equation 1:

Gradeijst = β1 T estScoreijst + β2 Gradeist−1 + β3 M athT estijst−1

(1)
mean
+β4 EnglishT estijst−1 + GIjt + Xit β + εijst

where i indexes student, j indexes teacher, s indexes school and t indexes year. The object of

interest is GIjt
mean which is the year-specific teacher fixed effect. This is the teacher’s contribution to

grades after controlling for several important factors. First we account for a student’s performance

in the subject as measured by the corresponding subject test score, T estScoreijst . Second, we

also control for Gradeijst−1 which is the student’s grade in the focal subject from the prior year.

Third, we control for prior test scores in both math and English, Additionally, we control for

student characteristics. The controls include school, grade, and year fixed effects, an indicator

for English language learner (ELL), previous year math test score, previous English test score, the

total academic GPA, the fraction of days absent, indicators for suspension, and held back status.19
c mean
For each teacher we calculate GI jt , which is our measure of grading leniency. We will

call this “mean grading leniency” or ”mean grade inflation” to distinguish it from passing grade

inflation which we discuss below. Mean grading leniency represents how much a teacher raises (or

lowers) their students’ average grades relative to their academic performance. We make several

adjustments to our measures of grade inflation for use in estimating the effect of grade inflation on
c mean
future outcomes. Following Chetty et al. (2014a), we estimate GI jt
c pass
and GI jt using a jackknife

empirical Bayes estimator. This approach uses data from surrounding years to estimate a teacher’s

propensity to grade inflate in year t, which avoids biasing estimates of the long-term effects of

teacher grade inflation on student outcomes (Jacob et al., 2010). Including year t in the prediction

would likely bias the estimates because unobservables in year t that are related to any dimension

of student performance would be captured in both the measure of grade inflation in year t and the

outcome of interest.
c mean is fundamentally defined as a residual. It is worth considering what
The estimate GI jt

c mean
GI jt could be capturing aside from grading leniency. We rule out some potential alternative

explanations by controlling for student characteristics. For example, student performance is ac-
19
For the Maryland sample, because students do not necessarily take math and English exams every year, we replace
the previous year test score with the most recent previous test score.

13
counted for in two ways. First, Equation 1 includes the student’s contemporaneous and past stan-

dardized test score. Second, we include the student’s grades from the previous year. This accounts

for students who might perform poorly on subject tests, but demonstrate their understanding of

the subject through their performance on non-test assessments.20

c mean could represent something a teacher does to improve their students’ grades
However, GI jt

in a way not captured by contemporaneous test scores. That could be a skill the teacher conveys

to their students that improves grades but not test scores, such as helping students learn to work

in groups. In our setting, a teacher who is very good at conveying skills not captured by contem-
c mean measure. As a result, we would expect higher
poraneous test scores would have a high GI jt

c mean
GI jt to improve future performance in the next year. Instead, we will show in our results that
c mean
high GI jt teachers reduce future performance, which lends more support for our interpretation

of this residual as a measure of grading leniency.

Our second measure of grading leniency replaces Gradeijst in Equation 1 with an indicator for

passing the class. This indicator is equal to 0 if a student received an F in the course and equal to

1 if the student received a grade of D or better. As discussed, we create this alternative measure of

grading leniency because some teachers may raise the grades of their students generally, whereas

others may only raise grades when students are on the margin of failing the class. We expect that

these two measures of leniency may have different effects on students’ future performance.21 For

our second measure, we are still interested in estimating the teacher effect in the modified Equation
pass
1, which we refer to as GIjt .

The ultimate goal is for our measures to characterize how much grade inflation a student
mean pass
experiences in a given year.22 To accomplish this, we calculate GI
c
jt and GI
c
jt for each teacher-

year-subject observation and standardize this to be mean zero, standard deviation one within year

and subject.23 We then characterize the grade inflation that a student experiences in a year by
20
Insofar as a student has a surge in performance in year t (above that expected by their contemporaneous test scores
and prior performance in that subject) our measure will not account for this.
21
We study two measures of how teachers map student performance into grades, but there are many potential
alternative measures of this mapping. For example, one teacher may be more likely to give Bs (and fewer Cs) than
another teacher, but not to inflate As or Ds. We focus on mean grade inflation to capture general leniency. We focus on
passing grade inflation because of the institutional importance of a student passing a class, and because some schools
may have formal or informal policies that pressure teachers not to fail their students.
22
Note that since we are using high school data, students have more than one math and English class per year, and
could even have multiple in the same semester.
23
Calculating at the subject, rather than course, level means that if a teacher teaches two different math classes in a
year, our measure of grade inflation for that teacher will be the same for those two classes.

14
averaging over all the student’s teachers in a given subject, weighted by the number of classes a

student takes with that same teacher.24 We then standardize the students’ subject-specific grade

inflation measures to be mean zero, standard deviation one within year. In most of our analysis, we

use a measure of grade inflation that combines math and English. We generate this by summing

the standardized grade inflation measures for each subject. We then standardize this sum so that

it is mean zero, standard deviation one within year. We do all of the above separately for mean

and passing grade inflation. As a result, in our estimates of the effect of grade inflation on future

performance, the coefficient on these measures represents a one standard deviation increase in the

grade inflation that a student experiences in a given year.

3.2 Validating Measures of Teacher Grade Inflation

Our measures of grade inflation face some of the same challenges faced by value-added measures.

First, these measures are estimated with noise and may not be predictive of actual student outcomes

out of sample. We refer to this lack of predictive power as estimation error. For example, if student

grades were very noisy year after year due to random variation in unobservables, the variation

could drown out the teacher signal, and teacher measures would not be very predictive of student

outcomes. Second, these measures may be biased due to student selection on some observable

omitted by our econometric model. We refer to this type of bias as selection. For example, if certain

students who earned high grades through some means not captured by our model consistently are

assigned to teachers who give high grades, other past measures of student ability not included in

our model would predict that those students would have higher grades.

Both of these concerns would result in forecast-biased measures of grade inflation. As in

Chetty et al. (2014a), forecast bias is measured by one minus the correlation between true the-

oretical teacher measures and the estimated teacher measure:

!
cov(GIjt , GI
c jt )
Bias = 1− .
var(GI
c jt )

Estimation error and selection would push forecast bias in different directions. Estimation error can

cause forecast bias because very noisy estimates of grade inflation would lead to a zero correlation
24
Weighting means that if a student takes three math courses and two are with the same teacher, that teacher’s grade
inflation measure will be used twice in the average.

15
on the right hand side of the expression and bias of 1. Selection can cause forecast bias because

it will misattribute some of the effect of the omitted variable on the outcome to the teacher effect,

leading to a lower covariance between true and estimated measures.

To address these concerns, we conduct two tests for forecast bias. Both tests use jackknifed

estimates of mean and passing grade inflation, which use all other years of data to estimate value-

added GI
b jt this year. The first, which we call the “individual test,” addresses the first concern,

by checking whether these teacher measures are predictive out of sample. To implement this test,

we regress student grades and passing indicators on grade inflation measures, controlling for the

same observables as in the estimating equations:

mean
Gradeijst = δ mean GI
c jt + α1 T estScoreijst + α2 Gradeist−1 + Xit α + ηijst
pass
P assijst = δ pass GI
c
jt + γ1 T estScoreijst + γ2 Gradeist−1 + Xit γ + ξijst .

In these regressions, 1 − δb will be our estimates of forecast bias. In this test, an estimate of δ close to

1 shows that grade inflation measures, though subject to estimation error, are actually predictive

of student outcomes out of sample. This predictive power results in lower forecast bias.

Results from the individual forecast bias tests are found in Table 3. The first two columns

show that our grade inflation measures show small levels of forecast bias across both subjects and

samples. Across both subjects, samples, and measures, we see estimates of forecast bias ranging

from 0.03 to 0.1, suggesting our measures are quite predictive. These estimates of forecast bias are

similar to estimates from the value-added literature.25

The second exercise, which we call the “prediction test,” addresses the second concern of

omitted observables. Specifically, we test whether selection on twice-lagged math and English test

scores leads to bias in our grade inflation measures.26 To conduct this test, we begin by residualiz-

ing the predictors (twice-lagged test scores) and the outcomes (grade and the passing indicator)

on the same set of controls that were used in estimation of our teacher measures, which include
25
For example, Kane and Staiger (2008) obtain forecast bias estimates of 0.06 to 0.17 when testing measures of value-
added of Kindergarten students. Chetty et al. (2014a) obtain forecast bias estimates of 0.02 to 0.05 for children in grades
3-8.
26
We only conduct this exercise in LAUSD because students take fewer standardized tests in Maryland.

16
school, grade and year fixed effects, English Language Learner status, and previous year values of

math test score, English test score, total GPA, fraction of days absent, suspension and held back.

We then construct predicted outcomes by regressing the residualized outcome on the residualized

predictors and construct fitted outcomes Grade

\ and P[
ass. Finally, we conduct the forecast bias test

by regressing the fitted outcomes on our teacher measures:

c mean + η ′ ,
\ ijst = ν mean GI
Grade jt ijst

with a similar equation for passing grade inflation. The estimates of ν are then estimates of forecast

bias. The closer νb is to zero, the less sorting due to the predictors, and the less forecast bias due to

sorting on the omitted observables.

Panel C from Table 3 shows the results of the prediction tests. Again, across both subjects and

outcomes, our grade inflation measures show low levels of forecast bias. The forecast bias estimates

range from 0.00 to 0.02 in absolute value, providing even stronger evidence than the estimates from

the individual tests.

Together, these tests suggest that our measures of grade inflation are predictive of instructor

grading behavior and do not suffer from unobserved selection. Establishing these patterns is

important for our use of this measure in our paper, but also for future work that use this measure.27

3.3 Constructing Other Measures of Teacher Quality

In order to understand how related our measures of grade inflation are to more common indica-

tors of teacher quality, we construct both test score and noncognitive value-added measures. We

estimate cognitive value-added using a jackknife empirical Bayes estimator following Chetty et al.

(2014a). We also estimate noncognitive valued-added using a jackknife empirical Bayes estimator

following Petek and Pope (2023). We create six different noncognitive value-added measures:

absences, suspension, grade retention, total academic GPA, cooperation GPA, and effort GPA. Due

to concerns about teachers affecting these noncognitive measures directly, we follow Petek and

Pope (2023) and use outcomes measured in the year after the student and teacher interact.

We then create student-year level measures of value-added in a similar way to our student-
27
Figlio and Lucas (2004); Mozenter (2019); Gershenson (2018) use related measures but we are unaware of prior
work showing that they are forecast unbiased.

17
year level measures of grade inflation. We first construct test score and noncognitive teacher

value-added measures for each teacher-subject-year observation, standardized to have mean zero,

standard deviation one within year. We then construct a student-level value-added measure that

averages the value-added from all the student’s teachers in a given subject/year, standardized to

be mean zero and standard deviation one. We combine math and English test score value-added

into a single “cognitive value-added” measure by summing these measures across subjects and

then standardizing to be mean zero, standard deviation one. Similarly, following Petek and Pope

(2023) we combine math and English course measures of days absent, suspension, grade retention,

cooperation GPA, total academic GPA, and effort GPA28 into a single “noncognitive” value-added

measure by summing the components and standardizing to be mean zero, standard deviation one.

We also create subject-specific “noncognitive” value-added measures by summing the components

from only that subject’s teachers and standardizing to be mean zero, standard deviation one.

3.4 Estimating Correlations Between Teacher Measures

In order to understand whether or not grade inflation is simply another measure of value-added

we construct correlations between both grade inflation measures and both value-added measures.

As described in Jackson et al. (2024), raw correlations between estimates of value-added may be

too small or too large relative to the cross-measure correlation between true value-added. First, the

correlations may be attenuated because of measurement error-induced noise in grade inflation and

value-added measures. Second, the correlations may be too large because if these measurement

errors are correlated across outcomes within the same year.

In order to address these challenges, we estimate the correlation between value-added mea-

sures using a split-sample, bootstrapping approach. Splitting the sample before estimation resolves

the concern of within-year correlations in measurement error across outcomes. We randomly split

the sample within classroom,29 then estimate teacher measures µA and µB , with the subscripts

denoting different outcome measures (e.g. grade inflation or cognitive value-added), separately

on both samples. This produces two estimates of value-added for each measure: µ0A , µ1A , µ0B , and

µ1B , with the superscripts denoting the separate samples. Next, we estimate the across-sample
28
The cooperation and effort GPAs are only available in LAUSD.
29
This within-classroom splitting approach was inspired by the correlation estimation of student-level measures in
Ayllón et al. (2025).

18
correlation between the measures as

ρb01 0 1


AB = corr µA , µB .

We then adjust this correlation for measurement error attenuation by dividing by the product

of the within-measure, across-sample correlations ρb01

A = corr µA , µA and ρ
0 1
B = corr µB , µB .
b01 0 1

This adjustment addresses the attenuation concern from measurement error in teacher measures.

The adjustment gives us rbAB as a unattenuated estimate of the correlation between the teacher

measures:
ρb01
rbAB = q AB .
ρb01
Aρ b01
B

Note that each sample split will give us two estimates of the correlation rbAB and rbBA , which need

not be numerically identical. To obtain our actual correlation estimates, we bootstrap this quantity

by splitting the sample in half 100 times, then averaging across the 200 resulting estimates.

3.5 Estimating the Effects of Grade Inflation

To explore the effects of grade inflation on future outcomes, we estimate specifications similar to

Chetty et al. (2014b), Petek and Pope (2023), Gershenson et al. (2022), Figlio and Lucas (2004), and

Mozenter (2019) where an observation is a student-year and we regress a longer-term outcome on

the two grade inflation measures, the test score and noncognitive value-added measures, and the

same set of controls used to construct these measures. In particular we estimate the following

equation:

mean pass test noncog

Yit = αmean GI
c it + θpass GI
c it + δcogV A V
d Ait + ψnoncogV A V
d Ait + Xit β + ηit (2)

where Yit is a future outcome, like graduation from high school or test score performance in the
c mean is the average measure of mean grade inflation a student experiences in
following year, GI it

c pass
year t, GI it is the average measure of passing grade inflation a student experiences in year t,
test noncog
V
d Ait and V
d Ait are the average measures of test score and noncognitive value-added a student

experiences in year t, and Xit is the same vector of individual level controls we use when estimating

19
grade inflation in Equation 1.30

In our main results we use measures of grade inflation and value-added which are averages

of the subject-specific measures, as described in Sections 3.1 and 3.3. We also explore how grade

inflation might be different in math and English classes and might have different effects on future

performance. In those specifications, we use the subject-specific versions of our grade inflation

and value-added measures. In both cases, the grade inflation and value-added measures have

been standardized to mean zero, standard deviation one.

In regressions based on Equation 2, an observation is a student-year. We cluster our standard

errors at the school level to account for within-school correlation of outcomes. The coefficients of

interest are αmean and θpass which estimate the effect of mean grade inflation and passing grade

inflation after accounting for other student and teacher characteristics. We also report δcogV A and

ψnoncogV A to verify that our measures of teacher value-added have the expected estimated effects

and to compare magnitudes. In all regressions, we limit our sample as described in Section 2

for Table 1, requiring students to have the necessary information to construct grade inflation and

value-added measures. In addition, we implement sample restrictions that vary by the availability

of the outcome. For example, the CSTs are only administered through 11th grade, so we do not

have future test scores for 11th and 12th graders, and we exclude them from the analysis when

future test score is the outcome variable.

4 Results

In this section, we find that grade inflation is a separate characteristic from value-added, and that

our two measures of grade inflation are related but distinct. We then find that grade inflation does

matter for students’ future outcomes. Mean grade inflation decreases students’ future standard-

ized test scores, probability of graduation, college enrollment, employment and earnings. Passing

grade inflation reduces the probability of being held back, increases the probability of graduation,

and increases postsecondary enrollment. We also find that the effects of grade inflation vary across

the ability distribution.

30
These controls include school, grade, and year fixed effects; an indicator for English language learner (ELL); and
previous year math test score, english test score, total academic GPA, fraction of days absent, an indicator for being
suspended, and an indicator for being heldback.

20
4.1 How is Grade Inflation Related to Other Teacher Characteristics?

Before exploring the relationship between grade inflation and students’ future academic and career

outcomes, we first characterize the relationship between our measures of teacher grade inflation

and other common measures of teacher quality. We show that the two measures we define in

Section 3.1 represent related, but distinct aspects of teacher grading standards. Furthermore, we

show that these grade inflation measures capture a separate aspect of teacher quality than that

captured by traditional cognitive and noncognitive value-added measures.

Table 4 presents correlations of our two measures of grade inflation with two measures of

teacher value-added: cognitive value-added and noncognitive value-added. All correlations are

calculated using the method described in Section 3.4 to avoid bias due to measurement error. These

correlations are inherently not causal, but rather reflect the extent to which grade inflation and

other teacher attributes covary in the observed in the data.

We would expect our two measures of grade inflation to be positively correlated, since a teacher

who raises the grades of all students will also increase the probability of a student passing. This

is confirmed in panels A and B of Table 4, which show that mean grade inflation and passing

grade inflation are positively correlated with each other in both the LAUSD and Maryland settings.

In Los Angeles, mean and passing grade inflation have a correlation of 0.86, while in Maryland

that correlation is notably lower, around 0.36. This lower estimated correlation in Maryland is

likely explained by the low fail rates in Maryland,31 which translates into less variation in the

passing grade inflation measure. Nevertheless, in both settings, the two measures are highly but

not perfectly correlated – passing grade inflation and mean grade inflation appear to be distinct

attributes. Therefore, we will be able to use both measures in Equation 2 to isolate the effects of

different types of grade inflation.

Table 4 also presents estimates of the correlation between teachers’ grade inflation and value-

added measures. Teachers who inflate grades tend to have lower cognitive (test score) value-

added. Mean grade inflation is negatively correlated with our cognitive value-added measure

(a combination of math and english test score value-added) at -0.41 for Los Angeles and -0.31 for

Maryland. This negative correlation is consistent with Betts and Grogger (2003), Figlio and Lucas
31
Only 3.1% of Math courses and 2.3% of English courses are failed by Maryland students, compared to 32% and 21%
in LAUSD.

21
(2004), Mozenter (2019), and Gershenson et al. (2022), who find that grade-inflating teachers

reduce test scores. Passing grade inflation is also negatively correlated with cognitive value-added,

though the correlation is weaker, with estimated correlations of -0.30 and -0.07 in Los Angeles and

Maryland respectively. Grade inflation tends to be weakly positively correlated with noncognitive

value-added (a combination of measures of value-added on absences, suspensions, grade reten-

tion, academic GPA, cooperation GPA, and effort GPA32 ). The correlation between mean grade

inflation and noncognitive value-added is estimated to be 0.16 in Los Angeles and 0.12 in Maryland.

For passing grade inflation, the estimated correlation is 0.18 in Los Angeles and 0.05 in Maryland.

The correlations between grade inflation and value-added measures suggest that grade in-

flation represents a distinct teacher characteristic from typical value-added metrics. Mean grade

inflation and passing grade inflation are correlated, but are separate attributes. Grade inflation

is moderately negatively correlated with cognitive value-added and weakly positively correlated

with noncognitive value-added. If these correlations contain a causal component, a teacher’s deci-

sion to inflate grades could carry a tradeoff because grade inflation is both positively and negatively

correlated with other desirable teacher characteristics. For instance, a grade-inflating teacher might

induce students to attend class more but reduce student learning as measured by test scores.

Inflating grades could even be the optimal choice to maximize certain student outcomes. As a

result, in the next section we explore whether grade inflation has a positive or negative impact on

students’ cognitive and noncognitive performance in high school, college, and their careers, while

holding other teacher characteristics constant.

4.2 Does Grade Inflation Matter for Future Outcomes?

Table 5 presents the results from a simple regression of the relevant student outcome on just one of

our measures of teacher characteristics (mean or passing grade inflation, or cognitive or noncog-

nitive value-added) at a time. Unless otherwise stated, all regression analysis discussed here and

in later sections controls for the same student characteristics used to create the grade inflation and

value-added measures (school, grade, year, ELL, previous math test score, previous English test

score, previous total academic GPA, previous fraction of days absent, previous suspension, and

previous grade retention). Standard errors are clustered at the school level.
32
The cooperation and effort GPAs are only available in LAUSD.

22
 Table 5 verifies that, when considered in isolation, the estimated associations of teacher value-

added measures with student outcomes have the expected sign. Having higher cognitive value-

added teachers is associated with increases in test scores, SAT scores, and the likelihood of gradu-

ating. Higher noncognitive value-added is associated with reductions in the likelihood a student

is held back and increases in the likelihood of graduating. The direction of these associations is

consistent with the existing literature. In addition, we estimate that grade-inflating teachers are

associated with reductions in future test scores, SAT scores, and the likelihood that a student is

held back. However, these results do not separately identify the impact of grade inflation while

holding other teacher characteristics (i.e., value-added) fixed. That is, the negative relationship

between grade inflation and student outcomes estimated in Table 5 could simply reflect the nega-

tive correlation between grade inflation and cognitive value-added rather than any causal effect of

grade inflation on outcomes.

To disentangle this, in our main results we explore whether grade inflation matters for future

outcomes conditional on other teacher characteristics by estimating Equation 2. Throughout we include

mean grade inflation, passing grade inflation, cognitive value-added, and noncognitive value-

added in each regression. This allows us to identify the effect of each measure of teacher quality

on student outcomes while holding constant the other teacher characteristics.

In Table 6 we first focus on cognitive and noncognitive value-added to verify that the estimated

coefficients have the expected magnitude and direction. In Los Angeles, having teachers with

higher cognitive value-added improves future end-of-course standardized test scores, SAT scores,

the likelihood of graduating, and the likelihood of taking the SAT. In Maryland, the estimated

effects of cognitive value-added are similar, though the relationships with graduation rates and

with SAT scores are not statistically significant. One exception is that in Maryland, cognitive

value-added increases the chance of being held back, although the effect size is very small (one

tenth of a percentage point). In both Los Angeles and Maryland, having teachers with higher

noncognitive value-added improves future English test scores, reduces the likelihood of grade

retention, increases graduation rates, and increases the likelihood of taking the SAT. In addition,

in Maryland, noncognitive value-added slightly reduces future SAT scores. These results closely

mirror those of Petek and Pope (2023).

23
High School Outcomes

Having determined that Equation 2 yields sensible estimates for well-established value-added

measures, we turn to estimating the effects of mean and passing grade inflation on high school

outcomes conditional on teacher value-added. Table 6 presents estimates for our main high school

outcomes (future math and English test scores, the likelihood of being held back, the likelihood

of graduating in five years, the likelihood of taking the SAT, and SAT score); Tables A.11 and A.12

presents estimates for additional high school outcomes such as absences, suspensions, and PSAT

scores.

When evaluating the effect of grade inflation on future test scores, we focus on the Los An-

geles sample, where standardized testing is more frequent. LAUSD administers standardized

tests yearly through 11th grade, whereas Maryland high school students take standardized math

and English tests in only about two years of high school.33 In Los Angeles, we find that mean

grade inflation reduces future test scores. A one standard deviation increase in the amount of

mean grade inflation a student is exposed to in one year reduces future math test scores by 0.024

standard deviations and future English test scores by 0.020 standard deviations. In comparison,

being exposed to higher cognitive value-added teachers increases future test scores by .113 for

math and .054 for English. Comparing the grade inflation and value-added estimates, a teacher’s

grading leniency has a meaningful impact on future test score performance – about 30 percent as

large as the effect of traditional value-added measures in math and 44 percent as large in English.

This finding closely mirrors that of Figlio and Lucas (2004) and Gershenson et al. (2022).

Beyond future test scores, mean grade inflation also impacts some other high school outcomes.

We estimate that mean grade inflation negligibly reduces the chance that a student takes the SAT,

which allows us to interpret the effect of grade inflation on an alternative measure of student

achievement, SAT scores, without worrying about selection into taking the test. However, in both

Los Angeles and Maryland, we find no effect of mean grade inflation on SAT test scores. Mean

grade inflation does have a small negative effect on PSAT scores (see Table A.11). The PSAT is de-

signed for a different purpose and test different skills than end-of-course standardized tests. Thus

the negative effects of mean grade inflation on both PSAT score and end-of-course standardized
33
Specifically, Algebra I, Algebra II, Geometry, and either 9th or 10th grade English are the standardized tests
administered in Maryland.

24
test scores suggest a more general reduction in human capital. Finally, we find that mean grade

inflation reduces the likelihood of graduating within 5 years by 0.8 percentage points in LAUSD

and 0.1 percentage points in Maryland. Five-year graduation rates are much higher in Maryland

than in Los Angeles (90 percent and 58 percent, respectively); the smaller effect of mean grade

inflation on graduation in Maryland could be explained by differences in the marginal student at

risk of not graduating. In Tables A.11 and A.12 we also show that grade inflation increases absences

and increases suspensions in both the LAUSD and in Maryland.

Unlike mean grade inflation, we find that passing grade inflation has no effect on future test

scores in math or English. However, passing grade inflation appears to be beneficial for some other

outcomes. We estimate that being exposed to more passing grade-inflating teachers reduces the

likelihood of being held back by 1.2 percentage points in Los Angeles and 0.4 percentage points in

Maryland. Being held back is defined as having the same administrative grade code in the current

and following year, and is a rare event, with 13 percent of students in our Los Angeles sample and

4 percent of students in our Maryland sample being held back. Thus this is a meaningful effect

size, representing about a nine to ten percent decline in the likelihood of repeating a grade. We

also estimate that passing grade inflation increases the likelihood of graduating within five years

by 0.6 percentage points in Los Angeles and 0.2 percentage points in Los Angeles. We find no effect

of passing grade inflation on taking the SAT, but a small reduction in SAT scores of about 3 points

in both Los Angeles and Maryland.

In summary, for high school outcomes we find that our more traditional measure of mean

grade inflation is reduces educational outcomes, evidenced by lower test scores in the following

year and a lower probability of graduating high school. This is consistent with reductions in

learning associated with muted incentives that arise from easier grading. In contrast, passing

grade inflation offers potential benefits for students by decreasing grade retention and increasing

the probability of graduating high school. However, from this evidence alone it is not clear whether

these higher rates of high school completion brought about by passing grade inflation are beneficial

to students in the longer term. We turn to our analysis of college and early career outcomes to

explore this further.

25
Longer Term Outcomes

Using Maryland’s linked K-12, college, and employment data, we are able to explore the impact

of mean and passing grade inflation experienced during high school on college enrollment and

graduation, earnings, and employment. In this analysis, we estimate the effect of grade inflation on

college and career outcomes at different times relative to the students’ expected high school gradu-

ation in order to capture any dynamic effects of the two grade inflation measures. This means that

the sample is composed of different cohorts for different outcome measures. For example, when

the outcome is being enrolled in any postsecondary education one year after expected high school

graduation, our data includes students who were in 12th grade in 2013 through 2022, whereas

when the outcome is measured eight years after expected high school graduation, only 12th-grade

cohorts 2013 through 2015 are included.

We find that the effects of grade inflation extend into early adulthood.34 Table 7 displays these

results. We find that mean grade inflation reduces enrollment in any postsecondary schooling by

about 0.5 percentage points, but has no measurable effect on the likelihood of graduating from

postsecondary schooling. This suggests that the students marginally induced to avoid college

enrollment would not have graduated if they had enrolled. We investigate this result further by

evaluating the effect of grade inflation on enrollment in two-year and four-year college separately

in Table 8. We find that mean grade inflation reduces enrollment in Associate’s degree programs

from one to seven years past expected high school graduation. For Bachelor’s degrees, we es-

timate reductions in enrollment from three to six years after expected high school graduation,

with negative but not statistically significant reductions in the first two years as well. Looking at

postsecondary graduation in Table 9, mean grade inflation reduces Associate’s degree completion

but has no effect on Bachelor’s degree completion. Passing grade inflation increases enrollment

in two-year schools one to five years after expected high school graduation. The point estimate

for passing grade inflation on enrollment in (and graduation from) four-year schools is negative,

perhaps because it reduces preparation as measured by the SAT, though it is only (marginally)

significant in the fourth year after expected high school graduation. The net effect is that passing

grade inflation brings about no change in any postsecondary enrollment, but as Table 7 shows,
34
This is similar to value-added. For example, Chetty et al. (2014b) showed that test-score value-added in elementary
and middle school affected student’s college-going and earnings.

26
passing grade inflation does decrease the likelihood of postsecondary degree completion by about

0.3 percentage points.

Grade inflation also affects students’ labor market outcomes, as Tables 7 and 10 show. Mean

grade inflation reduces employment one to four and six years after expected high school grad-

uation, whereas passing grade inflation has no measureable effect on employment. Our main

measure of employment is having any positive wages in at least one quarter of the year, but in 10 we

also impose a more strict criterion of working at least two quarters of the year to exclude those who

only work a summer job while in college. We estimate very similar effects of mean grade inflation

on employment; exposure to one standard deviation higher mean grade inflation in high school

reduces the probability of employment by 0.2 to 0.3 percentage points in the first four years after

expected high school graduation, or 0.3 to 0.4 percentage points when our measure is employment

for two or more quarters. Passing grade inflation has no statistically significant effect on any

employment, but does have a statistically significant positive effect (0.2 to 0.3 percentage points) on

employment for two or more quarters in the first four years after expected high school graduation.

This is consistent with the evidence we see on enrollment, which suggests passing grade inflation

contributes to a small shift away from four-year to two-year programs, where working throughout

the school year is more common.

Given the changes in employment due to mean grade inflation, we focus on unconditional

earnings (winsorized at the 99th percentile) as our measure of earnings. We find that mean grade

inflation reduces earnings by about $42 to $133 a year from one through six years after expected

high school graduation. We estimate a small positive relationship between passing grade inflation

and earnings of similar magnitude to the negative relationship between mean grade inflation and

earnings, though it is only statistically significant three and four years after expected high school

graduation.

Our exploration of the longer-term impact of mean and passing grade inflation suggests that

the negative effects of mean grade-inflating teachers persist into young adulthood. However, pass-

ing grade inflation shows more muted long-term effects, with some suggestion of a shift away from

four-year college enrollment and toward two-year. Despite the increased probability of graduating

high school, passing grade inflation does not result in an appreciable increase in earnings. These

results are consistent with the hypothesis that mean grade inflation reduces human capital accu-

27
mulation. They also suggests that the signaling value of a high school diploma may be minimal,

or offset by reductions in human capital as measured by SAT scores.

4.3 Heterogeneity

In the previous sections, we have focused on two forms of grade inflation and have shown they have

different average effects on students. However, these effects may be different for different students.

For instance, passing grade inflation is more relevant to students who are likely on the margin of

passing a class (holding mean grade inflation fixed). Additionally, mean grade inflation might

reduce human capital formation to a larger degree for top students who can reduce their effort

and still receive an A. On the other hand, the incentive effects induced by mean grade inflation

likely apply to all students, since students now need to study less to earn a B as well.

We test the hypothesis that mean and passing grade inflation impact different students dif-

ferently by estimating Equation 2 on separate samples split by a variety of student and school

characteristics. In Tables 11 through 13, we explore whether lower-achieving students are more

or less impacted by grade inflation than higher-achieving students. We use 8th grade GPA as

a rough proxy for achievement that is, importantly, measured before students enter high school

and therefore not impacted by high school teacher grade inflation. Panel C of the table presents

p-values from tests of the difference between the below-median and above-median estimates for

each measure and outcome.

First, we find that that mean grade inflation reduces future test scores for both above- and

below-median students. Our tests, with p-values between 0.38 to 0.97 in Los Angeles and 0.17 to

0.77 in Maryland, give no evidence that the estimated effect of grade inflation on future test scores

differs by student achievement. We conclude, therefore, that the reduction in human capital due

to mean grade inflation occurs at all parts of the student achievement distribution. Mean grade in-

flation also does not have a differential effect on the SAT score, but does for the likelihood of taking

the SAT. In Los Angeles we estimate that mean grade inflation is associated with larger declines in

SAT test-taking for above-median students (p-value: 0.06) while in Maryland we estimate that the

declines in SAT test-taking are larger for below-median students (p-value: 0.0001).

Second, we find heterogeneity in the effects of grade inflation on graduation and persistence

among students of different abilities. Our tests suggest that the estimated effects of passing grade

28
inflation on grade retention, with p-values of 0.2 in Los Angeles and 0.0001 in Maryland, and

graduation, with p-values of 0.01 in Los Angeles and 0.02 in Maryland, are slightly larger for below-

median students than for above-median students. This lines up with intuition since above-median

students are not close to the passing margin. We also estimate that mean grade inflation reduces

grade retention more among below-median students, with p-values of 0.017 in Los Angeles and

0.0002 in Maryland. Looking at longer-term outcomes, the main difference in the effect across the

academic achievement distribution is in enrollment in postsecondary education: we find that mean

grade inflation reduces enrollment more among below-median students, with p-values from 0.007

to 0.010. However, the effect does not persist into graduation or employment, where we see no

statistically different effects of grade inflation for below- and above-median students.

We examine several other dimensions of heterogeneity. First, we explore heterogeneous effects

by school-level grading patterns, proxied by school average GPA, to account for different grading

contexts. Second, we explore heterogeneity by student characteristics in Maryland.35 We split the

sample by English Language Learner status, Free and Reduced Price Meal status, gender, and race.

We do not find strong evidence of heterogeneity across either the school or student characteristics36 .

4.4 Robustness

Carrying out our analysis in both Los Angeles and Maryland provides an effective robustness check

for our main results. These two settings differ in many ways, including the demographic makeup

of students, curriculum, and the amount of funding, and yet we find similar patterns. Between the

two, we have evidence from a large urban school district on the West Coast with a large Hispanic

population, and a populous East Coast state with a range of school districts from small rural areas

to larger urban areas. We find that mean and passing grade inflation behave similarly in both

LAUSD and Maryland, with mean grade inflation reducing high school achievement and passing

grade inflation showing evidence of benefits for high school grade retention and graduation. We

view this as the strongest evidence that our results are not the result of a particular institutional

context.

As a supplement to carrying out our analysis in two settings, we explore the robustness of
35
The LAUSD data does not contain the rich student demographics necessary for this analysis.
36
The results of these heterogeneity analyses are all available upon request.

29
our specification in two additional ways. First, we explore whether our choice of controls impacts

our results. In Maryland, we have rich demographic data on students, which we use to augment

our controls in both the construction of the value-added and grade inflation measures as well as

the estimation of the effects of those measures on high school, college, and career outcomes. As

Tables A.15 and A.16 show, the estimates remain very similar when we control for these additional

student characteristics.

Second, we explore whether our results are consistent across math and English courses. Our

main analysis is based on a combined sample of math and English teachers. We create the value-

added and grade inflation measures separately for math and English teachers, then combine them

to form a single measure of the average cognitive value-added, noncognitive value-added, and

grade inflation experienced by a given student over all math and English courses in a given year

(see Section 2 for more details). Because in high school teachers specialize in one subject and

not all students take math and English courses in all years, combining the measures across math

and English greatly increases our sample size. However, as a robustness check, we also carry out

analysis with measures of grade inflation and value-added that are specific to math or English. The

main results from these analysis are found in Tables A.8, A.9, and A.10. We find generally similar

patterns, but with slightly less precision. The effects of grade inflation do not appear to be driven

primarily by either math or English teachers.

By testing the robustness of our results to different school districts, different data, different

model specifications, and different subjects, we have shown that the relationship between grade

inflation and student outcomes in the short and long term is unlikely to be driven by any particular

choice made by the research team.

5 Conclusion

In this paper, we demonstrate that teachers’ grading practices affect students in important and

heterogeneous ways. A teacher who generally inflates grades has a negative effect on students’

future performance as measured by test scores, high school graduation, college enrollment, and

earnings. In contrast, a teacher who inflates grades so that students are less likely to fail improves

high school graduation rates. The magnitude of these grade inflation effects varies depending on

30
student characteristics and whether a student is on the margin of passing a course.

Our results align with the concept that grades serve as a potent motivator for educational

engagement. In light of increasing grade inflation in the United States, our findings suggest that

students may be learning less as a result. While our paper does not directly address the change

in grading practice over time, our results highlight the potential detrimental impact of increasing

grades over time.

Overall, our results demonstrate an additional channel by which teachers affect their students’

future outcomes–grading practices. While most teacher training programs that improve teacher

quality are high-cost and time-consuming (Taylor and Tyler, 2012), changes in grading policies are

a potential low-cost strategy that could be implemented by teachers, schools, and school districts

to improve teacher quality and students’ long-term outcomes.

31
References

Alex, P. (2022). Time to pull the plug on traditional grading? Education Next, 22(4).

Ayllón, S., Lefgren, L. J., Patterson, R. W., Stoddard, O. B., and Urdaneta, N. (2025). ‘sorting’

out gender discrimination and disadvantage: Evidence from student evaluations of teaching.

Working Paper 33911, National Bureau of Economic Research.

Betts, J. R. and Grogger, J. (2003). The impact of grading standards on student achievement,

educational attainment, and entry-level earnings. Economics of Education Review, 22(4):343–352.

Black, S. E., Denning, J. T., and Rothstein, J. (2023). Winners and losers? the effect of gaining and

losing access to selective colleges on education and labor market outcomes. American Economic

Journal: Applied Economics, 15(1):26–67.

Bowden, A. B., Rodriguez, V., and Weingarten, Z. (2023). The unintended consequences of

academic leniency. EdWorkingPaper, pages 23–836.

Carlana, M. (2019). Implicit stereotypes: Evidence from teachers’ gender bias. The Quarterly Journal

of Economics, 134(3):1163–1224.

Chetty, R., Friedman, J. N., and Rockoff, J. E. (2014a). Measuring the impacts of teachers I:

Evaluating bias in teacher value-added estimates. American economic review, 104(9):2593–2632.

Chetty, R., Friedman, J. N., and Rockoff, J. E. (2014b). Measuring the impacts of teachers II: Teacher

value-added and student outcomes in adulthood. American economic review, 104(9):2633–79.

Costrell, R. M. (1994). A simple model of educational standards. The American Economic Review,

84(4):956–971.

Denning, J. T., Eide, E. R., Mumford, K. J., Patterson, R. W., and Warnick, M. (2022). Why have

college completion rates increased? American Economic Journal: Applied Economics, 14(3):1–29.

Figlio, D. N. and Lucas, M. E. (2004). Do high grading standards affect student performance?

Journal of Public Economics, 88(9-10):1815–1834.

Gershenson, S. (2018). Grade inflation in high schools (2005-2016). Thomas B. Fordham Institute.

Gershenson, S., Holt, S. B., and Tyner, A. (2022). Making the grade: The effect of teacher grading

standards on student outcomes. Contemporary Economic Policy.

Graham, K. A. (2017). Does Philly’s new grading policy level the playing field or

lower standards? — [Link]. [Link]

32
 does-phillys-new-grading-policy-level-the-playing-field-or-lower-standards-20170908.

html. [Accessed 03-10-2023].

Griffith, D. and Tyner, A. (2025). ‘Equitable’ Grading Through the Eyes of Teachers. Technical

report, Thomas B. Fordham Institute, Washington, DC.

Hurwitz, M. and Lee, J. (2018). Grade inflation and the role of standardized testing. Measuring

success: Testing, grades, and the future of college admissions, pages 64–93.

Insler, M., McQuoid, A. F., Rahman, A. S., and Smith, K. (2021). Fear and loathing in the classroom:

Why does teacher quality matter? Technical report, IZA Discussion Papers.

Jackson, C. K. (2018). What do test scores miss? the importance of teacher effects on non–test score

outcomes. Journal of Political Economy, 126(5):2072–2107.

Jackson, C. K., Kiguel, S., Porter, S. C., and Easton, J. Q. (2024). Who benefits from attending

effective high schools? Journal of Labor Economics, 42(3):717–751.

Jacob, B. A., Lefgren, L., and Sims, D. P. (2010). The persistence of teacher-induced learning. Journal

of Human resources, 45(4):915–943.

Kane, T. J. and Staiger, D. O. (2008). Estimating teacher impacts on student achievement: An

experimental evaluation. Working Paper 14607, National Bureau of Economic Research.

Koedel, C. and Rockoff, J. E. (2015). Value-added modeling: A review. Economics of Education

Review, 47:180–195.

Kohn, A. (2002). The dangerous myth of grade inflation. The chronicle of higher education, 49(11):B7.

Landaud, F., Maurin, É., Willage, B., and Willén, A. (2024). The value of a high school gpa. Review

of Economics and Statistics, pages 1–24.

Las Vegas Review-Journal Editorial (2023). EDITORIAL: ‘Too dumb to fail’ grading policies aren’t

working — [Link]. [Link]

editorial-too-dumb-to-fail-grading-policies-arent-working-2773256/. [Accessed 03-

10-2023].

Mozenter, Z. D. (2019). Essays on the Effects of Teacher Grading Standards and Other Teaching Practices.

PhD thesis, The University of North Carolina at Chapel Hill.

Papageorge, N. W., Gershenson, S., and Kang, K. M. (2020). Teacher expectations matter. Review

of Economics and Statistics, 102(2):234–251.

Petek, N. and Pope, N. G. (2023). The multidimensional impact of teachers on students. Journal of

33
 Political Economy.

Randazzo, S. (2023). Schools Are Ditching Homework, Deadlines in Fa-

vor of ‘Equitable Grading’ — [Link]. [Link]

schools-are-ditching-homework-deadlines-in-favor-of-equitable-grading-dcef7c3e.

[Accessed 03-10-2023].

Sanchez, E. I. and Moore, R. (2022). Grade inflation continues to grow in the past decade. ACT,

Inc.

Staiger, D. O. and Rockoff, J. E. (2010). Searching for effective teachers with imperfect information.

Journal of Economic Perspectives, 24(3):97–118.

Taylor, E. S. and Tyler, J. H. (2012). The effect of evaluation on teacher performance. American

Economic Review, 102(7):3628–3651.

Wright, B. (2019). The dangerous myth of grade inflation. Thomas Fordham B. Institute.

Zhang, Q. and Sanchez, E. I. (2013). High school grade inflation from 2004 to 2011. act research

report series, 2013 (3). ACT, Inc.

34
6 Tables and Figures

Figure 1: Average GPA over Time

Notes: This figure plots average total GPA over time for students in our Los Angeles and
Maryland samples. The data from the LAUSD span 2004 to 2013 and the data from Maryland
span 2013 to 2023. GPA is measured on a 0 to 4 scale (e.g., A = 4.0, B = 3.0).

35
 Table 1: Student-level Summary Statistics in Los Angeles

Mean Std. Dev. N

Math CST Score 0.06 1.02 706,922
English CST Score 0.10 1.00 752,431
Math GPA 1.75 1.24 910,194
English GPA 2.18 1.20 985,020
Math Pass Rate 0.68 0.46 986,396
English Pass Rate 0.79 0.41 1,086,116
GPA 2.29 0.98 985,020
Effort GPA 2.16 0.53 985,020
Cooperation GPA 2.39 0.45 985,020
Fraction of Days Absent 0.07 0.09 985,020
Ever Suspended 0.06 0.24 985,020
Held Back 0.12 0.32 985,020
English Learner 0.20 0.40 985,020
Average Teacher Experience 6.48 2.83 985,020
Don’t Graduate in LAUSD 0.31 0.46 733,949
Leave Dataset 0.27 0.45 733,949
Graduate on Time 0.49 0.50 833,266
Graduate within 5 Years 0.58 0.49 733,949
Number of AP Courses 1.11 2.04 985,020
Ever Took SAT 0.36 0.48 833,266
Ever Took PSAT 0.43 0.49 646,242
SAT Score 1335.38 299.80 363,022
PSAT Score 1114.21 258.12 393,366
English CAHSEE Score 0.13 0.99 854,094
Math CAHSEE Score 0.12 1.01 856,998
10th Grade Science CST Score 0.12 1.01 656,977
11th Grade Social CST Score 0.11 1.01 674,673
Notes: This table reports summary statistics (mean, standard deviation, and number of
observations) for all high school students enrolled in LAUSD between 2004-2013. The sample
is restricted to students with non-missing test scores, grades, and behavioral outcomes. CST
scores (end of course standardized tests) are standardized before making sample restrictions.
“Held back” is measured as whether we see the student with the same administrative grade
code in the following year. “Average teacher experience” is the average number of years a
student’s teachers have taught for. “Don’t graduate in LAUSD” and “Leave Dataset” capture
the degree to which the graduation outcomes are censored by the sample time period ending
or students moving outside of LAUSD. “CAHSEE” refers to the California High School Exit
Examination administered from 2006 to 2014.

36
 Table 2: Student-level Summary Statistics in Maryland

Mean Std. Dev. N

Math Test Score 0.03 0.98 322,700
English Test Score 0.04 0.97 356,336
Math GPA 2.59 1.08 1,542,447
English GPA 2.76 1.05 1,647,988
Math Pass Rate 0.97 0.17 1,692,638
English Pass Rate 0.98 0.14 1,786,392
GPA 2.89 0.85 1,647,988
Frac. Days Absent 0.07 0.10 1,647,988
Suspended 0.05 0.21 1,647,988
Held Back 0.04 0.19 1,541,341
English Learner 0.03 0.16 1,647,988
Free and Reduced Price Meals 0.34 0.47 1,647,988
Black 0.34 0.47 1,647,988
White 0.41 0.49 1,647,988
Asian 0.07 0.26 1,647,988
Hispanic 0.14 0.34 1,647,988
Female or Non-Binary 0.50 0.50 1,647,988
AP Classes 3.51 5.04 1,647,988
Honors Classes 8.34 8.16 1,647,988
Ever Took SAT 0.66 0.47 1,587,134
Ever Took SAT or ACT 0.67 0.47 1,647,988
Ever Took PSAT 0.29 0.45 1,596,066
SAT Score 1174.54 315.67 410,547
PSAT Score 128.38 32.43 156,007
Graduate On Time 0.86 0.35 1,587,134
Graduate within 5 Years 0.92 0.27 1,403,078
Enrolled in Higher Ed 1 year Post HS 0.65 0.48 1,403,078
Enrolled in Higher Ed 2 years Post HS 0.58 0.49 1,216,803
Graduated from Higher Ed 4 years Post HS 0.27 0.45 837,086
Graduated from Higher Ed 6 years Post HS 0.41 0.49 461,864
Employed 6 years Post HS 0.68 0.47 1,403,078
Employed 6 years after Expected HS Graduation 0.62 0.48 461,864
Earnings 1 year Post HS 7901.18 7755.60 954,844
Earnings 6 years Post HS 27789.25 21293.59 288,268

Notes: This table reports summary statistics (mean, standard deviation, and number of obser-
vations) for all high school students enrolled in Maryland between 2013-2023. The sample is
restricted to students with non-missing test scores, grades, and behavioral outcomes. Math tests
are from Algebra I, Algebra II, and Geometry courses; English tests are from grades 9 and 10. Test
scores are standardized before making sample restrictions. “Held back” is measured as whether
we see the student with the same administrative grade code in the following year. The timing of
the college and career outcomes are measured relative to expected high school graduation as of
9th grade, so “1 year post HS” refers to 5 years after we see the student in 9th grade. Earnings are
conditional on having observed any earnings and are winsorized at the 99th percentile.

37
 Table 3: Forecast Bias Tests of Grade Inflation

Math ELA
Panel A: Individual Test, LAUSD Course Grade Pass Indicator Course Grade Pass Indicator
Corresponding GI Measure 1.07 1.07 1.03 1.04
(0.01) (0.02) (0.01) (0.01)
N 474,686 474,839 545,631 545,773
Panel B: Individual Test, Maryland Course Grade Pass Indicator Course Grade Pass Indicator
Corresponding GI Measure 1.08 1.11 1.03 0.96
(0.02) (0.07) (0.02) (0.21)
N 204,994 204,994 250,734 250,734
Panel C: Prediction Test, LAUSD Course Grade Pass Indicator Course Grade Pass Indicator
Corresponding GI Measure -0.02 -0.01 -0.00 0.00
(0.00) (0.00) (0.00) (0.00)
N 538,525 538,542 603,581 603,585

Notes: This table presents results from two different forecast bias tests (the second test is only
practical in LAUSD where there are more frequent standardized tests). For the “individual” test,
an estimate close to 1 shows that these measures, though subject to estimation error, are actually
predictive of student outcomes out of sample. For the ‘prediction’ test, an estimate close to 0 shows
that these measures are not biased by sorting on omitted variables. Standard errors are clustered
at the school level. * p < 0.1, ** p < 0.05, *** p < 0.01.

38
Table 4: Correlations between Grade Inflation and Value-Added Mea-
sures

Panel A: LAUSD Mean GI Passing GI Cog. VA Noncog. VA

Mean GI 1.0000 . . .
Passing GI 0.8599 1.0000 . .
Cog. VA -0.4070 -0.3047 1.0000 .
Noncog. VA 0.1550 0.1787 0.0879 1.0000
Panel B: Maryland Mean GI Passing GI Cog. VA Noncog. VA
Mean GI 1.0000 . . .
Passing GI 0.3601 1.0000 . .
Cog. VA -0.3108 -0.0743 1.0000 .
Noncog. VA 0.1163 0.0542 -0.1241 1.0000
Notes: Within-teacher-year correlations are estimated using an error correction pro-
cedure based on randomly splitting the data within classroom. Reported correlations
are the average across 200 bootstrapped estimates. The grade inflation and value-added
measures used in estimating these correlations are constructed without shrinkage.

39
 Table 5: Regressions of Single Teacher Measures on High School Outcomes

Panel A: LAUSD Future Test (math) Future Test (ela) Held Back Graduate in 5 Years Took SAT SAT Score
Mean GI -0.052*** -0.029*** -0.007*** -0.004** -0.009*** -11.234***
(0.007) (0.003) (0.001) (0.002) (0.001) (2.629)
Passing GI -0.043*** -0.022*** -0.009*** -0.001 -0.007*** -10.919***
(0.007) (0.003) (0.001) (0.001) (0.001) (2.824)
Cog. VA 0.114*** 0.058*** 0.001 0.005* 0.030*** 25.004***
(0.011) (0.002) (0.001) (0.003) (0.003) (3.257)
Noncog. VA 0.008 0.016*** -0.006*** 0.012*** 0.013*** 0.594
(0.007) (0.003) (0.001) (0.002) (0.002) (1.508)
Outcome Mean 0.115 0.235 0.106 0.645 0.405 1,368.568
Observations 199,038. 215,559. 411,275. 327,528. 323,635. 173,195.
Panel B: Maryland Future Test (math) Future Test (ela) Held Back Graduate in 5 Years Took SAT SAT Score
Mean GI -0.003 -0.011*** -0.003*** -0.000 -0.004*** -1.115
(0.004) (0.003) (0.001) (0.001) (0.001) (1.244)
Passing GI -0.003 -0.003 -0.004*** 0.001+ -0.000 -3.051**
(0.004) (0.003) (0.001) (0.001) (0.002) (1.335)
Cog. VA 0.016*** 0.033*** 0.001** 0.001+ 0.005*** -0.521
(0.004) (0.003) (0.000) (0.000) (0.001) (1.300)
Noncog. VA 0.007+ 0.015*** -0.004*** 0.007*** 0.003+ -6.423***
(0.004) (0.003) (0.001) (0.001) (0.002) (1.517)
Outcome Mean -0.20 0.03 0.04 0.90 0.65 1158.08
Observations 123,205 279,802 1,117,227 979,425 1,117,945 252,950

Notes: This table presents estimated coefficients from regressions of the relevant outcome (column) on the relevant
teacher measure (row) and our main vector of controls. Thus, each cell reports an estimate from a different
regression. For this student-year level analysis, value-added and grade inflation are measured as averages across
the student’s math and English instructors in each year. The controls include school, grade, and year fixed effects,
ELL, and previous year values of math test score, English test score, total GPA, fraction of days absent, suspension,
and held back. Standard errors are clustered at the school level. * p < 0.1, ** p < 0.05, *** p < 0.01.

40
 Table 6: Effects of Teacher Grade Inflation and Value-Added on High School Outcomes

Panel A: LAUSD Future Test (math) Future Test (ela) Held Back Graduate in 5 Years Took SAT SAT Score
Mean GI -0.024*** -0.020*** 0.002 -0.008*** -0.005+ -1.546
(0.006) (0.003) (0.002) (0.003) (0.003) (1.790)
Passing GI -0.001 0.003 -0.012*** 0.006** 0.003 -3.315+
(0.006) (0.003) (0.002) (0.003) (0.002) (1.896)
Cog. VA 0.113*** 0.054*** 0.001 -0.000 0.029*** 24.748***
(0.010) (0.002) (0.001) (0.002) (0.003) (2.919)
Noncog. VA -0.004 0.011*** -0.005*** 0.011*** 0.010*** 1.625+
(0.004) (0.002) (0.001) (0.002) (0.001) (0.938)
Outcome Mean -0.02 0.04 0.13 0.58 0.32 1327.79
Observations 391,782 432,534 832,002 733,946 680,305 186,350
R2 0.522 0.659 0.167 0.296 0.364 0.738
Panel B: Maryland Future Test (math) Future Test (ela) Held Back Graduate in 5 Years Took SAT SAT Score
Mean GI 0.001 -0.006+ -0.001*** -0.001** -0.004*** -0.427
(0.004) (0.003) (0.000) (0.000) (0.001) (1.262)
Passing GI -0.003 0.002 -0.004*** 0.002** 0.001 -2.953**
(0.004) (0.003) (0.001) (0.001) (0.002) (1.416)
Cog. VA 0.015*** 0.031*** 0.001** 0.000 0.005*** -0.509
(0.004) (0.003) (0.000) (0.000) (0.001) (1.270)
Noncog. VA 0.006 0.013*** -0.004*** 0.007*** 0.003 -6.432***
(0.004) (0.003) (0.001) (0.001) (0.002) (1.512)
Outcome Mean -0.20 0.03 0.04 0.90 0.65 1158.08
Observations 123,205 279,802 1,117,227 979,425 1,117,945 252,950
R2 0.356 0.592 0.154 0.239 0.242 0.772

Notes: This table presents estimates from regressions of the relevant outcome (column) on mean grade inflation,
passing grade inflation, cognitive value-added, and noncognitive value-added, as well as our main vector of
controls. For this student-year level analysis, value-added and grade inflation are measured as averages across
the student’s math and English instructors in each year. The controls include school, grade, and year fixed effects,
ELL, and previous year values of math test score, English test score, total GPA, fraction of days absent, suspension,
and held back. Standard errors are clustered at the school level. * p < 0.1, ** p < 0.05, *** p < 0.01.

41
 Table 7: Effects of Teacher Grade Inflation and Value-Added on College and Career Outcomes in
Maryland

Enrolled in Postsecondary Graduated from Postsecondary Employed Unconditional Winz. Earnings

1 year 2 years 4 years 6 years 1 year 6 years 1 year 6 years
Mean GI -0.006*** -0.005*** 0.000 -0.002 -0.004*** -0.003** -49.052*** -165.851***
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (12.232) (51.212)
Passing GI 0.002** 0.001 -0.002** -0.003** 0.001 0.000 12.637 -10.113
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (11.698) (41.285)
Cog. VA 0.001 0.001 -0.003*** -0.001 -0.001 -0.001 -33.437*** -47.525
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (12.618) (41.879)
42

Noncog. VA 0.007*** 0.006*** -0.001 0.001 0.002*** 0.002** 32.501*** 40.429

(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (9.593) (44.426)
Outcome Mean 0.64 0.56 0.27 0.40 0.68 0.62 5444.70 17801.59
Observations 979,425 838,616 552,566 274,354 979,425 274,354 979,425 274,354
R2 0.287 0.279 0.237 0.338 0.037 0.050 0.080 0.041

Notes: This table presents estimates from regressions of the relevant outcome (column) on mean grade inflation, passing
grade inflation, cognitive value-added, and noncognitive value-added, as well as our main vector of controls. The outcomes
in this table are only available in the Maryland data, not in LAUSD. For this student-year level analysis, value-added and grade
inflation are measured as averages across the student’s math and English instructors in each year. The controls include school,
grade, and year fixed effects, ELL, and previous year values of math test score, English test score, total GPA, fraction of days
absent, suspension, and held back. Standard errors are clustered at the school level. * p < 0.1, ** p < 0.05, *** p < 0.01.
 Table 8: Dynamic Effects of Grade Inflation on College Enrollment in Maryland

Panel A: Enrolled in Associate’s X Years Post Expected HS Graduation

X=1 X=2 X=3 X=4 X=5 X=6 X=7 X=8
Mean GI -0.004*** -0.004*** -0.003*** -0.002*** -0.001*** -0.001*** -0.001** -0.000
(0.001) (0.001) (0.001) (0.001) (0.000) (0.000) (0.000) (0.001)
Passing GI 0.003*** 0.002*** 0.001** 0.001** 0.001+ 0.000 -0.000 -0.001
(0.001) (0.001) (0.001) (0.000) (0.000) (0.000) (0.001) (0.000)
Outcome Mean 0.24 0.21 0.15 0.10 0.07 0.05 0.04 0.03
Observations 1,403,076 1,216,801 1,026,593 837,085 652,118 461,860 276,317 135,233
R2 0.032 0.031 0.022 0.015 0.010 0.009 0.007 0.008
Panel B: Enrolled in Bachelor’s X Years Post Expected HS Graduation
X=1 X=2 X=3 X=4 X=5 X=6 X=7 X=8
Mean GI -0.001 -0.001+ -0.002** -0.002** -0.003*** -0.002*** -0.000 -0.000
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Passing GI -0.001 -0.001 -0.001 -0.001+ 0.000 0.001 0.000 0.000
(0.001) (0.001) (0.001) (0.001) (0.001) (0.000) (0.000) (0.001)
Outcome Mean 0.22 0.22 0.24 0.25 0.13 0.07 0.05 0.03
Observations 1,403,076 1,216,801 1,026,593 837,085 652,118 461,860 276,317 135,233
R2 0.137 0.144 0.155 0.145 0.032 0.012 0.008 0.007
Panel C: Enrolled in Any College X Years Post Expected HS Graduation
X=1 X=2 X=3 X=4 X=5 X=6 X=7 X=8
Mean GI -0.005*** -0.004*** -0.005*** -0.005*** -0.005*** -0.003*** -0.002*** -0.001
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Passing GI 0.001 0.000 0.000 -0.000 0.001 0.000 -0.000 0.000
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Outcome Mean 0.65 0.58 0.50 0.49 0.29 0.16 0.12 0.10
Observations 1,403,076 1,216,801 1,026,593 837,085 652,118 461,860 276,317 135,233
R2 0.285 0.278 0.254 0.268 0.069 0.025 0.017 0.014

Notes: This table presents estimates from regressions of being enrolled in the relevant postsecondary
degree program (panel), measured X years (column) after expected high school graduation, on mean
grade inflation and passing grade inflation as well as cognitive value-added, noncognitive value-added,
and our main vector of controls. The outcomes in this table are only available in the Maryland data,
not in LAUSD. The timing of enrollment is measured relative to expected high school graduation as of
9th grade; for example, “1 year post HS” refers to 5 years after we see the student in 9th grade. For
this student-year level analysis, value-added and grade inflation are measured as averages across the
student’s math and English instructors in each year. The controls include school, grade, and year fixed
effects, ELL, and previous year values of math test score, English test score, total GPA, fraction of days
absent, suspension, and held back. Standard errors are clustered at the school level. * p < 0.1, ** p <
0.05, *** p < 0.01.

43
Table 9: Dynamic Effects of Grade Inflation on College Graduation in
Maryland

Panel A: Graduated with Associate’s X Years Post Expected HS Graduation

X=4 X=5 X=6
Mean GI -0.001*** -0.002*** -0.002***
(0.000) (0.001) (0.001)
Passing GI 0.001 0.001+ 0.001
(0.000) (0.000) (0.001)
Outcome Mean 0.08 0.09 0.10
Observations 837,085 652,118 461,860
R2 0.044 0.041 0.041
Panel B: Graduated with Bachelor’s X Years Post Expected HS Graduation
X=4 X=5 X=6
Mean GI 0.001 0.000 -0.000
(0.001) (0.001) (0.001)
Passing GI -0.002*** -0.003** -0.003**
(0.001) (0.001) (0.001)
Outcome Mean 0.20 0.31 0.35
Observations 837,085 652,118 461,860
R2 0.242 0.325 0.341
Panel C: Graduated from Any College X Years Post Expected HS Graduation
X=4 X=5 X=6
Mean GI -0.000 -0.001 -0.002
(0.001) (0.001) (0.001)
Passing GI -0.002+ -0.002 -0.002+
(0.001) (0.001) (0.001)
Outcome Mean 0.27 0.37 0.41
Observations 837,085 652,118 461,860
R2 0.239 0.324 0.339
Notes: This table presents estimates from regressions of graduating from the relevant
postsecondary degree program (panel), measured X years (column) after expected
high school graduation, on mean grade inflation and passing grade inflation as well
as cognitive value-added, noncognitive value-added, and our main vector of controls.
The outcomes in this table are only available in the Maryland data, not in LAUSD. The
timing of college graduation is measured relative to expected high school graduation
as of 9th grade; for example, “1 year post HS” refers to 5 years after we see the student
in 9th grade. For this student-year level analysis, value-added and grade inflation are
measured as averages across the student’s math and English instructors in each year.
The controls include school, grade, and year fixed effects, ELL, and previous year values
of math test score, English test score, total GPA, fraction of days absent, suspension, and
held back. Standard errors are clustered at the school level. * p < 0.1, ** p < 0.05, *** p
< 0.01.

44
 Table 10: Dynamic Effects of Grade Inflation on Labor Market Outcomes in Maryland

Panel A: Unconditional Winz. Earnings X Years Post Expected HS Graduation

X=1 X=2 X=3 X=4 X=5 X=6 X=7 X=8
Mean GI -42.286*** -47.760*** -71.814*** -74.685*** -50.079+ -133.192*** -89.455+ -32.829
(10.820) (15.545) (19.821) (23.099) (27.227) (40.489) (46.940) (79.257)
Passing GI 13.452 22.284 47.988** 54.321** 19.158 4.432 -7.119 63.311
(12.857) (16.902) (19.031) (24.755) (27.346) (38.167) (45.439) (89.074)
Outcome Mean 5377.04 7486.49 8765.06 10121.78 14222.43 17344.43 19799.54 22105.47
Observations 1,403,076 1,216,801 1,026,593 837,085 652,118 461,860 276,317 135,233
R2 0.083 0.084 0.079 0.069 0.038 0.043 0.045 0.039
Panel B: Employed 2 or More Quarters X Years Post Expected HS Graduation
X=1 X=2 X=3 X=4 X=5 X=6 X=7 X=8
Mean GI -0.003*** -0.003*** -0.004*** -0.003*** -0.001 -0.002** -0.001 0.001
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002)
Passing GI 0.002** 0.002** 0.003*** 0.002** 0.001 0.001 0.000 0.001
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002)
Outcome Mean 0.40 0.45 0.46 0.45 0.48 0.49 0.49 0.48
Observations 1,403,076 1,216,801 1,026,593 837,085 652,118 461,860 276,317 135,233
R2 0.060 0.066 0.070 0.066 0.046 0.048 0.051 0.049
Panel C: Any Employment X Years Post Expected HS Graduation
X=1 X=2 X=3 X=4 X=5 X=6 X=7 X=8
Mean GI -0.003*** -0.002*** -0.002*** -0.002*** -0.001 -0.002** -0.000 0.002
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002)
Passing GI 0.001 0.001 0.001 0.001 0.000 0.000 0.001 -0.000
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002)
Outcome Mean 0.68 0.70 0.68 0.65 0.65 0.62 0.61 0.59
Observations 1,403,076 1,216,801 1,026,593 837,085 652,118 461,860 276,317 135,233
R2 0.039 0.037 0.041 0.045 0.044 0.051 0.055 0.056

Notes: This table presents estimates from regressions of labor market outcomes (panel), measured X years
(column) after expected high school graduation, on mean grade inflation and passing grade inflation as well
as cognitive value-added, noncognitive value-added, and our main vector of controls. The outcomes in this
table are only available in the Maryland data, not in LAUSD. Earnings are winsorized at the 99th percentile.
The timing of college graduation is measured relative to expected high school graduation as of 9th grade; for
example, “1 year post HS” refers to 5 years after we see the student in 9th grade. For this student-year level
analysis, value-added and grade inflation are measured as averages across the student’s math and English
instructors in each year. The controls include school, grade, and year fixed effects, ELL, and previous year
values of math test score, English test score, total GPA, fraction of days absent, suspension, and held back.
Standard errors are clustered at the school level. * p < 0.1, ** p < 0.05, *** p < 0.01.

45
Table 11: Effects of Teacher Measures Among Low- and High-Achieving Students in LAUSD

Panel A: Below Median

Future Test (math) Future Test (ela) Held Back Graduate in 5 Years Took SAT SAT Score
Mean GI -0.026*** -0.022*** -0.004 -0.007*** -0.000 -0.535
(0.006) (0.005) (0.003) (0.003) (0.002) (1.535)
Passing GI 0.001 0.003 -0.012*** 0.007*** 0.001 -4.821***
(0.005) (0.005) (0.003) (0.002) (0.002) (1.705)
Outcome Mean -0.33 -0.31 0.23 0.47 0.14 1213.85
Observations 161,249 181,382 343,437 263,059 272,601 55,051
R2 0.324 0.522 0.164 0.289 0.223 0.651
Panel B: Above Median
Future Test (math) Future Test (ela) Held Back Graduate in 5 Years Took SAT SAT Score
Mean GI -0.026*** -0.018*** 0.002+ -0.005** -0.005+ -0.955
(0.007) (0.003) (0.001) (0.002) (0.003) (1.854)
Passing GI 0.001 0.004 -0.009*** 0.001 0.003 -3.341+
(0.007) (0.003) (0.001) (0.002) (0.003) (1.929)
Outcome Mean 0.27 0.40 0.05 0.81 0.51 1371.51
Observations 189,020 202,392 337,968 283,415 268,546 178,682
R2 0.568 0.683 0.082 0.181 0.287 0.752
Panel C: P-values from Tests of Coefficient Equality
Future Test (math) Future Test (ela) Held Back Graduate in 5 Years Took SAT SAT Score
Mean GI 0.9677 0.3786 0.0178 0.3330 0.0562 0.8006
Passing GI 0.9658 0.8216 0.2015 0.0103 0.3594 0.3417

Notes: This table presents estimates from regressions of the relevant outcome (column) on mean grade inflation,
passing grade inflation, cognitive value-added, and noncognitive value-added, as well as our main vector of
controls. The sample is split into students whose 8th grade GPA is below and above median as a proxy for low-
and high-achieving students. For this student-year level analysis, value-added and grade inflation are measured
as averages across the student’s math and English instructors in each year. Panel C shows p-values of tests for
coefficient equality that were constructed using seemingly unrelated regressions. The controls include school,
grade, and year fixed effects, ELL, and previous year values of math test score, English test score, total GPA,
fraction of days absent, suspension, and held back. Standard errors are clustered at the school level. * p < 0.1,
** p < 0.05, *** p < 0.01.

46
Table 12: Effects of Teacher Measures Among Low- and High-Achieving Students in Mary-
land

Panel A: Below Median

Future Test (math) Future Test (ela) Held Back Graduate in 5 Years Took SAT SAT Score
Mean GI -0.017*** -0.007** -0.003*** -0.000 -0.004** -2.751***
(0.005) (0.003) (0.001) (0.001) (0.002) (0.963)
Passing GI -0.003 0.002 -0.005*** 0.002** 0.001 -1.162
(0.003) (0.004) (0.002) (0.001) (0.002) (0.865)
Outcome Mean -0.33 -0.36 0.08 0.87 0.51 958.20
Observations 130,241 128,690 461,942 527,128 462,508 71,299
R2 0.524 0.481 0.160 0.249 0.203 0.645
Panel B: Above Median
Future Test (math) Future Test (ela) Held Back Graduate in 5 Years Took SAT SAT Score
Mean GI -0.016*** -0.005 -0.001** -0.000 -0.003** 1.190
(0.005) (0.004) (0.000) (0.000) (0.001) (1.118)
Passing GI 0.001 0.001 -0.002+ 0.001 0.002 -1.250
(0.005) (0.004) (0.001) (0.001) (0.002) (1.027)
Outcome Mean 0.45 0.41 0.01 0.96 0.77 1146.36
Observations 134,573 137,366 497,777 557,599 497,894 116,390
R2 0.624 0.604 0.125 0.147 0.156 0.707
Panel C: P-values from Tests of Coefficient Equality
Future Test (math) Future Test (ela) Held Back Graduate in 5 Years Took SAT SAT Score
Mean GI 0.1690 0.5521 0.0002 0.8490 0.0001 0.7023
Passing GI 0.6052 0.7727 0.0001 0.0220 0.9266 0.2160

Notes: This table presents estimates from regressions of the relevant outcome (column) on mean grade inflation,
passing grade inflation, cognitive value-added, and noncognitive value-added, as well as our main vector of
controls. The sample is split into students whose 8th grade GPA is below and above median as a proxy for low-
and high-achieving students. For this student-year level analysis, value-added and grade inflation are measured
as averages across the student’s math and English instructors in each year. Panel C shows p-values of tests for
coefficient equality that were constructed using seemingly unrelated regressions. The controls include school,
grade, and year fixed effects, ELL, and previous year values of math test score, English test score, total GPA,
fraction of days absent, suspension, and held back. Standard errors are clustered at the school level. * p < 0.1,
** p < 0.05, *** p < 0.01.

47
 Table 13: Effects of Teacher Measures Among Low- and High-Achieving Students in Maryland

Panel A: Below Median

Enrolled in Postsecondary Graduated from Postsecondary Employed Unconditional Winz. Earnings
1 year 2 years 4 years 6 years 1 year 6 years 1 year 6 years
Mean GI -0.005*** -0.005*** -0.001 -0.002 -0.004*** -0.001 -57.674*** -92.938
(0.001) (0.001) (0.001) (0.002) (0.001) (0.002) (16.326) (84.959)
Passing GI 0.000 -0.001 -0.001+ -0.004*** 0.001 0.003+ 15.162 85.255
(0.001) (0.001) (0.001) (0.001) (0.001) (0.002) (15.274) (66.475)
Outcome Mean 0.50 0.42 0.13 0.24 0.69 0.66 6468.54 18343.79
Observations 527,128 436,928 261,427 85,800 394,905 64,182 394,905 64,182
R2 0.255 0.240 0.152 0.256 0.035 0.041 0.060 0.045
Panel B: Above Median
Enrolled in Postsecondary Graduated from Postsecondary Employed Unconditional Winz. Earnings
1 year 2 years 4 years 6 years 1 year 6 years 1 year 6 years
Mean GI -0.002** -0.002 -0.001 -0.001 -0.003*** -0.002 -33.472** -150.227
(0.001) (0.001) (0.001) (0.002) (0.001) (0.002) (15.764) (103.565)
Passing GI 0.002+ 0.000 -0.001 -0.002 0.000 -0.001 -2.636 -27.763
48

(0.001) (0.001) (0.001) (0.002) (0.001) (0.001) (15.197) (76.457)

Outcome Mean 0.78 0.69 0.39 0.58 0.67 0.59 4813.85 19676.29
Observations 557,599 461,734 270,874 89,935 427,243 68,924 427,243 68,924
R2 0.205 0.217 0.196 0.306 0.044 0.076 0.096 0.044
Panel C: P-values from Tests of Coefficient Equality
Enrolled in Postsecondary Graduated from Postsecondary Employed Unconditional Winz. Earnings
1 year 2 years 4 years 6 years 1 year 6 years 1 year 6 years
Mean GI 0.0066 0.0095 0.5658 0.9195 0.3386 0.5940 0.2302 0.6388
Passing GI 0.1059 0.2680 0.8109 0.4088 0.4917 0.0892 0.3562 0.2403

Notes: This table presents estimates from regressions of the relevant outcome (column) on mean grade inflation, passing
grade inflation, cognitive value-added, and noncognitive value-added, as well as our main vector of controls. The sample
is split into students whose 8th grade GPA is below and above median as a proxy for low- and high-achieving students.
For this student-year level analysis, value-added and grade inflation are measured as averages across the student’s math and
English instructors in each year. Panel C shows p-values of tests for coefficient equality that were constructed using seemingly
unrelated regressions. The controls include school, grade, and year fixed effects, ELL, and previous year values of math test
score, English test score, total GPA, fraction of days absent, suspension, and held back. Standard errors are clustered at the
school level. * p < 0.1, ** p < 0.05, *** p < 0.01.
Table A.1: Student-level Summary Statistics (Sample for Separate Math
and English Measures)

Mean Std. Dev. N

Math CST Score 0.18 1.08 364,658
English CST Score 0.30 1.00 376,540
Math GPA 1.93 1.27 468,907
English GPA 2.33 1.22 468,907
GPA 2.44 0.98 468,907
Effort GPA 2.24 0.53 468,907
Cooperation GPA 2.45 0.44 468,907
Fraction of Days Absent 0.06 0.09 468,907
Ever Suspended 0.05 0.22 468,907
Held Back 0.10 0.29 468,907
English Learner 0.14 0.35 468,907
Average Teacher Experience 6.79 2.81 468,907
Don’t Graduate in LAUSD 0.27 0.44 327,535
Leave Dataset 0.24 0.43 327,535
Graduate on Time 0.54 0.50 381,249
Graduate within 5 Years 0.65 0.48 327,535
Number of AP Courses 1.54 2.38 468,907
Ever Took SAT 0.45 0.50 381,249
Ever Took PSAT 0.45 0.50 328,059
SAT Score 1372.40 300.98 212,460
PSAT Score 1161.57 263.83 218,950
English CAHSEE Score 0.32 0.97 414,660
Math CAHSEE Score 0.32 1.02 415,862
10th Grade Science CST Score 0.30 1.04 334,061
11th Grade Social CST Score 0.28 1.02 329,346
Notes: This table reports summary statistics (mean, standard deviation, and number of
observations) for all high school students enrolled in LAUSD between 2004-2013. The sam-
ple is restricted to students with non-missing test scores, grades, and behavioral outcomes
and requires students to have taken both math and English in a given year. CST scores
(end of course standardized tests) are standardized before making sample restrictions.
“Held back” is measured as whether we see the student with the same administrative grade
code in the following year. “Average teacher experience” is the average number of years
a student’s teachers have taught for. “Don’t graduate in LAUSD” and “Leave Dataset”
capture the degree to which the graduation outcomes are censored by the sample time
period ending or students moving outside of LAUSD. “CAHSEE” refers to the California
High School Exit Examination administered from 2006 to 2014.

49
Table A.2: Student-level Summary Statistics, Maryland (Sample for Separate
Math and English Measures)

Mean Std. Dev. N

Math Test Score 0.06 0.98 151,929
English Test Score 0.07 0.94 174,079
Math GPA 2.54 1.09 631,853
English GPA 2.74 1.06 667,585
GPA 2.90 0.84 667,585
Frac. Days Absent 0.07 0.09 667,585
Suspended 0.05 0.21 667,585
Held Back 0.03 0.16 636,207
English Learner 0.02 0.13 667,585
Free and Reduced Price Meals 0.34 0.47 667,585
Black 0.36 0.48 667,585
White 0.42 0.49 667,585
Asian 0.07 0.25 667,585
Hispanic 0.12 0.32 667,585
Female or Non-Binary 0.52 0.50 667,585
AP Classes 3.07 4.36 667,585
Honors Classes 6.90 6.76 667,585
Ever Took SAT 0.68 0.46 652,155
Ever Took SAT or ACT 0.69 0.46 667,585
Ever Took PSAT 0.29 0.45 656,237
SAT Score 1163.84 302.15 166,923
PSAT Score 127.18 30.99 73,049
Graduate On Time 0.88 0.32 652,155
Graduate within 5 Years 0.93 0.25 588,912
Enrolled in Higher Ed 1 year Post HS 0.66 0.47 588,912
Enrolled in Higher Ed 2 years Post HS 0.58 0.49 520,522
Graduated from Higher Ed 4 years Post HS 0.27 0.44 359,479
Graduated from Higher Ed 6 years Post HS 0.41 0.49 189,756
Employed 6 years Post HS 0.69 0.46 588,912
Employed 6 years after Expected HS Graduation 0.63 0.48 189,756
Earnings 1 year Post HS 7945.15 7708.00 407,761
Earnings 6 years Post HS 28132.63 21325.57 120,397

Notes: This table reports summary statistics (mean, standard deviation, and number of obser-
vations) for all high school students enrolled in Maryland between 2013-2023. The sample is
restricted to students with non-missing test scores, grades, and behavioral outcomes and requires
students to have taken both math and English in a given year. Math tests are from Algebra
I, Algebra II, and Geometry courses; English tests are from grades 9 and 10. Test scores are
standardized before making sample restrictions. “Held back” is measured as whether we see the
student with the same administrative grade code in the following year. The timing of the college
and career outcomes are measured relative to expected high school graduation as of 9th grade, so
“1 year post HS” refers to 5 years after we see the student in 9th grade. Earnings are conditional
on having observed any earnings and are winsorized at the 99th percentile.

50
 Table A.3: Correlations between Grade Inflation and Value-Added Measures in LAUSD (Math Teachers)

Mean GI Passing GI Cog. VA GPA Total Effort GPA Coop. GPA Frac. Absent Held Back Suspended
VA VA VA VA VA VA
Mean GI 1.0000 . . . . . . . .
Passing GI 0.8863 1.0000 . . . . . . .
Cog. VA -0.4289 -0.3492 1.0000 . . . . . .
GPA Total VA 0.2610 0.2902 0.0166 1.0000 . . . . .
51

Effort GPA VA 0.0698 0.1290 0.1881 0.8296 1.0000 . . . .

Coop. GPA VA -0.1749 -0.0772 0.2790 0.5862 0.8293 1.0000 . . .
Frac. Absent VA 0.0882 0.0780 -0.0525 0.0974 -0.1770 -0.2266 1.0000 . .
Held Back VA 0.0597 0.0194 -0.0177 -0.0485 -0.0689 -0.1040 0.0463 1.0000 .
Suspended VA -0.0046 0.0037 -0.0732 -0.3544 -0.3381 -0.2945 0.0754 -0.0775 1.0000
Notes: Within-teacher-year correlations are estimated using an error correction procedure based on randomly splitting the data within classroom. Reported
correlations are the average across 200 bootstrapped estimates. The grade inflation and value-added measures used in estimating these correlations are constructed
without shrinkage. This sample includes only math teachers.
 Table A.4: Correlations between Grade Inflation and Value-Added Measures in LAUSD (English Teachers)

Mean GI Passing GI Cog. VA GPA Total Effort GPA Coop. GPA Frac. Absent Held Back Suspended
VA VA VA VA VA VA
Mean GI 1.0000 . . . . . . . .
Passing GI 0.8379 1.0000 . . . . . . .
Cog. VA -0.3747 -0.2431 1.0000 . . . . . .
GPA Total VA 0.2526 0.2518 0.0333 1.0000 . . . . .
52

Effort GPA VA 0.1819 0.2008 0.1641 0.8424 1.0000 . . . .

Coop. GPA VA 0.0213 0.1020 0.3720 0.5920 0.8228 1.0000 . . .
Frac. Absent VA 0.0879 0.0407 -0.1814 0.0710 -0.1826 -0.2101 1.0000 . .
Held Back VA 0.0498 0.0063 -0.1324 -0.0521 -0.0782 -0.1023 0.0806 1.0000 .
Suspended VA 0.0171 0.0168 -0.1531 -0.3366 -0.3128 -0.3074 0.0636 -0.0538 1.0000
Notes: Within-teacher-year correlations are estimated using an error correction procedure based on randomly splitting the data within classroom. Reported
correlations are the average across 200 bootstrapped estimates. The grade inflation and value-added measures used in estimating these correlations are constructed
without shrinkage. This sample includes only English teachers.
 Table A.5: Correlations between Grade Inflation and Value-Added Measures in Maryland (Math Teachers)

Mean GI Passing GI Cog. VA GPA VA Frac. Absent Held Back Suspended

VA VA VA
Mean GI 1.0000 . . . . . .
Passing GI 0.3850 1.0000 . . . . .
Cog. VA -0.3316 -0.1282 1.0000 . . . .
53

GPA VA 0.0168 0.2013 0.1377 1.0000 . . .

Frac Absent VA 0.1961 -0.1912 -0.2528 -0.3812 1.0000 . .
Held Back VA 0.0906 -0.0457 -0.1426 -0.2121 0.3704 1.0000 .
Suspended VA 0.0667 0.1637 -0.0952 -0.0586 -0.0896 -0.1802 1.0000
Notes: Within-teacher-year correlations are estimated using an error correction procedure based on randomly splitting the data
within classroom. Reported correlations are the average across 200 bootstrapped estimates. The grade inflation and value-added
measures used in estimating these correlations are constructed without shrinkage. This sample includes only math teachers.
 Table A.6: Correlations between Grade Inflation and Value-Added Measures in Maryland (English Teach-
ers)

Mean GI Passing GI Cog. VA GPA VA Frac. Absent Held Back Suspended

VA VA VA
Mean GI 1.0000 . . . . . .
Passing GI 0.3309 1.0000 . . . . .
54

Cog. VA -0.2961 -0.0104 1.0000 . . . .

GPA VA 0.0500 0.1934 0.0810 1.0000 . . .
Frac Absent VA 0.0983 -0.2060 -0.2042 -0.3634 1.0000 . .
Held Back VA 0.0220 -0.0988 -0.0702 -0.2486 0.3711 1.0000 .
Suspended VA 0.0615 0.1802 -0.1169 -0.0216 -0.1231 -0.1791 1.0000
Notes: Within-teacher-year correlations are estimated using an error correction procedure based on randomly splitting the data
within classroom. Reported correlations are the average across 200 bootstrapped estimates. The grade inflation and value-added
measures used in estimating these correlations are constructed without shrinkage. This sample includes only English teachers.
 Table A.7: Regressions of Single Teacher Measures on College and Career Outcomes in Maryland

Enrolled in Postsecondary Graduated from Postsecondary Employed Log Conditional Earnings Unconditional Winz. Earnings
1 year 2 years 4 years 6 years 1 year 6 years 1 year 6 years 1 year 6 years
Mean GI -0.005*** -0.005*** 0.000 -0.003*** -0.003*** -0.002** -0.003 -0.006** -39.118*** -163.664***
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002) (0.003) (12.443) (44.686)
Passing GI -0.000 -0.001 -0.001+ -0.003*** -0.001 -0.000 0.003 -0.002 -4.630 -71.329**
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002) (0.003) (11.847) (35.540)
Cog. VA 0.002*** 0.002** -0.003*** -0.000 -0.000 -0.001 -0.006*** -0.002 -25.438** -18.649
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002) (0.003) (12.737) (42.155)
55

Noncog. VA 0.007*** 0.006*** -0.001 0.001 0.002*** 0.002** 0.006*** 0.006+ 31.558*** 39.826
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002) (0.003) (9.536) (44.529)
Outcome Mean 0.64 0.56 0.27 0.40 0.68 0.62 8.39 9.78 5444.70 17801.59
Observations 979,425 838,616 552,566 274,354 979,425 274,354 664,075 170,840 979,425 274,354
Notes: This table presents estimated coefficients from regressions of the relevant outcome (column) on the relevant teacher measure (row) and our main vector of
controls. Thus, each cell reports an estimate from a different regression. The outcomes in this table are only available in the Maryland data, not in LAUSD. The timing
of enrollment is measured relative to expected high school graduation as of 9th grade; for example, “1 year post HS” refers to 5 years after we see the student in 9th
grade. For this student-year level analysis, value-added and grade inflation are measured as averages across the student’s math and English instructors in each year. The
controls include school, grade, and year fixed effects, ELL, and previous year values of math test score, English test score, total GPA, fraction of days absent, suspension,
and held back. Standard errors are clustered at the school level. * p < 0.1, ** p < 0.05, *** p < 0.01.
Table A.8: Effects of Subject-Specific Teacher Measures on High School Outcomes in Los Angeles

Future Test (math) Future Test (ela) Held Back Graduate in 5 Years Took SAT SAT Score
Mean GI (math) -0.046∗∗∗ -0.022∗∗∗ 0.000 -0.011∗∗∗ -0.009∗∗ -1.129
(0.010) (0.005) (0.003) (0.003) (0.004) (2.369)
Mean GI (ela) -0.005 -0.011∗∗∗ 0.003 -0.006 0.001 0.679
(0.008) (0.004) (0.002) (0.004) (0.004) (2.131)
Passing GI (math) 0.011 0.007 -0.011∗∗∗ 0.006+ 0.003 -2.459
(0.009) (0.005) (0.003) (0.004) (0.003) (1.960)
Passing GI (ela) -0.001 -0.006 -0.012∗∗∗ 0.005 -0.001 -6.945∗∗
(0.009) (0.005) (0.002) (0.004) (0.004) (2.644)
Cog. VA (math) 0.126∗∗∗ 0.017∗∗∗ 0.002+ -0.010∗∗∗ 0.011∗∗∗ 21.299∗∗∗
(0.013) (0.003) (0.001) (0.003) (0.004) (2.862)
Cog. VA (ela) 0.058∗∗∗ 0.066∗∗∗ -0.002 0.013∗∗∗ 0.033∗∗∗ 19.492∗∗∗
(0.008) (0.003) (0.001) (0.002) (0.002) (3.287)
Noncog. VA (math) -0.005 0.005∗∗∗ -0.003∗∗∗ 0.007∗∗∗ 0.007∗∗∗ -0.306
(0.003) (0.002) (0.001) (0.001) (0.001) (0.605)
Noncog. VA (ela) 0.004 0.008∗∗∗ -0.004∗∗∗ 0.005∗∗∗ 0.004∗∗∗ 0.492
(0.004) (0.002) (0.001) (0.001) (0.001) (1.166)
Outcome Mean 0.11 0.23 0.11 0.65 0.40 1368.58
Observations 199,038 215,559 411,275 327,528 323,635 173,195
R2 0.564 0.679 0.160 0.307 0.379 0.746

Notes: This table presents estimates from regressions of the relevant outcome (column) on mean grade inflation,
passing grade inflation, cognitive value-added, and noncognitive value-added, as well as our main vector of controls.
For this student-year level analysis, value-added and grade inflation are measured as averages across the student’s
math and English instructors in each year. The controls include school, grade, and year fixed effects, ELL, and
previous year values of math test score, English test score, total GPA, fraction of days absent, suspension, and held
back. Standard errors are clustered at the school level. * p < 0.1, ** p < 0.05, *** p < 0.01.

56
Table A.9: Effects of Subject-Specific Teacher Measures on High School Outcomes in Maryland

Future Test (math) Future Test (ela) Held Back Graduate in 5 Years Took SAT SAT Score
Mean GI (math) 0.001 -0.007 -0.002** -0.001 -0.003 -0.502
(0.008) (0.005) (0.001) (0.001) (0.003) (2.372)
Mean GI (ela) -0.000 -0.001 -0.001** -0.001 -0.004+ -0.138
(0.007) (0.007) (0.001) (0.001) (0.003) (2.308)
Passing GI (math) -0.010 -0.003 -0.007*** 0.004** 0.004 -8.151**
(0.012) (0.007) (0.002) (0.002) (0.004) (4.003)
Passing GI (ela) -0.013 0.009 -0.010** 0.002 0.008 -2.334
(0.015) (0.013) (0.005) (0.003) (0.009) (4.603)
Cog. VA (math) 0.004 0.036*** -0.001 0.002+ 0.008** 14.186***
(0.011) (0.007) (0.001) (0.001) (0.003) (3.911)
Cog. VA (ela) 0.034*** 0.043*** 0.002*** -0.001 0.006+ -11.621***
(0.006) (0.006) (0.001) (0.001) (0.003) (1.779)
Noncog. VA (math) 0.002 0.009** -0.002*** 0.004*** 0.000 -3.382**
(0.005) (0.004) (0.000) (0.000) (0.002) (1.379)
Noncog. VA (ela) 0.010+ 0.002 -0.002*** 0.004*** 0.001 -3.843***
(0.005) (0.004) (0.001) (0.001) (0.002) (1.373)
Outcome Mean -0.19 0.05 0.03 0.92 0.68 1150.20
Observations 65,988 130,701 491,914 444,650 492,054 111,200
R2 0.383 0.578 0.145 0.209 0.239 0.760

Notes: This table presents estimates from regressions of the relevant outcome (column) on mean grade inflation,
passing grade inflation, cognitive value-added, and noncognitive value-added, as well as our main vector of controls.
For this student-year level analysis, value-added and grade inflation are measured as averages across the student’s
math and English instructors in each year. The controls include school, grade, and year fixed effects, ELL, and
previous year values of math test score, English test score, total GPA, fraction of days absent, suspension, and held
back. Standard errors are clustered at the school level. * p < 0.1, ** p < 0.05, *** p < 0.01.

57
 Table A.10: Effects of Subject-Specific Teacher Measures on College and Career Outcomes in Maryland

Enrolled in Postsecondary Graduated from Postsecondary Employed Log Conditional Earnings Unconditional Winz. Earnings
1 year 2 years 4 years 6 years 1 year 6 years 1 year 6 years 1 year 6 years
Mean GI (math) -0.006*** -0.004** 0.001 0.000 -0.005*** -0.005** -0.010** 0.008 -74.643*** -110.426
(0.002) (0.002) (0.002) (0.002) (0.001) (0.002) (0.005) (0.007) (24.071) (108.179)
Mean GI (ela) -0.006*** -0.005*** 0.002 -0.002 -0.003** -0.001 -0.005 -0.010+ -44.788** -181.627**
(0.001) (0.002) (0.001) (0.002) (0.001) (0.002) (0.004) (0.006) (21.835) (81.448)
Passing GI (math) 0.002 -0.002 -0.005** -0.007** 0.002 0.003 0.010+ -0.005 45.384 -65.554
(0.002) (0.002) (0.002) (0.003) (0.002) (0.003) (0.006) (0.010) (34.094) (125.684)
Passing GI (ela) 0.005 0.005 -0.001 -0.003 0.000 -0.003 0.013 -0.019 19.245 -136.213
(0.004) (0.004) (0.004) (0.005) (0.002) (0.004) (0.008) (0.014) (43.935) (247.180)
Cog. VA (math) -0.002 -0.003 -0.003 -0.003 -0.003 -0.007*** -0.004 -0.005 -10.622 -227.701**
(0.002) (0.002) (0.002) (0.003) (0.002) (0.002) (0.007) (0.008) (38.391) (110.234)
58

Cog. VA (ela) 0.002+ 0.002+ -0.004*** 0.000 -0.001 0.000 -0.015*** -0.001 -79.798*** 46.044
(0.001) (0.001) (0.002) (0.002) (0.001) (0.002) (0.003) (0.006) (16.232) (80.076)
Noncog. VA (math) 0.004*** 0.004*** 0.000 0.000 0.001 0.001 0.006*** 0.003 22.707** -14.261
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002) (0.004) (11.518) (53.920)
Noncog. VA (ela) 0.004*** 0.004*** -0.001 0.000 0.000 0.001 -0.001 0.003 -3.207 57.061
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002) (0.004) (13.620) (57.590)
Outcome Mean 0.65 0.57 0.26 0.40 0.69 0.63 8.40 9.79 5546.57 18175.75
Observations 444,650 388,229 260,877 127,076 444,650 127,076 307,329 80,360 444,650 127,076
R2 0.269 0.268 0.230 0.329 0.032 0.048 0.089 0.049 0.078 0.041

Notes: This table presents estimates from regressions of the relevant outcome (column) on mean grade inflation, passing grade inflation, cognitive value-
added, and noncognitive value-added, as well as our main vector of controls. The outcomes in this table are only available in the Maryland data, not in
LAUSD. For this student-year level analysis, value-added and grade inflation are measured as averages across the student’s math and English instructors in
each year. The controls include school, grade, and year fixed effects, ELL, and previous year values of math test score, English test score, total GPA, fraction
of days absent, suspension, and held back. Standard errors are clustered at the school level. * p < 0.1, ** p < 0.05, *** p < 0.01.
 Table A.11: Effects of Teacher Measures on Additional High School Outcomes in Los Angeles

Future Held Back Future Frac. Days Absent Future Suspension Graduate On Time Don’t Graduate Leave Dataset Next Year PSAT Score Took PSAT
Mean GI 0.003+ 0.001∗∗∗ 0.001∗∗ -0.007∗∗ 0.009∗∗∗ 0.002 -3.956∗∗ -0.004
(0.002) (0.000) (0.001) (0.003) (0.002) (0.001) (1.629) (0.003)
Passing GI -0.004∗∗∗ -0.001+ 0.000 0.004 -0.008∗∗∗ -0.004∗∗∗ 2.235 0.002
(0.002) (0.000) (0.001) (0.003) (0.002) (0.001) (1.784) (0.004)
Cog. VA 0.002∗∗ 0.000 -0.001 -0.001 0.002 -0.001 22.855∗∗∗ 0.012∗∗∗
(0.001) (0.000) (0.000) (0.003) (0.002) (0.001) (3.083) (0.003)
59

Noncog. VA -0.011∗∗∗ -0.002∗∗∗ -0.004∗∗∗ 0.012∗∗∗ -0.012∗∗∗ 0.001 2.441∗∗∗ -0.003

(0.001) (0.000) (0.001) (0.002) (0.001) (0.001) (0.806) (0.003)
Outcome Mean 0.13 0.08 0.05 0.49 0.37 0.08 1097.94 0.40
Observations 630,184 731,638 763,350 833,265 597,851 832,002 261,265 562,777
R2 0.129 0.299 0.057 0.284 0.306 0.085 0.727 0.338

Notes: This table presents estimates from regressions of the relevant outcome (column) on mean grade inflation, passing grade inflation, cognitive value-added, and
noncognitive value-added, as well as our main vector of controls. For this student-year level analysis, value-added and grade inflation are measured as averages across the
student’s math and English instructors in each year. The controls include school, grade, and year fixed effects, ELL, and previous year values of math test score, English
test score, total GPA, fraction of days absent, suspension, and held back. Standard errors are clustered at the school level. * p < 0.1, ** p < 0.05, *** p < 0.01.
 Table A.12: Effects of Teacher Measures on Additional High School Outcomes in Maryland

Future Held Back Future Frac. Days Absent Future Suspension Graduate On Time PSAT Score Ever Took PSAT Ever Took SAT or ACT
Mean GI 0.001*** 0.001*** 0.002*** -0.001** -0.015 -0.003+ -0.004***
(0.000) (0.000) (0.000) (0.001) (0.128) (0.001) (0.001)
Passing GI -0.001 0.000 -0.000 0.002** -0.194 0.001 0.002
(0.001) (0.000) (0.000) (0.001) (0.130) (0.001) (0.002)
Cog. VA -0.000 -0.000 0.000 0.000 0.163+ -0.004*** 0.004***
(0.000) (0.000) (0.000) (0.001) (0.098) (0.001) (0.001)
Noncog. VA -0.004*** -0.001*** -0.004*** 0.010*** -0.742*** -0.001 0.005***
60

(0.000) (0.000) (0.001) (0.001) (0.143) (0.001) (0.002)

Outcome Mean 0.04 0.08 0.04 0.84 128.39 0.25 0.65
Observations 1,059,643 1,059,391 1,059,643 1,117,945 155,708 1,126,690 1,178,611
R2 0.108 0.366 0.065 0.342 0.738 0.645 0.327

Notes: This table presents estimates from regressions of the relevant outcome (column) on mean grade inflation, passing grade inflation, cognitive value-
added, and noncognitive value-added, as well as our main vector of controls. For this student-year level analysis, value-added and grade inflation are
measured as averages across the student’s math and English instructors in each year. The controls include school, grade, and year fixed effects, ELL, and
previous year values of math test score, English test score, total GPA, fraction of days absent, suspension, and held back. Standard errors are clustered at
the school level. * p < 0.1, ** p < 0.05, *** p < 0.01.
 Table A.13: Forecast Bias Tests of Value-Added in the LAUSD

Panel A: Individual Test, Math Test Score GPA Effort GPA Coop GPA Frac Days Absent Suspended Heldback
Corresponding VA Measure 1.11 0.68 0.85 1.12 -0.63 -1.79 -1.99
(0.02) (0.05) (0.06) (0.05) (0.09) (0.15) (0.13)
N 479,237 701,518 699,042 699,043 691,505 701,698 619,341
Panel B: Individual Test, ELA Test Score GPA Effort GPA Coop GPA Frac Days Absent Suspended Heldback
Corresponding VA Measure 1.16 0.59 0.68 0.92 -0.29 -1.77 -1.68
(0.04) (0.04) (0.04) (0.03) (0.04) (0.20) (0.16)
N 548,508 778,299 775,450 775,448 766,664 779,165 688,482
Panel C: Prediction Test, Math Test Score GPA Effort GPA Coop GPA Frac Days Absent Suspended Heldback
Corresponding VA Measure 0.10 0.00 0.01 0.01 -0.00 -0.00 -0.00
(0.01) (0.01) (0.00) (0.00) (0.00) (0.00) (0.00)
N 538,602 541,501 539,501 539,501 541,102 541,564 540,674
Panel D: Prediction Test, ELA Test Score GPA Effort GPA Coop GPA Frac Days Absent Suspended Heldback
Corresponding VA Measure 0.26 0.01 0.01 0.01 0.00 -0.00 -0.00
(0.03) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
N 603,609 607,046 604,596 604,596 605,768 607,479 603,901

Notes: This table presents results from two different forecast bias tests (the second test is only practical in LAUSD where
there are more frequent standardized tests). For the “individual” test, an estimate close to 1 shows that these measures,
though subject to estimation error, are actually predictive of student outcomes out of sample. For the ‘prediction’ test, an
estimate close to 0 shows that these measures are not biased by sorting on omitted variables. Standard errors are clustered
at the school level. * p < 0.1, ** p < 0.05, *** p < 0.01.

61
 Table A.14: Forecast Bias Tests of Value-Added in Maryland

Panel A: Individual Test, Math Test Score GPA Frac Days Absent Suspended Heldback
Corresponding VA Measure 1.07 0.85 -1.13 -1.32 -0.68
(0.09) (0.06) (0.09) (0.09) (0.06)
N 226,260 1,143,672 1,144,560 1,144,560 1,069,973
Panel B: Individual Test, ELA Test Score GPA Frac Days Absent Suspended Heldback
Corresponding VA Measure 0.88 0.88 -1.19 -1.40 -0.43
(0.11) (0.06) (0.10) (0.10) (0.06)
N 254,944 1,163,448 1,167,090 1,167,186 1,101,485

Notes: This table presents results from the forecast bias test that is relevant in Maryland. For the
“individual” test, an estimate close to 1 shows that these measures, though subject to estimation
error, are actually predictive of student outcomes out of sample. * p < 0.1, ** p < 0.05, *** p < 0.01.

62
Table A.15: Effects of Measures of Teacher Grade Inflation and Value-Added with Additional
Controls on High School Outcomes in Maryland

Future Test (math) Future Test (ela) Held Back Graduate in 5 Years Took SAT SAT Score
Mean GI 0.001 -0.007+ -0.001*** -0.001*** -0.004*** -0.407
(0.004) (0.003) (0.000) (0.000) (0.001) (1.254)
Passing GI -0.003 0.002 -0.004*** 0.002** 0.001 -2.793**
(0.004) (0.003) (0.001) (0.001) (0.002) (1.400)
Cog. VA 0.016*** 0.031*** 0.001+ 0.000 0.004*** -0.705
(0.004) (0.003) (0.000) (0.000) (0.001) (1.268)
Noncog. VA 0.001 0.011*** -0.004*** 0.007*** 0.002 -6.083***
(0.004) (0.003) (0.001) (0.001) (0.002) (1.440)
Outcome Mean -0.20 0.03 0.04 0.90 0.65 1158.08
Observations 123,205 279,802 1,117,227 979,425 1,117,945 252,950
R2 0.356 0.591 0.153 0.238 0.241 0.772

Notes: This table presents estimates from regressions of the relevant outcome (column) on alternative measures
of mean grade inflation, passing grade inflation, cognitive value-added, and noncognitive value-added. The
alternative measures are created using additional demographic controls available only in Maryland: race, sex,
and use of Free and Reduced Price Lunch. The regressions also include those additional controls along with
our standard controls. For this student-year level analysis, value-added and grade inflation are measured as
averages across the student’s math and English instructors in each year. The controls include school, grade, and
year fixed effects, ELL, and previous year values of math test score, English test score, total GPA, fraction of days
absent, suspension, and held back. Standard errors are clustered at the school level. * p < 0.1, ** p < 0.05, *** p
< 0.01.

63
 Table A.16: Effects of Teacher Grade Inflation and Value-Added with Additional Controls on College
and Career Outcomes in Maryland

Enrolled in Postsecondary Graduated from Postsecondary Employed Unconditional Winz. Earnings

1 year 2 years 4 years 6 years 1 year 6 years 1 year 6 years
Mean GI -0.006*** -0.005*** 0.000 -0.002+ -0.004*** -0.002+ -48.557*** -168.062***
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (12.284) (51.641)
Passing GI 0.002** 0.001 -0.002+ -0.002** 0.000 0.000 11.538 -6.060
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (11.629) (41.400)
Cog. VA 0.001 0.001 -0.004*** -0.001 -0.000 -0.001 -29.773** -53.385
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (12.433) (41.588)
Noncog. VA 0.006*** 0.005*** -0.001 0.000 0.002*** 0.001 33.330*** -24.619
64

(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (9.711) (43.546)

Outcome Mean 0.64 0.56 0.27 0.40 0.68 0.62 5444.70 17801.59
Observations 979,425 838,616 552,566 274,354 979,425 274,354 979,425 274,354
R2 0.285 0.278 0.237 0.338 0.034 0.049 0.080 0.040

Notes: This table presents estimates from regressions of the relevant outcome (column) on alternative measures of mean
grade inflation, passing grade inflation, cognitive value-added, and noncognitive value-added. The alternative measures are
created using additional demographic controls available only in Maryland: race, sex, and use of Free and Reduced Price
Lunch. The regressions also include those additional controls along with our standard controls. The outcomes in this table
are only available in the Maryland data, not in LAUSD. For this student-year level analysis, value-added and grade inflation
are measured as averages across the student’s math and English instructors in each year. The controls include school, grade,
and year fixed effects, ELL, and previous year values of math test score, English test score, total GPA, fraction of days absent,
suspension, and held back. Standard errors are clustered at the school level. * p < 0.1, ** p < 0.05, *** p < 0.01.
 Figure A.1: Average GPA over Time by Subject

(a) Math (b) ELA

Notes: This figure plots average math and English GPA over time for students in our Los
Angeles and Maryland samples. The data from the LAUSD span 2004 to 2013 and the data
from Maryland span 2013 to 2023. GPA is measured on a 0 to 4 scale (e.g., A = 4.0, B = 3.0).

65