University of Southern California Dissertations and Theses

School resource allocation in times of economic boom and bust

(USC Thesis Other)

School resource allocation in times of economic boom and bust

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

SCHOOL RESOURCE ALLOCATION IN TIMES OF ECONOMIC BOOM AND BUST

by

David S. Knight

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(EDUCATION)

May 2016

Copyright 2016 David S. Knight

ii
ACKNOWLEDGEMENTS
Read any statistic on economic inequality in the United States. Students born into poverty
today have a lower chance of entering the middle class than almost any other time in history. I
entered an Education Policy graduate degree program because I believe the nation’s public
institutions play a vital role in providing equitable opportunities for youth. Over the past four
years, I have encountered many hard working educators, mentors, and students who share my
passion for economic and educational justice. These individuals have continually inspired and
supported me throughout my graduate studies.
Completing a graduate degree is not something one accomplishes alone. My experience
is a perfect example. I have had the privilege of being surrounded by and receiving support from
an outstanding group of people throughout my life. Communicating my gratitude to these
individuals is a life-long duty. I will do my best in this space to acknowledge their many
contributions.
My faculty mentors all played major roles in my intellectual and personal growth. My
advisor Dr. Katharine O. Strunk has worked tirelessly to help support me and guide my
development. I have learned a tremendous amount from her and I continue to learn from her
everyday. Thank you Katharine. I thank Dr. Larry Picus, who is one of the most supportive and
friendliest mentors I have ever known. Dr. Picus models how scholars can be extremely engaged
in their work while maintaining close relationships among colleagues and a robust sense of
humor. His Hawaii Five-0 ringtone is but one example. I owe a special share of gratitude to Dr.
William G. Tierney, who provided mentorship throughout my graduate school experience and
has helped me navigate the academic world. I wish to thank Dr. Gary Painter, who has patiently

iii
offered his wisdom and experience to me during my four years at the University of Southern
California.
Many other individuals in the USC family have offered their support. I owe much
appreciation to my officemates in the data room, my classmates and fellow cohort members, and
our program directors Katie, Laura, Diane, and Aba—thank you.
I thank my parents who have always supported my education, and my siblings who lift
my spirits whenever we are together. Lastly, and most importantly, I want to thank my wife,
Julia. I thank her for always believing in me and making me feel like I can accomplish anything.
Thank you Julia. I love you and value your love and support deeply.

iv
TABLE OF CONTENTS

Acknowledgements ii

Introduction 1

Essay 1: Are There Hidden Costs Associated with Conducting Teacher Layoffs?
The Impact of RIFs and Layoffs on Teacher Effectiveness 4
Tables and Figures for Introduction and Essay 1 31
References for Essay 1 36
Appendix A-1: Additional Tables for Essay 1 42

Essay 2: Who Bears the Cost of District Funding Cuts: Equity Implications of
Teacher Layoffs 43
Tables for Essay 2 83
References for Essay 2 92
Appendix A-2: Additional Tables and Figures for Essay 2 95

Essay 3: Cost-Effectiveness in the Context of School Budget Restoration:
Comparing the Cost and Effects of Teacher Salary Increases and Class Size
Reduction in the Los Angeles Unified School District 104
Tables and Figures for Essay 3 149
References for Essay 3 155
Appendix A-3: Additional Tables for Essay 3 178
Appendix B: Additional information on cost methods 181
Tables and Figures for Appendix B 208

Appendix C: Estimating Value-Added Measures of Effectiveness 213

1
School Resource Allocation in Times of Economic Boom and Bust:
Three Essays on District Budget Management
Introduction
The Great Recession of 2008 devastated funding for public education in the United
States, forcing districts around the country to cut budgets and conduct layoffs. In times of
financial duress, when local districts are forced to make difficult choices about resources, limited
evidence is available to help guide decisions. Similarly, as districts have begun to restore funding
levels, there is a lack of research identifying the most cost-effective resource allocation strategies
in K-12 public schools. Thus a crucial contribution for educational research is to document: (a)
how teacher layoffs stemming from budget cuts impact students and teachers; (b) which students
and teachers are affected by these layoffs; and (c) the relative cost and benefits of alternate
strategies for restoring school budgets. This three-essay dissertation studies each of these
important topics in turn.
I explore these issues within the context of the Los Angeles Unified School District
(LAUSD) over a seven-year period from school years 2006-07 to 2012-13. Following the Great
Recession, the district underwent a wide-scale budget reduction leading to four years of teacher
layoffs from 2008-09 to 2011-12. Over 14,000 Reduction-in-Force (RIF) notices were
distributed to teachers (an average of 13.3% of all teachers each year) and approximately 4,500
of those notices resulted in a teacher layoff. The passing of a state tax initiative and the
recovering economy helped LAUSD begin to restore their budget back to pre-recession levels.
These events are depicted in Figure 1, which shows when the initial impact of the recession took
place (2008-09), the period of teacher layoffs (2008-09 to 2011-12), and the two years of budget
restoration and spending negotiations (2012-13 to 2013-14). Throughout this dissertation, I draw

2
on a unique seven-year panel of administrative student and teacher data provided by LAUSD,
which is linked to school data from the California Department of Education.
In the first essay, I document the detrimental effects on teachers’ classroom performance
when they are laid off and rehired to the district. I find that current year RIFs and layoffs had no
significant impact on teacher effectiveness (as measured by valued-added modeling of teachers’
contributions to student achievement, or VAMs), but being laid off in the previous year
significantly and substantially reduced teacher effectiveness in the following school year. I
present evidence to warrant a causal interpretation of these results and discuss implications for
policy.
The second essay reports the extent to which teacher layoffs are distributed inequitably
across students. First, I show that teacher layoffs disproportionately impact traditionally
underserved students (students of color, low income students and English language learners).
However, as a result of a lawsuit (Reed v. California, 2011) filed by the American Civil Liberties
Union against the state of California because of this disproportionate impact of layoffs, LAUSD
instituted a reform – called the Reed protections – that shielded 45 high-need schools from
disproportionate staff reductions during the district’s second two years of layoffs (2010-11 and
2011-12). I find that the Reed settlement lessened the impact of layoffs on students in protected
schools. Because these schools educate primarily low-income and underrepresented minority
students, the Reed protections substantially mitigated the inequitable consequences of seniority-
based teacher layoffs.
Last, in Essay 3, I turn to the post-recession era, when California and LAUSD were
emerging from the worst of the fiscal crunch and working with new funds to improve
educational programs and services. In particular, LAUSD’s fiscal crisis eased in 2012 when

3
California voters passed Proposition 30, a state tax initiative that increased funding for local
school districts. Around that same time, California overhauled its school finance system and
replaced it with the Local Control Funding Formula, a model that emphasizes equity and local
spending flexibility. During school years 2012-13 and 2013-14, the district was embroiled in
negotiations around how new money should be invested. In the third essay, I estimate the costs
and effects of spending proposals that LAUSD administrators and local stakeholders negotiated
during this time period. I focus on two polices in particular: class size reduction and teacher
salary increases. I show that targeting spending to lower-performing schools, lower grades, or
novice teachers is generally more cost-effective than across-the-board spending proposals.
Second, implementing policies in unison is generally more cost-effective than focusing solely on
one strategy or the other (i.e., class size reduction or salary increases alone). Finally, I find that
when compared to the spending agreement that was ultimately reached, the district’s plan
compares favorably to many, but not all, of the spending proposals examined in this study.
In summary, this dissertation follows a large urban school district through a major
economic recession and into a period of budget restoration. By assessing the negative impacts of
district budget cuts, documenting the types of students that bear the cost of these funding cuts,
and assessing cost-effective strategies for restoring budgets, this dissertation offers important
policy lessons for the field while providing timely policy recommendations.

4
Essay 1. Are There Hidden Costs Associated with Conducting Teacher Layoffs?
The Impact of RIFs and Layoffs on Teacher Effectiveness
The Great Recession of 2008 caused state and local governments to drastically reduce
funding allocated to public education. These funding cuts then forced school districts around the
country to reduce budgets and lay off educators in quantities that far exceeded anything in recent
history. National estimates of teacher layoffs stemming from the Great Recession range from
170,000 to 240,000 by the 2011-12 school year (Bureau of Labor Statistics, 2012; National
Education Association, 2010).
1
In many districts, and especially in large urban centers, such as
Los Angeles, New York City, and Chicago, budget shortfalls necessitated particularly deep
reductions in the labor force. For example, the Los Angeles Unified School District laid off over
4,500 teachers over the four years ranging from 2008-09 through 2011-12 school years.
2

Importantly for this study, the majority of laid off teachers were rehired the following school
year, after spending the summer unsure of whether they would be recalled to their position.
Scholars are beginning to examine the effects of these layoffs by studying their impacts
on teacher mobility patterns and changes in the composition of the teaching workforce (Boyd,
Lankford, Loeb, & Wyckoff, 2011; Kraft, 2015). Across the country and in LAUSD, teacher
layoff policies typically rely primarily on seniority (Last-in-First-Out, or LIFO), causing high-
performing junior faculty to be targeted for layoffs and removed from the classroom while
potentially lower performing senior faculty are retained (Goldhaber & Theobald, 2013). Some

1
Note that the lower number, provided by the National Education Association is based on the layoffs that took place
from 2009 to 2010, and is likely a lower bound estimate of the total number of teachers laid off since the Great
Recession.
2
Based on Author’s calculations using LAUSD administrative teacher data. The exact number of layoff notices
distributed to teachers depends on how classroom teachers are defined. I define teachers as any employee listed in
the non-administrator demographic file with the job title of teacher who can be linked to a single school. In my main
analyses, I exclude itinerant teachers, returned retirees, and long-term substitutes, but include these employees in
various model specification checks. Under this definition, a total of 4,455 layoff notices were distributed. A total of
5,876 notices were distributed to certificated staff members districtwide.

5
scholars argue that students will suffer because of the distributional shift in teacher effectiveness
caused by LIFO layoff policies.
However, what has been less reported in the media, and less studied by researchers
attempting to understand the impacts of layoffs on the education workforce, is the impact of the
broader layoff process. In particular, the way layoffs are conducted in most districts around the
country requires that initial layoff warning notices – called Reduction-in-Force (RIF) notices –
are sent to a far greater number of teachers than will actually need to be let go to meet budget
projections. In California, for example, state legislation requires public school districts to send
RIF notices by March 15 to any employee who may conceivably be let go for the following
school year.
3
Then, California districts must send final layoff notices by May 15. However, state
budgets – and therefore district budgets – are often not finalized until June or later, requiring that
school districts send out notification of RIFs and layoffs without complete information about
their budget for the coming year. This process results in significantly more teachers receiving
RIF notices than will eventually be laid off, and, in turn, more teachers being laid off then will
eventually need to be removed from the district in order to meet budget realities. The end result
of the layoff process, then, is that many teachers face substantial job threat but do not eventually
lose their jobs.
Districts bear significant personnel costs and legal fees in conducting layoffs (Estrada,
2012). State law requires that laid off teachers have a right to a hearing before an Administrative
Law Judge (Cal. Educ. Code Sec. 44955) while a substitute teacher covers their classroom. In a
survey of 230 California districts (44% of all students in California), administrators estimated

3
California Education Code (§44955-9) allows districts to conduct emergency layoffs in August, just before the
school year begins if the state’s revenue limit allocations to school districts are not increased by at least 2%.
However, over the four years of layoffs, California districts very rarely used this option in part because the state
budget was enacted close to or after the August window for school years 2008-09 and 2010-11, and the state
prohibited use of layoff window in 2011-12 (Estrada, 2012).

6
that personnel time, legal fees, substitute costs, and mailings added up to a total of $706 per
noticed teacher (Estrada, 2012). Combining this estimate and the 14,142 RIF notices distributed
in LAUSD over four years, the total explicit costs of the layoff process for LAUSD was
approximately $9.98 million.
In addition to the explicit costs involved in the layoff process, the shuffling of teachers
who results from over-notification causes substantial additional costs for the district. Until
recently, little research has examined these implicit or hidden costs of conducting widescale
teacher layoffs. For example, even teachers who have their RIF rescinded are more likely to exit
their school than otherwise similar teachers who are not RIFed (Goldhaber, Strunk, Brown, &
Knight, 2015). Wide distribution of layoffs thus contributes to teacher churn, which harms the
school culture, the professional environment of the school, and student achievement (e.g., Guin,
2004; Ronfeldt, Loeb, & Wyckoff, 2014). Moreover, schools’ organizational context plays a
significant role in shaping teachers’ performance (e.g., Johnson, Kraft & Papay, 2012). Finally,
literature from the field of psychology and behavioral economics suggests that public sector
workers may be more risk averse, and in general, workers may be less motivated and thus less
productive when their jobs are threatened (e.g., Brockner, Grover, O’Malley, Reed & Glynn,
1993; Buurman, Delfgaauw, Dur, & Van den Bossche, 2012). Together, this research suggests
that employment threat and distressed professional environment induced by the layoff process
may have severe implications for the teacher workforce that remains in the district, and as such,
for their students. In particular, the layoff process may harm teachers’ instructional effectiveness.
This study is the first that I know of to examine how teachers’ own job performance
changes as a result of exposure to layoffs and layoff threat. Using longitudinal teacher-level data
from the Los Angeles Unified School District (LAUSD) from the 2006-07 through 2012-13

7
school years (before, during, and after the period in which teacher layoffs were implemented), I
assess how teachers’ effectiveness (measured by value-added measures of teachers’ contributions
to their students’ test scores) changes following exposure to layoffs or layoff threat in the
previous year. In particular, I ask: Do value-added measures of teacher effectiveness change in
response to layoff-induced unemployment threat?
I find that teachers who are laid off but return to the district the following year are
substantially less effective. Teachers who receive a RIF notice that is rescinded in the prior year,
however, experience no significant changes in their measure of effectiveness. I argue that these
results are driven by teachers’ job insecurity and uncertainty associated with the layoff process
and I present evidence to rule out alternate explanations including selection bias, regression to
the mean of value-added scores, or transitions to a new school, grade level, or teaching position
caused by the layoff process.
The results from this study are instructive for policymakers as they look to the future.
Although Great Recession-era layoffs have ceased, recessionary trends are cyclical, and layoffs
will occur again. LAUSD, for example, distributed RIF notices in March of 2015, despite
additional state funding provided through a recent tax initiative (Blume, 2015). For
policymakers, it is important to consider how policies that lead to more teachers being impacted
by the layoff process than are necessary might be changed. Some possibilities include (a)
requiring that school districts and states have more information about budgets before sending out
RIF notices or laying off personnel; (b) appropriating education funding one year in advance to
ameliorate district budget uncertainties, as is done in Alaska, Wyoming, and other states (Brown,
2015); or (c) distributing layoffs based on local school needs rather than districtwide relative
seniority. Finally, it stands to reason that policies that evaluate teachers based on their

8
effectiveness should consider organizational context and job security during times of stress and
strife, such as reductions in force.
The remainder of the paper proceeds as follows. In the next section, I provide background
on the impact of layoffs, both on the structure of the teaching force and on the potential
responses of teachers to employment threat. The following section outlines the layoff process in
LAUSD and explains why the district is a particularly relevant context for a study of the impacts
of the layoff process on teacher effectiveness. I then describe my data and the analytic methods
used to assess these data. In the next section, I provide findings, and the final section discusses
these results, offers policy implications, and concludes.
Literature on the Impacts of Layoffs and the Layoff Process
As discussed above, there are several ways that the layoff process might impact teachers
and students. First, layoffs can directly impact the teacher workforce and student learning by
removing laid off teachers from their districts, thereby increasing class sizes. Second, LIFO
policies can remove promising early career teachers who may have been protected under other
variants of layoff policies. Third, the broader layoff process can indirectly impact the teacher
workforce by causing teachers to switch schools within the district at greater rates than they
otherwise would have (Goldhaber et al., 2015). Fourth, the employment threat introduced by the
layoff process may impact individual teachers’ organizational commitment and productivity
throughout the school year and professional development activities during summer months. Last,
layoffs and the layoff process can adversely impact teachers’ collective productivity by harming
the professional environments in which they work. I briefly review the literature addressing each
of these mechanisms in turn.

9
Direct Impacts of Layoffs on Class Size and Workforce Composition
Research examining the direct impacts of layoffs on classrooms and the teaching
workforce has grown in the wake of the Great Recession. The increase in class size is the most
obvious impact of layoffs and some studies suggest seniority-driven layoffs result in a greater
proportion of teachers leaving the workforce, and larger increases in class size (Boyd et al.,
2011; Goldhaber & Theobald, 2013). For example, simulations based on New York City data
show that a 5% reduction in salaries would require laying off 5% of teachers under a value-added
method and 7% using LIFO. Class sizes would increase by less than two students in each case,
and the increase in class sizes is about half a student greater under seniority-based layoffs. Along
the same lines, simulations based on actual layoffs in Washington State suggest that across the
state, districts would have saved about 13 teaching positions (about 9% fewer RIFed teachers) by
using VAMs to determine layoffs (Goldhaber & Theobald, 2013). Differences in the impacts on
class sizes between the two systems were likely small, though not reported. Analysis presented in
the next chapter shows that dismissal of about 2,308 teachers (2.5% of all teachers) results in an
increase in average class sizes districtwide of approximately 2 students. In sum, teacher layoffs
lead to class sizes increases, but by smaller amounts than might be expected.
Scholars are generally more concerned with the impact of layoffs, especially LIFO-based
layoffs, on the composition of the teaching workforce. The two studies cited above each find that
layoffs based solely on VAMs would have resulted in far fewer highly effective teachers being
laid off, while a far greater proportion of the least effective teachers would have been targeted
(Boyd et al., 2011; Goldhaber & Theobald, 2013). According to authors’ simulations, laid off
teachers are substantially less effective if selected using value-added measures of effectiveness
instead of seniority. The authors argue that these differences in the types of teachers targeted for

10
layoffs have major implications for student achievement. Although their results are speculative,
recent evidence from the Charlotte-Mecklenburg school district suggest that the specific teachers
selected to be laid off affects average student achievement the following school year (Kraft,
2015). Kraft (2015) finds that while layoffs generally lowered achievement the following school
year, the decline was reduced when principals laid off a less effective teacher.
Indirect Impact of the Layoff Process on Teachers and Students
Although the work reviewed above is critical for understanding how layoffs impact the
quality and size of the workforce, none of these papers assesses the impacts of the layoff process
beyond what occurs as a result of actual layoffs. In particular, little attention has been paid to the
impact of “surviving” a layoff. In other words, how do teachers who are faced with the layoff-
induced threat of job loss react in terms of their mobility, and how does layoff threat impact
teachers’ effectiveness and student outcomes?
A recent paper by Goldhaber and colleagues (2015) assesses the first part of this
question. They ask how layoff threat impacts teachers’ propensities to exit their schools, and find
that in Washington State and LAUSD, teachers who receive a RIF notice are significantly more
likely to exit their schools. However, it is unclear if this mobility is caused by teachers’ voluntary
responses to layoff threat or by the structural requirements stemming from seniority-based
layoffs (i.e., teachers who are RIFed but not let go are shuffled around schools to fill vacancies
left by laid off teachers). The authors also assess teachers’ responses to layoff threat stemming
from teachers’ colleagues being laid off, and find that, in LAUSD, teachers with greater peer-
induced layoff threat are more likely to exit their schools. Regardless of the underlying causal
mechanism of increased teacher turnover, the study shows that layoffs indirectly lead to a

11
substantial increase in school-level churn, which typically hurts school culture and student
outcomes.
While no prior work directly assesses the impact of the layoff process on teacher
effectiveness, the fact that layoff threat induces teacher churn may harm both student and teacher
outcomes. Sepe and Roza (2010), in their study of layoffs on low-income students of color, argue
that teacher churn in schools brought on by layoffs destroys established relationships between
teachers and students, families and teachers, and teachers and administrators. The authors further
contend that turnover destabilizes schools and negatively impacts learning environments. These
arguments follow a set of studies documenting the potential negative impacts of teacher churn
school climate and student achievement (Guin, 2004; Ronfeldt, Loeb, & Wyckoff, 2013;
Hanushek & Rivkin, 2013). For example, Guin conducts case studies of the perceived impact of
teacher turnover. Teachers and principals report that high rates of teacher turnover make it more
difficult for staff to build relationships with students, harm teacher collaboration, and result in
the loss of investments in teacher professional development. More recent work based on New
York City and Texas schools directly links grade- and school-level teacher churn to decreases in
student achievement on standardized exams (Hanushek & Rivkin, 2013; Ronfeldt et al. 2013).
In a similar vein, poor school working conditions negatively impact teacher productivity
(e.g., Hannaway, Sass, Figlio & Feng, 2009; Kraft & Papay, 2014). For example, Johnson, Kraft
and Papay (2012) find that teachers exhibit lower returns to experience in schools they perceive
to have poor working conditions. Other studies have documented that teachers prefer working in
schools where they perceive a supportive working environment and strong administrative
support (Boyd et al., 2011; Ladd, 2011). In short, this research suggests that, by causing churn

12
and poor working environments, the layoff process is likely detrimental to teacher productivity
and student outcomes.
Literature from the field of psychology also suggests that teachers may react to
employment threat in ways that could diminish their effectiveness and, as a result, student
achievement. In particular, research has found that workers are less motivated and thus less
productive when their jobs are threatened (e.g., Allen, Freeman, Russell, Reizenstein & Rentz,
2001; Brockner, Grover, Reed, & DeWitt, 1992). For instance, Brockner, Davy and Carter
(1985) conduct psychological studies of simulated layoffs in lab settings and find that exposure
to layoff threat decreases motivation and self-esteem. Research participants reported high levels
of “worry” when they had already been exposed to past layoffs and when they perceived that
layoff processes were conducted “unfairly.” Using surveys of workers, Brockner et al. (1993)
find that employees report the highest levels of motivation when they perceive moderate levels
of layoff-induced job threat, but report lower levels of motivation when faced with high levels of
job insecurity.
Altogether, the literature that can be applied to the research question at hand suggests that
teacher effectiveness may diminish when teachers are faced with layoff-induced job insecurity.
In addition, there are pragmatic reasons that go beyond the literature reviewed above that may
explain why teacher productivity would fall as a result of RIFs and layoffs. In many cases, laid
off teachers did not receive indication that they would be offered a job the following school year
until just before the beginning of the year, or, in some cases, several weeks into the year.
Consider that teachers typically spend at least some of their summer months preparing for the
next school year, attending district professional development, and engaging in other useful
activities intended to improve their instruction in the following year. Teachers participate in

13
these activities with the comfort of knowing they will have a job when the following school year
begins. Teachers who believe that they do not have a job in the coming year may spend less time
in preparation and improvement activities, and more time searching for new employment. Job
search activities likely do not contribute to improvements in teaching performance in the
following year.
If the productivity of teachers who have been RIFed or laid off does indeed diminish in
the year after they have received RIF and layoff notification, then policymakers will need to
consider how to amend current layoff policies and structures to decrease the harm caused to
teachers and their students by layoffs and the larger layoff process. In the next section, I describe
how policymakers can learn from the LAUSD context, in which layoffs–and more importantly
the layoff process–were implemented for four years as a result of the Great Recession.
The Layoff Process in LAUSD: 2008-09 through 2011-12
Table 1 shows the impact of the layoff process in LAUSD across the four years of
layoffs, from 2008-09 through 2011-12. LAUSD sent RIF notices to an average of 13.3% of
teachers each year, with the lowest proportion (8.2%) occurring in 2009-10, the year that the
federal government disbursed relief funds through the American Reinvestment and Recovery Act
and the Education Jobs Fund. Importantly given the topic under study in this paper, Table 1
shows that the majority of teachers who received RIF notices were eventually not released from
their positions. In particular, 68.6% of RIF notices that were distributed were rescinded and
48.1% of teachers who were actually laid off returned to the district the following year.
4
These

4
It is not possible to differentiate between teachers who were laid off and offered a job back, but declined to accept
the position, and those who were laid off, offered a job back, and accepted the position. The data show which
teachers received RIF notices and which teachers were laid off, but not which teachers were rehired. If teachers
appear in the dataset as a classroom teacher in the year following a layoff, I assume they were offered a teaching job
and that it was accepted. If teachers do not return to the dataset in the year following a layoff, I assume they were
not offered a position. It is possible, however, that some teachers who were laid off and do not show up in the data

14
rates of rescission are not substantially dissimilar to those of districts across California. As
reported by the Legislative Analyst’s Office, approximately 9% of California teachers received
initial RIF notices in school years 2007-08 to 2011-12, 51% of those were rescinded, and while
the other 49% were laid off, 57% of those laid off teachers were rehired the following year
(Estrada, 2012).
As noted earlier, California state law requires districts facing budget constraints to lay off
teachers in order of reverse seniority within teaching area. The law holds that seniority must
dictate layoffs unless the district can demonstrate “a specific need for personnel to teach a
specific course of study” and that the more junior employee “has special training and experience
necessary to teach that course or course of study” (California Education Code Sections 44955-9).
In other words, district administrators may take into account specific programmatic needs
beyond basic teaching credentials, but they cannot consider other factors such as teachers’
effectiveness, evaluations, or rates of absenteeism. I return to this point in the discussion of
analytic methods, as the LIFO process in use in LAUSD enables me to assume that teachers are
not selected for RIF or layoff due to their effectiveness, and especially not due to their
effectiveness as measured by VAMs.
These requirements are reflected in the distribution of RIFs and layoffs shown in Table 2.
As expected, the mean experience level of teachers who were laid off was substantially below
the mean experience of teachers who received a RIF but had it rescinded, which was in turn less
than the experience level of teachers who were unaffected by RIFs altogether. Similarly,
elementary teachers were more likely to receive RIFs as compared to middle and high school
teachers. Those with special education, math, and science credentials were less likely to receive

in the following year were offered reemployment and declined. However, this limitation has no bearing on the
current study, as I am interested in what happens to teachers who return to the district after a RIF or a layoff.

15
RIF or layoff notices.
5
Table 2 also demonstrates that the distribution of RIFs was not generally
related to value-added measures of effectiveness, after controlling for teacher experience. The
first value-added measure, one-year fixed effects, (used in the regression analyses) is based on
student achievement gains attributable to teachers, estimated separately in each year, without
controls for teacher experience. Consistent with the literature (e.g., Harris & Sass, 2011), I find
that this measure of effectiveness is correlated with experience, especially in the early years of
teaching. As a result, teachers who were laid off and not rehired have significantly lower value-
added scores compared to all other teachers. Both the second and third value-added measures
pool multiple years of data and use Empirical Bayes shrinkage. The third set of estimates shown
in Table 2 is also adjusted for experience. When I compare measures of teacher effectiveness
using pooled, shrunken VAMs across layoff conditions.
6
As I describe in further detail below, I
use the non-shrunken one-year fixed effects VAM estimates in all of my analyses. I include
experience as a predictor in all models (so that the distribution of RIF notices is unrelated to
VAM, conditional on observables) and weight observations by the standard error of the VAM
estimate.
Data and Analytic Approach
As is made clear above, the context in LAUSD affords a rich case in which to explore the
impacts of the layoff process on teacher effectiveness. I estimate a teacher fixed-effects
regression model to examine how individual teachers’ effectiveness changes after the receipt of a
RIF notice that is rescinded or receipt of a layoff notice, in both the current and prior school year.
I measure teacher effectiveness, my outcome variable, through value-added measures of

5
Although not shown here, my sample of teachers with value-added measures includes a greater proportion of
elementary teachers compared to the overall composition of teachers, reflecting the distribution of teachers for
whom VAM scores can be generated.
6
Note however that even in the experience-adjusted VAMs, teachers who are RIF-rescinded have slightly higher
(though not significantly different) VAM scores.

16
teachers’ contributions to their students’ achievement on standardized tests. I elaborate on the
data and identification strategies in the remainder of this section.
Data
For this essay, I draw on all seven years of the panel administrative data from school
years 2006-07 to 2012-13, which includes information about students, teachers, and schools in
LAUSD. These data consist of anonymized employee demographic information including
race/ethnicity, gender, years of experience, highest-degree earned, school and classroom
placement, layoff status, job title, and all credentials and certificates held by each certificated
non-administrative employee. All employee data are linked to anonymized student data that
includes California Standardized Test (CST) scores in math and English Language Arts,
race/ethnicity, gender, grade level, school and classroom placement, free lunch status, disability
(if any), English language proficiency, home language, and course enrollment information.
I supplement these LAUSD administrative data with public school-level data from the
California Department of Education. In this essay, I use school-level measures of total student
enrollment, and the percent of students that identify as the following race/ethnicities: African
American; Asian; Filipino; Hawaiian / Pacific Islander; Latino/a; and White. I limit the dataset to
K-12 district schools and dependent charter schools that operate within the district’s collectively
bargained employee contract (and are thus subject to LAUSD RIF and layoff processes). I
exclude non-traditional schools including community day schools, alternative schools, early
education centers, and special education centers because teachers at these schools were generally
protected from layoffs and because it is difficult to derive comparable VAM estimates for many
of the teachers in these schools.

17
I estimate VAMs for teachers in grades 4-7 because I can run the same model
specification for all teachers in this grade span.
7
I run my main analysis on school years 2007-08
through 2012-13 and use the 2006-07 school year data for specification checks.
8
This sample
consists of 31,160 teacher-year observations (19.57% of all teachers-year observations) and for
school year 2012-13, the most recent year of data used in the analysis, the sample includes 4,722
unique teachers.
9
In the following subsection I describe my value-added modeling approach and
the model specification for my primary analysis.
Analytic Approach
Although there is no universally accepted method for estimating teacher value-added
effects (e.g., McCaffrey et al., 2009, Ishii & Rivkin, 2009; Kane & Staiger, 2008; Rothstein,
2009), there are several approaches that represent those commonly utilized in the value-added
literature. I estimate the following model separately for each school year from 2006-07 to 2012-
13, indexing for student i, teacher j, and school s:
(1)

is a measure of student i’s achievement on the math CST, standardized within year and
test. I control for lagged test scores in both math (

) and ELA (

), and a vector of
student characteristics (Xijs) that includes family income, as measured by eligibility for free and
reduced price lunch, English language proficiency, whether English is spoken in the home,
whether the student is enrolled in any additional math courses (or English course for the ELA

7
Because of K-3 class size reduction in California, third grade teachers have classes with significantly lower class
sizes compared to those in other grades. Beginning in grade 8, students in the same course can take different exams
and I therefore omit teachers from the sample.
8
The district only has teacher demographic data available beginning in school year 2007-08. I am able to impute
experience, education, and credential information one year back for the majority of teachers; however, I exclude
these data in my main analysis because I am unable to impute these variables for all 2006-07 teachers.
9
I consider teachers to be those personnel who are labeled as “teachers” in the district datasets, who are recorded
teaching at least one class in the school year, and who teach at least eight students with current and prior year test
score data available.
A
ijs
Math
 a
1
A
ijs(t 1)
Math
 a
2
A
ijs(t 1)
ELA
 a
3
X
ijs
 
js
 
ijs

18
model), whether the student is in her or his first year in the district, and indicators for students’
grade level. Teachers’ VAM scores are observed in the term js, a vector of teacher fixed effects.
The error term, εijs, is assumed independently identically distributed with respect to the other
variables in the model. I run an identical model for ELA (again using both math and ELA lagged
test scores as controls). For teachers that have value-added scores in both subjects in the same
year, I take the average of the two scores.
I use one-year models rather than pooling data over time because I am interested in
teachers’ measure of effectiveness in each year they teach, and the possibility that their yearly
performance is affected by layoff threats. In order to test the robustness of this value-added
model to alternative specifications commonly employed in the value-added literature (e.g.,
Ehlert, Koedel, Parsons, & Podgursky, 2014; Koedel, Mihaly, & Rockoff, 2015), I run several
variations of this basic model and assess the correlations among these various specifications. The
alternate specifications include: 1) pooling all years of data and including year fixed effects
(Koedel & Betts, 2007; McCaffrey, Sass, Lockwood, & Mihaly, 2009); 2) removing the student
covariates and adding a twice-lagged measure of student achievement, as suggested by Rothstein
(2010); and 3) adding classroom and school covariates to control for peer effects (Lubienski &
Lubienski, 2013). Consistent with other studies (Chetty, Friedman & Rockoff, 2014; Ehlert et al.,
2014; Goldhaber, Walch & Gabele, 2014), the teacher effects generated by these models are
highly correlated (0.90 or above). The standard deviation of the one-year model is 0.274, which
is in line with past VAM estimates based on LAUSD data (Kane & Staiger, 2008).
My primary research question asks how teachers’ job performance changes after
experiencing a layoff or threat of layoff. I begin the analysis by fitting teacher fixed effects
models that predict teachers i’s value-added effectiveness – derived from equation (1) above – in

19
school s in year t (denoted VAist), based on both their current and prior year RIF and layoff
experience:

(2)
10

My four variables of interest are RIFreist and Layoffist and their lagged versions, RIFreist-1 and
Layoffist-1, which indicate whether teacher i received a RIF notice that was later rescinded, or
received both an initial RIF notice and a final layoff notice, respectively, in the current prior year
of teaching.
11
The coefficients β1 – β4 address my primary research question of how exposure to
threat of layoff or actual layoff is associated with teaching performance in the current and
subsequent year of teaching.
Finally, the inclusion of teacher fixed effects, τis, in equation (2) allows me to examine
changes in teacher effectiveness within teachers over time. Thus estimates of a teacher’s job
performance are relative to that same teacher’s performance in a typical year. The use of teacher
fixed effects helps control for any unobserved time-invariant factors related to teacher
effectiveness that are not included in equation (2). By comparing teachers to their own
performance in the years they are not RIFed, the coefficients β1 – β4 provide arguably causal
estimates of the effects of RIFs and layoffs because their distribution to teachers over time is
random after controlling for teaching area and experience level.
The vectors Xist and Sst include control variables for time-variant teacher and school
characteristics, respectively. At the teacher level, I control for indicator variables for experience
(1, 2, 3, 4, 5, 6-7, 8-10, 11-13 and the reference group is 14 or more years), whether the teacher

10
For simplicity, I show only the equation used to estimate Model 1, while noting that I run a number of additional
specifications discussed in the text.
11
For lagged RIF and layoff variables, the reference category is teachers who did not receive a RIF notice in year t-1
because I also control for teachers who were not present in the prior year). The reference category for the current
year RIF and layoff is teachers who did not receive a RIF notice in year t.

20
acquired any endorsements such as special education (SPED), science or math (STEM), or other
non-elementary endorsements, and whether the teacher acquired a master’s degree. My school-
level controls include the log of school enrollment, the percent of students that identify as an
underrepresented minority, and the school type (elementary, middle, high school, or span
school).
12
Finally, I include year fixed effects to remove any changes in teacher effectiveness that
are idiosyncratic to a particular year.
13
In all models, I weight observations by the inverse of the
standard error of the value-added estimate and cluster standard errors at the teacher level.
Weighting observations allows teachers with more precise estimates of their teaching
effectiveness (i.e. those with lower standard errors) to contribute more towards the estimation of
my parameters of interest, while those with less precisely estimated measures of effectiveness
contribute less to the parameter estimates.
I conduct a number of robustness tests to support a causal interpretation of my findings.
The first assesses whether the negative effects associated with being laid off and rehired are
driven by teachers being placed in a new grade level, school, or job category (i.e., full-time
teacher vs. long-term substitute). To run these tests I first run three separate models controlling
for whether the teacher is in a new grade, whether she or he is in a new school, and whether the
teacher returned from the prior year as a long-term substitute. Next, I run three additional models
in which I interact each of these variables (new grade, new school, new job category), with the
lagged layoff variables. These interactions models assess whether the effect of being laid off and
returning the following year differs for teachers who return to a new grade, school, or job
category.

12
While most teachers remain at the same school type over their window of observation, some switch school types
and these dummies remove any bias associated with these school-level switches.
13
Note that my preferred value-added measures are one-year teacher fixed effects, which further controls for time
trends that may affect measures of effectiveness.

21
Another explanation for why teachers’ measure of effectiveness declines following a
layoff is reversion to the mean. That is, teachers may have a year with an exceptionally high
value-added score, receive a layoff notice, and upon returning to the district the following year,
see their value-added score revert back toward their mean. In order to test for this reversion to
the mean, I run models that control for the number of students from which value-added measures
are based and then interact this variable with the lagged layoff indicators.
14
As an additional test
of internal validity, I run all models with and without teacher fixed effects.
I also conduct a placebo test in which I estimate the effect of a pseudo RIF or layoff in
the current and prior year, for teachers who were likely to be assigned to these conditions before
the district began implementing layoffs. To do this, I use the following procedure: first, I predict
the likelihood that a teacher receives an initial RIF notice during the years in which layoffs
actually took place (2008-09 to 2011-12). For teachers who did receive an initial RIF notice, I
also predict their likelihood of receiving a layoff notice. I then make an out-of-sample prediction
of each teacher’s likelihood of receiving an initial RIF and final layoff notice, based on their
observable characteristics in the years before layoffs (2006-07 and 2007-08). These results show
each teacher’s likelihood of receiving a RIF or layoff notice both before layoffs took place and
during the period of layoffs. Next, I generate a random number and assign teachers to either the
No RIF or the RIF notice condition, based on whether their random number is above their
predicted likelihood of receiving a RIF notice. For those teachers who are predicted to have

14
As described in greater detail in Kane and Staiger (2002), there is greater variability in estimates derived from
value-added models when they are based on fewer observations. The same is true in my data. Teachers are more
likely to have extremely high or low measures of effectiveness if their value-added estimate is based on fewer
students. Therefore, reversion to the mean would most likely happen for teachers with fewer students (Kane &
Staiger, 2002). In models that predict a teachers’ value-added score, the number of students from which the VAM is
based (usually between 25 and 30 students for elementary teachers and between 75 and 200 students for middle
school teachers), is positive and significant, but that the magnitude is zero up to the thousandth decimal. As I
describe further below, none of these additional covariates substantially alter the significant negative effects I find
for layoff in t-1.

22
received a RIF notice, I assign them to the layoff condition if their likelihood of receiving a
layoff is greater than their random number. Following these steps allows me to identify, based on
observable characteristics, which teachers would have been RIF-rescinded and laid off in the
years prior to layoffs. The final step to this procedure is to estimate the effect of receiving a
pseudo RIF or layoff notice in year t and t-1, for the two years prior to layoffs actually taking
place (2006-07 and 2007-08).
15

Before running the actual placebo test, I examine how well the assignment to these
pseudo RIF and layoff conditions in the pre-layoff years matches the actual likelihood that
teachers would have been RIF-rescinded or laid off. There are several ways to test the validity of
assignment to pseudo RIF and layoff conditions. First, I can compare the proportion of teachers
assigned to pseudo RIF and layoff conditions in the pre-layoff years to the proportion of teachers
who actually fell into these conditions during the layoff years. I find that during the two years
preceding layoffs, 2006-07 and 2007-08, an average of 12.4% and 11.8% of teachers would have
received a RIF notice and had it rescinded, respectively, while an average of 9.5% and 7.7%
would have been laid off. These figures, shown in Panel B of Appendix Table A1, are roughly in
line with the actual percent of teachers RIF-rescinded and laid off, which is shown in Panel A of
Appendix Table A1. For example, during the first year of actual layoffs, 2008-09, 10.5% of
teachers were RIF-rescinded and 6.2% were laid off. A slightly larger proportion of teachers are
assigned placebo RIF and layoff conditions in the years prior to layoffs compared to the 2008-09
proportions because the district’s teaching workforce was generally more junior, on average, in
the 2006-07 and 2007-08 school years (and therefore more likely to be RIFed).

15
The placebo test supports a causal interpretation of the findings if the coefficients for the pseudo RIF and layoff
variables show no relationship with teachers’ predicted measure of effectiveness. This placebo test procedure was
suggested by Dr. Dan Goldhaber as part of a related research project.

23
A second way to assess the validity of assignment to the pseudo RIF and layoff
conditions is to compare the proportion of teachers who were actually RIF-rescinded and laid off
during the layoff years to the proportions that were assigned to those pseudo conditions during
the same years. As shown in Appendix Table C1, approximately the same proportion of teachers
are assigned to the placebo RIF and layoff conditions each year as were actually RIF-rescinded
and laid off. For example, over the four years of layoffs, this process assigns 9.6% and 4.4% of
teachers to the RIF-rescinded and layoff conditions, respectively, while 9.1% and 4.2% were
actually RIF-rescinded and laid off during the same time period.
Finally, I can examine the overlap between the pseudo RIF-rescinded and layoff
conditions and the actual RIF and layoff conditions during the years in which layoffs actually
took place. There appears to be substantial overlap between the real and placebo RIF and layoff
conditions. For example, I find that between one third to one half of teachers who received an
actual RIF notice each year were also assigned a placebo RIF notice. To summarize, this
procedure appears to accurately predict the individual teachers who would have received a RIF
and layoff notices in the years prior to layoffs taking place. In the final step of the placebo test, I
estimate the effect of receiving a pseudo RIF or layoff notice in year t and t-1, for the two years
prior to layoffs actually taking place (2006-07 and 2007-08).
The models I use in this study provide unbiased estimates of the impact of RIFs and
layoffs if the selection of teachers into these conditions is unrelated to teachers’ measures of
effectiveness, after controlling for observable teacher and school characteristics. I argue that the
models provide unbiased estimates because the distribution of RIF notices and layoffs, as well as
the selection of which teachers to rehire, are highly regulated by the state’s Education Code.
When I run separate models that predict RIFs and layoffs (not shown), I find that the likelihood

24
of treatment is related only to experience and credential area, both of which I control for in all
models.
As described earlier, the California Education Code provides clear guidelines on how
layoffs are to be determined and the order in which laid off teachers are offered their jobs back.
Administrators have limited flexibility in deciding which teachers are laid off beyond
consideration of credential and teaching position. In addition, principals had no way of
ascertaining their teachers’ VAMs. LAUSD provided teachers with their VAMs (estimated by
the Value-Added Research Center at the University of Wisconsin) for a subset of the years in
which RIFs and layoffs were implemented, but explicit regulations prevented the district from
providing principals with their teachers’ VAMs. The data confirm that LAUSD complied with
the California Education Code in implementing seniority-based layoffs.
One final threat to the internal validity of my identification strategy is selection on which
laid off teachers return to the district. This form of selection could bias estimates in one of two
ways. On the one hand, if administrators deviated from state-mandated re-hiring policies, for
example, by offering jobs back only to teachers who were expected to have better than average
performance in the following year, such discrepancies would most likely bias estimates upward.
Therefore, any negative relationship I identify between past layoffs and current job performance
will be underestimated if teachers were offered their job back based on expected future
effectiveness.
On the other hand, selection bias could result from teachers’ own selection back into the
LAUSD workforce once they have been laid off. If laid off teachers are offered their job back,
they have discretion in whether to return to the district. My estimates of the impact of layoff on
effectiveness could be negatively biased if teachers’ choice to return to the district is related to

25
their future effectiveness. However, even if selection is driving the fact that teachers who are laid
off and then return to LAUSD are less effective in the following year (relative to their typical
year), this adverse selection is still relevant from a policy perspective. That is, if the layoff
process creates a sorting mechanism in which teachers who are more likely to have a below
average year are more likely to return to the district, while more effective teachers become more
likely to exit, then the layoff process still causes negative consequences for students.
16
With
these caveats in mind, I now turn to my results.
Findings
Table 3 shows the results of my preferred model (Model 1, henceforth the baseline
model) as well as models that include controls for teacher mobility (Models 2 and 3). All models
include the covariates discussed above, but I only report the variables of interest as well as the
coefficients for experience (results of the full models showing all covariates are available from
the author upon request).
The coefficients on my control variables are consistent with the literature, (e.g., Feng &
Sass, 2011; Goldhaber, Gross, & Player, 2011; Hanushek, Kain & Rivkin, 2004; Harris & Sass,
2011; Ost, 2014). Experience is positively related to measures of effectiveness, especially in the
early years of teaching. The returns to experience continue at least until the 10th year of
teaching. I find that teachers who acquire graduate degrees (and in models without fixed effects,
those that have graduate degrees) are only slightly more effective than those with bachelor’s
degree and these results are generally not significant. Finally, in the years in which a teacher
switches schools or grade levels, skips a school year, or is new to the district, they are generally

16
Conversations with district personnel in the Human Resources division note that most laid off teachers who are
offered reemployment accept these offers and return to the district the following school year (personal
communication, 2015).

26
less effective compared to the years in which those same teachers are in the same classroom
placement as the prior year.
The first model shown in Table 3 (the baseline model) predicts a teacher’s measure of
effectiveness, based on her or his current and prior year RIF or layoff status. The results of the
baseline model are consistent when I control for teacher mobility in Models 2 and 3. Current
year RIFs and layoffs have no relationship with teacher productivity. However, compared to
teachers who were not RIFed the prior year, those who are laid off, but rehired have lower
classroom effectiveness, relative to their typical year. The magnitudes on these coefficients are
both statistically and educationally significant. Given the overall standard deviation of value-
added of 0.276 for years 2007-08 to 2012-13, the coefficient of -0.061 shown in the baseline
model implies that teachers laid off and rehired experienced a decline in effectiveness of 22.1%
of a student-level standard deviation, an amount equivalent to approximately two months of lost
learning time (Lee & Finn, 2010). Teachers threatened with a layoff in the prior year, i.e., those
that received a RIF notice in March of the prior year, but had it rescinded before mid-May of that
year, did not experience changes in their effectiveness compared to those who were not RIFed.
In Table 4, I present a number of additional specifications and placebo tests to assess the
robustness of these results. I begin by showing the baseline model from Table 3 for comparison.
Then in Model 2, I run an identical specification, but remove the teacher fixed effects. My results
are similar in direction and statistical significance, although the magnitude on layoff in t-1
declines by about a third.
17
In Models 3 and 4, I run the same specification with and without
teacher fixed effects, respectively, but this time I do not control for current year RIFs and

17
Note that RIF-rescinded in year t becomes positive and significant when I remove the teacher fixed effects. This
happens because, on average and even after adjusting for experience, RIF-rescinded teachers are more effective than
non-RIFed and laid off teachers. This can also be seen in Table 2, which reports descriptive statistics for one-year
VAMs, pooled-year VAMs, and experience-adjusted pooled-year VAMs.

27
layoffs. Again, my result for layoff in t-1 is significant and the non-fixed effects model shows a
similar drop in magnitude, this time declining by about 23%. Models 5 and 6 control for the
number of students from which the teacher’s VAM is based. As I discussed in the section above,
these models test whether reversion to the mean is driving the negative effects for teachers who
were laid off and rehired. Again these models are consistent with the main results found in the
baseline model. I also run models that control for whether the teacher returns to the district in a
reduced FTE position or as a long-term substitute. For all of the covariates just mentioned
(indicators for teacher mobility, current year RIF and layoff, and number of students), I also
interact each with the lagged layoff variables and the results remain consistent (not shown, but
available from the author upon request). Finally, results are also consistent when I substitute a
continuous experience variable and its square for the categorical experience variables.
The last four models shown in Table 4 (Models 7 through 10) are placebo tests based on
the years prior to layoffs (2006-07 and 2007-08). In each of these models, the lagged layoff
coefficient is not significantly different from zero. In models that remove the teacher fixed effect
(Models 8 and 10), I do find a positive and significant correlation between pseudo-RIF-rescinded
in year t-1and current year VAM. However, because this model is only run on two years of data,
and because teachers must be employed in the district the prior year in order to have been RIFed
in the prior year, first year teachers in this model (those with zero years of experience) are far
more likely to be new to the district in year t, rather than pseudo-RIFed in t-1 (in the main
models, many teachers are new to the district, but enter with five or more years of teaching).
Because teachers make their largest gains in VAM after their first year on the job, the positive
coefficient shown on pseudo-lagged-RIF-rescinded in Models 8 and 10 likely reflects positive

28
gains from the first year of teaching. In sum, the overall null results of the placebo tests suggest
that the findings shown in the baseline model reflect an underlying causal relationship.
Discussion and Policy Implications
To my knowledge, this paper is the first to assess the impact of the layoff process on
teachers’ own effectiveness. Previous literature has examined how layoffs based on LIFO impact
the overall effectiveness of the teacher workforce, but none to date has examined how the receipt
of a RIF or a layoff followed by a return to the workforce impacts teachers’ contributions to their
students’ test scores. The results show that the layoff process does more than simply remove
effective teachers from schools and increase class sizes. Rather, the general lack of job security
brought on by the layoff process negatively affects teachers’ productivity.
These results are not surprising. For all the reasons I laid out based on the extant
literature, as well as mere common sense, one should expect to see diminishing productivity
from teachers who lose their jobs and remain unemployed for a period of time before being
rehired to the school district. Perhaps what is more surprising is policymakers’ failure to forecast
or recognize this impact.
There are multiple ways that policy might be amended to mitigate the potentially harmful
and disparate effects of RIFs and layoffs on teacher effectiveness. The simplest way would be to
maintain a source of additional revenue that can be used in times of emergency. Alternatively, as
the California Legislative Analyst’s Office has recommended (Estrada, 2012), the state could tie
the layoff notification deadline (currently May 15) to the state’s budget release, as is done in
several other states.
18
The underlying challenge for districts is that they do not receive the final

18
Districts in New Hampshire are required to notify teachers of a possible layoff within 15 days of the adoption of
the district budget, although the latest possible date of notification is no later than approximately mid-May (N.h. rev.
stat. ann. S 189:14a). Similarly, Virginia requires districts to notify teachers of a possible layoff within two weeks of
their Board’s approval of the final district budget, but not later than June 1 (Vir. admin. code S 22.1-304). Finally,

29
state budget until June or July. Thus another potential policy intervention would be to require
states to develop their budget projections by a specified time, enabling districts to have better
information before they make staffing decisions. Of course, states are supposed to release final
budgets early in the year, but California, like other states, often misses this deadline with little or
no resulting consequence (Estrada, 2012). By enabling districts to develop more accurate budget
projections, and as a result more accurate staffing requirements, fewer teachers would be
impacted by the layoff process. At the same time, it should be noted that it is not clear from the
data presented in this study whether earlier notification of layoffs is beneficial to teachers.
Policymakers can also generate new policies to improve the inequitable distribution of
layoffs on traditionally disadvantaged students. One such policy was implemented in LAUSD as
a result of the lawsuit Reed vs. California. Plaintiffs in Reed sued LAUSD because particular
classifications of students (low-income and underrepresented minority) were adversely affected
by layoffs. The settlement resulting from Reed required that LAUSD protect teachers in a set of
schools from layoffs. In this way, students in the most impacted schools were protected from the
churn, instability, and, based on the results of this study, the diminished effectiveness associated
with the layoff process.
19

Finally, policymakers might soften requirements to RIF teachers in order of reverse
districtwide seniority and to re-hire in the opposite order. This will help assuage the negative
impact of the layoff process for two reasons. First, LIFO processes by necessity require laying
off more teachers than would be necessary under other selection regimes (junior teachers are less
expensive than more senior teachers). If districts used alternative selection mechanisms for

Mississippi requires districts to notify teachers of a budget-induced layoff within ten days after the Governor
approves the appropriation bill(s) for the state’s K-12 education budget (Miss. code ann. S 37-9-105).
19
In essay 2, I examine the inequitable distribution of RIF and layoffs notices in the first two years of layoffs and
then analyze the extent to which Reed increased equity in layoff distribution across student subgroups.

30
layoffs, fewer teachers, and as a result fewer students, would be impacted. Second, and perhaps
more importantly, because schools with greater proportions of low-income students and students
of color attend schools with more junior teachers, the LIFO process causes such students to be
disproportionately impacted by the layoff process. By diminishing the reliance on seniority in
layoff processes, policymakers might also reduce the inequitable impact of layoffs and RIFs on
low-income students of color.
The results from this paper also provide lessons for policymakers and researchers as we
consider policies that rely on accurate measures of teacher effectiveness. For instance, popular
teacher evaluation and tenure reforms require staffing decisions to be made in part based on
measures of teacher effectiveness. Given the results presented in this study, policymakers and
administrators should account for these dips in effectiveness when considering how to evaluate
and assess teacher performance in high-stakes settings. Similarly, researchers evaluating the
impact of reforms on teacher effectiveness may need to include controls for teachers’ prior RIF
and layoff status.
In all, this paper investigates whether there are hidden costs of the layoff process that
must be considered in the larger policy conversation around RIFs, layoffs, and teacher
effectiveness. It is my hope that researchers, policymakers, and administrators can learn from the
Great Recession and improve upon policies enacted during this time in ways that can lessen the
negative impacts of the next financial crisis.

31
Tables and Figures for the Introduction and Essay 1

FIGURE 1
Timeline of Events Affecting Los Angeles Unified School District, 2008 - 2015

32
TABLE 1
Proportion of teachers exposed to various layoff conditions, by year
2008-09 2009-10 2010-11 2011-12 Total
No-RIF 24,212 24,577 22,070 21,259 92,118
RIF-rescinded 3,064 1,826 2,492 2,315 9,697
Layoff-return 456 212 770 699 2,137
Layoff-no return 1,356 143 500 309 2,308
Total 29,088 26,758 25,832 24,582 106,260

% of teachers RIFed 16.8% 8.2% 14.6% 13.5% 13.3%
% laid off of those RIFed 37.2% 16.3% 33.8% 30.3% 31.4%
% let go of those RIFed 27.8% 6.6% 13.3% 9.3% 16.3%
% let go of those laid off 74.8% 40.3% 39.4% 30.7% 51.9%
Teachers in this table (and throughout the analysis) are defined as employees that are included the non-administrator
demographic file and are assigned a job title of K-12 teacher. In alternate specification and robustness checks, I add
to this group all employees that are responsible for reporting grades for students (i.e., all employees listed in the
district’s Marks dataset). The proportion in each RIF and layoff categories, as well as the overall findings, are
similar to the more conservative sample represented here. Those results are available from the author upon request.

33
TABLE 2
Proportion of teachers with various experience levels, credential profiles, and mobility
outcomes, by RIF/layoff category, 2008-09 to 2011-12

% Overall No RIF
RIF

RIF-
rescinded
Laid off-
return
Laid off-
no return
All Teachers 106,260
92,118 9,697 2,137 2,308
86.7% 9.1% 2.0% 2.5%
Experience / Education
Novice teachers (1st - 3rd year) 7.0% 61.0% 15.9% 6.9% 16.2%
Mid - career teachers (4th - 8th year) 22.1% 68.3% 23.6% 4.9% 3.2%
Veteran teachers (9th year or above) 70.9% 95.0% 3.9% 0.6% 0.5%
Mean years of experience 10.1 10.9 6.0 4.8 3.4
Master’s degree or higher 36.5% 87.2% 9.3% 1.8% 1.7%
Endorsement Area
Special Ed 14.7% 96.9% 1.7% 0.8% 0.7%
Health/PE 5.5% 92.3% 2.9% 1.1% 3.7%
Science 5.9% 92.3% 6.5% 0.5% 0.7%
Foreign Languages 3.1% 91.5% 4.0% 2.1% 2.4%
Math 7.3% 89.8% 9.2% 0.5% 0.6%
Agriculture/Tech/Other elective 4.4% 88.7% 5.9% 3.1% 2.3%
Social Studies 9.1% 87.6% 7.4% 2.7% 2.2%
English/LA 11.4% 85.4% 9.3% 2.3% 2.9%
Elementary Ed 42.7% 82.8% 12.7% 2.4% 2.2%
Arts 6.9% 81.8% 10.3% 4.5% 3.4%
Value-added measures
One-year fixed effects estimates
-0.02 -0.02 0.01 -0.01 -0.08
(0.274) (0.276) (0.274) (0.286) (0.228)
Pooled estimates with EB shrinkage
0.01 0.01 0.02 0.00 -0.06
(0.205) (0.207) (0.201) (0.195) (0.174)
Experience adjusted pooled estimates with EB
shrinkage
0.00 -0.01 0.04 0.03 0.01
(0.205) (0.207) (0.201) (0.198) (0.178)
Mobility outcomes
Stay at school 84.8% 88.2% 79.1% 55.4% 0.0%
Switch schools in district 7.2% 6.0% 12.5% 44.6% 0.0%
Leave teaching/district 8.0% 5.8% 8.4% 0.0% 100.0%
Switch to a non-teaching position (same school) 1.1% 1.1% 1.4% 0.0% 0.2%
Switch to a non-teaching position (diff school) 2.9% 2.4% 3.0% 0.0% 26.2%
Return as a long-term substitute (same school) 0.6% 0.2% 0.0% 0.0% 22.0%
Return as a long-term substitute (diff school) 0.3% 0.1% 0.0% 0.0% 9.9%
Leave district 3.0% 2.0% 3.8% 0.0% 41.8%
Note: except for the value-added panel, the % overall column sums to 100% within each panel. For teacher
experience/ education and endorsement by layoff status, rows sum to 100% within the four layoff categories. In the
mobility outcomes panel, columns sum to 100% for stay, switch, and leave teaching/district. The final five rows of
the table provide additional information about teachers who leave teaching and/or the district. Each of these columns
sums to the total amount that left teaching and/or the district. For example, 8.0% overall left teaching/the district,
and the percentages shown in the first column (% overall) add up vertically to 8.0%.

34
TABLE 3
Regression coefficients predicting teacher effectiveness based on prior year layoff experiences,
2007-08 to 2012-13

(1) (2) (3)
Year t-1 layoff threat
RIF-rescinded in t-1
-0.009 -0.008 -0.009
(0.006) (0.006) (0.006)
Laid off in t-1
-0.061*** -0.050*** -0.060***
(0.015) (0.015) (0.015)
Skipped year t-1
-0.059+ -0.037 -0.058+
(0.031) (0.032) (0.031)
Year t is first year in district
-0.031+ -0.035* -0.036*
(0.016) (0.016) (0.016)
Prior year mobility
New to school (but not district)
-0.034***
(0.006)
New to grade
-0.015***
(0.004)
Teacher characteristics
1st year (ref is >13th year)
-0.145*** -0.144*** -0.147***
(0.030) (0.030) (0.030)
2nd year
-0.099*** -0.099*** -0.100***
(0.023) (0.023) (0.023)
3rd year teacher
-0.095*** -0.096*** -0.096***
(0.021) (0.020) (0.020)
4th year teacher
-0.053** -0.054** -0.053**
(0.018) (0.018) (0.018)
5th year teacher
-0.047** -0.047** -0.046**
(0.017) (0.016) (0.016)
6th or 7th year teacher
-0.036** -0.036** -0.036**
(0.014) (0.014) (0.014)
8th - 10th year teacher
-0.020+ -0.020+ -0.020+
(0.011) (0.011) (0.011)
11th - 13th year teacher
-0.011 -0.010 -0.011
(0.008) (0.008) (0.008)
Master's or higher degree
0.008 0.009 0.008
(0.012) (0.012) (0.012)
Constant
0.388*** 0.438*** 0.397***
(0.074) (0.074) (0.074)
R-squared 0.727 0.728 0.728
Note: n=31,160 in all models. All models shown here control for time-varying teacher characteristics including
current year layoff status and teaching credentials as well as school-level covariates including the log of enrollment,
percent of students that identify as an underrepresented minority, and school level (elementary, middle, or high
school). Models also include year fixed effects. Robust standard errors shown in parentheses (clustered at the
teacher level).

35
TABLE 4
Additional specification checks of models predicting teacher effectiveness based on prior year layoff experiences, 2007-08 to 2012-13,
and placebo tests, 2006-07 and 2007-08

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Year t layoff threat
RIF-rescinded in t
0.003 0.013* 0.004 0.006 -0.005 0.012+
(0.006) (0.005) (0.006) (0.006) (0.016) 0.007
Laid off in t
0.000 -0.012 0.001 0.005 0.002 -0.009
(0.010) (0.008) (0.010) (0.010) (0.023) 0.009
Year t-1 layoff threat
RIF-rescinded in t-1
-0.009 0.009+ -0.009 0.014* -0.005 -0.004 -0.005 0.020* -0.010 0.020*
(0.006) (0.005) (0.006) (0.006) (0.006) (0.006) (0.016) 0.01 (0.021) 0.009
Laid off in t-1
-0.061*** -0.041*** -0.062*** -0.047*** -0.050*** -0.045** 0.013 -0.004 0.015 -0.003
(0.015) (0.012) (0.015) (0.013) (0.015) (0.015) (0.021) 0.011 (0.029) 0.011
Skipped year t-1
-0.059+ -0.125*** -0.059+ -0.129*** -0.049 -0.046
(0.031) (0.020) (0.031) (0.021) (0.031) (0.031)
Year t is first year in
district
-0.031+ -0.065*** -0.031* -0.065*** -0.030+ -0.048*** -0.050*** -0.051***
(0.016) (0.011) (0.016) (0.011) (0.016) (0.013) 0.011 0.011
N 31,160 31,160 31,160 31,160 31,160 31,160 10,782 10782 10,782 10782
R-squared 0.727 0.123 0.727 0.123 0.728 0.728 0.898 0.364 0.898 0.364
Teacher fixed effects X X X X X
Control for num. of stu. on
which VAM is based
X X
Placebo test X X X X
Note: Model 1 is the baseline model shown in Table 4 and Model 2 is an identical model that removes teacher fixed effects. Models 3 and 4 remove year t layoff
controls, with and without teacher fixed effects, respectively. Models 5 and 6 control for the number of students on which each teacher’s VAM is based, again
with and without teacher fixed effects. Models 7 through 10 are placebo tests showing that the lagged layoff coefficient is insignificant, with or without fixed
effects. Because the placebo tests are run only on two years of data, 2006-07 and 2007-08, no teachers could have skipped the prior year (and that variable drops
out of the model). For a similar reason, in placebo models with fixed effects, the coefficient for new to the district drops out because teachers who were new in
2007-08 were, by definition not present in 2006-07 and therefore there is no variation from which to estimate this coefficient. The reference category for year t
layoff threat is not RIFed in year t and the reference category for year t-1 layoff threat is not RIFed in year t-1.

36
References for Essay 1
Aaronson, D., Barrow, L., & Sander, W. (2007). Teachers and student achievement in the
Chicago Public High Schools. Journal of Labor Economics, 25(1), 95–135.
Allen, T. D., Freeman, D. M., Russell, J. E. A., Reizenstein, R. C. & Rentz, J. O. (2001).
Survivor reactions to organizational downsizing: Does time ease the pain?. Journal of
Occupational and Organizational Psychology, 74, 145–164. DOI:
10.1348/096317901167299
Angrist, J. D., & Lavy, V. (1999). Using Maimonides’ rule to estimate the effect of class size on
scholastic achievement. Quarterly Journal of Economics, 114(2), 533–575.
Blume, H. (2015, June 23). LAUSD board OKs $7.8-billion budget that includes hundreds of
layoffs. Los Angeles Times. Retrieved July 31, 2015 from:
http://www.latimes.com/local/education/la‐me‐lausd‐budget‐20150624‐story.html.
Boyd, D., Grossman, P., Ing, M., Lankford, H., Loeb, S. & Wyckoff, J. (2011). The influence of
school administrators on teacher retention decisions. American Educational Research
Journal, 48(2), 303-333. DOI: 10.3102/0002831210380788
Boyd, D. J., Langford, H., Loeb, S., & Wyckoff, J. H. (2011). Teacher layoffs: An empirical
illustration of seniority vs. measures of effectiveness. Education Finance and Policy,
439-454
Brockner, J., Davy, J., & Carter, C. (1985). Layoffs, self-esteem, and survivor guild:
Motivational, affective, and attitudinal consequences. Organizational Behavior and
Human Decision Processes, 36, 229-244.
Brockner, J., Grover, S., Reed, T., Dewitt, R.L. (1992). Layoffs, job insecurity, and survivors'
work effort: Evidence of an inverted-U relationship. Academy of Management Journal,

37
2(35), 413-425.
Bureau of Labor Statistics. (2012). Employment situation archived news releases. Table B-1.
Retrieved from http://www.bls.gov/schedule/archives/empsit_nr.htm.
Chetty, R., Friedman, J., & Rockoff, J. (2014). Measuring the impacts of teachers II: Teacher
value-added and student outcomes in adulthood. American Economic Review, 104(9),
2633–2679. DOI: 10.1257/aer.104.9.2633.
Cho, H., Glewwe, P., & Whitler, M. (2012). Do reductions in class size raise students’ test
scores? Evidence from population variation in Minnesota’s elementary schools.
Economics of Education Review, 31, 77–95.
Estrada, J. (2012). A review of the teacher layoff process in California. California Legislative
Analyst’s Office. Retrieved August 4, 2013 from
http://www.lao.ca.gov/reports/2012/edu/teacher-layoffs/teacher-layoffs-032212.aspx
Feng, L., & Sass, T. (2011). Teacher quality and teacher mobility. Working Paper 57.
Washington, D. C.: National Center for Analysis of Longitudinal Data in Education.
Goldhaber, D., Gross, B., & Player, D. (2011). Teacher career paths, teacher quality, and
persistence in the classroom: Are public schools keeping their best? Journal of Policy
Analysis and Management, 30, 57–87.
Goldhaber, D., & Hansen, M. (2010). Using performance on the job to inform teacher tenure
decisions. American Economic Review, 100(2), 250-255.
Goldhaber, D., Strunk, K. O., Brown, N. & Knight, D. S. (2015, March). Lessons learned from
the Great Recession: Layoffs and the RIF-induced teacher shuffle. Paper presented at the
Annual Conference for the Association for Education Finance and Policy, Washington,
D.C.

38
Goldhaber, D., & Theobald, R. (2013). Managing the teacher workforce in austere times: The
determinants and implications of teacher layoffs. Education Finance and Policy, 8(4),
494-527.
Gordon, L. (2011, April 3). Today's teacher layoffs threaten tomorrow's college classrooms. Los
Angeles Times. Retrieved June 23, 2015 from:
http://articles.latimes.com/2011/apr/03/local/la-me-teaching-20110404.
Green, E. (2015, July 3). Rahm Emanuel: ‘I Am Not an Education Reformer’. The Atlantic.
Retrieved July 31, 2015 from:
http://www.theatlantic.com/education/archive/2015/07/rahm-emanuel-chicago-education-
reform/397673/.
Guin, K. (2004). Chronic teacher turnover in urban elementary schools. Education policy
analysis archives, 12(42), Retrieved May 17, 2013 from
http://epaa.asu.edu/epaa/v12n42/.
Hanushek, E. A., Kain, J. F., & Rivkin, S. G. (2004). Why public schools lose teachers. Journal
of Human Resources, 39(2), 326-354.
Hanushek, E. A., & Rivkin, S. G. (2013). Teacher turnover and the quality of instruction in urban
schools. Unpublished Manuscript.
Hanushek, E. A., & Rivkin, S. G. (2012). The distribution of teacher quality and implications for
policy. Annual Review of Economics, 4(1), 131-157.
Harris, D. N. & Sass, T. (2011). Teacher training, teacher quality, and student achievement.
Journal of Public Economics, 95, 798-812.
Ishii, J., & Rivkin, S. (2009). Impediments to the estimation of teacher value added. Education
Finance and Policy, 4(4), 520-536.

39
Jepsen, C. & Rivkin, S. (2009). Class size reduction and student achievement: The potential
tradeoff between teacher quality and class size. Journal of Human Resources, 44(1), 223-
250.
Johnson, S., Kraft, M., & Papay, J. (2012). How context matters in high-need schools: The
effects of teachers’ working conditions on their professional satisfaction and their
students’ achievement. Teachers College Record, 114(10), 1-39.
Kahn, C. (2009, June 17). L.A.'s urban schools hardest hit by teacher layoffs. National Public
Radio. Retrieved April 22, 2015 from:
http://www.npr.org/templates/story/story.php?storyId=105492482.
Kane, T. & Staiger, D. (2008). Estimating teacher impacts on student achievement: An
experimental evaluation. NBER Working Paper 14607. Cambridge, MA: National
Bureau of Economics Research.
Kane, T. & Staiger, D. (2002). The promise and pitfalls of using imprecise school accountability
measures. Journal of Economic Perspectives, 16(4), 91-114.
Koedel, C. & Betts, J. R. (2007). Re-examining the role of teacher quality in the educational
production function. National Center on Performance Incentives, Vanderbilt, Peabody
College.
Knight, D. S. (2015, April). Is class size reduction cost-effective? Comparing the cost-
effectiveness of class size reduction and teachers’ salary increases under various policy
designs. Paper presented at the Annual Conference of for the American Educational
Research Association, Chicago, IL.
Knight, D. S. & Strunk, K. O. (2015). Who bears the costs of district funding cuts? Equity
implications of teacher layoffs. Unpublished Manuscript.

40
Kraft, M. (In Press). Teacher layoffs, teacher quality and student achievement: Evidence from a
discretionary layoff policy. Education Finance and Policy.
Krueger, A. B. (1999). Experimental estimates of education production functions. Quarterly
Journal of Economics, 114(2), 497-532.
Ladd, H. F. (2011). Teachers' perceptions of their working conditions: How predictive of
planned and actual teacher movement? Educational Evaluation and Policy Analysis,
33(2), 235–261. DOI: 10.3102/0162373711398128.
Lee, J. & Finn, J. (2010, June). Developing time-indexed effect size metrics for K-12 reading
and math educational evaluation. Poster presented at the Institute of Education Sciences
(IES) annual conference, Washington, DC.
Lubienski, C. A., & Lubienski, S. T. (2013). The public school advantage: Why public schools
outperform private schools. Chicago, IL: University of Chicago Press.
McCaffrey, D. F., Sass, T. R., Lockwood, J. R., & Mihaly, K. (2009). The intertemporal
variability of teacher effect estimates. Education Finance and Policy, 4(4), 572-606.
Ost, B. (2014). How do teachers improve? The relative importance of specific and general
human capital. American Economic Journal: Applied Economics, 6(2), 127-151.
Rivkin, S. G., Hanushek, E. A., & Kain, J. F. (2005). Teachers, schools, and academic
achievement. Econometrica, 73(2), 417–458.
Ronfeldt, M., Loeb, S., & Wyckoff, J. (2013). How teacher turnover harms student achievement.
American Educational Research Journal, 50(1), 4–36.
Rothstein, J. (2010). Teacher quality in educational production: Tracking, decay and student
achievement. Quarterly Journal of Economics, 125(1), 175-214.
Sass, T., Hannaway, J., Xu, Z., Figlio, D., & Feng, L. (2012). Value added of teachers in high-

41
poverty schools and lower poverty schools: Implications for management and policy.
Journal of Urban Economics, 72(2), 104-122.
Sepe, C. & Roza, M. (2010, May). The disproportionate impact of seniority-based layoffs on
poor, minority students. Seattle, WA: Center on Reinventing Public Education
Todd, P., & Wolpin, K. (2003). On the specification and estimation of the production function
for cognitive achievement. Economic Journal, 113(485), F3–F33.
UCLA/IDEA (2009). Sharing the burden? The impact of proposed teacher layoffs across
LAUSD. Los Angeles, CA: Institute for Democracy, Education, and Access.

42
APPENDIX A-1: Additional Tables for Essay 1
TABLE A1
Number of teachers who received actual RIF and layoff notices and the number assigned to
placebo RIF and layoff conditions,

2006-07 2007-08 2008-09 2009-10 2010-11 2011-12
Total

2008-09 to
2011-12
All
years
Panel A: Teachers who received actual RIF and layoff notices
No RIF 26,601 29,716 24,212 24,577 22,070 21,259 92,118 148,435
RIF-re 0 0 3,064 1,826 2,492 2,315 9,697 9,697
Layoff 0 0 1,812 355 1,270 1,008 4,445 4,445
Total 26,601 29,716 29,088 26,758 25,832 24,582 106,260 162,577
% Not RIFed 100.0% 100.0% 83.2% 91.8% 85.4% 86.5% 86.7% 91.3%
% RIF-re 0.0% 0.0% 10.5% 6.8% 9.6% 9.4% 9.1% 6.0%
% laid off 0.0% 0.0% 6.2% 1.3% 4.9% 4.1% 4.2% 2.7%
Panel B: Teachers who are assigned to placebo RIF and layoff conditions
No RIF 20,774 23,927 24,212 24,646 21,684 20,824 91,366 136,067
RIF-re 3,296 3,511 3,108 1,733 2,758 2,608 10,207 17,014
Layoff 2,531 2,278 1,768 379 1,390 1,150 4,687 9,496
Total 26,601 29,716 29,088 26,758 25,832 24,582 106,260 162,577
% Not RIFed 78.1% 80.5% 83.2% 92.1% 83.9% 84.7% 86.0% 83.7%
% RIF-re 12.4% 11.8% 10.7% 6.5% 10.7% 10.6% 9.6% 10.5%
% laid off 9.5% 7.7% 6.1% 1.4% 5.4% 4.7% 4.4% 5.8%
Note: Figures shown Panel B for school years 2008-09 to 2011-12 are displayed only for comparison purposes. For
all placebo tests included in this study, I run models only on the pre-layoff school years, 2006-07 and 2007-08. For
this reason, the number and proportion of teachers who are assigned to placebo RIF and layoff conditions during the
layoff years are displayed in grey text. The number of teachers in 2006-07 is lower than the proximate school years
because for all years prior to 2007-08, the district only has information on teachers of record (those that appear in
the student-level Marks data), and does not have information on other teachers who are not directly responsible for
reporting grades. The final analytic sample (N=31,160) includes only teachers of record and only those in tested
grades and subjects for whom VAMs can be estimated (and spans school years 2007-08 to 2012-13), I include all
employees classified as classroom teachers to estimate placebo RIF and layoff conditions because these teachers are
subject to the RIF and layoff processes.

43
Essay 2. Who Bears the Costs of District Funding Cuts?
Equity Implications of Teacher Layoffs
The Great Recession of 2008 led to unprecedented reductions in funding for public
education in the United States. As a direct result of recessionary spending cuts, a greater number
of teacher layoffs occurred in the years during and following the Recession than at any other
time in recent history (Goldhaber & Theobald, 2013; Shierholz, 2013). Research has documented
that layoffs have harmful effects on both teachers and students. Apart from increasing class sizes,
layoffs contribute to school-level teacher churn, lower individual teachers’ productivity, likely
damage school culture and morale, and harm the composition of the teaching workforce by
removing some high-potential early-career teachers (Boyd et al., 2012; Goldhaber et al., 2015;
Kraft, 2015; Stunk et al., 2015).
It is likely that educator layoffs can disproportionately affect historically disadvantaged
students (e.g., UCLA/IDEA, 2009), causing such students to bear an uneven share of the harmful
consequences of recessionary staffing reductions. Studies show that across various contexts,
layoffs impact students inequitably both across school districts within a state and within school
districts (Goldhaber & Theobald, 2013; Dowell, Hodgman & Littlefield, 2011). State-level
analyses have found wide variation in the number of budget-based layoffs conducted across
districts following the great recession (Estrada, 2012; Plecki, Elfers & Finster, 2010). Because
many state school finance systems do not provide equitable levels of funding across districts
(Baker, Sciarra, & Farrie, 2015; Education Trust, 2015), layoffs can be concentrated in districts
with lower allocations of state financial support and greater proportions of low-income students
and students of color.

44
District policies governing staffing reductions can exacerbate these inequities by
concentrating layoffs at particular schools within districts (e.g., Sepe & Roza, 2010). When
school districts conduct budget-based layoffs, most state and district policies require that
districtwide seniority is a primary determinant of which teachers are laid off, rather than
individually assessing the enrollment and staffing needs at each school (National Council on
Teacher Quality, 2010). Although other factors including special skills or training and high need
credentials or endorsement areas are taken into account, teaching experience is typically the most
important factor (Dowell, Hodgman & Littlefield, 2011; Goldhaber & Theobald, 2013).
Seniority-based layoff policies in particular have been criticized for contributing to inequitable
distributions of layoffs within districts (Hahnel, Ramanathan & Barondess, 2011; The New
Teacher Project, 2010). Layoffs may be distributed inequitably under seniority-based policies
because in many districts novice teachers are concentrated in the highest-poverty schools
(Goldhaber, Lavery & Theobald, 2015; Clotfelter, Ladd & Vigdor, 2005; Darling-Hammond,
2004; 2000).
Some districts and states have attempted to mitigate the potential inequitable results of
layoffs by implementing policy reforms intended to protect low-income and minority students,
and the teachers who teach them, from the consequences of layoffs. For instance, the layoff
process in the Charlotte-Mecklenburg was conducted on a school-by-school basis during the
2008-09 and 2009-10 staff reductions (Sawchuk, 2015). By first taking into account changes in
enrollment and natural teacher attrition at each school, district administrators created a system
that allocated budget-based layoffs equitably across all schools in the district (Kraft, 2015).
A different type of policy intervention was implemented in the Los Angeles Unified
School District (LAUSD) during the layoffs that took place in 2010-11 and 2011-12. After an

45
unprecedented staffing reduction in which almost 5,000 reduction-in-force (RIF) notices were
distributed to teachers in March of 2009, the American Civil Liberties Union (ACLU) filed a
class action lawsuit alleging that the layoff process in LAUSD disproportionally affected
students in South Los Angeles, where schools have the highest concentrations of poverty,
English language learners, and students of color (Reed v. State of California, 2010). A second
round of teacher layoffs occurred the following school year, before State Supreme Court made
any decision in the Reed case. Ultimately, the parties involved in the case agreed on a settlement
that required LAUSD to redirect layoffs in a set of 45 high-need schools for the third and fourth
years of layoffs (2010-11 and 2011-12). While some analyses of the disproportionate distribution
of layoffs were conducted for the Reed court case, and literature on the negative impact of
recessionary layoffs is growing (e.g., Goldhaber et al., 2015; Kraft, 2015; Strunk et al., 2015), no
prior research has drawn on longitudinal student-level data to study the distribution of teacher
layoffs across students in a large school district. Moreover, past research has not analyzed the
efficacy of policy interventions designed to mitigate the disparate impacts of widescale staffing
reductions. The lack of research in these areas highlights an important gap in the literature that
this study seeks to address.
The LAUSD reforms stemming from the Reed settlement, and the layoffs that took place
prior to the intervention, provide an ideal setting to evaluate how teacher layoffs are distributed
across students and schools and how districts can address potential inequities. Within this
context, I pose the following two research questions that align with the gaps in the research:
1. To what extent were teacher layoffs disproportionately distributed across students in
LAUSD during the first two years of layoffs?

46
2. To what extent did the Reed policy intervention impact the distribution of layoffs
across low-income students, students of color, and English language learners?
Using longitudinal data from LAUSD, which instituted layoffs for the four years
immediately following funding cuts associated with the Great Recession, I document how
layoffs differentially affected students across various demographic indicators. I then show how
the policy implemented as a result of the Reed settlement dramatically reduced the extent to
which students of color and low-income students were disproportionately impacted. I find that
while the Reed policy significantly reduced inequities in the way layoffs were distributed across
students and schools, the small scale of the Reed intervention (8.7% of students) meant that the
district did not ultimately achieve equity in the distribution of layoff. The results of this study
have important policy implications as districts around the country continue to struggle to meet
budget obligations and teacher layoffs have continued into the 2014-15 school year (Bosman,
2015; Schulte, 2015).
In the next section I provide additional background on the implications of widescale
layoffs for schools, teachers, and students. I then review the policy context under which large-
scale teacher layoffs took place in LAUSD as well as the Reed v. State of California settlement
that prevented budget-based layoffs in high-need schools. The following section provides an
overview of my data and analytic approach and in the final two sections, I discuss findings and
policy implications.
Background Literature
How layoffs are distributed across students is important because past research suggests
teacher layoffs hurt students’ educational experiences. Throughout the remainder of the paper, I
distinguish between reduction-in-force (RIF) notices, which provide teachers with an early

47
warning that their position is at risk of being eliminated, and layoff notices, which are distributed
later in the school year and imply that the teacher has officially lost her or his teaching position.
Effects of Teacher Layoffs
One of the lessons learned from the Great Recession is that school district funding cuts
that require teacher layoffs substantially harm educational processes and student learning (Boyd
et al., 2012; Goldhaber et al., 2015; Goldhaber & Theobald, 2013; Kraft, 2015; Stunk et al.,
2015). As recent research shows, layoffs have a number of negative consequences. First, staffing
schools with fewer teachers per student requires increasing class sizes (Kraft, 2015) or reducing
the amount of teacher planning time. For example, when LAUSD laid off 2,308 classroom
teachers and did not rehire them the following year, average class size across the district
increased by 2.15 students. Estimates of the achievement effects of increased class size taken
from Kruger (1999) and other studies (e.g., Cho, Glewwe & Whitler, 2012; Jepsen & Rivkin,
2009; Rivkin, Hanushek, & Kain, 2005) suggest this class size increase lowers achievement by
between 0.015 to 0.060 standard deviations, depending on the estimate of class size effects. At
the same time, middle and secondary teachers in LAUSD also saw an increase in the proportion
of the day spent teaching, which is likely to lower the quality and coherence of their instruction
(e.g., DuFour, DuFour, Eaker & Many, 2010).
Second, teacher layoffs change the composition of the teaching workforce, and scholars
argue that such changes negatively impact the overall quality of teachers in a district (Boyd et al.,
2011; Goldhaber & Theobald, 2013). Because most union contracts and many state laws require
that districts rely primarily on seniority (Thomsen, 2014), principals often have little or no
autonomy over which teachers are let go during budget reductions. High-performing or
promising early career teachers can be targeted for layoffs simply as a function of their

48
experience in the district, while other potentially less effective teachers are protected through
seniority (Boyd et al., 2011). Scholars have argued that by removing highly effective early career
teachers, overall average teacher effectiveness in the district declines (Boyd et al., 2011;
Goldhaber & Theobald, 2013).
Third, the layoff process can directly impact teachers’ classroom performance and
student achievement on standardized exams. Strunk and colleagues (2015) find that LAUSD
teachers who are laid off and rehired to the district have lower value-added scores in the year
they are rehired, compared to their typical year when they were not RIFed or laid off. The
authors also show that in Washington State, receipt of an initial RIF notice that is ultimately
rescinded early in the school year lowers teachers’ value-added scores the same year. Only one
prior study directly links teacher layoffs to measures of student achievement. Kraft (2015)
examines variation in teacher layoffs across grades within schools in the Charlotte-Mecklenburg
district, where principals have discretion over which teachers to layoff. He finds evidence that
layoffs generally lower student achievement in the subsequent year, but that achievement
declines by more when the teacher laid off is rated highly effective (either through observation or
student performance measures).
Finally, widescale layoffs also indirectly contribute to teacher turnover by introducing job
insecurity and requiring teachers to be shuffled across schools. In a recent study of teacher
layoffs in Los Angeles and Washington State, Goldhaber et al. (2015) find that teachers in
LAUSD who feel threated by the possibility of a layoff are more likely to leave their district. In
addition, receipt of a RIF notice that is later rescinded increases the likelihood teachers switch
schools in both LAUSD and Washington State. The authors’ findings suggest that layoffs
contribute to school-level teacher churn both by introducing layoff threat and through district

49
structural changes that require teachers to switch schools to fill positions. In short, researchers
have learned a great deal about the negative consequences associated with the Great Recession
era teacher layoffs. Less empirical work has examined how these layoffs are distributed across
districts and schools, a topic to which I now turn.
The Distribution of Teachers and Teacher Layoffs
Decades of research has documented the inequitable access to highly qualified and
effective teachers across students’ race/ethnicity, socioeconomic status, English language
proficiency, and achievement levels (e.g., Darling-Hammond, 2004; 2000; 1998; Goldhaber,
Lavery & Theobald, 2015; Ingersoll, 1999; Isenberg et al., 2013; Lankford, Loeb & Wyckoff,
2002). Past studies focus particularly on the high concentrations of novice teachers in schools
with greater proportions of low-income students and students of color (e.g., Clotfelter, Ladd &
Vigdor, 2005).
Given that most districts lay off in reverse order of seniority (due to “last-in-first-out” or
LIFO regulations), many researchers and policy organizations caution that LIFO-based layoffs
lead to an inequitable distribution of layoffs as the less experienced teachers (who are placed in
low-income/high minority schools) are let go (e.g., Hahnel, Barondess & Ramanathan, 2011;
National Center on Teacher Quality, 2010; Sepe & Roza, 2011; The New Teacher Project,
2010).
20
For example, a report from the University of California Los Angeles Institute for
Democracy, Education and Access (UCLA/IDEA, 2009) shows that 46% of LAUSD high-
poverty middle schools employ a “high proportion” of first and second year teachers (more than
20%), whereas fewer than 10% of low-poverty middle schools employ a high proportion of

20
Note that the term last-in-first out is misleading as no layoffs are ever strictly based solely on seniority and most
districts maintain policies that allow for exceptions to strict seniority-based layoff policies (see DiCarlo, 2013 for
more discussion on this topic).

50
novice teachers.
21
That study found similar patterns in high schools, but did not include data on
elementary schools in LAUSD. The same trends exist in the 15 largest districts in California.
Sepe and Roza (2010) find that, given average teacher experience levels across schools within
districts, a greater proportion of teachers were likely to be laid off in schools with higher
concentrations of low-income students and students of color. Both of these studies use publicly
available school-level data to analyze the distribution of teacher experience without actual
information on assignment of RIFs and layoffs. Their conclusions are thus based on likely
outcomes given seniority-based layoff policies. That said, numerous media stories have reported
that RIF notices were in fact disproportionately distributed to high poverty schools in LAUSD
and other California districts (e.g., Frey, 2012). One story for example documented that in 2009,
more than half of the staff in a single high-poverty LAUSD school received RIF notices (Felch,
Song, & Smith, 2010), while the district average that year was approximately 17%.
22

Other studies simulate distributions of layoffs under various layoff policies. Boyd et al.
(2011) simulate layoffs in New York City and find that the average laid off teacher comes from a
school where approximately 80% of students are eligible for FRL, regardless of whether teachers
are laid off by seniority or by value-added, and this figure is roughly equal to the districtwide
mean. Simulations of layoffs based on value-added measures reported in Goldhaber and
Theobald (2013) suggest that such a layoff policy would result in a more equitable distribution of
layoffs than in a LIFO-based policy, but their analysis looks across rather than within districts
and is largely influenced by the types of districts conducting layoffs.
23
In sum, studies that

21
The authors define high poverty as schools with three quarters or more students eligible for free and reduced price
lunch (FRL) and low-poverty as fewer than half of students eligible for FRL. Data used in the study were based on
the 2009-10 school year.
22
This figure is based on the author’s calculations.
23
The authors’ simulations are based on changes in layoff policies within districts; however, the summary statistics
presented examine the distribution of layoffs across districts.

51
simulate layoffs or estimate how layoffs might be distributed given average teacher experience
across schools reach different conclusions depending on the context, with studies based in
California and LAUSD suggesting that LIFO-based policies create inequitable distributions of
layoffs, and evidence from New York City showing fairly equitable distributions across students.
A surprisingly small number of studies use actual data on teacher layoffs to examine how
student subgroups are differentially impacted. Like simulation-based studies, the results across
various studies and contexts are mixed (Goldhaber & Theobald, 2013; Kraft, 2015; Stunk et al.,
2015). Plecki, Elfers, and Finster (2010) use actual layoff data from the 2008-09 school year in
Washington State to show that the number of RIF notices distributed across districts varied
greatly; however, districts conducting layoffs and those that avoided layoffs had similar
characteristics in terms of student race/ethnicity and poverty levels. Drawing on an additional
year of data for the same state, Goldhaber and Theobald (2013) find that Black students were
more likely to see their teacher laid off, while Latina/o students were less likely to experience a
layoff. However, these results are based on classroom averages across the state, so the
differences in the likelihood of experiencing a layoff reflect both the types of districts conducting
layoffs and policies governing layoffs within those districts.
24

Only one study examines how layoffs are distributed across students and schools when
districts use selection criteria other than seniority and teaching credentials. Kraft (2015)
examines layoffs in Charlotte Mecklenburg Schools, where district administrators had discretion
over which teachers would be laid off. As part of the layoff procedures proposed by the district’s

24
Goldhaber and Theobald (2013) use regression-based analyses that allow for within-district comparisons of the
likelihood that particular student groups see their teacher laid off (using district fixed effects). These models show
generally no relationship between student race/ethnicity or family income and the likelihood of experiencing a
layoff. However, because the authors control for teacher seniority, the interpretation of these models simply suggest
that students of teachers with similar levels of experience have approximately equal likelihood of seeing their
teacher laid off.

52
then-Superintendent, the district first calculated the number of layoffs at each school, based on
the difference between (a) the full-time equivalent (FTE) staffing levels generated from the
district’s student-teacher ratio policies and (b) the number of FTE teachers employed at each
school. Given the number of layoffs for each school, principals then selected which position
categories would be targeted (e.g., art, history, or physical education). Finally, district
administrators selected individual teachers within position categories to be targeted for layoffs,
based in part on teachers’ performance evaluations. Kraft finds that while high school teachers
were approximately twice as likely as elementary teachers to be laid off, and 40% more likely
than middle school teachers, there were not substantial differences in the likelihood of being laid
off by student race/ethnicity or students’ family income level. Although teachers in schools with
greater proportions of African-American students, lower achievement scores, and higher rates of
student absenteeism were slightly more likely to be laid off, the school-by-school determinations
of the number of teachers targeted for RIFs and layoffs likely prevented the substantial inequities
observed in studies based in California or LAUSD (i.e. Sepe & Roza, 2010; UCLA/IDEA,
2009).
All of the studies discussed above use teacher-level analyses to study the distribution of
layoffs across student subgroups. I build off this past work by using a student-level analysis,
which has two benefits over prior research. First, a student-level analysis uncovers potential
within-school differences in the likelihood particular students’ teachers are laid off. This
distinction is important because some past research shows that students have inequitable access
to experienced teachers within schools (Kalogrides & Loeb, 2013). Second, a student-level
analysis of layoff distributions in middle and high schools recognizes that students have multiple
teachers throughout the school day. I can therefore examine how my results may differ when I

53
consider the proportion of a student’s teachers that are laid off, the likelihood a student has at
least one teacher laid off, or the likelihood a student has the majority of her or his teachers laid
off. In the next section, I describe the local policy context in LAUSD during the period of
layoffs.
Policy Context in LAUSD
During the four years following the onset of the Great Recession, from 2008-09 to 2011-
12, 13.7% of teachers received initial RIF notices and 4.5% were laid off.
25
Like most districts
around the country, LAUSD generally uses seniority within teaching area to determine layoffs.
26

The California Education Code requires that teachers receive a RIF notice by March 15 that
warns the teacher of a possible layoff. By May 15, districts must notify teachers of whether their
RIF notice was rescinded, or whether they will be laid off at the end of the school year. As I
highlighted in the section above, whether a teacher is not RIFed, RIF-rescinded, or laid off is
important because past research shows that RIF-rescinded and laid off teachers in LAUSD have
higher school attrition and laid off teachers that are rehired to the district the following year are
less productive in the year they return from a layoff (Goldhaber et al., 2015; Strunk et al., 2015).
In top two panels of Table 1, I report the percent of teachers that (a) did not receive a RIF
notice; (b) received a RIF notice, but had the notice rescinded before May 15; and (c) received an
initial RIF notice and a final layoff notice. The proportion of teachers receiving RIF and layoff
notices was generally stable over the four years, with the exception of 2009-10, when federal

25
These figures differ slightly from those reported elsewhere (i.e., Goldhaber et al., 2015 and Strunk et al., 2015)
because in prior work, we included all teachers that were subject to the RIF process, had teacher job titles, and were
covered under teacher contract. In the current study, I consider only teachers of record that I can link to students.
26
A review of teacher layoff policies conducted for this study and related work finds that as of the 2011-12 school
year, 17 states allowed local districts to determine layoff policies, 22 required that preference be given to seniority in
some way, and 11 required districts to use measures other than seniority to determine layoffs. This situation has
changed recently with 12 states now requiring that teacher performance be considered, and 10 other states explicitly
prohibiting the use of tenure or seniority as a primary factor in layoff decisions (as of the 2015-16 school year;
Thomsen, 2014). When states allow districts to determine their own layoff policies, most districts use seniority as
the primary factor (National Center on Teacher Quality, 2010).

54
stimulus funding lowered the proportion of teachers that the district RIFed or laid off.
Approximately 10% to 13% of elementary teachers received RIF notices that were rescinded,
and other than the 2009-10 school year about 5% to 8% were laid off each year (shown in the top
panel of Table 1). A slightly lower proportion of middle and high school teachers were RIF-
rescinded or laid off each year (shown in the second panel of Table 1). The bottom two panels
show the extent to which students’ teachers were affected by RIFs and layoffs. Because RIF-
rescinded and laid off elementary teachers had slightly larger class sizes than non-RIFed
teachers, a greater proportion of elementary students were exposed to the layoff process
compared to the proportion of teachers. In the bottom panel of Table 1, I show the average
percent of middle and high school students’ teachers that were RIF-rescinded and laid off. For
example, in 2008-09, students in middle and high schools saw an average of 11.4% of their
teachers RIF-rescinded and 5.6% laid off. Again, because non-RIFed secondary teachers
generally taught smaller class sizes, the proportion of middle and high school students exposed
to RIFs and layoffs is greater than that of teachers.
27

Recessionary layoffs in LAUSD can be split into two phases. In Phase I, 2008-09 and
2009-10, layoffs were dictated solely by state law. Then in Phase II, 2010-11 and 2011-12,
LAUSD implemented an intervention as a result of Reed settlement that was designed to prevent
budget-based layoffs at the most vulnerable schools. Phase I demonstrates how layoffs are
distributed in a large district under a typical seniority-based system; Phase II shows the impact of
an intervention intended to stem the inequities resulting from LIFO layoff processes. I first turn

27
Non-RIFed teachers generally have smaller class sizes in part because special education teachers are typically
assigned to smaller class sizes at all school levels, and these teachers are more protected from layoffs than other
teachers. A second factor relates to California’s K-3 class size reduction program. At the elementary level, teachers
in grades K-3 generally have more experience compared to teachers in grades 4-5, which may happen if more senior
teachers transfer from grades 4-5 to K-3 over time (because they prefer the smaller class sizes in K-3, compared to
4-5). As a result, when elementary teachers are targeted for layoffs, they tend to fall disproportionately on grades 4-5
because those teachers are slightly less senior. Thus grade 4-5 teachers teach larger class sizes and are more likely to
be RIFed than K-3 teachers.

55
to the Phase I layoff process and then discuss the policy context under which Phase II layoffs
took place.
Phase I: Layoffs During the Pre-Intervention Period
The California laws dictating how teachers are laid off are reflected in the LAUSD
administrative data. In Table 2, I provide descriptive statistics of teachers for each layoff
category outlined above, for the first two years of layoffs in LAUSD (before the Reed policy was
implemented). The first panel shows the strong correlation between seniority and teachers’
likelihood of receiving a RIF or layoff notice. Only 53% of novice teachers (with between 0-2
years experience), compared to 97% of veteran teachers (9 or more years experience), did not
receive a RIF notice. Approximately 28% of novice teachers received layoff notices, while 3.4%
of mid-career teachers (3-8 years) and 0.6% of veteran teachers were laid off. Similarly, while
the overall average experience in the district is 9.3 years, RIF-rescinded and laid off teachers had
an average experience of 4.7 and 2.5 years, respectively. Because layoffs in LAUSD are
determined by seniority within teaching area, teachers’ endorsement area is also a strong
predictor of receiving a RIF or layoff notice, echoing the results in Goldhaber and Theobald
(2013).
The second panel of Table 2 shows that special education teachers were largely protected
from layoffs, with only 4% of special education teachers receiving either a RIF notice and having
it rescinded or receiving a layoff notice. Teachers with math or science credentials were slightly
more likely to be RIF-rescinded, compared to the overall average, but far less likely to be laid
off. In addition to the most novice teachers within each teaching area, the district also targeted
teachers who did not have the appropriate credentials for their teaching area during the first year
of layoffs (i.e., were not NCLB-compliant). This is reflected both by the fact that teachers with

56
other non-elementary credentials had the highest proportion of layoffs and by the fact that a
considerable amount of mid-career teachers (with between three and seven years of experience)
received RIF and layoff notices.
Phase II: Layoffs During the Post-Intervention Period
In February of 2010, after just the first year of layoffs, the ACLU filed a class action
lawsuit asserting that three middle schools in LAUSD incurred a disproportionate level of layoffs
(ACLU, 2011). I consider the second two years of layoffs Phase II because the settlement
reached in Reed v. State of California was implemented during these two years of layoffs. The
parties involved in the lawsuit agreed to prevent budget-based layoffs at 45 high-need schools
during the third and fourth years in which layoffs took place in LAUSD (2010-11 and 2011-12).
The selection process for Reed schools was determined during a series of meetings
between district administrators and representatives of United Teachers Los Angeles (UTLA, the
local teachers’ union). The parties agreed that there would be two ways in which a school could
be selected for Reed protection. The first included the 25 schools with the highest teacher
turnover rates that were also in the bottom 30% of Academic Performance Index (API, a
composite measure of student test score performance), but demonstrating positive API growth
over the past three years, and with at least 15 teachers. The second set included 20 schools that
were established within two years of September 1 of the current school year that would be most
adversely affected by layoffs.
28
To ensure that no non-Reed school experienced a substantially
greater proportion of layoffs as a result of the Reed policy, a final provision of the Reed decision
required that redirected RIF notices could only be sent to schools in which the proportion of
teachers receiving notices was below the district-average (see ACLU, 2011).

28
Although information on the Reed selection criteria is available publically, meetings with human resource
directors were held to confirm how the selection process was implemented and how comparison schools could be
identified.

57
For the first year Reed was implemented, the district and union ultimately agreed to
protect 35 schools under the first set of criteria and 10 schools under the “new school” criteria.
During the second year of Reed, 36 and 9 schools were selected under each set of criteria,
respectively.
29
Of the 45 schools that were protected in the first year, 32 were again selected
during the second year, while 13 schools were only targeted during either the first or second year
of Reed. Based on the selection criteria, greater proportion of middle and high schools were
targeted for Reed protection.
Because the Reed intervention took place over two years and a different set of schools
was selected in each year, there are three sets of treatment schools: those treated in just the first
year (2011 only schools), those treated in just the second year (2012 only schools), and those
treated in both years (2011/2012 schools). Similarly, students fall into one of three “treatment”
groups: students affected by Reed in both years, just the first year, or just the second year. Table
3 shows the how students were sorted across these three treatment groups, a set of comparison
schools (I explain in the Methods section below how I identified comparison schools), and all
other schools in LAUSD. Table 3 focuses on differences in race/ethnicity, English language
status, and family income across Reed and non-Reed schools because the purpose of the Reed
intervention was to protect schools with greater concentrations of these student populations.
As Table 3 shows, 1.7% of all elementary students were treated by Reed in both 2010-11
and 2011-12, which included 3.0% of Black elementary students and 2.0% of Latina/o
elementary students, but only 0.1% of White elementary students. Non-native English speaking
students were also more heavily represented in Reed schools at the elementary level, while
elementary students eligible for FRL were roughly split between Reed and non-Reed schools

29
By September 1, 2010, a total of 49 schools qualified as a “new school” (i.e., opened within the prior two years).
A total of 38 schools were considered new for the second year of Reed protection.

58
(with slightly more in Reed schools). For middle and high school grades, 7.1% of students
attended Reed schools in both years, while 4.0% and 6.9% were enrolled in Reed schools in just
2011 and 2012, respectively. Secondary students affected by Reed included a larger proportion
of students of color, non-native English speakers, and low-income students.
In summary, the policy context in LAUSD following the Great Recession was not unlike
other large urban districts around the country in that a substantially greater proportion of teachers
were laid off than any time in recent history. Like other districts, layoffs were primarily guided
by teacher seniority within teaching area. However, the district is unique in the sense that, mid-
way through the four years of layoffs, a policy was implemented that dramatically altered the set
of teachers receiving RIF and layoff notices. Whether layoffs were inequitably distributed during
the Phase I period of layoffs and whether the Reed policy was successful in reducing layoff
notices at targeted schools and increasing the overall equity in the distribution of layoffs are
important empirical questions, which I address throughout the remainder of this paper.
Data and Analytic Approach
In this section, I describe the longitudinal data used for this study and the analytic
approach I use to analyze my research questions.
Data
I draw on data that links students to teachers and schools over a five-year window of
observation. The student-level data include information on students’ race/ethnicity, gender,
eligibility for free/reduced price lunch programs (FRL), English language learner status, and
special education classification. Students are linked to the teacher-level datasets, which include
information on teachers’ experience, educational attainment, endorsement areas, contract status
(e.g., temporary, probationary, permanent/tenured, etc.), courses taught each semester, and layoff

59
status. These administrative data are then merged with public-use school-level data accessed
through the California Department of Education. These data include each school’s yearly
Academic Performance Index (API, and school performance measure that combines average
scores across several standardizes tests), school level (elementary, middle, high school, or span
school), total enrollment, and the mean percent of students eligible for FRL and that identify in
each race/ethnicity category.
Analytic Approach
I draw on several outcome measures to address the research questions that include (a)
whether a student’s teacher did not receive a RIF notice, (b) whether a student’s teacher received
a RIF notice that was rescinded (i.e. whether the student’s teacher was “RIF-rescinded”), (c)
whether a student’s teacher received a layoff notice, and (d) whether a student’s current teacher
was not RIFed, RIF-rescinded, or laid off in the prior year but rehired to teach in the current
year. Again, how these outcomes were distributed across students and schools is critical because
past research demonstrates that each one exacerbates teacher turnover or harms teacher
productivity (Goldhaber et al., 2015; Strunk et al., 2015). At the elementary level, I examine the
likelihood that various student groups see their teacher impacted by RIFs or layoffs. For middle
and high school students, the outcome measures are the proportion of a student’s teachers who
are impacted by RIFs or layoffs. Outcomes for middle and high school students are calculated
differently because these students attend non-self-contained classes and therefore have multiple
teachers throughout the school day.
For the first research question, I use descriptive analyses to assess the extent to which
layoffs disproportionately affected various student groups in the years prior to the Reed
intervention. I disaggregate the outcome measures by race/ethnicity, English language status,

60
FRL status, and special education enrollment, and report outcome measures for first two years of
layoffs (2008-09 and 2009-10, before the Reed intervention). To simplify the presentation of
results for research question one, I first present mutually exclusive current year RIF and layoff
categories that include: not RIFed, RIF-rescinded, and laid off. I then discuss mutually exclusive
prior year RIF and layoff categories including not RIFed, RIF-rescinded, laid off, and not
present in the prior year. Because these measures are calculated differently for elementary
students (in self-contained classrooms) and middle and high schools students (in non-self-
contained classrooms), I report results separately for elementary students and then for middle and
high schools students.
I use several strategies to address the second research question of how Reed impacted the
distribution of RIFs and layoffs across students. First, I calculate all of the outcome measures for
each of the student groups listed above for the two years after the Reed intervention was
implemented (2010-11 and 2011-12). I then compare the odds ratios that particular students’
teachers were exposed to RIFs or layoffs before and after implementation of Reed. This
approach shows whether, for example, low-income students or students of color became less
likely to have their teacher laid off, compared to their higher income and/or white peers when
Reed layoff protections were in place.
This first approach is limited because it does not explicitly compare students in Reed
schools to those in non-Reed schools, and I am therefore unable to attribute changes in the
outcomes of interest to just the Reed policy. Another straightforward method would be to assess
the outcomes of interest before and after implementation of Reed for just students attending Reed
schools. Both of these two approaches would fail to consider confounding factors that take place
within the same timeframe as the policy intervention. For example, federal stimulus funding was

61
exhausted the same school year Reed was implemented (2010-11), which increased the
proportion of teachers receiving layoff notices that year. Then as the economy continued to
recover and state funding levels in California increased (Bunting, Kueneman, Louttit, Park &
Parker, 2014; Leachman & Mai, 2013), the district distributed fewer RIF and layoff notices
during in the final year of layoffs (2011-12), compared to the first two years. Finally, over the
course of the period of layoffs, teacher demographics and the distribution of teachers across
schools was changing dramatically in large part because so many teachers were being laid off,
and those that remained were being shuffled around the district to fill empty vacancies
(Goldhaber et al., 2015).
In order to account for these confounding factors, I use a difference-in-difference (DID)
analysis to answer the second research question. The logic behind a DID analysis is that a
comparison group of schools that do not receive the treatment can be identified to serve as a
counterfactual for the treatment group (Murnane & Willet, 2010). Because the treatment schools
and comparison schools had approximately the same proportion of teachers RIFed and laid off
before the Reed intervention, the comparison schools provide a counterfactual for what would
have happened in the treatment schools in the absence of treatment.
30
In short, the DID analysis
compares the pre- and post-intervention outcomes in treated schools (the first difference) with
the pre- and post-intervention outcomes in non-treated schools (the second difference).
Because the Reed intervention took place over two years and a different set of schools
was selected in each year, there are three different sets of treatment schools: those treated in just
the first year (2011 only schools), those treated in just the second year (2012 only schools), and
those treated in both years (2011/2012 schools). I selected comparison schools using the same

30
It is important to note that as part of the Reed court decision, the district could not simply move all redirected
layoffs to a small set of non-Reed schools. Instead, redirected RIF notices could only be sent to schools in which the
proportion of teachers receiving notices was below the district-average (see ACLU, 2011).

62
selection criteria used to identify Reed schools. Recall that Reed schools were selected through
two possible sets of criteria, new schools that would be most heavily impacted by RIF notices
and low-performing schools that were showing improvement over the prior three years, but had
the highest rates of teacher turnover. I chose the first set of comparison schools by including the
10 new schools that were most heavily impacted by RIF notices, but were not selected for Reed
protection. I chose to include 10 schools under this definition because the district had originally
selected the same number of new schools for Reed protection in 2010-11. To identify
comparison schools to match the second set of selection criteria, I also considered only schools
showing growth in API over the prior three years, but expanded the other criteria to all schools in
the bottom four deciles of API (as opposed to three), and with teacher turnover above the district
average. Therefore, my comparison schools represent the set of schools that would have been
selected for Reed if a greater number were protected in the original agreement (see Appendix
Table A3 for summary statistics of Reed and comparison schools). In addition, any school ever
selected for Reed, but not treated during one of the two years in which Reed took place (i.e.,
“2012 only schools” during the 2010-11 school year and “2011 only schools” during the 2011-12
school year) is included as comparison group schools.
For research question two, I focus on the three RIF and layoff outcomes that prior
research has linked to higher teacher turnover and lower teacher effectiveness in LAUSD (i.e.,
Goldhaber et al., 2015; Strunk et al., 2015): (a) whether a teacher receives an initial RIF notice in
the current year, (b) whether a teacher is laid off in the current year, and (c) whether a teacher
was laid off in the prior year, but rehired to teach for the current year.
31
Because Reed was

31
The first outcome, whether a student’s teacher received a RIF notice (or for secondary students, the proportion
receiving RIF notices) includes teachers who received a RIF notice that was rescinded and teachers who received a
final layoff notice. I combine the RIF-rescinded and layoff outcomes because no teachers in Reed schools were RIF-
rescinded – of the teachers that received a RIF notice, all received a final layoff notice. Discussions with district

63
implemented over two years, with a different set of schools selected each year, students fall into
one of four categories: those in Reed schools in both years, in 2011 only, in 2012 only, or never
in a Reed school (i.e., in a comparison school).
32
Given these treatment groups, the difference-in-
difference (student-level) equation that creates the appropriate comparison groups for each
school is specified using the following model:
Pr(RIFit) = β0 + β1 REED_2011it + β2 REED_2011it * YEARit + β3 REED_2012it
+ β4 REED_2012it * YEARit + β5 SCHOOL_LEVELit + εi + μit (1)
Where Pr(RIFit) is the probability that student i's teacher received a RIF notice in year t. I run the
same models predicting the likelihood a students’ teacher is laid off or was laid off the prior year,
but rehired. For middle and high school students, the outcomes are continuous variables showing
the proportion of a student’s teachers that received a RIF notice, received a layoff notice, or were
laid off in the prior year. The Reed indicators (REED_2011 and REED_2012) are time-invariant
variables indicating whether the student will be enrolled in a Reed school in 2011 and in 2012,
respectively.
33
YEARit and SCHOOL_LEVELit are sets of dummy variables indicating the school
year and the school level (elementary, middle, high school, or span school), respectively. Finally,
the error term is divided into two components, εi and μit, which represent unobserved factors
related to individual students and individual student-observations, respectively.

administrators revealed that layoffs at Reed schools teachers took place due to changes in enrollment. I do not
include models that predict whether a student’s teacher was not RIFed because these results would simply be the
inverse of RIF notice. Finally, I exclude models that use RIF-rescinded in the prior year because past research shows
this particular variable is unrelated to changes in teacher turnover or effectiveness.
32
Because students switch schools over time, these treatment groups would emerge even if Reed targeted the same
set of schools in both years.
33
Note that most Reed schools were treated in both 2010-11 and 2011-12 (35 of the 58 ever selected). The DID
estimate for students in schools treated in both years is given by β
1
+ β
2
, while the effect of Reed for students treated
only in 2010-11 or only in 2011-12 is given by β
1
and β
2
, respectively. For models in which the outcome measure is
whether the student’s teacher was laid off and rehired in the prior year, the Reed indicators are time-invariant
variables indicating whether the student’s current year school was a Reed school in the prior year. I make this slight
adjustment because the likelihood of being taught by a teacher that was laid off in the prior year (but rehired) is most
affected by whether the student’s school was protected by Reed in the prior year.

64
The models described above provide causal estimates of the extent to which Reed
impacted the likelihood that students’ teachers were affected by layoffs. In order to test the
specific impact of Reed on student subgroups, I rerun each of the Reed models for each student
subgroup by race/ethnicity, FRL status, and language status. Models run for each student
subgroup show how the general effects of Reed varied across students.
Findings
My main results are shown in Tables 4-9. In Tables 4 and 5, I provide results to answer
research question 1 – how layoffs are distributed under typical LIFO layoff process (i.e., during
the Phase I layoffs). I show how RIF and layoff notices were distributed across elementary
students in Table 4 and across middle and high school students in Table 5, both prior to
implementation of Reed. Tables 6-9 address research question 2 of how the Reed decision
altered the distribution of layoffs across students generally, and for particular student subgroups.
First in Tables 6 and 7, I present the same odds ratios as in Tables 4 and 5, this time covering the
two years of Phase II layoffs, in which the Reed policy was in place. Then in Tables 8 and 9, I
show the effects of Reed protections on each outcome of interest and how those effects were
distributed across student subgroups. Findings are discussed in three sections: (a) the distribution
of RIF and layoff notices during the Phase I layoffs, prior to the Reed settlement; (b) the
distribution of RIF and layoff notices during Phase II when the Reed intervention was in place;
and (c) the effects of Reed protection on the outcomes of interest.
Teacher Layoffs Prior to the Reed Intervention
During the first phase of layoffs – the two years prior to Reed intervention – students of
color and English language learners in all school levels and low-income students in middle and

65
high schools were far more likely to see their teacher laid off than White students, native English
speaking students, and students from middle and upper-income families.
Elementary students. In Table 4, I show the distribution of RIF and layoff notices for
elementary students overall and by race/ethnicity, English language status, family income, and
special education classification, during the first two years of layoffs. The first column shows the
demographic makeup of all elementary students over this time period. The next three columns
show, for each student group, the percent whose current teachers’ fell into each layoff category:
not RIFed, RIF-rescinded, and laid off. For example, Black students made up 8.7% of all
students in the district in the 2008-09 and 2009-10 and of those, 78.6% had teachers who were
not RIFed, 15.5% who were RIF-rescinded, and 5.9% who were laid off. For Latina/o students,
who comprised 75.6% of elementary students, 84.7%, 10.9% and 4.4% had teachers who were
not RIFed, RIF-rescinded, and laid off, respectively (during the first two years of layoffs).
The final three columns show odd ratios of these outcomes, which allow for comparisons
to a reference group. I chose reference categories to reflect the group within each classification
that is generally most privileged in educational settings: White students, native English speaking
students, those never eligible for FRL, and students with no special education eligibility.
Significance tests show whether odds ratios are significantly different than one (an odds ratio of
one implies the student group is equally as likely as the reference group to have their teacher not
RIFed, RIF-rescinded, or laid off). For example, Black students in elementary grades were
41.4% and 67.9% more likely than White students to have their teacher RIF-rescinded or laid off,
respectively, during the first two years of layoffs. Latina/o elementary students had
approximately the same likelihood of having a teacher receive a RIF notice that was rescinded as
compared to White students (10.9%), but were 24.5% more likely than White students to have

66
their teacher laid off. Pacific Islander students in the elementary grades were slightly more likely
to see their teacher RIF-rescinded and less likely to have a teacher laid off compared to White
students, but these results are not substantial and not statistically significant. Finally, students
who identify as Native American and as Asian were both more likely to see their teacher laid off,
but less likely to have their teacher RIF-rescinded, compared to White students.
English language learners in elementary grades also received a disproportionate level of
RIF and layoffs notices in the years before Reed. The district has three classifications for non-
native English speakers. Student with limited English proficiency (LEP) are considered still
learning to speak English, students who were reclassified as fluent are those who initially entered
the district as English language learners, but have since gained proficiency in English, and fluent
non-native English speakers are those who entered the district fluent in English, but do not speak
English as their native language. At the elementary level, 65.5% of LAUSD students were non-
native English speakers in 2008-09 and 2009-10 and about 40% were classified as LEP. Table 4
shows that before the Reed intervention, relative to native English speakers, LEP students were
6.2% more likely to see their teacher laid off, while those reclassified as English proficient and
those who are fluent non-native English speakers were 4.5% and 1.3% more likely than native
English speakers to see their teacher laid off, respectively. All together, non-native English
speaking students in the elementary grades were 2.5% and 4.9% more likely to see their teacher
RIF-rescinded or laid off, respectively, compared to native English speakers.
Surprisingly, I do not find the same trends by FRL at the elementary level. As shown in
Table 4, students ever eligible for FRL were 6.9% less likely to see their teacher laid off,
compared to students never eligible for FRL. When I compare elementary students eligible for
FRL in a particular year to those not eligible that year (instead of ever eligible and never

67
eligible), FRL students are even less likely to see their teacher RIFed or laid off compared to
students not eligible for FRL that year. These results echo those found in Boyd et al. (2012), who
examined the distribution of RIFs by FRL in New York City elementary grades. This finding
emerges largely because of the way teacher experience is distributed across the 86% of
elementary students ever eligible for FRL and the 14% never eligible.
34
In elementary schools
with the highest quintile of FRL (where 100% of students are ever eligible for FRL), the average
teacher experience was 9.5 years in 2008-09 and 10.5 years in 2009-10. These numbers are not
far off from average experience in the lowest quintile FRL schools, where the respective
numbers for the same years are 10.0 and 11.2 years. Moreover, when I look within schools across
classrooms, the average teaching experience for low- and high-FRL classrooms is even more
similar, with average experience of 9.7 years in the lowest FRL classrooms and 9.5 in the highest
FRL classrooms for school year 2008-09. In 2009-10, the difference in average experience
between the lowest and highest quintile of percent FRL for elementary classrooms was only 0.4
years (11.2 years for low-FRL classrooms and 10.8 for high-FRL classrooms). I report the
average experience across FRL quintiles for schools and classrooms in Appendix Figure A1. To
sum up, average teaching experience at the elementary level is more equitably distributed across
FRL status compared to student race/ethnicity and English language status, and as a result, RIF
and layoff notices were distributed more evenly in elementary grades with respect to FRL,
compared to other student categories.
35


34
Note that because of the very high rate of eligibility for FRL in the district, this particular indicator variable does
not provide a substantial degree of variation across schools and classrooms and is therefore not an ideal measure of
students’ family income.
35
Examining the exposure to RIF and layoff notices by FRL status within race/ethnicity provides additional
clarification. I find that in the elementary grades, Black students ever eligible for FRL are 19% and 8% more likely
than Black students never eligible for FRL to see their teacher RIF-rescinded or laid off, respectively. However,
Latina/o students in elementary grades (who make up over three-quarters of all elementary students), along with
White and Asian students are all more likely to see their teacher RIF-rescinded or laid off if they were never eligible
for FRL, compared to if they were ever eligible.

68
The final panel in Table 4 shows how RIF and layoff notices were distributed across
special education enrollment. Elementary students eligible for special education were far less
likely to see their teacher affected by RIFs or layoffs across any of the measures shown. This
finding is not surprising given the degree to which special education teachers were protected
from the layoff process, as shown in Table 2.
Because past research based in LAUSD identifies a negative relationship between lagged
layoff indicators and teacher effectiveness (Strunk et al., 2015), I also examine the distribution of
teachers who fall into the following lagged RIF and layoff categories: (a) not RIFed in the prior
year, (b) RIF-rescinded in the prior year, (c) laid off in the prior year, but rehired to the district,
and (d) not present in the prior year. I report these results for elementary students in Appendix
Table A4 for school years 2009-10 and 2010-11, reflecting the first years in which Phase I
layoffs would have effects on lagged RIF and layoff outcomes. I find that the RIF process
disproportionately impacted disadvantaged students across many, but not all of the lagged RIF /
layoff categories. For example, while Black students were 32% more likely to have a teacher
who was laid off and rehired in the prior year, Latina/o students were 36% less likely, compared
to White students. Non-native English speakers and students ever eligible for FRL were also less
likely to have teachers who were either RIF-rescinded or laid off in the prior year, compared to
native English speakers and non-FRL students, respectively.
However, in elementary grades during the Phase I layoffs, students of color across all
race/ethnicity categories, non-native English speakers, and students ever eligible for FRL were
all more likely than their more advantaged peers to have a teacher who was not teaching in the
district the prior year. In particular, Latina/o students were 63.5% more likely and Black students
were 2.3 times more likely than White students to have such teachers. Non-native English

69
speakers and students ever eligible for FRL were 17.1% and 2.1 times more likely, respectively
to have a teacher who not teaching the prior year. While the RIF process contributed
significantly to teacher turnover, some of the inequitable distribution of elementary teachers who
were not teaching in the district the prior year may be due to the historic rates of high teacher
churn in LAUSD schools with greater proportions of disadvantaged students. However, as past
research has found, the RIF process also contributed to the high turnover at these schools.
Finally, special education students in elementary schools were less likely to have a teacher who
was exposed to any of the RIF or layoff conditions including not having taught in the prior year.
Middle and high school students. The Phase I layoff process was far more inequitable
in middle and high schools compared to elementary schools. In Table 5, I report the same
summary statistics as in Table 4, this time for students in middle and high schools, again in the
2008-09 and 2009-10 school years. Overall, secondary students had an average of 7.5% of their
teachers RIF-rescinded over the first two years of layoffs and 4.0% laid off. RIF and layoff
notices were more disproportionately distributed at the secondary level compared to the
elementary level. For example, on average Black students in the middle and high school grades
saw 7.7% of their teachers RIF-rescinded and 4.3% laid off. Latina/o students had 7.8% of their
teachers RIF-rescinded on average and 4.2% laid off, while the same figures for White students
were 5.8% and 2.1%. Thus Black and Latina/o students in middle and high schools on average
had over twice the proportion of their teachers laid off compared to White secondary students, as
shown by the odds ratios in the final three columns of Table 5.
The second panel of Table 5 shows how RIF and layoff notices were distributed to
teachers of English language learners. Middle and high school students with LEP had, on
average, 8.6% and 4.8% of their teachers RIF-rescinded and laid off, respectively, while native

70
English speakers had 6.9% and 3.4% of their teachers RIF-rescinded and laid off. The
discrepancies for non-native English speakers reclassified as fluent and initially fluent were not
quite as strong, although they still experienced a disproportionate level of RIFs and layoffs. As
shown in the final three columns of Table 5, all together non-native English speakers had 11.5%
more of their teachers RIF-rescinded, and 23.9% more laid off, compared to native English
speakers. The third panel shows results for students ever eligible for FRL. On average, middle
and high school low-income students had 19.9% and 52.0% more of their teachers RIF-rescinded
and laid off, compared to non-low-income students. Finally, as with elementary students, those
enrolled in special education were far more protected for RIFs and layoffs compared to students
not enrolled, and these discrepancies hold for each special education category (not shown).
36

I also examine outcomes for lagged layoff categories for the two years in which Phase 1
layoffs take effect (i.e., 2009-10 and 2010-11, the years following the first two years of layoffs).
Results for middle and high school students are shown in Appendix Table A6. Across all lagged
RIF and layoff indicators, every category of disadvantaged student had a greater proportion of
teachers who were RIF-rescinded in the prior year, laid off in the prior year (but rehired), and not
teaching in LAUSD in the prior year, compared to more advantaged students in the district.
Black and Latina/o students were assigned to teachers who were laid off in the prior year at 3.0
and 2.9 times the rate as White students. Native American students, Pacific Islander students, and
those reporting multiple race/ethnicities across years experienced approximately the same level

36
I also consider alternate definitions of exposure to RIF and layoff notices at the middle and high school level,
shown in Appendix Table A1. While middle and high school students had 7.5% and 4.0% of their teachers RIF-
rescinded and laid off on average during the first two years of layoffs, only 2.5% and 0.9% had the majority of their
teachers fall into these respective categories. However, 31.9% and 20.8% of middle and high school students had at
least one teacher RIF-rescinded or laid off respectively (thus over half had at least one teacher RIFed at all). For
students of color, non-native English speakers, and low-income students, the discrepancies in the distribution of
RIFs and layoffs are even more inequitable when considering students for whom a majority of teachers were RIFed
or laid off, but slightly less inequitable when considering students for whom at least one teacher was RIFed or laid
off.

71
of disproportionate exposure to previously laid off teachers as Black and Latina/o students.
Compared to White students, each of these students groups, on average, was assigned to between
1.6 to 2 times as many teachers who were not teaching in LAUSD the prior year.
Non-native English speakers had about 9%, 31%, and 11% more of their teachers RIF-
rescinded in the prior year, laid off in the prior, or not teaching in the prior, relative to White
students. Low-income students were assigned to 1.23, 2.04 and 1.64 times the proportion of
teachers that fell into each of these lagged RIF / layoff categories, as compared to non-FRL
students. Finally, students enrolled in special education were assigned to a lower proportion of
teachers impacted by RIFs or layoffs, but slightly more (8.4%) of their teachers were not
teaching in the district the prior year (compared to non-SPED students).
Teacher Layoffs During the Reed Intervention
During the second phase of layoffs – the two years during which the Reed intervention
was in place – RIF and layoff notices were distributed more equitably than in the prior two years.
The changes in the distribution of RIF and layoffs notices were larger for middle and high school
students, but Reed increased the equity of layoffs across students at all schools levels.
Elementary students. Results for the Phase II layoffs for elementary students are shown
in Table 6. During the second phase of layoffs, a slightly greater proportion of elementary
students had teachers who received RIF and layoff notices compared to the first phase (18.4%
compared to 15.7% received initial RIF notices). To review, I previously showed that in
elementary schools during Phase I layoffs, Black students and students reclassified as fluent in
English had a disproportionate level of their teachers RIF-rescinded compared to White and
native-English speaking students. In addition, Black, Latina/o, Native American, Asian, limited

72
English proficient, and students reclassified as fluent English speakers all had a greater
proportion of teachers laid off than their more advantaged counterparts.
For every one of the elementary groups who were disproportionately affected during the
Phase I layoffs, I find that the distribution of teachers who were RIF-rescinded and laid off was
more equitable during the Phase II layoffs, when Reed protections were in place. For example, as
shown in Table 6, Latina/o students in elementary grades were actually less likely than White
students to see their teacher RIF-rescinded or laid off. Although Black students in elementary
grades were still 15.4% and 17.4% more likely to have their teacher RIF-rescinded or laid off in
the years in which Reed was in place, the respective figures for the Phase I layoffs were 41.4%
and 67.9%. Students who identify in other race/ethnicity categories and non-native English
speakers also experienced dramatic improvements in the extent to which they were
disproportionately impacted by the RIF process.
Middle and high school students. Changes in the distribution of RIF and layoff notices
between the first and second phase of layoffs were even more dramatic for secondary students. In
Table 7, I report the same summary statistics as before for the Phase II layoff process, this time
for middle and high school students. The first panel in Table 7 shows that during the second two
years of layoffs, every race/ethnicity category had a lower proportion of teachers RIF-rescinded
compared to White students (with the exception of Pacific Islanders, who had a roughly equal
proportion). In contrast, each student group in the first panel maintained a higher proportion of
teachers laid off compared to White students during the second phase of layoffs, but the degree
of inequality reduced substantially. For example, Black and Latina/o students had an average of
16.5% and 21.0% more of their teachers laid off than White students, but these figures were
down from over double the number of layoffs as White students during the Phase I layoffs

73
(during which time Black and Latina/o students had 106.5% and 102.5% more of their teachers
laid off, respectively, than White students). In addition, compared to the first phase of layoffs,
Native American, Pacific Islander, and East Asian students all experienced a decrease in the
relative proportion of their teachers RIF-rescinded or laid off.
The distribution of RIF and layoff notices for non-native English speakers and FRL
students in middle and high schools also became more equitable during the second phase of
layoffs. Both groups had fewer teachers RIF-rescinded than their more advantaged peers during
the Phase II layoffs, but still had a slightly greater proportion of their teachers laid off. While
non-native English speakers saw 11.5% and 23.9% more of their teachers RIF-rescinded and laid
off during the Phase I layoffs (compared to native English speakers), these figures declined to
5.3% less teachers RIF-rescinded and only 6.4% more teachers laid off during the second phase
of layoffs. During the Phase II layoffs, low-income secondary students had 18.5% fewer RIF-
rescinded teachers compared to non-FRL students, but 10% more teachers who were laid off.
Thus compared to the Phase I period, low-income students in middle and high schools
experienced a dramatic decrease in their relative rate of exposure to layoffs. Finally, secondary
students enrolled in special education were still far less likely to have a teacher RIF-rescinded or
laid off during the second two years of layoffs; however, special education students’ relative
likelihood of exposure (compared to non-special education students) increased slightly compared
to the Phase I layoff process.
For students at all levels of education – elementary, middle, and high schools – much of
the changes in the relative exposure to the layoff process between the first and second phases of
layoffs are likely the product of the Reed intervention. In the next subsection, I present estimates
of the effect of Reed on the likelihood of having a teacher impacted by layoffs.

74
The Effect of Reed Layoff Protections
In Table 8, I report the DID estimates of the effects of Reed on the likelihood a student’s
teacher is RIFed, laid off, and laid off in the prior year but rehired. I show results for elementary
students and for middle and high school students in accordance with each of the three Reed
treatment groups (those in Reed schools for both years, those in Reed school for the first year
only, and those in Reed school for the second year only). Whether the observed effects were
consistent across student race/ethnicity, English language status, and income levels in a critical
component of the Reed intervention. I discuss heterogeneous effects in the following subsection
below.
Overall effects of Reed. I find that for elementary students in Reed schools, the
likelihood of having their teacher received a RIF notice decreased by between 24.7 and 27.8
percentage points and the likelihood their teacher received a layoff notice declined by between
7.0 and 8.7 percentage points for each year students were enrolled in Reed schools. Finally, Reed
lowered the likelihood that elementary students were taught by a teacher who was laid off in the
prior year, but rehired by between 2.5 and 3.8 percentage points.
37
The results are generally
similarly whether I focus on students treated in just the first year (shown in the first row of Table
8), just the second year (shown in the second row of Table 8), or in both years (shown in the
third row).
For middle and high school students, the outcomes are the proportion of a student’s
teachers that were RIFed, laid off, and laid off in the prior year and rehired. Much like students
in elementary schools, I find that Reed substantially lowered middle and high school students’

37
Recall that the DID treatment indicators are slightly different for this third outcome, although the interpretation
does not change. I use time-invariant variables indicating whether the student’s current year school was a Reed
school in the prior year (as opposed to the current year). Also note that the estimated Reed effects for students
treated in both years are the average for the two years, while the estimated Reed effects for students treated in just
one year is the effect of Reed based on that one year of treatment.

75
exposure to the layoff process. In particular, Reed reduced the average proportion of a student’s
teachers that were RIFed, laid off, and laid off in the prior year but rehired, by 15.6, 6.3, and 3.1
percentage points, respectively (for students enrolled in Reed schools in both years). As shown in
the bottom two rows of Table 8, the effects of Reed for middle and high school students enrolled
in Reed for only one year were similar to the yearly effects for students treated in both years.
Heterogeneous effects of Reed across students. In Table 9, I show how the main results
for Reed varied across student race/ethnicity. I find that Reed had the intended effects each
individual race/ethnicity at both elementary schools and middle and high schools. For example,
for elementary students enrolled in Reed schools for two years, Reed lowered the likelihood of
having a teacher who received a RIF notice by 27.8 percentage points overall, and by 30.9
percentage points for Black students, 27.5 percentage points for Latina/o students, 15.2
percentage points for other non-White/non-Asian students, and by 24.3 percentage points for
White /Asian students (shown in the first column of Table 9).
38
The effects for other non-White /
non-Asian students are estimated with far less precision, though still statistically significantly
different than zero. These results are shown in Column 1 of Table 9, in the first row of each
panel.
I also find similar effects across elementary student race/ethnicity for the other two
outcomes, which are the likelihood students’ teachers were laid off and the likelihood students’
teachers were laid off in the prior year but rehired. I show these results in the second and third
column of Table 9. While Reed lowered the likelihood that elementary students saw their teacher
laid off by between 7.5 and 8.7 percentage points overall, the estimated effects were slightly
larger for Black students (between 8.9 and 10.5 percentage points), similar for Latina/o students

38
Throughout this discussion I sometimes refer to students who identify as Native American, Pacific Islander, and
students reporting multiple race/ethnicities across years as “other non-White/non-Asian students.”

76
(between 7.1 and 8.5 percentage points), and similar for White / Asian students (between 7.5 and
8.2 percentage points.
Interestingly, I find that Reed increased the likelihood of having one’s teacher laid off for
elementary students that identify as either Pacific Islander, Native American, or who reported
multiple race/ethnicities across years (for students effected by Reed in both years).
39
This finding
can be explained as follows. In the two years prior to Reed, an average of 7.5% of other non-
White / non-Asian students who would ultimately be enrolled in Reed schools during the years
of treatment had their teacher laid off. Similarly, in the years prior to Reed, other non-White /
non-Asian students who would ultimately be enrolled in non-Reed comparison schools during
the years of treatment had an average 7.7% of their teachers laid off. Thus the students in
comparison schools appear to provide an accurate counterfactual for students treated in Reed,
given the similarities in their pre-treatment exposure to layoffs. However, during the years Reed
was in effect, an average of 9.1% of other non-White / non-Asian students in Reed schools saw
their teacher laid off – which were all likely enrollment based layoffs – while only 6.5% of other
non-White / non-Asian students in comparison schools saw their teacher laid off (giving a DID
estimate of positive 2.7 percentage points). Because of the very low proportions of such students
in each school, it is possible that principals of comparison schools (who were just barely not
selected for Reed) assigned other non-White / non-Asian students to teachers that were not likely
to be laid off, thus causing only 6.5% of such students to see their teacher laid off.
40


39
Some of the models shown in Table 9 could not be estimated because of “empty cells” in the dataset. This
happens when, for example, no White / East Asian elementary students who were enrolled in Reed schools in just
2011-12 had a teacher RIFed or laid off in 2011-12 (i.e., one of the values required for estimation is zero, so it is
impossible to invert the matrix that calculates the variance in covariates and therefore impossible to estimate beta).
40
Note that Pacific Islander students, Native American students, and students who reported multiple race/ethnicities
in across years make up about 1% of elementary students. When limiting the sample to just Reed schools and
comparison schools, this amounts to 2,633 student observations or about 650 per year, spread over about 150 Reed
and comparison schools each year.

77
Reed effects are also fairly consistent for elementary students across English language
status and family income. The effect on the likelihood of having a teacher RIFed for students
classified as limited English proficiency, reclassified fluent, and initially fluent were 28.1, 27.5,
and 26.3 percentage points, respectively, compared to 27.8 percentage points overall (for
students in Reed schools in both years). Similarly, Reed lowered the probability of having a
teacher RIFed, laid off, or laid off in the prior year for FRL students by 28.0, 7.6, and 3.0
percentage points and by 27.5, 7.3, and 2.9 percentage points for non-FRL students, respectively
(not shown in Tables). These figures are very similar to those for all other students and the
estimated effects were not substantially different for students treated in just one year as opposed
to both years.
The estimated Reed effects by student race/ethnicity for middle and high school students
are shown in columns 4-6 of Table 9. As with elementary students, the average effects for Black
and Latina/o students are very similar to the overall effects. While Reed lowered the proportion
of a student’s teachers that received a RIF notice by between 24.7 and 27.8 percentage points.
The ranges for Black and Latina/o students, for whom Reed lowered the proportion of RIFed
teachers by between 25.7 and 30.9 percentage points and 24.3 and 27.5 percentage points,
respectively, are similar to the overall effects. Although the ranges of estimated effects were
wider for other non-White / non-Asian students and for White / Asian students, they center on
the same general values as the overall estimated effects. Finally, although not shown, I find that
the effects for Reed across middle and high school English language learners and those eligible
for FRL are roughly in line with the overall effects, with Reed substantially lowering exposure to
RIF and layoff notices across all student subgroups enrolled in Reed schools.

78
In sum, with few exceptions, none of the Reed effects were substantially different across
models run for different student subgroups, indicating that Reed had the intended effect on all
students enrolled in Reed schools. Across all school levels, these results suggest that the Reed
policy was implemented as intended: students in schools selected for Reed had substantially less
likelihood of having their teacher RIFed or laid off. The Reed intervention also lowered the
probability students would be taught by teachers laid off and rehired the prior year – an outcome
of the RIF process that past research links to lower teacher productivity.
Discussion and Implications for Policy
This paper makes a number of important contributions to understanding of education
funding cuts and teacher layoff policies. First, the seniority-based layoff policies used in LAUSD
led to an inequitable distribution of RIF and layoff notices for students of color, low-income
students (in middle in high schools), and students with limited English proficiency. The
disparities were especially substantial at middle and high schools. This finding is inconsistent
with that found in other contexts such as Washington State or New York City, where studies
have generally found that layoffs conducted under LIFO did not disproportionately harm
disadvantaged students. In short, the mechanism for determining the order of teacher layoffs has
implications for the distribution of layoffs across students and schools. Importantly, given the
well-documented disparities in access to well-qualified and effective teachers across many
measures of teachers quality (e.g., Goldhaber Lavery, & Theobald, 2015), any teacher
characteristic used to determine layoffs at a districtwide level runs the risk of concentrating
layoffs in high-need schools.
Second, and more importantly, this paper demonstrates that districts can work within the
context of seniority-based layoff policies to create more equitable distributions of layoffs across

79
student characteristics. The result that the Reed policy lowered the likelihood that teachers in
Reed schools received RIF and layoff notices is not surprising. However, the findings document
the extent to which implementation of the Reed settlement helped reduce the disproportionate
impact of layoffs on students of color, low-income students, and English language learners.
These findings are important because prior research based in LAUSD (Goldhaber et al., 2015;
Strunk et al., 2015) has shown that how layoffs are distributed across student groups has
significant implications for students’ equal access to educational opportunity.
The results of this paper also highlight an important, but often overlooked aspect of
layoff policies and high-need schools. In times of fiscal crisis, when districts must conduct
budget-based layoffs, protection from layoffs in a subset of schools can serve as a recruitment or
retention incentive in hard-to-staff schools. This study demonstrates that schoolwide layoff
protection policies can be implemented with fidelity in large urban districts. Thus, district
administrators can make agreements with staff to implement layoffs protections in high-turnover
or otherwise high-need schools, providing teachers in those schools with increased job security.
Indeed, one of the key findings in Strunk et al. (2015) was that the Reed layoff protections
increased retention regardless of whether teachers were likely to be laid off. Teachers in Reed
schools, because of the schoolwide protections from budget-based layoffs, were more likely to
return to their teaching positions the following year than otherwise similar teachers in non-Reed
schools. School administrators can also use layoff protections as a potential recruitment tool for
otherwise difficult-to-staff schools. These types of recruitment and retention incentives would
not only reduce teacher turnover at little or not direct expense for the district, but, as this paper
shows, would also dramatically improve access to equal educational opportunities for
disadvantaged students.

80
In addition to providing a relatively expenditure-free retention and recruitment incentive,
layoff protections in high-need schools could actually reduce the costs associated with teacher
layoffs. By targeting teachers with the least experience in a district, and therefore, the lowest
salaries, layoffs conducted under LIFO may increase the number of teachers that must be let go
in order to achieve a given reduction in district expenditures. The Reed intervention could reduce
the total number of teachers that must be laid off for a given budget cut because many layoff
notices skipped over the most junior teachers and were instead redirected to more experienced
teachers, who earn higher salaries. However, I conducted an analysis of potential cost saving
associated with Reed and found that the district did not ultimately reduce the number of teachers
who were laid off through implementation of Reed. Because LAUSD is such a large district,
with over 23,000 teachers in 2011-12 across 692 schools, redirecting layoffs in just 45 schools to
more senior teachers does not have a substantial impact on the average experience of laid off
teachers.
41

An important lesson learned from LAUSD’s experience with Reed is that the policy did
not ultimately go far enough to protect vulnerable schools. While the policy was a substantial
improvement over the first two years of layoffs, this analysis shows that even with Reed
protections, most traditionally disadvantaged student groups still faced a disproportionate
amount of RIF and layoff notices. Again, these results are not surprising given the scale of the
Reed intervention. Over the two years the policy was implemented, Reed included only 3.6% of

41
In total, the Reed policy prevented 385 teachers from being laid off in 2010-11 and an additional 415 in 2011-12.
These teachers had an average experience level of 5.36 and average total compensation of $83,709. Teachers that
received layoff notices as a result of Reed had 5.44 years of experience on average and earned $83,914 in total
compensation. Importantly, some of the Reed-protected teachers still received layoff notices due to declining
enrollment in their school, and many of the teachers that received RIF notices because of Reed had their notices
rescinded, as the district gained more information about the next year’s budget. While these RIF-rescissions make it
difficult to pin down the exact cost-saving that resulted from Reed, the differences in experience and salaries
between Reed-protected and Reed-RIFed teachers suggest that cost-saving associated with Reed were likely
minimal.

81
elementary students and 14.0% of middle and high school students (8.7% of all students).
Because the settlement only called for layoff protection in 45 schools in a district with 692
traditional K-12 public schools, the overall impact of the Reed policy was not as substantial as it
could have been if a larger number of schools were targeted. Although the Reed policy did not
ultimately create districtwide equity in the distribution of layoffs, the intervention dramatically
improved on a system that had created substantial inequities in terms of which students were
bearing the costs of district budget cuts.
A final implication pertains to teacher protections. Students whose teachers were
protected were obviously less likely to experience the negative effects of layoff-induced teacher
turnover. Three groups of teachers that were protected were those credentialed in special
education, math, and science. The district may consider protecting teachers with experience
working with students of color or in high-poverty schools. As a result, low-income students of
color would be less likely to experience a disparate impact of layoffs. Such an intervention may
also help districts recruit and retain teachers with experience working in high-poverty / high-
minority schools, which is a challenge in many large urban districts across the country (Darling-
Hammond, 2004; Ingersoll & May, 2011; Quiocho & Rios, 2000; Villegas & Lucas, 2004).
Conclusion
The analyses shown here demonstrate important facets of layoff polices generally, and
LIFO-based policies in large urban districts in particular. Although LIFO can create inequitable
distributions of layoffs, a relatively simple mechanism to address this problem is to prevent
budget-based layoffs from taking place at the highest need schools. By selecting schools with the
greatest needs and ensuring that a sufficient number of schools are protected, districts can
prevent particular students groups from bearing an uneven share of the costs associated with

82
district funding cuts. In the often-unavoidable circumstances that force districts to conduct
budget-based layoffs, averting inequitable distributions is an important and laudable objective.
This study demonstrates the potential for substantial inequities to occur under standard layoff
policies, but shows how districts can ameliorate this problem.

83
Tables for Essay 2
TABLE 1
Total students and proportion affected by layoffs over time

2008-09 2009-10 2010-11 2011-12
Elementary teachers 13,815 12,646 12,521 12,138
No RIF 81.7% 88.7% 82.0% 82.9%
RIF-rescinded 10.7% 10.5% 12.8% 12.5%
Laid off 7.6% 0.8% 5.2% 4.6%
Middle and high school teachers 13,388 12,564 12,152 11,340
No RIF 83.6% 94.5% 88.0% 89.8%
RIF-rescinded 11.0% 3.3% 6.4% 6.1%
Laid off 5.4% 2.1% 5.6% 4.2%
Elementary students 275,161 276,438 274,416 269,527
No RIF 80.5% 88.0% 81.0% 82.2%
RIF-rescinded 11.4% 11.1% 13.6% 13.1%
Laid off 8.0% 0.9% 5.5% 4.7%
Middle and high school students 289,993 281,368 267,586 256,965
Average % not RIFed 83.0% 94.5% 86.4% 88.8%
Average % RIF-rescinded 11.4% 3.3% 7.0% 6.4%
Average % Laid off 5.6% 2.2% 6.6% 4.8%
Note: “No RIF” refers to a student whose teacher did not receive a reduction-in-force (RIF) notice. “RIF-rescinded”
implies the student’s teacher received a RIF-notice, but it was later rescinded. Laid off means the student’s teacher
was laid off. For non-self-contained classrooms in middle and high schools, because each student has multiple
teachers, I report the average proportion of a student’s teacher that is RIF-rescinded, laid off, or not RIFed. The
sample of teachers shown here includes only classroom teachers that report grades for students. The total sample
size differs from that reported in other studies based on the same datasets (e.g., Goldhaber, Strunk, Brown & Knight,
2015) because in that work, we include any employee classified as a teacher who was subject to the RIF process.
Enrollment for elementary students as a proportion of total enrollment increases beginning in 2009-10 because in
that year it became more common for grade 6 to be housed in elementary schools, rather than middle schools.

84
TABLE 2
Summary statistics for teachers by layoff threat (teacher-year observations), 2008-09 to 2009-10

% Overall No RIF
RIF

RIF-
rescinded
Laid off-
All Teachers 52,406
45,571 4,688 2,147
87.0% 8.9% 4.1%
Experience / Education
Novice teachers (1st - 3rd year) 10.4% 52.7% 19.7% 27.5%
Mid - career teachers (4th - 8th year) 24.3% 75.4% 21.2% 3.4%
Veteran teachers (9th year or above) 65.3% 96.7% 2.7% 0.6%
Mean years of experience 9.3 10.1 4.7 2.5
Master’s degree or higher 34.2% 88.5% 8.4% 3.0%
Endorsement Area
Special Education 12.6% 96.0% 1.7% 2.2%
Math or Science 12.9% 86.8% 11.8% 1.4%
Other non-elementary 29.3% 87.2% 7.0% 5.8%
Elementary 45.2% 84.3% 11.4% 4.3%
Note: “No RIF” refers to teacher who did not receive a reduction-in-force (RIF) notice; “RIF-rescinded” implies the
teacher received a RIF-notice, but it was later rescinded; “Laid off-return” means the teacher was laid off, but
returns in the following year; and “Laid off-no return” refers to teachers that were laid off and do not return the
following year. The % overall column shows the overall proportion for each teacher characteristic districtwide
(within categories, columns sum to 100%). The next four columns show how those characteristics are distributed
across four RIF/layoff categories (rows sum to 100%). For example, 10.4% of teachers are novice and of those,
52.7% were not RIFed.

85
TABLE 3
Summary statistics of student characteristics by Reed treatment group, 2010-11 to 2011-12

Overall
Reed 2011
/ 2012
Reed 2011
schools
Reed 2012
schools
Comparable
schools
All other
schools
Elementary student
observations
543,943
9,516 11,203 11,744 127,152 384,328
1.7% 2.1% 2.2% 23.4% 70.7%
Race / Ethnicity

Black 8.2% 3.0% 4.2% 3.4% 30.2% 59.2%
Latina/o 75.3% 2.0% 2.1% 2.4% 25.3% 68.2%
Other race/ethnicity 7.0% 0.2% 1.3% 0.7% 12.3% 85.5%
White 9.5% 0.1% 0.3% 0.2% 10.7% 88.7%
ELL Status
Non-native Eng. speakers 63.1% 2.1% 2.3% 2.6% 26.1% 66.9%
Native English speaker 36.9% 1.2% 1.6% 1.4% 18.7% 77.1%
Family income
Ever enrolled in FRL 88.6% 1.5% 2.1% 2.3% 23.6% 70.5%
Never enrolled in FRL 11.4% 1.8% 2.1% 2.1% 23.3% 70.7%
Special education
Any SPED classification 10.8% 8.0% 4.0% 7.2% 33.9% 46.9%
No SPED classification 89.2% 7.0% 4.0% 6.9% 31.9% 50.2%
Middle and high school
student obs.
524,551
37,185 21,109 36,295 168,370 261,592
7.1% 4.0% 6.9% 32.1% 49.9%
Race / Ethnicity
Black 9.1% 9.9% 5.5% 10.8% 34.4% 39.3%
Latina/o 75.9% 7.8% 4.3% 7.5% 34.0% 46.4%
Other race/ethnicity 7.6% 2.5% 2.9% 2.4% 28.8% 63.5%
White 7.4% 0.7% 0.9% 1.3% 12.8% 84.4%
ELL Status
All non-native Eng. speakers 70.8% 7.8% 4.3% 7.4% 34.1% 46.4%
Native English speaker 29.2% 5.3% 3.4% 5.8% 27.2% 58.3%
Family income
Ever enrolled in FRL 90.1% 2.0% 2.3% 2.4% 25.3% 68.1%
Never enrolled in FRL 9.9% 0.0% 0.4% 0.2% 8.7% 90.7%
Special education
Any SPED classification 11.0% 7.8% 4.3% 7.4% 34.0% 46.5%
No SPED classification 89.0% 0.7% 1.2% 2.9% 14.9% 80.3%
Note: The Reed 2011/2012 schools column indicates students that were enrolled in Reed school during both the
2010-11 and 2011-12 school years. Reed 2011 schools and Reed 2012 schools refer to students enrolled in Reed
school only in 2010-11 or 2011-12, respectively. A total of 32 schools were selected for Reed protection in both
years, while 13 schools were selected for Reed in just the 2010-11 and 2011-12 school years. I identified
comparison schools using the same criteria the district used to identify Reed schools. Thus comparison schools were
next in line for Reed protection and would have been targeted for Reed if the intervention included more schools.
The first column, labeled “overall,” shows the proportion of elementary and secondary students that fall into each
category. In the Reed treatment categories, shown in next five columns, rows sum to 100%. For example, 8.2% of
elementary students are Black and of those, 1.4% were enrolled in Reed schools during both the 2010-11 and 2011-
12 school years. Other race/ethnicity includes the categories shown in Tables 4 and 5.

86
TABLE 4
Percent of elementary students in each demographic category (% overall column) and the
percent of those students whose teachers fall into each RIF / layoff category, 2008-09 to 2009-10

%
Overall
No RIF
RIF Odds ratios

RIF-
rescinded
Laid off No RIF
RIF-
rescinded
Layoff
Total
student-obs.
551,599
464,864 62,234 24,501

84.3% 11.3% 4.4%
Race / Ethnicity Relative to White
Black 8.7% 78.6% 15.5% 5.9% 0.919*** 1.414*** 1.679***
Latina/o 75.6% 84.7% 10.9% 4.4% 0.990*** 1.000 1.245***
Multiple 0.5% 85.7% 10.4% 3.9% 1.002 0.952 1.095
Nat. Am. 0.3% 86.1% 9.3% 4.6% 1.007 0.849* 1.295*
Pac. Isl. 0.3% 85.5% 11.7% 2.8% 1.000 1.073 0.785
Asian 5.9% 85.5% 10.3% 4.2% 1.000 0.942** 1.184***
White 8.8% 85.5% 10.9% 3.5% 1.000 1.000 1.000
English language status Relative to NES
LEP 39.1% 84.3% 11.1% 4.6% 0.997* 1.002 1.062***
Reclassified 14.6% 82.9% 12.6% 4.5% 0.980*** 1.132*** 1.045***
Fluent, NNE 11.8% 84.9% 10.8% 4.4% 1.003 0.971* 1.013
All NNE 65.5% 84.1% 11.4% 4.5% 0.994 1.025** 1.049***
Native Eng. 34.5% 84.6% 11.1% 4.3% 1.000 1.000 1.000
Family income Relative to non-FRL
Ever FRL 86.0% 84.4% 11.2% 4.4% 1.015*** 0.925*** 0.931***
Never FRL 14.0% 83.2% 12.1% 4.7% 1.000 1.000 1.000
Special Education (SPED) Relative to non-SPED
Any SPED 10.6% 89.0% 7.9% 3.1% 1.063*** 0.673*** 0.679***
Non-SPED 89.4% 83.7% 11.7% 4.6% 1.000 1.000 1.000
Note: Multiple refers to students who report multiple race/ethnicities across years. LEP stands for limited English
proficiency; reclassified refers to students who entered the district with LEP, but were reclassified as fluent; fluent
NNE refers to students who are non-native English speakers (NNE), but entered the district fluent in English; NES
refers to native English speakers. FRL refers to students eligible for free and reduced price meals. ED stands for
emotional disturbance; SLD stands for specific learning disability, and the "other" disability category includes
deafness, developmental delay, established medical disability, hard of hearing, mentally retarded, multiple
disabilities, orthopedic impairment, other health impairment, traumatic brain injury, and visual impairment; any
SPED refers to student enrolled in special education. +p<0.10, * p<0.05, ** p<0.01, *** p<0.001.

87
TABLE 5
Percent of middle and high school students in each demographic category (% overall column)
and the average percent of their teachers in each RIF / layoff category, 2008-09 to 2009-10

%
Overall
No RIF
RIF Odds ratios

RIF-
rescinded
Laid off No RIF
RIF-
rescinded
Layoff
Total
student-obs.
571,361
506,144 42,601 22,616
88.6% 7.5% 4.0%
Race / Ethnicity Relative to White
Black 9.7% 88.0% 7.7% 4.3% 0.955*** 1.334*** 2.065***
Latina/o 75.8% 88.0% 7.8% 4.2% 0.955*** 1.355*** 2.025***
Multiple 0.3% 88.0% 8.0% 4.0% 0.955*** 1.390*** 1.913***
Nat. Am. 0.3% 90.2% 6.3% 3.5% 0.978*** 1.098 1.685***
Pac. Isl. 0.3% 89.4% 7.5% 3.2% 0.970*** 1.297*** 1.507***
Asian 6.4% 92.8% 5.1% 2.1% 1.007*** 0.889*** 1.003
White 7.2% 92.2% 5.8% 2.1% 1.000 1.000 1.000
English language status Relative to NES
LEP 23.5% 86.6% 8.6% 4.8% 0.965*** 1.252*** 1.419***
Reclassified 39.8% 88.8% 7.2% 4.0% 0.989*** 1.048*** 1.191***
Fluent, NNE 9.5% 89.4% 7.3% 3.3% 0.996*** 1.058*** 0.990
All NNE 72.8% 88.2% 7.7% 4.2% 0.982*** 1.115*** 1.239***
Native Eng. 27.2% 89.7% 6.9% 3.4% 1.000 1.000 1.000
Family income Relative to non-FRL
Ever FRL 85.6% 88.2% 7.6% 4.2% 0.970*** 1.199*** 1.520***
Never FRL 14.4% 90.9% 6.4% 2.7% 1.000 1.000 1.000
Special Education (SPED) Relative to non-SPED
Any SPED 89.2% 92.0% 5.3% 2.8% 1.043*** 0.682*** 0.675***
Non-SPED 10.8% 88.2% 7.7% 4.1% 1.000 1.000 1.000
Note: Multiple refers to students who report multiple race/ethnicities across years. LEP stands for limited English
proficiency; reclassified refers to students who entered the district with LEP, but were reclassified as fluent; fluent
NNE refers to students who are non-native English speakers (NNE), but entered the district fluent in English; NES
refers to native English speakers. FRL refers to students eligible for free and reduced price meals. ED stands for
emotional disturbance; SLD stands for specific learning disability, and the "other" disability category includes
deafness, developmental delay, established medical disability, hard of hearing, mentally retarded, multiple
disabilities, orthopedic impairment, other health impairment, traumatic brain injury, and visual impairment; any
SPED refers to any student enrolled in special education. +p<0.10, * p<0.05, ** p<0.01, *** p<0.001.

88
TABLE 6
Percent of elementary students in each demographic category (% overall column) and the
percent of those students whose teachers fall into each RIF / layoff category, 2010-11 to 2011-12

%
Overall
No RIF
RIF Odds ratios

RIF-
rescinded
Laid off No RIF
RIF-
rescinded
Layoff
Total
student-obs.
543,943
443,779 72,571 27,593

81.6% 13.3% 5.1%
Race / Ethnicity Relative to White
Black 8.2% 77.8% 15.6% 6.6% 0.962*** 1.154*** 1.174***
Latina/o 75.3% 82.2% 12.9% 4.8% 1.017*** 0.956*** 0.862***
Multiple 0.6% 81.2% 13.2% 5.6% 1.004 0.973 1.005
Nat. Am. 0.3% 82.4% 12.6% 5.0% 1.019 0.933 0.887
Pac. Isl. 0.3% 81.3% 15.0% 3.7% 1.005 1.109+ 0.663***
Asian 5.8% 79.8% 14.9% 5.3% 0.987*** 1.103** 0.940*
White 9.5% 80.9% 13.5% 5.6% 1.000 1.000 1.000
English language status Relative to NES
LEP 36.6% 82.5% 12.3% 5.1% 1.011* 0.926 1.019***
Reclassified 16.5% 79.1% 15.6% 5.3% 0.968*** 1.175*** 1.042
Fluent, NNE 10.0% 82.1% 13.4% 4.6% 1.005*** 1.004*** 0.906**
All NNE 63.1% 81.6% 13.4% 5.1% 0.999 1.004 1.009
Native Eng. 36.9% 81.6% 13.3% 5.0% 1.000 1.000 1.000
Family income Relative to non-FRL
Ever FRL 88.6% 81.8% 13.2% 5.0% 1.025*** 0.904*** 0.888***
Never FRL 11.4% 79.8% 14.6% 5.6% 1.000 1.000 1.000
Special Education (SPED) Relative to non-SPED
Any SPED 10.8% 86.4% 9.7% 3.9% 1.067*** 0.762*** 0.860***
Non-SPED 89.2% 81.0% 13.8% 5.2% 1.000 1.000 1.000
Note: Multiple refers to students who report multiple race/ethnicities across years. LEP stands for limited English
proficiency; reclassified refers to students who entered the district with LEP, but were reclassified as fluent; fluent
NNE refers to students who are non-native English speakers (NNE), but entered the district fluent in English; NES
refers to native English speakers. FRL refers to students eligible for free and reduced price meals. ED stands for
emotional disturbance; SLD stands for specific learning disability, and the "other" disability category includes
deafness, developmental delay, established medical disability, hard of hearing, mentally retarded, multiple
disabilities, orthopedic impairment, other health impairment, traumatic brain injury, and visual impairment; any
SPED refers to student enrolled in special education. +p<0.10, * p<0.05, ** p<0.01, *** p<0.001.

89
TABLE 7
Percent of middle and high school students in each demographic category (% overall column)
and the average percent of their teachers in each RIF / layoff category, 2010-11 to 2011-12

%
Overall
No RIF
RIF Odds ratios

RIF-
rescinded
Laid off No RIF
RIF-
rescinded
Layoff
Total
student-obs.
524,551
500,181 38,501 32,679
87.5% 6.7% 5.7%
Race / Ethnicity Relative to White
Black 9.1% 88.5% 5.8% 5.7% 1.016*** 0.728*** 1.165***
Latina/o 75.9% 87.4% 6.7% 5.9% 1.003*** 0.835*** 1.210***
Multiple 0.4% 88.6% 6.3% 5.1% 1.017*** 0.782*** 1.048
Nat. Am. 0.3% 87.2% 7.4% 5.5% 1.001 0.918* 1.124*
Pac. Isl. 0.4% 86.0% 8.0% 6.0% 0.987** 1.002 1.224***
Asian 6.6% 88.4% 7.0% 4.6% 1.15*** 0.878*** 0.935***
White 7.4% 87.1% 8.0% 4.9% 1.000 1.000 1.000
English language status Relative to NES
LEP 19.0% 87.9% 6.3% 5.8% 1.004*** 0.906*** 1.062***
Reclassified 40.3% 87.5% 6.6% 5.9% 1.000 0.939*** 1.085***
Fluent, NNE 11.5% 87.2% 7.3% 5.5% 0.996*** 1.049** 0.998
All NNE 70.8% 87.5% 6.6% 5.8% 1.000 0.947*** 1.064***
Native Eng. 29.2% 87.5% 7.0% 5.5% 1.000 1.000 1.000
Family income Relative to non-FRL
Ever FRL 90.1% 87.6% 6.6% 5.8% 1.011*** 0.815*** 1.100***
Never FRL 9.9% 86.7% 8.1% 5.2% 1.000 1.000 1.000
Special Education (SPED) Relative to non-SPED
Any SPED 89.0% 90.7% 5.0% 4.3% 1.041*** 0.725*** 0.724***
Non-SPED 11.0% 87.2% 6.9% 5.9% 1.000 1.000 1.000
Note: Multiple refers to students who report multiple race/ethnicities across years. LEP stands for limited English
proficiency; reclassified refers to students who entered the district with LEP, but were reclassified as fluent; fluent
NNE refers to students who are non-native English speakers (NNE), but entered the district fluent in English; NES
refers to native English speakers. FRL refers to students eligible for free and reduced price meals. ED stands for
emotional disturbance; SLD stands for specific learning disability, and the "other" disability category includes
deafness, developmental delay, established medical disability, hard of hearing, mentally retarded, multiple
disabilities, orthopedic impairment, other health impairment, traumatic brain injury, and visual impairment; any
SPED refers to any student enrolled in special education. +p<0.10, * p<0.05, ** p<0.01, *** p<0.001

90
TABLE 8
Difference-in-difference estimates of the likelihood a students’ teacher receives a RIF notice,
receives a layoff notice, or received a layoff notice in the prior year and was rehired to the
district

RIF Layoff Layoff in t-1
Elementary students

Students in Reed schools in 2010-11 and 2011-12 -27.8%*** -7.5%*** -3.0%***
2012 only Reed schools -24.7%*** -7.0%*** -3.8%***
2011 only Reed schools -24.9%*** -8.7%*** -2.5%***
Middle and high school students
2011/2012 Reed schools -15.6%*** -6.3%*** -3.1%***
2012 only Reed schools -12.0%*** -4.4%*** -3.2%***
2011 only Reed schools -16.3%*** -7.1%*** -5.2%***
Note: RIF stands for reduction-in-force and refers to students whose teacher received a RIF notice, regardless of
whether the notice was rescinded. Layoff refers to students whose teacher received both a RIF notice and a final
layoff notices. The outcome “Layoff in t-1” is defined as having a teacher who was laid off in the prior year and
returned to the district. For middle and high school students, who have more than one teacher throughout the day, I
use the average proportion of a student’s teacher that was RIFed, laid off, or laid off in t-1. For both elementary and
middle and high school students, there are three different treatment effects for each of the three different treatment
groups: (a) students affected by Reed in both years (2011/2012 Reed schools), (b) those affected by Reed in just
2012 (2012 only Reed schools); and (c) those affected by Reed in just 2011 (2011 only Reed schools). Treatment
indicators for the RIF and layoff variable identify students who are in Reed schools while Reed is in place. In
models predicting whether a student’s teacher was laid off in t-1 (i.e. laid off and rehired in the prior year), the
treatment indicators are changed to reflect students who were ever enrolled in a school that was classified as a Reed
school in the prior year. *** p<0.001.

91
TABLE 9
Difference-in-difference estimates of the likelihood a students’ teacher receives a RIF notice, receives a layoff notice, or received a
layoff notice in the prior year (and was rehired to the district), by student demographic categories

Elementary students

Middle and high school students

RIF Layoff Layoff in t-1 RIF Layoff Layoff in t-1
All students
Students in Reed schools in 2010-11 and 2011-12
-27.8%*** -7.5%*** -3.0%*** -15.6%*** -6.3%*** -3.1%***
Students in Reed schools in 2011-12 only
-24.7%*** -7.0%*** -3.8%*** -12.0%*** -4.4%*** -3.2%***
Students in Reed schools in 2010-11 only
-24.9%*** -8.7%*** -2.5%*** -16.3%*** -7.1%*** -5.2%***
Black students
Students in Reed schools in 2010-11 and 2011-12
-30.9%*** -8.9%*** -4.5%*** -15.0%*** -7.7%*** -3.0%***
Students in Reed schools in 2011-12 only
-25.7%*** -6.7%*** -4.4%*** -9.8%*** -4.6%*** -0.4%
Students in Reed schools in 2010-11 only
-28.3%*** -10.5%*** -2.9%** -15.2%*** -7.0%*** -3.8%***
Latina/o students
Students in Reed schools in 2010-11 and 2011-12
-27.5%*** -7.4%*** -2.9%*** -15.3%*** -6.0%*** -3.0%***
Students in Reed schools in 2011-12 only
-24.9%*** -7.1%*** -3.8%*** -12.2%*** -4.3%*** -3.6%***
Students in Reed schools in 2010-11 only
-24.3%*** -8.5%*** -2.4%*** -16.2%*** -7.2%*** -5.5%***
Native American, Pacific Islander, and students reporting multiple race/ethnicities across years
Students in Reed schools in 2010-11 and 2011-12
-15.2%* 2.7% -1.1% -20.7%*** -8.7%*** -8.0%***
Students in Reed schools in 2011-12 only
-12.8% -2.7% 6.9% -11.7%*** -5.2%*** -2.6%*
Students in Reed schools in 2010-11 only
N/A N/A -1.2% -20.5%*** -7.4%*** -6.8%***
White / Asian students
Students in Reed schools in 2010-11 and 2011-12
-24.3%*** -7.5%*** 1.6%* -20.7%*** -5.2%*** -3.1%***
Students in Reed schools in 2011-12 only
N/A N/A -2.3% -13.6%*** -5.4%*** -2.4%***
Students in Reed schools in 2010-11 only
-26.6%*** -8.2%*** -3.8%*

-21.5%*** -7.3%*** -4.3%***
Note: Each column and each panel refers to a different difference-in-difference regression. For example, the first regression shown is for all elementary student
and that model provides results for students treated in both years, in 2011-12 only, and in 2010-11 only. The second regression is for just Black students in
elementary grades. N/A refers to models that STATA can not estimate because of empty cells. This happens when, for example, no White / East Asian
elementary students who were enrolled in Reed schools in just 2011-12 had a teacher RIFed or laid off in 2011-12. *** p<0.001, ** p<.0.01, * p<.05

92
References for Essay 2
American Civil Liberties Union. (2011, January). Judge approves landmark settlement in Reed v.
State of California. Retrieved June 13, 2014 from https://www.aclusocal.org/judge-
approves-landmark-settlement-in-reed-v-state-of-california/.
Bosman, J. (2015, June 7). States confront wide budget gaps even after years of recovery. New
York Times. Retrieved June 13, 2015 from:
http://www.nytimes.com/2015/06/08/us/states-confront-wide-budget-gaps-even-after-
years-of-recovery.html.
Clotfelter, C. T., Ladd, H. F., & Vigdor, J.L. (2005). Who teaches whom? Race and the
distribution of novice teachers. Economics of Education Review, 24, 377–392.
Darling-Hammond, L. (2004). Inequality and the right to learn: Access to qualified teachers in
California's public schools. The Teachers College Record, 106(10), 1936-1966.
Darling-Hammond, L. (2000). Teacher quality and student achievement: A review of state policy
evidence. Education Policy Analysis Archives, 8(1). Retrieved September 5, 2015 from
http://epaa.asu.edu/epaa/v8n1.
Darling-Hammond, L. (1998). Unequal opportunity: Race and education. The Brookings Review,
16(2), 28.
DuFour, R., DuFour, R., Eaker, R. & Many, T. (2010). Learning by doing: A handbook for
professional communities at work. Bloomington, IN: Solution Tree Press.
Felch, J., Song, J., & Smith, D. (2010, December). When layoffs come to L.A. schools,
performance doesn't count. The Los Angeles Times. Retrieved October 17, 2014 from
http://articles.latimes.com/2010/dec/04/local/la-me-1205-teachers-seniority-20101204.

93
Goldhaber, D., Lavery, L., & Theobald, R. (2015). Uneven playing field? Assessing the teacher
quality gap between advantaged and disadvantaged students. Educational Researcher,
44(5), 293-307.
Goldhaber, D. & Theobold, R. (2014). Managing the teacher workforce in austere times: The
determinants and implications for teacher layoffs. Education Finance and Policy, 8(4),
494-527.
Hacker, J. S., Huber, G. A., Nichols, A., Rehm, P., & Craig, S. (2012). Economic insecurity
across the American states: New state estimates from the economic security index. New
York, NY: The Rockefeller Foundation.
Hahnel, C., Barondess, H., & Ramanathan, A. (2011). Victims of the churn: The damaging
impact of California's teacher layoff policies on schools, students, and communities in
three large school districts. K-12 Policy. Oakland, CA: Education Trust-West.
Ingersoll, R. M. (1999). The problem of underqualified teachers in American secondary schools.
Educational Researcher, 28(2), 26-37.
Isenberg, E., Max, J., Gleason, P., Potamites, L., Santillano, R., Hock, H., & Hansen, M. (2013).
Access to effective teaching for disadvantaged students. Washington, D.C.: National
Center for Education Evaluation and Regional Assistance, U.S. Department of Education.
Kalogrides, D., & Loeb, S. (2013). Different teachers, different peers: The magnitude and effects
of student sorting within schools. Educational Researcher, 42, 304-316.
Lankford, H., Loeb, S., & Wyckoff, J. (2002). Teacher sorting and the plight of urban schools: A
descriptive analysis. Educational Evaluation and Policy Analysis, 24, 37–62.

94
Shierholz, H. (2013). The teacher gap: More students and fewer teachers. Washington, D. C.:
Economic Policy Institute. Retrieved on June 13, 2015 from
http://www.epi.org/publication/teacher-gap-students-teachers/.
Schulte, S. (2015, July 2). CPS teacher, parents protest 1,400 layoffs, $200M. ABC 7 Chicago
News. Retrieved July 14, 2015 from: http://abc7chicago.com/politics/cps-teachers-parents-
protest-1400-layoffs-$200m-in-cuts/823307/.
United States Department of Education. (2015). Teacher shortage areas nationwide listing 1990-
1991 through 2015-2016. Office of Postsecondary Education. Washington, D.C.: Author.
Retrieved September 5, 2015 from http://www2.ed.gov/about/offices/list/ope/pol/tsa.doc.
Ingersoll, R. & May, H. (2011a). Recruitment, retention, and the minority teacher shortage.
Philadelphia, PA: Con- sortium for Policy Research in Education, University of Pennsylvania
and Center for Educational Research in the Interest of Underserved Students, University of
California, Santa Cruz.
Quiocho, A. & Rios, F. (2000). The power of their presence: Minority group teachers and
schooling. Review of Educational Research, 70(4), 485-528.
Villegas, A.M. & Lucas, T. (2004). Diversifying the teacher workforce: A retrospective and
prospective analysis. In M.A. Smylie & D. Miretky (Eds.), Developing the teacher workforce
(103rd Yearbook of the National Society for the Study of Education, Part 1) (pp. 70-104).
Chicago, IL: University of Chicago Press.

95
APPENDIX A-2: Additional Tables and Figures for Essay 2
APPENDIX FIGURE A1
Panel A: Average experience by percent of students eligible for free/reduced price lunch (FRL)
at the school level (with percent FRL in quintiles), elementary and secondary levels

Panel B: Average experience by percent of students eligible for free/reduced price lunch (FRL)
at the classroom level (with percent FRL in quintiles), elementary and secondary levels

96
APPENDIX TABLE A1
Distribution of RIFs and layoffs across secondary students, using three definitions, 2008-09 to
2009-10

Proportion of
teachers

Majority of
teachers

At least one
teacher

RIF-
rescinded
Layoff

RIF-
rescinded
Layoff

RIF-
rescinded
Layoff
Total 7.5% 4.0% 2.5% 0.9% 31.9% 20.8%
Race / Ethnicity
Black 7.7% 4.3% 2.5% 0.9% 33.6% 22.9%
Latina/o 7.8% 4.2% 2.7% 1.0% 32.9% 22.2%
More than one race/ethnicity 8.0% 4.0% 2.5% 1.2% 33.0% 19.0%
Native American 6.3% 3.5% 1.8% 0.7% 28.4% 19.1%
Pacific Islander 7.5% 3.2% 3.1% 0.7% 29.3% 17.7%
Asian 5.1% 2.1% 1.4% 0.3% 23.2% 11.7%
White 5.8% 2.1% 1.5% 0.3% 26.9% 11.6%
ELL Status
Limited English proficiency 8.6% 4.8% 3.3% 1.2% 35.7% 24.2%
Reclassified English proficient 7.2% 4.0% 2.3% 0.9% 31.3% 21.4%
Fluent, non-native English 7.3% 3.3% 2.7% 0.7% 30.0% 17.5%
Native English speaker 7.7% 4.2% 2.6% 1.0% 32.6% 21.8%
All non-native English speakers 6.9% 3.4% 2.2% 0.7% 30.0% 18.2%
Family income
Ever enrolled in FRL 7.6% 4.2% 2.7% 1.0% 32.1% 21.7%
Never enrolled in FRL 6.4% 2.7% 1.5% 0.3% 30.3% 15.8%
Special education
Any SPED classification 5.3% 2.8% 1.6% 0.5% 24.2% 16.0%
No SPED classification 7.7% 4.1% 2.6% 0.9% 32.8% 21.4%
Note: More than one race/ethnicity refers to students who report multiple race/ethnicities across years. Reclassified
English proficient refers to students who entered the district with limited English proficiency, but were reclassified
as fluent; and fluent non-native English refers to students who are non-native English speakers, but entered the
district fluent in English. FRL refers to students eligible for free and reduced price meals. The "other disability”
category includes deafness, developmental delay, established medical disability, hard of hearing, mentally retarded,
multiple disabilities, orthopedic impairment, other health impairment, traumatic brain injury, and visual impairment;
any SPED refers to any student enrolled in special education.

97

APPENDIX TABLE A2
Descriptive statistics for Reed schools, comparable schools, and all other schools
2009-10 School year 2010-11 School year 2011-12 School year

Reed
2011
schools
Reed
2011 /
2012
Reed
2012
schools
Comp-
arable
schools
All
other
schools
Reed
2011
schools
Reed
2011 /
2012
Reed
2012
schools
Comp-
arable
schools
All
other
schools
Reed
2011
schools
Reed
2011 /
2012
Reed
2012
schools
Comp-
arable
schools
All
other
schools
Student char.
% FRL 88.5% 92.3% 85.9% 88.1% 78.0% 87.4% 90.2% 84.0% 88.4% 78.1% 88.3% 92.6% 87.2% 89.5% 78.3%
% LEP 44.5% 31.3% 31.5% 34.7% 29.6% 44.5% 29.9% 29.3% 32.5% 28.1% 44.8% 28.0% 28.9% 30.6% 26.7%
% non-White 96.4% 98.6% 97.4% 95.9% 87.8% 97.1% 98.8% 98.1% 96.3% 87.9% 97.3% 98.8% 98.6% 96.2% 88.2%
API 723 627 605 682 756 716 661 619 702 770 737 680 655 718 782
% top quin API 6.7% 0.0% 0.0% 1.5% 19.9% 5.1% 0.0% 0.0% 1.5% 20.6% 5.0% 0.0% 0.0% 2.2% 21.6%
% bot. quin API 24.7% 88.5% 91.1% 50.9% 16.7% 39.9% 69.8% 91.0% 44.4% 17.1% 38.0% 66.5% 100.0% 40.1% 15.7%
Mean enroll. 551 1,895 1,920 1,213 1,047 556 1,406 1,791 1,081 994 549 1,205 1,430 956 921
Teacher char.
Mean exp. 9.5 8.6 8.6 9.5 10.5 9.5 8.7 9.0 10.0 11.1 9.7 9.1 9.0 10.7 11.7
% novice tchrs 8.6% 13.5% 10.7% 7.8% 4.7% 8.3% 12.9% 7.4% 5.1% 3.0% 3.2% 9.1% 11.4% 3.0% 1.8%
% MA / Doc 43.5% 41.1% 42.4% 37.8% 35.2% 44.9% 41.6% 42.4% 38.8% 36.3% 47.4% 42.1% 41.6% 40.1% 37.5%
% NBPTS 3.5% 2.3% 2.0% 3.2% 3.8% 5.4% 2.5% 2.1% 3.3% 4.3% 6.7% 2.7% 2.5% 3.9% 4.6%
RIF variables
% Not RIFed 16.9% 8.3% 5.5% 8.7% 7.9% 0.0% 0.5% 15.8% 17.3% 14.9% 26.6% 0.7% 0.7% 15.4% 14.2%
% RIF-re. 14.1% 5.9% 3.2% 6.9% 7.0% 0.0% 0.5% 8.3% 10.3% 10.7% 12.6% 0.4% 0.3% 9.9% 10.6%
% Laid off 2.7% 2.4% 2.3% 1.8% 0.9% 0.3% 2.3% 7.5% 7.1% 4.2% 14.0% 2.2% 1.4% 5.6% 3.6%
Num of tch. 255 1,992 951 7,122 16,438 336 1,819 868 6,902 15,907 342 1,655 734 6,484 15,367
Num of sch. 9 27 10 143 420 13 32 10 155 427 13 32 13 164 436
Note: FRL stands for free/reduced price lunch and indicates students participating in the federal school lunch program. LEP indicates students with limited
English proficiency. API stands for Academic Performance Index, a measurement of a school’s academic achievement on reading and math standardized exams.
This table demonstrates that comparable schools had similar characteristics as Reed schools.

98
APPENDIX TABLE A3
Difference-in-difference estimates of the likelihood a students’ teacher receives a RIF notice,
layoff notice, or was laid off in the prior year and returned to the district

Elementary Middle & high school
RIF Layoff Layoff in t-1 RIF Layoff Layoff in t-1
Treatment indicators
2012 Reed school x
2013 year FE
-1.106*** -0.024***
(0.141) (0.001)
2011 Reed school x
2013 year FE
0.343** -0.011***
(0.109) (0.002)
2012 Reed school x
2012 year FE
-3.005*** -2.135*** 0.270* -0.116*** -0.039*** 0.022***
(0.085) (0.101) (0.111) (0.002) (0.001) (0.002)
2011 Reed school x
2012 year FE
0.253*** 0.759*** -0.688*** -0.056*** -0.039*** -0.058***
(0.041) (0.055) (0.108) (0.002) (0.002) (0.002)
2012 Reed school x
2011 year FE
0.065* 0.207*** 0.601*** 0.006** 0.009*** 0.004*
(0.032) (0.046) (0.157) (0.002) (0.002) (0.001)
2011 Reed school x
2011 year FE
-4.185*** -3.311*** 0.738*** -0.185*** -0.096*** -0.011***
(0.118) (0.141) (0.151) (0.002) (0.002) (0.002)
2012 Reed school x
2010 year FE
0.030 -0.337*** 0.005* 0.009***
(0.031) (0.097) (0.002) (0.001)
2011 Reed school x
2010 year FE
0.002 0.361*** -0.043*** -0.048***
(0.033) (0.094) (0.002) (0.002)
2012 Reed school
0.310*** 0.212*** -0.274** 0.005** -0.004** (0.001)
(0.020) (0.029) (0.097) (0.002) (0.001) (0.001)
2011 Reed school
0.136*** 0.145*** 0.384*** 0.052*** 0.047*** 0.018***
(0.022) (0.031) (0.091) (0.002) (0.002) (0.002)
School types
Middle school
0.490*** 0.172 0.800*** -0.059*** 0.013*** 0.004
(0.070) (0.121) (0.218) (0.006) (0.003) (0.003)
High school
-0.102*** 0.006+ 0.001
(0.006) (0.003) (0.003)
K-12 school (ref.
cat. is Elementary)
0.225*** 0.274*** 0.623*** -0.070*** 0.007+ 0.004
(0.037) (0.057) (0.061) (0.007) (0.004) (0.003)
Year fixed effects
2013
1.196*** -0.002***
0.04 0.001
2012
-0.025+ -0.349*** 1.394*** -0.067*** -0.010*** 0.011***
0.014 0.022 0.038 0.001 0.001 0.001
2011
0.138*** -0.041* -1.137*** -0.044*** 0.012*** -0.022***
(0.013) (0.020) (0.067) (0.001) (0.001) (0.001)
2010
-0.533*** -2.198*** -0.143*** -0.046***
(0.014) (0.041) (0.001) (0.001)
Constant
-1.330*** -2.386*** -4.104*** 0.287*** 0.064*** 0.041***
(0.010) (0.014) (0.034) (0.006) (0.003) (0.003)
N 335,689 335,689 258,289 457,458 457,458 356,618
R-squared - - - 0.154 0.055 0.023
Note: Models for elementary grades are logisitic regressions and models for secondary grades are OLS regressions
(predicting the percent of a student’s teachers that were RIFed or laid off. . +p<0.10, * p<0.05, ** p<0.01, ***
p<0.001

99
APPENDIX TABLE A4
Difference-in-difference estimates of the likelihood a students’ teacher receives a RIF notice,
receives a layoff notice, or received a layoff notice in the prior year (and was rehired to the
district)
RIF Layoff Layoff in t-1
Elementary students
2011/2012 Reed schools
2009-10 to 2010-11 (2010-11 to 2011-12 for layoff in t-1) -29.2%*** -8.1%*** -3.3%***
2009-10 to 2011-12 (2010-11 to 2012-13 for layoff in t-1) -24.7%*** -4.3%*** -4.2%***
Mean of two years of pre-treatment to mean of two years of treatment -27.8%*** -7.5%*** -3.0%***
2012 only Reed schools
2009-10 to 2011-12 (2010-11 to 2012-13 for layoff in t-1) -23.4%*** -5.1%*** -4.1%***
2010-11 to 2011-12 (2011-12 to 2012-13 for layoff in t-1) -26.2%*** -8.9%*** -3.9%***
Mean of three years of pre-treatment to treatment year -24.7%*** -7.0%*** -3.8%***
2011 only Reed schools
2009-10 to 2010-11 (2010-11 to 2011-12 for layoff in t-1) -24.5%*** -8.4%*** -2.7%***
2010-11 to 2011-12 (2011-12 to 2012-13 for layoff in t-1) 29.9%*** 15.5%*** 6.6%***
Mean of two years of pre-treatment to treatment year -24.9%*** -8.7%*** -2.5%***
Middle and high school students
2011/2012 Reed schools
2009-10 to 2010-11 (2010-11 to 2011-12 for layoff in t-1) -14.0%*** -4.7%*** -2.8%***
2009-10 to 2011-12 (2010-11 to 2012-13 for layoff in t-1) -13.4%*** -3.8%*** -2.7%***
Mean of two years of pre-treatment to mean of two years of treatment -15.6%*** -6.3%*** -3.1%***
2012 only Reed schools
2009-10 to 2011-12 (2010-11 to 2012-13 for layoff in t-1) -12.1%*** -4.7%*** -2.8%***
2010-11 to 2011-12 (2011-12 to 2012-13 for layoff in t-1) -12.3%*** -4.8%*** -4.6%***
Mean of three years of pre-treatment to treatment year -12.0%*** -4.4%*** -3.2%***
2011 only Reed schools
2009-10 to 2010-11 (2010-11 to 2011-12 for layoff in t-1) -14.2%*** -4.7%*** -4.6%***
2010-11 to 2011-12 (2011-12 to 2012-13 for layoff in t-1) 12.9%*** 5.6%*** 4.7%***
Mean of two years of pre-treatment to treatment year -16.3%*** -7.1%*** -5.2%***
Note: the outcome “Layoff in t-1” is defined as having a teacher who was laid off in the prior year and returned to
the district. For middle and high school students, who have more than one teacher throughout the day, I use the
average proportion of a student’s teacher that was RIFed, laid off, or laid off in t-1. +p<0.10, * p<0.05, ** p<0.01,
*** p<0.001

100
APPENDIX TABLE A5
Odds ratios predicting whether each elementary student’s teacher falls into each lagged RIF /
layoff category, 2009-10 to 2010-11 (next year results of Phase I layoffs)

No RIF in prior
year
RIF-rescinded in
prior year
Layoff in prior
year
Not present in
prior year
550,854
477,099 62,761 5,538 5,456
86.6% 11.4% 1.0% 1.0%
Race / Ethnicity
Black 0.938*** 1..358*** 1.320*** 2.339***
Latina/o 0.1006*** 0.949*** 0.742*** 1.635***
Multiple 1.005 0.961 0.619* 1.741**
Native American 1.015 0.860* 0.794 1.973**
Pacific Islander 0.996 1.034 0.863 1.239
Asian 1.010*** 0.937** 0.837** 1.124
White 1.000 1.000 1.000 1.000
English language status
LEP 1.001 0.966*** 0.968 1.358***
Reclassified 0.991*** 1.107*** 0.809*** 0.771*
Fluent, NNE 1.013*** 0.937*** 0.691*** 0.907+
All NNE 1.001 0.996 0.854*** 1.171***
Native Eng. 1.000 1.000 1.000 1.000
Family income
Ever FRL 1.006*** 0.935*** 0.784*** 2.073***
Never FRL 1.000 1.000 1.000 1.000
Special Education
Any SPED 1.027*** 0.698*** 1.440*** 1.864***
Non-SPED 1.000 1.000 1.000 1.000
Note: Odds ratios are relative to the group shown in the bottom row of each panel (i.e., White students, native
English speaking students, never FRL students, and non-SPED students).

101
APPENDIX TABLE A6
Odds ratios predicting the proportion of each middle and high school student’s teachers that fall
into each lagged RIF / layoff category, 2009-10 to 2010-11 (next year results of Phase I layoffs)

No RIF in prior
year
RIF-rescinded in
prior year
Layoff in prior
year
Not present in
prior year
548,954
478,968 38,546 14,530 16,911
87.3% 7.0% 2.6% 3.1%
Race / Ethnicity
Black 0.942*** 1.204*** 3.048*** 2.036***
Latina/o 0.945*** 1.256*** 2.879*** 2.003***
Multiple 0.958*** 1.182*** 2.635*** 1.699***
Native American 0.975*** 1.071 1.840*** 1.645***
Pacific Islander 0.965*** 1.224*** 2.033*** 1.562***
Asian 1.003+ 0.870*** 1.284*** 1.148***
White 1.000 1.000 1.000 1.000
English language status
LEP 0.962*** 1.171*** 1.551*** 1.377***
Reclassified 0.987*** 1.054*** 1.279*** 1.051***
Fluent, NNE 1.002* 1.075*** 0.954** 0.799
All NNE 0.982*** 1.092*** 1.311*** 1.110***
Native Eng. 1.000 1.000 1.000 1.000
Family income
Ever FRL 0.956*** 1.230*** 2.040*** 1.642***
Never FRL 1.000 1.000 1.000 1.000
Special Education (SPED)
Any SPED 1.033*** 0.678*** 0.724*** 1.084***
Non-SPED 1.000 1.000 1.000 1.000
Note: Odds ratios are relative to the group shown in the bottom row of each panel (i.e., White students, native
English speaking students, never FRL students, and non-SPED students).

102
APPENDIX TABLE A7
Odds ratios predicting whether each elementary student’s teacher falls into each lagged RIF /
layoff category, 2011-12 to 2012-13 (next year results of Phase II layoffs)

No RIF in prior
year
RIF-rescinded in
prior year
Layoff in prior
year
Not present in
prior year
534,961
433,633 71,801 24,458 5,069
81.1% 13.4% 4.6% 0.9%
Race / Ethnicity
Black 0.984*** 1.017 1.125*** 1.496***
Latina/o 1.033*** 0.859*** 0.872*** 1.134*
Multiple 1.000 0.976 1.0403 1.220
Native American 1.011 0.941 0.989 1.072
Pacific Islander 1.001 1.073 0.835 0.584
Asian 0.985*** 1.054** 1.042 1.182*
White 1.000 1.000 1.000 1.000
English language status
LEP 1.013 0.890*** 1.051 1.216
Reclassified 0.977 1.159*** 0.995 0.695
Fluent, NNE 1.012 0.974* 0.915 0.775
All NNE 1.003* 0.978** 1.015 1.005
Native Eng. 1.000 1.000 1.000 1.000
Family income
Ever FRL 1.038*** 0.836*** 0.883*** 1.292***
Never FRL 1.000 1.000 1.000 1.000
Special Education
Any SPED 1.046*** 0.695*** 0.745*** 3.210***
Non-SPED 1.000 1.000 1.000 1.000
Note: Odds ratios are relative to the group shown in the bottom row of each panel (i.e., White students, native
English speaking students, never FRL students, and non-SPED students).

103
APPENDIX TABLE A8
Odds ratios predicting the proportion of each middle and high school student’s teachers that fall
into each lagged RIF / layoff category, 2011-12 to 2012-13 (next year results of Phase II layoffs)

No RIF in prior
year
RIF-rescinded
in prior year
Layoff in prior
year
Not present in
prior year
503,953
478,396 36,311 21,512 12,735
87.1% 6.6% 3.9% 2.3%
Race / Ethnicity Relative to White
Black 0.999 0.721*** 1.073*** 2.448***
Latina/o 0.996*** 0.837*** 1.178*** 1.699***
Multiple 1.005 0.730*** 1.132*** 1.913***
Native American 0.996 0.881** 1.135** 1.563***
Pacific Islander 0.995 0.995 1.120 1.022
Asian 1.017*** 0.843*** 0.970*** 0.910*
White 1.000 1.000 1.000 1.000
English language status Relative to NES
LEP 0.999+ 0.902*** 1.081*** 1.204***
Reclassified 1.002*** 0.964*** 1.062*** 0.914***
Fluent, NNE 1.006*** 1.015+ 0.979+ 0.787***
All NNE 1.002*** 0.957*** 1.052*** 0.963***
Native Eng. 1.000 1.000 1.000 1.000
Family income Relative to non-FRL
Ever FRL 1.010*** 0.779*** 1.049*** 1.485***
Never FRL 1.000 1.000 1.000 1.000
Special Education (SPED) Relative to non-SPED
Any SPED 1.021*** 0.753*** 0.737*** 1.404***
Non-SPED 1.000 1.000 1.000 1.000
Note: Odds ratios are relative to the group shown in the bottom row of each panel (i.e., White students, native
English speaking students, never FRL students, and non-SPED students).

104
Essay 3. Cost-Effectiveness in the Context of School Budget Restoration:
Comparing the Cost and Effects of Teacher Salary Increases and Class Size Reduction
in the Los Angeles Unified School District
The global recession that began in 2008 led to an unprecedented decline in public school
funding in the United States. The recession had drastic effects on labor markets across the
country. Employment declined by 8.7 million jobs (6%) from 2007 to 2011 (Federal Reserve
Bank, 2013), a time span during which the economy should have added 7 million jobs (Stiglitz,
2012). The teaching profession was particularly hard hit. Over the same four-year period,
employment in local educational agencies declined by 141,500 positions (Bureau of Labor
Statistics, 2012).
While progress toward recovery has been slow, especially for public education, the
economy has improved since the 2008 recession (Oliff & Leachman, 2011). According to a
report on school funding for the 2013-14 school year, only 14 states had fully restored per-pupil
funding in K-12 education back to pre-recession levels, but 29 states increased funding in 2013-
14 relative to the previous year (Leachman & Mai, 2013). In addition, rising income and
property tax revenues provided a majority of states with budget surpluses in fiscal year 2014
(Lyman, 2014). Many state governors appear poised to increase funding for K-12 education over
the next several years (Bunting, Kueneman, Louttit, Park & Parker, 2014).
The context in the state of California has mirrored national trends. The state’s teacher
workforce decreased by about 32,000 (11%) between 2008 and 2011, with the greatest declines
concentrated in Los Angeles County (Estrada, 2012). Following budgetary shortfalls, the Los
Angeles Unified School District (LAUSD) conducted a reduction-in-force in each school year
from 2008-09 to 2011-12. However, two state-level changes in to school finance in California

105
specifically have helped pull districts out of financial crisis. First, voters passed Proposition 30 in
November 2012, a state tax initiative that increased state funding to local school districts.
Second, the state replaced its school finance system with the Local Control Funding Formula
(LCFF), which allocates supplementary funding to districts serving higher proportions of English
language learners, socioeconomically disadvantaged students, and foster care youth. LCFF also
provides districts with greater flexibility over how money is spent. Due to Proposition 30 and the
LCFF, administrators in LAUSD have estimated that the district will receive approximately $195
million of additional funding per school year, beginning in 2014-15.
42

As district budgets stabilize, administrators in LAUSD, like their counterparts around the
country, need to decide how to allocate new resources. For example, one possibility is to restore
funding according to previous allocation patterns, while another is to consider new allocations
that may better promote student outcomes. Two of the most common uses of educational funding
are (a) reducing class size and (b) raising teachers’ salaries (Baker, 2012; Chingos, 2013; Hoxby,
2000). Whether to hire more teachers, thereby reducing class size, or increase teacher salaries,
thereby leaving class sizes in tact, represents a fundamental tradeoff for researchers and
policymakers working to ensure equitable and adequate school funding (Baker, 2012; Chingos,
2013; Gilpin & Kaganovich, 2012; Hanushek & Luque, 2000; Odden & Picus, 2013). Any
reduction in the average class size will make it more difficult for districts to increase teacher pay,

42
District funding through LCCF began in 2013-14 school year and will be phased in gradually over eight years (Ed
Source, 2013). Estimates from LAUSD suggest the district received approximately $332 million in additional
funding beginning in school year 2014-15 (LAUSD, 2014a). However, $137 million of this amount is generated
through supplemental and concentration funding and must be targeted to high need students as part of the Local
Control Accountability Plan (LCAP) (see Taylor, 2013 and LAUSD, 2014a for information on the LCAP). The
district will receive another increase in state funding for the 2015-16 school year and will continue to receive
additional funding until LFCC is fully implemented in school year 2020-21 (and the amount of additional funding
over the previous school year will decrease each year as the amount of state funding approaches the LCFF funding
target and the formula is fully implemented). The district was provided complete flexibility over $195 million in
new, annual funding for the 2014-15 school year. Therefore, the analysis described here demonstrates potential
outcomes of alternative uses of receiving $195 million of additional funding in perpetuity. As I expand upon below,
I also consider other proposals not directly tied to a particular dollar figures (e.g., restoring class size to pre-layoff
levels).

106
while raising teachers’ salaries diminishes the extent to which class sizes can be reduced within a
particular budget.
Variations of these policies have been implemented and studied at the district, state, and
federal levels over the past decades. However, very few studies examine the policies’ impacts on
student achievement and their relative cost simultaneously, or their cost-effectiveness (Harris,
2009; Levin et al, 2012). Cost-effectiveness methods allow researchers to determine the effects
of various policies on student outcomes, relative to resource use. Such analyses provide guidance
to local practitioners and policymakers, as they can shed light on the most productive and, in
some cases, most equitable uses of limited funds.
The extent to which one policy (i.e., decreasing class size or raising teacher salaries)
represents a better use of funds is therefore an empirical question, and motivates the current
study. The purpose here is to examine the cost-effectiveness of variations of class size reduction
and teachers’ salary increases in LAUSD, in order to provide insight into the best uses of limited
educational resources. I base this study in the context of LAUSD for several reasons. The
district’s experience with several years of severe budget cuts, followed by a significant budget
restoration, is representative of the situation facing districts across the country. LAUSD is also
the second largest district in the nation, serving a diverse student population with respect to
race/ethnicity, family income, English proficiency, and overall student need (Advancement
Project, 2014; Strunk, Marsh, Hashim, Bush & Weinstein, 2014). This particular point in
LAUSD’s history represents a unique opportunity to analyze the costs and benefits of proposals
that were actually considered and compare those proposals to the spending policies that were
ultimately agreed upon. Findings from this study have implications locally and nationally. As
district and state leaders strive to make effective resource allocation decisions, determining the

107
cost-effectiveness of class size and teacher salary policies is particularly salient. The analyses
add to understanding of school resource allocation by highlighting the conditions under which
each of these policies is more cost-effective. Finally, the study contributes to literature
employing cost-effectiveness analysis by demonstrating how different permutations of policy
designs can alter the balance of costs and effects.
I examine several permutations of teachers’ salary increases and class size reduction
policies in order to estimate which of the two uses of resources will have the largest impact on
student achievement per dollar spent. For each policy, I consider both across-the-board spending
proposals as well as proposals targeted at specific grades, teachers, and schools. I also consider
policies the combine salary increases with class size reduction and compare their cost-
effectiveness to proposals that involve implementing just one of these policies in isolation. The
district reached a spending agreement in April 2015. I estimate the costs and effects of this final
agreement and compare those estimates to the alternate proposals described above.
Overall, I find that rehiring teachers and reducing class size is more cost-effective than
raising teachers’ salaries. However, this finding only holds under certain class size reduction
policies. Lowering class sizes across-the-board is less cost-effective than salary increase policies
that are targeted to specific teachers or schools. Moreover, my results are sensitive to the
projected effect on student achievement of salary increases. If I assume salaries will have the
magnitude of effects found in North Carolina, where salary incentives were provided to
encourage teachers to work in particular schools within districts (Clotfelter, Glennie, Ladd, &
Vigdor, 2008), then salary increases are generally more cost-effective than class size reduction
policies.
Regardless of the parameters used, I find that targeting funding at specific grades or

108
schools yields higher student achievement gains per dollar, compared to across-the-board
spending policies. Further, I find policies that combine salary increases with class size reduction
are generally more cost-effective than implementing either of these policies in isolation. I find
that the final spending agreement that the district and teachers’ union reached in April 2015
compares favorably to many, but not all, of the spending proposals examined in this study.
43

In the following section, I provide a brief overview of the budgetary policies that were
under consideration in LAUSD over the time period of negotiations (2012-13 to 2014-15 school
years), the stakeholders involved in the process, and the final agreement that was reached in
April of 2015. The subsequent section reviews literature employing cost-effectiveness analysis in
educational settings. I then describe the data and cost-effectiveness methods I use in my policy
simulations. The final three sections review methods for projecting policy effects, present
findings and sensitivity analyses, and discuss implications, policy recommendations, and
directions for further research.
Spending Proposals in LAUSD
For over two years, the LAUSD Board of Education, district administrators,
representatives of United Teachers of Los Angeles (UTLA, the union representing all non-
administrator certificated staff), and local community members negotiated over the terms of new
spending proposals. The proposals were based on additional funding allocated to the district
through Proposition 30 and LCFF as well as the additional flexibility afforded through this new
funding formula.
Most LAUSD school board members expressed a preference for using the funds to lower
class sizes. During a June 2013 LAUSD Board meeting, three board members co-sponsored a

43
Importantly, my estimates are necessarily limited by the narrow focus on student achievement; considering all
outcomes associated with class size reduction and salary increases (e.g., students’ non-cognitive outcomes, teacher
morale, etc.) may change the balance of costs and benefits for these two policies.

109
resolution requiring the superintendent to examine the feasibility of restoring class sizes to pre-
layoff levels (LAUSD, 2013). Similarly, a fourth board member stated in January 2014 that new
funds should be used to “lower class sizes, rehire nurses, librarians, counselors and other critical
staff members, and provide raises to all LAUSD employees who have made sacrifices during the
recession” (LAUSD, 2014b). Other board members supported providing raises for staff
members, extending the school year, expanding the arts program, and restoring the summer
intervention and enrichment program (LAUSD Budget Realities, 2013).
During this same time period, then-superintendent John Deasy expressed concern that the
district could not afford to rehire all previously laid off staff (LAUSD, 2013). Instead, Deasy
pushed for salary increases for all certificated staff members and suggested incentives for
teachers to work in harder-to-staff schools (Blume, 2014). Since the time current superintendent
Ramon Cortines was hired, the district has adjusted its salary proposal from 8.6 percent over
three years to 6 percent over a one-year period.
44

UTLA announced on January 16, 2014 their request for a 17.6% raise for its members
over two years (Fletcher, 2014). During ongoing negotiations with the district, UTLA adjusted
its salary increase proposal to 10 percent in one year (Romo, 2014) and then to 8.5 percent over
two years (UTLA, 2015).
45
The union has also continually supported rehiring previously laid off
staff members, and its most recent spending proposal included $525 million for class size
reduction (LAUSD, 2015b).

44
The district’s most recent salary proposal (before the final agreement was reached) was made on January 14,
2015. The proposal includes a 4 percent increase to all salary schedules, retroactive to July 1, 2014, and an
additional 2 percent annual salary increase through additional responsibilities and professional development
activities that equate to four additional work days, beginning in the 2015-16 school year (LAUSD, 2015a). The
district’s proposal is outlined on their website (home.lausd.net/ourpages/auto/2015/1/14/40061727/UTLA-
DistrictProposal-ArticleXIV-RevisedSalaryIncrease-01-14-15.pdf).
45
UTLA’s most recent salary proposal, which was made prior to reaching a final agreement with the district, was
released to the public on January 22, 2015 (UTLA, 2015b) and is outlined in further detail on their website
(www.utla.net/system/files/UTLASalaryProposal.pdf).

110
As part of the regulations included in the Local Control Funding Formula, the district was
also required to seek input from the local community on what educational programs new funds
would support. During the 2012-13 school year, the district conducted approximately 100 town
hall-style meetings to elicit input from the local community. Parents and community members at
these meetings echoed the support for reductions in class size, staff hiring, and raising salaries,
among other initiatives (Jones, 2013).
In short, the two main bargaining parties, UTLA and LAUSD administrators, moved
closer and closer to an agreement from 2012 to 2014. However, despite continual compromises
over two years, up until mid-April of 2015, there remained an $833 million gap between the
UTLA and LAUSD proposals (LAUSD, 2015b). At the heart of the disagreements were the
specific details of spending proposals for two commonly implemented policies, class size
reduction and teacher salary increases. Both parties agreed negotiations had reached an impasse
and a third party mediator was appointed in February 2015 (Blume, 2015; Oshiro, 2015). Finally
on April 14, 2015 the LAUSD Board approved a spending agreement that included a salary
increase of 10.4% and set maximum average class sizes across all grades in the district.
The class size reduction plan included maximum average class sizes of 24 for all
classrooms in transition kindergarten, kindergarten, and grades one through three, for all schools.
The class size reduction plan also mandated lower average class sizes in schools categorized as
“predominantly Hispanic, Black, Asian, or other Non-Anglo” (PHBAO), which in 2012-13,
comprised 584 of the 692 active schools or about 83.9% of students. Finally, the plan called for a
two-student reduction in grades seven and eight English and math classes, which the district
estimated would cost $13 million. The parties agreed on a salary increase phased in over time,
with 4% retroactive to July 1, 2014, 2% retroactive to January 1, 2015, a 2% increase effective

111
July 1, 2015 and a 2% increase effective January 1, 2016. Each salary increase compounds so
that the actual total salary increase, when fully phased in, represents a 10.4% increase at each
point of the current salary schedule. I describe the final agreement in greater detail and how I
estimate its cost in the Methods section below.
In summary, despite the considerable disagreement around the use of new state funding
coming to LAUSD through Proposition 30 and LCFF, the parties involved ultimately reached an
agreement. In the following section, I discuss past studies of the cost-effectiveness of educational
interventions generally and discuss in greater detail how the current study builds off these past
efforts and brings new insights to the difficult budgetary decisions that districts across the
country are currently facing.
Literature Review of Cost-Effectiveness Analysis in Education
In this section I first discuss how the tools of cost-effectiveness analysis have been
applied in educational research. I then present an overview of research on class size reduction
and teacher salary increases focusing specifically on studies that assess both the cost and effects.
Application of Cost-Effectiveness Analysis in Educational Research
Cost-effectiveness analysis ranks alternative policy options according to a measure of
their effect on outcomes relative to their resource use (Levin, 2001). Measures of educational
effectiveness are typically based on the findings of past studies and expressed in units of
standard deviations of student achievement (Levin & Belfield, 2015). Past studies define costs as
the total value of resources used to achieve the intended outcome (Levin & McEwan, 2000). For
example, one of the earliest applications of cost-effectiveness analysis in education compared
teacher-hiring decisions based on verbal scores to those based on experience (Levin, 1970). In
this work, Levin (1970) showed that hiring teachers based on verbal scores was five to ten times

112
as cost-effective as hiring teachers based on experience. More recently, cost-effectiveness
analysis has been applied to a broad range of educational interventions, including comprehensive
school reform (Borman & Hewes, 2002), dropout prevention programs (Levin et al., 2012), class
size reduction (Aos & Pennucci, 2013; Levin, Glass, & Meister, 1987), full day kindergarten
(Aos, Miller, & Mayfield, 2007), and pre-school (Barnett & Masse, 2007; Belfield, Nores,
Barnett, & Schweinhart, 2006).
46

Cost-effectiveness analysis is severely underutilized in educational evaluation literature
(Levin & Belfield, 2015; Ross, Barkaoui, & Scott, 2007). Despite the large sums of money
invested in funding public education, past reviews of extant literature have revealed few studies
of educational programs or policies that take into account the cost (e..g, Clune, 2000; Monk &
King, 1993; Ross, Barkaoui, & Scott, 2007). Levin and Belfield (2015) and Levin (2001)
provides several explanations for the scarcity of such research in the education literature
including the lack of demand from policymakers, lack of expertise among educational evaluation
specialists in the use of cost-effectiveness analysis, and the few number of evaluations in
education that rigorously measure program or policy impacts. In short, while some exemplar
studies in the educational evaluation literature that employ cost-effectiveness methods, this work
is relatively rare and most cost analyses do not move beyond rhetorical claims of cost-
effectiveness.
Class Size Reduction and Teacher Salary Increases
A limited number of studies have assessed both the costs and effects of class size
reduction and teacher compensation policies. However, past studies suffer from limitations that

46
Levin and McEwan (2000) distinguish cost-benefit analysis from cost-effectiveness analysis, where the former
converts outcome measures to dollar figures, while the latter measures outcomes in non-monetary terms. Aos and
Pennucci (2013) estimate the cost-benefit of class size reduction in the state of Washington and find that class size
reduction yields positive monetary net benefits to society. However, their study was not designed to examine
alternate uses of educational resources.

113
make them less useful for district policymaking. First, prior studies do not base cost estimates on
empirical data. For example, Greenwald, Hedges, and Laine (1994) examine the cost-
effectiveness of several policies, including increasing teachers’ salaries and decreasing class size.
The authors find that salary increases are more cost-effective that class size reduction; however,
they base their cost estimates on broad assumptions of the number of students per teacher, the
average teacher salary, and the percent of per-pupil expenditures allocated to teacher salaries,
rather than on an empirical analysis of resource use within a school or district. Levin, Glass, and
Meister (1987) compared the cost-effectiveness of class size reduction to peer tutoring, computer
assisted instruction, and lengthening the school day. The authors employ a more thorough cost
analysis of class size reduction, accounting for the cost of teachers’ salaries, physical classroom
space, insurance, energy needs, and classroom furnishings. However, given data constraints, cost
estimates are based on a hypothetical classroom using national average prices. While this
approach is useful for measuring the average cost of replicating a class size reduction policy,
potentially increasing the generalizability of the results, the findings do not directly apply to any
particular school or district. Finally, past studies of the cost of class size reduction do not take
into account shocks to the local labor market or changes in costs and effects over time (e.g., Aos
& Pennucci, 2013; Reichardt, 2000).
In this study, I am able to build on previous research by using administrative student-
level data from the district under study (LAUSD) to assess the cost of class size reduction for
various grades and schools. Because I have data on teachers’ salaries and fringe benefits,
including pension contributions, I am able to contrast cost estimates of class size reduction with
those of salary increases. I extend past research on the cost-effectiveness of class size reduction
and teachers’ salary increases by allowing costs and effects to change over time and accounting

114
for shocks in the local labor market. Using the methods commonly employed in cost-
effectiveness analyses, I combine cost estimates with measures of effectiveness to show how
salary increases and class size reduction compare in terms of their effect on student achievement
relative to cost. By using a specific example of spending proposals within a large school district,
this study also demonstrates how other districts can use the tools of cost-effectiveness analysis to
guide local decision-making in the future. The following section describes the administrative
data and outlines my analytic approach.
Data and Analytic Approach
In this section, I first describe the student-, teacher-, and school-level data used in the
analysis. I then describe the simulated policy variations, the approach to estimating costs, and the
methods used to project policy effects.
Data and Measures of Class Size
This study draws on a panel of LAUSD teacher- and student-level administrative data for
school years 2007-08 to 2012-13. These data consist of: 1) anonymized employee demographic
information including race/ethnicity, gender, years of experience, highest-degree earned, school
and classroom placement, layoff status, job title, contract status, salaries, fringe benefits
(including pension contributions), and all credentials and certificates held by each employee; as
well as 2) anonymized student data including California Standardized Test (CST) scores in math
and English Language Arts, race/ethnicity, gender, grade level, school and classroom placement,
free lunch status, disability (if any), English language proficiency, home language, course
enrollment, teacher assignment, individual grades for each class and overall grade point average,
date of birth, zip code, and parental education. These administrative data are combined with
school-level data from the California Department of Education that include each schools’

115
Academic Performance Index score (API, a composite score that combines various student
performance measures into a single weighted measure), total student enrollment, number of
students tested, and the percent of students who are classified in the following categories: special
education; enrolled in free or reduced price lunch programs; reclassified English proficient;
African American; Asian; Filipino; Hawaiian / Pacific Islander; Latino/a; and White. I limit the
dataset to K-12 traditional district schools and dependent charter schools that operate within the
district’s collectively bargained employee contract.
I exclude non-traditional schools including independent charters, community day schools,
alternative schools, early education centers, and special education centers because these schools
are not governed by negotiations and salary schedules or were largely protected from recent
budget cuts. Overall, these data include 3,314,993 student-year observations, 149,080 teacher-
year observations, and 224,747 employee-year observations.
47
For school year 2012-13, these
data include 517,180 unique students, 22,258 unique classroom teachers, 32,828 total employees,
and 692 schools.
Because several of the policy proposals at hand would require LAUSD to decrease class
sizes by hiring additional staff, and in some proposals to pre-recession levels, it is important to
correctly measure class sizes before the recession (2007-08), during the recessionary era (2008-
09 through 2011-12) and after the budget crisis eased (2012-13). The methods for estimating
average class sizes over time are described in Appendix B. Appendix Table B1 reports average
class sizes from 2007-08 to 2012-13 across various grades and schools.
Class sizes in LAUSD are negotiated as part of the collective bargaining agreement with
UTLA. Language within the union contract allows for class size increases during fiscal crises,

47
I include non-teaching staff in some projections of the cost of salary increases because, like many other districts,
LAUSD’s contract with UTLA requires that salary increases are applied to all certificated non-administrator
employees (what the district refers to as “me too” provisions).

116
which became necessary during the period of teacher layoffs. As Table B1 demonstrates, class
sizes increased in almost every grade level, school type, and department beginning in school year
2009-10. Average class sizes increased by 2.15 students across all K-12 grades in the district, by
3.15 in grades K-3, 1.78 in lower-performing schools, and by 1.80 students in PHBAO schools.
The relatively lower increase in average class size in lower-performing and PHBAO schools
resulted from greater access to categorical funding sources (also note that the district sets lower
class size norms in PHBAO schools; LAUSD, 2013a).
48

Defining the Final Spending Agreement and Alternate Spending Proposals
I describe here the policy variations simulated in this study, which I base on (a) polices
proposed by various LAUSD stakeholders during the budget negotiation period; (b) evidence of
effective spending policies identified in prior research; and (c) the final LAUSD spending
agreement reached in April of 2015. In addition to the final spending agreement, I simulate a
total of 16 alternative class size reduction policies, 16 alternative compensation policies, and
every possible combination of these two policies (resulting in a total of 256 “combination”
policies). For each simulated policy, I project the yearly costs and effects over a seven-year
period, beginning in the 2014-15 school year and extending to the 2020-21 school year.
49

Class size reduction policy simulations. I begin the analysis of class size reduction
policies by estimating the costs and effects of the district’s final agreement on class size
reduction. The agreement for class size reduction establishes mandated averages and caps for

48
In middle and high schools, the proportion of the school day in which teachers are teaching also contributes to the
magnitude of the impact of layoffs on class size increases. Throughout the period of layoffs, the proportion of the
day teachers spent teaching increased, which helped lower the impact of layoffs on class size increases. For
example, in 2008-09 the average high school teacher taught for 68.9% of the day and by 2012-13, the average high
school teacher taught for 74.9% of the day. Notably, even after several rounds of teacher layoffs, average class sizes
for grades K-3are still below the state-mandated average class size of 24 students.
49
In my primary analysis, I project costs and effects over seven years (school years 2014-15 to 2020-21) because the
California Legislative Analyst’s office projects that LCFF would be implemented over this time period (Taylor,
2013). I also rerun all analyses using a 30-year projection. All projections are based on data from the 2012-13 school
year and future costs are discounted using a 3% discount rate.

117
various schools and grade levels, as outlined in Article XVIII and Section 1.5 of the agreement
(see LAUSD, 2015d for a description of the full agreement).
50
The district class size plan is
further outlined in Appendix Table A1. As that table shows, maximum and average class sizes
vary by grades level, whether a school is PHBAO, and in some upper grades, whether the
particular course is considered academic or non-academic. The maximum average class size for
grades K-3 is set to 24 students in all schools. In elementary grades 4-6, maximum average class
sizes are 30.5 students in PHBAO schools and 36 students in non-PHBAO schools. For grades 6-
10, maximum average class sizes in all academic courses are set to 34 in PHBAO schools and
39.5 in non-PHBAO schools.
51
All non-academic courses in grades 6-10 and all courses in
grades 11-12 have a maximum average class size of 42.5 in all schools. Finally, the district
agreed to allocate additional resources to reduce class sizes in all grade 8 and 9 math and ELA
classrooms by two students.
Next I consider a set of 16 alternate strategies for reducing class size. Table 1 illustrates
the alternate permutations of class size reduction strategies. I estimate the impact and cost of
reducing class size to four possible levels, listed in Column 1 of Table 1. For example, the first
policy option – reducing class sizes to pre-layoff levels – requires that LAUSD generate class
sizes identical to the 2008-09 class sizes (shown in Table B1). The second policy option requires
the district to hire enough teachers to staff classes at levels recommended by an evidence-based

50
As in the prior union contract, the new agreement allows the district to increase class sizes in times of fiscal
emergency and the district maintains unilateral authority to declare such fiscal emergency (as outlined in Section 1.5
of the prior contract and current agreement). During the years of and immediately following the Great Recession, the
district used this authority, referred to as the “1.5 process” to increase class sizes to levels that fit the budgetary
needs of the district. However, the current agreement is unique in that a separate set of average and maximum class
sizes are established for fiscal emergencies. Because the district has already declared a fiscal emergency for the first
year in which the new agreement was active, 2015-16, the costs of the plan that I model in this study are based on
the class sizes agreed upon as part of Section 1.5 of the new contractual agreement.
51
The district defines academic courses as those in the departments of English, English as a Second Language,
Reading, Math, Social Studies, Science, and Foreign Language (LAUSD, 2015d).

118
funding model (Odden & Picus, 2014).
52
The third policy option requires the district to lower
class sizes by the maximum amount possible using the estimated $195 million in additional
funding that LAUSD will receive through Proposition 30 and the new Local Control Funding
Formula (LCFF) allocations.
The fourth policy option requires the district to rehire all previously laid off teachers with
above average “effectiveness,” as measured by a value-added model. Under current California
law, tenured teachers who are laid off are placed on a reemployment list for 39 months. The
district must rehire these teachers in order of seniority within teaching area before hiring other
teachers, except in cases where specialized skills are needed. Alternatively, the district could
rehire teachers according to test-based measures of effectiveness. Although currently not legal in
California, the decision in Vergara vs. California could ultimately overturn state statute requiring
seniority-based rehiring. That decision will undergo an appeals process that could take up to
several years from the time of this writing (California Teachers Association, n.d.). If the appeal
process is unsuccessful, California districts could rehire teachers based on measures other than
seniority. Rehiring teachers based on their past evaluations is currently legal in 35 other states
across the country. Thus, I estimate the costs associated with a policy in which only teachers on
the reemployment list that have measures of effectiveness above the district average are
rehired.
53
Methods for estimating teacher effectiveness using value-added modeling (VAM) are

52
Evidence-based funding models review extant literature to determine optimal class sizes for different grade levels.
The Odden-Picus funding model, used to help make resource allocation decisions in Arkansas (Odden, Picus, &
Goetz, 2006), Wyoming (Picus, Hayward & Ehlers, 2002; Seder, Picus, & Smith, 2002), Kentucky (Odden, Picus, &
Fermanich, 2003), and other states and districts, recommends class sizes of 15 in grades K-3 and 25 in grades 4-12
(Odden & Picus, 2013).
53
The policy of rehiring only teachers with above average “effectiveness,” as measured solely through value-added
measures does not necessarily reflect a viable policy option for LAUSD or other districts. This particular policy
simulation is meant to demonstrate the potential impact of rehiring teachers based on a multiple measure teacher
evaluation system.

119
described in Appendix C.
54

I include two specification checks of this fourth policy. The first extends the
reemployment list to any teacher laid off in the past four years (rather than just the past 39
months) and simulates district hiring of any previously laid off teacher with above average
effectiveness. Because districts can only generate VAMs for teachers in tested grades and
subjects (3rd–11th grade math and ELA teachers in California, with the 2nd grade CSTs serving
as pre-scores), hiring teachers back in order of VAM-estimated effectiveness necessarily limits
the reemployment pool to a limited set of teachers. Therefore, in the second specification check I
impute value-added measures for all teachers on the reemployment list. Measures of teacher
effectiveness are imputed based on teachers’ experience. As shown in Appendix Table C1,
teachers with greater levels of experience generally have higher VAM estimates, although the
returns to experience diminish over time.
For each of these four possible class size reduction policies, I estimate the impact and
cost of reducing class sizes: (a) for all grades (K-12), (b) for early elementary grades only (K-3),
(c) in lower-performing schools (bottom 20th percentile of API), and (d) for early elementary
grades in lower-performing schools. These policy permutations are shown in the second and
third columns of Table 1. I consider targeted reductions for just grades K-3 because the majority
of statewide initiatives to lower class sizes have focused on these grade levels (Education
Commission of the States, 2005; Chingos, 2013). I include policies that target funding increases
at lower-performing schools because a primary focus of the Local Control Funding Formula is to
target funding at previously underserved students and schools (Taylor, 2013). These policy
variations result in a total of 16 potential strategies for reducing class size.

54
While there is no consensus in the literature on the ideal value-added model (see, for example McCaffrey et al.,
2009, Kane and Staiger, 2008, and Rothstein, 2009), I use the methods commonly employed in past research. My
model is pooled across years and shrunken using Empirical Bayes methods.

120
Teacher compensation policy simulations. Simulations of teacher compensation
policies begin with the estimation of the costs and effects of the districts’ final agreement on
salary increases. The final agreement calls for an increase in salary of 10.37%, implemented
incrementally over time.
55
The first pay increase of 4% is retroactive effective July 1, 2014.
Because all of the policies simulated in this study are implemented for the 2013-14 school year, I
assume this first pay raise would begin during what is considered year 1 of all other policies,
school year 2014-15. The second pay raise of 2% was also retroactive at the time of the
agreement, and was effective January 1, 2015. Thus I assume that teachers receive this raise for
half of their annual salary during the 2014-15 school year, and receive the remaining amount in
the 2015-16 school year and every school year thereafter. The next salary increase of 2% was
scheduled for July 1, 2015, so I model this raise as beginning in the 2015-16 school year. The
final pay increase of 2% was set for January 1, 2016, so I again assume that teachers receive this
salary increase for half of their annual salary during the 2015-16 school year, and receive the
remaining amount in the 2016-17 school year and then every year after.
Next, I simulate a set of 16 alternate approaches to increasing teachers’ salaries. Table 2
outlines the different permutations of compensation policies, which vary by the extent of the
salary raise, the teachers targeted by the raises, and the kinds of schools targeted for raises. I
estimate the impact and cost of increasing salaries by 6%, 8.5%, 17.6%, and by the maximum
amount possible using the estimated $195 million in additional funding. The first alternate salary
increase of 6% was selected based on the district’s last compensation proposal, prior to reaching
the final agreement (LAUSD, 2015a). Under the district’s proposal, teachers would have

55
While the total pay increases sum to 10%, the parties agreed that each salary raise would compound over the
original amount (LAUSD, 2015d). For example, the second salary increase of 2% would be based on salaries that
had already been increased by 4%. Thus the 2% raise effective January 1, 2015 would effectively increase salaries
by 2.08%, compared to the salaries in place prior to the initial 4% raise. As a result, the total increase in pay from
the original amount would total 10.37%.

121
received a 4% raise, retroactive to July 1, 2014, and an additional 2% effective in the 2015-16
school year. The second two salary increase policies are based on UTLA’s original and most
recent proposals, which the union suggested could be implemented over two years (Blume, 2014;
UTLA, 2015). For each of these levels of teacher salary increases, I consider the impact and cost
of increasing salaries for (a) all teachers, (b) novice teachers only (teachers with 0-3 years of
experience), (c) all teachers in lower-performing schools, and (d) novice teachers in lower-
performing schools.
For targeted salary increases, I choose to focus on novice teachers because research
suggests that these teachers (a) are more likely than experienced teachers to leave their district
and the profession, (b) may be especially swayed to stay by increases in their compensation, and
(c) see greater returns to experience for each year they teach (Boyd, Lankford, Loeb, & Wyckoff,
2005; Dolton & van der Klaauw, 1999; Grissom & Strunk, 2012; Hannaway, Xu, Sass, Figlio, &
Feng, 2009; Hanushek, Kain & Rivkin, 2004).
56
I consider targeting compensation increases to
teachers in lower-performing schools because several recent district, state, and federal policies
have used compensation bonuses targeted to teachers in lower-performing schools, with varied
levels of success (e.g., Clotfelter et al., 2008; Glazerman, & Seifullah, 2012; Springer et al.,
2012; Yuan et al., 2013). Taken together, these policy variations result in 16 distinct strategies
for increasing teacher salaries (four amounts targeted to four different teacher populations), in
addition to the districts final agreement for raising salaries.


56
Research is less consistent with respect to the relationship between teacher experience and the effect of salaries on
retention. Clotfelter et al. (2008a) and Borman and Dowling (2008) find that salaries have larger effects on retention
for teachers with more experience. Hanushek et al. (2004) find that salaries have the largest effects on retention for
mid-career teachers, compared to those in their first three years and those with six or more years. Many other studies
of the effects of salaries on retention focus solely on new or beginning teachers (e.g., Imazeki, 2005; Murnane &
Olsen, 1989).

122
Combination policy simulations. During the negotiation process, most stakeholders
showed support for both salary increases and reductions in class size. I model these policy
proposals by combining each of the 16 class size reduction policies with each of the 16 salary
increase policies. For example, I consider the costs and effects of providing a 6% raise to all
teachers and all schools and restoring class size for all grades and all schools, for lower grades
and all schools, for just lower-performing schools, and for lower-grades in lower-performing
schools. I then repeat all of these variations in class size reduction when the 6% raise is targeted
only to novice teachers in all schools, all teachers in lower-performing schools, and novice
teachers in lower-performing schools. I also simulate policies that restore class sizes and then
raise salaries by the maximum amount using the remaining funding. Conversely, I include
combination policies that reduce class sizes by the greatest amount possible after a 6% raise for
all teachers, and a 6% raise targeted at each subgroup of teachers.
Because there are 16 variations of each policy, a total of 256 separate combination
proposals arise from these policy simulations. Because it is not possible to raise salaries by the
maximum amount possible using all additional funding and also lower class size by the
maximum amount using all new funding, I simulate policies that would evenly split the new
funding between various class size reduction and salary increase policies. The combination
policies are shown in Table 3.
Estimating the Cost of the Final Spending Agreement and Alternate Policy Proposals
This section reviews the methods I use to estimate the cost of resources required to
reduce class sizes and / or increase teachers’ salaries. Because these methods are fairly complex,
I present a brief overview of the methods and provide additional information in Appendix B.
Included in Appendix B are specific examples of how I estimate the cost of LAUSD’s final

123
spending agreement and several of the alternative spending proposals simulated in this study.
Cost of class size reduction. The per-pupil cost of reducing class sizes depends on a
number of factors, including the initial class size, the amount of class size reduction, the number
of teachers that would need to be rehired, and the experience and educational level of rehired
teachers. In the following three subsections, I first explain how I estimate the number of teachers
that need to be rehired for various class size reduction policies. I then show how I model the
district’s reemployment list and the characteristics of teachers that would likely apply for
LAUSD teaching positions.
Number of new teachers required to reduce class sizes by various amounts. Note that
class size reduction policies in this study are defined from three perspectives: (a) reducing class
size by hiring a specified number of teachers, (b) reducing class size to specified levels; and (c)
reducing class size by the maximum amount possible with a given amount of funding. Reducing
class size by only rehiring teachers with above average measures of effectiveness simply requires
hiring the 33 teachers on the reemployment list with above average VAM scores. Thus the cost
of this policy is the sum of these teachers’ most recent total salary and fringe benefits.
In contrast to the selective rehiring policies, estimating the cost of reducing class size to
specified levels, for instance, by restoring them to pre-layoff levels, or by reducing class sizes to
those recommended in an Evidence-Based model, requires calculating the actual number of
teachers that must be rehired. I do this by multiplying the current number of teachers by the ratio
of initial to final class sizes minus one. For example, suppose the districts wanted to restore class
sizes to pre-layoff levels in just grades K-3. Given that there were in 7,914 K-3 teachers
employed in LAUSD in 2012-13, it would take 1,315 new teachers to reduce K-3 class sizes

124
from their current (2012-13) level of 22.12 to their pre-recession (2008-09) level of 18.97.
57
For
policies that reduce class sizes in the upper grades, where classrooms are non-self-contained, I
make additional adjustments to account for teachers’ planning periods. These details are further
outlined in Appendix B.
The cost of hiring teachers on the reemployment list. Under current state policy, to
rehire teachers and reduce class size, the district must first hire appropriately credentialed
teachers on the reemployment list in the reverse order of which they were laid off (most senior
laid off teachers are rehired first). Teachers remain on the reemployment list for 39 months, after
which they have no reemployment rights. Therefore, all teachers who were laid off at the end of
the 2010-11 or 2011-12 school years and never rehired were on the reemployment list for the
2014-15 school year. Although the district laid off a substantial number of teachers, 51.9% of
those laid off were rehired to the district as classroom teachers the very next school year, and an
additional 28.3% were rehired as teachers in later years, or rehired immediately into non-
teaching positions. Only 1,050 of the 4,445 laid off teachers were never rehired (23.6%). Given
that no teacher layoffs took place during the 2012-13 and 2013-14 school years, the
reemployment list for 2014-15 includes only those teachers laid off in 2010-11 or 2011-12 and
never rehired. In total, there were 103 elementary teachers and 304 secondary teachers, for a total
of 444 teachers that fit this description and are therefore included on the 2014-15 reemployment
list. In sum, despite the fact that 4,445 teachers were laid off over four years, only about 10% of
those were on the reemployment list during the 2014-15 school year, since 76.4% had already
been rehired and the other 13.6% were never rehired, but were laid off over 39 month before the

57
That is, 7,914 * ([22.12 / 18.97] – 1) = 1,315. Formally, the number of teachers in year t required to return class
size to the year t-1 levels is given by the following equation: ([Class size
t
/ Class size
t-1
] -1 ) * Teachers
t
. I cannot
simply subtract the number of teachers in 2012-13 from the number of teachers in 2007-08 because enrollments in
LAUSD decreased over this time period. Therefore, simply hiring teachers back to pre-layoff levels would actually
reduce class sizes below their 2007-08 levels.

125
2014-15 rehiring period (and would therefore not be eligible for reemployment rights).
Because I observe laid off teachers in the years they were still employed with the district,
I have data on their total compensation at the time they were laid off. LAUSD teachers receive
both fixed benefits, which total $17,134 for each fulltime staff member, and variable benefits,
which total 12.71% of base salary.
58
For the most recent year of data available for this study,
2012-13, teachers with zero years of experience and Bachelor’s degree earned an average of
$43,430 base salary and $66,084 when including fringe benefits and pension contributions.
Those with 8 years of experience and a master’s degree earned $66,036 base salary and $91,564
total compensation. When teachers are hired (or rehired from the reemployment list), the district
is responsible for paying teachers’ base salary and both fixed and variable benefits.
Determining the cost for school year 2014-15 of rehiring all 444 teachers on the
reemployment list simply requires summing the total compensation of these teachers. The sum of
the total compensation for teachers on the rehire list is $32.6 million for the first year of
employment ($7.4 million for the 103 elementary teachers and $25.2 million for the 341
secondary teachers. On average these teachers earned a total compensation of $71,559 and
$73,969, respectively). Each year thereafter, newly hired teachers gain an additional year of
experience and (over time) additional education credits, which increases their total
compensation. In all class size reduction simulations, I account for these additional costs over
time, adjusting all future costs to present value costs using a discount rate of 3% (I describe the

58
Fixed benefits include health, dental, and life insurance, benefits for retirees, and other post-employment benefits.
Variable benefits include workers’ compensation, state unemployment insurance, and payments to the state teachers’
retirement system. Note that compared to the national average, teachers in LAUSD receive a larger proportion of
their compensation in fringe benefits (Allegretto, Corcoran & Mishel, 2004; Richwine & Biggs, 2011). In 2012-13,
LAUSD teachers received, on average, fringe benefits of 27.4% of their total compensation or 37.7% of their base
salary. Variable benefits include 2.95% for workers compensation, 0.06% for state unemployment insurance, 8.25%
for the state’s retirement system, and 1.45% for Medicare. Fixed benefits include $10,468 for medical, dental,
vision, and life insurance, $5,382 for benefits for retirees, and $1,284 for “other postemployment” benefits. These
figures were provided by the district and reflect the cost to the district, not the value of obtaining such benefits in the
private market.

126
cost methods in greater detail in Appendix B). Most class size reduction strategies I simulate
require hiring more than 444 teachers; therefore, some assumptions must be made regarding the
characteristics of teachers likely to apply for and be hired to teaching positions in LAUSD.
Modeling the rehiring of teachers on the reemployment list requires an additional and
more fundamental assumption. Once laid off, I assume these teachers have not found
employment elsewhere, or would be willing to leave their current positions to return to LAUSD.
Notably, the LAUSD Board of Education urged administrators during 2014 meetings to examine
the feasibility of rehiring previously laid off teachers. While I am unable to discern whether
previously laid off teachers would accept employment at LAUSD if offered a position, the
Board’s recommendation suggests that they believe many laid off teachers would be willing to
return to the district.
The cost of hiring additional teachers beyond the reemployment list. In my main
specification, I assume rehired teachers (beyond the initial 444 on the reemployment list) are, on
average, similar to the teachers hired during the pre-recession years (2007-08 and 2008-09).
59

These teachers had between zero and eight years of experience with a mean of 1.90 years.
Therefore, I assume newly hired teachers are drawn from a distribution of teachers with these
same characteristics. That is, each newly hired teacher beyond those on the rehire list is
randomly drawn from a distribution of teachers that mirrors those hired in the pre-recession
period, 2007-08 and 2008-09 (using a random number generator). To do this, I create a database
of 10,000 “new hire” teachers for which the proportion of each experience level is the same as
the group of teachers hired during the 2007-08 and 2008-09 school years. For example, 56.95%

59
I use the characteristics of teachers hired during the pre-recessionary period to model the characteristics of those
hired during the post-recessionary period because recent evidence suggests teachers hired during recessionary
periods are more experienced and more effective on average than teachers hired during times of macroeconomic
expansion (Nagler, Piopiunik, & West, 2015). Moreover, LAUSD had severely limited the number of new hires
during the recessionary period.

127
of teacher hired during the pre-recession period had zero years of teaching experience, while
6.18% had five years of experience. Therefore, the new hire database contains 5,695 teachers
with zero years of experience and 618 with five years of experience. I then use a random number
generator to place these teachers in the order in which they will be hired. Teachers in the “new
hire” database are hired in the same order regardless of the particular class size reduction policy.
Finally, to gauge the extent to which the characteristics of newly hired teachers determine the
cost of class size reduction, I re-run all simulations with upper and lower bounds in which newly
hired teachers earn the average salary of teachers with (a) a bachelor’s degree and zero years of
experience; and (b) a master’s degree and eight years of experience. The first year cost of hiring
teachers from the “new hire” database is the sum of the total compensation of all teachers
hired.
60

Cost of teachers’ salary increases. Estimating the cost of increasing salaries is more
straightforward than estimating the cost of class size reduction. Compensation policies are
defined from two different perspectives: increasing salaries by a specified level and increasing
salaries by the maximum level possible using a specified amount of funding.
The estimated cost of increasing salaries by a specified level is calculated by multiplying
the total base salaries and variable benefits by the amount of salary increase. For example, the
sum of base salaries and variable benefits for all 22,258 classroom teachers in 2012-13 is $1.7
billion. Thus the cost of the union’s proposed salary increase of 8.5%, implemented over two

60
As I describe in more detail in Appendix B, all newly hired teachers gain an additional year of experience each
year after their initial hiring date. I account for these future costs and project all policy simulations over a seven-year
period (the timespan over which LCFF will be fully implemented). I exclude the cost of interviews, recruitment,
screening, and induction mentoring and professional development, which will underestimate the costs of class size
reduction. Milanowski and Odden (2008), for example, estimate that the central office and school staff time required
for each new teacher hire costs approximately $1,136 and $2,588 respectively (Table 3), and a total of $22,434 when
including induction mentoring and ongoing professional development over the first five years of employment. Thus
my cost estimates of large scale class size reduction may be underestimated when considering additional costs
related to newly hired teachers.

128
years, is about $123.6 million, or $239 per student per year when projecting over seven years and
discounting all future costs to present value.
61

A number of the policy simulations also require that I measure the total amount salaries
can be increased for a giving spending amount. For example, I consider the cost and effects of
allocating all additional funding the district will receive through Proposition 30 and LCFF
allocations to increase teachers’ salaries and in combination policies, the impacts of raising
salaries by the maximum amount possible after implementing various class size reduction
policies. To measure the extent to which salaries can be increased for a given spending level, I
divide the total amount allocated for salary increases by the sum of the salaries and variable
benefits. For example, allocating all $195 million to increase salaries for all teachers would raise
salaries by approximately 11.3% ($195 million / $1.7 billion), at a cost of $346 per pupil. If the
$195 million is allocated to just novice teachers or low-performing schools, the total cost
remains constant, but the percent increase in salaries and the per-pupil cost are both higher. For
each of these salary increases, I assume no transaction costs, such as administrative personnel
time or legal fees.
I estimate the cost of combination policies by summing the total and per-pupil costs of
each individual policy. For example, the cost to restore class size to pre-layoff levels in lower
grades in all schools is $98.9 million and the cost of providing an 8.5% raise for novice teachers
in lower-performing schools is $1.4 million, thus the total cost of this policy is about $100.3
million. Because each policy affects a different number of students, the yearly per-pupil cost is

61
LAUSD administrators have noted in personal conversation that most district labor contracts contain provisions
that require any salary increases for one labor unit to apply to all labor units (what the district calls “me too”
provisions). Administrators in the budget and finance department use a general rule of thumb that a 1% increase in
salaries for all labor units in LAUSD costs about $40 million. Thus a 4.25% raise would cost $170 million in the
first year of implementation. In my baseline analyses, I consider salary increases for only classroom teaching staff
and then assess how the results would change under LAUSD’s calculation of the cost of salary increases.

129
calculated by summing the yearly per-pupil cost of each individual policy. Thus I assume that
from a cost perspective, the district does not necessarily save money by implementing policies
together, as opposed to individually.
Estimating the Effects of the Final Spending Agreement and Alternate Policy Proposals
As is common in educational cost-effectiveness studies, I use effect sizes on student
achievement estimated in the available research literature to project policy effects (Levin &
McEwan, 2000; Levin & Belfeild, 2015).
62
Effect sizes allow for comparisons of achievement
gains on different types of standardized tests at different grade levels (Cohen, 1994). Because
each policy alters the districtwide composition of the teaching workforce, I also estimate the
impact of each policy on the distribution of teacher effectiveness across the district. While this
approach is far from perfect, several recent studies have simulated similar policies and projected
effects by calculating the change in the distribution of teacher effectiveness (Boyd, Lankford,
Loeb & Wyckoff, 2010; Goldhaber & Hansen, 2010; Rothstein, 2012; Winters & Cowen, 2013).
Due to the complexity involved in projecting policy effects, I provide a brief overview here and
include greater detail and specific examples in Appendix B.
As I discuss further in the limitations section below, focusing solely on student
achievement necessarily limits consideration of other outcomes. Excluding students’ non-
cognitive outcomes, for example, may bias estimates of cost-effectiveness toward zero. By
including only measures of student achievement, the assumption is that districts are primarily
concerned with improving students’ cognitive development, as measured by standardized exams,
which is at least weakly correlated with other outcomes including civic engagement, student
health, or graduation rates (Dee & West, 2011; Heckman & Rubenstein, 2001).

62
Effect sizes are measures of student achievement gains associated with an educational intervention that are
calculated by dividing the difference in mean achievement between a group of students that receive an intervention
and a group of control students by the standard deviation of achievement in both groups (Cohen, 1994).

130

The Effect of Class Size Reduction
A great deal of research examines the effect of class size reduction on student
achievement (e.g., Aos & Pennucci, 2013; Aos, Miller, & Mayfield, 2007; Angrist & Lavy,
1999; Glass & Smith, 1979; Hoxby, 2000; Nye, Hedges & Konstantopoulos, 1999; 2000; 2002).
Studies generally show that while students perform better in smaller class sizes, particularly
students of color and those from lower-income families, large scale class size reductions can
lower the quality of the overall teaching workforce in a school district (Jepsen & Rivkin, 2009;
Krueger, 1999). I describe below how I estimate the main effects of reducing class sizes, the
impact of changes to the composition of the teaching workforce, and the policy effects of
alternate hiring strategies.
Main effects of class size reduction. To estimate the average effect of a one-student
decrease in class size, I use the average effect size found in a limited set of peer-reviewed
studies. I use the following criteria to narrow the number of included effect size estimates. First,
I consider only studies that drew on experimental or quasi-experimental designs and were
published within the past 20 years. These criteria reduce the total number of potential studies to
37. I then exclude studies that did not examine student achievement as the outcome measure (or
did not report findings in a way that could be translated to achievement effect sizes), were not
published in peer-reviewed journals, or were analyses of data from past studies. This reduces the
total number of studies to eight, within which there are 25 estimates of class size reduction
effects across kindergarten through eighth grade. Following Aos and Pennucci’s (2013)
approach, I use effects for grades four through eight to estimate the effect of class size reduction

131
in grades 9-12.
63

To project the average effect of class size reduction in LAUSD, I use the average of the
effect sizes found across these 25 estimates (see Table A3 in the Appendix for effect sizes,
standard errors, and citations for each individual study). Given past research on the
heterogeneous effects of class size reduction, I assume larger effects for lower-income students
and students of color (and provide greater detail on these methods in Appendix B). These
estimated effects assume that there are no major shocks to the local labor market that would alter
the quality of the district’s teaching workforce, a topic to which I turn next.
Modeling changes in teacher effectiveness. One of the key determinants of the impact
of class size reduction is the effectiveness of newly hired teachers (Chingos, 2013; Jepsen &
Rivkin, 2009; Stecher & Sims, 2002). To address the concern of shocks to the teaching
workforce, I estimate for each class size reduction policy the decline in overall teacher
effectiveness that results from hiring a large number of inexperienced teachers. To do this, I first
estimate value-added measures of effectiveness for all teachers in tested grades and subjects
employed during the window of observation (2007-08 to 2012-13). For teachers that are not in
tested grades and subjects and for teachers on the “new hire” database, I impute measures of
effectiveness based on years of experience and educational degree attainment.
64
The change in
the overall district average teacher effectiveness in the first year of class size reduction is the
difference between the average teacher effectiveness before the reduction in class size and the
average effectiveness of teachers after hiring new teachers. Thus the effect of a class size

63
No studies that I reviewed that met the criteria for inclusion had estimates of the effects of class size reduction on
grades 9-12; therefore, I assume that the average effects for grades 4-8 will be the same as those fore grades 9-12.
As I note in the text, I test a number of specifications of class size effects to gauge the sensitivity of my overall
results to the estimated effects.
64
Appendix C outlines the method I used to generate teachers’ value-added measures of effectiveness (see Table C1
for average predicted value-added measures across teacher experience)

132
reduction on the composition of the teaching workforce depends on which grades or schools are
targeted, the size of the class size reduction (i.e., the number of teachers rehired), and the
particular teachers who are rehired. In Appendix B, I further explain how I make these
calculations and provide specific examples.
Rehiring policies. One of the class size reduction policies I consider requires that the
district rehire only teachers on the reemployment list in the top half of value-added measures of
effectiveness. Of the 444 teachers on the reemployment list, 33 teachers have value-added
measures above the district mean. These teachers have an average value-added measure of 0.149
SD and their total value-added is equal to 4.91 SD. If this group of teachers were rehired to the
district, average effectiveness across the district would increase by .0002 standard deviations
(4.91 / [33 + 22,258], given that there were 22,258 teacher employed in 2012-13). Hiring 33
teachers would also lower class sizes by 0.03 students districtwide, which would raise student
achievement by 0.0002 SD. Thus the total yearly effect on achievement of class size reduction
under selective rehiring is 0.0004 SD.
The Effect of Teacher Salary Increases
In contrast to research on class size reduction, which has primarily examined the impact
of smaller class sizes on student achievement, past studies of the effects of teacher salaries tend
to focus on teacher retention as the outcome (e.g., Hanushek, Kain, & Rivkin, 2004; Imazeki,
2005; Ingersoll, 2001; Murnane & Olsen, 1989). More recent studies assess the effect of salary
bonuses on retention for teachers that receive positive evaluations or that agree to work in
harder-to-staff schools (Clotfelter et al., 2008; Glazerman, & Seifullah, 2012; Glazerman,
Chiang, Wellington, Constantine, & Player, 2011; Goodman & Turner, 2013; Springer et al.,
2012; Yuan et al., 2013). The extant literature demonstrates that in general, monetary bonuses do

133
not significantly improve teacher performance, but they may induce teachers to remain in or
transfer to specific types of schools. Additionally, higher salaries attract more qualified and,
presumably, more effective teachers (e.g., Andrews, Fayissa, & Tate, 1991; Card & Krueger,
1992; Dolton & Marcenaro-Gutierrez, 2011; Figlio, 2000; 2002; Hanushek, Kain, & Rivkin,
1999; Heckman, Layne-Farrar, & Todd, 1995; Hoxby, 1996; Ladd, 2007; Loeb & Page, 2000).
In sum, higher teacher salaries contribute to student achievement both by reducing harmful and
disruptive teacher turnover and by attracting better-qualified applicants. I model each of these
effects to project the outcomes of raising teacher salaries, while noting that research on the
impact of salaries on student achievement is less developed than that of class size reduction.
Salary effects of decreased teacher turnover. The first mechanism through which
salaries may improve student outcomes is through decreases in disruptive teacher turnover
(Ingersoll, 2001; Guin, 2004; Ronfeldt, Loeb, & Wyckoff, 2013), defined as teacher attrition that
harms school culture and student learning. To project the impact of salaries on student
achievement, I use a two-step process. I begin with findings reported in Ronfeldt, Loeb, and
Wyckoff (2013), which show that achievement on standardized tests is about 0.073 SD lower in
the years with 100% grade-level teacher turnover, compared to years with zero turnover.
65
I then
combine this parameter with estimates of wage elasticities – the effect of a 1% increase on
salaries on the percent decrease in teacher turnover. I use as my primary estimate the wage
elasiticies found in Imazeki (2005), which indicated that a 1% increase in district-wide teacher
salary was associated with a 0.926% decrease in the likelihood a female teacher leaves the
district, and a 1.804% decrease for male teachers.
66
Using a weighted average of the percent of

65
This figure is the average effect of teacher turnover on math and English scores (shown in Model 2 of Table 4).
66
This estimate is taken from Table 4 of Imazeki (2005). Using a weighted average for female and male teachers in
LAUSD (69.6 and 30.4% in the 2012-13 school year, respectively), a one percent increase in wage is associated
with a 1.193% decrease in the likelihood a teacher leaves the district.

134
male and female teachers in LAUSD, this wage elasticity is –1.193. Scaling this parameter down,
a decrease in turnover of 1.193% would raise average achievement by 0.00087 SD. Therefore, a
6% increase in salary would raise achievement by 0.0052 standard deviations and a 17.6%
increase would raise achievement by 0.0154 SD. I use these figures in my primary analysis.
I use several alternate estimates of wage elasticities from other studies as specification
checks of my primary estimate. These include estimates taken from Clotfelter et al. (2008a)
Hanushek, Kain, and Rivkin (2004), and Greenwald, Hedges, and Laine (1996). These alternate
estimates are used to test the sensitivity of compensation policy effect projections (and explained
in greater detail in Appendix B). All sensitivity analyses are available from the author upon
request.
Salary effects of improved quality of teacher applicants. The second mechanism
through which salary increases may improve educational outcomes is by attracting a more
qualified teaching workforce into the profession. Past studies find that a 1% increase in salaries
increases the probability of hiring a teacher from a selective undergraduate institution by 0.75%
to 1.58% (Figlio, 1997) and a teacher that majored in her or his teaching area by 1.13% (Figlio,
2002). Studies also find that higher salaries attract teachers with higher test scores: a 1% increase
in salaries increases average SAT scores by 0.27% (Figlio, 2002) and scores on the licensure
exams by 1.2% (Ferguson, 1991). Importantly, studies have linked these teacher characteristics
to impacts on student achievement (Ehrenberg & Brewer, 1994; Goldhaber, Gross, & Player,
2013). For my preferred estimate, I assume a 10% increase in salaries would raise teacher
effectiveness by use 0.0193 SD, which is based on estimates from Figlio (2002).
67
So a 6% raise

67
Although past studies have linked these teacher characteristics to positive student outcomes (e.g., Ehrenberg &
Brewer, 1994; Goldhaber, Gross, & Player, 2013), no studies link salaries directly to measures of teaching
effectiveness. I therefore use a range of estimates of the impact of salaries on the quality of new hires. In various
model specifications, I assume that each 10% increase in salaries will improve the effectiveness of newly hired

135
in salaries increases the quality of new hires by 0.0116 SD while a 17.6% salary raise would
increase the new hire effectiveness by 0.0339 standard deviations.
For policies that raise teachers’ salaries only (without reducing class size), new teachers
are hired only to replace the natural teacher attrition that takes place over time through career
transitions and retirements. I base the natural rates of teacher attrition over time on the attrition
rates of teachers during the pre-recessionary period. Early career teachers and those approaching
retirement were more likely to leave the district, while mid-career teachers had lower rates of
attrition. On average, 7.5% of teachers exited the district or teaching positions each year during
the pre-recession period.
68
As before, I assume that while individual teachers gain an additional
year of experience each year, the districtwide average experience level and effectiveness is
constant over time in the absence of any policy intervention. My workforce projections suggest
the district will need to hire 1,667 new teachers per year to replace the 7.5% that leave (0.075 *
22,258), or about 13,332 over seven years, to maintain 2012-13 staffing ratios (assuming
constant enrollment).
For combination policies that increase salaries and reduce class sizes, the positive effects
associated with more highly qualified teacher applicants is magnified by the large number of
newly hired teachers. For example, as I highlighted above, compensation policies only affect the
quality of new hires. Based on pre-recession attrition and retirement rates, about 7.5% of all
teachers are replaced each year through natural turnover. However, if the district provides both a
salary increase and hires additional teachers through class size reduction, the district will attract
more highly qualified teachers than it otherwise would in the absence of a salary increase. In this

teachers by between 0% and 2% of a student-level standard deviation (i.e., between 0 and 0.02 SD), for a given level
of teacher experience and educational attainment.
68
Note that school-level attrition during the pre-recession period was 11.5%; however, much of this turnover was
across schools within the district and therefore would not by itself require the hiring of a new teacher.

136
way, salary increases and class size reduction have synergistic effects on student achievement.
Findings
I present findings from simulations in three sections. First, I describe the effectiveness-
cost ratios for the baseline policies—those that require the district to either reduce class size or
increase teachers’ salaries (but not both). These results are shown in Table 4. Second, I review
the findings on combination polices in which the district reduces class sizes and increases
teachers’ salaries. Table 5 shows selected results for combination policies. Finally, I compare
cost-effectiveness of various policy proposals to the final LAUSD spending agreement.
Baseline policies
Table 4 shows effectiveness-cost ratios for all 16 teacher salary increase policies and all
16 class size reduction policies, ranked in order of cost-effectiveness. For ease of interpretability,
I report findings in effectiveness-cost ratios, which show the expected gain in standard deviations
of student achievement for each $1,000 per pupil per year (2015 dollars). Thus larger
effectiveness-cost ratios imply that the policy is more cost-effective. The relative cost-
effectiveness of each baseline policy can be compared both across the two general policies—
comparing class size reduction to teachers’ salary increases—and within each policy (comparing
across-the-board spending policies to targeted spending policies). I discuss each of these
comparisons in the following two subsections.
Comparing class size reduction to teachers’ salary increases. I find that class size
reduction is more cost-effective than raising teachers’ salaries in many, but not all of the
permutations of each separate policy. As shown in Table 4, when ranked in order of cost-
effectiveness, the first nine policies are all class size reduction policies. The most cost-effective
policy is rehiring teachers based on measures of prior effectiveness in just lower-performing

137
schools in grades K-3. The second most cost-effective proposal is the same policy, except that
teachers would be rehired to teach in lower grades across all schools. Note however, that both of
these policies would involve hiring 21 teachers at a cost just over $1.5 million. When I expand
the reemployment list to all teachers ever laid off and not rehired, and impute value-added
measures based on teachers’ observable characteristics, the total number of teachers that could be
rehired under this policy never reaches beyond 72 teachers, and the resulting class size reduction
is typically less than half a student. Because the number of teachers with prior measures of
effectiveness is severely limited, it is not possible to implement selective rehiring policies on a
wide scale.
The third most cost-effective policy in Table 4 requires the district to restore class sizes
to pre-layoff levels in K-3 classrooms in lower-performing schools. This policy would cost
approximately $20 million per year or about $533 per student per year and would result in a 2.99
student reduction in class sizes for the schools and grades impacted. By implementing this
policy, the district would raise student achievement for students in lower-performing schools in
grades K-3 by an estimated 0.05 SD, providing approximately one and a half months of
additional learning (Lee & Finn, 2010). The effectiveness-cost ratio for this policy suggests an
increase in student achievement of 0.09 SD for each $1,000 per pupil, which is less than the
selective rehiring policies, but more than any other policy shown in Table 4.
These baseline results rely on a number of assumptions, many of which can be tested
using sensitivity analysis and specification checks. For instance, I assumed that when the district
hired additional teachers to reduce class size, newly hired teachers would be drawn from a
random distribution of teachers with similar characteristics as those hired during the pre-layoff
period (2007-08 and 2008-09). My general findings do not change substantially if I assume all

138
newly hired teachers earn average compensation equivalent to teachers with zero years of
experience and a bachelor’s degree and eight years of experiences and a master’s degree. The
per-pupil yearly cost of across-the-board class size restoration increases by 17.75% when I
assume newly hired teachers hold a master’s degree and have eight years of experience, whereas
the cost decreases by 7.29% when I assume that teachers would hold a bachelor’s degree and
have zero years of experience at the time they are hired. Given that the most cost-effective class
size reduction place is 35% more cost-effective than the most cost-effective salary increase plan,
changing the assumed salary of newly hired teachers does not significantly alter the relative cost-
effectiveness of class size reduction and teacher compensation policies.
The results are also generally consistent when I substitute the projected salary effects
found in Imazeki (2005) with estimates found in Hanushek, Kain, and Rivkin (2004) and
Greenwald, Hedges and Laine (1996). Considering the vastly different contexts and
methodologies used in these studies, the proximity of their findings provides further evidence
that the effects of salary increases were accurately estimated for this study. However, when I use
the estimates of teacher salary effects found in Clotfelter et al. (2008a), I find the opposite trend:
increasing teachers’ salaries is consistently more cost-effective than class size reduction. In
discussing their findings, Clotfelter et al. (2008a) highlight a number of reasons why their study
represents an outlier within research on the effects of salaries on teacher attrition. First, both the
Hanushek, Kain, and Rivkin (2004) and the Imazeki (2005) studies draw on cross-district
variation in teachers’ salaries. That is, the authors compare outcomes based on differences in
teacher salaries across districts.
69
In contrast, Clotfelter et al.’s estimates of salary effects are

69
The longitudinal data used in Hanushek, Kain and Rivkin (2004) allow the authors to observe changes in district
salaries over time. The authors are therefore able to use district fixed effects to estimate the effect of changes in
teachers’ salaries within the same district over time. However, these models result in lower estimates of salary
effects than those presented in their main findings.

139
based on bonuses provided to teachers within districts. The larger effects found in Clotfelter et al.
may therefore be more applicable to targeted salary increases in LAUSD as compared to across-
the-board salary increases. Second, the teachers in the Clotfelter et al. study could only receive
salary bonuses if they taught high need subjects in lower-performing schools. These teachers
may not be representative of all teachers within a district. While I am unable to estimate the
effects of teachers’ salary increases with absolute certainty, three of the four studies I include
yield similar estimates, providing some indication that the primary estimate used (Imazeki, 2005)
is appropriate.
One way to address uncertainty in parameter estimation is to conduct a break-even
analysis (Boardman, Greenberg, Vining, & Weimer, 2011). In general, break-even analyses
show the values of an unknown parameter that correspond to a positive return on investment for
a particular policy intervention. In the case of a cost-effectiveness study, a break-even analysis
shows the values of an unknown parameter for which one policy becomes more cost-effective
than another policy. Figure 1 displays a break-even analysis comparing the cost-effectiveness of
restoring K-3 class sizes to pre-layoff levels, to a 6% salary increase for all teachers. These two
policies are chosen because they are relatively comparable in terms of total and average cost per
pupil and because both have received support from the LAUSD teachers’ union, administration,
and school board.
70
The y-axis in Figure 1 is the difference in cost-effectiveness between
restoring K-3 class sizes to pre-layoff level (i.e., lowering average class size in the lower grades
by 3.15 students) and providing all teachers with a 6% raise. Positive y-values indicate that class
size reduction is more cost-effective, while negative y-values indicate that salary increases are
more cost-effective. The x-axis plots various levels of effect sizes on student achievement for a

70
Not all policies considered in my analysis received support from all three of these stakeholders. For instance, the
teachers’ union did not publicly support targeting funding to low-performing schools and LAUSD administrators
argued that a 17.6% salary increase would have been cost prohibitive (LAUSD, 2013).

140
6% increase in teachers’ salaries. The dashed lines indicate the effect sizes of salary increases
found in each of the four studies mentioned above. As is clear, three of the four estimates of
salary increase effects suggest that class size reduction is more cost-effective than raising
teachers’ salaries. If the effect of a 6% teacher pay increase is above approximately 0.012
standard deviations, then raising teachers’ salaries will be more cost-effective than reducing class
size.
Comparing targeted spending to across-the-board policies. One of the important
lessons that emerges from the comparisons shown in Table 4 is that some class size reduction
policies are more cost effective than salary increases, while others are less cost-effective. This is
because both class size reduction and salary increases are more cost-effective when targeted at
specific teachers or students, compared to interventions implemented across all schools and
teachers in the district. For example, class size reduction in the lower grades is found to be more
cost-effective than class size reduction in all grades. Table 4 shows that the effectiveness-cost
ratio for restoring class size to pre-layoff levels in just the lower grades is 0.070 SD/$1,000,
whereas the effectiveness-cost ratio for restoring all class sizes is 0.039 SD/$1,000. These figures
translate to approximately 39.5 and 22.0 days of learning for each $1,000 per pupil, respectively.
Thus for the same level of spending per pupil per year, class size reduction has a larger effect in
lower grades. This finding is driven by the fact that the cost of reducing class size is only
moderately higher for lower grades than for upper grades, while class size reduction has a
substantially larger effect in grades K-3 (in the previous section, I discuss in greater detail why
the costs of class reduction varies by grade). Class size reduction is also generally more cost-
effective when targeted to lower-performing schools.
71


71
This trend holds for all comparisons except for one. Restoring class size to pre-layoff levels across all grades and
all schools is actually more cost-effective than restoring class sizes across all grades in just lower-performing

141
Table 4 also indicates that teacher compensation policies are more cost-effective when
targeted at novice teachers, compared to across-the-board salary increases. As shown in Table 4,
providing a 6% raise to all teachers in the district results in an effectiveness-cost ratio of 0.054
SD/$1,000, whereas targeting those salary increases to just novice teachers raises the
effectiveness-cost ratio to 0.061.
Combination policies and the LAUSD Spending Plan
In this section I discus policies that combine class size reduction with salary increase
policies. I then compare how each of the simulated proposals compare with the final spending
agreement that LAUSD reached in April 2015.
Cost-effectiveness of combination policies. I estimate the costs and effects of a total of
256 policies that combine teacher salary increases with class size reduction. In order to simply
the presentation of results, I selected three combination policies that exemplify the benefits of
combining class size reduction and salary increase policies. These policies are shown in Table 5.
The first policy requires the district to restore class sizes to pre-layoff levels in lower grades of
lower-performing schools and simultaneously provide a 6% raise for novice teachers in all
schools in the district. I chose these policies because of the viable policy options, each one
represents the most cost-effective policy within its broad category. In order to restore K-3 class
sizes in lower-performing schools, the district would need to hire 270 elementary teachers of
which 103 would be pulled from the reemployment list and the other 167 would come from the
new hire database. Because salary increases improve the quality of newly hired teachers, the 167
teachers who are pulled from the new hire list are more effective than if these teachers were

schools. This is due to the patterns of class size increases took place during the recession—while all schools saw
greater increases in class sizes in grades 6-12, lower-performing schools experienced a relatively larger proportion
of class size increases in grades 4-6. Thus restoring class sizes in just lower-performing schools would require
greater decreases in class sizes in grades 4-6 (where class sizes are already lower than in upper grades, which
contributes to a higher per-pupil cost for each one-student reduction).

142
hired as part of a class size reduction policy that did not include a salary increase. In this way,
class size reduction is more effective when coupled with an increase in salaries.
In the second row of Table 5, I consider a policy that combines a larger class size
reduction – restoring K-3 class sizes in all grades – and a larger salary increase of 8.5%. Finally,
in the third policy, I show the cost-effectiveness of a proposal that would provide all teachers
with a salary increase of 8.5% and use all remaining funding to reduce class sizes in the lower
grades of lower-performing schools.
The first combination policy is the most cost-effective of the three because it provides a
modest targeted class size reduction (a reduction of 2.99 students), while offering higher salaries
to all newly hired teachers (thereby increasing their average quality). The effectiveness cost-ratio
of 0.086 SD/$1,000 suggests that for each $1,000 per pupil per year, achievement would increase
by 0.086 SD per student or about two and a half months of additional learning (Lee & Finn,
2010). The second combination policy is slightly less cost-effective because the class size
reduction is not targeted to lower-performing schools. In addition, because the salary increase is
targeted to lower-performing schools, only the new hires in such schools have greater
effectiveness. Thus the second combination policy does not fully benefit from the synergistic
effect of lower class sizes and raising salaries. Finally, the third policy is the least cost-effective
of the three policies shown in Table 5 because it requires such a substantial reduction in class
size. Specifically, the district would have $91.8 million remaining in additional funding after
providing all teachers with a salary increase of 6%, which, under this combination policy, it
would use to lower class sizes in the lower-grades of lower-performing schools by 6.1 students.
Such a substantial reduction would lower class sizes to an average from 22.0 to 15.8 in these
grades and schools. As a result, this combination policy is not as cost-effective as other

143
combination policies that include more modest reductions in class size.
Differences in the cost-effectiveness across combination policies arise because a) each
requires a different balance of resource use between class size reductions and salary increase; b)
class size reduction becomes more costly (and less cost-effective) as the resulting class size
decreases; and c) the negative impact of hiring a large number of novice (and less effective)
teachers to reduce class size changes depending on the size of the salary increase (because salary
increase attract higher quality teachers). In short, policy makers who want to maximize the cost-
effectiveness of combination policies should allocate more resources to salary increases if class
size reductions are implemented across-the-board. However, combination policies should
allocate more resources to reducing class size if reductions are targeted to lower-performing
schools or lower grades (or both). Finally, because class size reduction becomes more costly and
less cost-effective as the resulting class size decreases, substantial reductions in class size such as
those required in the EB model are less cost-effective than more modest reductions, even when
class size reduction are targeted.
Comparing the LAUSD spending plan to simulated policies. I find that the
effectiveness-cost ratio for the LAUSD spending plan is 0.073 SD / $1,000. The LAUSD
spending plan compares favorably to most of the baseline policies shown in Table 4 (the
district’s plan would rank 7th place out of 32 policies). This happens for two reasons. First, the
district’s class size reduction plan targets the schools with the highest proportions of students of
color (PHBAO schools). Because effects of class size reduction are larger for disadvantaged
students, this targeting of resource allocation reduces cost and improves cost-effectiveness.
Second, the salary is increased over time, which significantly reduces the cost of the salary
increase relative to the benefit. In each of the salary increase policies, I assume that benefits

144
accrue immediately because research shows teachers take into account both current and future
earnings potential when deciding where and for how long to teacher (Imazeki, 2005). As a result,
the salary increase policy that LAUSD agreed on is more cost-effective than any other salary
increase policy estimated in this study. According to the results of my simulations, had the
district targeted that salary increase to novice teachers, the policy would be even more cost-
effective.
Limitations of Cost-Effectiveness Analysis
Two limitations to the current study that are common to most cost-effectiveness analyses
warrant discussion. First, the literature demonstrates that both policies examined in this study,
class size reduction and teachers’ salary increases, have other positive outcomes beyond
improving student achievement (e.g., Dee & West, 2011; Chetty et al., 2011; Loeb & Page,
2000). The desired outcomes of an educational system change over time and vary for different
stakeholders (Labaree, 1997). Focusing on short-term student achievement gains necessarily
limits consideration of loftier goals such as social cohesion, capacity to participate in a
democratic society, or opportunities for social mobility. Scholars have found increases in
students’ non-cognitive outcomes associated with class size reduction (Dee & West, 2011) and
higher teachers’ salaries over a longer time period have been shown to raise graduation rates and
students’ labor market earnings (Card & Krueger, 1992; Heckman, Layne-Farrar & Todd, 1995;
Loeb & Page, 2000). Incorporating multiple outcomes into a cost-effectiveness analysis is
possible; however, each outcome must be assigned a specific weight and it is unclear how
stakeholders, including students, parents, educators, and society at large, might weight various
outcomes (Levin, 2002).

145
A more fundamental problem with incorporating additional student outcomes is the lack
of comparable studies for each policy. For example, studies might define high school graduation
as receiving any type of diploma including a General Educational Development (GED)
certificate, or high school graduates may be defined as having graduated through the traditional
route (Levin et al., 2011). Such differences can potentially bias comparisons of outcomes across
studies (Levin & McEwan, 2000). Effect sizes of student achievement partly address this
problem by comparing the relative achievement on standardized exams of students that receive
an intervention to similar students that do not receive an intervention (Cohen, 1994; Lee & Finn,
2010). Additionally, short-term student achievement may serve as a weak proxy for many other
desirable outcomes (Harris, 2009; Yeh, 2010). A particular policy that is relatively more cost-
effective at raising student achievement may also be more cost-effective at achieving other
intended outcomes (Harris, 2009).
72
At the very least, cost-effectiveness studies, including the
current study, can be viewed as one data point that local school decision-makers can take into
account when determining the best uses of funding.
A second limitation of cost-effectiveness studies, including the current study, is that not
all potential uses of funds can be assessed. Despite the vast array of policies implemented around
the country every year, very few undergo rigorous evaluation that provides causal estimates of
the impact on student outcomes (Mervis, 2004). Without a valid estimate of the effects of an
intervention or policy, based on at least one experimental or quasi-experimental study,
calculating cost-effectiveness is not possible. For example, various LAUSD stakeholders have
maintained that rehiring health and human service employees that were laid off in previous years
is essential for ensuring a safe and healthy schooling environment (LAUSD Board of Education,

72
This assumption may be less tenable depending on the policy interventions under consideration. For instance,
comparing the cost-effectiveness of art enrichment programs to after school tutoring may not be appropriate if
outcomes include only measures of student achievement (Levin & McEwan, 2000).

146
2014b; UTLA, 2014b). However, the available literature does not provide a strong evidence base
for this intervention. This does not mean that placing additional nurses in schools is a bad policy;
rather, there is only limited evidence quantifying the individual impact of nurses, counselors, and
other non-teaching school staff members (Lightfoot & Bines, 2000; Perna et al., 2008; Puskar &
Marie Bernardo, 2007; Reback, 2010a; 2010b). Similarly, as part of the implementation of
Common Core State Standards, the district approved the hiring of about 160 instructional
coaches to provide intensive training for teachers (LA Daily News, 2013). While there is
growing empirical evidence of the impact of intensive professional development on student
achievement, including instructional coaching, past studies have examined specific coaching
programs such as the Literacy Collaborative and the “Just Read, Florida!” literacy coaching
project (Biancarosa, Bryk & Dexter, 2010; Marsh, McCombs & Martorell, 2009). Determining
the potential impact and per-pupil cost of hiring instructional coaches requires some information
about what model of coaching these teacher specialists would implement, and how their time
would be allocated across schools. District administrators and board members might consider
conducting their own cost-effectiveness analyses of additional uses of funding, based on their
own policy proposals, the literature supporting those proposals, and estimates of cost.
Discussion and Policy Implications
District administrators are well aware of the importance of investing limited educational
resources into strategies that are cost-effective. Even as LAUSD has reached a spending
agreement around salary increases and class size reduction, it will be important moving forward
to ensure that these policies are implemented with both costs and likely effects in mind. As I
noted earlier, this study represents one piece of evidence that might aid in the decisions around
how funding should be allocated within school districts. In this discussion, I highlight three key

147
contributions this study makes to research on school resource allocation.
First, this study is the first that I know of to draw on empirical data to compare the cost-
effectiveness of salary increases and class size reduction within the same context. The goal of
this study is not necessarily to declare definitively that one policy is more cost-effective than the
other, but rather to understand the conditions in which each of the two policies is likely to be
worth the investment. Many educational evaluations focus solely on the impacts of programs and
policies without taking into account the costs. Analysis of costs alone can provide useful
information for researchers and policymakers. For example, the maximum amount salaries can
be raised using the $195 million dollars in new funding is 11.3%, while the largest class size
reduction possible would lower class sizes by an average of 2.48 students. Because these policies
have equal cost in the first year, the approach that is more cost-effective depends solely on which
has a larger effect on the intended outcome. When policymakers consider educational costs, it
can be useful to move beyond the concept of dollar resources and instead consider opportunity
costs. This study shows, for instance, that the opportunity cost of a 2.48 reduction in districtwide
class sizes is an 11.3% salary increase.
Second, by modeling the cost-effectiveness of combination policies, this analysis
demonstrates the potential benefit of implementing policies together, rather than in isolation. For
example, researchers have argued that basic educational reforms such as class size reduction
often fail because they are implemented without attention to larger issues surrounding teacher
labor markets (Grubb, 2008; Jepsen & Rivkin, 2002; Shapson et al., 1980)
Finally, this study demonstrates how cost-effectiveness analyses can be conducted in
other local settings. Scholars have increasingly supported districts’ use of data driven decision-
making to guide policy and practice (Marsh, McCombs & Martorell, 2009) and many districts

148
are finding strategic ways to base policies on their own empirical evidence (LaFee, 2002;
Mandinach,

2012). By using the tools of cost-effectiveness analysis and sharing the findings with
interested stakeholders, districts can help various constituencies come to the table with more
information, which may foster more informed decision-making processes. Readers of this study
can use the methods described here to generate their own estimates of cost. Measures of policy
effects can also be gleaned from this study, allowing the reader to make cost-effectiveness
comparisons in their own local context. Improving the use of resources in districts over time is
an important step toward ensuring equal educational opportunity for all students.

149
Tables and Figures for Essay 3

FIGURE 1
Breakeven analysis comparing the cost-effectiveness of restoring K-3 class size to pre-layoff
levels and providing all teachers with a 6% salary increase, as a function of the estimated effect
of a 6% raise in teachers’ salaries on student achievement

Note: the solid line shows the relationship between the estimated effect of teachers’ salary increases and the extent
to which class size reduction is more cost-effective than raising teachers’ salaries. The break-even point shown in
this graph indicates that a 6% salary raise is equally as cost-effective as restoring K-3 class sizes to pre-layoff levels
if the effect on student achievement of the 6% raise is 0.012 standard deviations per $1,000 per pupil per year. Thus
at salary effect sizes larger than 0.012 (to the right of the break-even point), teachers’ salary increases are more cost-
effective than class size reduction. Conversely, if we assume the effect on student achievement of a 6% increase in
teachers’ salaries is less than 0.012 (to the left of the break-even point), class size reduction is more cost-effective
that salary increases. The dashed lines show estimated effect sizes from four studies of the effects of teachers’ salary
increases. Results from three of the four studies suggest that class size reduction is more cost-effective than raising
teachers’ salaries. With the exception of Greenwald, Hedges, and Laine (1996), effect sizes for each of these studies
are found by combining findings with Ronfeldt, Loeb, and Wyckoff (2013).

150
TABLE 1
Potential strategies for reducing class size, considering four different amounts of class size
reduction, two sets of targeted grade levels, and two sets of targeted schools, for a total of 16
approaches to reducing class size
Class Size Reduction Policy Grades Schools
Restoring class size to pre-layoff (2009) level
All grades

Lower grades
All schools

Lower-performing
schools
Class sizes recommended in an evidence-based funding model
Using all additional funding provided through Prop 30 / LCFF
Rehiring all previously laid off teachers with above average
evaluation scores
Note: “Lower grades” refers to kindergarten through grade 3. Lower-performing schools are defined as those in the
bottom 20% of Academic Performance Index. Additional funding provided through Proposition 30 (“Prop 30”) and
the Local Control Funding Formula (LCFF) total $100 million per year, based on LAUSD administrators’ estimates
(see the first footnote of Essay 1; LAUSD Board of Education, 2014a).

151
TABLE 2
Potential strategies for increasing teachers’ salaries, considering four different amounts of
salary increase, two sets of targeted teachers levels, and two sets of targeted schools, for a total
of 16 approaches to raising salaries
Salary Increase Policy Teachers Schools
6 percent
All teachers

Novice teachers
All schools

Lower-performing schools
8.5 percent
17.6 percent
Using all additional funding provided through Prop 30 / LCFF

152
TABLE 3
Potential strategies for combining class size reduction and salary increases
Compensation
policy
Class size policy
Teachers
targeted for
salary
increases
Grades
targeted for
class size
reduction
Schools
6 percent
Restoring class size to pre-layoff
(2008-09) level
All teachers

Novice
teachers
All grades

Lower grades
All schools

Lower-
performing
schools
8.5 percent
Class sizes recommended in
Evidence-Based funding model
17.6 percent
Rehiring all previously laid off
teachers with above average
evaluation scores
(in the top half of teachers
ranked by value-added to student
achievement)
Increase salaries by
the greatest amount
possible, after
implementing each
class size reduction
policies
Reduce class size by the greatest
extent possible (using all
remaining funding), after
implementing salary increase
policy
Note: a total of 256 potential spending policies are depicted in this table. For example, the first policy would provide
a 6% raise for all teachers and all schools and restore class sizes to pre-layoff levels for all grades and all schools.

153
TABLE 4
Cost-effectiveness of baseline policies for reducing class size or increasing teachers’ salaries
Policies
Number
of
students
Total cost
Cost
per
pupil
Salary
increase /
class size
reduction
Effect
size
Effectiveness-
cost ratio (SD
/ $1,000)
Rehire teachers with above
average "effectiveness," lower
grades, LP schools 37,640 $1,535,386 $41 0.23 0.0069 0.169
Rehire teachers with above
average "effectiveness," lower
grades, all schools 175,043 $1,574,864 $9 0.05 0.0011 0.125
Restore class size to pre-layoff
level, lower grades, LP schools 37,640 $20,067,200 $533 2.99 0.0495 0.093
Rehire teachers with above
average "effectiveness," all
grades, all schools 517,180 $2,448,646 $5 0.03 0.0004 0.088
Reduce class size by maximum
amount possible, lower grades,
LP schools 37,640 $208,592,046 $5,542 29.45 0.4877 0.088
Rehire teachers with above
average "effectiveness," all
grades, LP schools 93,439 $2,387,252 $26 0.12 0.0022 0.086
Reduce class size to levels
recommended in an EB model,
lower grades, LP schools 37,640 $60,031,138 $1,595 6.97 0.1137 0.071
Restore class size to pre-layoff
level, lower grades, all schools 175,043 $98,854,673 $565 3.15 0.0397 0.070
Reduce class size by maximum
amount possible, lower grades,
all schools 175,043 $208,592,046 $1,192 6.37 0.0802 0.067
6% raise, novice teachers, all
schools 24,909 $3,726,378 $150 6.0% 0.0091 0.061
8.5% raise, novice teachers, all
schools 24,909 $5,279,035 $212 8.5% 0.0129 0.061
17.6% raise, novice teachers, all
schools 24,909 $10,930,708 $439 17.6% 0.0268 0.061
6% raise, novice teachers, LP
schools 6,682 $1,022,028 $153 6.0% 0.0091 0.060
8.5% raise, novice teachers, LP
schools 6,682 $1,447,873 $217 8.5% 0.0129 0.060
17.6% raise, novice teachers, LP
schools 6,682 $2,997,950 $449 17.6% 0.0268 0.060
8.5% raise, all teachers, all
schools 517,180 $123,629,891 $239 8.5% 0.0129 0.054
6% raise, all teachers, all schools 517,180 $87,268,159 $169 6.0% 0.0091 0.054

154
17.6% raise, all teachers, all
schools 517,180 $255,986,598 $495 17.6% 0.0268 0.054
17.6% raise, all teachers, LP
schools 93,439 $47,673,218 $510 17.6% 0.0268 0.053
8.5% raise, all teachers, LP
schools 93,439 $23,023,998 $246 8.5% 0.0129 0.053
6% raise, all teachers, LP
schools 93,439 $16,252,234 $174 6.0% 0.0091 0.053
Raise salaries by the maximum
amount possible for all teachers
in all schools 517,180 $178,764,619 $346 11.3% 0.0173 0.050
Reduce class size to levels
recommended in an EB model,
lower grades, all schools 175,043 $384,245,938 $2,195 7.12 0.0877 0.040
Restore class size to pre-layoff
level, all grades, all schools 517,180 $173,627,328 $336 2.15 0.0130 0.039
Reduce class size by maximum
amount possible, all grades, all
schools 517,180 $208,571,176 $403 2.48 0.0149 0.037
Restore class size to pre-layoff
level, all grades, LP schools 93,439 $35,181,919 $377 1.78 0.0139 0.037
Reduce class size to levels
recommended in an EB model,
all grades, LP schools 93,439 $58,869,402 $630 2.88 0.0223 0.035
Reduce class size by maximum
amount possible, all grades, LP
schools 93,439 $208,571,176 $2,232 10.14 0.0789 0.035
Reduce class size to levels
recommended in an EB model,
all grades, all schools 517,180 $396,214,479 $766 3.81 0.0219 0.029
Raise salaries by the maximum
amount possible for novice
teachers in all schools 93,439 $178,764,619 $1,913 60.9% 0.0494 0.026
Raise salaries by the maximum
amount possible for all teachers
in LP schools 24,909 $178,764,619 $7,177 265.4% 0.1823 0.025
Raise salaries by the maximum
amount possible for novice
teachers in LP schools 6,682 $178,764,619 $26,754 967.7% 0.6386 0.024
Note: policies reported in grey text are considered not feasible because they either require extremely high increases
in salaries, or extremely large reductions in class sizes.

155
TABLE 5
Cost-effectiveness of selected combination policies for reducing class size and increasing
teachers’ salaries
Policies
Cost
per
pupil
Salary
increase
Class size
reduction
Effect
size
Effectiveness-
cost ratio (SD /
$1,000)
Restore class size to pre-layoff level in
lower grades in LP schools and provide
a 6% raise for novice teachers in all
schools
$683 6.0% 2.99 0.059 0.086
Restore class size to pre-layoff level in
lower grades in all schools and provide
a 8.5% raise for novice teachers in LP
schools
$781 8.5% 3.15 0.053 0.067
Reduce class sizes by maximum amount
possible in the lower grades of LP
schools, after providing all teachers with
an 8.5% salary raise
$1,543 8.5% 6.14 0.061 0.040
Note: LP stands for lower-performing.

156
References for Essay 3
Advancement Project. (2014). An equity aggregate student need index for LCFF in the Los
Angeles Unified School District. Los Angeles, CA: Author. Retrieved April 20, 2014
from:
http://www.advancementprojectca.org/?q=LCFF_Equity&utm_source=LA+School+Rep
ort+Daily+Email&utm_campaign=c68d3ce820-
Daily_Newsletter_June_18_6_18_2013&utm_medium=email&utm_term=0_2ad2e2e3a4
-c68d3ce820-86102573#STUDENT%20INDEX.
Achilles, C. M. (2012). Class-size policy: The STAR experiment and related class-size studies.
NCPEA Policy Brief, 1.2. Retrieved February 12, 2014 from:
http://files.eric.ed.gov/fulltext/ED540485.pdf.
Angrist, J. D., & Lavy, V. (1999). Using Maimonides’ rule to estimate the effect of class size on
scholastic achievement. Quarterly Journal of Economics, 114(2), 533–575.
Aos, S. & Pennucci, A. (2013). K–12 Class Size Reductions and Student Outcomes: A Review of
the Evidence and Benefit-Cost Analysis (Document No. 13-01-2201). Olympia:
Washington State Institute for Public Policy.
Aos, S., M. Miller, and J. Mayfield. (2007). Benefits and costs of K–12 educational policies:
Evidence-based effects of class size reductions and full-day kindergarten. Document 07-
03-2201. Olympia, WA: Washington State Institute for Public Policy.
Akerhielm, K. (1995). Does class size matter? Economics of Education Review, 14(3), 229-241.
Allegretto, S. A., Corcoran, S. P. Mishel, L. (2004). How does teacher pay compare?
Methodological challenges and answers. Washington, D.C.: Economic Policy Institute.
Retrieved June 13, 2015 from http://www.epi.org/publication/books_teacher_pay/.

157
Baker, B. D. (2012). Revisiting the age old question: Does money matter in Education?.
Washington, D.C.: The Albert Shanker Institute.
Barnes, G., Crowe, E., & Schaefer, B. (2007). The cost of teacher turnover in five school
districts: A pilot study. New York, NY: National Commission on Teaching and
America’s Future. Retrieved April 21, 2015 from: http://nctaf.org/wp-
content/uploads/CTTFullReportfinal.pdf.
Barnett, W. S., & Masse, L. N. (2007). Early childhood program design and economic returns:
Comparative benefit-cost analysis of the Abecedarian program and policy implications,
Economics of Education Review, 26, 113-125.
Belfield, C. R., Nores, M., Barnett, S., & Schweinhart, L. (2006). The High/Scope Perry
Preschool Program cost-benefit analysis using data from the age-40 follow-up. Journal of
Human Resources, 41(1), 162-190.
Blume, H. (2015, March 10). L.A. Unified to send layoff notices to hundreds of teachers. Los
Angeles Times. Retrieved March 21, 2015 from: www.latimes.com/local/lanow/la-me-ln-
school-layoff-notices-20150310-story.html.
Blume, H. (2014, January 17). L.A. teachers union calls for 17.6% raise. Los Angeles Times.
Retrieved February 2, 2014 from: www.latimes.com/local/lanow/la-me-ln-utla-calls-for-
176-raise-20140116,0,6920288.story#axzz2sDsLrdb4.
Blume, H. (2009, August 15). Los Angeles school district will sharply raise taxes on property
owners. Los Angeles Times. Retrieved on February 2, 2014 from
http://articles.latimes.com/2009/aug/15/local/me-lausd-bonds15.
Boardman, A. Greenberg, D. Vining, A. & Weimer, D. (2011). Cost-benefit analysis: Concepts
and practice (4th Ed.). Upper Saddle River, NJ: Prentice Hall.

158
Bohrnstedt, G. W. & Stecher, B. M. (2002). What we have learned about class size reduction in
California. Sacramento, CA: California Department of Education.
Borman, G. D., & Hewes. G. M. (2002). The long-term effects and cost-effectiveness of Success
for All. Educational Evaluation and Policy Analysis, 24(4), 243–266.
Boyd, D., Grossman, P., Ing, M., Lankford, H., Loeb, S. & Wyckoff, J. (2010). The influence of
school administrators on teacher retention decisions. American Educational Research
Journal, 48(2), 303-333. DOI: 10.3102/0002831210380788
Boyd, D. J., Langford, H., Loeb, S., & Wyckoff, J. H. (2010). Teacher layoffs: An empirical
illustration of seniority vs. measures of effectiveness. Washington, D.C.: The Urban
Institute.
Boyd, D., Lankford, H., Loeb, S., & Wyckoff, J. (2005). The draw of home: How teachers’
preferences for proximity disadvantage urban schools. Journal of Policy Analysis and
Management, 24(1), 113-132.
Brewer, D. J., Krop, C., Gill, B. P., & Reichardt, R. (1999). How much does it cost to reduce
class size? A national perspective. Educational Evaluation and Policy Analysis, 21(2),
179-192.
Brown, E. (2015, January 15). In Alaska and other energy states, low oil prices put pressure on
public schools. The Washington Post. Retrieved June 22, 2015 from:
http://www.washingtonpost.com/local/education/in-alaska-and-other-energy-states-low-
oil-prices-put-pressure-on-public-schools/2015/01/14/8ac55004-9b39-11e4-bcfb-
059ec7a93ddc_story.html.
Bunting, L., Kueneman, A, Louttit, M., Park, H. & Parker, D. (2014, January 28). State of the
States. New York Times. Retrieved April 20, 2014 from:

159
http://www.nytimes.com/interactive/2014/01/28/us/28-stateofstates.html.
Bureau of Labor Statistics. (2012). Employment situation archived news releases. Table B-1.
Retrieved from http://www.bls.gov/schedule/archives/empsit_nr.htm.
California Department of Education. (n. d.). Academic Performance Index (API). Retrieved
April 19, 2014 from http://www.cde.ca.gov/ta/ac/ap/.
California State Board of Education. (2014). Agenda for the California State Board of Education
(SBE) meeting on January 15-16, 2014. Retrieved on March 6, 2014 from
http://www.cde.ca.gov/be/ag/ag/yr14/agenda201401.asp
California Teachers Association. (n.d.). Vergara lawsuit targets teaching profession – California
Teachers Association. Retrieved July 11, 2014 from: http://www.cta.org/en/Issues-and-
Action/Ongoing-Issues/Vergara-Trial1.aspx.
CBS Los Angeles (2014, March 4). UTLA encourages parents to demand pay raises, smaller
class sizes. Retrieved on March 31, 2014 from:
http://losangeles.cbslocal.com/2014/03/04/utla-hands-out-leaflets-to-lausd-parents-
demanding-raises-smaller-classes/.
Chambers, J. (2010). Strategic school funding for results (SSFR): An overview of the project in
LAUSD. Retrieved on March 29, 2014 from:
http://www.schoolfundingforresults.org/reports/CHAMBERS_PACE_PIVOT_2010_4_2
7.pdf.
Chetty, R., Friedman, J. N., Hilger, N., Saez, E., Schanzenbach, D. W., & Yagan, D. (2011).
How does your kindergarten classroom affect your earnings? Evidence from Project
STAR. Quarterly Journal of Economics, 126(4), 1593-1660.
Chingos, M. M. (2013). Class size and student outcomes: Research and policy implications.

160
Journal of Policy Analysis and Management, 32(2), 411-438.
Chingos, M. M. (2012). The impact of a universal class-size reduction policy: Evidence from
Florida’s statewide mandate. Economics of Education Review, 31(5), 543-562.
Clotfelter, C., Glennie, E., Ladd, H., & Vigdor, J. (2008). Would higher salaries keep teachers in
high-poverty schools? Evidence from a policy intervention in North Carolina. Journal of
Public Economics, 92(5), 1352-1370.
Clune, W. H. (2002). Method strength and policy usefulness of cost-effectiveness research. In H.
M. Levin & P. McEwan (Eds.), Cost-effectiveness and educational policy (pp. 187–203).
New Rochelle, NY: Eye on Education.
Costrell, R. M., Hanushek, E. A., & Loeb, S. (2008). What do cost functions tell us about the
cost of an adequate education?. Peabody Journal of Education, 83(2), 198-223.
Costrell, R. M. & Podgursky, M. (2009). Teacher retirement benefits: Are employer
contributions higher than for private sector professionals? Education Next, 9(2), 58-63.
Condie, S., Lefgren, L., & Sims, D. (In press). Teacher heterogeneity, value added and education
policy. Economics of Education Review. Retrieved on February 1, 2014 from
http://dx.doi.org/10.1016/j.econedurev.2013.11.009
Dee, T., & West, M. (2011). The non-cognitive returns to class size. Education Evaluation and
Policy Analysis, 33(1), 23–46.
Dolton, P. & Marcenaro-Gutierrez, O. D. (2011). If you pay peanuts do you get monkeys? A
cross-country analysis of teacher pay and pupil performance. Economic Policy, 26, 5-55.
DOI: 10.1111/j.1468-0327.2010.00257.x
Education Commission of the States. (2005). State class-size reduction measures. Denver, CO:
Author.

161
Edwards, V. & Prichard, M. E. (2006). Why is LAUSD’s enrollment declining if the Los
Angeles region’s population is growing? Master Planning and Demographics, Los
Angeles Unified School District. Retrieved on February 2, 2014 from
http://paa2006.princeton.edu/papers/61492.
Estrada, J. (2012). A review of the teacher layoff process in California. California Legislative
Analyst’s Office. Retrieved August 4, 2013 from
http://www.lao.ca.gov/reports/2012/edu/teacher-layoffs/teacher-layoffs-032212.aspx
Evans, L. (1997). Understanding teacher morale and job satisfaction. Teaching and Teacher
Education, 13(8), 831-845.
Felch, J., Song, J., & Smith, D. (2010, December). When layoffs come to L.A. schools,
performance doesn't count. Los Angeles Times. Retrieved April 21, 2015 from:
http://articles.latimes.com/2010/dec/04/local/la-me-1205-teachers-seniority-20101204.
Figlio, D. N. (1997). Teacher salaries and teacher quality. Journal of Economic Letters, 55(1),
267-271.
Figlio, D. N. (2002). Can public schools buy better-qualified teachers? Industrial and Labor
Relations Review, 55(4), 686-699.
Finn, J. D. & Achilles, C. M. (1990). Answers and questions about class size: A statewide
experiment. American Educational Research Journal, 27(3), 557-577. DOI:
10.3102/00028312027003557.
Fletcher, W. (2014). Letter to Dr. Deasy. Los Angeles, CA: United Teachers of Los Angeles.
Retrieved March 4, 2014 from:
http://www.utla.net/system/files/Letter_to_Dr.DeasyRESalaryIncreaseBargaining012914.
pdf.

162
Folger, J., & Breda, C. (1989). Evidence from project STAR about class size and student
achievement. Peabody Journal of Education, 67(1), 17–33.
Frey, S. (2012). Beyond personal pain, teacher layoffs have rippling effects on schools. Oakland,
CA: EdSource. Retrieved February 12, 2013 from: http://edsource.org/2012/beyond-
individual-pain-teacher-layoffs-have-rippling-effects-on-schools/6893#.VTyeolqaT8F.
Gilpin, G., & Kaganovich, M. (2012). The quantity and quality of teachers: Dynamics of the
trade-off. Journal of Public Economics, 96(3), 417-429.
Goodman, S. F., & Turner, L. J. (2013). The design of teacher incentive pay and educational
outcomes: Evidence from the New York City bonus program. Journal of Labor
Economics, 31(2), 409-420.
Glass, G. V. & Smith, M. L. (1979). Meta-analysis of research on class size and achievement.
Educational Evaluation and Policy Analysis, 31, 77-95
Glazerman, S., Chiang, H., Wellington, A., Constantine, J., & Player, D. (2011). Impacts of
performance pay under the Teacher Incentive Fund: Study design report. Mathematica
Policy Research, Inc. Retrieved on February 13, 2014 from http://www.mathematica-
mpr.com/publications/PDFs/education/performpay_TIF.pdf?spMailingID=4078907.
Glazerman, S., & Seifullah, A. (2012). An evaluation of the Chicago Teacher Advancement
Program (Chicago TAP) after four years. Mathematica Policy Research, Inc. Retrieved
on February 13, 2014 http://www.mathematica-
mpr.com/publications/pdfs/education/tap_year4_impacts.pdf.
Goldhaber, D., DeArmond, M., Player, D., & Choi, H. J. (2008). Why do so few public school
districts use merit pay?. Journal of Education Finance, 33(3), 262-289.

163
Goldhaber, D., Gross, B., & Player, D. (2011). Teacher career paths, teacher quality, and
persistence in the classroom: Are public schools keeping their best? Journal of Policy
Analysis and Management, 30, 57–87.
Goldhaber, D., & Hansen, M. (2010). Using performance on the job to inform teacher tenure
decisions. American Economic Review, 250-255.
Goldhaber, D., Lavery, L., & Theobald, R. (2015). Inconvenient truth? Do collective bargaining
agreements help explain the mobility of teachers within school districts?. CEDR Working
Paper 2015-1. University of Washington, Seattle, WA.
Goldhaber, D., & Theobald, R. (2013). Managing the teacher workforce in austere times: The
determinants and implications of teacher layoffs. Education Finance and Policy, 8(4),
494-527.
Greenwald, R., Hedges, L. V., & Laine, R. D. (1996). The effect of school resources on student
achievement. Review of Educational Research, 66, 361-396.
Grissmer, D. (2002). Cost-effectiveness and cost-benefit analysis: The effect of targeting
interventions. In H. M. Levin and P. J. McEwan, (Eds.). Cost-effectiveness and
educational policy. Larchmont, NY: Eye on Education.
Grissom, J. A., & Strunk, K. O. (2012). How should school districts shape teacher salary
schedules? Linking school performance to pay structure in traditional compensation
schemes. Educational Policy, 26(5), 663-695. DOI: 10.1177/0895904811417583
Guin, K. (2004). Chronic teacher turnover in urban elementary schools. Education policy
analysis archives, 12(42), Retrieved May 17, 2013 from
http://epaa.asu.edu/epaa/v12n42/.

164
Hannaway, J., Sass, T., Figlio, D., & Feng, L. (2009). Value added of teachers in high-poverty
schools and lower poverty schools: Implications for management and policy. CALDER
Working Paper. Washington, DC: The Urban Institute.
Hanushek, E. A. (2003). The failure of inputbased schooling policies. The Economic Journal,
113(485), F64-F98.
Hanushek, E. A. (1998). The evidence on class size. Occasional Paper Number 98-1. Rochester,
NY: University of Rochester.
Hanushek, E.A. (1997) Assessing the effects of school resources on student performance: An
update. Educational Evaluation and Policy Analysis, 19(2), 141-164
Hanushek, E. A., Kain, J. F., & Rivkin, S. G. (2004). Why public schools lose teachers. Journal
of Human Resources, 39(2), 326-354.
Hanushek, E. A., Kain, J. F., & Rivkin, S. G. (1999). Do higher salaries buy better teachers?.
NBER working paper no. 7082. Retrieved on January 4
th
, 2014 from:
http://www.nber.org.libproxy.usc.edu/papers/w7082.pdf?new_window=1.
Hanushek, E. A. & Luque, J. (2000). Smaller classes, lower salaries? The effects of class size on
teacher labor markets. In Sabrina W. M. Laine and James G. Ward (Eds.). Using what we
know: A review of the research on implementing class-size reduction initiatives for state
and local policymakers. Oak Brook, IL: North Central Regional Educational Laboratory.
Hanushek, E. A., & Rivkin, S. G. (2012). The distribution of teacher quality and implications for
policy. Annual Review of Economics, 4(1), 131-157.
Harris, D. N. (2009). Toward policy-relevant benchmarks for interpreting effect sizes:
Combining effects with costs. Educational Evaluation and Policy Analysis, 31(1), 3-29.
Harris, D. N., Ingle, W. K., & Rutledge, S. A. (2014). How teacher evaluation methods matter

165
for accountability: A comparative analysis of teacher effectiveness ratings by principals
and teacher value-added measures. American Educational Research Journal, 51(2), 73-
112. doi: 10.3102/0002831213517130.
Harris, D. N. & Sass, T. (2011). Teacher training, teacher quality, and student achievement.
Journal of Public Economics, 95, 798-812.
Heckman, J. J., & Rubenstein, Y. (2001). The importance of non-cognitive skills: Lessons from
GED testing program. American Economic Review, 91(2), 145–149.
Heckman, J., Layne-Farrar, A. & Todd, P. (1995). Does measured school quality really matter?
An examination of the earnings-quality relationship. NBER working paper no. 5274
Hedges, L. V., Laine, R. D. & Greenwald, R. (1994). An exchange: Part I: Does money matter?
A meta-analysis of studies of the effects of differential school inputs on student
outcomes. Educational Researcher, 23(3), 5-14.
Honig, M. I. (2006). New directions in education policy implementation: Confronting
complexity. Albany, NY: State University of New York Press
Hoxby, C. M. (2000). The effects of class size on student achievement: New evidence from
population variation. Quarterly Journal of Economics,
Hoxby, C. M. (1996). How teachers’ unions affect education production. Quarterly Journal of
Economics, 111(3), 671–718.
Imazeki, J. (2005). Teacher salaries and teacher attrition. Economics of Education Review, 24,
431–449.
Ingersoll, R. M. (2001). Teacher turnover and teacher shortages: An organizational analysis.
American Educational Research Journal, 38, 499–534.

166
Jacob, B. A. (2011). Do principals fire the worst teachers? Educational Evaluation and Policy
Analysis, 33, 403–434.
Jackson, C. K. (2009). Student demographics, teacher sorting, and teacher quality: Evidence
from the end of school desegregation. Journal of Labor Economics, 27(2), 213–56.
Janofsky, M. (2014a, May 27). JUST IN: LAUSD offers teachers 2 percent raise, union insulted.
LA School Report. Retrieved from: http://laschoolreport.com/lausd-offers-teachers-
salary-2-percent-raise-union-insulted/.
Janofsky, M. (2014b, April 7). Unions have lukewarm response to Deasy’s new budget proposal.
LA School Report. Retrieved on April 8, 2014 from http://laschoolreport.com/unions-
have-lukewarm-response-to-deasys-2014-15-budget-
proposal/?utm_source=LA+School+Report+Daily+Email&utm_campaign=b17496f67c-
Daily_Newsletter_June_18_6_18_2013&utm_medium=email&utm_term=0_2ad2e2e3a4
-b17496f67c-68246533#more-21948.
Jepsen, C. & Rivkin, S. (2009). Class size reduction and student achievement: The potential
tradeoff between teacher quality and class size. Journal of Human Resources, 44(1), 223-
250.
Jones, B. (2013, June 1). LAUSD Prop 30 money: School board members divided over how to
spend money. Los Angeles Daily News. Retrieved on March 22, 2014 from:
http://www.huffingtonpost.com/2013/06/01/lausd-prop-30-money_n_3371549.html
Kahn, C. (2009, June). L.A.'s urban schools hardest hit by teacher layoffs. National Public
Radio. Retrieved April 22, 2015 from:
http://www.npr.org/templates/story/story.php?storyId=105492482.
Kane, T. & Staiger, D. (2008). Estimating teacher impacts on student achievement: An

167
experimental evaluation. NBER Working Paper 14607.
King, J. A. (2010). The impact of teacher experience: Examining the evidence and policy
implications. National Center of Longitudinal Data in Education Research. Retrieved
form http://www.urban.org/UploadedPDF/1001455-impact-teacher-experience.pdf
Koedel, C. & Betts, J. R. (2007). Re-examining the role of teacher quality in the educational
production function. National Center on Performance Incentives, Vanderbilt, Peabody
College.
Krieg, J. M. (2006). Teacher quality and attrition. Economics of Education Review, 25, 13–27.
Krueger, A. B. (1999). Experimental estimates of education production functions. The Quarterly
Journal of Economics, 114(2), 497-532.
LA Daily News. (2013, September 17). LAUSD approves $113M budget to train teachers for
Common Core standards. Los Angeles, CA: Author. Retrieved from
http://www.dailynews.com/social-affairs/20130917/lausd-approves-113m-budget-to-
train-teachers-for-common-core-standards.
Ladd, H. F. (2007). Teacher labor markets in developed countries. Future of Children, 17(1),
201–217.
LaFee, S. (2002). Data-driven districts. School Administrator, 59(11), 6-15.
Lankford, H., Loeb, S., & Wyckoff, J. (2002). Teacher sorting and the plight of urban schools: A
descriptive analysis. Educational Evaluation and Policy Analysis, 24(1), 37–62.
Leachman, M. & Mai, C. (2013). Most states funding schools less than before the recession.
Washington, D.C.: Center on Budget and Policy Priorities. Retrieved April 17, 2014
from: http://www.cbpp.org/files/9-12-13sfp.pdf.

168
Lee, J. & Finn, J. (2010, June). Developing time-indexed effect size metrics for K-12 reading
and math educational evaluation. Poster presented at the Institute of Education Sciences
(IES) annual conference, Washington, DC.
Levin, H. M. (2002). The cost-effectiveness of whole school reforms (ERIC Clearinghouse on
Urban Education, Urban Diversity Series 114). New York, NY: Columbia University,
Teachers College.
Levin, H. M. (2001). Waiting for Godot: Cost-effectiveness analysis in education. In Richard
Light (Ed.) Evaluations that Surprise. San Francisco, CA: Jossey-Bass.
Levin, H. M. (1983). Cost-effectiveness. Thousand Oaks, CA: Sage.
Levin, H. M., & Belfield, C. (2015). Guiding the development and use of cost-effectiveness
analysis in education. Journal of Research on Educational Effectiveness, 8(3), 400-418,
DOI: 10.1080/19345747.2014.915604
Levin, H. M., Belfield, C., Hollands, F., Bowden, B. A., Cheng, H., Shand, R., Pan, Y., &
Hanisch-Cerda, B. (2012). Cost-effectiveness analysis of interventions that improve high
school completion. Center for Benefit-Cost Studies of Education, Teachers College,
Columbia University. Retrieved on January 4, 2014 from
http://www.ny.frb.org/research/education_seminar_series/IESHighSchoolCompletion.pdf
Levin, H. M., Glass, G. V., & Meister, G. (1987). A cost-effectiveness analysis of computer-
assisted instruction. Evaluation Review, 11, 50-72.
Levin, H. M. & McEwan, P. J. (2000). Cost-effectiveness analysis: Methods and applications.
Thousand Oaks, CA: Sage Publications, Inc.
Lightfoot, J., & Bines, W. (2000). Working to keep school children healthy: the complementary
roles of school staff and school nurses. Journal of Public Health, 22(1), 74-80.

169
Llanos, C. (2010, November 9). LAUSD's numbers keep on dropping. Los Angeles Daily News.
Retrieved February 2, 2014 from: http://www.dailynews.com/20101110/lausds-numbers-
keep-on-dropping.
Loeb, S. & Page, M. E. (2000). Examining the link between teacher wages and student
outcomes: The importance of alternative labor market opportunities and non-pecuniary
variation. The Review of Economics and Statistics, 82(3), 393-408.
Los Angeles Unified School District. (2015a, January 14). Revised LAUSD salary increase
proposal. Los Angeles, CA: Author. Retrieved February 2, 2015 from:
home.lausd.net/ourpages/auto/2015/1/14/40061727/UTLA-DistrictProposal-ArticleXIV-
RevisedSalaryIncrease-01-14-15.pdf.
Los Angeles Unified School District. (2015b, January 26). Letter to employees & labor partners
– UTLA Negotiations. Los Angeles, CA: Author. Retrieved February 2, 2015 from:
home.lausd.net/ourpages/auto/2015/1/26/54959551/DOC012615-01262015143542.pdf.
Los Angeles Unified School District. (2015c, February 18). LAUSD agrees with UTLA’s
Declaration of Impasse. Los Angeles, CA: Author. Retrieved February 2, 2015 from:
home.lausd.net/ourpages/auto/2015/2/18/50272463/NS-
LAUSD_UTLA%20impasse_021815d-ds_Final.pdf.
Los Angeles Unified School District. (2015d, April 17). Los Angeles Unified School District –
United Teachers Los Angeles tentative agreement 2014-2017. Los Angeles, CA: Author.
Retrieved September 5, 2015 from:
http://achieve.lausd.net/cms/lib08/CA01000043/Centricity/domain/368/utla/utla%20side
%20letters/LAUSD-UTLA%20-%202014-2017%20Tentative%20Agreement%2004-17-
15.pdf.

170
Los Angeles Unified School District Board of Education. (2014a, April 4). LAUSD’S local
control accountability plan (LCAP) - 1st draft. Los Angeles, CA: Author. Retrieved July
11, 2014 from:
http://achieve.lausd.net/cms/lib08/CA01000043/Centricity/Domain/358/Board%20Presen
tation%20LCAP%20Draft%204414%20FINAL.pdf.
Los Angeles Unified School District. (2014b, January 9). Statement from board member Tamar
Galatzan on Gov. Jerry Brown's proposed budget. News Releases. Los Angeles, CA:
Author. Retrieved March 3, 2014 from
http://home.lausd.net/apps/news/show_news.jsp?REC_ID=350706&id=27.
Los Angeles Unified School District. (2014c). Predominantly Black, Hispanic, Asian & other
non-Anglo (PHBAO) school program. Los Angeles, CA: Author. Retrieved April 20,
2014 from:
http://notebook.lausd.net/portal/page?_pageid=33,1122877&_dad=ptl&_schema=PTL_E
P.
Los Angeles Unified School District. (2013a). Superintendent’s 2013-14 final budget. Budget
Services and Financial Planning Division. Los Angeles, CA: Author. Retrieved February
12, 2014 from: http://laschoolboard.org/sites/default/files/LAUSD2013-
14FinalBudget.pdf.
Los Angeles Unified School District. (2013b, June 4). Special meeting order of business. Los
Angeles, CA: Author. Retrieved June 15, 2013 from: http://laschoolboard.org/06-04-
13SpclBd.
Lyman, R. (2014, February 2). Battles loom in many states over what to do with budget
surpluses. New York Times. Retrieved April 20, 2014 from:

171
http://www.nytimes.com/2014/02/03/us/battles-loom-in-many-states-over-what-to-do-
with-budget-surpluses.html?_r=0.
Mandinach,

E. B. (2012). A perfect time for data use: Using data-driven decision making to
inform practice. Educational Psychologist, 47(2), 71-85. DOI:
10.1080/00461520.2012.667064
Marsh, J. A., McCombs, J. S. & Martorell, F. (2010). How instructional coaches support data-
driven decision making: Policy implementation and effects in Florida middle schools.
Educational Policy, 24(6), 872-907.
Mihaly, K., McCaffrey, D. F., Staiger, D. O., & Lockwood, J. R. (2013). A composite estimator
of effective teaching. Technical Paper. Retrieved from:
http://www.metproject.org/downloads/MET_Composite_Estimator_of_Effective_Teachi
ng_Research_Paper.pdf.
McCaffrey, D. F., Sass, T. R., Lockwood, J. R., & Mihaly, K. (2009). The intertemporal
variability of teacher effect estimates. Education Finance and Policy, 4(4), 572-606.
McEwan, P. J. (2002). Are cost-effectiveness methods used correctly? In H. M. Levin and P. J.
McEwan. (Eds). Cost-effectiveness and educational policy. Yearbook of the American
Education Finance Association. Larchmont, NY: Eye on Education.
Milanowski, A. T. & Odden, A. R. (2007). A new approach to the cost of teacher turnover.
CRPE Working Paper No. 13. Seattle, WA: Center on Reinventing Public Education.
Miles, K. H., Odden, A., Fermanich, M., & Archibald, S. (2004). Inside the black box of school
district spending on professional development: Lessons from five urban districts. Journal
of Education Finance, 30 (1), 1–26.
Monk, D. H., & King, J. A. (1993). Cost analysis as a tool for education reform. In S. L.

172
Jacobson and R. Berne (Eds.). Reforming education: The emerging systemic approach.
Thousand Oaks, CA: Corwin Press.
Mosteller, F. (1995). The Tennessee study of class size in the early school grades. The future of
children, 113-127.
Murnane, R. J. & Olsen, R. J. (1989). The effects of salaries and opportunity costs on duration in
teaching: Evidence from Michigan. The Review of Economics and Statistics, 71(2), 347-
352.
Nye, Hedges, L. V. & Konstantopoulos. S. (2002). Do low-achieving students benefit more from
small classes? Evidence from the Tennessee class size experiment. Educational
Evaluation and Policy Analysis, 24(3), 201-217.
Nye, B., Hedges, L.V., & Konstantopoulos, S. (2004). Do minorities experience larger lasting
benefits from small classes? Evidence from a five-year follow-up of the Tennessee class
size experiment. Journal of Educational Research, 98, 94-100.
Nye, B., Hedges, L.V., & Konstantopoulos, S. (2001). Are the effects of small classes
cumulative? Evidence from the Tennessee class size experiment. Journal of Educational
Research, 94, 336-345.
Nye, B., Hedges, L.V., & Konstantopoulos, S. (2000). Effects of small classes on academic
achievement: The results of the Tennessee class size experiment. American Educational
Research Journal, 37, 123-151.
Nye, B., Hedges, L.V., & Konstantopoulos, S. (1999). The long-term effects of small classes: A
five-year follow-up of the Tennessee class size experiment. Educational Evaluation and
Policy Analysis, 21, 127-142.

173
Odden, A. R. (2012). Improving student learning when budgets are tight. Thousand Oaks, CA:
Corwin Press
Odden, A., & Kelley, C. (2002). Paying teachers for what they know and do: New and smarter
compensation strategies to improve schools (2nd ed.). Thousand Oaks, CA: Corwin
Press.
Odden, A. R. & Picus, L. O. (2008). School Finance: A Policy Perspective. (4
th
Ed). New York:
McGraw-Hill.
Odden, A.R., Picus, L. O. & Fermanich, M. (2003). A state of the art approach to school finance
adequacy in Kentucky. Kentucky Department of Education.
Oliff, P. & Leachman, M. (2011). New school year brings steep cuts in state funding for schools.
Washington, D.C.: Center on Budget and Policy Priorities. Retrieved from
http://www.cbpp.org/files/9-1-11sfp.pdf.
Perna, L. W., Rowan-Kenyon, H. T., Thomas, S. L., Bell, A., Anderson, R., & Li, C. (2008). The
role of college counseling in shaping college opportunity: Variations across high schools.
The Review of Higher Education, 31(2), 131-159.
Picus, L.O., Hayward, G.C. & Ehlers, J. (2002). Wyoming education finance: Small schools
report. Management Analysis and Planning, Inc., Davis, CA. Submitted to Wyoming
State Legislature.
Picus, L. O., Monk, D. H. & Knight, D. S. (2012). Measuring the cost effectiveness of rich
clinical practice in teacher preparation: Part one, understanding the problem. Paper
prepared for the National Council for Accreditation of Teacher Education.
http://www.caepsite.org/events/CostEffectiveness.pdf.

174
Plecki, M., Elfers, A., & Finster, M. (2010). Examining the impact of Reduction in Force (RIF)
notices in Washington School Districts: 2009-2010. Report for the Center for
Strengthening the Teaching Profession, March 2010. Podgursky, M., & Springer, M.
(2007). Teacher performance pay: A review. Journal of Policy Analysis and
Management, 26, 909-949.
Puskar, K. R., & Marie Bernardo, L. (2007). Mental health and academic achievement: Role of
school nurses. Journal for Specialists in Pediatric Nursing, 12(4), 215-223.
Reback, R. (2010a). Noninstructional spending improves noncognitive outcomes: Discontinuity
evidence from a unique elementary school counselor financing system. Education
Finance and Policy, 5(2), 105-137.
Reback, R. (2010b). Schools' mental health services and young children's emotions, behavior,
and learning. Journal of Policy Analysis and Management, 29(4), 698-725.
Reichardt, R. E. (2000). The cost of class size reduction: Advice for policymakers. Santa
Monica, CA: RAND Graduate School of Policy Studies.
Richwine, J. & Biggs, A. G. (2011). Assessing the compensation of public-school teachers. A
report of the Heritage Center for Data Analysis. CDA 11-03. Washington, D.C.: The
Heritage Foundation and American Enterprise Institute.
Rivkin, S. G., Hanushek, E. A., & Kain, J. F. (2005). Teachers, schools, and academic
achievement. Econometrica, 73(2), 417–458.
Romo, V. (2014, October 3). Teachers union revises wage demand to 10% raise for a year.
Retrieved December 28, 2014 from: http://laschoolreport.com/teachers-union-revises-
wage-demand-to-10-raise-for-a-year-lausd/.
Ross, J. A., Barkaoui, K., & Scott, G. (2007). Evaluations that consider the cost of educational

175
programs: The contribution of high quality studies. American Journal of Evaluation, 28,
477–492.
Rothstein, J. (2012). Teacher quality policy when supply matters. NBER Working Paper No.
18419. Retrieved December 19, 2013, from http://www.nber.org/papers/w18419.
Rothstein, J. (2010). Teacher quality in educational production: Tracking, decay and student
achievement. Quarterly Journal of Economics, 125(1), 175-214.
Seder, R.C., Picus, L.O. & Smith, J.R. (2002). Wyoming education finance: Estimating the costs
of services for at risk students. Management Analysis and Planning, Inc., Davis, CA.
Submitted to Wyoming State Legislature.
Shapson, S. M., Wright, E. N., Eason, G., & Fitzgerald, J. (1980). An experimental study of the
effects of class size. American Educational Research Journal, 17(2), 141-152.
Springer, M. G., Pane, J. F., Le, V. N., McCaffrey, D. F., Burns, S. F., Hamilton, L. S., &
Stecher, B. (2012). Team pay for performance: Experimental evidence from the round
rock pilot project on team incentives. Educational Evaluation and Policy Analysis, 34(4),
367-390.
Staiger, D. O., & Rockoff, J. E. (2010). Searching for effective teachers with imperfect
information. The Journal of Economic Perspectives, 24, 97–117.
Stiglitz, J. E. (2012). The price of inequality: How today’s divided society endangers our future.
New York, NY: W. W. Norton & Company.
Stinebrickner, T. R. (2002). An analysis of occupational change and departure from the labor
force: Evidence of the reasons that teachers leave. Journal of Human Resources, 37(1),
192–216.
Stinebrickner, T. R. (1998). An empirical investigation of teacher attrition. Economics of

176
Education Review, 17(2), 127–136.
Strunk, K. O., McEachin, A., & Westover, T. N. (July, 2012). The use and efficacy of capacity-
building assistance for low performing districts: The case of California’s district
assistance and intervention teams. Journal of Policy Analysis and Management. Advance
online publication. doi: 10.1002/pam.21658.
Strunk, K. O., Marsh, J. A., Hashim, A., Bush, S., & Weinstein, T. (2012). The efficacy of the
Los Angeles Unified School District Public School Choice Initiative for student
achievement outcomes: Early evidence from the first year. Presented at the 2014 Annual
Conference for the American Educational Research Association, Philadelphia, PA.
Strunk, K. O., Weinstein, T. L., & Makkonen, R. (forthcoming). New evidence on the
relationship between multiple measures of teacher effectiveness in practice: Evidence
from the Los Angeles Unified School District’s Educator Growth and Development
Cycle Initial Implementation Phase. Education Policy Analysis Archives, 22(75).
Summers, A. A., & Wolfe, B. (1977). Do schools make a difference? American Economic
Review, 639–652.
Taylor, M. (2013). Updated: An overview of the Local Control Funding Formula. California
Legislative Analyst’s Office. Retrieved on January 4, 2014 from:
http://www.lao.ca.gov/reports/2013/edu/lcff/lcff-072913.pdf.
Tyack, D. & Cuban, L. (1995). Tinkering toward utopia: A century of public school reform.
Cambridge, MA: Harvard University Press.
Wenglinsky, H. (1997) How money matters: The effect of school district spending on academic
achievement. Sociology of Education, 70(3), 221-237.
West, M. R., & Chingos, M. M. (2009). Teacher effectiveness, mobility, and attrition in Florida.

177
In M. G. Springer (Ed.). Performance incentives: Their growing impact on American K-
12 education (pp. 251–271). Washington, DC: Brookings Institution Press.
Winters, M. A., & Cowen, J. M. (2013). Would a value-added system of retention improve the
distribution of teacher quality? A simulation of alternative policies. Journal of Policy
Analysis and Management, 32(3), 634-654.
Word, E., Johnston, J., Bain, H. P., Fulton, D. B., Zaharias, J. B., Achilles, C. M., Lintz, M. N.,
Folger, J., & Breda, C. (1990). The state of Tennessee’s Student/Teacher Achievement
Ratio (STAR) Project: Technical report 1985–1990. Nashville: Tennessee State
Department of Education, 1990.
Yeh, S. S. (2012). The reliability, impact and cost-effectiveness of value-added teacher
assessment methods. Journal of Education Finance, 37(4).
Yeh, S. S. (2010). The cost effectiveness of 22 approaches for raising student achievement.
Journal of Education Finance, 36(1), 38-75.
Yeh, S. S., & Ritter, J. (2009). The cost-effectiveness of replacing the bottom quartile of novice
teachers through value-added teacher assessment. Journal of Education Finance, 34(4),
426-451.
Yuan, K., Le, V. N., McCaffrey, D. F., Marsh, J. A., Hamilton, L. S., Stecher, B. M., & Springer,
M. G. (2013). Incentive pay programs do not affect teacher motivation or reported
practices: Results from three randomized studies. Educational Evaluation and Policy
Analysis, 35(1), 3-22.
U.S. Department of Education, National Center for Education Statistics. (2013). The condition of
education 2013 (NCES 2013-037), Public School Expenditures. Retrieved from
http://nces.ed.gov/pubsearch/pubsinfo.asp?pubid=2013037.

178
UCLA/IDEA (2009). Sharing the burden? The impact of proposed teacher layoffs across
LAUSD. Institute for Democracy, Education, and Access.
United Teachers of Los Angeles. (2015a, January 22). Bargaining proposal, Article XIV,
salaries. Los Angeles, CA: Author. Retrieved February 2, 2015 from:
www.utla.net/system/files/UTLASalaryProposal.pdf.
United Teachers of Los Angeles. (2014a, December 16). UTLA adjusts salary demand to 9%
increase 2014-15. Los Angeles, CA: Author. Retrieved December 28, 2014 from:
http://www.utla.net/bargainingproposals121614.
United Teachers of Los Angeles. (2014b). Tuesday talking points. Los Angeles, CA: Author.
Retrieved March 4, 2014 from: http://www.utla.net/system/files/TTP_2014_2_11.pdf.
United Teachers of Los Angeles. (2014c). We demand LAUSD use new money for our kids:
Parents, teachers & school staff working together for our schools. Los Angeles, CA:
Author. Retrieved March 4, 2014 from:
http://www.utla.net/system/files/UTLA_Leafleting_Parent_Flier.pdf.
United Teachers of Los Angeles (n. d.). UTLA: District offer not what L.A. students and
educators deserve. Retrieved from: http://laschoolreport.com/wp-
content/uploads/2014/05/Union-News-Release-response-to-2-percent-teacher-raise-
LAUSD.pdf.

179
APPENDIX A-3: Additional Tables for Essay 3
APPENDIX TABLE A1
Class size maximums based on April 2015 agreement between the Los Angles Unified School
District and the United Teachers of Los Angeles, for schools designated as Predominantly
Hispanic, Black, Asian, or other Non-Anglo (PHBAO) and all other schools
Grade
PHBAO schools Non-PHBAO schools
Academic Non-academic

Academic Non-academic
K - 3rd grade 24 24 24 24
Grade 4 - 5 30.5 30.5 36 36
Grade 6 - 8 34 42.5 39.5 42.5
Grade 9 - 10 34 42.5 39.5 42.5
Grade 11 - 12 42.5 42.5 42.5 42.5
Note: The district refers to PHBAO schools (predominantly Hispanic, Black, Asian or other) as schools with 70% or
more students of color. Academic classes include those in the departments of English, English as a Second
Language, Reading, Math, Social studies, Science, and Foreign language. Non-academic classes include all others,
with no class size caps for physical education and activity classes such as Band or Drill team. The LAUSD class size
reduction plan also requires a two-student reduction in all math and English language arts classes in grades 7 and 8.

180
APPENDIX TABLE A3
Predicted salaries by teacher experience
Experience
Total
compensation
Std. Err. z P>|z| [95% Conf. Interval]
0 67845.4 54.70823 1240.13 0.000 67738.17 67952.62
1 71123.25 47.14822 1508.5 0.000 71030.84 71215.66
2 74256.58 40.29196 1842.96 0.000 74177.61 74335.55
3 77245.38 34.20041 2258.61 0.000 77178.35 77312.41
4 80089.67 28.96184 2765.35 0.000 80032.9 80146.43
5 82789.42 24.69611 3352.33 0.000 82741.02 82837.83
6 85344.66 21.54148 3961.88 0.000 85302.44 85386.88
7 87755.37 19.60049 4477.2 0.000 87716.95 87793.79
8 90021.56 18.84805 4776.17 0.000 89984.62 90058.5
9 92143.22 19.08155 4828.92 0.000 92105.82 92180.62
10 94120.36 19.98875 4708.67 0.000 94081.18 94159.54
11 95952.98 21.26795 4511.62 0.000 95911.29 95994.66
12 97641.07 22.69616 4302.1 0.000 97596.59 97685.55
13 99184.64 24.13597 4109.41 0.000 99137.33 99231.94
14 100583.7 25.51823 3941.64 0.000 100533.7 100633.7
15 101838.2 26.82399 3796.53 0.000 101785.6 101890.8
16 102948.2 28.07193 3667.3 0.000 102893.2 103003.2
17 103913.7 29.31021 3545.31 0.000 103856.2 103971.1
18 104734.6 30.61081 3421.49 0.000 104674.6 104794.6
19 105411.1 32.06408 3287.51 0.000 105348.2 105473.9
20 105943 33.77202 3137 0.000 105876.8 106009.2

181
APPENDIX TABLE A4
Summary of class size reduction effect sizes
Author Grades
Average Effect size

Math Reading
Num. of
students
Krueger (1999) 0 0.200 0.200 8
Krueger (1999) 1 0.280 0.280 8
Jepsen and Rivkin (2009) 2 0.040 0.025 10
Krueger (1999) 2 0.220 0.220 8
Angrist & Lavy (1999) 3 0.000 0.000 10
Chingos (2012) 3 0.000 0.000 3
Cho, Glewwe, and Whitler (2012) 3 0.045 0.045 10
Jepsen and Rivkin (2009) 3 0.040 0.025 10
Krueger (1999) 3 0.190 0.190 8
Angrist & Lavy (1999) 4 0.000 0.100 10
Chingos (2012) 4 0.000 0.000 3
Hoxby (2000) 4 0.000 0.000 n/a
Jepsen and Rivkin (2009) 4 0.040 0.025 10
Rivkin, Hanushek, and Kain (2005) 4 0.095 0.095 10
Angrist & Lavy (1999) 5 0.150 0.220 10
Rivkin, Hanushek, and Kain (2005) 5 0.095 0.035 10
Chingos (2012) 5 0.000 0.000 3
Cho, Glewwe, and Whitler (2012) 5 0.045 0.045 10
Rivkin, Hanushek, and Kain (2005) 6 0.035 0.000 10
Chingos (2012) 6 0.000 0.000 3
Hoxby (2000) 6 0.000 0.000 n/a
Rivkin, Hanushek, and Kain (2005) 7 0.000 0.000 10
Chingos (2012) 7 0.000 0.000 3
Dee & West (2011) 8 0.025 0.025 10
Chingos (2012) 8 0.000 0.000 3
Average for all grades 0.060 0.061 7.826
Average for K-3 0.113 0.109

182
APPENDIX B: Additional Information on Cost Methods
This appendix is included to provide additional information on how average class sizes
are measured, how I estimate costs of various policy simulations, and how I project policy
effects. In the first section, I describe how I use the administrative data to estimate average class
sizes and then briefly describe how LAUSD class sizes changed over the course of the Great
Recession. I then explain how I estimate the costs of class size reduction under various policy
designs. In the third section of this appendix, I provide greater detail on how I estimate the costs
of increasing teacher salaries. In the final two sections, I describe my methods for projecting the
policy effects associated with class size reduction and teachers’ salary increases, respectively.
Measuring Class Sizes
Because several of the policy proposals estimated in this study require LAUSD to
decrease class sizes by hiring additional staff, and in some proposals to pre-recession levels, it is
important to correctly measure class sizes before the recession (2007-08), during the
recessionary era (2008-09 through 2011-12) and after the budget crisis eased (2012-13). The data
described in the main text allow me to estimate average class sizes for all grade levels and
schools. Elementary students and teachers are observed in the data three times per year, in
accordance with academic terms for the fall, winter, and spring. I measure elementary class sizes
annually for each teacher as the average class size at each of these three different time periods.
Measuring class size in middle and secondary schools is less straightforward because
classrooms are not self-contained and students’ schedules change between the fall and spring
semesters. The data include outlier classrooms in activity-related departments that are extremely
small (less than five students) or extremely large (more than 50 students). I therefore exclude
classrooms in the non-academic departments “School Service,” “School Offices,” and “Driver

183
Education” and for similar reasons further restrict the sample to class periods zero through
nine.
73
Once these departments and class periods are excluded, I use the average class size across
both fall and spring semester for each year, grade level, and school type. I measure the average
class sizes across all schools and grades, and then disaggregate average class size between grades
K-3 and all other grades and between lower-performing schools (those with an Academic
Performance Index in the bottom 20
th
percentile) and all other schools.
74
The district-wide
average class size is calculated as the average size of all classes, giving equal weight to the
number of students in elementary and secondary classrooms.
Class sizes in LAUSD are negotiated as part of the collective bargaining agreement with
UTLA. Language within the union contract allows for class size increases during fiscal crises,
which became necessary during the period of teacher layoffs. Recall that the district sets lower
class sizes at what it refers to as predominantly Hispanic, Black, Asian, or Other (PHBAO)
schools, defined as schools in which 70 percent or more students identify as a non-White
race/ethnicity (LAUSD, 2014c). Table B1 shows the average class sizes each year by grade
level, department, and school type. Because the first round of layoffs took place at the end of the
2008-09 school year, the first two school years shown in Table B1 (2007-08 and 2008-09) reflect
class sizes before the onset of teacher layoffs. Class sizes increased in almost every grade level,
school type, and department beginning in school year 2009-10. Average class sizes increased by
2.15 students across all K-12 grades in the district, by 3.15 in grades K-3, 1.78 in lower-

73
Class periods refer to blocks of instructional time allocated to each subject in middle and high school grades,
where classrooms are not self-contained. I exclude class periods labeled e.g., “homeroom” or “basketball” because
these class periods are either not official classes (in the case of homeroom) or after school programs. Periods zero
through nine thus refer to all periods during the regular school day, beginning with the first period of the day (zero
period), and what is typically the last period of the day (ninth period).
74
In general, an Academic Performance Index (API) below 800 suggests that the school has not reached proficiency,
as defined by the California Department of Education (California Department of Education, n. d.). Lower-
performing schools in this study are defined in relative terms as the bottom 20th percentile of API within each
school level, across the district. Thus an equal number of elementary, middle, and high schools are included in the
list of lower-performing schools.

184
performing schools, and by 1.80 students in PHBAO schools. The relatively lower increase in
average class size in lower-performing and PHBAO schools resulted from greater access to
categorical funding sources (LAUSD, 2013a).
75

Estimating the Cost of Class Size Reduction
State policy requires LAUSD to rehire appropriately credentialed teachers on the
reemployment list in the reverse order of which they were laid off (most senior laid off teachers
are rehired first). Teachers remain on the reemployment list for 39 months, after which they have
no reemployment rights. In the main text, I discuss which teachers are on the reemployment list
and how I estimate the cost of rehiring these teachers. In this section, I provide further detail and
specific examples of how I estimate: (a) the cost of reducing class sizes to particular levels, (b)
the cost of reducing class size in non-self-contained classrooms, and (c) the amount class size
can be reduced for a given cost, as well as the extent to which class sizes can be reduced after
raising teachers’ salaries by various amounts (i.e., class size reduction for combination policies),
beginning in the 2014-15 school year. Then in the fourth subsection, I present my estimates of
total and per-pupil cost and explain why they differ across policies (shown in Table B3). Finally,
I describe how I use these methods to estimate the cost of the district’s final class size reduction
agreement.
Estimating the cost of reducing class size to particular levels. I use a three-step
process to estimate the total cost of reducing class sizes to one of the four levels outlined in
Table 1 of the main text. First, I estimate the number of teachers that need to be rehired for each
policy. I then calculate the total compensation (i.e., salaries, benefits, and pension contributions)

75
In middle and high schools, the proportion of the school day in which teachers are teaching also contributes to the
magnitude of the impact of layoffs on class size increases. Throughout the period of layoffs, the proportion of the
day teachers spent teaching increased, which helped lower the impact of layoffs on class size increases. For
example, in 2008-09 the average high school teacher taught for 68.9% of the day and by 2012-13, the average high
school teacher taught for 74.9% of the day.

185
of these teachers.
76
Last, I project costs over a seven-year period. To explain how this works, I
provide an exposition of the costs associated with just one of the policies discussed above: hiring
enough teachers to return the district to pre-layoff levels of class size, assuming the district
wishes to focus on reducing K-3 class sizes.
77

Returning class size to pre-layoff levels in kindergarten through third grade requires
hiring a total of 1,315 teachers. To compute this value, I first calculate the ratio of 2008-09 class
sizes to 2012-13 class sizes for grades K-3. The average class size in theses grades went from
18.97 in 2008-09 to 22.12 in 2012-13, so the ratio is equal to 1.166 (22.12 / 18.97 = 1.166). The
second step is to subtract one from this ratio and multiply by the number of K-3 teachers
employed during the 2012-13 school year, which is 7,914. I do this because it is mathematically
equivalent to restoring pupil-teacher ratios to pre-layoff levels and this method takes into account
the declining enrollment in LAUSD. The following equation gives the number of teachers
required to restore class sizes in just grades K-3 to pre-layoff levels: (1.166 – 1) * 7,914 =
1,315.
78

In order to hire 1,315 teachers, the district would have to rehire all teachers on the rehire
list with credentials to teach at the elementary level (103) plus an additional 1,212 teachers. The
total cost to rehire the 103 elementary teachers on the rehire list is equal to the sum of their total

76
I exclude the cost of interviews, recruitment, screening, and induction mentoring and professional development,
which will underestimate the costs of class size reduction. Milanowski and Odden (2008), for example, estimate that
the central office and school staff time required for each new teacher hire costs approximately $1,136 and $2,588
respectively (Table 3 of Milanowski & Odden, 2008), and a total of $22,434 when including induction mentoring
and ongoing professional development over the first five years of employment. Thus my cost estimates of large
scale class size reduction may be underestimated when considering additional costs related to newly hired teachers.
77
Throughout this discussion, I report the number of rehired teachers in whole numbers; however, all calculations
are made based on full-time equivalent (FTE) teachers and due to rounding of average class sizes, the numbers I
discuss may be one teacher off from what the arithmetic shown would suggest.
78
Formally, the number of teachers in year t required to return class size to the year t-1 levels is given by the
following equation: ([Class size
t
/ Class size
t-1
] -1 ) * Teachers
t
. I cannot simply subtract the number of teachers in
2012-13 from the number of teachers in 2007-08 because enrollments in LAUSD decreased over this time period.
Therefore, simply hiring teachers back to pre-layoff levels would actually reduce class sizes below their 2007-08
levels.

186
yearly compensation, $7.4 million. The total cost to hire the remaining 1,237 teachers is the sum
of the newly hired teachers’ total yearly compensation. The average salary of the first 1,237 on
the new hire database is $73,465 and the total cost of rehiring these teachers is $89.1 million for
the first year. Adding the cost of the 103 teachers on the reemployment list brings the total cost
to $96.5 million ($7.4 million plus $89.1 million). Therefore, the total cost to restore class sizes
to pre-layoff levels for grades K-3 (by hiring 1,315 teachers and lowering average class size by
3.15 students) is $96.5 million in the first year of implementation.
The final step in estimating the total cost of reducing class sizes is to project the first year
of costs forward over the next seven years. Present value costs of class size reduction change
over time for two reasons. First, future costs are discounted to their present value. Because
dollars today can be invested to yield a positive return in future years, under typical conditions
money today is more valuable than the same amount of money in the future. Conversely, costs
borne in the future are preferred to costs incurred today (Levin & McEwan, 2000). I discount
future costs at 3%, which is in line with cost-effectiveness literature and roughly approximates
the public return on dollar resources (Boardman, Greenberg, Vining, & Weimer, 2010).
79
Thus
the first year costs of hiring 1,315 teachers is $96.5 million, but the second year cost, discounted
to present value is $93.7 million, and the present value cost in the seventh year is $80.8 million.
Discounting over seven years lowers the yearly total cost from $96.5 million to $88.5 million.
The second reason costs of reducing class sizes vary over time is because the large
number of teachers hired in the first year gain experience and education credits over time and as

79
Future costs are discounted to present value costs using the following formula: present value costs in year t =
future cost in year raised to the power of (t + n) * (1+ δ)
n
, where δ is the discount rate and n is the number of years
into the future that costs are incurred. For example, a policy that will cost $1,000 in five years from today would be
discounted using the formula: 1,000 / (1 + d)
5
, where d is the discount rate. Using a discount rate of 3% would imply
that a cost of $1,000 five years from today has a present value cost of $863. As an additional specification check, I
also calculate costs assuming a 5% discount rate and assuming a 0% discount rate (no discounting).

187
a result earn higher salaries. I calculate costs of class size reduction allowing for newly hired
teachers to a higher salary over time as they gain experience and education credits, for each year
that costs are projected forward. This process requires assumptions about the number and type of
newly hired and currently employed teachers that would leave the district at the end of each year.
First, I assume that in the absence of any policy intervention, the average experience level of the
district’s teacher workforce is stable over time at its 2012-13 level of 11.51 years (and 13.00
years for all teachers in grades K-3).
80
I also assume that through natural attrition and gains in
experience over time, the workforce gradually returns to this “steady state” level of experience
whenever there are a large number of teachers added to the workforce. These two assumptions
allow me to model changes over time in the overall districtwide experience and education level
of the teaching workforce, as well as the overall average experience level in just the lower
grades, in lower-performing schools, and in lower grades in lower-performing schools (i.e, for
each of the class size reduction policy simulations).
To see how these assumptions work in practice, I describe here the projected costs of
hiring 1,319 teachers to restore just K-3 class sizes. First, given the average experience level of
teachers on the reemployment list and the “new hire” database (1.90 years), hiring 1,319 teachers
would initially lower average experience for teachers in grades K-3 from 13.00 years to 11.42
years.
81
Each subsequent year, average experience would increase as all of the newly hired

80
Thus retirements and natural turnover are forecasted by assuming that the district replaces retiring teachers and
those that leave the district for other reasons with new teachers. However, in the absence of a policy intervention,
average experience across the district is stable over time because all remaining teachers gain an additional year of
experience each school year and teachers’ growth in experience over time offsets the decrease in experience that
results from retirements.
81
This figure can be calculated by dividing the total cumulative years of experience for all K-3 teachers, by the total
number of K-3 teachers. I use the following equation to calculate the new average experience for all K-3 teachers,
after hiring 1,319 teachers: [(current average experience * current number of teachers) + (average experience of
newly hired teachers * number of newly hired teachers)] / (current teachers + newly hired teachers). For K-3 class
size reduction in which 1,319 teachers are hired, this equates to: [(13.00*7,914) + (1.90*1,319)] / (7,914 + 1,319) =
11.42. In the second year of implementation, I make the same calculation except that newly hired teachers now have

188
teachers gain an additional year of experience and commensurate education credits. In the second
year of implementing K-3 class size reduction, for example, average experience in the district
would increase to 11.56, and by 2020-21, the year LCFF is expected to be fully implemented,
average experience for K-3 teachers would be 12.27.
82
I use predicted salaries at each level of
experience, shown in Appendix Table A2, to estimate the additional compensation teachers
receive for each additional year of experience.
83
As shown in Table A2, the increase in average
compensation associated with moving from zero to one year of experience is $3,278. This figure
is larger than the amount reported in district’s the salary schedules because it reflects teachers’
total compensation, rather than just base salary, and because it includes increases in
compensation related to additional education credits. The salary increase from 1 to 2 years of
experience is $3,133, and the weighted average salary increase for each level of experience is
$3,003 (weighted by the experience levels of newly hired teachers). Accounting for changes in
experience increases the total present value yearly cost of K-3 class size reduction from $96.4
million to $108.3 million.
Table B2 shows how average experience and the costs of class reduction change over
time, using as an example, the policy outlined above in which all K-3 class sizes are restored to
pre-layoff levels. The first column shows the school year, the second column shows the number

2.90 years of experience. Over time, the average experience of all K-3 teachers approaches to original level of 13.00.
It is necessary to calculate changes in average experience for all K-3 teachers because when the average experience
returns to its steady state, I assume that average experience is stable, and no additional costs are incurred as a result
of teachers gaining years of experience.
82
Note that teachers with fewer years of experience tend to have higher than average attrition rates (e.g, Ingersoll,
2001), and my method of projecting changes to districtwide experience does not directly take this into account.
Higher attrition of newly hired teachers would slow down the rate at which average districtwide experience returns
to the steady state. However, when I correct for differential attrition by experience (based on pre-layoff trends), the
changes to my projections of districtwide average experience over time is trivial, so I ignore the potential effect of
differential attrition rates of newly hired teachers. In effect, I assume that following a class size reduction and
widescale hiring of teachers, and recently hired teachers that leave the district are replaced with teachers who have
the same level of experience as their replacement.
83
I generate predicted salaries by fitting an OLS model that uses educational attainment, endorsement area, and
years of experience and it’s square to predict salaries for all teachers in my sample.

189
of teachers in grades K-3, and the third column shows the number of teachers hired each year.
Prior to implementation of the class size reduction, there are 7,914 K-3 teachers, and in each year
after hiring the initial 1,315 teachers for class size reduction, there are 9,229 teachers. The fourth
column shows the average experience for all K-3 teachers, which starts at 13.00 years, drops
immediately to 11.42 years of experience in the first year of implementation (when many early
career teachers are hired), then increases gradually over time. In the final four columns of Table
B2, I show how the costs of restoring class sizes in grades K-3 change over seven years, and the
bottom row shows the average costs per year. Column 6 shows undiscounted costs without
adjusting for changes in teacher experience (which average to $96.5 million), Column 7
discounts costs to present value, Column 8 shows undiscounted costs that are adjusted for
changes in average teacher experience, and the final column shows the adjusted and discounted
costs. I use in my final calculation of cost-effectiveness ratios the adjusted and discounted
estimates, which equate to an average total yearly cost of $98.9 million.
Estimating the cost of reducing class size in non-self-contained classrooms.
Classrooms in elementary schools (grades K-5, and in some cases, grades K-6) are considered
“self-contained” because teachers spend the entire day with students. In contrast, classrooms in
middle and high schools (grades 6-12) are considered non-self-contained because teachers only
teach each student for a portion of the day. Because teachers are given a planning period in
which they are not actively teaching students, the process for calculating the number of teachers
that need to be rehired to restore class sizes to pre-layoff levels in grades 6-12 includes a slight
modification. Most schools in the district have six to eight class periods, and teachers teach an
average of five classes during the school day. As a result, more teachers need to be rehired to
lower class size in grades with non-self-contained classrooms, compared to grades with self-

190
contained classrooms. Based on the number of class periods per day each teacher was assigned,
and the total number of class periods in each school, teachers in grades 6-12 spent, on average,
79.3% of the day teaching during the 2012-13 school year. This figure is approximately
equivalent to teaching an average of 5.5 class periods in a seven-period school day.
84
Therefore,
each teacher the district hires in non-self-contained grades requires an additional 19.7% of a full-
time equivalent (FTE) teacher to make up for the planning periods. The cost of class size
reduction in non-self-contained classrooms is therefore 19.7% higher than in elementary grades,
all else equal. Importantly, teachers in both middle schools and high schools experienced an
increase in the average proportion of the school day spent teaching during the recessionary
period, which lowered the impact of layoffs on class size increases. In my calculations, restoring
class sizes to pre-layoff levels requires only restoring average class sizes and not lowering the
proportion of the day spent teaching.
To see how I make this adjustment for planning periods, I provide an example of how I
calculate the number of teachers that must be rehired to lower class size to pre-layoff levels in
grades 6-12. Average class sizes in grades 6-12 increased from 26.79 to 28.50 during the period
of layoffs, so the ratio of class sizes over this period is about 1.06 (28.50 / 26.79 = 1.0639).
Given that there were 10,188 teachers in grades 6-12 during the 2012-13 school year, and using
the methods described previously, 651 additional teachers would need to be hired to restore class
sizes in grades 6-12 to the pre-layoff level (10,188 * [1 - 1.0639] = 651). However, because these
teachers will only teach a portion of the day, I adjust this figure up by an additional 20.7%, so

84
The district’s collective bargaining agreement requires that middle and high school teachers be assigned at least
five planning periods per week, but the number of class periods per day differs across schools.

191
that the final estimate is 786 additional teachers.
85
Thus the total number of teachers that need to
be rehired to restore all class sizes to pre-layoff levels is 2,306, which is found by summing the
total teachers needed to restore class sizes in grades K-3 (1,315), grades 4-6 (205), and grades 6-
12 (786). Hiring this number of teachers reduces the average class size across the district by 2.15
students.
86

Estimating the amount class size can be reduced for a given cost. The method for
estimating the extent to which class sizes can be reduced if the district allocates all additional
$195 million in new funding to class size reduction is similar to the one described above for
restoring class sizes to particular levels. To demonstrate how these calculations are made, I
describe how I estimate cost for just one policy variation: allocating all additional funding to
class size reduction for all grade levels. I first estimate the total number of teachers that could be
hired using these funds. I assume that all 444 teachers on the rehire list would be rehired (at a
cost of $32.6 million in the first year) and that the residual $162.4 million would be used to hire
additional teachers. As before, I use the “new hire” database to identify the compensation of
newly hired teachers. Given the average salary of teachers on the rehire list of $73,457, a total of
2,211 teachers could be hired using this amount of funding, so when including the teachers
rehired off the reemployment list, the district could hire an additional 2,655 teachers for the
2014-15 school year using the approximate $195 million it received in new funding. The
discounted average yearly cost over seven years of this spending policy would be $178.8 million.
When I account for the fact that these teachers gain an additional year of experience each year,

85
I cannot simply use pupil-teacher ratios for these calculations because at least one of my policy simulations
requires restoring class sizes in upper grades to a particular level (the Odden-Picus funding model, which suggests
25 students per classroom in grades 4-12). Pupil-teacher ratios cannot be directly translated to average class sizes.
86
This figure is the average change in class size in grades K-6 (self-contained) and 6-12 (non-self-contained),
weighted by the number of students in 2012-13 in each of these grade ranges. Similar calculations for lower-
performing schools are weighted by the number of students in each grade range in lower-performing schools.

192
and earn higher salaries and pension contributions, the long-term yearly cost of hiring these
teachers is $218.9 million (before discounting), which is discounted to a present value cost of
$208.5 million per year.
The final step in measuring the amount class size can be reduced using all $195 million
of additional funding requires calculating the extent to which class size would be reduced if
2,655 teachers are hired. Above, I showed that the district could restore average class sizes
districtwide to pre-layoff levels by hiring 2,306 teachers, which would lower average class size
by 2.15 students. I use this ratio to calculate the extent to which average class sizes would be
reduced if 2,655 teachers are hired. Thus, for district-wide class size reduction, hiring 2,655
teachers reduces average class size by 2.48 students.
87
If the district were to allocate all $195
million to just K-3 class size reduction, it would first need to hire the 103 elementary teachers on
the reemployment list at a yearly cost of $7.4 million in the first year. The remaining $187.6
million could then be used to hire 2,544 teachers for a total of 2.657 additional teachers, which
would lower average class sizes in grades K-3 by 6.37 students.
88

Final estimates of the cost of alternate class size reduction policies. The costs of all
class size reduction policies are summarized in Table B3. The first three columns show the class
size policy and the grades and schools targeted. The fourth column shows the number of students

87
That is, (2.15/2,306) * 2,655 = 2.48. Note that this approach implicitly assumes that the district would use the
$195 million to reduce class sizes along the same pattern used in the first policy in which class sizes are restored to
pre-layoff levels. That is, given the changes in class size that took place during the recessionary period, 60.6% of
newly hired teachers would be allocated to grades K-3 (1,509 teachers), 9.4% to grades 4-6 (235 teachers) and the
remaining 30.0% to the non-self-contained classrooms in grades 6-12 (793 teachers). This is important because my
cost methods allow for the cost of class size reduction to vary across grade levels, so any policy in which class size
reduction occurs across all K-12 grades should reflect the same pattern of class size reduction across grades.
88
Note that slightly more teachers can be rehired using the $195 million of additional funding if the district focuses
on class size reduction in grades K-3. This is because elementary teachers on the reemployment list, on average,
earn lower salaries than secondary teachers on the reemployment list ($71,559 in total compensation, compared to
$73,969 for secondary teachers). Once all elementary teachers on the reemployment list are hired (103 total) the rest
of the newly hired teachers are employed from the new hire database, on which teachers earn an average total
compensation of $73,457. If using the $195 million for class size reduction in grades K-3 only, the district would be
required to hire the 341 secondary teachers on the reemployment list.

193
affected by the policy. For example, across-the-board spending policies involve all 517,180
students enrolled in the district in 2012-13, while policies targeted at lower-performing schools
involve 93,439 students, slightly fewer than 20% of enrollment. The fourth and fifth columns of
Table B3 show the total yearly cost and the total yearly cost per student to the school district for
each class size reduction strategy. Lowering all class sizes to pre-layoff levels would require an
expenditure of $173.6 million per year or $336 per student per year, while restoring class sizes in
just lower-performing schools would cost $35.1 million or $337 per student per.
Differences in the patterns of class size increase during recessionary period lead to
differences in the per-pupil cost of restoring class sizes to previous levels. Since the 2008-09
school year (the first year of teacher layoffs), class sizes in lower grades increased by a larger
amount than in upper grades; therefore, the per-pupil cost to restore class sizes in lower grades is
larger than for all grades. The increase in K-3 class sizes was similar between lower-performing
schools and all other schools; however, lower-performing schools saw a relatively greater
increase in grades 4-6, whereas class size increases were relatively more concentrated in middle
and high school grades in all other schools (where class sizes on average and are larger than in
elementary grades in both lower-performing schools and all others). Note that all else equal,
class size reduction is more expensive on a per-student basis at lower average class sizes
(Riechardt, 2015). This basic rule of thumb and the pattern of class sizes increases during the
recessionary period explain why restoring class sizes is more expensive on a per-student basis for
lower-performing schools than for all schools and for lower-grades compared to all grades.
The second four rows in Table B3 show the costs of reducing class sizes to levels
recommended in an Evidence-Based (EB) funding model (Odden & Picus, 2013). The total and
per-pupil costs of class size reduction for the EB model are higher than the costs of restoring

194
class sizes to pre-layoff levels because the Odden-Picus funding model recommends class sizes
that are below the pre-recessionary levels in LAUSD. The costs of restoring class sizes by only
hiring previously laid off teachers with above average value-added scores are shown in the next
four rows. Because the reemployment list for 2013-14 consisted of only 444 teachers, 78 of who
have value-added measures available (with 33 teachers showing above average measures), the
selective reemployment policy results in very minor class size reduction and similarly low total
and per-pupil costs. The last four rows in Table B3 show the costs of restoring class sizes if the
district allocated all of the additional $195 million in additional funding. Note that the total cost
of this policy exceeds $195 million because after this funding is allocated to hiring teachers, the
cost of these teachers increases as they gain teaching experience and earn higher wages (which I
described above). Along those same lines, the district could hire more teachers if allocating all
new funding to just the lower grades because teachers in self-contained classrooms do not have
planning periods. The total yearly cost of allocating all funding to reduce class sizes in just the
lower grades is therefore slightly larger (by about $20,000) because a greater number of teachers
would be gaining experience over time.
The final two columns in Table B3 show the amount that each policy would reduce class
sizes and the per-pupil cost for each one-student reduction in class size. The cost of a one-
student reduction in class size differs across policies for several reasons. Frist, class size
reductions are more expensive on a per-student basis when either the initial or final average class
size is lower, all else equal. As a result, policies that require larger decreases in class size, all else
equal, will have higher costs for each one-student reduction. Second, different sets of teachers
are rehired from the reemployment list. For instance, targeting class size in grades K-3 implies
that only the elementary teachers on the reemployment list would be rehired. Compared to

195
secondary teachers, elementary teachers on the reemployment list have less experience on
average earn slightly lower salaries ($73,969 for elementary and $71,559 for secondary, although
currently employed elementary teachers have slightly higher levels of experience compared to all
employed teachers in the district). Similarly, above high value-added teachers have lower total
average total compensation than the average for all other teachers on the reemployment list. Last,
as mentioned above, a greater number of teachers must be hired in non-self-contained classrooms
for the same level of class size reduction because teachers have planning periods throughout the
week.
Estimating the cost off the district plan to reduce class size. Given the current average
class sizes (shown in Table B1), the district would not need to spend additional money to achieve
the mandated maximum average class sizes, as they are described in Section 1.5 of the new
agreement with UTLA. Importantly, the class sizes listed in the agreement may require
rebalancing of teachers in particular schools, but given the districtwide average class sizes in
2012-13, no additional teachers would need to be hired to meet the new mandated maximum
average class sizes. However, part of the district’s final agreement also included a two-student
reduction in all math and ELA courses in grades 8 and 9. The 2012-13 average class size in
grade 8 and 9 math courses was 26.3 and there are 677 teachers who are assigned to these
courses. Given the ratio of class sizes of 1.08 (26.3 / 24.3 = 1.081), the district would need to
hire about 55 FTE teachers for grades 7 and 8 class size reduction in math classes. Because these
teachers would be assigned to non-self-contained classrooms and would be allocated planning
periods, the final adjustment (discussed above) increases the number of newly hired teachers to
66. The 2012-13 average class sizes in English for grades 7 and 8 were 26.7, which included 665
teachers, so using the same calculations, the district would also need to hire 66 teachers to reduce

196
seventh and eighth grade English classes by two students.
89
Given that the reemployment list for
2013-14 hiring includes 341 secondary teachers, all 132 math and English teachers could by
pulled from the reemployment list of previously laid off teachers.
The sum of the total compensation of the 132 math and English teachers is $9.5 million,
and when I factor in the increases in experience over time (using the methods described
previously), the total first year cost before discounting is $10.6 million. In sensitivity analyses, I
recalculate the cost of the district’s class size reduction plan by altering the salary of newly hired
teachers. If the teachers hired had the average salary equal to teachers currently employed in the
district with zero years of experience and a bachelor’s degree, the total undiscounted cost would
be $9.8 million. If newly hired teachers had the salary equivalent to teachers with five years of
experience and a master’s degree, the total cost would be $12.3 million.
90
Accounting for gains
in experience over time and discounting to present value increases the total cost to $10.7 million.
Comparing the district’s plan to alternative class size reduction plans solely on the basis
of total and per-pupil cost provides some but not all of the necessary information for assessing
their efficiency. Given current and past budgetary constraints in LAUSD, it is important to take
into account the per-pupil costs of policies before implementation. However, without examining
measures of effectiveness, it is impossible to assess the relative cost-effectiveness of alternative
policy options. Because the aim of this study is to combine measures of cost and effects to gain
insights into the most efficient use of new funding, I combine all cost estimates with measures of

89
Note the exact number of FTE math and English classes are actually 66.25 and 65.97, respectively, and I use these
more precise estimates in all calculations of cost.
90
As noted previously, the LAUSD administrative data for students in middle and high school includes individual
observations for each course that each student takes each semester. I am therefore able to determine which courses
are coded as academic and which are coded as non-academic. The final UTLA-LAUSD agreement (LAUSD, 2015d)
defines academic courses as those in the following departments: English, English as a Second Language, Reading,
Math, Social Studies, Science, and Foreign Language. Non-academic courses include those in the departments of
Art, Business Education, Computer Science, Environmental / Agricultural Education, Health Careers, Health
Education, Home Economics, Industrial Education, Military Science, Music and Theatre. No Class size caps are
placed on classes in the Physical Education, Band, or Drill Team departments.

197
effectiveness based on extant literature. These results are discussed in the main text. In the
section below, I describe the methods used to estimate the cost of the district’s compensation
plan, as well as a set of alternative policy options for raising teacher salaries.
Estimating the Cost of Raising Teacher Salaries
In the following four subsections, I explain how I estimate costs of increasing teachers’
salaries by specified amounts and by using a set amount of funding. Then in the third subsection,
I present my estimates of the total and per-pupil cost of alternate compensation policies and
explain why they differ across policies (shown in Table B4). Finally, I use these same methods to
show how I estimate the costs of the district’s final compensation plan.
Estimating the cost of increasing salary by various amounts. As noted in the main
text, increases in salary only affect variable benefits and base salary, while fixed benefits remain
constant at $17,134 per year in FY 13-14. Therefore, to estimate the costs of salary increases, I
first calculate the sum of base salaries and variable benefits for each teacher group (novice
teachers and all teachers; teachers in lower-performing schools and all schools). The sum of base
salaries and variable benefits for these teacher groups is: (a) $1.7 billion for all 22,258 classroom
teachers in 2012-13, (b) $320.4 million for the 4,205 teachers in lower-performing schools, (c)
$73.4 million for the 1,267 novice teachers, and (d) $20.2 million for the 390 novice teachers in
lower-performing schools. To estimate the first year cost of each salary increase, I multiply the
sum of base salaries and variable benefits by the percent salaries are increased. For example, the
cost of the union’s proposed salary increase of 8.5% (implemented over two years, with 4.25% in
the first year and the remaining 4.08% in the second year) is equal to about $73.1 million in each
of the first two years ($1.7 billion * 0.085, divided over two years) for a total of $146.3 million.
After discounting these estimates to present value and averaging over seven years, the total cost

198
of the 8.5% salary increases is $123.6 million, or $239 per student per year.
91

Estimating the cost of salary increases using a set amount of funding. I also consider
the cost and effects of allocating all additional funding the district will receive through
Proposition 30 and LCFF allocations to increase teachers’ salaries. To measure the cost salaries
can be increased for a given spending amount, I divide the total amount allocated for salary
increases by the sum of the salaries and variable benefits. For example, allocating all $195
million to increase salaries for all teachers would raise salaries by approximately 11.3% ($195
million / $1.7 billion), at a cost of $346 per pupil. If the $195 million is allocated to just novice
teachers or low-performing schools, the total cost remains constant, but the percent increase in
salaries and the per-pupil cost are both higher (these examples are included in the main text and
highlighted here for clarity). For each of these salary increases, I assume no transaction costs,
such as administrative personnel time or legal fees.
To measure the cost of policies that combine salary increases with class size reductions, I
can simply add the cost of each individual salary increase policy to each individual class size
reduction policy in most cases. The only exceptions are policies that increase salaries by
specified amounts and use the remaining amount of funding for class size reduction. For these
policies, the amount of additional costs associated with experience gains over time of newly
hired teachers depends on the number of newly hired teachers, which is dependent on the
remaining funding after each salary increase policy. An example helps clarify this calculation. If
the district provided a 6% raise for all teachers, $91.8 million would be leftover from the $195

91
LAUSD administrators have noted in personal conversation that most district labor contracts contain provisions
that require any salary increases for one labor unit to apply to all labor units (what the district calls “me too”
provisions). Administrators in the budget and finance department use a general rule of thumb that a 1% increase in
salaries for all labor units in LAUSD costs about $40 million. Thus a 4.25% raise would cost $170 million in the
first year of implementation. In my baseline analyses, I consider salary increases for only classroom teaching staff
and then assess how the results would change under LAUSD’s calculation of the cost of salary increases.

199
million in additional funding, which could be used to hire 1,249 teachers (and lower average
class size by 1.17 students). Hiring this number of teachers would lower average experience in
the district from 11.51 to 10.99 in the first year of implementation. As the newly hired teachers
gained experience over time, the district would incur additional costs over seven years that
would total an additional average yearly cost of $11.3 million over and above the $195 million.
The total cost of this combination policy, before discounting to present value, would be $206.3
million per year over seven years. Discounting this figure to present value reduces it to $188.7
million
Final estimates of the cost of alternate compensation policies. Table B4 shows the cost
of each salary increase policy. The first three columns show the particular compensation policy
and the fourth row shows the number of students impacted.
92
The next two columns show the
total and per-pupil cost for each policy and the final two rows show the total amount of salary
increase and the per-pupil cost for each 1% increase in salaries. Not surprisingly, the total costs
of targeted salary increases are far lower than across-the-board salary increases because fewer
teachers receive raises under targeted policies. The per-pupil costs of salary increases are lower
for novice teachers because these teachers have lower average total compensation ($75,122
compared to $94,438 for all teachers districtwide), and therefore a 6% raise, for example,
requires less spending overall. Because class sizes are slightly lower in lower-performing
schools, the per-pupil cost of salary increases in lower-performing schools is slightly higher.


92
Determining the number of students taught by novice teachers in self-contained classrooms (elementary) is
straightforward. To determine whether a student is taught by a novice teacher in the upper grades, where classrooms
are not self-contained, I weighted each student according to the proportion of the school day she or he is taught by a
novice teacher (i.e., if a students spends half the school day in classrooms with novice teachers, that student counts
for 0.5 of a student). The total number of students taught by novice teachers in non-self-contained classrooms is the
sum of these weighted student counts.

200
The last column in Table B4 shows the cost of salary raises per pupil, for each 1%
increase in salaries. Note that regardless of the amount of salary increase, each 1% increase
would cost $28 if allocated to teachers in all schools, $29 in lower-performing schools, and about
$25 for novice teachers in all schools and in lower-performing schools. The one exception is the
set of policies in which all new funding is devoted to increasing salaries in the first year. All
other salary increase policies are phased in over two years, while the final policy would raise
salaries immediately. As a result, the per-pupil cost for each 1% increase is slightly larger
because costs are not spread over multiple years, and all costs are discounted to present value
using a 3% discount rate.
93

Estimating the cost off the district plan to increase teacher salaries. After
approximately two years of negotiations, the district and teachers union reached a final
agreement in April of 2015 that included both class size reduction and a salary increase. The
salary agreement calls for an increase in salary of 10.37%, implemented incrementally over time.
The first pay increase of 4% is retroactive effective July 1, 2014. Because all of the policies
simulated in this study are implemented for the 2013-14 school year, I assume this first pay raise
would begin during what is considered year 1 of all other policies, school year 2014-15. The
second pay raise of 2% was also retroactive at the time of the agreement, and was effective
January 1, 2015. Thus I assume that teachers receive this raise for half of their annual salary
during the 2014-15 school year, and receive the remaining amount in the 2015-16 school year
and every school year thereafter. The next salary increase of 2% was scheduled for July 1, 2015,

93
As with the cost of class size reduction, the cost of raising salaries would grow over time as teachers gain
additional years of experience and move up the salary schedule. However, it is not clear how average experience
might change over time following an increase in salaries. I assume that following an increase in the salary, all else
equal, the district maintains its steady state average experience level of 11.51 years (and 13.00 for teachers in grades
K-3). As a result, the costs of a salary increase do not grow over time in the same way as a large-scale class size
reduction.

201
so I model this raise as beginning in the 2015-16 school year. The final pay increase of 2% was
set for January 1, 2016, so I again assume that teachers receive this salary increase for half of
their annual salary during the 2015-16 school year, and receive the remaining amount in the
2016-17 school year and then every year after.
While the total pay increases sum to 10%, the parties agreed that each salary raise would
compound over the original amount (LAUSD, 2015d). For example, the second salary increase
of 2% would be based on salaries that had already been increased by 4%. Thus the 2% raise
effective January 1, 2015 would effectively increase salaries by 2.08%, compared to the salaries
in place prior to the initial 4% raise. As a result, the total increase in pay from the original
amount would total 10.37%. Because each pay increase is applied to all 22,258 teachers
employed in 2012-13, the total undiscounted cost of the total salary increase is $178.3 million.
When this cost is discounted to present value projected over seven years, the total average yearly
cost is $75.5 million per year.
To summarize, after about two years of negotiations between UTLA and LAUSD that
ultimately required a third part mediator, the final settlement involved both class size reductions
and an across-the-board salary increase. In total, approximately $10.6 million were allocated to
class size reduction and $75.5 million to salary increases. This spending agreement represented a
compromise from what each of the two parties originally proposed. For example, the agreed-
upon salary increase was significantly lower than the union’s original proposal of a 17.6%, but
much greater than the district’s first offer of a 6% raise. This final spending plan provides a
useful comparison for each of the alternate budgeting options that I simulate in this study.
In the next two sections, I provide additional information on how I project the effects of
class size reduction policies and teachers’ salary increase policies, respectively.

202
Estimating the Effects of Class Size Reduction
Estimating the main effects of class size reduction. Scholars argue that smaller class
sizes benefit students by creating more intimate environment, providing the teacher with more
time for each individual student (e..g, Achilles, 2012; Krueger, 1999). I refer to the direct
impacts of lower class sizes as the main effects and differentiate these from the impacts of class
size reduction on the quality of the overall teaching workforce (which I describe below). As
stated in the main text, I project the main effects of class size reduction in LAUSD by using the
average of the effect sizes found across the 25 estimates shown in Table A3. I also average the
95% confidence intervals for each estimate, which give upper and lower bounds for the
estimated effect of class size reduction. Estimates from these studies suggest that a one-student
reduction in K-3 class sizes increases student achievement by 0.014 standard deviations (SD),
while the average effect across all grade levels is 0.007 SD.
94
Thus restoring all class sizes to
pre-layoff levels, thereby reducing the average class size by 2.15 students, would increase
student achievement by an estimated 0.015 standard deviations. For perspective, Lee and Finn
(2010) estimate that an effect size of 0.1 is equivalent to approximately 2.8 months of learning
for the average K-12 student, so an effect size of 0.015 implies approximately 0.4 months of
additional learning per year, or about eight and a half days assuming 20 schools days per month.
Some studies of class size reduction find larger effects for low-income students, students
of color, and students in urban schools (Angrist & Lavy, 1999; Dee & West, 2011; Krueger,

94
Other meta-analyses of class size reduction have found that a one-student reduction in class size has a greater
effect when the initial class size is smaller. Glass and Smith (1979) conducted a meta-analysis of class size reduction
and found that, across 80 total studies on class size reduction, the effect on achievement can be summarized in the
following equation: 
S – L
=  log
e
(L/S), where 
S – L
is the effect on student achievement of a change in class size
from a large class size of L students to a small class size of S students, and  is a constant showing the linear effect
of a one-student reduction class size. Glass and Smith assigned  an average value for math and reading of 0.30. For
instance, a reduction in class size from 31 to 30 results in an effect size of 0.0098, while reducing class sizes from 21
to 20 gives an effect size of 0.0146. I use this alternative measure of the effects of class size reduction on student
achievement in a sensitivity analysis.

203
1999; Summers & Wolfe, 1977). Few studies disaggregate class size effects by ethnicity beyond
simple binary comparisons of Black and White, or non-White and White; however, most studies
report differences in class size effects for low-income students, compared to all other students
(Chingos, 2013; Krueger, 1999; Nye, Hedges & Konstantopulos, 2000). As one of the only
large-scale randomized experiments of class size ever undertaken, the results of the Tennessee
STAR experiment, as reported in Krueger (1999), provide arguably the most definitive evidence
of heterogeneous effects of class size reduction with respect to family income. The author finds
that an eight-student reduction in class size raised achievement for students eligible for free or
reduced price lunch (FRL) by 0.0110 SD more than for students not eligible for FRL,
representing a 10.2% greater gain from class size reduction. Kruger finds that Black students and
students in inner-city schools gained 0.0476 and 0.0310 SD more than White students and
student from suburban neighborhoods, respectively. On average, students from traditionally
disadvantaged backgrounds made 29.0% greater gains from class size reduction (Krueger, 1999,
also reported in Chingos, 2013). I therefore assume that each one-student reduction in class size
targeted to all grades in lower-performing or PHBAO schools raises achievement by 0.009 SD
(0.007 + 0.007 * 0.290 = 0.009) and by 0.018 SD if targeted to just the lower grades within those
schools. In alternative specifications, I assume that class size effects are log-linear (rather than
linear) across various initial class sizes by averaging estimated effect sizes of a 1% decrease in
class size (rather than a one-student decrease). Class size effects for various grades and student
subgroups are summarized in Table A3.
These estimated effects assume that there are no major shocks to the local labor market
that would alter the quality of the district’s teaching workforce, a topic to which I turn next.

204
Modeling changes in teacher effectiveness. As described in the main text, one of the
key determinants of the impact of class size reduction is the effectiveness of newly hired teachers
(Chingos, 2013; Jepsen & Rivkin, 2009; Stecher & Sims, 2002). Because of space limitations, I
present only a brief overview of this method in the main text. To further explain how I model the
changes in districtwide teacher effectiveness, I use as an example here the policy in which all
class sizes are restored to their pre-layoff levels.
Assuming the district wants to hire 2,306 teachers in order to restore class size to their
pre-layoff level, the majority of newly hired teachers would be drawn from the local labor
market (with the initial 444 teachers being hired from the reemployment list). As I described
earlier, teachers on the “new hire” database have, on average, 1.9 years of experience. Of the
first 1,862 teachers that would be hired from the new hire database, 56% are first year teachers,
7% are second year teachers and the remaining teachers have between three and eight years of
experience (recall that these figures are based on new hires in LAUSD during the pre-
recessionary period). While the districtwide average measure of effectiveness is zero, first and
second year teachers have, on average, a value-added measure of -0.046 and -0.040 SD,
respectively, and these measures generally increase with greater levels of experience (see Table
C1). The overall average value-added scores for teachers on the reemployment list is -0.019 SD
and the average for teachers on the new hire database is -0.039 SD. All together, the 2,306
teachers that would need to be rehired in order to restore all class sizes have an average value-
added of -0.036 SD and a total value-added of -83.04 standard deviations. Considering the
average value-added for all 22,258 teachers in the district during the 2013-14 school year is zero,
the impact of adding 2,306 teachers would be to lower the average teacher effectiveness by -
0.0034 SD ([-83.04 + 0] / [2,306 + 22,258] = -0.0034) in the first year of implementing a class

205
size reduction. The following year, each of the newly hired teachers gains a year of experience
and their value-added increases by average returns to experience (shown in Appendix Table C1).
Thus while a 2.15 reduction in class size was initially was predicted to raise achievement
by 0.015 SD, my projections suggest that achievement would only increase by 0.012 SD in the
first year of class size reduction (.0150 - .0034 = .0116, as shown in Figure B1). However, the
following year, all of the newly hired teachers would gain an additional year of experience.
According to my results of VAM estimates (which are consistent with the broader literature, e.g.,
Harris & Sass, 2011), these teachers would also improve their teaching effectiveness. Using the
same calculations as before to calculate district-wide VAM, and assuming teachers’ value-added
increases each year by the amounts shown in Table C1, I find that in the second year of class size
reduction, the detrimental effects of hiring 2,306 relatively inexperienced teachers is -0.0029 SD
and the overall impact of the class size reduction that year is 0.0121 SD. Projecting these effects
over seven years, I find that the average yearly effect of hiring the younger cohort of teachers is -
0.0020 SD and the average effect per year of a 2.15 student reduction in class size is 0.0130
(student-level) SD. Figure B1 displays the projected total effect over time of reducing class sizes
by 2.15 students (dashed line), which is the sum of the main effects (solid line) and the
workforce composition effects (dotted line).
Estimating the Effects of Increasing Teachers’ Salaries
In this study, there are two mechanisms through which higher salaries impact student
achievement. The first is through decreases in disruptive teacher turnover (Ingersoll, 2001; Guin,
2004). The second is by increasing the quality of new teacher applicants to the district. I discuss
each of these in the main text, but provide further detail here.

206
Salary effects of decreased teacher turnover. To estimate salary effects through the
first mechanism, I combine estimates of wage-elasticities (the percent decrease in turnover
associated with a 1% increase in wages) with an estimate of the impact of turnover on
achievement. Ronfeldt, Loeb, and Wyckoff (2013) show that achievement on standardized tests
is about 0.073 SD lower in the years with 100% grade-level teacher turnover, compared to years
with zero turnover.
95
My preferred estimate of the wage-elasticity with respect to turnover comes
from Imazeki (2005), who found that a 1% increase in district-wide teacher salary was associated
with a 0.926% decrease in the likelihood a female teacher leaves the district, and a 1.804%
decrease for male teachers.
96
Using a weighted average of the percent of male and female
teachers in LAUSD, this wage elasticity is -1.193. Scaling this parameter down, a decrease in
turnover of 1.193% would raise average achievement by 0.00087 SD. Therefore, a 6% increase
in salary would raise achievement by 0.0052 standard deviations and a 17.6% increase would
raise achievement by 0.0154 SD.
I use several parameters for wage elasticities from other studies as specification checks of
my primary estimate. Clotfelter et al. (2008a) examined the impact of a bonus program in North
Carolina schools in which teachers received a bonus of $1,800 to teach particular subjects in
harder-to-staff schools. Given the range of salaries for teachers involved in the program, this
salary raise represented between a 3% and 7% increase in salaries and on average, the bonus
provided teachers with a 4.2% increase in salaries. The authors found that this increase in
salaries reduced turnover by 17.3% (Model 4 of Table 5). These findings suggest a wage
elasticity of -4.119. Hanushek, Kain, and Rivkin (2004) used data from schools in Texas and

95
This figure is the average effect of teacher turnover on math and English scores (shown in Model 2 of Table 4).
96
This estimate is taken from Table 4 of Imazeki (2005). Using a weighted average for female and male teachers in
LAUSD (69.6 and 30.4% in the 2012-13 school year, respectively), a one percent increase in wage is associated
with a 1.193% decrease in the likelihood a teacher leaves the district.

207
found that a 10% increase in teacher salaries reduced the likelihood that a teacher leaves their
district by between zero and 3.4 percentage points, depending on gender and experience level.
Based on the descriptive statistics from Table 1 in Hanushek, Kain, and Rivkin (2004), this
figure correlates to a wage elasticity of -0.904. Finally, Greenwald, Hedges, and Laine (1996)
used meta-analysis of studies published between 1980 and 1993 to measure the direct effect of
teachers’ salary increases on student achievement. In total, the authors included 13 estimates in
their research synthesis. They found that, on average, raising teacher salaries by approximately
$1,625 in 2015 dollars was associated with a 0.0007 SD increase in student achievement. Based
on the average total compensation of teachers in LAUSD, $94,438, the Greenwald, Hedges, and
Laine study suggests that a 6% increase in teachers’ salaries would increase student achievement
by 0.0024 SD and a 17.6% increase would raise achievement by 0.0072 SD, or about 4.1 days of
learning (Lee & Finn, 2010). These alternate parameters will be used to test the sensitivity of
policy effect projections. All sensitivity analyses are available from the author upon request.
Salary effects of improved quality of teacher applicants. The second mechanism
through which salary increases improve educational outcomes is by attracting more effective
teachers into the profession. As I describe in the main text, based on past work by Figlio (2002;
1997), I assume a 10% increase in salaries would raise the value-added measure of newly hired
teachers by 0.0193 SD, so a 6% salary increase would raise the quality of new hires by 0.0116
SD. I assume that in the absence of any policy, the district hires 1,667 teachers per year with
average effectiveness, to replace the 7.5% that exit through retirements and natural attrition
(which is based on pre-recessionary trends in LAUSD).
I project changes in districtwide teacher effectiveness using the same general methods
described in the section on class size effects. To explain this process more specifically, I use as

208
an example, the impacts on the workforce composition of a salary increase of 6%. In the first
year of a 6% salary increase, teachers hired to replace those that leave through natural attrition
would have an average of 0.0116 SD value-added. The 1,667 newly hired teachers would
increase districtwide average teacher effectiveness from zero to 0.0009 SD ([0.0116 * 1,667] /
22,258). The next year, an additional 1,667 teachers would be hired and average teacher
effectiveness would increase again. I find that a 6% increase in salaries would increase
districtwide teacher effectiveness by 0.0009, 0.0017, and 0.0026 SD in the first, second, and
third years of the policy. Over seven years of implementation, this salary increase would improve
the overall average teacher effectiveness by 0.0039 SD of student achievement, when averaged
over these seven years. For policies that involve both class size reduction and salary increases, I
model both the decline in average teacher experience brought on by the large-scale hiring of new
teachers, as well as the increase in average effectiveness that results from higher salaries.

209

FIGURE B1
Average effect on student achievement of restoring all class sizes to pre-layoff levels, taking into
account both the smaller class size and the impact on teacher experience in the district

Note: restoring class size to pre-layoff levels requires that the district reduce class size by 2.15 students. Measures of
teacher effectiveness are imputed based on predicted values of teachers with similar characteristics. CSR stands for
class size reduction.

210
TABLE B1
Average class sizes in LAUSD, 2007-08 to 2012-13
2007-08 2008-09 2009-10 2010-11 2011-12 2012-13
Districtwide 26.2 26.0 26.5 26.3 26.9 27.8
Grades K-6 20.0 20.1 22.1 22.2 22.5 22.7
Grades K-6, Lower-
performing schools
20.0 20.1 21.5 21.7 21.9 22.4
Grades 6-12 27.1 26.8 27.1 26.8 27.5 28.5
Grades 6-12, Lower-
performing schools
26.1 25.5 25.2 24.5 25.2 26.3
Elementary schools (self-contained)
SPED classrooms 10.0 10.2 10.5 11.1 11.3 11.3
Kindergarten 19.0 19.2 22.0 21.9 21.9 22.3
Grade 1 19.0 18.9 21.4 21.5 21.9 22.2
Grade 2 18.7 18.9 21.6 21.5 21.7 21.9
Grade 3 18.7 18.9 21.5 21.6 22.0 22.0
Grade 4 25.5 26.1 26.4 26.3 26.7 27.3
Grade 5 26.3 26.3 26.6 26.7 27.3 27.5
Grade 6 26.9 25.8 25.1 25.3 26.4 26.3
Middle and high schools (not self-contained)
Grade 6 27.5 27.1 26.8 26.6 27.2 28.1
Grade 7 28.3 27.3 26.2 26.3 26.8 28.0
Grade 8 27.1 26.7 26.6 25.9 26.7 27.4
Grade 9 25.6 25.6 26.8 26.2 27.3 28.6
Grade 10 28.1 28.2 28.7 28.2 28.5 29.9
Grade 11 25.8 25.8 28.1 28.1 28.7 29.4
Grade 12 27.0 26.8 27.5 27.1 27.9 28.5
Departments (Grades 6-12)
Math 26.7 26.2 26.2 25.4 25.9 27.1
Science 28.0 27.6 27.4 26.7 27.4 28.5
English 24.1 24.1 25.7 25.4 26.4 27.5
Social studies 28.2 27.6 27.8 27.1 27.7 28.7
Art 30.7 30.0 30.8 31.7 32.6 34.1
Theatre 28.6 27.4 28.3 29.6 30.4 32.4
Music 28.3 28.2 29.0 29.6 30.7 31.4
Eng. as a second language 22.6 21.8 20.0 19.2 18.5 18.4
Foreign language 28.5 28.3 29.2 29.0 30.6 32.2
Physical education 34.9 35.2 34.0 34.7 34.0 34.2
Note: self-contained classrooms are those in which students spend the entire day with the same teacher. The district
has grade 6 classrooms housed in self-contained classrooms (in elementary schools) and non-self-contained
classrooms (in middle schools). Because the first year of layoffs in LAUSD was 2008-09, the first year in which
class sizes were increased as a result of layoffs was 2009-10. I limit the dataset to K-12 district schools and
dependent charter schools that operate within the district’s collectively bargained employee contract for reasons
described in the text. In total, I include 692 (88%) schools out of the 782 schools in the LAUSD administrative
datasets. Of the 154,009 teacher-year observations from all schools, 151,366 (98.3%) are in traditional district
schools and dependent charters, while 2,643 (1.7%) are in non-traditional district schools or independent charters.
Source: Author’s calculations based on LAUSD administrative data.

211
TABLE B2
Costs over time of restoring all class sizes to pre-layoff levels
School year
Total teachers
(all grades)
New
teachers
Average experience
Total cost of newly hired teachers
All teachers Newly hired
2012-13 22,258 0 11.51 -
$0 $0 $0 $0
2014-15 24,564 2,306 10.60 1.90 $169,442,835 $169,442,835 $169,442,835 $169,442,835
2015-16 24,564 0 10.70 2.90 $169,442,835 $164,507,607 $176,366,617 $171,229,725
2016-17 24,564 0 10.79 3.90 $169,442,835 $159,716,123 $183,290,399 $172,768,780
2017-18 24,564 0 10.89 4.90 $169,442,835 $155,064,197 $190,214,181 $174,072,921
2018-19 24,564 0 10.98 5.90 $169,442,835 $150,547,764 $197,137,963 $175,154,527
2019-20 24,564 0 11.07 6.90 $169,442,835 $146,162,878 $204,061,745 $176,025,454
2020-21 24,564 0 11.17 7.90 $169,442,835 $141,905,707 $210,985,527 $176,697,057
Average cost per year $169,442,835 $155,335,301 $190,214,181 $173,627,328
Discounted to present value: No Yes No Yes
Accounting for changes in experience of newly hired teachers: No No Yes Yes
Note: This table shows how the costs of class size reduction are projected over time. The district starts with 22,258 classroom teachers with average experience
of 11.51 year. The average experience of the 2,306 newly hired teachers is 1.90. In the first year of class size reduction, average experience decreases to 10.60. In
each subsequent year, the teachers hired for class size reduction gain a years of experience, which increase districtwide average experience.

212
TABLE B3
Costs of class size reduction policies
Class size reduction policies
Number
of
students
Total yearly
cost
Cost
per
pupil
Amount
of class
size
reduction
Cost per
pupil for
each one-
student
reduction
Restoring class
size to pre-layoff
(2008) level
All
grades
All schools 517,180 $173,627,328 $336 2.15 $156
LP schools 93,439 $35,181,919 $377 1.78 $211
Lower
grades
All schools 175,043 $98,854,673 $565 3.15 $179
LP schools 37,640 $20,067,200 $533 2.99 $178
Class sizes
recommended in
Evidence-Based
funding model
All
grades
All schools 517,180 $396,214,479 $766 3.81 $201
LP schools 93,439 $58,869,402 $630 2.88 $219
Lower
grades
All schools 175,043 $384,245,938 $2,195 7.12 $308
LP schools 37,640 $60,031,138 $1,595 6.97 $229
Rehiring all
previously laid
off teachers
classified as
"highly effective"
All
grades
All schools 517,180 $2,448,646 $5 0.03 $166
LP schools 93,439 $2,387,252 $26 0.12 $219
Lower
grades
All schools 175,043 $1,574,864 $9 0.05 $179
LP schools 37,640 $1,535,386 $41 0.23 $175
Using all
additional
funding provided
through Prop 30 /
LCFF
All
grades
All schools 517,180 $208,571,176 $403 2.48 $163
LP schools 93,439 $208,571,176 $2,232 10.14 $220
Lower
grades
All schools 175,043 $208,592,046 $1,192 6.37 $187
LP schools 37,640 $208,592,046 $5,542 29.45 $188
Note: LP schools stands for lower-performing schools, defined as schools in the bottom quintile of API.

213
TABLE B4
Cost of salary increase policies
Compensation policy
Number
of
students
Total cost
Cost
per
pupil
Amount
of
salary
increase
Cost per
pupil per
1%
increase
in salary
6 percent
All
teachers
All schools 517,180 $87,268,159 $169 6.0% $28
LP schools 93,439 $16,252,234 $174 6.0% $29
Novice
teachers
All schools 24,909 $3,726,378 $150 6.0% $25
LP schools 6,682 $1,022,028 $153 6.0% $25
8.5 percent
All
teachers
All schools 517,180 $123,629,891 $239 8.5% $28
LP schools 93,439 $23,023,998 $246 8.5% $29
Novice
teachers
All schools 24,909 $5,279,035 $212 8.5% $25
LP schools 6,682 $1,447,873 $217 8.5% $25
17.6 percent
All
teachers
All schools 517,180 $255,986,598 $495 17.6% $28
LP schools 93,439 $47,673,218 $510 17.6% $29
Novice
teachers
All schools 24,909 $10,930,708 $439 17.6% $25
LP schools 6,682 $2,997,950 $449 17.6% $25
Using all
additional funding
provided through
Prop 30 / LCFF
All
teachers
All schools 517,180 $178,764,619 $346 11.3% $30
LP schools 93,439 $178,764,619 $1,913 60.9% $31
Novice
teachers
All schools 24,909 $178,764,619 $7,177 265.4% $27
LP schools 6,682 $178,764,619 $26,754 967.7% $28
Note: LP schools stands for lower-performing schools, defined as schools in the bottom quintile of API.

214
APPENDIX C: Estimating Value-Added Measures of Effectiveness
For elementary teachers who teach in self-contained classrooms (with the same group of
students all day), I estimate the following model, for each school year from 2006-07 to 2012-13:
,
where Aijs is a measure of student i’s achievement on either the math or English subjects of the
California Standardized Test (CST), standardized within year and test, Ais(t-1) and the alternate
(“alt”) represent lagged test scores in the current subject (math or English) and in the alternate
subject (math if the model is estimating English achievement, English if the model is estimating
math achievement). Xit represents vectors of student characteristics that include family income,
as measured by eligibility for free and reduced price lunch, English language proficiency,
whether English is spoken in the home, whether the student is enrolled in any additional courses
in math or English (depending on which test score is being predicted), whether the student is in
his or her first year in the district, and dummy indicators for students’ grade level. Finally, js
represents a teacher fixed effect and the error term, εijs, is assumed independently identically
distributed with respect to the other variables in the model. I also test several variations of this
basic model including: 1) pooling all years of data and estimating teacher effectiveness over
multiple years and including year fixed effects (Koedel & Betts, 2007; McCaffrey, Sass,
Lockwood, & Mihaly, 2009); 2) removing the student covariates and adding a twice-lagged
measure of student achievement, as suggested by Rothstein (2010); and 3) adding classroom and
school covariates to control for peer effects (Lubienski & Lubienski, 2013)
Standard testing in California spans from grade two through grade eleven, thus it is
possible to generate VAM for teachers in grades three through eleven (grade two teachers’
students do not have pretests and therefore must be excluded). I test whether the estimates of
A
ijs
 a
1
A
ijs(t 1)
 a
2
A
ijs(t 1)
alt
 a
3
X
ijs
 
js
 
ijs

215
value-added measures are sensitive to the particular students and teachers included in the model.
In particular, I generate VAMs for just students in grades three and four, four and five, five and
six, six and seven, seven and eight, eight and nine, nine and ten, ten and eleven, and every
combination thereof (three through five, three through six, three through seven, etc., four through
five, four through six, etc.). I do this because there are idiosyncrasies specific to certain grades.
For example, because of regulations in California Education Code, class sizes in grades
Kindergarten through third grade are significantly smaller compared to those in grades four and
higher. Beginning in grade eight, students enrolled in the same course can take different math
tests, which complicates estimation of VAMs beyond grade seven.
In general, all of the models described here are highly correlated, usually at 0.96 or
higher. However, in all analyses presented in this dissertation, I use the most conservative value-
added estimate, which limits the sample of teachers to just those in grades four through seven
(omitting grade three teachers due to their smaller class sizes and teachers in grades eight and
above because in these grades, students in the same class can take different standardized exams).
For the analysis presented in Essay 2, I test my preferred baseline model using all of the grade
spans described above. My results are generally consistent across all of these specifications. In
particular, the coefficient on layoff in year t-1 remains negative and significant across all models
(results are available from the author upon request).
For models in which I include middle and high school students (which serve as
specification checks to the preferred grade four through seven model), a few adjustments are
necessary because classrooms are not self-contained. In these models, I add dummy variables
indicating the post-test that each student took and interact these indicators with students’ pre-test
score. This approach controls for the potential non-random selection of students to particular

216
tests as well as the measurement error associated with each test (Isenberg & Hock, 2010).
Although my preferred VAM estimates for this study are the raw (non-shrunken) measures, in
alternate specifications, I conduct a Bayesian shrinkage procedure (Herrmann, Walsh, Isenberg,
& Resch, 2013). The purpose of shrinking teacher effects is to adjust toward the mean those
teachers with extreme scores (either high or low) but imprecise measures, while making only a
trivial adjustment to those with precisely measured extreme scores (Goldhaber & Theobald,
2013). In the main analyses for Essay 1, I do not use shrunken estimates and instead weight
observations in the regression by the standard error of each teachers’ estimate value-added. Table
A1 below shows the average value-added measure for teachers at various experience and
educational levels, based on the pooled model with shrunken estimates (which I use in Essay 3).

217
APPENDIX TABLE C1
Average value-added score for teachers by experience and highest level of educational
attainment (grand mean = 0.00, SD =.22 after Baysian shrinkage)
Experience Predicted VAM SE Z P<|Z| Lower bound Upper bound
1 -0.107 0.0232 -4.61 0.000 -0.152 -0.061
2 -0.096 0.0206 -4.68 0.000 -0.137 -0.056
3 -0.086 0.0181 -4.77 0.000 -0.122 -0.051
4 -0.077 0.0157 -4.89 0.000 -0.107 -0.046
5 -0.067 0.0134 -5.05 0.000 -0.094 -0.041
6 -0.059 0.0111 -5.28 0.000 -0.081 -0.037
7 -0.051 0.0090 -5.63 0.000 -0.068 -0.033
8 -0.043 0.0070 -6.16 0.000 -0.057 -0.029
9 -0.036 0.0051 -6.96 0.000 -0.046 -0.026
10 -0.029 0.0037 -7.79 0.000 -0.036 -0.022
11 -0.023 0.0033 -6.78 0.000 -0.029 -0.016
12 -0.017 0.0043 -3.9 0.000 -0.025 -0.008
13 -0.012 0.0061 -1.91 0.056 -0.024 0.000
14 -0.007 0.0082 -0.84 0.403 -0.023 0.009
15 -0.003 0.0105 -0.24 0.808 -0.023 0.018
16 0.001 0.0130 0.1 0.923 -0.024 0.027
17 0.005 0.0156 0.3 0.768 -0.026 0.035
18 0.007 0.0183 0.41 0.684 -0.028 0.043
19 0.010 0.0212 0.46 0.643 -0.032 0.051
20 0.012 0.0242 0.48 0.629 -0.036 0.059
Note: VAM refers to teachers’ value-added measure of effectiveness, SD stands for standard deviation, and N refers
to teacher-year observations for school years 2007-08 to 2012-13. A teacher’s predicted VAM is the predicted value
of their VAM, based on their experience, educational attainment, and school level (elementary or middle school).

Abstract (if available)

Abstract The Great Recession of 2008 devastated funding for public education in the United States, forcing districts around the country to cut budgets and conduct layoffs. In times of financial duress, when local districts are forced to make difficult choices about resources, limited evidence is available to help guide decisions. Similarly, as districts have begun to restore funding levels, there is a lack of research identifying the most cost-effective resource allocation strategies in K-12 public schools. Thus a crucial contribution for educational research is to document: (a) how teacher layoffs stemming from budget cuts impact students and teachers

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Are there hidden costs associated with teacher layoffs? The impact of job insecurity on teacher effectiveness in the Los Angeles Unified School District

PDF

District allocation of human resources utilizing the evidence based model: a study of one high achieving school district in southern California

PDF

Educational resource allocation at the middle school level: a case study of six middle schools in one California district

PDF

Personnel resource allocation in a Hawaii school complex

PDF

Personnel resource allocation strategies in a time of fiscal stress

PDF

Personnel resource allocation strategies in a time of fiscal crisis: case study of elementary schools in a California school district

PDF

Levels of interest: the effects of teachers' unions on Congress, statehouses, and schools

PDF

Educational resource allocation at the high school level: a case study of high schools in one California district

PDF

Navigating troubled waters: case studies of three California high schools' resource allocation strategies in 2010-2011

PDF

Resource allocation and educational adequacy: case studies of school-level resource use in southern California with budget reductions

PDF

School funding and the evidence based model: an examination of high school budget allocation in Hawaii

PDF

Allocation of resources to improve student achievement

PDF

The local politics of education governance: power and influence among school boards, superintendents, and teachers' unions

PDF

Resource allocation practices in relation to identified school reform strategies

PDF

Teacher evaluation reform in situ: three essays on teacher evaluation policy in practice

PDF

Educational resource allocation at the elementary level: a case study of one elementary school district in California

PDF

An examination of resource allocation strategies and finance adequacy: case studies of American Samoa Department of Education secondary schools

PDF

A case study of an evidence-based approach to resource allocation at a school for at-risk and incarcerated students in Hawaii

PDF

Resource allocation and instructional improvement strategies in rural single-school elementary districts

PDF

School-level resource allocation practices in elementary schools to increase student achievement

Asset Metadata

Creator Knight, David S. (author)

Core Title School resource allocation in times of economic boom and bust

School Rossier School of Education

Degree Doctor of Philosophy

Degree Program Urban Education Policy

Publication Date 02/02/2016

Defense Date 11/02/2015

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag budget cuts,class size reduction,cost-effectiveness analysis,economics of education,OAI-PMH Harvest,school finance,school research allocation,teacher effectiveness,teacher labor markets,teacher layoffs,teacher salaries

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Strunk, Katharine O. (committee chair), Lawrence, Picus O. (committee member), Painter, Gary D. (committee member)

Creator Email davidkni@usc.edu,dsknight@utep.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-204624

Unique identifier UC11278105

Identifier etd-KnightDavi-4077.pdf (filename),usctheses-c40-204624 (legacy record id)

Legacy Identifier etd-KnightDavi-4077.pdf

Dmrecord 204624

Document Type Dissertation

Format application/pdf (imt)

Rights Knight, David S.

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA