You are on page 1of 16

The Challenge of Teacher Retention in Urban Schools: Evidence of Variation From a Cross-

Site Analysis
Author(s): John P. Papay, Andrew Bacher-Hicks, Lindsay C. Page and William H. Marinell
Source: Educational Researcher, Vol. 46, No. 8 (NOVEMBER 2017), pp. 434-448
Published by: American Educational Research Association
Stable URL: https://www.jstor.org/stable/44971877
Accessed: 20-03-2024 07:51 +00:00

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

American Educational Research Association is collaborating with JSTOR to digitize,


preserve and extend access to Educational Researcher

This content downloaded from 152.118.24.31 on Wed, 20 Mar 2024 07:51:29 +00:00
All use subject to https://about.jstor.org/terms
The Challenge of Teacher Retention in
Urban Schools: Evidence of Variation From
a Cross-Site Analysis
John P. Papay1, Andrew Bach er- Hicks2, Lindsay C. Page3, and William H. Marinell4

Substantial teacher turnover poses a challenge to staffing public schools with effective teachers. The scope of the teacher
retention challenge across school districts, however, remains poorly defined. Applying consistent data practices and analytical
techniques to administrative data sets from 16 urban districts, we document substantial cross-district variation in teacher
retention rates. Observable characteristics do not easily explain this cross-district variation. We also find considerable cross-
district variation in key results from the retention literature, including the relationship between retention and both experience
and estimated effectiveness. Finally, we explore the influence of temporary leaves of absence and cross-district, within-state
movement on retention estimates. Accounting for cross-district movement matters little, while accounting for temporary
leaves matters a great deal in many districts and tends to exacerbate cross-district differences in retention rates.

Keywords: educational policy; longitudinal studies; multisite studies; retention; urban education

the only available data sources to quantify the teacher retention


door" in the teaching profession, researchers and policy- challenge nationally, he and coauthors acknowledge that the
Since door" makersmakersRichard in have thethatrecognized
have recognized high levels ofteaching Ingersoll profession, that first high wrote levels researchers about of teacher the and "revolving turnover policy-
teacher turnover SASS and TFS have notable limitations, namely, that they do
pose a critical challenge to the goal of staffing U.S. public school not track individual teachers over a long period of time (Ingersoll,
classrooms with effective teachers. Ingersoll's (2003) estimate that Merrill, & Stuckey, 2014).
40% to 50% of new teachers leave the classroom within the first A more recent, five-year longitudinal administration of the
five years of their careers has been widely cited in academic papers, SASS and TFS (from 2007-2008 through 2011-2012)- called
policy proposals, and the broader press since he first published the Beginning Teacher Longitudinal Study (BTLS) - partially
these findings over a decade ago (e.g., Hanna & Pennington, resolves this concern. It finds substantially lower turnover rates
2015; Johnson, Berg, & Donaldson, 2005; Keller, 2007; Kim & than Ingersoll's earlier work, suggesting that only 17% of new
Barg, 2010; National Commission on Teaching and America's teachers have left the profession after five years (Gray & Taie,
Future, 2003; Riggs, 2013; Ronfeldt, Loeb, & Wyckoff, 2013; 2015). These results have led some analysts to argue that teacher
Toppo, 2012). In fact, it has been repeated so regularly that it is retention is less of a concern than policymakers have thought
often reported without citation, particularly in the popular press (Hanna & Pennington, 2015; Zinshteyn, 2015).
and by policymakers, who have seized on these findings to focus We provide additional evidence about the true extent of the
attention on efforts to improve teacher retention. teacher turnover challenge in large urban school districts. Unlike
Most evidence on teacher turnover patterns comes from the SASS/TFS and the BTLS, which rely on survey data, we use
either national cross-sectional analysis of the Schools and Staffing
Survey (SASS) /Teacher Follow-up Survey (TFS) (e.g., Ingersoll,
1 Brown University, Providence, RI
2001, 2002, 2003) or individual studies examining turnover in 2Harvard University, Cambridge, MA
a particular state or school district (as summarized in Appendix 3University of Pittsburgh, Pittsburgh, PA
Table Al). While Ingersoll's seminal work uses the SASS, one of 4One8 Foundation, Boston, MA

Educational Researcher, Vol. 46 No. 8, pp. 434-448


DOI: 10.3102/0013189X17735812
© 2017 AERA, http://edr.aera.net
434 I EDUCATIONAL RESEARCHER

This content downloaded from 152.118.24.31 on Wed, 20 Mar 2024 07:51:29 +00:00
All use subject to https://about.jstor.org/terms
administrative human resources records provided by states and when evaluating retention, suggests that commonly cited teacher
districts themselves. With these data, we can track all employed turnover statistics may be inflated, and highlights the need to
teachers rather than relying on survey responses. The longitudi- account for returners when estimating teacher turnover costs in
nal nature of our data enables us to follow individual teachers specific districts. Differences in the proportion of teachers who
over time. Thus, we eliminate two important sources of bias in return tend to exacerbate cross-district differences in retention.
past estimates - the lack of longitudinal data in the SASS/TFS Third, using data from four states, we quantify within-state,
and the low response rate in the BTLS. cross-district turnover. Anecdotally, administrators in urban districts
Furthermore, while national teacher retention rates may be express concern about losing teachers to surrounding suburban dis-
useful for informing large-scale policy discussions, retention varies tricts. Interestingly, cross-district movement is not a substantial
widely across states, districts, and schools. The SASS/TFS and driver of teacher turnover in the districts we examine. This finding
BTLS sampling strategy and small sample size make them of lim- may have important implications for strategies that state education
ited use for examining variation in teacher retention rates across agencies propose as part of their Educator Equity Plans to ensure
localities. Urban school systems in particular struggle to retain that students have equal access to effective educators.1
teachers (e.g., Marinell & Coca, 2013; Newton, Rivero, Fuller, & Lastly, we examine how retention varies by teacher experience
Dauter, 2011; Useem, Offenberg, & Farley, 2007). Several high- and effectiveness. Past research has found that teachers are more
quality studies have examined teacher retention at the state and likely to leave early or late in their careers or if they are not hav-
district levels; however, each of these studies examines only one ing success with their students (Boyd et al., 201 1; Feng & Sass,
site, and variation in their analytic approaches, data sources, peri- 2012; Goldhaber, Gross, & Player, 2011; Hanushek, Kain, &
ods of observation, and specific research questions hinders direct Rivkin, 2004). On average, we find that these general trends
cross-site comparison. It is not clear whether differences in calcu- hold in our data, confirming what others have found.
lated retention rates across districts in these studies reflect true Nevertheless, we also document quite substantial differences in
variation or are simply an artifact of variation in the methods used. magnitude, with some districts doing a much better job of
Therefore, despite a substantial literature, policymakers are left retaining their more effective teachers than others.
with limited information about the extent of the teacher retention Taken together, our results add considerable nuance to our
challenge and how it varies across districts. understanding of the teacher retention challenge in the United
We make four main contributions. First, we quantify the extent States today. Although in some cases they confirm prior evidence,
of and variation in the teacher retention challenge across 16 urban our results reveal a more complex reality than the newer teacher
public school districts in seven states using longitudinal, teacher- retention estimates from the BTLS, providing important insights
level administrative data and common data practices. We examine into the nature of the teacher retention challenge and how it varies
retention at the school, district, and state levels for two categories across districts. Recendy, school systems have once again been strug-
of teachers: all teachers and teachers in the first year of their career gling with substantial teacher shortages (e.g., DeAvila & Hobbs,
(i.e., novices). We find that context matters - there are large differ- 2017; Rebora, 2016; Sutcher, Darling-Hammond, & Carver-
ences in retention rates across districts. Ingersoll (2003) reported Thomas, 2016). While much of the media attention has focused on
that nationwide, 14% of novices left teaching within one year, efforts to attract new teachers to the classroom, teacher retention
33% within three years, and 46% within five years. We find that will likely return to a central place in the policy spotlight.
teacher retention and therefore teacher turnover vary substantially Finally, our study provides a useful example of the value of
by district. Across the districts examined, 13% to 35% of novices replication and in particular, conducting comparable analyses
left their district after one year, while 44% to 74% left within five across different contexts. Given data availability, educational
years. Thus, if anything, Ingersolls estimates understate the research tends to focus either on the nation overall or a single
teacher turnover challenge in large urban school districts, and the state or district. Thus, much of what researchers "know" about
hiring needs to consistently fill a single teaching slot are as much the teacher labor market or the effects of educational policies
as 40% greater from one district to the next due to differences in (particularly those concerning teachers) derives from a small
teacher retention rates. This variation implies that districts may subset of districts and states with robust data systems, such as
face substantially different costs to staff their schools fully. For New York, North Carolina, Florida, and Texas. Yet, context
example, in the sample district with the lowest rate of retention, clearly matters. Had we focused only on one district, our results
our conservative, back-of-the-envelope calculation suggests that would have been strikingly different depending on which district
annual hiring costs for novice teachers might be as much as $4 we chose. The burgeoning amount of school and district admin-
million lower if the district retained novices at the highest rate we istrative data and the increasing consistency in data collected
observe. across sites facilitate important comparisons and allow research-
Second, we find that the methods used matter. The analytical ers to develop a better understanding of which patterns hold true
approaches in past studies have varied in many ways, including universally and which are specific to a given context.
how they treat "returners" - teachers who take temporary leaves
of absence before returning to their school or to another school
The Magnitude and Costs of Teacher Turnover:
in the same district. We find that in some districts, very few
Existing Evidence
teachers leave and return. By contrast, in half of the districts we
examine, accounting for returners raises the five-year within- High rates of teacher turnover impose several costs on school dis-
district retention rate by at least 10 percentage points. This find- tricts. First, when teachers leave, schools and districts must replace
ing underscores the importance of using a long panel of data them. Estimates of the financial costs of hiring vary widely, but in

NOVEMBER 201 7 | 435

This content downloaded from 152.118.24.31 on Wed, 20 Mar 2024 07:51:29 +00:00
All use subject to https://about.jstor.org/terms
all cases are quite high, ranging from $10,000 to $20,000 in urban TFS annually for five years to the 1,990 novice teachers who par-
districts (Birkeland & Curtis, 2006; Milanowski & Odden, 2007). ticipated in the 2007-2008 SASS. Using the two most recent
Second, the total cost of turnover must factor in both the net dif- administrations of the SASS (2007-2008 and 2011-2012) and
ference in salary and the net difference in productivity between a the BTLS, Hanna and Pennington (2015) estimate lower levels of
departing teacher and his or her replacement. Thus, examining attrition in recent years, with 30% of novice teachers leaving the
teacher retention rates by years of experience (the primary salary profession within five years. Gray and Taie (2015) use five waves
determinant) and level of effectiveness is important; we take up of data from the BTLS to estimate that 83% of the novice teach-
these questions in the following sections. ers who began teaching in 2007-2008 remained in the profession
Of course, some turnover may be beneficial, particularly if five years later.3 The higher national retention rates that these
less effective teachers leave the district.2 On average, teachers studies identify may be due in part to the recession that occurred
who leave are indeed less effective than their peers (Boyd et al., during the BTLS 's period of observation (Hanna & Pennington,
2011; Feng & Sass, 2012; Goldhaber et al., 2011), and in fact, 2015).
teachers' lack of success is a key determinant in their decisions to Although the BTLS 's longitudinal data are useful for estimat-
leave (Johnson & Birkeland, 2003). Yet, departing teachers often ing national retention rates, the data set has a number of notable
are replaced by novice teachers who are substantially less effec- limitations. The initial cohort of novice teachers is small (1,990
tive than more experienced teachers, on average, largely because teachers), so retention estimates are relatively imprecise, particu-
teachers' effectiveness improves rapidly early in the career (Harris larly for teacher subgroups. Further, response rates to the TFS
& Sass, 2011; Papay & Kraft, 2015; Rockoff, 2004). In sum, were low in some years (e.g., 57% response rate for the fifth
turnover negatively affects the composition of a school's teaching wave), which could generate bias in retention estimates because
force (Ronfeldt et al., 2013). Further, high levels of turnover can teachers who left the profession may be less likely to respond.
reduce student achievement, even among students whose teach- Perhaps most importantly, given the nature of the SASS sample
ers have remained in the school and especially in schools with design, it is only possible to compare estimated retention rates
higher proportions of low-performing and African American across broadly defined community types, such as "city/subur-
students (Ronfeldt et al., 2013). Thus, turnover hurts students ban" and "town/rural." Thus, the teacher retention estimates
both by changing the composition of the teaching force and dis- drawn from the BTLS, like those from the SASS/TFS, obscure
rupting the school organization. important variation across states and school districts, which face
The substantial financial and educational costs of chronically dramatically different labor markets and policy contexts.
high teacher turnover underscore the importance of understand- In contrast to these national estimates, numerous studies
ing the magnitude of the problem. The most prominent account- have examined teacher turnover in an individual state or district
ing of teacher turnover in the U.S. context comes from Ingersoll's (we summarize these results in Appendix Table Al). Although
(2001, 2002, 2003) analyses of the Schools and Staffing Survey/ these studies broadly agree with Ingersoll's estimates, turnover
Teacher Follow-Up Survey, which conclude that "after just five rates are substantially higher in large urban districts, where
years, between 40 and 50 percent of all beginning teachers have approximately two-thirds of new teachers leave their school
left teaching altogether" (Ingersoll, 2003, p. 13). Using within the first five years. Even among large urban districts,
Baccalaureate and Beyond data, Perda (2013) similarly estimates these studies collectively signal substantial cross-site variation in
that approximately 40% of beginning teachers leave the profes- turnover rates. For example, Newton et al. (201 1) find that 57%
sion within five years. With data from the Current Population of new elementary school teachers in Los Angeles leave their
Survey, Harris and Adams (2007) find that nearly 8% of all school within five years, while Useem et al. (2007) find that 84%
teachers exit the profession each year, comparable to other pro- of new teachers in Philadelphia leave within six years.
fessions, such as nursing. Characterizing such variation from these studies is complicated,
Policymakers have seized on these national figures, often however, by the fact that they differ along several dimensions. As
using them to justify policy efforts designed to stem teacher we document in Appendix Table Al, these studies examine
turnover (e.g., Keller, 2007; National Commission on Teaching teachers at different grade levels, during different time periods,
and America's Future, 2003; Riggs, 2013; Toppo, 2012). and over different time spans. While many studies, including
However, as Ingersoll and colleagues (2014) have acknowledged, Ingersoll's, focus on novice teachers, others look at all new hires
their work provides "only a rough estimate using cross-sectional or all teachers. Studies also differ in whether they examine teach-
national data" (p. 24). The central drawback involves the survey's ers who leave and return; as Grissom and Reininger (2012) find,
limited follow-up, which prevents tracking individual teachers reentrants represent a substantial supply of teacher labor. Finally,
over a several-year period. Rather, to approximate the five-year they differ in the perspective from which turnover is measured:
retention rate, he estimates the cumulative probability that a Some examine school-level turnover, while others focus on turn-
teacher leaves in each of the first five years of his or her career. over from a district, a state, or the entire profession. From a
Ingersoll also notes that these estimates may overstate the true school's perspective, the effect of a teacher transferring to another
extent of turnover because they do not account for teachers who school in the district or leaving the district/profession entirely is
leave temporarily and then return (e.g., after parental leave). essentially the same. From a district's perspective, however, only
Recognizing the limitations of the cross-sectional nature of the the latter necessitates a new hire. We hold these variations con-
SASS andTFS, in 2007-2008, the National Center for Education stant across sites, compiling comparable data from multiple
Statistics launched the BTLS, which entailed administering the urban school districts and using consistent analytical approaches.

436 I EDUCATIONAL RESEARCHER

This content downloaded from 152.118.24.31 on Wed, 20 Mar 2024 07:51:29 +00:00
All use subject to https://about.jstor.org/terms
Data and Methods In addition to the main retention outcomes, we also derive
measures of teaching experience and teachers' value-added to
We constructed a rich longitudinal data set from 16 midsized to
examine variation in retention across these key characteristics.
large urban school districts across seven states. These administra-
We draw our measure of teaching experience from human
tive data were acquired by Strategic Data Project (SDP), housed
resource files, based on the teacher's step on the district's salary
within the Center for Education Policy Research at Harvard
scale, and define a novice teacher as a teacher on the first step
University, as part of research collaborations with local and state
of the salary scale. We estimate value-added for math teachers
education agencies.4 These data typically entail annual files
using administrative data from the four districts for which we
including (a) teacher human resource data, which provide the
have links between students and their teachers of record. We
key information for our analysis; (b) student demographic infor-
control for students' prior year test scores and demographic
mation; (c) student test score data; and (d) student-teacher links
characteristics in our value-added model; we provide more
(although these links are not available in all districts).
detail on the value-added model in Appendix B. While value-
We focus only on urban or county-wide districts that have a
added is clearly a limited measure of teacher effectiveness, it is
large, urban core.5 The 16 districts we study collectively serve
a useful proxy measure that is available in at least some of the
more than 2.4 million students annually, or approximately 5%
districts we study.
of the U.S. public school population. In general, schools in large
urban districts serve a large share of disadvantaged students, who
Results
may stand to gain the most from increased teacher quality and
stability. In Appendix Table A2, we present summary statistics
Retention Rates Across Districts
on a variety of characteristics for the 16 school districts included
Across
in our study. As two points of comparison, we also provide aver-our sample, 13% of all teachers leave their district each
age characteristics for all public school districts in the year,
United and 45% do so within five years. By comparison, 19% of
States and for the 100 largest urban school districts. The average leave their school within one year and 58% within five
teachers
district in our analysis serves nearly 1 50,000 students peryears.
year,Of course, some turnover is expected given that many
the majority of whom are eligible for subsidized lunch (61%) teachers were nearing the age of retirement. Across dis-
sampled
and are African American or Hispanic (62%). Although the tricts,
stu-between 6% and 17% of teachers have more than 25 years
of experience.
dents in our sample are markedly different than the average stu- As a result, it can be useful to examine novice
teachers, in line with Ingersoll's focus on "beginning teachers."
dent in the United States, the students in our sample are similar
to those in the 100 largest U.S. districts. Compared to teachers overall, retention rates for novice teachers
Across the 16 districts in our sample, we constructed a
are somewhat lower. Within five years, 55% of novice teachers
leave their
teacher-year-level longitudinal data set, spanning 1 5 school years district, and 70% leave their school. These overall
(1998-1999 to 2012-2013) and including over 200,000 unique
figures largely corroborate Ingersoll's work assessing the magni-
tude
teachers.6 Not all districts are represented in all years of our of the teacher turnover problem and are substantially larger
data.
In addition to the summary statistics described previously,than we
the more recent national teacher turnover statistics, perhaps
unsurprising
also provide an overview of the years included for each district in given our focus on urban districts.
Appendix Table A2. Using this teacher year-level longitudinal Importantly, these figures mask substantial cross-site varia-
data set, we derive our variables of interest. We thus ensure that
tion, even for demographically similar districts. In Table 1 , we
report teacher retention rates at the school and district levels for
consistent definitions and analytic methods are used, allowing
all teachers and for novice teachers. Across districts, one-year
for more accurate comparisons across sites. Our main variables
of interest indicate whether a teacher remains in a teaching
retention rates for all teachers range from a low of 83% (District
assignment in the same school (or district or state) after aC)given
to a high of 90% or more in several districts. Three-year and
period of time.7 We count a teacher as being retained evenfive-year
if he retention rates vary more substantially; three-year rates
rangetofrom 58% to 75% and five-year from 43% to 64%. In
or she changes grades or subjects but not if the teacher moves
short, in some districts, nearly 60% of all teachers leave within
a nonteaching role.8 For all teachers in all districts, we created
the following indicators of same-school teacher retention: five years, while in others only one in three leaves.9
From the perspective of a school, district-level retention rates
• One-year same-school retention: Teacher remains
do in
not afully represent the challenge. As expected, rates of same-
teaching assignment in the same school from year t to school
t + 1 . retention fall below those of same-district retention
• Three-year same-school retention: Teacher remainsbecause
in a teachers transfer between schools. Across the districts,
five-year
teaching assignment in the same school from year t to t + 3. school-level retention rates for all teachers range from
• Five-year same-school retention: Teacher remains in a 29% to 47%. These results are striking for two reasons. First, in
teach-
ing assignment in the same school from year t to t + 5.some districts, school-level retention is quite low - in District C,
less than one-third of all teachers remain in the same school for
five years. Second, even in the districts with the highest retention
We identify analogous one-year, three-year, and five-year district-
level retention variables. For the districts within four states,
rates, substantial instability exists in staffing at the building level.
For example, even in the districts with the highest district-level
we are also able to take advantage of the state-level administra-
retention rates, only about half of teachers remain in the same
tive files to identify a set of analogous variables for within-state
retention. school for at least five years.

NOVEMBER 2017 | 437

This content downloaded from 152.118.24.31 on Wed, 20 Mar 2024 07:51:29 +00:00
All use subject to https://about.jstor.org/terms
Table 1
Within-District and Within-School Teacher Retention Rates Across Districts (%)
All Teachers Novice Teachers

Within-District Within-School Within District Within School

One- Three- Five- One- Three- Five- One- Three- Five- One- Three- Five-
District Year Year Year Year Year Year Year Year Year Year Year Year

A 90 74 63 82 59 45 77 61 55 67 47 37
B 88 72 58 77 52 37 82 66 56 61 40 25
C 83 58 43 76 45 29 73 40 26 66 29 14
D 89 65 47 81 54 34 72 49 31 59 33 21
E 90 74 63 76 48 31
F 92 74 59 85 64 42 83 64 56 67 49 33
G 88 72 80 56 75 64 60 43
H 87 73 79 58 86 69 77 54
I 87 69 83 63
J 90 74 84 61
K 87 69 57 82 62 47 82 57 53 75 46 38
L 87 68 54 82 60 45 87 62 47 78 47 33
M 88 71 58 78 55 41 76 61 52 59 40 29
N 90 74 62 83 62 47 83 67 54 72 51 35
0 89 75 64 80 59 45 65 55 50 46 29 19
P 87 69 57 81 57 43 77 58 66 45

Note. For the columns corresponding to all


the five-year retention. For the novice teac
the 10 districts with five years of data and i

Retention within-statefor
rates transfers. While novice
we cannot examine cross-state teac
than those for moves,
all these data provide a broader picture of retention
teachers and in the ar
In most districts, more
profession. In Figure than
1, we present three-year retention rates at on
schools after a the school, district, and year.
single state levels for all teachers.
Three-Interestingly, a
also quite
low, thewith
within-state retention some distri
rates are not substantially larger than
retaining new district retention rates. Thus, the large urban
teachers. For districts we exam
study
novice teachers remain
do not appear into other
to be losing large shares of teachers their dis-
remain in the tricts in the same state. This is true
district) regardless of district
after size as
five
nearly 40% of our
novice
sample includes some states teachers
with large, county-wide districts re
years after theyand others
start. with many smaller school districts. These patterns
also hold for novice teachers.
Second, accounting for teachers who leave temporarily may
The Role of Cross-District Tran
Leaves affect retention rates. In Figure 2, we present five-year within-
district retention rates adjusted for teachers who returned within
Our analyses thus
five years.10 All far count
districts had teachers teac
who subsequendy returned,
have never leftbut thea given
proportion district
of returners varies substantially. For example, in
likely understate retention
Districts A and E, in
accounting for returners only increases five-year t
they do not account for
retention rates by 5 percentage those
points. In contrast, in Districts L, M, w
district lines. and O, rates increase
Such by 12 or more percentage points. Teachersma
movement
comparisons if in
who leave and returnsome
are important to considerlocation
because these teach-
move from one district
ers do not to
represent lost human capital for the anoth
school district and do
Florida and Maryland,
not require the distria to undertake which
cosdy recruitment efforts. hav
ers may be less likely
Interestingly, to
districts with the highest retentiontransit
rates tend to have
Massachusetts,
the largestwhich has
adjustments, suggesting that these sma
districts do a better job
Second, our estimates
of retaining teachers on leave as well. Indo
sum, adjustingnot
for returners ac
temporarily (e.g., for
increases the cross-distria parental
range in five-year retention rates, suggest- lea
To examine the prevalence
ing that of
the retention picture is comparatively even bleaker cro
in dis-
on four states tricts
in which
with high we
turnover once we account for returners. can o

438 I EDUCATIONAL RESEARCHER

This content downloaded from 152.118.24.31 on Wed, 20 Mar 2024 07:51:29 +00:00
All use subject to https://about.jstor.org/terms
FIGURE 1 . Three-year within-school, -district, and -state retention rates for all teachers
The sample includes the 12 districts for which we have data on within-state three-year retention.

FIGURE 2. Five-year within-district retention rates for all teachers, adjusting for teachers who leave and then subsequently return
The sample includes the 10 districts in which we observe six years of data (one year to first observe those who leave and then five
additional years to observe retention and returns).

How Teacher Experience and Effectiveness Relate to teachers leave. Here, we consider two factors that have garnered
Teacher Turnover substantial interest: teacher experience and teacher effectiveness.
Clearly, attrition from retirements suggests a challenge different
The full costs of teacher turnover, including the impact of turn- from rapid churn among early- and midcareer teachers.
over on school effectiveness, depends in large part on which Furthermore, schools and districts unsuccessful in retaining their

NOVEMBER 2017 | 439

This content downloaded from 152.118.24.31 on Wed, 20 Mar 2024 07:51:29 +00:00
All use subject to https://about.jstor.org/terms
FIGURE 3. Five-year within-district retention rates for all teachers across years of teaching experience
The sample includes the 10 districts for which we have data on five-year retention rates and teacher experience. In this figure, we
present the smoothed values of a kernel-weighted local regression of retention rates on teacher experience. We use the Epanechnikov
kernel function and a bandwidth for each regression that minimizes the conditional weighted mean integrated squared error.

most effective teachers face greater costs in terms of lost student across the districts. In District H, retention rates for high and low
learning when these teachers leave. These analyses broadly con- value-added teachers differ substantially. Among the top third
firm the patterns that we expect concerning the relationship novice teachers, 50% remained in their same school three years
between retention and both experience and teacher effectiveness later, compared to just 29% of teachers at the bottom of the distri-
(as measured by value-added). However, we also examine and bution. By contrast, Districts C, K, and P exhibited much smaller
document substantial variation across districts in the magnitude differences among high and low value-added teachers. For exam-
of these relationships; the patterns hold much more strongly in ple, in District C, 36% of higher performing teachers and 29% of
some districts than others. lower performing teachers remained in their same school after
In Figure 3, we present average teacher retention rates over three years. Thus, while Districts C and H had quite similar reten-
teachers' careers, plotting the percentage of teachers at each tion rates for less effective novice teachers, they had vasdy different
experience level who remain teaching for the next five years in success in retaining their most effective novices.11 These differ-
each district. As would be expected, within each district, a dis- ences imply that the effects of turnover on student achievement
tinctly nonlinear relationship exists between teacher retention might differ substantially between the two districts. Furthermore,
and years of experience such that retention is lowest among nov- efforts to retain high-performing teachers might be particularly
ice teachers and veteran teachers and comparatively higher useful in reducing the negative effects of turnover.
among midcareer teachers. At the same time, at each level of
teacher experience, substantial cross-district variation exists.
Predictors of Cross-District Variation in Teacher
Consider teachers with 1 5 years of experience. In some districts, Retention
these teachers exhibit five-year retention rates near 80%. In oth-
ers, however, analogous retention rates fall below 60%. One obvious question that arises from our findings is what drives
We also investigate cross-site variation in retention by esti- the substantial variation in teacher retention rates that we docu-
mated teacher effectiveness. Using the same value-added method- ment across districts? Why do some districts have much higher
ology across sites, we generated estimates of math teachers' retention rates than others? While a robust literature has exam-
effectiveness in raising student test scores. While value-added ined determinants of teacher workforce stability at the individ-
clearly does not capture all aspects of a teacher s effectiveness, it is ual and school levels (for reviews, see Ingersoll, 2001; Johnson et
a useful proxy, particularly for examining relative differences, as we al., 2005; Simon & Johnson, 2015), there is less evidence on the
do here. In Figure 4, we plot three-year retention rates for teachers predictors of cross-district differences.
in the top and bottom third of their district s value-added distribu- The literature on individual- and school-level factors driving
tion. Across all four districts we can examine, the general patterns teacher turnover has largely focused on three areas. The first
present in the literature hold: More effective teachers are more involves traditional features of the labor market commonly
likely to remain in their districts and schools than their less effec- explored by economists, including pay, benefits, pensions, geo-
tive colleagues. Here again, however, considerable variation exists graphic location, job amenities, and outside labor market

440 I EDUCATIONAL RESEARCHER

This content downloaded from 152.118.24.31 on Wed, 20 Mar 2024 07:51:29 +00:00
All use subject to https://about.jstor.org/terms
FIGURE 4. Three-year within-district and within-school retention rates for novice teachers by value-added tercile
The sample includes the four districts for which we have data on value-added and teacher experience. Within each district, novice
teachers are divided into three equally numbered groups based on value-added estimates. Retention rates for the teachers in the top
third and bottom third within each district are presented.

opportunities (e.g., Brewer, 1996; Gritz & Theobald, 1996; the district-level determinants of turnover is a potentially fruitful
Jacob, 2007; Mont & Rees, 1996; Murnane & Olsen, 1990; area of research because district-level policies may reflect levers
Murnane, Singer, & Willett, 1989; Shen, 1997; Stinebrickner, that are more readily malleable than, for example, the organiza-
1998; Theobald, 1990). These factors all contribute to teachers tional context of a school.
decisions to leave the profession, although they are less relevant While rigorous analysis of the determinants of cross-district
in teachers' decisions to move schools given that such moves variation falls beyond the scope of this paper and the data avail-
tend to be local and many of these features are consistent within able to us, we present some initial descriptive results. Specifically,
districts. A second area has tended to explore the types of stu- we examined the relationship between district retention and a
dents with whom a teacher works in the classroom (e.g., Boyd et parsimonious set of predictors identified in existing studies that
al., 2005; Clotfelter, Ladd, & Vigdor, 2011; Hanushek et al., we could measure reliably and consistently across the 16 dis-
2004). The consistent finding in this literature is that teacher tricts, including student demographic characteristics, available
retention is lower in schools that serve high proportions of measures of basic working conditions (i.e., district average
minority students, low-achieving students, or students with low student-to-teacher ratio and the number of student support ser-
family income. Third, researchers have begun more recently to vices staff per teacher), and economic conditions (i.e., measures
examine the role of teacher working conditions in teachers' of the regional unemployment rate and average district teacher
retention decisions, focusing on features of the organizational salaries). We estimate a separate district-level ordinary least
context rather than the students served (Simon & Johnson, squares (OLS) regression of retention on each predictor and
2015). An emerging finding from this literature is that by and present the results of this analysis in Appendix Table A3.
large, teachers prefer to teach in schools that have strong school Our primary conclusion is that these readily observable dis-
cultures, effective leadership, and supportive colleagues (e.g., trict characteristics do not appear to be systematically related to
Goldhaber, Lavery, & Theobald, 2016; Johnson, Kraft, & Papay, teacher retention rates, at least among the sample of urban dis-
2012; Ladd, 201 1; Sutcher et al., 2016). tricts that we study. In total, we examined nine predictors across
Cross-district differences in each of these factors may produce six measures of teacher retention: the one-year, three-year, and
the variation in turnover we see here. However, certain factors five-year rates for both novice teachers and all teachers. Across
may relate more directly to specific types of turnover than others. these 54 relationships, only 1 is statistically significant at the 5%
For example, differences in district policies such as pay scales level, which is less than what we would expect solely by chance.
may motivate teachers to move across district boundaries but Of course, our relatively small sample of only 16 school districts
likely do not contribute to their decisions to transfer schools likely limits our ability to detect significant relationships. Thus,
within a district. Or, local labor market conditions or the state it is perhaps more compelling that most of the relationships are
accountability context may influence teachers' decisions to leave quite small in magnitude and several in fact change sign as we
the profession altogether but are less likely to affect whether examine different teacher retention outcomes (e.g., one-year vs.
teachers move across schools or districts. Further exploration of five-year retention rates). Furthermore, some have a sign

NOVEMBER 2017 | 441

This content downloaded from 152.118.24.31 on Wed, 20 Mar 2024 07:51:29 +00:00
All use subject to https://about.jstor.org/terms
opposite from what we would expect, suggesting that factors we our highest retention rate), this figure is 1.61 teachers. That is,
have not observed likely play an important role. For example, for each open teaching slot, one district s hiring needs are nearly
the positive coefficients associated with the share of students 40% greater than the others due to differences in retention.15
receiving free or reduced price meals would indicate higher rates Assuming that a district spends $10,000 to hire a novice teacher
of teacher retention in lower-income districts. We do note that (at the low end of the estimates generated by Milanowski &
one variable follows a consistent and sensible pattern across all Odden, 2007), District C must spend $6,300 more than District
outcomes: Regional unemployment is positively associated with B to keep each novice teacher slot filled for five years. Scaling
all six retention measures. this cost by the District C s annual novice teacher cohort of
Although our analysis is only exploratory, we conclude that approximately 650, the implied cost of filling slots for five years
there is little evidence of strong relationships between these pre- from a given novice teacher cohort is thus $4. 1 million greater in
dictors and teacher retention.12 Thus, the key factors driving District C than in District B.16
variation in teacher retention across these districts likely are Comprehensively discussing the variation in policies, prac-
much more nuanced than can be observed in our measures of tice, and context that leads to these striking differences in reten-
student demographics, working conditions, or regional labor tion falls beyond the scope of this article. But they are not easily
markets; that retention rates vary so dramatically across observ- explained by observable characteristics such as student demo-
ably similar districts suggests that hard to measure differences - graphics, unemployment rates, salaries, and so on, at least among
in policy, school work environments, and so on - contribute to the sample of large, urban school districts that we study.
this variation more than these easy to observe characteristics. Researchers have begun to document in more detail why turn-
Two of these potentially more nuanced factors involves differ- over occurs, investigating, for example, the role of the schools
ences in the prevalence of teachers who enter the profession from organizational context (Boyd et al., 201 1; Johnson et al., 2012;
alternative routes and the prevalence of teachers in charter schools, Ladd, 201 1; Marinell & Coca, 2013). Our results suggest that in
both of whom may have shorter stints in the classroom. Of the addition to these school-level factors, district policies and prac-
districts we study, Teach For America (TFA) is the largest alterna- tices may also contribute to teacher retention. Policy efforts to
tive pathway to teaching, with TFA corps members comprising improve teacher stability likely will benefit from further under-
from 0% to 19% of all novice teachers in these districts. However, standing the factors that contribute to the substantial variation
we see no evidence of a positive relationship between the preva- we have documented.
lence of TFA or charter school teachers and higher teacher turn- A second finding is that even the districts with the highest
over in the districts we study, suggesting that these explanations levels of retention struggle to retain teachers, particularly at the
are not drivers of the cross-district variation we find.13 school level. While retention represents a crisis for some districts,
it remains a serious challenge for all. Our findings suggest that
Discussion and Conclusion Ingersolls estimates represent lower bounds on the retention
rates of large urban districts, while the more recent BTLS esti-
This article is the first, to our knowledge, to use administrativemates
data substantially understate them. Thus, these results can
help, once again, push teacher retention more firmly into the
from multiple states and districts and a consistent analytic approach
policy spotlight and motivate districts and schools to develop
to examine the challenge of teacher retention. Unlike past studies
that focus on national cross-sectional samples or longitudinal data
strategies to retain teachers.
from a single locale, we combine information from 16 urban schoolHere in particular, districts must attend to efforts to identify
and in
districts in seven states to document the level of and variation retain their most effective teachers. Substantial differences
exist
teacher retention as well as the relationship between retention andin retention rates by estimated effectiveness across the four
districts we examine, as shown in Figure 4. In two of the four
both teacher experience and effectiveness. On average, our results
are of a similar magnitude to retention patterns observed in districts,
prior Districts C and K, the discrepancy between the retention
ratestoof teachers with value-added estimates in the top and bottom
work. For example, whereas Ingersoll (2003) estimated that 40%
50% of beginning teachers were no longer teaching within five thirds
years of the distribution is less than 10 percentage points. These
findings
of starting, we find a five-year turnover rate of 55% across the dis- confirm that teacher retention requires not only general
solutions
tricts that we examine. Looking beyond such averages, however, we designed to stem the flow of early-career teachers from
the profession but also specific solutions targeted to support and
derive several key lessons for researchers and policymakers regarding
cross-site variation. engage a districts most effective teachers.
Most strikingly, districts have substantially different degrees A third finding to underscore is our lack of evidence, at least in
of success in retaining teachers. For example, some districts the four states we study, that teachers leave urban districts to move
retain only one-quarter of their new teachers for five years, while to neighboring suburban schools - a concern that policymakers
others keep more than half. To help interpret the magnitude of and district administrators often voice. Adjusting retention rates
this cross-district variation in teacher retention, we note that for in-state movers boosts districts' retention rates by only a per-
these differences in retention rates imply that the hiring needs centage point or two. Instead, comparatively more movement
and financial costs associated with turnover vary substantially occurs within these large districts as teachers transfer from one
across school districts. Our back-of-the envelope calculations school to another. From the school perspective then, teacher reten-
suggest that District C (with our lowest retention rate) would tion is even more of a problem - in some districts fewer than one
need to hire 2.24 teachers to fill a single teaching slot consis- in five new teachers remain in their initial school for five years.
tently over a five-year period.14 By contrast, in District B (with While within-district movement avoids some administrative costs

442 I EDUCATIONAL RESEARCHER

This content downloaded from 152.118.24.31 on Wed, 20 Mar 2024 07:51:29 +00:00
All use subject to https://about.jstor.org/terms
of turnover, schools still must invest resources to find and train 2Teacher employment, particularly early in the career,
new candidates. Furthermore, students may suffer because of the way matching process where both the district and the teach

resulting organizational instability. evaluate the match. Thus, some teacher turnover is volunta
ers transfer across schools or leave the profession, and som
This work also raises important questions for further inquiry
tary, as teachers are removed from their positions. Although
into teacher retention. Extensions might investigate what
policy efforts have sought to strengthen teacher evaluation
factors - beyond the easily observable characteristics that we and remove less effective teachers from the classroom, suc
consider - explain the variation in retention rates we describe.
were not widespread during the period (and within the
Clearly, the data requirements and research design challenges in study, and few teachers were dismissed for performanc
conducting such work across a range of sites are daunting. But ferences in teacher dismissal rates do not explain the hete
developing more nuanced understandings of, for example, the retention rates we see.

variation in the organizational characteristics of schools or dis- 3This estimate includes reentrants (i.e., those who left but then
trict policies is a clear next step. Relatedly, the field needs a better returned to the profession during the period of observation). Using the
understanding of effective strategies for retaining teachers, par- same data set, Raue and Gray (2015) estimate the five-year national
ticularly top-performing ones. Here, we also need better infor- retention rate to be slightly lower, at 77%.
4Founded in 2008 with funding from the Bill and Melinda Gates
mation about what stages in a teacher s career are most efficient
Foundation, Strategic Data Project (SDP) collaborates with education
for intervention to boost retention; for example, perhaps highly
effective later-career teachers who intend to retire can be more agencies across the country with the goal of transforming the use of
data in education to improve student achievement. In each agency, SDP
readily encouraged to remain in the classroom for a few addi-
gathers de-identified longitudinal data on both students and teachers
tional years than their less experienced colleagues who may to conduct diagnostic reports about human capital and college-going.
intend to change careers or leave the district for external personal Many of the analyses that we feature here were initially conducted as
reasons.
part of SDP s Human Capital Diagnostics with state and district part-
Finally, our study has important lessons for policymakers
ners. Consistent with SDPand
s policy, we mask agencies' identities to pre-
researchers concerning the creation and use serve their confidentiality.
of research evidence.
Context matters. That national estimates of teacher turnover 5The National Center for Education Statistics (NCES) identifies

mask important cross-district variation suggests similar variation all 16 districts as cities or large suburbs.
6Our estimate of the number of unique teachers in our com-
in other policy areas. Our understanding of the turnover chal-
bined data set relies on two related assumptions. First, we assume that
lenge would have been much different had we done a compre-
the same individual teacher is not included in more than one locality's
hensive study in District B instead of in District C. However,
data set. We believe that this is a reasonable assumption given the geo-
because of data limitations, most studies can only examine one
graphic dispersion of the sites included in our analysis. Second, specific
specific site. In education in particular (and especially for studies to the state-level data files we include, we assume that teacher ID num-
of teachers), much of the evidence generated to answer policybers do not change if/when they move across districts within the same
questions of importance relies on data from relatively few sites, state. Given that teachers' records are administratively linked by social
such as New York, Florida, Texas, and North Carolina. Our security number and/or other state-assigned identifiers such as teacher
study illustrates the importance of replication across sites to license number, we expect that these types of potential errors in the
investigate cross-context variation more fully. Recent promising matching of teacher records across districts are extremely limited.
efforts, including a major initiative funded by the Spencer and 7Of course, we are not able to observe certain types of teacher
turnover given the data on which we rely. For example, we are not able
William T. Grant foundations, have focused on exploring het-
to observe teachers who leave public school teaching in favor of a private
erogeneity of causal effects from multisite randomized controlled
school context. Similarly, we are not able to track teachers from one
trials. We hope that the increased availability of data can facili-
state to the next. Our estimates are best seen as examining the retention
tate additional research to explore patterns and variation across
of public school teachers within a given school, district, or state.
sites in both causal and descriptive analyses. Such understanding 8Such movement appears to be fairly rare in our sample.
coupled with information on the contexts in which policies and 9We hypothesize that some combination of district policies and
interventions are most successful will contribute to improvedlocal labor market conditions produce the substantial cross-district vari-
targeting of resources and interventions to better educationalation that we observe in retention. While differences in the experience
outcomes for students. distribution could drive our results, we observe substantial variation
in retention even among similarly experienced teachers, as we describe
later in the paper. Furthermore, we use different years of data for differ-
NOTES
ent districts, and our data panel includes 2007-2008 (the beginning of
The authors thank Jon Fullerton and the Center for Education
the U.S. financial crisis). Therefore, differences across time rather than
Policy Research at Harvard University as well as the educational agen-
true differences across districts may influence our results. To investigate
cies who provided data for this project. We thank Raegen Miller
this and
possibility, we examined the extent of variation in teacher retention
Rachel Perera from Teach For America (TFA) for providing informa-
across sites within each year. Even using this approach, which controls
tion on TFA teachers in the districts we study. We also thank for
Richard
across-year differences, we still observe substantial variation across
Murnane, AEFP and APPAM conference participants, and anonymoussites. We also focused on a subset of districts with relatively long panels
reviewers for helpful feedback on the paper. Any and all errors and
are examined
our retention rates in a set of common years; separately, we
own.
examined retention rates in data from before the recession. Both analy-
^ee http://www2.ed.gov/policy/elsec/guid/secletter/l40707
ses showed substantial cross-district variation, confirming the pattern
.html for more information about the motivation of
and
our requirements
main findings. Wefor
take up the question of other predictors of
Educator Equity Plans. variation in teacher retention rates in the following sections.

NOVEMBER 2017 | 443

This content downloaded from 152.118.24.31 on Wed, 20 Mar 2024 07:51:29 +00:00
All use subject to https://about.jstor.org/terms
10Among teachers who leave, male and female teachers return administrative costs of hiring teachers in this analysis as a way to contex-
at statistically indistinguishable rates. White teachers are less likely to tualize the differences that we observe between districts.

return than their non-White counterparts, first-year teachers are less 17To do this, we estimate our model separately for every two-year
likely to return than more experienced teachers, and teachers who hold period in our data. Each two-year model is estimated using all teachers
alternative teaching certifications are less likely to return than those present in the data during that two-year period. As a result, we generate a
with traditional certifications. different 'ik for every two-year period. We retain only the corresponding
11 Note that the terciles of teacher value-added are not directly to each teachers first two years in the district. If a teacher leaves after one
comparable across districts given that our estimates of value-added are year, then the estimate is only based on the first year of teaching.
relative to the pool of teachers within the same district.
12We complement our analysis with a more "kitchen sink" REFERENCES
approach to examining these cross-district predictors, estimating the
relationship between our six outcomes and 41 district-average mea- Aliensworth, E., Ponisciak, S., & Mazzeo, C. (2009). The schools teach-
sures. Across these 246 relationships, only 10 are statistically significant, ers leave: Teacher mobility in Chicago public schools . Chicago, IL:
similar to what we would expect by chance. Here we also conducted a Consortium on Chicago School Research at the University of Chicago.
permutation test in which we generated a simulated null distribution, Birkeland, S. E., & Curtis, R. (2006). Ensuring the support and devel-
assigning the values of the predictors at random. The results of this test opment of new teachers in the Boston Public Schools : A proposal to
confirm that our small number of significant findings are due to chance. improve teacher quality and retention. Boston, MA: Boston Public
The full results from these analyses are available on request. One excep- Schools.

tion to this pattern is an indicator for whether the district bargains col- Boyd, D., Grossman, P., Ing, M., Lankford, H., Loeb, S., & Wyckoff, J.
lectively. This predictor is strongly related to teacher retention, echoing (201 1). The influence of school administrators on teacher retention
results from Goldhaber, Lavery, and Theobald (2016) that collective decisions. American Educational Research Journal. 48. 303-333.

bargaining may play a role. However, the limited number of districts Boyd, D., Lankford, H., Loeb, S., & Wyckoff, J. (2002). Initial matches,
without teacher collective bargaining makes these conclusions quite transfers, and quits: Career decisions and the disparities in aver-
tenuous as they are driven entirely by District C, which has the lowest age teacher qualifications across schools. Stanford CEPA Working
retention rates across the board and happens to be in a state without Paper. Retrieved September 30, 2014, from https://cepa.stanford.
bargaining. Thus, we do not make too much of this result. edu/content/initial-matches-transfers-and-quits-careerdecisions-
13Three results suggest that the differential presence of Teach For and-disparities-average-teacher-qualifications-across-schools.
America (TFA) teachers is not an important threat (we thank Raegen Miller Boyd, D., Grossman, P., Lankford, H., Loeb, S., & Wycoff, J. H.
and Rachel Perera from TFA for providing data on TFA corps members). (2008). Who leaves ? Teacher attrition and student achievement
First, particularly early in the career, the retention rates for TFA teachers (NBER Working Paper No. W14022). Cambridge, MA: NBER.
are not substantially different from those of novice teachers in our sample; Boyd, D., Lankford, H., Loeb, S., & Wyckoff, J. (2005). Explaining the
this echoes results from other studies (e.g., Donaldson & Johnson, 201 1). short careers of high-achieving teachers in schools with low-perform-
Second, we see substantial variation in retention rates among the nine dis- ing students. The American Economic Review, 95(2), 166-171.
tricts that did not employ any TFA teachers over the period we studied. Brewer, D. J. (1996). Career paths and quit decisions: Evidence from
Third, even among districts with TFA teachers, we see no relationship teaching . Journal of Labor Economics, 14(2), 313-339.
between the prevalence of TFA and novice retention rates. Similarly, we Clotfelter, C. T., Ladd, H. F., & Vigdor, J. L. (201 1). Teacher mobil-
find limited evidence that differences in charter school prevalence explain ity, school segregation, and pay-based policies to level the playing
cross-district variation in turnover. In 2014-2015, the percentage of stu- field. Education Finance and Policy, 6(3), 399-438.
dents enrolled in charter schools varied from less than 10% to more than DeAngelis, K. J., & Presley, J. B. (2011). Toward a more nuanced
33% across the districts in our study (National Alliance for Public Charter understanding of new teacher attrition. Education and Urban
Schools, 2015). Although teacher turnover rates are typically higher in Society , 43(5), 598-626.
charter schools than traditional public schools (Stuit & Smith, 2012), we DeAvila, J., & Hobbs, T. D. (2017, September 6). Teacher short-
find a positive (albeit not statistically significant) correlation between the age prompts some states to lower the bar. Wall Street Journal.
percentage of students in each district enrolled in charter schools and aver- Retrieved from https://www.wsj.com/articles/teacher-shortage-
age within-district retention rates. prompts-some-states-to-lower-the-bar- 1 504699200
14For simplicity, we assume that districts maintain staffing lines for Donaldson, M. L., & Johnson, S. M. (2011). Teach For America
departing teachers and thus that districts' student enrollment and work- teachers: How long do they teach? Why do they leave? Phi Delta
force size remain constant. We also assume that teachers are replaced Kappan, 93(2), 47-51.
by novices. Feng, L., & Sass, T. R. (2012). Teacher quality and teacher mobility
15Across the years we examine, District C hired an average of 650 (Andrew Young School of Policy Studies Research Paper Series
novice teachers annually. To keep those 650 slots filled for five years, 12-08). Retrieved from https://papers.ssrn.com/sol3/papers.cfm?
the district would need to hire 1 ,456 teachers, compared to only 1 ,047 abstract id=2020373&rec= 1 &srcabs=902032&ale=7&pos=8
hires if their retention rates were similar to those in District B. Goldhaber, D., Gross, B., & Player, D. (2011). Teacher career paths,
16Clearly, from a true cost-benefit perspective, we must also account teacher quality, and persistence in the classroom: Are public schools
for the costs associated with changing a district s retention rate, the net keeping their best? Journal of Policy Analysis and Management,
difference in productivity, and the net difference in salary. In other words, 30(1), 57-87.
we might expect the lower productivity of a replacement teacher to be Goldhaber, D., Lavery, L., & Theobald, R. (2016). Inconvenient truth?
offset by lower salary. However, as Milanowski and Odden (2007) note, Do collective bargaining agreements help explain the mobility
providing relative estimates of these differences is quite difficult because of teachers within school districts? Journal of Policy Analysis and
doing so relies on quite substantial assumptions about the dollar value Management, 35(4), 848-880.
of productivity differences (not to mention the technical challenges Gray, L., & Taie, S. (2015). Public school teacher attrition and mobil-
involved with estimating these productivity differences). As such, rather ity in the first five years: Results from the first through fifth waves
than provide a cost-benefit analysis, we simply focus on differences in the of the 2007-08 Beginning Teacher Longitudinal Study (NCES

444 I EDUCATIONAL RESEARCHER

This content downloaded from 152.118.24.31 on Wed, 20 Mar 2024 07:51:29 +00:00
All use subject to https://about.jstor.org/terms
2015-337). Washington, DC: National Center for Education Marinell, W. H., & Coca, V. M. (2013). Who stays and who leaves? Findings
Statistics. Retrieved from http://nces.ed.gov/pubs2015/2015337.pdf from a three-part study of teacher turnover in NYC middle schools. New
Grissom, J. A., & Reininger, M. (2012). Who comes back? A longitu- York, NY: Research Alliance for New York City Schools.
dinal analysis of the reentry behavior of exiting teachers. Education Milanowski, A. T., & Odden, A. R. (2007). A new approach to the
Finance and Policy, 7(4), 425-454. cost of teacher turnover (Working Paper No. 13). Seattle, WA:
Gritz, R. M., & Theobald, N. D. (1996). The effects of school district School Finance Redesign Project, Center on Reinventing Public
spending priorities on length of stay in teaching. Journal of Human Education.
Resources , 31, 477-512. Mont, D., & Rees, D. I. (1996). The influence of classroom charac-
Hanna, R., & Pennington, K. (2015). Despite reports to the con- teristics on high school teacher turnover. Economic Inquiry, 34(1),
trary, new teachers are staying in their jobs longer. Retrieved from 152-167.
https :/ / www. americanprogress .org / issues/ education/ news/ 2015/ Murnane, R. J., & Olsen, R. J. (1990). The effects of salaries and
01/08/1 0342 1 / despite-reports-to-the-contrary-new-teachers-are- opportunity costs on length of stay in teaching: Evidence from
staying-in-their-jobs-longer/ North Carolina. Journal of Human Resources, 25, 106-124.
Hanushek, E. A., Kain, J. F., & Rivkin, S.G. (2004). Why public Murnane, R., Singer, J., & Willett, J. (1989). The influences of salaries
schools lose teachers .Journal of Human Resources, 39(2), 326-354. and "opportunity costs" on teachers' career choices: Evidence from
Harris, D. N., & Adams, S. J. (2007). Understanding the level and North Carolina. Harvard Educational Review, 59(3), 325-347.
causes of teacher turnover: A comparison with other professions. National Alliance for Public Charter Schools. (2015). A growing move-
Economics of Education Review, 26(3), 325-337. ment: America's largest charter school communities. Washington,
Harris, D., & Sass, T. (2011). Teacher training, teacher quality, and DC: Author.
student achievement. Journal of Public Economics, 95, 7 98-812. National Commission on Teaching and America's Future. (2003). No
Ingersoll, R. M. (2001) Teacher turnover and teacher shortages: An dream denied: A pledge to America's children (Summary report).
organizational analysis. American Educational Research Journal, Washington, DC: Author.
38(3), 499-534. Newton, X. A., Rivero, R., Fuller, B., & Dauter, L. (201 1). Teacher sta-
Ingersoll, R. M. (2002). The teacher shortage: A case of wrong diag- bility and turnover in Los Angeles: The influence of teacher and school
nosis and wrong prescription. NASSP Bulletin, 56(631), 16-31. characteristics (Working paper). Stanford, CA: Policy Analysis for
Ingersoll, R. M. (2003). Is there really a teacher shortage? Retrieved from California Education.
http : / / repository, upenn . edu/ cgi / viewcontent. cgi?article= 1 1 33 & Papay, J. P., & Kraft, M. A. (2015). Productivity returns to experience
context=gse_pubs in the teacher labor market: Methodological challenges and new
Ingersoll, R. M., Merrill, L., & Stuckey, D. (2014). Seven trench: The evidence on long-term career growth. Journal of Public Economics,
transformation of the teaching force, updated Apńl 2014 (CPRE 130, 105-119.
Report No. RR-80). Retrieved from http://www.cpre.org/sites/ Perda, D. (2013). Transitions into and out of teaching: A longitudinal
default/ flles/workingpapers/ 1 506_7 trendsapril20 1 4.pdf analysis of early career teacher turnover (Unpublished doctoral dis-
Jacob, B.bA. (2007). The challenges of staffing urban schools with sertation). University of Pennsylvania, Philadelphia.
effective teachers. Future of Children, /7(1), 129-153. Raue, K., & Gray, L. (2015). Career paths of beginning public school teach-
Johnson, S. M., Berg, J. H., & Donaldson, M. L. (2005). Who stays ers: Results from the first through fifth waves of the 2007-08 Beginning
in teaching and why: A review of the literature on teacher retention. Teacher Longitudinal Study (NCES 2015-196). Washington, DC:
Cambridge, MA: The Project on the Next Generation of Teachers, National Center for Education Statistics. Retrieved from http://
Harvard Graduate School of Education. nces.ed.gov/pubs20 1 5/20 151 96.pdf
Johnson, S. M., & Birkeland, S. E. (2003). Pursuing a "sense of suc- Rebora, A. (2016). Faced with deep teacher shortages, Clark County,
cess": New teachers explain their career decisions. Amerìcan Nev., district looks for answers. Education Week, 35(19). Retrieved
Educational Research Journal, 40, 581-617. from http://www.edweek.org/ ew / articles/ 20 1 6/0 1 / 27/faced-with-
Johnson, S. M., Kraft, M. A., & Papay, J. P. (2012). How context deep-teacher-shortages-clark-county.html
matters in high-need schools: The effects of teachers' working Riggs, L. (2013, October 18). Why do teachers quit? The Atlantic.
conditions on their professional satisfaction and their students' Retrieved from http://www.theatlantic.com/education/
achievement. Teachers College Record, 114( 10), 1-39. archive/20 13/1 0/why-do-teachers-quit/280699/
Keller, B. (2007). Oft-cited statistic likely inaccurate. Education Rockoff, J. (2004). The impact of individual teachers on student
Week, 26(41), 30. Retrieved from http://www.edweek.org.ezp- achievement. Evidence from panel data. American Economic
prodl.hul.harvard.edu/ew/articles/2007/06/13/4l5year-sl.h26 Review, 94(2), 247-252.
.html?qs=Oft-Cited+Statistic Ronfeldt, M., Loeb, S., & Wyckoff, J. (2013). How teacher turn-
Kim, S. E., & Barg, D. (2010). Reducing music teacher turnover and its over harms student achievement. Amerìcan Educational Research
consequences (Boston University Music Education Policy Brief No. Journal, 50(1), 4-36.
2). Retrieved from http://www.bu.edu/muedpolicyproject/brief2 Shen, J. (1997). Teacher retention and attrition in public schools:
.pdf Evidence from SASS91 . The Journal of Educational Research, 91(2),
Kirby, S. N., Berends, M., & Naftel, S. (1999). Supply and demand of 81-88.

minority teachers in Texas: Problems and prospects. Educational Simon, N., & Johnson, S. M. (2015). Teacher turnover in high-
Evaluation and Policy Analysis, 2/(1), 47-66. poverty schools: What we know and can do. Teachers College Record,
Ladd, H. (2011). Teachers' perceptions of their working condi- 117(3), 1-36.
tions: How predictive of planned and actual teacher movement? Stinebrickner, T. R. (1998). An empirical investigation of teacher attri-
Educational Evaluation and Policy Analysis, 33(2), 235-261. tion. Economics of Education Review, 17(2), 127-136.
Lankford, H., Loeb, S., & Wyckoff, J. (2002). Teacher sorting and Stuit, D. A., & Smith, T. M. (2012). Explaining the gap in charter
the plight of urban schools: A descriptive analysis. Educational and traditional public school teacher turnover rates. Economics of
Evaluation and Policy Analysis, 24(1), 37-62. Education Review, 31(2), 268-279.

NOVEMBER 2017 | 445

This content downloaded from 152.118.24.31 on Wed, 20 Mar 2024 07:51:29 +00:00
All use subject to https://about.jstor.org/terms
Sutcher, L., Darling-Hammond, L., & Carver-Thomas, D. (2016). A ANDREW BACHER-HICKS, EdM, is a PhD student in public policy
coming crisis in teaching ? Teacher supply , demand, and shortages in at the Harvard Kennedy School, 79 JFK St., Cambridge, MA 02138;
the U.S. Palo Alto, CA: Learning Policy Institute. abacherhicks@g. harvard.edu. His research focuses on teacher labor mar-
Theobald, N. D. (1990). An examination of the influence of personal, kets, with particular emphasis on questions regarding equitable access
professional, and school district characteristics on public school to high-quality educators in public K-12 systems.
teacher retention. Economics of Education Review , 9(3), 241-250.
LINDSAY C. PAGE, EdD, is an assistant professor of research meth-
Toppo, G. (2012). More teachers green in the classroom. USA Today.
odology at the University of Pittsburgh School of Education, 5918
Retrieved from http://usatoday30.usatoday.com/news/health/
WWPH, 230 South Bouquet Street, Pittsburgh, PA 1 5260; lpage@pitt
backtoschool/story/20 1 2-09-05/new-teachers/ 5758 1 638/ 1
.edu. Her research focuses on quantitative methods and their applica-
Useeem, E., Offenberg, R., & Farley, E. (2007). Closing the teacher
tion to questions regarding the effectiveness of educational policies and
quality gap in Philadelphia: New hope and old hurdles. Philadelphia,
PA: Research for Action. programs across the preschool to postsecondary spectrum.

Zinshteyn, M. (2015). Contrary to conventional wisdom, new teachers WILLIAM H. MARI NELL, EdD, is the director of education research at
are staying on the job. The Atlantic. Retrieved from http://www the One8 Foundation, 240 Newbury Street, Second Floor, Boston, MA
.theatlan tic. com/ education/ archive/20 1 5/04/contrary-to-conven 021 16; wmarinell@verizon.net. His research focuses on the careers of teach-
tional-wisdom-new-teachers-are-staying-on-the-job/39 1 985/ ers, policies and practices related to human capital, and the impact of par-
ticipating in K-12 applied learning educational opportunities.

AUTHORS
Manuscript received October 6, 2016
JOHN P. PAPAY, EdD, is an assistant professor of education and
Revisions received February 24, 2017,
economics at Brown University, 340 Brook Street, Box 1938,
and September 14, 2017
Providence, RI 02912 'john _papay@hrown.edu. His research focuses on
Accepted September 14, 2017
teacher policy, the economics of education, and teacher labor markets.

Appendix A
Table A1
Rates of Teacher Turnover From Existing State and District Studies
Turnover Time Grade Type of Include Definition of
Study Site Years Rate (%) Span Level Teacher Returners? Retention
Aliensworth, Ponisciak, and Chicago, IL 2002 to 2007 51 5-year rate Elementary All teachers No School level
Mazzeo (2009)
54 5-year rate High All teachers No School level
66 5-year rate Elementary First-year No School level
teachers

63 5-year rate High First-year No School level


teachers

Boyd, Grossman, Lankford, New York, NY 2000 to 2004 51 3-year rate 4 to 8 Entering No School level
Loeb, and Wyckoff (2008) teachers
Boyd et al. (2002)a New York 1990 cohort 54 3-year rate All grades New teachers No School level
34 3-year rate All grades New teachers No District level
26 3-year rate All grades New teachers No State level
DeAngelis and Presley (201 1 ) Illinois 1 987 to 2001 40 5-year rate All grades New teachers No School level
27 5-year rate All grades New teachers Yes School level
Klrby, Berends, and Naftel Texas 1979 to 1996 40 4-year rate All grades New teachers No State level
(1999)
~50 6-year rate All grades New teachers No State level
Lankford, Loeb, and New York 1993 to 1998 60 5-year rate All grades New teachers No School level
Wyckoff (2002)
Marinell and Coca (201 3) New York, NY 2000 to 2010 59 5-year rate Elementary All teachers No School level
66 5-year rate Middle All teachers No School-level
65 5-year rate High All teachers No School level
Newton, Rivero, Fuller, and Los Angeles, CA 2002 to 2009 57 5-year rate Elementary New teac
Dauter (2011)
62 5-year rate Secondary New teachers No School level
Useem, Offenberg, and Philadelphia, PA 1999 to 2005 70 6-year rate All grades New teachers No District level
Farley (2007)
84 6-year rate All grades New teachers No School level

''Authors' calculations from Table 7, assuming 20% male and 80% female teachers in the state.

446 I EDUCATIONAL RESEARCHER

This content downloaded from 152.118.24.31 on Wed, 20 Mar 2024 07:51:29 +00:00
All use subject to https://about.jstor.org/terms
Table A2
Summary of the Strategic Data Project Administrative Data Sets
School Years Average Percentage Percentage
Included Number of Eligible English
District (Spring Students Per for Free or Language Percentage Percentage Percentage
Name Year) Year Reduced Lunch Learner Hispanic Black White
District A 2004 to 2011 10,000 64 4 11 63 21
District B 2004 to 2011 40,000 77 8 14 57 25
District C 1999 to 2007 110,000 41 5 8 44 43
District D 2007 to 2011 20,000 50 8 14 42 37
District E 2000 to 201 3 60,000 78 5 10 69 17
District F 2007 to 2011 10,000 49 9 16 45 35
District G 2008 to 2011 30,000 49 8 24 10 60
District H 2008 to 2011 80,000 69 28 56 16 22
District I 2009 to 2011 40,000 47 7 10 24 58
District J 2009 to 2011 100,000 59 5 5 36 52
District K 2005 to 2010 200,000+ 74 37 73 11 9
District L 2004 to 2011 200,000+ 71 14 39 31 14
District M 2004 to 2011 30,000 79 8 21 65 12
District N 2004 to 2011 20,000 74 8 11 53 31
District 0 2004 to 2011 30,000 70 15 49 26 18
District P 2003 to 2011 130,000 28 8 10 29 54
Average of Districts A through P 1999 to 201 3 149,288 61 11 23 39 32
Average of 100 largest urban districts 201 5 to 201 5 108,806 56 13 22 33 33
Average of all U.S. public districts 201 5 to 201 5 2,696 51 9 25 16 50

Note. All information comes from National Center for Education Statistics (http://nces.ed.gov/ccd/elsi/). To mas
approximate number of students per year student, rounded to the nearest 1 0,000 students. All averages are

Table A3
Predictors of Teacher Retention Outcomes

All Teachers Novice Teachers

1-Year 3-Year 5-Year 1-Year 3-Year 5-Year


Retention Retention Retention Retention Retention Retention

Student characteristics
Percentage of English language learner students -0.429 0.208 0.388 2.032 1 .41 2 2.359
(0.536) (1.177) (2.071) (1.665) (2.213) (3.501)
Percentage of students eligible for free/reduced price 0.554 2.082 3.1 05 1 .744 3.541 8.587*
lunch

(0.527) (1.038) (1.594) (1.747) (2.056) (3.098)


Percentage of students Hispanic -0.407 0.41 0.848 1.064 1.202 2.266
(0.537) (1.173) (2.013) (1.786) (2.276) (3.479)
Percentage of students Black 0.722 0.56 1.58 -0.174 -0.09 1.755
(0.513) (1.169) (2.127) (1.908) (2.424) (4.197)
Percentage of students White -0.232 -0.916 -3.453 -1.661 -1.554 -9.065
(0.544) (1.152) (2.282) (1.977) (2.548) (4.553)
Organizational characteristics
Student-teacher ratio -0.472 -0.741 -1.518 1.034 -0.907 -1.209
(0.533) (1.161) (1.820) (1.687) (2.162) (3.149)
Student support services per teacher -0.044 -0.388 -3.484 1.057 0.44 -5.937
(0.548) (1.173) (2.481) (1.688) (2.177) (4.163)
Salary and local labor market features
Regional unemployment rate 0.29 1.064 0.423 2.432 4.242 4.074
(0.542) (1.143) (3.113) (2.371) (2.881) (5.304)
Salary 0.735 1.815 2.53 -0.344 1.307 4.365
(0.511) (1.074) (1.670) (1.758) (2.201) (3.291)

Note. Each cell represents a separate bivariate regression, where the sample is the 16 districts in this study
(district-wide retention rates). Rows correspond to nine different independent variables, which are standard
our sample of 16 districts. Standard errors are displayed in parentheses below coefficients. With the except
Center for Educational Statistics. Unemployment rates come from the Bureau of Labor Statistics.
*p < .05.

NOVEMBER 2017 | 447

This content downloaded from 152.118.24.31 on Wed, 20 Mar 2024 07:51:29 +00:00
All use subject to https://about.jstor.org/terms
Appendix B cohort (Ssg), including average baseline test scores and averages
of Xit. We decompose the idiosyncratic student-level error, vi k t ,
Value-Added Model
into teacher-level random effects, 'ik, classroom-level random
effects, ' and residual error, eř- ¿ ř . The teacher-level random
We estimate value-added for math teachers using student-level
administrative data from four districts, which include students'
effects, represent our empirical Bayes estimate of a teacher's
effectiveness
end-of-year test scores, students' demographic information, and in raising student test scores. Separate regression
the links between students and their teachers of record. We use are fit for each district, so teachers are only compared to
models
other teachers within their district. To avoid differences in value-
these data to estimate a student-level equation of the following
general form: added being driven by differences in how long teachers remain
in a district, our estimate of value-added for a given teacher, [ik>
A At - 4>-ia + Xiß + Pc, fi + s s, g, ti + <P g, t + va,i >
is based only on each teacher's first two years in the district.17
To explore how turnover rates differed by estimated teacher
where va,ř = ^ + e^ + £a,ř effectiveness, we focus on the four districts where we have linked
student-teacher data and at least 100 novice teachers in tested
gradeskand subjects. We then examine three-year retention rates
and Aikt is the outcome score for student i taught by teacher
for novice teachers whose estimated effectiveness was in the top
during school year t. In addition to grade by year fixed effects,
Wg,t , we include the following control variables: Ait_v and bottom thirds of the value-added distribution for their dis-
a cubic
polynomial of student is prior achievement; Xi ť a vector trict. We focus on novice teachers to avoid the issues involved
of indi-
with
cators for gender, race and ethnicity, free or subsidized disentangling the relationship between teacher experience,
school
lunch eligibility, English language learner status, andturnover,
special and effectiveness. We also focus on three-year year
turnover rates (as opposed to five-year) to ensure that we have a
education status; and Pc t and Ssgt, vectors of average characteris-
sufficiently large sample of teachers with value-added estimates.
tics of student is peers in the same class [Pc t) and school-grade

448 I EDUCATIONAL RESEARCHER

This content downloaded from 152.118.24.31 on Wed, 20 Mar 2024 07:51:29 +00:00
All use subject to https://about.jstor.org/terms

You might also like