Grade Assignments and the Teacher Pipeline: A Low-Cost Lever to Improve Student Achievement?

Abstract

Research on teacher stability typically focuses on the extent to which teachers remain in the same school, district, or the teaching profession from one year to the next. I investigate another facet of stabilitywhether teachers remain in the grade they teach. Drawing on administrative data from a large district in California, I find that high shares of teachers switch grades. Disproportionately, these are early career teachers who come from low-achieving or high-minority schools. Teachers who switch grades leave schools at higher rates than their colleagues and exhibit lower impacts on their students' achievement. For teachers who switch to a nonadjacent grade, these negative effects can wipe out any gains due to increased experience and can persist in the year after the switch occurs.

Full text

Educational Researcher, Vol. 44 No. 4, pp. 213 –227
DOI: 10.3102/0013189X15580944
© 2015 AERA. http://er.aera.net
MAY 2015
213
Introduction
Each year, school principals, leadership, and teaching teams
make decisions about how best to allocate their scarce resources
in the service of improving student outcomes. They may decide,
for example, on how small or large to make classes, whom to
hire, and what type of professional development to provide their
staff. Despite fierce debates about which—if any—of these
resources matter most (e.g., Greenwald, Hedges, & Lain, 1996;
Hanushek, 1997), a broad consensus has emerged on the relative
importance of investing in high-quality teachers (Kyriakides &
Creemers, 2008; Nye, Konstantopoulos, & Hedges, 2004;
Rivkin, Hanushek, & Kain, 2005). A key question, then, is how
schools can allocate resources to get these high-quality teachers
into classrooms. Research indicates that paying for education
master’s degrees may not be a wise investment (Wayne & Youngs,
2003), but focusing instead on developing teacher productivity
through experience, coaching, and relationships with high-quality
colleagues can improve teacher quality and student achievement
(Allen, Pianta, Gregory, Mikami, & Lun, 2011; Jackson &
Bruegmann, 2009; Rockoff, 2004).
Another potential avenue not yet fully explored is teacher
grade assignments. To the extent that researchers have studied
teacher assignments, they have focused primarily on whether or
not teachers hold qualifications in the subject they teach (e.g.,
Ingersoll, 2002). However, recent analyses indicate that teachers
switch grades at high rates; in multiple districts across the United
States, over 20% of teachers switch grades from one year to the
next (Jacob & Rockoff, 2011). In addition, these trends may
have negative consequences for teachers and students. Studies
indicate that accountability policies likely incentivize school
leaders to keep their most effective teachers in tested grades
(Boyd, Lankford, Loeb, & Wyckoff, 2008; Chingos & West,
2011; Fuller & Ladd, 2012) and that switching grades is associ-
ated with declines in teachers’ effectiveness at raising student
achievement (Jacob & Rockoff, 2011; Ost, 2014) and increases
in teacher turnover (Ost & Schiman, 2015).
Attention to and modification of grade assignment policies
and practices has the potential to be quite valuable to schools,
particularly when compared to other programs seeking to har-
ness teacher productivity to improve educational outcomes for
students. Although individualized coaching can cost around
$4,000 per teacher (Allen et al., 2011) and increasing the quality
of teachers’ peers requires large-scale recruitment efforts, the
decision of whether or not to have a teacher switch grades may
have no direct and few indirect costs.
To explore the extent to which grade assignments can be a
lever for change in the teacher pipeline and for student
achievement, I draw on a 10-year panel of administrative data
580944EDRXXX10.3102/0013189X15580944Educational ResearcherMonth XXXX
research-article2015
1Center for Education Policy Research, Harvard Graduate School of
Education, Cambridge, MA
Grade Assignments and the Teacher Pipeline:
A Low-Cost Lever to Improve Student Achievement?
David Blazar1
Research on teacher stability typically focuses on the extent to which teachers remain in the same school, district, or the
teaching profession from one year to the next. I investigate another facet of stability—whether teachers remain in the
grade they teach. Drawing on administrative data from a large district in California, I find that high shares of teachers
switch grades. Disproportionately, these are early career teachers who come from low-achieving or high-minority schools.
Teachers who switch grades leave schools at higher rates than their colleagues and exhibit lower impacts on their students’
achievement. For teachers who switch to a nonadjacent grade, these negative effects can wipe out any gains due to
increased experience and can persist in the year after the switch occurs.
Keywords: descriptive analysis; econometric analysis; educational policy; grade assignments; retention; returns to
experience; school/teacher effectiveness; teacher education/development; teacher retention; teacher stability
FEATURE ARTICLES
at Harvard Libraries on August 11, 2015http://er.aera.netDownloaded from

214 EDUCATIONAL RESEARCHER
from a large urban school district in California. I focus on ele-
mentary school teachers who are most likely to teach one self-
contained grade in a given year. In order to understand the
context within which grade switching occurs, I first explore
trends in teacher grade assignments. I focus on the extent to
which grade reassignments occur more frequently for inexperi-
enced teachers and those with low value-added scores in years prior
to the switch, or for those teachers who work in high-risk schools
(i.e., low-achieving, low-income, and/or high-minority schools).
Second, I explore the relationship between grade switching and
teachers’ long-term career trajectories. Specifically, I examine
whether there is a relationship between switching grades and
teachers’ growth in effectiveness at raising student achievement or
their retention in a school or in the district.
In order to help inform decision-making processes, I focus on
two additional features of grade reassignments: switching to an
adjacent versus a nonadjacent grade and the lasting effects of
switching grades. It is possible that teachers who switch to a
grade far from their original assignment experience more disrup-
tion than those who switch to an adjacent grade. If this is the
case, then school leaders may aim to avoid the latter more than
the former. Relatedly, if disruptions due to grade reassignment
last for multiple years, this would be much more problematic
than a scenario where disruptions fade out quickly.
Background
Recent research highlights a variety of explanations for why
teachers switch grades from one year to the next. Administrators
may reassign teachers to different grades based on need, due
either to changes in cohort size or teacher turnover (Jacob &
Rockoff, 2011). Relatedly, administrators may aim to match
teachers to a specific group of students. An array of studies also
suggests that grade assignments are related to accountability poli-
cies that emphasize tested grades over untested ones. Low-quality
teachers are more likely than their colleagues to be moved to an
untested grade, and these trends are more pronounced following
the enactment of accountability policy (Boyd et al., 2008;
Chingos & West, 2011; Fuller & Ladd, 2012). Furthermore, the
decision to switch grades may be a voluntary one made by teach-
ers who choose to work with a specific age group or want a change
of pace (Jacob & Rockoff, 2011).
The multitude of factors for grade reassignment suggests no
single outcome for teachers who experience this event. Many of
the scenarios described above have theoretical benefits for teach-
ers and their students. At the same time, qualitative research,
theory, and intuition highlight a number of potential conse-
quences. Interviews with teachers indicate that understanding of
curricula and content is a primary factor in their job satisfaction,
perceptions of their teaching skill, and decisions to stay in their
school and in the profession (Johnson & Birkeland, 2002;
Kauffman, Johnson, Kardos, Liu, & Peske, 2002). Therefore,
adapting to new curricula, content, and possibly a new set of
grade-level colleagues as a result of switching grades may create
disruptions in their satisfaction and professional growth.
Relatedly, research on teaching across subject areas highlights a
need for teachers to negotiate differences in how content is deliv-
ered, including the level of cognitive demand (Graeber, Newton,
& Chambliss, 2012). Differences in content standards and expec-
tations across grade levels—even within the same subject—likely
bring similar challenges. Finally, teachers who switch to a grade
far from their original one also must adapt to differences in devel-
opmental needs of students at different ages (Fischer, 1980).
A handful of quantitative analyses indicate that these and other
potential challenges associated with switching grades are negatively
related to teacher effectiveness and retention. Drawing on data
from North Carolina, Ost (2014) examines differences in elemen-
tary teachers’ returns to experience—that is, how much they con-
tribute to gains in student achievement over time—for those who
remain in the same grade versus those who switch. His value-added
model with teacher fixed effects attempts to account for sorting of
students into classrooms. Results indicate that teachers who repeat-
edly teach the same grade from one year to the next have returns to
experience roughly one-third to one-half larger (i.e., 0.01 to 0.04
standard deviations in student achievement growth) than general
returns to experience. Jacob and Rockoff (2011) discuss similar
findings from unpublished analyses of New York City administra-
tive data. However, in this policy-oriented discussion paper, the
authors do not describe their sample or estimation strategy. Using
similar data as above, Ost and Schiman (2015) find that teachers
who switch grades also leave their school at the end of that year over
3 percentage points higher than teachers who do not switch grades.
These negative trends may be most problematic when viewed
from an educational equity perspective. Brummet, Gershenson,
and Hayes (2013) find that grade switchers in Michigan tend to
come from a unique subset of the workforce. Teachers who
switch grades are more likely to work in schools that are low
performing and have high percentages of minority students.
Therefore, students in these schools who are assigned to a teacher
who recently switched grades may be even worse off than stu-
dents in other schools.
Despite a growing literature highlighting suboptimal outcomes
associated with switching grades, questions remain about trends in
grade reassignment and its effect on teachers and students. There
may be additional predictors of grade reassignment, such as income
level, which also are related to issues of equity. In addition, studies
typically have focused on rates of switching from a tested to an
untested grade because of its relevance to policy; however, they
have not explored switching from adjacent versus nonadjacent
grades, which may be more relevant to teachers’ experiences and
their ability to grow as educators. Teachers who switch to a grade
far away from their original one may experience a much steeper
learning curve than those who switch to a grade close to their origi-
nal assignment. Relatedly, it is not clear the extent to which teach-
ers recover losses to productivity in years after the switch occurs, or
whether higher rates of turnover fade out over time.
Therefore, I build on existing work by asking three sets of
related research questions: (1) Do inexperienced teachers, those
with low value-added scores, or those who work in high-risk schools
(i.e., high-turnover, low-achieving, and/or low-income schools)
switch grades at higher rates than their colleagues in a way that
may exacerbate inequality? (2) Is grade reassignment related to
teachers’ long-term career trajectories—namely, their productiv-
ity or retention in their school or in the district? (3) Do these
trends differ for those who switch to a grade adjacent to their
original assignment versus those who switch to a grade farther
at Harvard Libraries on August 11, 2015http://er.aera.netDownloaded from

MAY 2015
215
away? An additional contribution of this work is to combine
these questions into a single set of analyses. As such, I am able to
draw on descriptive patterns in grade reassignments to inform
results exploring teachers’ longer term trajectories.
Methods
Data and Sample
In order to answer my research questions, I draw on administra-
tive records from a large urban school district in California. This
10-year panel of data beginning in the 2002-03 school year
includes human resource information for all teachers, demo-
graphic and test-score data (where relevant) for all students, and
course files that allow me to connect teachers to students and
identify the grades they teach.
Although I have access to information on all teachers and
students in the district, I limit this sample in ways that are most
conducive to answering my research questions. Across all analy-
ses, I focus on elementary teachers who are most likely to teach
one self-contained grade in a given year. Based on grade identifi-
cation that I describe below, over 90% of all teacher-year obser-
vations at the elementary level are attached to just one grade,
compared to roughly 60% and 20% of observations at the mid-
dle- and high-school levels, respectively. I also limit the sample
to teachers in core academic subject areas (i.e., English language
arts [ELA], math, science, social studies), excluding those who
teach self-contained special education, physical education, art,
music, or supplemental courses. I describe below additional
restrictions for individual analyses, such as teachers who are
observed as novices at some point in the dataset and those with
test-score data. These samples include between roughly 2,000
and 22,000 teachers and between roughly 7,000 and 129,000
teacher-year observations.
For each of these teachers, I identify the grade(s) taught in
each school year based on course rosters and student grade infor-
mation provided by the district. Where applicable, I verify stu-
dents’ grade level against information from the standardized tests
they took at the end of each school year. For 89% of core subject
courses, all students within a given course have the same grade
designation. Therefore, I am confident that this grade level is
attached correctly to the assigned teacher. An additional 8% of
courses include students at different grade levels. Although some
of these may be mixed-level courses (e.g., a combined fourth-
and fifth-grade class), two-thirds are heavily skewed toward one
grade. That is, the average grade level across individual students
in the class is within 0.33 of the integer value. Therefore, in these
instances, I use the modal value of student grade levels in the
course as the class grade. I also consider alternative ways of
addressing this issue—using smaller bandwidths around the
integer value and excluding these cases altogether—and find that
rates of switching do not change substantively. Finally, 3% of
courses are missing grade data for all students. However, in all of
these cases, I am able to determine teachers’ grade level in that
year through a course in another subject.
I define grade switches in instances where teachers move to a
self-contained grade in the current year from a different self-
contained grade in the prior school year. As the vast majority of
elementary teachers teach just one grade in both the current and
prior years, these switches are straightforward to identify. For
those few teachers who teach multiple grade levels either in the
prior or current year, I identify switches in instances where a
teacher adds a grade level to his or her teaching load. For exam-
ple, a teacher who teaches fourth grade one year and both fourth
and fifth the next is identified as switching grades. A teacher who
teaches fourth and fifth grade in one year but then only teaches
fourth grade the next year is not identified as switching grades,
as no new grades are added to their course load.
I exclude from this definition “loopers” (i.e., teachers who
move with students from one grade to the next), teachers who
switch both grades and schools, and teachers who return to a
grade that they taught in any prior year because of the unique
nature of these types of switches; in particular, these switches
may not lead to the same challenges associated with grade reas-
signment described above. This third exclusion also ensures that
teachers who leave teaching for a short period of time (e.g.,
maternity leave) but then return to the classroom are not identi-
fied as switching grades. For those teachers who I identify as
switching grades in a given year, I further classify these as switches
to an adjacent versus a nonadjacent grade.
Analysis
Estimating Trends in Grade Reassignment
My first research question aims to understand trends in grade
reassignments and the extent to which experiencing this event
may be inequitably distributed amongst certain types of schools
and teachers. First, I present rates of switching by school year,
grade level that teachers switch from, and experience level. I also
disaggregate rates by those who switch to an adjacent versus non-
adjacent grade.
Next, I explore whether teachers who work in high-risk
schools (i.e., high-turnover, low-achieving, and/or low-income
schools) switch grades at higher rates than teachers who work in
other schools. I do so in a regression framework in order to con-
trol for differential rates of switching by teaching experience,
given evidence that inexperienced teachers are more likely to
work in high-risk schools (Clotfelter, Ladd, & Vigdor, 2005;
Lankford, Loeb, & Wyckoff, 2002). Specifically, I estimate the
following equation using Ordinary Least Squares (OLS):
SWITCH SCHOOL CHARACTERISTIC
f EXPERIENCE
jt
qj
t
q
jt
=+
()
=
∑
κ
ψ
π
2
4
()++
ε
jt , (1)
where SWITCHjt is an indicator for whether or not teacher
j switches grades in year t (from year t – 1), which I also
disaggregate by adjacent and nonadjacent switches in separate
models. I estimate expected values of rates of grade
reassignment by quartiles of school characteristics,
ψ
qj
t
qSCHOOL CHARACTERISTIC
=
∑2
4
. These main pre-
dictor variables include average school achievement in math
(correlated with average achievement in ELA above 0.95); per-
cent low-income, operationalized as the percent of students
at Harvard Libraries on August 11, 2015http://er.aera.netDownloaded from

216 EDUCATIONAL RESEARCHER
eligible for free- or reduced-price lunch; and percent minority,
operationalized as the percent of non-White students. I calculate
these averages/percents using all available data, given limited
variation in these school characteristics across years. I include a
time subscript t on these characteristics—indicating time vari-
ance—to account for teachers who switch schools at some point
in the dataset. For teachers who stay in the same school, these
school characteristics are constant across time points. I control
for teaching experience, f(EXPERIENCEjt ), modeled flexibly as
a series of dummy variables that are absorbed in the model.
Finally, I examine whether teachers with low value-added
scores switch grades at higher rates than their higher quality
peers. To do so, I estimate the relationship between switching
grades and teachers’ prior value-added score using a similar
model as above:
SWITCH PRIOR VALUE ADDED
jt
qj
ttjt
q
=+ ++
=
∑
κζ
ε
2
4
.
(2)
Here, I replace school characteristics with indicators for teach-
ers’ prior value-added quartile [see equation (3) below for the
sort of value-added model from which scores are derived]. To
increase the precision of my estimates, I utilize all years of data
prior to t (Goldhaber & Hansen, 2012; Koedel & Betts 2011;
Schochet & Chiang, 2013). I focus on value-added scores in
prior years in order to ensure appropriate directionality of this
relationship. That is, I explore whether teachers who are
observed as being ineffective at raising student achievement in
all prior years switch grades at the end of that year at higher
rates than their more effective colleagues. This analysis does not
aim to estimate the effect of switching grades on teachers’ effec-
tiveness. Because this sample consists of elementary teachers,
many of whom teach both math and reading, I average value-
added scores across subjects to create a composite measure of
teachers’ overall effectiveness at raising student achievement. I
do not run analyses separately by subject area, given that teach-
ers who switch grades almost always do so in both subjects
simultaneously. I control for a vector of school year indicators,
ηt, given that value-added scores are normed within a given year.
I also estimate models that include school fixed effects in order
to account for various factors at the school level that contribute
to grade reassignment.
For all of these analyses, I only include teachers in their sec-
ond year of teaching or higher, as these are teachers for whom it
is possible to switch grades. As teachers in their first year in the
classroom could not have switched grades from the prior year,
including them in these analyses would artificially attenuate
rates of switching. I also exclude school year 2002-03, as I am
not able to observe a switch in this first year of available data.
Estimating the Relationship Between Switching Grades
and Student Achievement
My second set of research questions examines the relationship
between switching grades and teachers’ long-term careers. First, I
estimate the relationship between switching grades and teachers’
effectiveness at raising student achievement over time. To do so, I
specify a value-added model similar to those used by Kane et al.
(2013) and Chetty, Friedman, and Rockoff (2014):
The outcome of interest is current-year test score, Aisgcjt, for stu-
dent i in school s, grade g, and class c with teacher j at time t. Test
scores are modeled as a function of students’ prior achievement,
Ait–1. I control for vectors of student covariates, Xit, and peer
covariates, Pct, for all students within classroom c at time t. I
include school fixed effects, φs, in order to account for some of
the school-level factors that are related to grade reassignments
and also affect student achievement. Results available upon
request are quite similar when I control instead for observable
school characteristics. I also include grade-by-year fixed effects,
ωgt, to account for scaling of tests at this level. I cluster standard
errors at the class level in order to account for the nested struc-
ture of the data.
Following Ost (2014) and Rockoff (2004), I include teacher
fixed effects, τj, to control for the fact that different teachers have
different underlying effectiveness, regardless of experience. As
Ost (2014) argues, teacher fixed effects account for the possibil-
ity that “unobserved teacher characteristics are correlated with
grade-specific experience” (p. 128).
I add parameters for years of experience, f(EXPERIENCEjt), to
examine the extent to which teachers’ ability to raise student
achievement improves over time. I include dummy variables for
Year 2 of teaching and up, with Year 1 as the reference group. This
allows me to identify average gains in student achievement attrib-
utable to each additional year of experience relative to average
gains for novice teachers. Others who have specified similar mod-
els note nearly perfect collinearity between experience and year
when teacher fixed effects are included (Harris & Sass, 2011; Ost,
2014; Rockoff, 2004). To address this, I replace experience dum-
mies with experience levels in some instances. Although the
authors cited above use a cutoff of 10 years of experience, I do so
for teachers with 7 years of experience given that I only am able to
follow novice teachers into their 9th year. Teachers who have 7 or
more years of experience are identified as having exactly 7 years of
experience.
I parameterize decrements to returns to experience for those
teachers who switch grades in a few key ways that align with my
research questions. The simplest way to estimate this relation-
ship would be to include a single dummy variable indicating
whether or not a teacher switched grades from the prior school
year. This is the same as the dummy variable SWITCHjt,
described above. For example, if a teacher switched grades from
Year 1 to Year 2, this variable would take on a value of 1 in that
teacher’s second year. One should interpret the coefficient on
this binary variable as any losses (or gains) to student achieve-
ment above and beyond factors already included in the model,
including teaching experience.
To examine whether these potential decrements vary by the
year that a teacher switches grades, I create a dummy variable for
each potential year of experience that a teacher could switch
grades. I do so by interacting SWITCHjt with experience dummy
variables. The result is a set of additional dummy variables,
(3)
AfAfEXPERIENCE
fEXPERIENCE
isgcjt it jt
jt
=+
()
()
+
()
()
+
()
−
κα π
β
1
*SSWITCH YEAR SWITCH YEAR
XP
jt jt
it ct sgtjis
27
+…
(
+
)
()
++++ ++
γυϕω τε
ggcjt
η
at Harvard Libraries on August 11, 2015http://er.aera.netDownloaded from

MAY 2015
217
SWITCH YEAR2jt through SWITCH YEAR7jt. This approach
also allows me to capture instances where teachers switch grades
more than once. As with teaching experience, teachers who
switch grades in their seventh year of teaching or higher are col-
lapsed into one category.
To examine whether these potential decrements are lasting or
fade out over time, I allow the indicators I just described to take
on a value of 1 in the year of the switch and all subsequent years.
Then, I interact them with the experience dummy variables.
Because these variables still take on a value of 0 in years prior to
the switch, I include the subscript t to indicate time variance.
These interactions are included in equation (3) and are my main
predictors of interest. For the sake of parsimony, I only follow
teachers for 2 years after the switch could occur. That is, if a
teacher switched grades in their second year, I estimate the decre-
ments to returns to experience in that year, in their third year, and
in their fourth year. These coefficients should be interpreted in the
same way that I describe above, with each indicating losses (or
gains) to student achievement above and beyond factors already
included in the model. I only include these interactions, and not
the main effect of switching grades, as they are exhaustive of all the
ways that teachers might switch grades. Finally, in separate models
I disaggregate these effects for teachers who switch to an adjacent
grade versus a nonadjacent grade, which are mutually exclusive for
each year of experience.
In this set of analyses, I restrict the sample to those teachers
who are observed as novices at some point in the dataset, which
ensures that I always am able to capture a grade switch if it
occurs. If teachers enter my dataset in their third year of teach-
ing, I cannot tell if they switched grades from their first to their
second year, or from their second to their third. I also am limited
to teachers who are observed teaching a tested subject and grade,
and whose students have prior-year test scores. This includes
teachers who teach math and/or ELA in Grades 3 through 5 in
school years 2003-04 through 2011-12.
Estimating the Relationship Between Switching Grades
and Retention
The hypothesized model to describe the relationship between
switching grades and teacher retention is given by equation (4):
LEAVE SWITCH fEXPERIENCE
COHORT
jstj
tj
t
yj
y
=+ +
()
+=
κ
βπ
δ
()
2004
20100
∑++
ϕε
sjst . (4)
Here, the outcome of interest, LEAVEjst, is one of two variables,
indicating whether teacher j left his or her school, s, or left the
district at the end of year t. In most instances, I identify these
teachers who leave their school or the district through human
resource records. As this information is not available in the first
2 years of available data, I define teachers who leave schools as
those who are observed in a different school year in the following
year, t + 1. In these same 2 years, I define teachers who leave the
district as those who are not observed in the dataset in any sub-
sequent year, T > t. Setting up the variable this way ensures that
teachers who take a leave for maternity or other reasons but then
return to the district are not identified as leavers.
Equations for leaving a school versus leaving the district are
estimated separately, with each modeling the linear probability
that teacher j leaves the school or district as a function of switch-
ing grades. In a teacher-by-year dataset, this setup is similar to
those used in discrete time hazard analyses (Singer & Willett,
1993). I estimate linear probability as opposed to hazard models
given empirical justification (Heckman & Snyder, 1996) and
ease of interpretation of estimates. As above, SWITCHjt is an
indicator for whether teacher j switches grades in year t (from
year t – 1) and is further disaggregated by adjacent and nonadja-
cent switches in separate analyses. I control for indicators for
teachers’ year of experience, with first year teaching as the refer-
ence group. I also include indicators for the year that a teacher
began teaching to control for idiosyncratic differences in leave
rates for a given cohort due to policy shocks such as the Great
Recession. Although experience dummies vary across years,
cohort indicators do not; therefore, these variables are not col-
linear. Finally, like Ost and Schiman (2015), I include school
fixed effects, φs, to account for differences across schools that
might be related both to grade switching and to turnover, such
as student and teacher composition.
The coefficient β on the indicator for whether or not teachers
switch grades, describes the percentage point difference of leav-
ing for those teachers who switch grades compared to those who
do not. Although this set of analyses only describes the differ-
ence in retention rates in the year of the switch itself, I also exam-
ine whether there is a relationship between grade switching and
retention in any year after the switch. To do so, I estimate addi-
tional models where I replace my main predictor with a new
variable that takes on a value of 1 in the year of a switch and all
subsequent years.
Restrictions to the sample are similar to those described above
in analyses that examine the relationship between switching and
student achievement. However, I am able to include teachers
outside of tested grades and subjects. In addition, I exclude
school years 2010-11 and 2011-12, given lack of human resource
data in these years and the fact that I am not able to impute data
by following teachers into subsequent years.
Results
Trends in Grade Reassignment
I begin the Results section by describing trends in grade reassign-
ment. Overall, I find that, year to year, elementary teachers
switch the grade in which they teach at high rates upwards of
17% (see Figure 1); lower rates around 12%-14% after the
2007-08 school year may reflect more general trends in transfer
and migration as a result of budget cuts and teacher firings
related to the Great Recession. Interestingly, switching to an
adjacent grade occurs slightly more frequently than switching to
a nonadjacent grade through the 2007-08 school year, but this
trend is reversed in the following school years.
Disaggregating grade switches by grade level—that is, the
grade teachers switch from—I find that rates steadily increase
across the elementary grades. Roughly 11% of kindergarten
teachers switch grades, compared to roughly 17% for fifth-grade
teachers. By design of schools, I find lower rates of adjacent
grade switching for kindergarten and fifth-grade teachers who
at Harvard Libraries on August 11, 2015http://er.aera.netDownloaded from

218 EDUCATIONAL RESEARCHER
teach at the extremes of the elementary grades and, therefore,
only can switch to one adjacent grade.
Disaggregating rates of switching by teachers’ level of experi-
ence, I find that rates of switching increase between the second
and fourth year of experience, from roughly 21% overall to
24%, and then decline steadily for teachers with more experi-
ence. By construction of my grade-switching variable, no teacher
can switch grades in his or her first year of experience; therefore,
this bin is excluded from the analysis. Teachers with 9 or more
years of experience switch grades at much lower rates of roughly
11%. Within each experience level, rates of adjacent and nonad-
jacent grade switching are roughly equal.
In addition, I find that rates of switching differ depending on
school and teacher characteristics. In Table 1, I present results for
school characteristics that are estimated from a regression frame-
work that controls for teaching experience, which is absorbed in
the model. Teachers who work in the lowest achieving schools,
operationalized as average prior-year achievement in math, switch
grades from one year to the next at higher rates than those who
work in the highest achieving schools. I find that roughly 16% of
teachers who work in bottom-quartile schools (i.e., low-achieving)
switch grades, compared to roughly 13% who work in top-quartile
schools (i.e., high-achieving). This differential of over 3 percent-
age points is statistically significant. Differentials between the
bottom quartile and the second and third quartiles are also statis-
tically significant but smaller in magnitude, of 1.3 and 1.7 per-
centage points, respectively. When examining rates of switching
to adjacent or nonadjacent grades, trends are similar, but differ-
entials between quartiles are smaller in magnitude given lower
rates of switching to an adjacent or nonadjacent grade overall.
When I examine trends for teachers who work in low-income
and high-minority schools, some differentials remain statistically
significant but are substantively smaller in magnitude. Roughly
13% of teachers who work in high-income schools (i.e., bottom
quartile of percent of students eligible for free- or reduced-price
lunch) switch grades, compared to roughly 15% of teachers who
work in low-income schools (i.e., top quartile of percent of stu-
dents eligible for free- or reduced-price lunch). However, these
differences disappear when controlling for average school math
achievement. Finally, 13% of teachers who work in low-minority
schools (i.e., bottom quartile of percent of non-White students)
switch grades, compared to 14% of those who work in high-
minority schools (i.e., top quartile of percent of non-White stu-
dents). As above, differentials between teachers who work in
top- versus bottom-quartile schools are similar when examining
rates of switching to adjacent versus nonadjacent grades.
In Table 2, I present differences in rates of switching by teach-
ers’ prior-year value-added quartile. Rates of switching are esti-
mated from a regression framework that controls for school year
given that value-added scores are normed within years, with 2003-
04 as the baseline year. For teachers who have a prior-year value-
added score in both math and ELA, I average these scores. Because
grade reassignment is a school-level decision, I look at trends both
across and within schools.
Here, I find that rates of switching are higher for teachers in
the bottom quartile of prior-year value-added than for their
more effective colleagues. Roughly 20% of bottom-quartile
teachers switch grades compared to roughly 15% of top-quartile
teachers. This trend remains when disaggregating by adjacent
versus nonadjacent grade switches and when comparing teachers
within schools.
Relationship Between Switching Grades and Teachers’
Long-Term Careers
In addition to observing differential rates of switching for spe-
cific types of teachers and schools, I examine the relationship
between switching grades and teachers’ longer term careers. This
is of particular importance given findings above that the most
vulnerable teachers—early career teachers, those with low prior
value-added scores, and those who work in high-risk schools—
switch grades at the highest rates.
FIGURE 1. Rates of grade switching by school year (top panel),
grade level teachers switch from (middle panel), and teaching
experience (bottom panel).
at Harvard Libraries on August 11, 2015http://er.aera.netDownloaded from

MAY 2015
219
I begin this analysis by examining the relationship between
grade switching and teachers’ effectiveness at raising student
achievement over time—often referred to as teachers’ “returns to
experience.” In Table 3, I present parameter estimates for average
returns to experience and deviations from this trajectory for
switching grades in a given year.
In order to interpret results related to switching grades, I first
describe average returns to experience. Importantly, I note two
trends. Consistent with prior research (Ost, 2014; Papay &
Kraft, in press), I find larger returns to experience in math than
in ELA. At the same time, the magnitude of these estimates are
substantively larger than those found in most analyses that
examine teacher productivity over time (Clotfelter, Ladd, &
Vigdor, 2006; Harris & Sass, 2011; Kraft & Papay, 2014; Ost,
2014; Papay & Kraft, in press; Rockoff, 2004; Wiswall, 2013).
In these studies, average returns to experience generally rise no
Table 1
Rates of Grade Switching by School Characteristics
Switch to Any Grade Adjacent Grade Nonadjacent Grade
By school achievement
Bottom quartile 15.9 7.9 8.0
Second quartile 14.6* 7.2* 7.5
Third quartile 14.2*** 7.0** 7.2*
Top quartile 12.5*** 6.0*** 6.5***
By school FRPL quartile
Bottom quartile 12.7 6.0 6.7
Second quartile 14.4*** 7.3*** 7.1
Third quartile 14.6*** 7.1*** 7.3*
Top quartile 15.4*** 7.6*** 7.9***
By school FRPL quartile, controlling for school achievement
Bottom quartile 14.1 6.7 7.4
Second quartile 14.3 7.2 7.0
Third quartile 14.0 7.0 7.0
Top quartile 14.2 7.4* 7.6
By school minority quartile
Bottom quartile 12.9 6.1 6.7
Second quartile 14.9** 7.4*** 7.4*
Third quartile 15.2** 7.5*** 7.7**
Top quartile 14.1* 7.0*** 7.0
Teacher observations 22,388 to 22,552
Teacher-year observations 127,251 to 128,657
Note. Estimates are calculated from regression models that include fixed effects for teaching experience, which are absorbed in the model. Sample includes all elementary
teachers who teach core academic subjects and are in their second year of teaching or higher.
*p < 0.05, **p < 0.01, ***p < 0.001, comparing the second through top quartiles to the bottom.
Table 2
Rates of Grade Switching by Teachers’ Prior Value-Added Quartile
Across Schools Within Schools
Switch to Any
Grade
Adjacent
Grade
Nonadjacent
Grade
Switch to Any
Grade
Adjacent
Grade
Nonadjacent
Grade
Bottom quartile 19.8 9.5 10.3 20.4 9.8 10.7
Second quartile 18.8~9.5 9.3* 19.4~9.8 9.8*
Third quartile 16.9*** 8.4** 8.4*** 17.4*** 8.7** 8.8***
Top quartile 14.6*** 8.2*** 6.4*** 15.0*** 8.4*** 6.8***
Teacher observations 10,944 10,944
Teacher-year observations 42,505 42,505
Note. Estimates are calculated from regression models that include fixed effects for school year, with 2003-04 as the baseline year. Sample includes all elementary
teachers who teach core academic subjects, are in their second year of teaching or higher, and have a value-added score in the year(s) prior to the switch. In instances
where a teacher has value-added data in both reading and mathematics, I average these scores.
~p < 0.10, *p < 0.05, **p < 0.01, ***p < 0.001, comparing the second through top quartiles to the bottom.
at Harvard Libraries on August 11, 2015http://er.aera.netDownloaded from

220 EDUCATIONAL RESEARCHER
Table 3
Returns to Experience for Teachers Who Do and Do Not Switch Grades
Math ELA
Switch to
Any Grade
Adjacent
Grade
Nonadjacent
Grade
Switch to
Any Grade
Adjacent
Grade
Nonadjacent
Grade
Second year teaching 0.099*** 0.102*** 0.102*** 0.047*** 0.046*** 0.046***
(0.014) (0.014) (0.014) (0.011) (0.011) (0.011)
Third year teaching 0.168*** 0.173*** 0.173*** 0.077*** 0.075*** 0.075***
(0.025) (0.025) (0.025) (0.019) (0.019) (0.019)
Fourth year teaching 0.217*** 0.225*** 0.225*** 0.086** 0.083** 0.083**
(0.036) (0.036) (0.036) (0.027) (0.027) (0.027)
Fifth year teaching 0.257*** 0.268*** 0.268*** 0.114** 0.111** 0.111**
(0.046) (0.047) (0.047) (0.035) (0.036) (0.036)
Sixth year teaching 0.283*** 0.295*** 0.295*** 0.119** 0.111** 0.111**
(0.057) (0.057) (0.057) (0.042) (0.043) (0.043)
Seventh year teaching 0.321*** 0.334*** 0.334*** 0.138** 0.126* 0.126*
(0.070) (0.071) (0.071) (0.052) (0.053) (0.053)
Switch in second year, second year teaching –0.026 –0.031 –0.007 0.003 –0.007 0.033
(0.020) (0.023) (0.040) (0.016) (0.018) (0.034)
Switch in second year, third year teaching –0.036~–0.018 –0.066 0.018 0.012 0.038
(0.022) (0.025) (0.041) (0.018) (0.021) (0.031)
Switch in second year, fourth year teaching –0.052* –0.044~–0.051 –0.018 –0.016 –0.016
(0.020) (0.024) (0.032) (0.018) (0.022) (0.029)
Switch in third year, third year teaching –0.036~–0.008 –0.078** –0.005 0.009 –0.030
(0.020) (0.025) (0.028) (0.017) (0.021) (0.024)
Switch in third year, fourth year teaching –0.029 0.011 –0.073** 0.016 0.026 0.006
(0.021) (0.031) (0.025) (0.017) (0.023) (0.022)
Switch in third year, fifth year teaching –0.008 –0.002 –0.016 –0.014 –0.010 –0.020
(0.023) (0.030) (0.030) (0.019) (0.025) (0.025)
Switch in fourth year, fourth year teaching –0.025 –0.017 –0.033 –0.003 –0.003 –0.002
(0.020) (0.024) (0.033) (0.018) (0.021) (0.028)
Switch in fourth year, fifth year teaching –0.022 –0.020 –0.018 –0.014 –0.041~0.035
(0.023) (0.027) (0.038) (0.021) (0.025) (0.032)
Switch in fourth year, sixth year teaching –0.015 –0.025 0.024 0.000 –0.005 0.032
(0.025) (0.029) (0.041) (0.021) (0.023) (0.035)
Switch in fifth year, fifth year teaching –0.059* –0.019 –0.118** –0.032 –0.018 –0.057
(0.024) (0.027) (0.041) (0.021) (0.022) (0.040)
Switch in fifth year, sixth year teaching 0.008 0.086** –0.076* –0.012 0.067** –0.093**
(0.027) (0.031) (0.039) (0.022) (0.026) (0.030)
Switch in fifth year, seventh year teaching
or higher
–0.015 0.022 –0.053 0.000 0.022 –0.033
(0.031) (0.039) (0.044) (0.024) (0.032) (0.032)
Switch in sixth year, sixth year teaching –0.070* –0.075* –0.025 –0.047* –0.039 –0.020
(0.031) (0.038) (0.046) (0.024) (0.031) (0.033)
Switch in sixth year, seventh year teaching
or higher
–0.061* 0.010 –0.029 –0.040 0.027 –0.032
(0.030) (0.050) (0.044) (0.024) (0.043) (0.029)
Switch in seventh year teaching or higher –0.036 –0.071 –0.091* –0.017 0.010 –0.008
(0.032) (0.057) (0.046) (0.023) (0.036) (0.031)
Teacher observations 1,657 1,657 1,657 1,663 1,663 1,663
Teacher-year observations 6,872 6,872 6,872 6,875 6,875 6,875
Student-year observations 159,808 159,808 159,808 159,305 159,305 159,305
Note. Estimates for adjacent and nonadjacent grade switches are from the same regression model. Each model controls for student and class characteristics, and includes
teacher fixed effects, school fixed effects, and grade-by-year fixed effects. Robust standard errors clustered at the class level are reported in parentheses. Sample includes
all elementary teachers who are observed as novices at some point in the data and whose students have test score data in the given subject in the current and prior year.
All cells have 89 teachers or more.
~p < .10. *p < .05. **p < .01. ***p < .001.
at Harvard Libraries on August 11, 2015http://er.aera.netDownloaded from

MAY 2015
221
higher than 0.17 SD by the 10th year of experience (often with
smaller returns in ELA). In comparison, I find average returns to
experience in the seventh through ninth year of teaching of
0.321 SD and 0.138 SD in math and ELA, respectively.
One reason for these differences in estimates is the unique
sample of teachers that I include in my analyses, which focuses
on elementary teachers who are observed as novices at some
point in my dataset. When I estimate returns to experience using
the same analytic models for all teachers with current- and prior-
year test-score data (third through eighth grade in math and
third through ninth grade in ELA), I find estimates that are
much more in line with other work (see Table A1 in the
Appendix). Returns to experience for teachers in their seventh
year of teaching or higher are 0.130 SD and 0.070 SD in math
and ELA, respectively. These estimates rise to 0.184 SD and
0.073 SD, respectively, when I further limit the sample to teach-
ers who are observed as novices at some point in the dataset.
With the exception of Ost (2014), most studies do not make this
same sample restriction, suggesting that including teachers with
censored data may attenuate results.
A second reason for larger average returns to experience in my
study is the fact that I account for the confounding effect of grade
reassignment on teacher productivity. Again, with the exception
of Ost (2014), other studies do not take this into account. The
fact that teachers with grade-specific experience have larger
returns than those who switch grades is consistent with theory,
prior work, and findings that I describe in detail below. As a point
of comparison, in Table A1, I show average returns to experience
for elementary teachers who are observed as novices at some point
in the dataset without controlling for switching grades. Indeed,
these estimates (0.270 SD and 0.098 SD in math and ELA,
respectively, for teachers in their seventh year of teaching or
higher) are smaller than those I observe in Table 3 when I do
control for switching grades (0.321 SD and 0.138 SD in math
and ELA, respectively, for this same group of teachers). In light of
these differences between my work and other studies, I describe
trends both in absolute terms (i.e., magnitude of effect sizes
related to switching grades) and relative terms (i.e., returns to
experience for those who switch grades compared to average).
Here, I observe some statistically significant and some margin-
ally significant decrements to average returns to experience in both
math and ELA. For example, average returns to experience are
0.099 SD and 0.168 SD in Years 2 and 3, respectively. In both
years, returns for those who switch grades are 0.036 SD, or one-
third to one-fifth, lower. I find larger decrements in math of 0.059
SD and 0.070 SD for teachers who switch grades in their fifth or
sixth year of teaching, respectively. These losses are one-and-a-half
to two times larger than average returns to experience from the
prior year (0.040 SD between Years 4 and 5, and 0.026 SD between
Years 5 and 6), meaning that teachers who switch grades in these
years lose ground relative to where they began before switching
grades. For teachers who switch grades in their second or sixth year
of teaching, these decrements persist into the following year.
In ELA, I only find decrements to returns to experience for
those who switch grades in their sixth year of teaching of 0.047
SD. One reason for this likely is that average returns to experience
are much smaller in ELA than they are in math. Thus, decrements
also are smaller in magnitude and harder to distinguish from zero.
I illustrate these trends for both ELA and math in Figure 2, only
presenting trend lines that deviate from average returns to experi-
ence at conventional levels of statistical significance.
I further explore these relationships by disaggregating returns
to experience for those who switch to adjacent versus nonadja-
cent grades. Two trends emerge. First, although I do observe
some statistically significant lower returns to experience for those
who switch to an adjacent grade, I also observe some higher
returns for these teachers. For teachers who switch grades in their
fifth year of teaching, returns to experience in both math and
ELA are statistically significantly larger than average returns in
the year following the switch (i.e., Year 6). This suggests that
there may be something unique about teachers who switch grades
in this year, such as a motivation to switch. This could be the case
for teachers who feel that they mastered teaching in one grade
over their first few years in the classroom and want to explore a
new grade level. Interestingly, though, teachers who switch to an
adjacent grade in their sixth year of experience have statistically
significant lower returns to experience in math.
A second key finding is that, on average, teachers who switch to
a nonadjacent grade experience substantively larger decrements to
returns to experience than those discussed earlier. These decre-
ments often persist into future years. In math, teachers who switch
to a nonadjacent grade in their third or fifth year exhibit returns to
experience roughly half as large as average returns to experience in
that year. For teachers in their third year, average returns are 0.173
SD, whereas those for teachers who switch to a nonadjacent grade
are 0.078 SD lower or 0.095 SD overall. For teachers in their fifth
year, average returns are 0.268 SD, whereas those for teachers who
switch to a nonadjacent grade are 0.118 SD lower or 0.150 SD
overall. In both cases, decrements of over 0.070 SD persist into the
year following the switch. Teachers who switch to a nonadjacent
grade in their seventh year of teaching or higher have returns 0.091
SD lower than average returns of 0.334 SD. As I cut off experience
at Year 7, I cannot follow these trajectories into future years. In
ELA, I find that teachers who switch to a nonadjacent grade in
their fifth year erase almost all returns to experience in the follow-
ing year. While average returns to experience in ELA in both Years
5 and 6 are 0.111 SD, returns for teachers who switch to a nonad-
jacent grade are 0.093 SD lower in Year 6, or 0.018 SD overall.
Finally, I examine the relationship between switching grades
and teacher retention, both in a given school (see Table 4) and in
the district (see Table 5). Findings indicate that teachers who switch
grades transfer schools at the end of that year 2.7 percentage points
higher than their colleagues who do not switch grades. This differ-
ential is almost 40% of the average transfer rate for this sample of
first- through eighth-year teachers, of 7.2%. These trends are simi-
lar for teachers who switch to adjacent and nonadjacent grades. I
also find that teachers who switch grades transfer schools in any
year after the switch roughly 1.7 percentage points higher than
those who do not do so. The differential for those who switch to an
adjacent grade is about half as large and no longer statistically sig-
nificant. I find similar results when I control for prior valued-added
scores but do not show these results in Table 4, as these value-added
scores are not a statistically significant predictor of school transfer
rates when also controlling for teaching experience.
at Harvard Libraries on August 11, 2015http://er.aera.netDownloaded from

222 EDUCATIONAL RESEARCHER
FIGURE 2. Returns to experience in math (left panel) and ELA (right panel) on average and for teachers who switch to any grade (top
panel) and to adjacent versus nonadjacent grades (bottom panel).
~p < .10. *p < .05. **p < .01. ***p < .001. Statistical significance refers to difference in returns to experience for switchers compared
to average.
at Harvard Libraries on August 11, 2015http://er.aera.netDownloaded from

MAY 2015
223
The relationship between switching grades and retention in
the district is less strong. Those teachers who switch to a nonad-
jacent grade leave the district at the end of the year 1.9 percent-
age points higher than those who do not switch grades. However,
given relatively high attrition rates overall—between roughly 8%
for first-year teachers and 19% for second-year teachers—this
differential is not as substantively significant as the results above.
Discussion and Conclusion
Research on teacher stability typically focuses on the extent to
which teachers remain in the same school, school district, or the
teaching profession from one year to the next. In this paper, I inves-
tigate another potential facet of stability—whether teachers experi-
ence stability in the grade in which they teach. General trends in
grade reassignment and the relationship to student achievement
and teacher turnover are consistent with prior research (Brummet,
Gershenson, & Hayes, 2013; Chingos & West, 2011; Jacob &
Rockoff, 2011; Ost, 2014; Ost & Schiman, 2015).
To my knowledge, this study is the first to distinguish between
adjacent and nonadjacent grade switching. I find that, in many
cases, teachers who switch grades exhibit smaller returns to expe-
rience in the year of the switch relative to average. For those who
switch to a nonadjacent grade, these decrements can wipe out
any gains due to increased experience and can persist in the year
after the switch occurs.
Though not a main focus of this paper, interestingly, I find
returns to experience that are substantively larger in magnitude
than those found in other studies (Clotfelter, Ladd, & Vigdor,
2006; Harris & Sass, 2011; Kraft & Papay, 2014; Ost, 2014;
Papay & Kraft, in press; Rockoff, 2004; Wiswall, 2013), driven
in large part by the estimation sample. This does not affect my
interpretation of the results, as I describe the relationship
between switching grades and returns to experience in both
absolute and relative terms. That said, my findings suggest that
this topic deserves further inquiry.
In light of disproportionate rates of switching that this and
other studies highlight (Brummet, Gershenson, & Hayes, 2013),
findings described above may be particularly troubling from an
equity perspective. In addition to having potentially negative
consequences for the most vulnerable teachers (i.e., early career
teachers and those with low prior value-added scores), high rates
of grade switching likely impact the most vulnerable students.
Teachers who switch grades—and thus exhibit decrements in
their returns to experience—are more likely to work in low-
achieving and high-minority schools. In addition, high rates of
switching to nontested grades (Boyd et al., 2008; Chingos &
West, 2011; Fuller & Ladd, 2012) may result in concentration of
Table 4
School Retention Rates for Teachers Who Do and Do Not Switch Grades
At End of Year of Switch In Any Year After Switch
Constant 0.035*** 0.035*** 0.035*** 0.035***
(0.006) (0.006) (0.006) (0.006)
Second year teaching –0.025*** –0.025*** –0.023*** –0.021***
(0.006) (0.006) (0.005) (0.005)
Third year teaching –0.022*** –0.022*** –0.023*** –0.021***
(0.006) (0.006) (0.006) (0.006)
Fourth year teaching –0.003 –0.003 –0.006 –0.003
(0.007) (0.007) (0.007) (0.007)
Fifth year teaching 0.005 0.005 –0.000 0.002
(0.008) (0.008) (0.008) (0.008)
Sixth year teaching 0.025** 0.025** 0.018~0.020*
(0.009) (0.009) (0.009) (0.009)
Seventh year teaching 0.044*** 0.044*** 0.035** 0.037**
(0.012) (0.012) (0.013) (0.013)
Eighth year teaching 0.092*** 0.092*** 0.083*** 0.084***
(0.020) (0.020) (0.020) (0.020)
Switch to any grade 0.027*** 0.017***
(0.005) (0.005)
Switch to adjacent grade 0.025*** 0.008
(0.007) (0.005)
Switch to nonadjacent grade 0.029*** 0.016**
(0.007) (0.006)
Teacher observations 5,203 5,203 5,203 5,203
Teacher-year observations 20,373 20,373 20,373 20,373
Note. Estimates in each column are from separate regression models that include fixed effects for entering cohort, with 2003-04 as the baseline cohort, and school fixed
effects. Robust standard errors clustered at the school level are reported in parentheses. Sample includes all elementary teachers who teach core academic subjects and
are observed as novices at some point in the data.
*p < .05. **p < .01. ***p < .001.
at Harvard Libraries on August 11, 2015http://er.aera.netDownloaded from

224 EDUCATIONAL RESEARCHER
low-quality teachers in the earliest elementary grades with poten-
tially lasting effects on students at a formative stage in their aca-
demic and emotional development (Jennings & DiPrete, 2012).
At the same time, it is important to note that one must be cau-
tious in interpreting results on the relationship between switching
grades and teachers’ longer term career trajectories, given the non-
causal nature of these analyses. Above, I describe a variety of fac-
tors that may motivate grade reassignments. In addition, the first
set of descriptive analyses suggests that grade reassignments likely
are not random. For example, findings that teachers with low
prior value-added scores switch grades at higher rates than their
more effective colleagues suggest that grade reassignments may be
a strategic decision on the part of school leadership.
Like Ost (2014), I attempt to limit the degree of bias through
strategic use of fixed effects that control for some of these factors.
However, although Ost predominantly is concerned with sorting
of students to teachers, I recognize that selection bias still likely
plays a role. For example, if teacher motivation to switch grades
is time invariant, then teacher fixed effects cannot account for
this. Similarly, school fixed effects control for observed and
unobserved differences across schools but cannot account for
within-school factors, such as grade reassignments motivated by
matching of teachers to students.
Not being able to distinguish between forced and chosen migra-
tion is a limitation of this work. That said, given that selection into
treatment can occur in multiple ways, the direction of bias is
unclear. Forced migration may make decrements to returns to
experience and retention rates larger than they would be otherwise,
whereas chosen migration likely makes decrements smaller.
Therefore, the extent of bias in my estimates depends in part on the
proportion of switches that are forced versus chosen. Although I
am unable to calculate or even estimate these proportions directly,
I argue that trends from this work are unlikely to disappear com-
pletely. Future research may attempt to isolate the causal effect of
switching grades on teacher- and student-level outcomes.
In addition, findings must be placed in a broader context of
teaching and schools as organizations. As noted by prior research, in
some instances switching grades likely is done strategically to avoid
accountability policy. However, in other cases, switching grades may
be done with the best interest of teachers in mind. For example,
working in different grades may help teachers develop a deeper
understanding of student development and learning trajectories. If
this is true, then teachers who switch grades may also be less likely
to experience “burn out.” This may be one reason why I observe
some positive returns to experience for teachers who switch to an
adjacent grade, relative to average returns to experience. It also may
Table 5
District Retention Rates for Teachers Who Do and Do Not Switch Grades
At End of Year of Switch In Any Year After Switch
Constant 0.074*** 0.075*** 0.075*** 0.075***
(0.007) (0.007) (0.007) (0.007)
Second year teaching 0.116*** 0.116*** 0.117*** 0.117***
(0.008) (0.008) (0.008) (0.008)
Third year teaching 0.088*** 0.088*** 0.087*** 0.087***
(0.008) (0.008) (0.008) (0.008)
Fourth year teaching 0.062*** 0.062*** 0.060*** 0.059***
(0.009) (0.009) (0.009) (0.009)
Fifth year teaching 0.033*** 0.033*** 0.029** 0.028**
(0.009) (0.009) (0.009) (0.009)
Sixth year teaching 0.008 0.008 0.004 0.003
(0.009) (0.009) (0.010) (0.010)
Seventh year teaching –0.015 –0.015 –0.020~–0.022*
(0.010) (0.010) (0.011) (0.011)
Eighth year teaching –0.021 –0.021 –0.026~–0.029*
(0.013) (0.013) (0.013) (0.013)
Switch to any grade 0.014~0.010
(0.007) (0.007)
Switch to adjacent grade 0.008 –0.001
(0.009) (0.007)
Switch to nonadjacent grade 0.019* 0.020**
(0.009) (0.007)
Teacher observations 5,203 5,203 5,203 5,203
Teacher-year observations 20,373 20,373 20,373 20,373
Note. Estimates in each column are from separate regression models that include fixed effects for entering cohort, with 2003-04 as the baseline cohort, and school fixed
effects. Robust standard errors clustered at the school level are reported in parentheses. Sample includes all elementary teachers who teach core academic subjects and
are observed as novices at some point in the data.
~p < .10. *p < .05. **p < .01. ***p < .001.
at Harvard Libraries on August 11, 2015http://er.aera.netDownloaded from

MAY 2015
225
be why, in some instances, decrements to returns to experience for
switching grades appear to fade in the year after the switch.
Taken together, these findings indicate a need for districts to
investigate why teachers are switching grades at high rates and to
consider the extent to which increasing stability in teachers’ grade
assignments may benefit schools, teachers, and students. Doing
so may be particularly relevant given that, compared to other
policies and programs focused on improving educational out-
comes for students, creating stability in teachers’ grade assign-
ments poses a potentially cost-effective way of shifting the needle.
Each year, districts spend millions of dollars recruiting, develop-
ing, and seeking to retain their teacher workforce (Barnes, Crowe,
& Schaefer, 2007; Darling-Hammond, Wei, Andree, Richardson,
& Orphanos, 2009; Miles, Odden, Fermanich, Archibald, &
Gallagher, 2011). Yet these investments—particularly around
teacher development—often do not correspond to significant
improvements in teacher quality or student achievement at scale
(Garet et al., 2011; Garet et al., 2008; Yoon, Duncan, Lee,
Scarloss, & Shapley, 2007). Even factors that are related to teach-
ers’ effectiveness and development trajectories come at high costs
for the size of their effects. For example, one-on-one teacher
coaching can raise student achievement by 0.22 SD in the postin-
tervention year but costs roughly $4,000 per teacher (Allen et al.,
2011). Interacting with high-quality teacher colleagues and
working in a strong school environment can produce annual
returns to experience of 0.04 SD and 0.003 SD, respectively
(Jackson & Bruegmann, 2009; Kraft & Papay, 2014) but require
large-scale recruitment efforts, building of school support net-
works, etc. Conversely, the decision of whether or not to have a
teacher switch grades has no direct cost. In fact, in light of the
relationship between grade reassignment and teacher retention in
schools, the small effort of keeping teachers in the same grade
may save money while also potentially raising student achieve-
ment. Therefore, continued research in this area may prove quite
valuable to schools.
NOTES
The research reported here was supported in part by the Dean’s
Summer Fellowship, Harvard Graduate School of Education. Additional
support comes from the Bradley Foundation. The opinions expressed
are those of the authors and do not represent views of the funders. I
thank Lindsay Page, Marty West, and John Willett for their guidance
and feedback throughout the study.
REFERENCES
Allen, J. P., Pianta, R. C., Gregory, A., Mikami, A. Y., & Lun, J. (2011).
An interaction-based approach to enhancing secondary school
instruction and student achievement. Science, 333, 1034–1037.
Barnes, G., Crowe, E., & Schaefer, B. (2007). The cost of teacher turn-
over in five school districts: A pilot study. Washington, DC: National
Commission on Teaching and America’s Future.
Boyd, D., Lankford, H., Loeb, S., & Wyckoff, J. (2008). The impact
of assessment and accountability on teacher recruitment and reten-
tion: Are there unintended consequences? Public Finance Review,
36(1), 88–111.
Brummet, Q., Gershenson, S., & Hayes, M. S. (2013). The frequency
and correlates of teachers’ grade-level reassignments: Evidence from
Michigan. Unpublished working paper, American University.
Chetty, R., Friedman, J. N., & Rockoff, J. E. (2014). Measuring the
impacts of teachers I: Evaluating bias in teacher value-added esti-
mates. American Economic Review, 104(9), 2593–2632.
Chingos, M. M., & West, M. R. (2011). Promotion and reassignment
in public school districts: How do schools respond to differences in
teacher effectiveness? Economics of Education Review, 30(3), 419–433.
Clotfelter, C. T., Ladd, H. F., & Vigdor, J. (2005). Who teaches
whom? Race and the distribution of novice teachers. Economics of
Education Review, 24(4), 377–392.
Clotfelter, C. T., Ladd, H. F., & Vigdor, J. L. (2006). Teacher-student
matching and the assessment of teacher effectiveness. Journal of
Human Resources, 41(4), 778–820.
Darling-Hammond, L., Wei, R. C., Andree, A., Richardson, N., &
Orphanos, S. (2009). Professional learning in the learning profes-
sion: A status report on teacher development in the United States and
abroad. Palo Alto, CA: National Staff Development Council and
The School Redesign Network, Stanford University.
Fischer, K. W. (1980). A theory of cognitive development: The control and
construction of hierarchies of skills. Psychological Review, 87, 477–531.
Fuller, S. C., & Ladd, H. F. (2012). School based accountability and
the distribution of teacher quality among grades in elementary schools.
Unpublished working paper, Duke University.
Garet, M. S., Cronen, S., Eaton, M., Kurki, A., Ludwig, M., Jones,
W., . . . Sztejnberg, L. (2008). The impact of two professional
development interventions on early reading instruction and achieve-
ment. Washington, DC: U.S. Department of Education, National
Center for Education Statistics.
Garet, M. S., Wayne, A. J., Stancavage, F., Taylor, J., Eaton, M.,
Walters, K., . . . Doolittle, F. (2011). Middle school mathemat-
ics professional development impact study: Findings after the second
year of implementation. Washington, DC: U.S. Department of
Education, National Center for Education Statistics.
Goldhaber, D., & Hansen, M. (2012). Is it just a bad class? Assessing the
long-term stability of estimated teacher performance. Economica,
80(319), 589–612.
Graeber, A. O., Newton, K. J., & Chambliss, M. J. (2012). Crossing
the borders again: Challenges in comparing quality instruction in
mathematics and reading. Teachers College Record, 114(4), 30.
Greenwald, R., Hedges, L. V., & Laine, R. D. (1996). The effect of
school resources on student achievement. Review of Educational
Research, 66(3), 361–396.
Hanushek, E. A. (1997). Assessing the effects of school resources on
student performance: An update. Educational Evaluation and
Policy Analysis, 19(2), 141–164.
Harris, D., & Sass, T. (2011). Teacher training, teacher quality, and
student achievement. Journal of Public Economics, 95, 798–812.
Heckman, J. J., & Snyder, J. M., Jr. (1996). Linear probability models of
the demand for attributes with an empirical application to estimating
the preferences of legislators (NBER Working Paper No. W5785).
Cambridge, MA: National Bureau of Economic Research.
Ingersoll, R. M. (2002). Out-of-field teaching, educational inequality, and
the organization of schools: An exploratory analysis. Seattle: Center
for the Study of Teaching and Policy, University of Washington.
Jackson, C. K., & Bruegmann, E. (2009). Teaching students and
teaching each other: The importance of peer learning for teachers.
American Economic Journal: Applied Economics, 1(4), 85–108.
Jacob, B. A., & Rockoff, J. E. (2011). Organizing schools to improve
student achievement: Start times, grade configurations, and teacher
assignments (Discussion Paper 2011–08). Washington, DC: The
Hamilton Project, Brookings Institution.
Jennings, J. L., & DiPrete, T. A. (2012). Teacher effects on social and
behavioral skills in early elementary school. Sociology of Education,
83(2), 135–159.
at Harvard Libraries on August 11, 2015http://er.aera.netDownloaded from

226 EDUCATIONAL RESEARCHER
Johnson, S. M., & Birkeland, S. E. (2003). Pursuing a “sense of suc-
cess”: New teachers explain their career decisions. American
Educational Research Association, 40(3), 581–617.
Kane, T. J., McCaffrey, D. F., Miller, T., & Staiger, D. O.
(2013). Have we identified effective teachers? Validating mea-
sures of effective teaching using random assignment. Seattle,
WA: Measures of Effective Teaching Project, Bill and Melinda
Gates Foundation.
Kauffman, D., Johnson, S. M., Kardos, S. M., Liu, E., & Peske, H. G.
(2002). “Lost at sea”: New teachers’ experiences with curriculum
and assessment. Teachers College Record, 104(2), 273–300.
Koedel, C., & Betts, J. R. (2011). Does student sorting invalidate
value-added models of teacher effectiveness? An extended analy-
sis of the Rothstein critique. Education Finance and Policy, 6(1),
18–42.
Kraft, M. A., & Papay, J. P. (2014). Do supportive professional envi-
ronments promote teacher development? Explaining heterogene-
ity in returns to teaching experience. Educational Evaluation and
Policy Analysis, 36(4), 476–500.
Kyriakides, L., & Creemers, B. P. M. (2008). A longitudinal study
on the stability over time of school and teacher effects on student
learning outcomes. Oxford Review of Education, 34(5), 521–545.
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.
Miles, K. H., Odden, A., Fermanich, M., Archibald, S., & Gallagher,
A. (2004). Inside the black box of professional development
spending: Lessons from comparing five urban districts. Journal of
Education Finance, 30(1), 1–26.
Nye, B., Konstantopoulos, S., & Hedges, L. V. (2004). How large are teacher
effects? Educational Evaluation and Policy Analysis, 26(3), 237–257.
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.
Ost, B., & Schiman, J. C. (2015). Grade-specific experience, grade reas-
signments, and teacher turnover. Economics of Education Review, 46,
112–126.
Papay, J. P., & Kraft, M. A. (in press). Productivity returns to experience
in the teacher labor market: Methodological challenges and new evi-
dence on long-term career improvement. Journal of Public Economics.
Rivkin, S. G., Hanushek, E. A., & Kain, J. F. (2005). Teachers, schools,
and academic achievement. Econometrica, 73(2), 417–458.
Rockoff, J. E. (2004). The impact of individual teachers on student
achievement: Evidence from panel data. American Economic
Review, 94(2), 247–252.
Schochet, P. Z., & Chiang, H. S. (2013). What are error rates for clas-
sifying teacher and school performance using value-added models?
Journal of Educational and Behavioral Statistics, 38(2), 142–171.
Singer, J. D., & Willett, J. B. (1993). It’s about time: Using discrete-
time survival analysis to study duration and the timing of events.
Journal of Educational and Behavioral Statistics, 18(2), 155–195.
Wayne, A. J., & Youngs, P. (2003). Teacher characteristics and student
achievement gains: A review. Review of Educational Research, 73(1),
89–122.
Wiswall, M. (2013). The dynamics of teacher quality. Journal of Public
Economics, 100, 61–78.
Yoon, K. S., Duncan, T., Lee, S. W. Y., Scarloss, B., & Shapley, K.
(2007). Reviewing the evidence on how teacher professional devel-
opment affects student achievement. Washington, DC: U.S.
Department of Education, Institute of Education Sciences,
National Center for Education Evaluation and Regional Assistance,
Regional Educational Laboratory Southwest.
AUTHOR
DAVID BLAZAR is a doctoral candidate in Quantitative Policy
Analysis in Education at the Harvard Graduate School of Education,
Center for Education Policy Research, 50 Church Street Fourth Floor,
Cambridge, MA 02138; david_blazar@mail.harvard.edu. His research
focuses on teacher and teaching quality, and the effects of policies
aimed at improving both.
Manuscript received June 26, 2014
Revisions received December 12, 2014,
and March 2, 2015
Accepted March 14, 2015
at Harvard Libraries on August 11, 2015http://er.aera.netDownloaded from

MAY 2015
227
Table A1
Sensitivity of Returns to Experience to Different Analysis Samples
Math ELA
Third Through
Eighth Grade
Teachers
Third Through
Eighth Grade
Teachers Who
Are Observed
as Novices
Third Through
Fifth Grade
Teachers Who
Are Observed
as Novices
Third Through
Ninth Grade
Teachers
Third Through
Ninth Grade
Teachers Who
Are Observed
as Novices
Third Through
Fifth Grade
Teachers Who
Are Observed
as Novices
Second year teaching 0.060*** 0.069*** 0.086*** 0.030*** 0.035*** 0.040***
(0.005) (0.009) (0.012) (0.003) (0.006) (0.010)
Third year teaching 0.080*** 0.100*** 0.136*** 0.040*** 0.045*** 0.064***
(0.005) (0.015) (0.022) (0.004) (0.010) (0.017)
Fourth year teaching 0.094*** 0.125*** 0.169*** 0.051*** 0.049*** 0.061**
(0.006) (0.022) (0.031) (0.004) (0.014) (0.024)
Fifth year teaching 0.106*** 0.136*** 0.203*** 0.058*** 0.059** 0.072*
(0.006) (0.029) (0.040) (0.004) (0.018) (0.031)
Sixth year teaching 0.122*** 0.158*** 0.229*** 0.063*** 0.055* 0.078*
(0.006) (0.035) (0.048) (0.004) (0.022) (0.037)
Seventh year teaching 0.130*** 0.184*** 0.270*** 0.070*** 0.073** 0.098*
(0.007) (0.044) (0.059) (0.005) (0.028) (0.045)
Teacher observations 16,094 2,986 1,920 16,795 3,137 1,922
Teacher-year observations 67,671 10,973 7,051 70,029 11,449 7,050
Student-year observations 2,166,595 455,008 159,903 2,082,156 407,204 159,400
Note. Estimates in each column are from separate regression models that control for student and class characteristics, and include teacher fixed effects, school fixed
effects, and grade-by-year fixed effects. Robust standard errors clustered at the class level are reported in parentheses.
*p < .05. **p < .01. ***p < .001.
Appendix
at Harvard Libraries on August 11, 2015http://er.aera.netDownloaded from