THE QUALIFICATIONS AND

CLASSROOM PERFORMANCE

OF TEACHERS MOVING TO

CHARTER SCHOOLS

Celeste K. Carruthers

经济系

田纳西大学

Knoxville, TN 37996-0550

carruthers@utk.edu

抽象的
Do charter schools draw good teachers from traditional,
mainstream public schools? Using a thirteen-year panel
of North Carolina public schoolteachers, I ﬁnd that less
qualiﬁed and less effective teachers move to charter
学校, particularly if they move to urban schools, 低的-
performing schools, or schools with higher shares of
nonwhite students. It is unclear whether these ﬁndings
reﬂect lower demand for teachers’ credentials and value
added or resource constraints unique to charter schools,
but the inability to recruit teachers who are at least as
effective as those in traditional public schools will likely
hinder charter student achievement.

This article is part of a series invited by this journal, in which authors present results of dissertation
research receiving the Jean Flanigan Outstanding Dissertation Award from the Association for Education
Finance and Policy.

C(西德:2) 2012 Association for Education Finance and Policy

233

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
e
d
p
A
r
t
我
C
e
–
p
d

我

F
/

7
3
2
3
3
1
6
8
9
3
4
2
e
d
p
_
A
_
0
0
0
6
7
p
d

乙
y
G
你
e
s
t

哦
n
0
8
S
e
p
e
米
乙
e
r
2
0
2
3

TEACHERS MOVING TO CHARTER SCHOOLS

介绍

1.
Charter schools, playing the role of competitive entrants in partially dereg-
ulated public education markets, can theoretically spur efﬁciency gains and
improve the quality of education overall by introducing innovative teaching
and leadership strategies, improving the match value between students and
学校, and forcing incumbents (这里, traditional public schools) to compete
for limited resources. Critics of charter schools argue that they skim good
students and remove much-needed funds from public schools without deliv-
ering unambiguously superior student achievement. While the effectiveness
with which charter schools raise student achievement has been studied at
length,1 much less is known about the ﬂow of resources between charter and
mainstream schools. Public funding transfers to charter schools are typically
proportionate to the number of students enrolled—that is, “the money follows
the students.” But might teachers follow the students as well? What can we
learn from the teachers who opt out of traditional public schools in favor of
the charter sector? In this study I use unprecedented detail on future charter
teachers in North Carolina to describe the qualiﬁcations and value added of
teachers moving out of traditional public schools and into charter schools. 我
show that North Carolina charter schools are drawing less qualiﬁed and less
effective teachers from traditional, mainstream public schools. This pattern
indicates that charters have lower demand for credentials and value added
or that charters do not have the resources, broadly speaking, to attract good
teachers from other schools.

Among North Carolina’s charter teachers, 36.1 percent previously taught
in a mainstream school. This subset of teachers represents a unique window
on the competition between public schooling sectors for teaching talent. 这些
“charter movers” are much more likely to have exercised revealed preference
for charter schools than teachers who are recruited directly from college or
nonteaching careers. The strength with which charters draw good teachers
from mainstream schools is a signal of charters’ workplace appeal but also
a likely precursor to higher student achievement. Teacher quality is a pro-
found factor in student achievement, and charters seeking to produce high
achievement will value teachers who have succeeded in other schools. 但
good teachers tend to gravitate to good schools, and this tendency may work

See Dee 1998, Hoxby 2003, and Booker et al. 2008 for empirical evidence that mainstream stu-
dent performance improves in light of competition from choice schools. Observational studies of
administrative data ﬁnd that enrolling in a new charter school has a negative impact on student
achievement growth, more so in newer schools (see Bifulco and Bulkley 2008 and Gleason et al.
2010 for reviews). 相比之下, recent studies of lottery-based admissions to urban, oversubscribed
charter schools ﬁnd large and positive impacts of charter attendance (Hoxby and Murarka 2009;
Abdulkadiroglu et al. 2011; Dobbie and Fryer 2011).

234

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
e
d
p
A
r
t
我
C
e
–
p
d

我

F
/

7
3
2
3
3
1
6
8
9
3
4
2
e
d
p
_
A
_
0
0
0
6
7
p
d

乙
y
G
你
e
s
t

哦
n
0
8
S
e
p
e
米
乙
e
r
2
0
2
3

Celeste K. Carruthers

against North Carolina’s charters, which typically underperform traditional
public schools (Bifulco and Ladd 2006; Carruthers 2012). 的确, results sug-
gest that North Carolina’s charter schools attract less talented teachers from
traditional public schools, a consequence and perhaps a cause of weaker per-
formance throughout the state’s charter sector. Teachers moving to charter
schools are less experienced, less likely to be licensed, and less likely to hold
graduate degrees than teachers who make similar shifts between mainstream
学校. In this sense they resemble teachers who leave North Carolina public
学校 (and likely leave the profession) more than other teachers who switch
学校. 此外, charter movers are less effective in terms of math and
reading value added, and teachers moving to less effective charter schools,
urban charter schools, or charters with higher shares of nonwhite students
have even lower reading value added.

This article proceeds as follows. 部分 2 outlines conceptual expectations
for the relative qualiﬁcations and effectiveness of teachers moving to char-
ter schools. 部分 3 describes pertinent features of North Carolina’s charter
system and the data used in this study. 部分 4 analyzes the credentials
of teachers moving to charter schools. In section 5 I describe measures of
classroom performance, evaluate charter movers’ value added relative to other
mobile teachers, and explore the possibility of observational bias from sam-
pling error and nonrandom student sorting. 部分 6 offers conclusions and
open questions.

2. CONCEPTUAL FRAMEWORK
A founding purpose of North Carolina’s charter legislation is to “create new
professional opportunities for teachers.”2 It remains to be seen if charters
capitalize on these opportunities to attract talented, effective teachers, nor is
it clear if charters have enough advantages in the teacher labor market to do
所以. This section illustrates why expectations about the relative qualiﬁcations
and value added of charter teachers are ambiguous a priori, which motivates
the empirical exercises to follow and frames results within the broader policy
implications of teacher quality in charter schools.

As a starting point, consider the motives of charter schools, which make
up a small part of the demand for teachers. Broadly speaking, charters seek to
optimize student enrollment (and proportionately budgets), subject to regula-
保守党, ﬁnancial, and physical constraints. Producing high student achievement
supports this objective by meeting the accountability thresholds necessary to
stay open and appealing to parents. Teachers are schools’ primary resource

2. North Carolina General Statutes,

115c-238.29a(4)

(1996;

see www.ncga.state.nc.us/

enactedlegislation/statutes/html/bysection/chapter 115c/gs 115c-238.29a.html).

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
e
d
p
A
r
t
我
C
e
–
p
d

我

F
/

7
3
2
3
3
1
6
8
9
3
4
2
e
d
p
_
A
_
0
0
0
6
7
p
d

乙
y
G
你
e
s
t

哦
n
0
8
S
e
p
e
米
乙
e
r
2
0
2
3

235

TEACHERS MOVING TO CHARTER SCHOOLS

for producing tested achievement as well as myriad other student virtues that
are less readily observable (例如, maturity, an appreciation for the arts, or other
outcomes unique to each charter school’s mission). Since charter schools are
subject to more oversight than traditional public schools (from authorizers, 在
至少, if not parents and the public at large) as well as the credible threat of
closure, charter administrators may place a higher value on teacher charac-
teristics that help raise student achievement and otherwise retain the support
of parents. 而且, charters are typically free to deviate from step-lane pay
scales that pay mainstream teachers according to strict functions of experi-
恩斯, 教育, and other credentials. Charter leaders can pay more or less
for these credentials if they wish, and they can also pay more for less tangible
elements of teacher quality, such as value added.

If we ﬁnd that charter teachers exhibit more of a particular characteristic,
this could be because charter schools have higher demand for that characteris-
tic, because charters have more favorable endowments in the labor market for
teachers with that characteristic, 或两者. Note that labor market endowments
encompass pecuniary resources with which to pay salaries but also nonpecu-
niary workplace attributes that teachers may value. Hoxby (2002) develops this
intuition further and, using a nationwide survey, ﬁnds that charter teachers
tend to have taken more college math and science courses, are more likely
to have graduated from a competitive college, but are no more or less likely
to have earned a graduate degree. Lower levels of certain teacher aspects in
the charter sector could reﬂect lower demand (perhaps because these aspects
do not advance charters’ objectives) or lower pecuniary and nonpecuniary re-
来源. Financial resource constraints stem from charter ﬁnance models that
allocate each school a per pupil rate roughly equal to the surrounding district’s
average per pupil cost. If a district enjoys substantial economies of scale, 它是
per pupil expenses will be less than a charter school’s average cost. Finan-
cial difﬁculties are common in North Carolina’s charter schools. Twenty-four
charters were relinquished or revoked between 1998 和 2006; of those, 九
cited ﬁnancial problems as a leading cause of failure.3

供给侧, teachers have been shown to vote with their feet in
favor of more effective schools and more socioeconomically advantaged stu-
dent populations,4 which works against charter schools that target urban or
at-risk students. Charter schools, and choice schools in general, may induce
sorting by race, 收入, or parental preferences for education (Hastings, Kane,

In a few cases, insolvency stemmed directly from embezzlement or negligent ﬁnancial management.
看, 例如, Lanier 2008.
看, 例如, Hanushek, Kain, and Rivkin 2004; Falch and Strøm 2005; Scaﬁdi, Sjoquist, 和
Stinebrickner 2007; and Loeb, Kalogrides, and B´eteille 2012.

236

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
e
d
p
A
r
t
我
C
e
–
p
d

我

F
/

7
3
2
3
3
1
6
8
9
3
4
2
e
d
p
_
A
_
0
0
0
6
7
p
d

乙
y
G
你
e
s
t

哦
n
0
8
S
e
p
e
米
乙
e
r
2
0
2
3

Celeste K. Carruthers

3
0
.

y
t
我
s
n
e
D
d
e
t
A
米

我
t
s
乙

2
0
.

1
0
.

100

Percent of a School’s Students Who Are Nonwhite

Mainstream Schools (n = 2375)
Charter Schools (n = 96)

kernel = epanechnikov, bandwidth = 2.5000

数字 1. Density Estimates—Percent of Students Who Are Nonwhite in Charter and Mainstream
学校, Academic Year 2009

and Staiger 2009). 的确, North Carolina’s charter schools are racially seg-
regated to a much starker degree than mainstream schools, as described by
Bifulco and Ladd (2007) and illustrated in ﬁgure 1. 数字 1 plots kernel
densities of nonwhite student shares for charter and mainstream schools
在里面 2009 school year.5 Whether because of location, pedagogical foci, 或者
other determinants of student sorting, North Carolina’s charter schools are
much more likely than mainstream schools to have very high or very low
concentrations of nonwhite students. Stuit and Smith (2012) ﬁnd that non-
white student shares explain part of the fact that charter schools tend to have
higher teacher turnover; therefore it would not be surprising to ﬁnd that
charters with more nonwhite students struggle to recruit mainstream teach-
呃. Setting aside the relative draw of different student populations, teach-
ers may be unwilling to accept lower pay in charter schools. 虽然
charters are free to exceed step-lane pay rates for teachers, charter teach-
ers tend to earn no more than their mainstream counterparts, and unlike
mainstream school districts, charters do not necessarily pay teachers more
for experience, licensure, or education (Podgursky and Ballou 2001; Hoxby
2002; 泰勒 2005). 即使是这样, nonpecuniary beneﬁts may offset disadvantaged
student populations and lower pay in charter schools. Early advocates of the
charter model stressed the professionalization and empowerment of teachers

I refer to school years by the year of their conclusion. 例如, 2009 references the 2008−9
school year.

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
e
d
p
A
r
t
我
C
e
–
p
d

我

F
/

7
3
2
3
3
1
6
8
9
3
4
2
e
d
p
_
A
_
0
0
0
6
7
p
d

乙
y
G
你
e
s
t

哦
n
0
8
S
e
p
e
米
乙
e
r
2
0
2
3

237

TEACHERS MOVING TO CHARTER SCHOOLS

as critical tenets of charter development (看, 例如, Budde 1988 和
Kolderie 1990), and modern charter schools often follow this course. In sur-
veys, charter teachers cite collegiality, common instructional philosophies,
and greater creative license as roots of job satisfaction (Malloy and Wohlstetter
2003).

To sum, charters’ autonomy, limited resources, and diversity lead to un-
certain expectations about the qualiﬁcations and effectiveness of teachers who
choose to work in the charter sector over other options. Charters will value
teacher characteristics that support school objectives, and in principle charters
can offer teachers possessing those characteristics higher pay and higher job
satisfaction. But in practice charters have limited salary funds, uncertain time
horizons, and varied student challenges that may conﬂict with teacher prefer-
恩塞斯, and we cannot be certain that charters value the same qualiﬁcations that
mainstream districts value, nor can we be certain where student achievement
ranks among charter school objectives. The net, reduced-form outcome of all
these factors is unclear; charter teachers may be more or less qualiﬁed than
their mainstream counterparts, and charter teachers may have higher or lower
value added in terms of student achievement.

In spite of ambiguous expectations overall, this conceptual framework
has implications for different types of teachers and different types of charter
学校. Nonpecuniary school conditions that teachers value should be posi-
tively associated with teacher characteristics that schools value. 例如,
Loeb, Kalogrides, and B´eteille (2012) show that more effective schools are bet-
ter able to recruit good teachers from other schools. So within the charter
sector, we would expect to see teachers with higher value added gravitating
toward more effective charters. (部分 5 demonstrates this very pattern of
positive matching between teacher and school quality.) 相似地, teacher pref-
erences for more socioeconomically advantaged schools are consistent with
teacher mobility patterns I describe, in that more effective teachers sort into
less urban charter schools as well as charters with fewer nonwhite students.

I approach the question of charter teacher quality empirically, identifying
relative qualiﬁcations and classroom performance for the subset of charter
teachers who moved from the mainstream sector. These mobile teachers pro-
vide valuable insight about how well charter schools compete with mainstream
schools for effective teachers who could likely work in either sector. 教师
who move from traditional public schools to charter schools (especially if they
are regularly licensed) are more likely to exhibit revealed preference in favor
of charter schools, in which case their mobility decisions involve demand-side
charter and mainstream offerings, rather than charter alone. When assess-
ing the qualiﬁcations and value added of charter movers, I compare them
with other teachers changing schools but staying in the mainstream sector,

238

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
e
d
p
A
r
t
我
C
e
–
p
d

我

F
/

7
3
2
3
3
1
6
8
9
3
4
2
e
d
p
_
A
_
0
0
0
6
7
p
d

乙
y
G
你
e
s
t

哦
n
0
8
S
e
p
e
米
乙
e
r
2
0
2
3

Celeste K. Carruthers

controlling for sending and receiving school characteristics.6 Results indicate
that charter schools attract less qualiﬁed and less effective teachers, 虽然
it remains to be seen if these ﬁndings are due to resource constraints, 降低
demand for these qualities, or supply-side preferences. Regardless of which
labor market factors dominate the sorting of teachers between charter and
mainstream schools, the inability to recruit teachers who are at least as effec-
tive as those in traditional public schools likely hinders student achievement
in charter schools.

3. CHARTER SCHOOLS IN NORTH CAROLINA
Background

The North Carolina legislature authorized the state’s system of charter schools
在 1996, and the ﬁrst charters opened for the 1997−98 school year. 经过 2009
有超过 35,000 students enrolled in charter schools, representing
2.4 percent of statewide enrollment. Charters are spread throughout urban,
rural, and socioeconomically diverse regions of North Carolina. The state’s
charter legislation and oversight bear many features in common with other
charter systems. North Carolina had a binding one hundred school cap on
the charter sector throughout the window of time this analysis considers;
因此, a very small percentage of teachers move to charter schools in
any given year.7 The comparison group—mainstream teachers moving to
other mainstream schools—is large and varied, as are the schools they move
到, so for any given mainstream-to-charter move, I can likely ﬁnd another
move between qualitatively similar mainstream schools. 那是, charter and
mainstream movers have common support for identiﬁcation of their relative
质量, controlling for sending and receiving school environments. Extensive
data covering a thirteen-year period are available for all teachers in the state.
These data allow me to characterize the qualiﬁcations of every teacher moving
to the state’s charter sector and to estimate the classroom performance of many
elementary charter movers.

The application, approval, and evaluation of charter schools is closely reg-
ulated, but the schools are given wide latitude in their personnel management
and daily operations. North Carolina charter schools are organized as private,
nonproﬁt organizations. They are allotted funding from state and local boards
of education on a per pupil rate, commensurate with district per pupil costs.

6. Evaluating charter movers against nonmobile teachers would impose an inappropriate benchmark
if stayers had higher value added than leavers (Hanushek and Rivkin 2010) or if charter movers
would have changed schools or left teaching regardless of charter opportunities (Jackson 2011).
Results to follow support these ideas.
The one hundred school cap was lifted in the summer of 2011.

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
e
d
p
A
r
t
我
C
e
–
p
d

我

F
/

7
3
2
3
3
1
6
8
9
3
4
2
e
d
p
_
A
_
0
0
0
6
7
p
d

乙
y
G
你
e
s
t

哦
n
0
8
S
e
p
e
米
乙
e
r
2
0
2
3

239

TEACHERS MOVING TO CHARTER SCHOOLS

Public funds can be used toward “operational and ﬁnancing leases for real
财产,” but the state does not take responsibility for any charter indebt-
edness.8 Charters can raise additional funds by winning grants or soliciting
donations, but they cannot charge tuition.

Charter schools are allowed great ﬂexibility in the recruitment, retention,
and pay of their faculties. At least 75 percent of charter teachers in kinder-
garten through ﬁfth-grade classrooms must hold teaching certiﬁcates. 这
number falls to 50 percent for charter teachers of grades 6−12. Only certiﬁed
teachers are eligible for tenure after four consecutive years of teaching in a
mainstream public school. Tenured mainstream teachers who wish to teach in
a charter school are granted one year’s leave, meaning that they can return to
their original school after a year, space permitting.9 Charters are not required
to offer tenure, nor are they required to participate in the state retirement
plan.

数据

I utilize a richly detailed panel describing teachers’ credentials, work envi-
罗蒙兹, and career paths over the years 1997−2009 within the universe
of North Carolina public schools, 学生, and teachers.10 Data are not
uniform across charter and mainstream teachers. 性别, 种族, and school
assignments are known for all teachers—including, 重要的是, charter
teachers—but teacher credentials and student-teacher linkages necessary
for value-added estimation are known only for mainstream teachers. 因此
I cannot characterize the qualiﬁcations or classroom performance of charter
teachers who are hired directly from college or other careers, and I cannot
observe teacher qualiﬁcations or value added after teachers move to charter
学校. These are irreconcilable limitations of the data. But I observe
mainstream teachers who move to charter schools, allowing me to evaluate
their credentials and, for the ﬁrst time, to estimate the value added of
个人 (albeit future) charter teachers. I exclude teaching assistants,
facilitators, and teachers simultaneously assigned to more than one school. 我
link each teacher’s school assignment to campuswide statistics derived from
the National Center for Education Statistics (NCES) Common Core of Data
(grades served, urbanicity of locale, percentage of students who are nonwhite,

8. North Carolina General Statutes, § 115C-238.29H(a1). See www.ncga.state.nc.us/enactedlegislation/

statutes/html/bysection/chapter 115c/gs 115c-238.29h.html.

9. North Carolina General Statutes, § 115C-238.29F(e3). See www.ncga.state.nc.us/enactedlegislation/

statutes/html/bysection/chapter 115c/gs 115c-238.29f.html.

10. Data are managed by the North Carolina Education Research Data Center (NCERDC) at Duke

大学. See Muschkin, Bonneau, and Dodge (2009) 欲了解详情.

240

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
e
d
p
A
r
t
我
C
e
–
p
d

我

F
/

7
3
2
3
3
1
6
8
9
3
4
2
e
d
p
_
A
_
0
0
0
6
7
p
d

乙
y
G
你
e
s
t

哦
n
0
8
S
e
p
e
米
乙
e
r
2
0
2
3

Celeste K. Carruthers

桌子 1.

In-Sample Mobility Patterns of Charter Teachers

Teacher Mobility Pattern

百分比

Started and ended in the charter system (right censored)

Started and ended in the charter system (uncensored)

Mainstream to charter

Mainstream to charter to mainstream

Charter to mainstream

Other patterns

18.8

33.2

9.1

11.1

<1.0 Notes: n = 6,823 teachers. The ﬁrst two mobility patterns apply to teach- ers who teach exclusively in charter schools. Right-censored charter teach- ers enter the sample in the charter system and are observed teaching there in 2009, the last year of the panel. Uncensored teaching spells end before 2009. The last four mobility patterns apply to teachers who teach in charter and mainstream schools. The percent of all charter participants who follow each pattern is indicated at right. and total enrollment) as well as schoolwide proﬁciency rates calculated by the North Carolina Department of Public Instruction.11 4. THE QUALIFICATIONS OF TEACHERS MOVING TO CHARTER SCHOOLS Table 1 describes teacher mobility patterns between charter and mainstream schools for the 6,823 teachers who are observed working in a charter school at some time between 1998 and 2009. The majority teach exclusively in charter schools. The results to follow focus on charter teachers who initially teach in a mainstream North Carolina school before moving to the charter sector, which accounted for 36.1 percent of all charter teachers. Table 2 lists summary statistics for the 1997−2008 panel of North Carolina’s mainstream public schoolteachers with known following-year teaching assignments. It is impor- tant to emphasize that the qualiﬁcations listed in table 2 are at best weakly linked to student achievement.12 Thus the relative frequency of these creden- tials among charter movers will do little to foreshadow student achievement 11. Common Core statistics and schoolwide proﬁciency data are available alongside teacher, student, and school microdata in the NCERDC data ﬁles. The percent of students performing at grade level, or the “performance composite,” is determined by the state as part of annual accountability procedures. 12. Graduate degrees appear to have no robust impact on teacher quality (see Goldhaber 2008 for a review). Licensure has been linked with higher student achievement (Clotfelter, Ladd, and Vigdor 2007; Goldhaber 2007), although there tends to be much more variation in teacher quality within licensure classes than between (Boyd et al. 2006; Kane, Rockoff, and Staiger 2008). The returns to teacher experience are initially steep, with signiﬁcant student achievement gains over the ﬁrst three to ﬁve years of a teacher’s career. Thereafter the impact of teacher experience plateaus (Murnane and Phillips 1981; Rockoff 2004; Clotfelter, Ladd, and Vigdor 2007). l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / f e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d . f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 241 TEACHERS MOVING TO CHARTER SCHOOLS Table 2. North Carolina Public School Teachers: Summary Statistics Teacher Qualiﬁcation Holds graduate degree (%) Attended competitive college (%) Mean licensure test score (∼N(0;1)) Regularly licensed (%) Teaching experience (years) Experience ≤ 3 years (%) Experience ≥ 25 years (%) Days absent Black (%) Hispanic (%) Other, nonwhite (%) Female (%) (1) All Teachers 30.7 (46.1) 76.3 (42.5) 0.031 (0.851) 90.5 (29.4) 12.6 (9.6) 20.7 (40.5) 14.0 (34.7) 11.7 (9.9) 14.3 (35.0) 0.9 (9.2) 1.2 (10.9) 80.0 (40.0) (2) Mainstream Movers 28.4∗ (45.1) 74.8∗ (43.4) 0.022∗ (0.828) 87.6∗ (32.9) 9.5∗ (9.0) 31.9∗ (46.6) 8.4∗ (27.8) 13.1∗ (11.2) 15.9∗ (36.6) 1.3∗ (11.4) 1.3 (11.1) 78.9∗ (40.8) n (teacher years) 867,019 84,222 (3) Charter Movers 25.2∗ (43.4) 69.5∗ (46.0) 0.039 (0.895) 78.7∗ (41.0) 8.3∗ (9.3) 39.4∗ (48.9) 9.4 (29.2) 13.4 (11.6) 23.1∗ (42.1) 1.9∗ (13.5) 1.9∗ (13.5) 78.3∗ (41.2) 1,926 Notes: The table lists summary statistics for all 1997−2008 North Carolina mainstream school teachers with known school assignments in the following year. Standard deviations are in paren- theses below each mean. Data for moving teachers reference the year immediately preceding a school change. Mainstream movers (column 2) are mainstream teachers who are next observed in a different mainstream school, and asterisks in column 2 indicate a signiﬁcant difference (at 95% conﬁdence or greater) between mainstream movers and nonmobile teachers. Charter movers (column 3) are next observed in a charter school, and asterisks indicate signiﬁcant differences between mainstream and charter movers. in charter schools. This analysis will determine if charters are drawing well- qualiﬁed teachers from mainstream schools in terms of qualiﬁcations like graduate education, licensure, and experience that are valued by mainstream districts and rewarded in step-lane pay scales. Teachers’ experience and higher education data are drawn from de- tailed personnel records.13 College competitiveness is inferred from the 1995 13. Experience is underestimated because personnel records do not necessarily account for accumulated teaching experience from outside North Carolina public schools. 242 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / f e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Celeste K. Carruthers edition of Barron’s Proﬁles of American Colleges, which roughly corresponds to the graduation date of mobile teachers with six years (the median) of experi- ence. North Carolina teachers take a variety of licensure exams, most of which are in the Praxis series. In order to include all available test information, I scale raw licensure test scores to have a standard normal distribution within each test code and test year. I calculate each teacher’s mean standardized licensure test score, equal to the average of all his or her unique exam records. Regularly li- censed teachers, who account for 90.5 percent of all teachers, have completed an approved teacher education and testing program or attained North Car- olina licensing by reciprocal or interstate agreement. Teachers without regular licensure are uncertiﬁed teachers holding temporary, emergency, or provi- sional licenses. Absenteeism ﬁgures exclude vacation days and obvious data errors.14 Mobile teachers,15 summarized in the second column of table 2, are on average less qualiﬁed than teachers who are not changing schools. Movers are earlier in their careers, less likely to have graduate degrees, and somewhat less likely to have graduated from a competitive college. Further, they have lower li- censure test scores and are absent more often in the year prior to moving. Main- stream teachers moving to charter schools, summarized in the third column of table 2, are even less credentialed than other moving teachers in terms of higher education, licensure, and experience. Strikingly, charter movers are 8.9 percentage points less likely to be regularly licensed. North Carolina’s policy of permitting more uncertiﬁed teachers in charter schools, which is intended to attract individuals from outside the traditional teacher education pipeline, may have the consequence of drawing untenured mainstream teachers nearing the expiration of temporary licenses. Additional descriptive evidence supports this notion. Tenure is typically granted to licensed teachers after four years. Among unlicensed school changers, the likelihood of moving to a charter school is 3.8 percent for ﬁrst-year teachers, increasing to 5−6 percent for second- and third- year teachers, and back down to 3.1 percent for teachers completing their fourth year. Finally, table 2 shows that teachers moving to charter schools are much more likely to be black and modestly more likely to be Hispanic or otherwise nonwhite. 14. Vacation days, which include all mandatory holidays, are recorded inconsistently in the absenteeism data. I exclude records with negative days absent, more than 25 absences in a month, or more than 150 in a year. 15. Throughout the article, mobile teachers are deﬁned as those who are next observed in a different school, with no more than a one-year gap between schools. Principal results are robust to more liberal deﬁnitions of mobility (allowing for longer gaps between schools) and to more conservative deﬁnitions (allowing for no gaps between schools). l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / f e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d . f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 243 TEACHERS MOVING TO CHARTER SCHOOLS Summary evidence suggests that charter movers are typically less creden- tialed than other mobile teachers, but it may be the case that charter movers are drawn from schools or districts with weaker qualiﬁcations throughout. I estimate a multinomial logit model of teacher mobility to assess how particular teacher and school factors relate to charter mobility. Pim = (cid:2) θ Zi βm l =1 θ Zi βl S m = 1, . . . , 5 l = 1, . . . , 5, (1) where Pim is the probability of choosing one of ﬁve mobility outcomes (m): move to a charter school, move to another mainstream school, temporarily leave the sample, leave the sample completely, or teach at the same school. Zi is a matrix of teacher characteristics summarized in table 2 and several variables describing teachers’ sending schools: the percent of students who are nonwhite, the percent who are proﬁcient, total enrollment, grade level(s) served, urbanicity indicators,16 and indicators for missing data. The baseline option—teach at the same school—is restricted to have β5 = 0, so βm estimates for other outcomes are interpreted as relative to β5. Relationships between control variables and the relative risk of making mobility choice m are listed in table 3. Signiﬁcant coefﬁcients greater than 1 indicate that a marginal increase in variable Zi is associated with a higher risk of making that particular mobility choice. Teachers with graduate degrees or degrees from competitive colleges are not signiﬁcantly more or less likely to move to a charter school relative to stayers. Teachers with higher licensure test scores are more likely to move to a charter school, as are unlicensed teachers, teachers with no more than three years’ experience, and teachers with more absences. In terms of credentials, charter movers resemble teachers leaving North Carolina public schools more than teachers changing schools or temporarily exiting the sample. Unlicensed and inexperienced teachers are more likely to make any move, as are teachers with higher licensure test scores (which may correlate with skills valued in other schools and careers), but these patterns are pronounced for charter movers and leavers. Echoing Jackson (2011), these ﬁndings may shed light on why teachers leave for charter schools if doing so is a close substitute to leaving the profession. In contrast to descriptive statistics in table 2, multinomial logit results suggest that black teachers are not signiﬁcantly more likely to move to charter schools, conditioning on teacher and school characteristics. In terms of school characteristics, teachers outside 16. Census Bureau classiﬁcations are used to classify schools as located in rural areas, towns, or urban areas. Rural areas generally include places with fewer than 2,500 inhabitants outside metropolitan statistical areas (MSAs). Large and small towns are incorporated places with at least 2,500 inhabitants outside any urban fringe or MSA. Urban areas encompass MSAs, including cities and urban fringes. 244 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / f e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Celeste K. Carruthers Table 3. Multinomial Logit Results: Teacher Mobility, Teacher Qualiﬁcations, and School Characteristics (1) (2) To a Charter To a Mainstream Temporarily School (4) Out of Out of Sample Sample School (3) 0.995 (0.07) 1.050 (0.67) 1.201 (9.23) 0.449 (9.45) 2.151 (11.24) 0.909 (0.93) 1.019 (8.66) 1.070 (0.78) 1.426 (1.27) 1.136 (0.59) 0.836 (2.27) 1.018 (0.20) 1.101 (1.43) 0.982 (6.37) 1.012 (8.70) 1.048 (4.75) 1.035 (3.03) 1.031 (5.15) 0.776 (15.59) 1.666 (49.43) 0.565 (36.63) 1.015 (41.07) 0.891 (7.73) 1.436 (7.96) 0.760 (6.69) 0.856 (13.72) 0.969 (2.49) 1.038 (3.77) 0.980 (49.34) 1.005 (22.30) 1.264 (13.59) 1.075 (3.50) 1.014 (1.37) 0.672 (14.37) 1.879 (32.83) 1.055 (2.24) 1.031 (62.70) 1.162 (5.95) 1.880 (8.63) 0.872 (1.83) 0.909 (4.41) 0.788 (9.85) 0.865 (7.99) 0.988 (16.57) 1.004 (9.31) 1.353 (33.93) 1.097 (8.39) 1.133 (21.52) 0.457 (56.42) 2.025 (67.98) 2.306 (78.20) 1.031 (93.54) 0.929 (5.21) 1.477 (9.06) 0.912 (2.41) 0.865 (13.54) 0.800 (17.81) 0.939 (6.80) 0.983 (41.90) 1.003 (14.48) Type of Move Teacher Qualiﬁcations Holds graduate degree Attended competitive college Mean licensure test score (∼N(0;1)) Regularly licensed Experience ≤ 3 years Experience ≥ 25 years Days absent Teacher Characteristics Black, non-Hispanic Hispanic Other nonwhite, non-Hispanic Female School and Student Characteristics Located in a large or small town Located in a rural area Percent performing at grade level Percent nonwhite Wald2 = 129,752 Pseudo R2 = 0.044 Notes: n = 707,110 teachers. The baseline category is no mobility—that is, staying at one’s current school. Relative risk ratios (and robust z-scores, in parentheses) are estimated by multinomial logit. Additional control variables include indicators for schools with elementary, middle, and/or high school grades as well as a set of indicators for missing data. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / f e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 245 TEACHERS MOVING TO CHARTER SCHOOLS urban areas are less likely to change schools or leave the sample (temporarily or otherwise), but no more or less likely to move to a charter school. Teachers in schools with more proﬁcient students are less likely to make any move. In agreement with the teacher mobility literature, higher shares of nonwhite students are associated with a higher likelihood of making any move, and charter mobility is somewhat more sensitive to nonwhite student shares than other types of mobility.17 Multinomial logit results conﬁrm many impressions from table 2 sum- mary statistics but do not resolve the question of whether charter schools are attracting more or less qualiﬁed teachers than similar mainstream schools. The charter sector as a whole is attracting less qualiﬁed teachers, but receiving charters may have been drawing relatively well, controlling for their own loca- tions and student populations. Toward that end I conduct further analyses of charter and mainstream movers’ credentials (Qk j sl t ) by estimating equation 2 via ordinary least squares (OLS) for each North Carolina teacher j observed in year t (1997−2008), school s, and county l: Qk j sl t = δm1(moving ) j + δc 1(tochar ter ) j + Cr sl (t+1) βr +C s sl t βs + αl (t+1) + v j sl t . (2) All mobile teachers have the indicator 1(moving) j equal to one. Teachers mov- ing to a charter school additionally have 1(tocharter) j equal to one. Equation 2 estimates regression-adjusted mean differences in qualiﬁcation k between mainstream movers and nonmovers (δm) and between mainstream movers and charter movers (δc ). This formulation evaluates charter movers against other mobile teachers, who are more likely to share some of the omitted variables that are correlated with teachers’ mobility decisions as well as their credentials (e.g., job satisfaction). I estimate equation 2 separately for each of the credentials summarized in table 2: graduate degree, competitive col- lege education, mean licensure test score, regular licensure, three measures of experience, and absenteeism. Controls include receiving and sending cam- sl t ) and receiving county-by-year effects (αl (t+1)). pus characteristics (Cr School characteristics include variables representing student body size and composition (percent nonwhite, percent proﬁcient, and total enrollment), ur- banicity indicators, range of grades served, and a set of dummy variables controlling for missing data. County-by-year effects control for unobserved heterogeneity in regional variables like nonteaching job opportunities. Robust sl (t+1), C s 17. The 95 percent conﬁdence interval for the relationship between percent nonwhite and charter mobility is (1.010, 1.015) versus (1.004, 1.005) for intra-mainstream mobility. 246 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / f / e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d . f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Celeste K. Carruthers standard errors allow for clustering within each sending school and year. If charter schools have higher demand for some qualiﬁcations and if they are able to outbid mainstream schools by manipulating employment terms and working conditions, δc will be positive. If charters have lower demand or if they are unable to realize an advantage in the teachers’ labor market, δc will be insigniﬁcant or negative. More experienced teachers may seek graduate degrees or additional certi- ﬁcations to increase their pay, so I control for teacher experience categories (indicators for fewer than three years’ experience or more than twenty-ﬁve years’ experience) when estimating equation 2 for licensure and education variables. Licensed and unlicensed teachers may have different incentives to consider charter schools. Unlicensed teachers likely have smaller choice sets in the mainstream sector, particularly if they are near the end of their proba- tionary period. We can infer more about the relative appeal of charter schools as workplaces from teachers who have the opportunity to work in either sec- tor. As such, I produce separate estimates of δm and δc for the subsample of licensed teachers. Table 4 lists estimates of δm and δc for each r´esum´e qualiﬁcation. Columns 1 and 2 present results from the full sample. Column 1 lists coefﬁcient estimates for δm, the difference in qualiﬁcation k between teachers moving to mainstream schools and nonmoving teachers. Estimates of δm serve as the baseline with which δc estimates are compared. Mobile teachers are modestly more likely to have graduate degrees and less likely to have graduated from a competitive college. They have lower licensure test scores than nonmovers, and they are 1.3 percentage points less likely to be licensed. Movers are much less experienced, by 3.2 years on average, than their nonmoving counterparts. They are 11.6 percentage points more likely to have three years’ experience or less and 5.8 percentage points less likely to have at least twenty-ﬁve years’ experience. Movers are absent an additional 1.7 days relative to nonmovers. Column 2 coefﬁcients in table 4 answer the question, “Are charter movers more or less qualiﬁed than teachers moving between comparable mainstream schools?” It is important, given the heterogeneity of mainstream opportunity costs and charter school environments, to control for sending and receiving school proﬁles. Nonetheless, results are robust to controls for receiving-school controls alone or to sending-school controls alone. With respect to graduate education, degrees from competitive colleges, licensure, and experience, char- ter movers are signiﬁcantly less qualiﬁed. They are 3.4 percentage points less likely to hold a graduate degree, in agreement with ﬁndings by Hoxby (2002) and Taylor (2005). Charter movers are much less likely to be licensed than other mobile teachers, by 7.2 percentage points. Charter movers are less experi- enced than mainstream movers by 1.2 years, and they are 7.2 percentage points l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / f e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 247 TEACHERS MOVING TO CHARTER SCHOOLS Table 4. Regression Results: Qualiﬁcations of Teachers Changing Schools, by Mainstream/Charter Des- tination Receiving School Type Sample (Equation 2 Coefﬁcient) Holds graduate degree Attended competitive college Mean licensure test score (∼N(0;1)) Regularly licensed Teaching experience (years) Experience ≤ 3 years Experience ≥ 25 years Days absent (1) Mainstream All (δm) (2) Charter All (δc) (3) Mainstream Licensed (δm) (4) Charter Licensed (δc) 0.007 (4.47) −0.005 (3.38) −0.016 (4.99) −0.013 (11.23) −3.2 (82.86) 0.116 (62.18) −0.058 (51.00) 1.7 (37.78) −0.034 (3.17) −0.044 (4.11) −0.020 (0.93) −0.072 (7.58) −1.2 (4.73) 0.072 (6.06) 0.010 (1.40) 0.3 (1.16) 0.012 (6.67) 0.001 (0.53) −0.011 (3.22) −3.2 (76.08) 0.106 (55.23) −0.061 (47.02) 1.7 (36.31) −0.027 (2.13) −0.026 (2.20) 0.052 (2.31) −0.5 (1.78) 0.049 (3.71) 0.028 (2.96) 0.6 (1.61) Notes: n = 867,019 teachers with known school assignments in the following year. Column 1 lists the regression-adjusted mean difference in each qualiﬁcation between teachers moving to traditional, mainstream public schools and nonmovers (δm in equation 2). Column 2 lists the regression-adjusted mean difference in each qualiﬁcation between charter and mainstream movers (δc). Columns 3 and 4 report δm and δc estimates when the analysis is limited to regularly licensed teachers. Control variables include receiving and sending school characteristics (percent nonwhite, performance composite, total enrollment, locale indicators, grade ranges served), a set of dummy variables for school missing data, and receiving county-by-year effects. The absolute values of t- statistics are reported in parentheses below each coefﬁcient. Robust standard errors are clustered within each school and year. more likely to have three or fewer years’ experience. There is no signiﬁcant gap in licensure test scores or absenteeism between charter and mainstream movers. Columns 3 and 4 list results for the subsample of licensed teachers. Lim- iting the sample has little effect on results for mainstream movers; point esti- mates are not economically different between columns 1 and 3. But excluding unlicensed teachers from the analysis narrows or reverses the qualiﬁcation gap between charter and mainstream movers, suggesting that uncertiﬁed main- stream teachers moving to charter schools attenuated the average qualiﬁca- tions of charter movers. The difference between the full and limited samples is particularly stark for licensure test scores and high levels of experience. Licensed teachers moving to charter schools have signiﬁcantly higher licen- sure test scores than other moving teachers, by 5.2 percent of a standard 248 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / f / e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d . f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Celeste K. Carruthers deviation, and they are 2.8 percentage points more likely to have twenty-ﬁve years or more of experience. These ﬁndings emphasize the possibility that teachers view the charter sector as a low-cost job change preceding retirement or permanent career changes. Sample attrition is high among new teachers, highly experienced teachers, uncertiﬁed teachers, and teachers with higher licensure test scores. These are the same groups I observe disproportionately ﬂowing to charter schools. Nonetheless charter schools do not appear to be a quick precursor to attrition for mainstream teachers. Charter movers have a typical post-move duration (uncensored) that is just 12 percent shorter than that of teachers moving to mainstream schools (3.24 years versus 3.67). 5. CLASSROOM PERFORMANCE As noted above, teacher credentials are limited proxies of teacher quality. Even if charter schools are drawing less qualiﬁed individuals, they may be attracting more effective teachers from traditional public schools, which could subsequently increase charter student achievement. North Carolina students in the third through eighth grades take end-of-grade (EOG) exams in math and reading each spring. I assess teachers’ value added using grades 3−5 student EOG records for 2.3 million student-years spanning 1996 to 2008, omitting grade repeaters and test exemptions. To compare teachers across time and grade levels, raw EOG scores are scaled to have a mean of 0 and standard deviation equal to 1 within each year and grade.18 North Carolina is one of the rare settings where teachers can be linked to their students over several years. I utilize this valuable feature of the data to describe the classroom performance of mainstream elementary teachers who ultimately move to charter schools. Exam proctors are linked to each student’s test scores and sociodemo- graphic data. For test takers in elementary grades, exam proctors are usually— but not always—their classroom teachers. To minimize the likelihood of in- valid teacher-student matches, I omit makeup tests, alternative tests, tests for severely disabled students, classrooms with fewer than ﬁve or more than thirty test takers, and tests that accommodate students’ need for multiple sessions, dictation, home testing, or separation from the rest of the class. I also focus on self-contained classrooms whose proctor is found in the assembled panel of teachers. Self-contained classrooms embody the traditional structure of ele- mentary education, where each class of students spends all or the majority of each day with one teacher. These limitations lend considerable validity to each 18. EOG exams are interval scaled across grades, but the range of raw scores shifts over time and tends to compress in higher grades. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . f / / e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 249 TEACHERS MOVING TO CHARTER SCHOOLS allowed teacher-student match. Of the 146,282 EOG test-taking classrooms with a known teacher, 87.3 percent are considered valid matches. I explore the soundness of observed teacher-student matching by cross-referencing EOG records with course membership ﬁles for 2007−9, the only years for which students can be linked with certainty to their teachers of record. Within these years I ﬁnd that 91.4 percent of student-teacher pairs in the analysis sample are matched to veriﬁed student-teacher pairs in course membership records, whereas a naive sample of all test records and all proctors yields just a 74.2 percent match rate. Classroom Performance—Main Results Consider the following model describing student i’s standardized, normalized test score Ak i j t in subject k (math or reading) in teacher j’s classroom, school s, year t: l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / f / e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d . f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Ak i j t = λAk it−1 + X i j t βk X + ¯X −i j t βk ¯X + Tj t βk T + C s t βk C + θ k j + αk s + εk i j t . (3) Equation 3 is an educational production function that controls for once-lagged student achievement (Ak it−1) in place of prior inputs and endowed ability. The model assumes that effects of prior inputs and endowments decay uniformly and geometrically (Todd and Wolpin 2003). These are strong assumptions, but because of the students’ short time series (three years at most), equation 3 is the best available value-added speciﬁcation for the purposes of this study. Variables in X i j t are student characteristics, including race, gender, parental education, and learning disability indicators. ¯X −i j t is a vector controlling for class size and average student characteristics in j’s classroom, excluding student i. T j t is an indicator equal to one if j is a new teacher. C s t contains campus-level variables, including urbanicity indicators and continuous variables for total enrollment, percent proﬁcient, and percent nonwhite. The coefﬁcients θ k j and αk s are teacher ﬁxed effects and school ﬁxed effects, respectively. The results to follow evaluate θ k j estimates under two variations of equation 3: (1) without school ﬁxed effects (assumes αk s is a mean-zero component of the error term) and (2) with school ﬁxed effects. Estimates of θ k j from each speciﬁcation are centered at zero so they can be interpreted as each teacher’s standing relative to the average instructor. Teacher ﬁxed effects from the speciﬁcation of equation 3 without school ﬁxed effects do not account for unobserved, inherent school quality, so any tendency of students experiencing steeper achievement growth to gravitate toward particular schools will bias teacher quality estimates. Teacher ﬁxed effects from the speciﬁcation controlling for school ﬁxed effects reﬂect teachers’ relative performance within their schools, which limits the scope of 250 Celeste K. Carruthers Table 5. Teacher Fixed Effect Estimates: Summary Statistics Teachers’ math ﬁxed effects Without school ﬁxed effects With school ﬁxed effects Teachers’ reading ﬁxed effects Without school ﬁxed effects With school ﬁxed effects (1) All Teachers (2) Mainstream Movers 0.013 (0.207) [8.52∗] 0.011 (0.144) [7.56∗] 0.018 (0.240) [3.09∗] 0.014 (0.171) [2.77∗] −0.007∗ (0.213) −0.002∗ (0.153) −0.005∗ (0.237) −0.002∗ (0.173) (3) Charter Movers −0.037∗ (0.228) −0.024∗ (0.170) −0.033∗ (0.263) −0.029∗ (0.200) n(teacher-years) 124,852 14,728 354 Notes: Teacher ﬁxed effects are estimated by regressing student achievement against current teacher indicators and other inputs in the educational production function, equation 3. Cells list mean teacher ﬁxed effects by subject, speciﬁcation, and mobility status. Standard deviations are in parentheses below each mean, and F-statistics from Wald tests of the joint signiﬁcance of teacher ﬁxed effects are in brackets below each standard deviation. Data for moving teachers reference the year immediately preceding a school change. Mainstream movers (column 2) are mainstream teachers who are next observed in a different mainstream school, and asterisks in column 2 indicate a signiﬁcant difference (at 95% conﬁdence or greater) between mainstream movers and nonmobile teachers. Charter movers (column 3) are next observed in a charter school, and asterisks indicate signiﬁcant differences between mainstream and charter movers. interpretation and understates the variance in teacher quality across schools but adequately addresses between-school sorting. Both models control for students’ prior place in their grade-cohort distribution, so teacher ﬁxed effect estimates represent the degree to which teachers are responsible for advancing their students through the distribution, conditioning on baseline achievement. Controlling for prior achievement necessarily limits the analysis to fourth- and ﬁfth-grade teachers. Teacher ﬁxed effects are estimated for 14,728 mobile individuals, 354 of whom are moving to a charter school. Table 5 summarizes the time-invariant teacher ﬁxed effects generated by each speciﬁcation of equation 3. Since stu- i j t ) is normalized to a ∼N(0,1) scale by grade and year, dent achievement (Ak teacher ﬁxed effects are in terms of student-level standard deviations. That is, a teacher with θ k = 1 would on average advance her students by 1 stan- j dard deviation in subject k relative to the rest of their cohort. The typical l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / f / e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d . f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 251 TEACHERS MOVING TO CHARTER SCHOOLS teacher in this panel advanced students by 0.011−0.018 standard deviations, as listed in the ﬁrst column of table 5, and these positive values are explained by longer teaching spells for above-average teachers. Moving teachers have some- what lower value added than nonmoving teachers, in agreement with recent work by Hanushek and Rivkin (2010). Relative to mainstream movers, charter movers have even lower value added, by 0.022−0.030 student-level standard deviations. Simple mean differences do not control for the type of schools teachers are leaving or moving to, and charter schools may have attracted relatively high-performing teachers, compared with mainstream schools with similar student populations. In parallel to the analysis of r´esum´e qualiﬁcations, I regress teacher ﬁxed effect estimates against mobility indicators, sending and receiving school characteristics, and receiving county-by-year effects: ˆθ k j = δm,k1(moving ) j + δc,k1(tochar ter ) j + Cr + vk . +C s sl t C s + αk βk l (t+1) j sl t sl (t+1) βk Cr (4) Subjects (math and reading) are again indexed by k, teachers by j, schools by s, counties by l, and years by t. Table 6 presents estimates of δm,k and δc,k. Column 1 lists the estimated difference in teacher ﬁxed effects between mainstream movers and nonmovers (δm,k in equation 4), and column 2 lists conditional mean differences in teacher ﬁxed effects between charter and mainstream movers (δc,k). The rate of licensure is nearly 100 percent among elementary teachers with ﬁxed effect estimates, and given the small number of charter movers for whom ﬁxed effects can be estimated, I do not produce separate estimates for the subsample of regularly licensed teachers. In agreement with unconditional mean differences in teacher ﬁxed ef- fects, charter movers have lower ﬁxed effects than teachers moving between mainstream schools, and mainstream movers have lower ﬁxed effects than nonmobile teachers. Point estimates are more precise for gaps in math value added and for gaps in teacher ﬁxed effects estimated without controls for school ﬁxed effects. Recall that without controls for school ﬁxed effects, teacher ﬁxed effects collectively represent that value-added distribution of teachers across schools. Table 6 results indicate that charter movers are on average drawn from the lower half of the teacher effectiveness distribution, by 2.6 percent of a student-level standard deviation in math and 1.8 percent in reading. Per- haps, however, charter schools tend to attract teachers from lower-performing schools, which could bias their value added down if equation 3 without school ﬁxed effects fails to adequately control for student sorting across schools. Table 6 shows that even within sending schools, teachers moving to char- ter schools are 1.9 percent of a standard deviation less effective in reading 252 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . f / / e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d . f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Celeste K. Carruthers Table 6. Regression Results: Math and Reading Fixed Effects of Teachers Changing Schools, by Mainstream/Charter Destination (Equation 4 coefﬁcient) Teachers’ math ﬁxed effects Without school ﬁxed effects With school ﬁxed effects Teachers’ reading ﬁxed effects Without school ﬁxed effects With school ﬁxed effects (1) Mainstream Movers (δm) −0.022 (11.13) −0.023 (10.20) −0.012 (8.97) −0.016 (9.62) (2) Charter Movers (δc) −0.026 (2.08) −0.014 (0.98) −0.018 (1.86) −0.019 (1.67) Notes: n = 124,852 teachers with known following year assignments, of whom 14,728 are changing schools and 354 are moving to a charter school. Column 1 lists regression-adjusted mean differences in teacher ﬁxed effects between movers and nonmovers (δm in equation 4), and column 2 lists regression-adjusted mean differences in teacher ﬁxed effects between charter and non-charter movers (δc). Control variables include receiving and sending school characteristics (percent nonwhite, performance composite, total enrollment, locale indicators, grade ranges served), a set of dummy variables for school missing data, and receiving county-by-year effects. The absolute values of t-statistics are reported in parentheses below each coefﬁcient. Robust standard errors are clus- tered within each school and year. instruction than other mobile teachers and are at best equivalent to other mobile teachers in terms of math value added. A 0.018−0.026 gap between these two groups of mobile teachers has statistical and practical signiﬁcance. Unreported coefﬁcient estimates from equation 3 (without school ﬁxed effects) suggest that ﬁrst-year teachers are associated with signiﬁcantly lower student achievement: 0.069 standard de- viations in math and 0.038 standard deviations in reading. These are very similar to returns to teacher experience estimated by Clotfelter, Ladd, and Vigdor (2007). Thus the difference between a teacher moving to the charter sector and a teacher moving elsewhere is 38−47 percent of the effectiveness gap between new and more experienced teachers. Knowing that many char- ter movers are themselves new or inexperienced teachers, we can conclude that North Carolina’s charter schools are not recruiting teachers with higher value added from traditional, mainstream public schools,19 although the causal 19. These ﬁndings are robust to several corrections for observational biases in the estimated gap between charter and mainstream movers’ value added. For instance, if mobile teachers tend to l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . f / / e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 253 TEACHERS MOVING TO CHARTER SCHOOLS mechanisms behind this ﬁnding remain to be seen. Charters may have lower demand for effective teachers, perhaps because of an emphasis on untested skills, or the schools may have insufﬁcient resources to attract more effec- tive teachers. It bears repeated emphasis that resources are not limited to salary funds. Nonpecuniary conditions like school quality, school location, in- stitutional and administrative experience, student and parental engagement, and regulatory compliance burdens combine with salary to form the utility a teacher expects to enjoy in a new charter school. If this utility cannot dominate other options, charters will have difﬁculty recruiting good teachers. Results in the following subsection indicate that the effectiveness of mobile teach- ers varies by some of these nonpecuniary conditions—speciﬁcally, schoolwide effectiveness, urbanicity, and student racial composition. Heterogeneous Teacher Flows between Mainstream and Charter Schools Given the idiosyncratic nature of charter schools and their students, it may be the case that some charters draw more qualiﬁed or more effective teachers than others. Moreover, heterogeneous teacher ﬂows have important policy im- plications if, for instance, more disadvantaged mainstream schools tend to lose more effective teachers to charters. To assess the relationships among teach- ers’ qualiﬁcations, value added, and features of their sending and receiving schools, I estimate additional speciﬁcations of equations 2 and 4, adding an interaction between the charter mobility indicator 1(tocharter) j and one send- ing or receiving school characteristic. In order to compare teacher effectiveness gaps across schools, I use teacher ﬁxed effect ( ˆθ k j ) estimated by speciﬁcations of equation 3 without school ﬁxed effect controls. School characteristics that are interacted with charter mobility include an indicator for urban locales, the share of students who are proﬁcient, the share of students who are nonwhite, or schools’ own math or reading value added. Schools’ math and reading value added are time-varying school ﬁxed effects estimated by the following: Ak i j t = λAk it−1 + X i j t βk X + ¯X −i j t βk ¯X + Tj t βk T + αk s t + εk i j t . (5) be more effective in their new schools (Jackson 2010), the time-invariant ﬁxed effects of charter movers could be biased downward, since their performance while in charter schools cannot be observed. I replicate the analysis after simulating these data limitations for all moving teachers. Speciﬁcally, I reestimate teacher ﬁxed effects, neglecting to observe any classroom performance following a school change. Results are less precise but nonetheless suggest a 0.012−0.022 standard deviation gap in teacher ﬁxed effects between mainstream movers and nonmovers (very similar to the ﬁrst column of table 6), and a large and signiﬁcant gap in math ﬁxed effects between charter and mainstream movers. Another possibility is that the comparison group (mainstream movers statewide) is inadequate because some of them presumably do not have charter schools in their choice set. If, for instance, mainstream movers in districts without charter schools tend to have higher value added, their inclusion would widen estimates of the charter-mainstream mover gap reported in table 6. I address this possibility by estimating equation 4 for a limited sample of teachers with at least one charter elementary school in their sending county. Results indicate an even larger gap in value added between mainstream and charter movers. 254 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . f / / e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d . f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Celeste K. Carruthers s t ) vary intertemporally. Estimates of αk Equation 5 is much like the educational production function represented by equation 3, with three differences. Most important for the current context, school ﬁxed effects (αk s t represent each school’s average value added in a given year. This allows me to compare each mobile teacher’s value added with that of the current quality of her next school and further to preclude bias from any effect she may have on that school’s average quality after moving there. Second, the sample of students is broader for equation 5, including students without self-contained classrooms or veriﬁed teachers. This allows me to estimate school ﬁxed effects for charter students, who are rarely linked to classroom teachers in these data. Finally, teacher ﬁxed effects are omitted from equation 5, in part because of weaker teacher-student linkages in this broader sample but also to identify variation in school quality across teachers. Results are reported in table 7. Each coefﬁcient represents the change in the gap between mainstream and charter movers’ qualiﬁcations or value added associated with a marginal increase in urbanicity, student proﬁciency, percent of students who are nonwhite, or schoolwide effectiveness. For instance, the upper panel of table 7, column 1, indicates that teachers leaving urban mainstream schools for the charter sector are even less experienced than other charter movers, by a weakly signiﬁcant 1.64 years. The lower panel shows that teachers moving to urban charter schools are not signiﬁcantly different from other charter movers in terms of credentials or math value added, but they have lower value added in reading by 3.6 percent of a student-level standard deviation. The rate of student proﬁciency in sending schools (i.e., the share of students at grade level according to the state accountability system) has little relationship with the type of teachers leaving for charters, but more proﬁcient charters draw teachers with higher licensure test scores and modestly less ex- perience. Mainstream schools with more nonwhite students tend to lose more licensed teachers to the charter sector as well as teachers with higher licensure test scores, but they do not differentially lose more or less effective teachers. Charter schools with more nonwhite students tend to attract less qualiﬁed teachers in terms of selective college education, licensure, and licensure test scores. Furthermore, the reading value added of teachers moving to charter schools with more nonwhite students is lower, by a weakly signiﬁcant 0.004 student-level standard deviations per 10 percentage point increase in nonwhite students. Recalling the bimodal distribution of nonwhite stu- dent shares illustrated in ﬁgure 1, this means that teachers moving to a 90 percent nonwhite school are typically 0.032 standard deviations less ef- fective in reading instruction than teachers moving to a 10 percent non- white school. The upper panel of table 7 shows that more effective main- stream schools tend to lose teachers with higher value added to the charter l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / f e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 255 TEACHERS MOVING TO CHARTER SCHOOLS Table 7. Regression Results: Qualiﬁcations and Classroom Performance of Teachers Moving to Charter Schools, by Characteristics of Sending and Receiving Schools (1) (2) Proﬁcient (3) (4) (5) Nonwhite School Math School Reading Urbana Students (%) Students (%) Fixed Effectb Fixed Effectb Teacher Credentials and Value Added by Characteristics of Sending Schools Holds graduate degree 0.008 (0.20) Attended competitive college 0.002 (0.06) Regularly licensed Mean licensure test score Experience (years) Days absent Teacher math ﬁxed effectc 0.009 (0.32) 0.125 (1.67) −1.64 (1.85) −0.24 (0.22) 0.004 (0.18) Teacher reading ﬁxed effectc −0.010 (0.55) 6.0E-5 (0.06) 3.9E-4 (0.31) 0.001 (1.72) 0.004 (1.56) 0.01 (0.32) 0.01 (0.26) −4.0E-5 (0.05) 1.4E-4 (0.27) 1.9E-4 (0.29) −0.001 (1.56) −0.001 (2.32) −0.004 (2.86) −0.01 (0.37) 0.00 (0.06) 1.3E-4 (0.29) −2.9E-4 (0.91) 0.017 (0.14) −0.218 (1.61) −0.016 (0.16) −0.438 (1.85) −3.54 (1.20) −4.16 (0.97) 0.230 (2.54) 0.003 (0.04) Teacher Credentials and Value Added by Characteristics of Receiving Schools Holds graduate degree −0.042 (1.00) −0.001 (1.13) Attended competitive college 0.004 (0.08) Regularly licensed Mean licensure test score Experience (years) Days absent 0.004 (0.15) 0.067 (0.77) −0.50 (0.47) 1.16 (0.84) Teacher math ﬁxed effectc −0.035 (1.47) Teacher reading ﬁxed effectc −0.036 (2.03) 0.001 (1.48) 8.0E-5 (0.17) 0.003 (2.04) −0.04 (2.03) 0.01 (0.39) −1.0E-4 (0.19) −2.5E-4 (0.68) 2.2E-4 (0.44) −0.001 (2.15) −0.001 (2.43) −0.005 (4.60) 0.01 (0.65) 0.01 (0.84) −4.7E-4 (1.37) −4.4E-4 (1.66) −0.159 (1.80) −0.091 (0.92) 0.015 (0.28) 0.035 (0.18) −1.25 (0.59) 3.68 (1.28) 0.280 (2.98) 0.128 (2.06) 0.101 (0.88) −0.015 (0.09) 0.125 (1.03) 0.351 (1.45) −3.16 (0.97) −1.74 (0.42) 0.194 (2.29) 0.170 (2.27) −0.081 (0.69) 0.053 (0.35) −0.048 (0.54) −0.019 (0.07) 3.25 (1.26) −3.78 (0.95) 0.112 (0.90) 0.106 (1.17) Notes: The table lists estimated coefﬁcients from regressions of teachers’ credentials and value added on mobility indicators and characteristics of their sending or receiving schools (i.e., equations 2 and 4), including interactions between the charter mobility indicator and one sending or receiving school characteristic. The absolute values of t-statistics are reported in parentheses below each coefﬁcient. Robust standard errors are clustered within each school and year. aAn urban area is deﬁned as an incorporated place inside a metropolitan statistical area with a population of at least 250,000. bSchool ﬁxed effects are estimated by equation 5. cTeacher ﬁxed effects—limited to elementary teachers with self-contained classrooms—are esti- mated by equation 3, omitting school ﬁxed effects. 256 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / f / e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d . f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Celeste K. Carruthers sector.20 The lower panel shows that charters with higher schoolwide value added in math tend to receive teachers with higher value added in both sub- jects. Charter school math ﬁxed effects have a standard deviation of 0.262. Thus increasing the schoolwide quality of math instruction by 1 standard de- viation (roughly the difference between a median charter school and a 90th percentile charter school) is associated with a 0.073 student-level standard deviation gain in recruited teachers’ math value added and a 0.034 standard deviation gain in their reading value added, which more than compensates for the 0.018−0.026 standard deviation gap in charter movers’ value added reported in table 7. The positive relationship between teachers’ value added and that of their receiving charter schools is illustrated in ﬁgure 2, where a mean-smoothing local polynomial plots the average value added of mobile teachers against the value added of receiving charters. The upward trend in matched teacher-school effectiveness is apparent for both subjects. Note, however, that even the most effective charter schools tend to draw teachers with below-average (i.e., below zero) value-added estimates. The tendency of more effective teachers to move to more effective char- ter schools has important implications for the state’s charter sector, chieﬂy because few North Carolina charters outperform traditional public schools. And given the well-documented importance of teacher quality in supporting student learning, recruiting less effective teachers from mainstream schools can reinforce the disadvantages charters face in raising student achievement. Robustness Check—Biases from Student Sorting Teacher ﬁxed effects are interpreted as each individual’s history of classroom performance relative to expectations, which should be important to schools looking to hire teachers with a record of success in raising student test scores, but nonrandom sorting of students between and within schools can lead to biased estimates of teacher effectiveness. School ﬁxed effects control for the tendency of high-growth students to group within particular schools, but they do not control for any systematic sorting of students within schools. An ex- ample of such sorting is “teacher shopping” by parents, which has a wealth of anecdotal support. In general, there are abundant practical reasons why school leaders might not want to randomly assign students to classrooms. Rothstein (2010), using a subset of the North Carolina data employed here, shows that 20. This result is in part, but not wholly, due to the exclusion of school ﬁxed effects in speciﬁcations estimating teacher ﬁxed effects. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / f e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d . f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 257 TEACHERS MOVING TO CHARTER SCHOOLS 5 0 . 0 5 0 . − 1 . − 2 0 . 0 2 0 . − 4 0 . − 6 0 . − 8 0 . − −.4 −.2 .2 0 Receiving Charter Value Added in Math .4 Mean Math Value Added of Charter Movers 95% Confidence Interval n=306 teachers moving to charter schools l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / f e d u e d p a r t i c e - p d l f / −.2 0 .2 .4 Receiving Charter Value Added in Reading Mean Reading Value Added of Charter Movers 95% Confidence Interval n=306 teachers moving to charter schools / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d . f Figure 2. Panels 1 and 2 illustrate mean-smoothing estimates and conﬁdence intervals for the relationship between charter movers’ value added and the schoolwide value added of the charters they are moving to. School ﬁxed effects are estimated by equation 5. Teacher ﬁxed effects—limited to elementary teachers with self-contained classrooms—are estimated by equation 3, omitting school ﬁxed effects. common value-added methodologies falsely ascribe signiﬁcant value to stu- dents’ future teachers, which is symptomatic of sorting biases.21 Sorting biases would affect the analysis only to the degree that charter movers are more affected by nonrandom student sorting than other mobile f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 21. Much attention has been devoted to resolving this issue, and I apply some of the lessons learned to the evaluation of charter movers. Koedel and Betts (2011), for instance, show that teacher effect estimates derived from multiple years of data are less subject to sorting biases than single-year classroom effects. Kane and Staiger (2008) conclude that controls for a student’s lagged achievement and observable characteristics of peers within his classroom are sufﬁcient to drive the bias from nonrandom sorting to zero. Here equation 3 is estimated for a multicohort panel, generating teacher ﬁxed effect estimates covering multiple school years. Lagged achievement and peer characteristics— at the classroom and school levels—are included in all speciﬁcations. 258 Celeste K. Carruthers teachers. I investigate this possibility directly by substituting following year teacher indicators for current teacher indicators in equation 3 and then comparing the false “effects” of movers and nonmovers and of charter and mainstream movers. Summary statistics for false effects are listed in table 8. Charter movers are not signiﬁcantly more connected to their students’ prior achievement than are mainstream movers. Controls for school ﬁxed effects and/or lagged achievement are not sufﬁcient to drive the joint signiﬁcance of false effects to zero (as indicated by small but signiﬁcant F-statistics for false effects), but they are sufﬁcient to eliminate any signiﬁcant, spurious difference between mainstream movers and charter movers. It should be noted that these false effects reﬂect multiple years of sorting, not trends in sorting that may have led a teacher to change schools. In a related test, I supplement the educational production function described by equation 3 with indicators for the mobility of students’ future teachers (i.e., indicating whether one’s year t + 1 teacher moved to a mainstream or charter school in t + 2). If charter movers are systematically assigned more or less proﬁcient students than other mobile teachers in the year immediately preceding a school change, the coefﬁcient on leading charter mobility should be signiﬁcantly different from that of leading mainstream mobility. Results are in agreement with those already discussed: lagged achievement and school ﬁxed effects adequately control for signiﬁcant pre-mobility differences in teachers’ students.22 Robustness Check—Biases from Sampling Error Sampling error from ﬁnite panel length and class size cause the variance of teacher ﬁxed effects to overstate the variance of true value added. Furthermore, if sampling error disproportionately affects certain groups of teachers who are more likely to transition to charter schools (new teachers, for instance), com- paring teacher ﬁxed effects may put charter movers at a disadvantage. Follow- ing several recent studies, I partition the variance in persistent teacher quality from that of sampling error in students’ residual achievement and construct estimates of teacher effectiveness that account for likely sampling error.23 These variance components are used to approximate teachers’ persistent value added. Drawing from Carrell and West (2008) and Kane and Staiger (2008), I construct teacher j’s value added by scaling her classes’ residual performance toward zero according to an estimate of her signal-to-total variance ratio. This Bayesian shrinkage estimator disproportionately attenuates the value added 22. I thank an anonymous referee for suggesting this test. 23. See, e.g., Carrell and West 2008; Kane, Rockoff, and Staiger 2008; Kane and Staiger 2008; and Hanushek and Rivkin 2010. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . f / / e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 259 TEACHERS MOVING TO CHARTER SCHOOLS Table 8. Future Teachers’ False Effects: Summary Statistics Teachers’ false math ﬁxed effects Without school ﬁxed effects With school ﬁxed effects Teachers’ false reading ﬁxed effects Without school ﬁxed effects With school ﬁxed effects (1) All Teachers (2) Mainstream Movers 0.002 (0.194) [2.97∗] 0.001 (0.188) [1.85∗] 0.004 (0.179) [1.71∗] 0.005 (0.185) [1.35∗] −1.87E-04 (0.202) 0.002 (0.198) −0.001∗ (0.192) 0.003 (0.199) (3) Charter Movers −0.001 (0.242) 0.004 (0.227) −0.005 (0.242) −0.007 (0.241) n(teacher-years) 66,478 8,073 184 indicators and other Notes: Future teacher “effects” are estimated by regressing student achievement against leading teacher inputs in the educational production function, equation 3. Column 1 lists average false ﬁxed effects for all teachers. Standard devia- tions are in parentheses below each mean, and F-statistics from Wald tests of the joint signiﬁcance of false teacher ﬁxed effects are in brackets below each standard deviation. Data for moving teachers are evaluated in the year immediately preceding a school change. Column 2 lists mean false ﬁxed effects for teachers moving to mainstream schools, with standard deviations in parentheses below each mean, and asterisks indicating statistically signiﬁcant differences (at 95% conﬁdence or greater) relative to nonmovers. Column 3 lists mean false ﬁxed effects for teachers moving to charter schools, with standard deviations in parentheses below each mean, and asterisks indicating signiﬁcant differences between mainstream and charter movers. of less experienced teachers who are expected to have been more affected by sampling error. Computational details are provided in the appendix. Figure 3 illustrates the effect of Bayesian scaling and affords a visual com- parison of future charter teachers’ value added versus that of their exclusively mainstream counterparts. Panel 1 plots comparative kernel densities of teach- ers’ math ﬁxed effects, controlling for students’ lagged achievement but not school ﬁxed effects. Panel 2 plots densities of teachers’ mathematics value- added estimates from Bayesian shrinkage estimators, again controlling for lagged achievement but not school ﬁxed effects.24 The distribution of future charter teachers’ value added is signiﬁcantly left of the distribution of other mainstream teachers’ value added, regardless of Bayesian scaling. Wilcoxon 24. Figures for reading value added are qualitatively equivalent to ﬁgure 3. 260 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / f e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Celeste K. Carruthers y t i s n e D d e t a m i t s E 2 5 . 1 1 5 . 0 y t i s n e D d e t a m i t s E 4 3 2 1 0 −1 −1.5 −1 −.5 0 Estimated Teacher Value Added .5 Never a charter teacher (n=20259) Future charter teacher (n=432) −.5 0 Estimated Teacher Value Added .5 1 1 1.5 Never a charter teacher (n=20259) Future charter teacher (n=432) Figure 3. Panels 1 and 2 illustrate kernel density estimates of future charter teachers’ math value added (solid line), relative to the same for exclusively mainstream teachers (dashed line). Panel 1: distribution of teacher ﬁxed effects, controlling for lagged student achievement. Panel 2: distribution of persistent value added (via Bayesian shrinkage estimators), controlling for lagged achievement. Densities are estimated using Epanechnikov kernel functions and halfwidths of 0.025 standard deviations. rank-sum tests reject the hypothesis that future charter teachers and exclu- sively mainstream teachers are drawn from the same distribution of value added. Figure 3 provides further evidence that teachers ﬂowing to the char- ter sector typically have lower classroom performance than other mainstream teachers, but also demonstrates considerable overlap and variation in teacher quality across groups. 6. CONCLUSIONS, IMPLICATIONS, AND CAVEATS In many ways, autonomous charter schools are well positioned to exploit any inefﬁciencies in monopsonistic markets for public teachers. Mainstream l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . f / / e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d . f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 261 TEACHERS MOVING TO CHARTER SCHOOLS teachers in North Carolina, as in other systems, are paid according to rigid salary schedules that climb steadily with experience and graduate degrees, despite compelling evidence that the returns to experience are largely ex- hausted after the ﬁrst few years of a teacher’s career and that the returns to graduate degrees are insigniﬁcant. Charter administrators are free to hire, compensate, and ﬁre according to merit and robust signals of teacher qual- ity. They also have the ﬂexibility to structure work environments that are more appealing to teachers, whereas many elements of mainstream employ- ment (especially curricula) are centrally managed. But do charter schools have the resources to exploit these inefﬁciencies and attract good teachers from the mainstream? Tighter budgets, institutional inexperience, and chal- lenging student populations may limit the appeal of working in a charter school. I ﬁnd that the North Carolina teachers who leave the mainstream pub- lic school sector for charter schools are less qualiﬁed and less effective than other mobile teachers, even in the presence of controls for sending and re- ceiving school environments. These results suggest that charters have lower demand for these teacher qualities or that charters have insufﬁcient resources to outbid competing mainstream schools, or both. The relative risk of charter mobility increased with nonwhite student shares in mainstream schools, so choice schools may exacerbate higher turnover in high-minority schools. It is important to note, however, that charters could reduce overall teacher turnover by offering a viable alternative to nonteaching careers. Charter movers resem- bled teachers leaving North Carolina public schools more so than other mobile teachers, but charter movers taught for another 3.24 years on average. Low- performing or high-minority mainstream schools do not lose substantially more effective or more qualiﬁed teachers to the charter sector, but among re- cipient charters better teachers gravitate to better schools, schools with fewer nonwhite students, and schools in less urban areas. These patterns will likely reinforce subpar achievement in North Carolina’s charter schools. Three important caveats and open questions must be emphasized along- side these results. First, charter teachers’ value added while in charter schools remains an important but unexplored topic, largely because of data limita- tions. It may be the case that charter teachers recruited straight from col- lege or other careers are much more effective than teachers who opt out of traditional public schools. Or it is possible that charter movers become much more effective upon entering a charter school. These two possibilities seem unlikely to reverse this study’s implication that North Carolina’s charter teachers have lower value added, given persistently low student achievement in the state’s charter system. Second, although North Carolina’s charter in- frastructure resembles that of many other systems in terms of ﬁnance and 262 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / f / e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d . f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Celeste K. Carruthers regulation, results may not generalize to other settings. Charters may be bet- ter suited to attract talented teachers in settings where charter schools out- perform traditional public schools (as in successful urban charter programs). Similarly, in areas with thicker choice markets (i.e., with more competitive choice systems or looser caps on the number of charter schools), the typical charter school may be more appealing to teachers with options in multiple sectors. Finally, this study joins many others in using “teacher quality” and econometric estimates of “value added” interchangeably. But rather than being below-average teachers, it is possible that charter movers are less devoted to standardized testing or that charters have lower relative demand for teachers who advance tested achievement rather than other school objectives. These possibilities are difﬁcult to explore with administrative data but present an opportunity for observational or qualitative research. Nonetheless, if North Carolina’s EOG exams reﬂect a set of appropriate cognitive standards, the relative ability of future charter teachers to raise student achievement has important implications for the successful application of school choice. Re- garding this dimension of teacher quality, North Carolina charter teachers compare unfavorably to other teachers while they are teaching in mainstream schools. I am grateful to the North Carolina Education Research Data Center at Duke University for data access and technical support. I am also grateful to the University of Florida Lockhart Endowment and Walter-Lanzillotti Award for research and travel support. I am indebted to many individuals for helpful comments and suggestions received throughout the evolution of this study: David Figlio, Larry Kenny, Sarah Hamersma, Paul Sindelar, Scott Carrell, Dan Goldhaber, Matthew Kim, Thomas Dee, two anony- mous reviewers, and participants of the 2008 American Education Finance Association meetings, 2008 Association for Public Policy and Management meetings, and 2008 Southern Economic Association meetings. Stata 12 programs used to generate results are available on request. All errors are my own. REFERENCES Abdulkadiroglu, Atila, Joshua Angrist, Susan Dynarski, Thomas J. Kane, and Parag Pathak. 2011. Accountability and ﬂexibility in public schools: Evidence from Boston’s charters and pilots. Quarterly Journal of Economics 126(2): 699–748. Bifulco, Robert, and Katina Bulkley. 2008. Charter schools. In Handbook of research in education ﬁnance and policy, edited by Helen F. Ladd and Edward B. Fiske, pp. 425–46. New York: Routledge. Bifulco, Robert, and Helen F. Ladd. 2006. The impacts of charter schools on student achievement: Evidence from North Carolina. Education Finance and Policy 1(1): 50– 90. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / f / e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 263 TEACHERS MOVING TO CHARTER SCHOOLS Bifulco, Robert, and Helen F. Ladd. 2007. School choice, racial segregation, and test- score gaps: Evidence from North Carolina’s charter school program. Journal of Policy Analysis and Management 26(1): 31–56. Booker, Kevin, Scott Gilpatric, Timothy Gronberg, and Dennis Jansen. 2008. The effect of charter schools on traditional public school students in Texas: Are children who stay behind left behind? Journal of Urban Economics 64(1): 123–45. Boyd, Donald, Pamela Grossman, Hamilton Lankford, Susanna Loeb, and James Wyckoff. 2006. How changes in entry requirements alter the teacher workforce and affect student achievement. Education Finance and Policy 1(2): 176–216. Budde, Ray. 1988. Education by charter: Restructuring school districts key to long-term continuing improvement in American education. Andover, MA: Regional Laboratory for Educational Improvement of the Northeast & Islands. Carrell, Scott E., and James E. West. 2008. Does professor quality matter? Evidence from random assignment of students to professors. NBER Working Paper No. 14081. Carruthers, Celeste K. 2012. New schools, new students, new teachers: Evaluating the effectiveness of charter schools. Economics of Education Review 31(2): 280–92. Clotfelter, Charles T., Helen F. Ladd, and Jacob L. Vigdor. 2007. Teacher credentials and student achievement: Longitudinal analysis with student ﬁxed effects. Economics of Education Review 26(6): 673–82. Dee, Thomas S. 1998. Competition and the quality of public schools. Economics of Education Review 17(4): 419–27. Dobbie, Will, and Roland G. Fryer. 2011. Are high-quality schools enough to increase achievement among the poor? Evidence from the Harlem children’s zone. American Economic Journal: Applied Economics 3(3): 158–87. Falch, Torberg, and Bjarne Strøm. 2005. Teacher turnover and non-pecuniary factors. Economics of Education Review 24(6): 611–31. Gleason, Philip, Melissa Clark, Christina Clark Tuttle, and Emily Dwoyer. 2010. The evaluation of charter school impacts: Final report. NCEE 2010-4029. Washington, DC: U.S. Department of Education, Institute of Education Sciences. Goldhaber, Dan. 2007. Everybody’s doing it, but what does teacher testing tell us about teacher effectiveness? Journal of Human Resources 42(4): 765–94. Goldhaber, Dan. 2008. Teachers matter, but effective teacher quality policies are elu- sive. In Handbook of research in education ﬁnance and policy, edited by Helen F. Ladd and Edward B. Fiske, pp. 146–65. New York: Routledge. Hanushek, Eric A., John F. Kain, and Steven G. Rivkin. 2004. Disruption versus Tiebout improvement: The costs and beneﬁts of switching schools. Journal of Public Economics 88(9–10): 1721–46. Hanushek, Eric A., and Steven G. Rivkin. 2010. Constrained job matching: Does teacher job search harm disadvantaged urban schools? NBER Working Paper No. 15816. 264 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / f e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Celeste K. Carruthers Hastings, Justine S., Thomas J. Kane, and Douglas O. Staiger. 2009. Heteroge- neous preferences and the efﬁcacy of public school choice. NBER Working Paper No. 12145. Hoxby, Carolyn M. 2002. Would school choice change the teaching profession? Journal of Human Resources 37(4): 846–91. Hoxby, Carolyn M. 2003. School choice and school productivity: Could school choice be a tide that lifts all boats? In The economics of school choice, edited by Carolyn M. Hoxby, pp. 287–342. Chicago: University of Chicago Press. Hoxby, Carolyn M., and Sonali Murarka. 2009. Charter schools in New York City: Who enrolls and how they affect their students’ achievement. NBER Working Paper No. 14852. Jackson, C. Kirabo. 2010. Match quality, worker productivity, and worker mobility: Direct evidence from teachers. NBER Working Paper No. 15990. Jackson, C. Kirabo. 2011. School competition and teacher labor markets: Evidence from charter school entry in North Carolina. NBER Working Paper No. 17225. Kane, Thomas J., Jonah E. Rockoff, and Douglas O. Staiger. 2008. What does certiﬁ- cation tell us about teacher effectiveness? Evidence from New York City. Economics of Education Review 27(6): 615–31. Kane, Thomas J., and Douglas O. Staiger. 2008. Estimating teacher impacts on student achievement: An experimental evaluation. NBER Working Paper No. 14607. Koedel, Cory, and Julian R. Betts. 2011. Does student sorting invalidate value-added models of teacher effectiveness? An extended analysis of the Rothstein critique. Educa- tion Finance and Policy 6(1): 18–42. Kolderie, Ted. 1990. Beyond choice to new public schools: Withdrawing the exclusive franchise in public education. Policy Report No. 8. Washington, DC: Progressive Policy Institute. Lanier, Josh. 2008. Teachers’ questions about pay unanswered: School’s ﬁnancial woes topic of meeting with faculty, parents. Independent Tribune (Concord, NC), 22 February. Loeb, Susanna, Demetra Kalogrides, and Tara B´eteille. 2012. Effective schools: Teacher hiring, assignment, development, and retention. Education Finance and Policy 7(3): 269–304. Malloy, Courtney L., and Priscilla Wohlstetter. 2003. Working conditions in charter schools: What’s the appeal for teachers? Education and Urban Society 35(2): 219–41. Murnane, Richard J., and Barbara R. Phillips. 1981. Learning by doing, vintage, and selection: Three pieces of the puzzle relating to teaching experience and teaching performance. Economics of Education Review 1(4): 453–65. Muschkin, Clara, Kara Bonneau, and Kenneth Dodge. 2009. North Carolina Edu- cation Research Data Center. Grant No. 200300138. Annual report to the Spencer Foundation. Available www.childandfamilypolicy.duke.edu/pdfs/projects/NCERDC SpencerReport2009 ﬁnal.pdf. Accessed 20 January 2012. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / f / e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 265 TEACHERS MOVING TO CHARTER SCHOOLS Podgursky, Michael, and Dale Ballou. 2001. Personnel policy in charter schools. Wash- ington, DC: Thomas B. Fordham Foundation. Rockoff, Jonah E. 2004. The impact of individual teachers on student achievement: Evidence from panel data. American Economic Review 94(2): 247–52. Rothstein, Jesse. 2010. Teacher quality in educational production: Tracking, decay, and student achievement. Quarterly Journal of Economics 125(1): 175–214. Scaﬁdi, Benjamin, David L. Sjoquist, and Todd R. Stinebrickner. 2007. Race, poverty, and teacher mobility. Economics of Education Review 26(2): 145–59. Stuit, David A., and Thomas M. Smith. 2012. Teacher turnover in charter schools. Economics of Education Review 31(2): 268–79. Taylor, Lori L. 2005. Teacher cream skimming: Do charter schools attract the most effective teachers? Unpublished paper, Texas A&M University. Todd, Petra E., and Kenneth I. Wolpin. 2003. On the speciﬁcation and estimation of the production function for cognitive achievement. Economic Journal 113(485): F3– F33. APPENDIX: CONSTRUCTION OF BAYESIAN SHRINKAGE ESTIMATORS Consider equation 3, omitting teacher ﬁxed effects (θ j ) and suppressing subject notation: Ai j t = λAit−1 + X i j t βX + ¯X −i j t β ¯X + Tj t βT + C s t βC + αs + εi j t . (A1) Residuals are ei j t = θ j + η j t + εi j t , where θ j is the persistent effectiveness with which teacher j can advance students’ place in their cohort distribution, η j t represents nonpersistent classroom shocks, and εi j t represents nonper- sistent student shocks. An example of η j t would be the shared effect of a dog barking outside the classroom on test day, and εi j t could be driven by student i having a uniquely bad morning prior to taking his end-of-grade exam. I estimate equation A1 with and without school ﬁxed effects (αs ) and decompose residual variance into the variance of each component: ˆσ 2 η , and ˆσ 2 ε . Of particular interest is ˆσ 2 θ , the standard deviation of persistent teacher quality. θ , ˆσ 2 The average student residual for each class can be expressed as: ˆ¯e j t = θ j + η j t + 1 Nj t (cid:3) Nj t i=1 εi j t , (A2) where N j t is class size for year t. If θ j , η j t , and ¯ε j t are independent, the vari- ance of ˆ¯e j t across teachers can be decomposed into the variance of persistent value added and the variance of nonpersistent error: E[ˆ¯e 2 η + σ 2 θ + σ 2 s ], j ct where σ 2 θ is the variance of persistent teacher quality, σ 2 η is the variance of = σ 2 266 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / f / e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d . f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Celeste K. Carruthers classroom-by-year residuals not attributable to teachers, and σ 2 is the vari- s ance of student-by-year residuals not attributable to teacher or classroom ef- fects. Consider two average residuals from two different classes taught by the same teacher: ˆ¯e j t and ˆ¯e j t (cid:6), where t (cid:7)= t (cid:6). If the three residual components are uncorrelated contemporaneously, and if nonpersistent shocks are uncorrelated intertemporally, then E[ˆ¯e j t ˆ¯e j t (cid:6) ] = σ 2 θ . (A3) The assumption that θ j , η j t , and ¯ε j t are uncorrelated is nontrivial: in fact, it is one of the assumptions that must be met in order to interpret estimated teacher ﬁxed effects as unbiased measures of teacher quality. Positive match- ing of better students with better teachers, for instance, will increase estimates of σ 2 θ . In addition, omitting teacher ﬁxed effects in equation A1 may bias other coefﬁcients if they are correlated with θ j ; this in turn will bias estimated resid- uals, ˆ¯e j t . Controlling for school ﬁxed effects in equation A1 limits biases from between-school sorting, but within-school assignment patterns may nonethe- less affect σ 2 θ estimates. Following Carrell and West (2008), I estimate σ 2 θ by computing the pair- wise covariance of classroom-averaged residuals between teacher j’s class in year t and all j classes in years t (cid:6) (cid:7)= t: ˆσ 2 θ = (cid:4)(cid:3) J j =1 (cid:3) C j t=1 (cid:5) ˆ¯e j t ˆ¯e j t (cid:6) /N, (A4) where J is the number of teachers in the sample, C j is the number of classes taught by teacher j, and N is the number of same-teacher pairs. The remaining steps follow Kane and Staiger (2008). The variance of student-by-year residuals is approximated by ˆσ 2 θ = var (ei j t − ¯e j t ), the variance of deviations from class means. The variance of class-by-year residuals is taken to be the gap between the total variance of errors and the sum of teacher- induced and student-by-year residual variance: ˆσ 2 θ + ˆσ 2 ε ). C j For each teacher j, I compute ˜e j = t=1 ω j t ¯e j t , a weighted average of her classroom-averaged residuals. Weights are as follows: η = var (ei j t ) − ( ˆσ 2 (cid:2) ⎛ ⎝ ω j t = ⎞ ⎡ 1 η + ˆσ 2 ˆσ 2 ε Nj t ⎠ ∗ (cid:3) ⎣ Tj s =1 ⎤ −1 ⎦ . 1 η + ˆσ 2 ˆσ 2 ε Nj s (A5) Note that weights favor classes with more students. As class size grows, sam- pling error is expected to diminish. Class size per se is included as a control variable in educational production function regressions, so losses from attend- ing larger classes are reﬂected in teachers’ value-added estimates. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . f / / e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 267 TEACHERS MOVING TO CHARTER SCHOOLS The empirical Bayes estimator of each teacher’s value added is computed by scaling ˜e j toward zero by the approximated signal-to-total variance ratio in residual classroom performance: ˆθ Bayes j = ˜e j ∗ (cid:15) (cid:14) σ 2 θ var ( ˜θ j ) ⎡ var (˜e j ) = ˆσ 2 θ + ⎣ (cid:3) C j s =1 1 η + ˆσ 2 ˆσ 2 ε Nj s ⎤ −1 ⎦ . (A6) (A7) Note that the only components of the scaling factor that are unique to teacher j are C j , the number of classes she taught in the panel, and N j t , the number of students in a particular class. The scaling factor multiplying ˆθ Bayes is increasing in N j t and C j , so ˆθ Bayes j or more experience. will be scaled by less for teachers with more students j l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / f / e d u e d p a r t i c e - p d l f / / / / / 7 3 2 3 3 1 6 8 9 3 4 2 e d p _ a _ 0 0 0 6 7 p d . f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 268 THE QUALIFICATIONS AND image

下载pdf