TEACHER MOBILITY, SCHOOL

SEGREGATION, AND PAY-BASED

POLICIES TO LEVEL THE

PLAYING FIELD

Charles T. Clotfelter

Sanford School of Public Policy

杜克大学

达勒姆, NC 27708

charles.clotfelter@duke.edu

Helen F. Ladd

(corresponding author)

Sanford School of Public Policy

杜克大学

达勒姆, NC 27708

hladd@duke.edu

Jacob L. Vigdor

Sanford School of Public Policy

杜克大学

达勒姆, NC 27708

jacob.vigdor@duke.edu

抽象的
Research has consistently shown that teacher qual-
ity is distributed very unevenly among schools, 到
clear disadvantage of minority students and those from
low-income families. Using North Carolina data on
the length of time individual teachers remain in their
学校, we examine the potential for using salary dif-
ferentials to overcome this pattern. We conclude that
salary differentials are a far less effective tool for retain-
ing teachers with strong preservice qualiﬁcations than
for retaining other teachers in schools with high propor-
tions of minority students. Consequently large salary
differences would be needed to level the playing ﬁeld
when schools are segregated. This conclusion reﬂects
our ﬁnding that teachers with stronger qualiﬁcations are
both more responsive to the racial and socioeconomic
mix of a school’s students and less responsive to salary
than are their less-qualiﬁed counterparts when making
decisions about remaining in their current school, mov-
ing to another school or district, or leaving the teaching
职业.

C(西德:2) 2011 Association for Education Finance and Policy

399

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
/

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

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

TEACHER MOBILITY, SEGREGATION, AND PAY

介绍

1.
Public schools that are segregated by the race or socioeconomic status of their
students raise many educational and societal concerns. Of central interest
for this article is that such segregation is typically associated with an uneven
distribution of resources, the most important of which is teacher quality (作为
measured here by teacher qualiﬁcations) across schools. Schools with large
proportions of nonwhite or low-income students tend to have teachers with
far weaker qualiﬁcations than those in schools serving whiter or more afﬂuent
学生 (Betts, Rueben, and Danenberg 2000; Lankford, Loeb, and Wyckoff
2002; Clotfelter, Ladd, and Vigdor 2006, 2007). This well-documented pattern
largely reﬂects the operation of a teacher labor market in which the distribution
of teachers across schools is determined not only by state or district policies
but also by the preferences of teachers.

For the purposes of this study, we deﬁne an equitable distribution of
teachers as one in which students of different racial and economic groups have
equal access to teachers with strong qualiﬁcations.1 This input-based deﬁnition
of equity is far less demanding than an outcome-based equity measure, 哪个
might well require that disadvantaged groups, often challenging to teach, 有
access to teachers with even stronger qualiﬁcations than those available to
other students (see Ladd 2008 and also Roemer 1998 in a broader policy
语境). 尽管如此, given the current uneven distribution of teachers in the
美国, attaining a level playing ﬁeld even in our more limited sense
still represents a challenging equity goal.

One way to assure an equitable distribution of teachers would be for a state
or district to require that students of different racial and economic groups be
evenly distributed across schools. 在这种情况下, members of each student group
would automatically have equal access to teachers with strong qualiﬁcations at
the school level.2 Although the school desegregation plans that were introduced
starting in the late 1960s pushed many districts in that direction with respect
to race, the Supreme Court’s 2007 decision in Parents Involved in Community
Schools v. Seattle School District No. 1, 551 我们. 701, has ruled out the explicit

Throughout, we use the expressions “teachers with strong qualiﬁcations” or “strong teachers” rather
than the more felicitous phrase “highly qualiﬁed teacher” so as not to confound our concept with
that embedded in the federal No Child Left Behind legislation, which requires that all teachers be
highly qualiﬁed but which in practice allows states to water down the requirement for many of
their established teachers. Ideally one might prefer a more encompassing but difﬁcult to observe
construct of teacher quality. For reasons we discuss further below, we take the more pragmatic
approach of focusing on teacher qualiﬁcations.

2. This even distribution of students across schools, 然而, would still not assure equal access
to quality teachers across student groups within schools at the classroom level to the extent that
classrooms are segregated. Clotfelter, Ladd, and Vigdor (2003, 2008) document for North Carolina
that at the elementary level, most of the racial segregation is between, not within, 学校. 在
high school level, within-school segregation plays a far larger role.

400

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
/

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

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

Charles T. Clotfelter, Helen F. Ladd, and Jacob L. Vigdor

use of the race of a student in assigning students to schools. The use of
socioeconomic rather than racial measures for the purposes of school balance
would address this legal hurdle. It would not, 然而, address the political
challenge posed by middle-class parents who tend to be highly protective
of their middle-class schools and strongly resistant to efforts to move their
children to schools with many minority or low-income students.3

In light of these obstacles, an alternative approach for promoting an equi-
table distribution of teachers would be for states or districts to pursue strate-
gies designed to counter what many teachers may view as the difﬁcult working
conditions associated with large proportions of educationally disadvantaged
学生. Policy makers could, 例如, invest in other components of
the working conditions in those schools, such as school safety or the quality
of school leadership (Milanowski et al. 2009; Ladd forthcoming). 其他
战略, and the one of central interest for this study, is for policy makers to
use salary differentials to help schools serving disadvantaged students attract
and retain teachers with strong qualiﬁcations. The success of such a strategy
depends largely on how responsive teachers are to salary differentials on the
one hand and to school demographic characteristics on the other as they make
their job market decisions.

We use longitudinal data for teachers in North Carolina to examine
these responses, with particular attention to the differential responses of
teachers with strong qualiﬁcations compared to those with average qualiﬁ-
阳离子. Speciﬁcally, we estimate probit models to examine how salary af-
fects the ability of schools to ﬁll vacancies with strong teachers, 和我们
estimate competing risk hazard models to determine how both salary and
school demographics affect teachers’ decisions to leave their current schools.
The results permit us to estimate the magnitudes of the salary differentials
that would be required to offset speciﬁed differences in segregation across
学校.

As discussed below, this research follows in a long tradition of studies that
examine the determinants of teacher mobility. The research reported here is
enriched by the following components. First is our explicit attention to whether
teachers with strong qualiﬁcations respond more or less strongly than other
teachers to salary incentives and to the demographic characteristics of schools.
Second is our detailed modeling of salary differentials that take account of the
local nature of many teacher labor markets. Third is our examination of teacher
movement at all three levels of schooling. Fourth is our ability to supplement

3. A clear example emerges in Wake County, North Carolina, which until recently served as a model
for socioeconomic balancing of its schools. The recent election of a new school board majority
openly opposed to busing to achieve socioeconomic balance now threatens that system.

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
/

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

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

401

TEACHER MOBILITY, SEGREGATION, AND PAY

our basic analysis of responses to salary differentials with a difference-in-
differences analysis of two speciﬁc policy interventions that include ﬁnancial
激励措施. One of these is a program used in two of the state’s districts
designed to attract or retain teachers in hard-to-staff schools, and the other is
a short-lived statewide bonus program for selected teachers at low-performing
middle and high schools.

2. THE TEACHER LABOR MARKET
Like those for many occupations, the labor market for teachers in this country
operates largely as a set of loosely connected regional or local labor markets.
The market for teachers differs from the typical labor market, 然而, 在一个
number of ways, the most important of which relates to salaries. In the typical
labor market, competitive pressures would force employers offering jobs with
poor working conditions to pay higher salaries than other employers to attract
equally qualiﬁed workers. Public schools, 相比之下, are typically bound by
contracts within districts that stipulate speciﬁc salary levels for teachers with a
given set of credentials. To the extent that teachers with strong qualiﬁcations
prefer to teach in schools with higher-achieving, more afﬂuent, and whiter
student bodies, schools with more disadvantaged students ﬁnd it difﬁcult to
attract and retain those teachers. Although not all teachers have such prefer-
恩塞斯, studies dating back to Becker (1952) indicate that many do.

The gravitation of teachers with strong qualiﬁcations away from schools
serving disadvantaged students to those serving more advantaged students
generates potentially large inequities across schools. This sorting process con-
sists of three identiﬁable processes: attrition (teachers leaving the profession),
movement (teachers changing schools, either within or across districts), 并重新-
placement (schools ﬁlling vacancies). Much has been written about the ﬁrst two
of these, and we build on that literature by estimating hazard models of teacher
departures from their current schools. The third process, in which schools ﬁll
vacancies, has received comparatively little attention, but it deserves to be exam-
ined because schools are not equally successful in attracting desirable teachers.
Examining these three processes elucidates not only how teacher labor mar-
kets lead to inequities in the matching of teachers to students in the public
schools but also the potential for salary policies to counteract these processes.
Our study builds on a substantial empirical literature that can be summa-

rized by two major conclusions.

1. 教师, like most other people, respond to ﬁnancial incentives in deciding where
去工作. When working conditions are controlled for, the evidence shows
that teachers are attracted to positions with higher salaries and are more

402

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
/

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

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

Charles T. Clotfelter, Helen F. Ladd, and Jacob L. Vigdor

inclined to leave their current post or to leave teaching altogether when
alternative salaries are higher. Thus higher teacher salaries in their current
positions tend to reduce attrition rates, and attrition is sensitive to wage
differences between teacher and nonteacher salaries.4 Teachers with the
best prospects outside teaching are generally most likely to leave teaching.
尤其, higher exit rates from the profession typically emerge for
teachers with high scores on achievement tests and for math and science
teachers than for other types of teachers who presumably have fewer good
alternatives.5

2. Teachers care about certain nonwage aspects of their jobs. As we have already
著名的, social science research going back at least ﬁfty years suggests that
教师, by and large, prefer to work in schools with students who are high
achieving, afﬂuent, and white.6 In studies of teacher attrition, these pref-
erences reveal themselves directly through the estimated effect of certain
school characteristics, and they also show up in comparisons of the origin
and destination schools between which teachers transfer. Racial composi-
tion is the school characteristic most consistently associated with teacher
mobility: teachers more often leave schools with higher concentrations of
nonwhite students, with a greater response among white than nonwhite
teachers.7 Particularly compelling evidence of teacher sorting by the racial

看, 例如, Murnane and Olsen 1989, 1990; Mont and Rees 1996; Podgursky, Monroe, 和
沃森 2004; Krieg 2006; and Reed, Rueben, and Barbour 2006. For the effect of nonteacher
salaries, see Baugh and Stone 1982; Rickman and Parker 1990; and Dolton and van der Klaauw
1995. Gritz and Theobald (1996) and Imazeki (2005) incorporate teacher salaries in both the current
and alternative districts, as well as nonteacher salaries. The former study ﬁnds signiﬁcant effects in
most speciﬁcations for nonteacher salaries. The latter includes both current and expected teacher
salaries in a teacher’s own district and neighboring ones, as well as average nonteaching salaries in
the local area. It ﬁnds signiﬁcant wage effects for current and expected teacher salaries and relative
teacher salaries for women, but no effects for men or women associated with nonteaching salaries.
See Murnane and Olsen 1990; Lankford, Loeb, and Wyckoff 2002; and Podgursky, Monroe, 和
沃森 2004. Stinebrickner (1998) ﬁnds that teachers with bachelor’s degrees in science were more
likely to quit, and Imazeki (2005) observes higher transfer rates among women teaching math and
special education. 相比之下, Krieg (2006) ﬁnds that highly effective female teachers, 测量的
by a long history of raising test scores, were less likely to quit.

6. Hollingshead (1949, p. 171) 报道, “Because the academic teachers believe that college preparatory
students have more ability, are more interested, and do better work than those in the general course,
they prefer to teach the former group.” See also Becker 1952.

7. Greenberg and McCall (1974, 桌子 3) show that teachers who changed schools within the San
Diego school district generally ended up in a school with a smaller proportion of minority students.
Lankford, Loeb, and Wyckoff (2002, 桌子 10) for New York and Hanushek, Kain, and Rivkin (2004,
桌子 4) for Texas show that this was also true for moves between districts. 同样地, Hanushek,
Kain, and Rivkin (2004) and Falch and Strom (2005) ﬁnd that higher concentrations of minority
students were associated with higher rates of attrition. Scaﬁdi, Sjoquist, and Stinebrickner (2007)
ﬁnd both transfers and exits to be higher from predominantly black schools in Georgia for nonblack
教师. In her study of Wisconsin teachers, Imazeki (2005) ﬁnds this aversion only for exits from
教学, and then only by white male teachers. Boyd et al. (2005) ﬁnd that white and Hispanic
teachers were more likely to quit or transfer from New York City elementary schools with lower
proportions of white students.

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
/

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

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

403

TEACHER MOBILITY, SEGREGATION, AND PAY

mix of a school’s students emerges from a recent study of how teachers re-
sponded to the resegregation of schools associated with the end of student
busing in Charlotte, North Carolina (Jackson 2009).
Evidence also supports the view that teachers prefer to teach high-achieving
students.8 Feng and Sass (2008) ﬁnd that effective teachers, 测量的
by a history of raising students’ test scores, are more likely to leave schools
where other teachers are generally less effective. Less strong is the evidence
for reluctance to teach low-income students. Although comparisons of ori-
gin and destination schools show that teachers tend to move to schools with
students from more afﬂuent families, the independent effect of income is
often not conﬁrmed statistically in equations that also include the racial
mix of a school’s students.9

An obvious policy response to these two stylized facts would be to use
salary differentials to offset the effect of racial composition, or other student
特征, on the perceived desirability of schools as workplaces. 康姆-
pensation policy, 然而, may or may not be a practical tool for equalizing
teacher quality across schools. 一方面, teachers may differ in their sensi-
tivity to working conditions and salary. To the extent that teachers with strong
qualiﬁcations are less responsive to salary and more responsive to student de-
mographics than other teachers, 例如, salary differentials might reduce
the overall rate of turnover at disadvantaged schools but do little to equalize
the proportion of high-quality teachers across schools. For another, salary dif-
ferentials may be more powerful tools for reducing exits from the teaching
profession than for altering the distribution of teachers across workplaces.
Hence the power of salary differentials to promote an equitable distribution of
teachers is an empirical question.

In comparisons of origin and destination schools, Greenberg and McCall (1974, 桌子 3) 和
Hanushek, Kain, and Rivkin (2004) show that teachers moved from less to more able student
身体, as measured by average standardized test scores. Evidence of this preference also appears
in two multivariate studies of attrition—Mont and Rees 1996 和, for female teachers only, Krieg
2006. Clotfelter et al. (2004) ﬁnd that the rate of exit from low-performing schools increased with
the advent of North Carolina’s assessment program, one that exposed teachers in low-rated schools
to fewer rewards and the prospect of punitive policies. Using data for New York City, Boyd et al.
(2008) ﬁnd that teachers in low-performing schools were more likely to leave than those in other
学校.

9. See Greenberg and McCall 1974 (桌子 3) and Hanushek, Kain, and Rivkin 2004 (桌子 4). In partic-
他们是, hazard models estimated by Imazeki (2005) and Krieg (2006) obtain statistically insigniﬁcant
coefﬁcients for percent of students receiving free lunch, and Reed, Rueben, and Barbour (2006)
obtain a negative coefﬁcient. Note that these insigniﬁcant coefﬁcients are not inconsistent with the
observation that teachers tend to move from poorer to richer schools because of the typically high
correlation between poverty and minority enrollment. There is also evidence that teachers are more
likely to quit when their classes are large (Mont and Rees 1996) or when they do not feel successful
or supported by school administrators (Johnson and Birkeland 2003).

404

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
/

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

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

Charles T. Clotfelter, Helen F. Ladd, and Jacob L. Vigdor

3. MAIN CONSTRUCTS AND BASIC EMPIRICAL PATTERNS
We use North Carolina data to examine teacher mobility over the period 1995–
2004, with attention to all three levels of schooling. Before describing the
general patterns in the data, we discuss three constructs that are central to
our empirical analysis: disadvantaged students, qualiﬁcations of teachers, 和
teacher pay, with attention to what they mean in North Carolina, a state with
多于 9 million people and diverse geographic areas that range from the
coast through the Piedmont in the center to mountains in the west.10

Student Disadvantage
Throughout the analysis, we use as our two proxies of student disadvantage
at the school level the percent of students who are nonwhite and the percent
who are eligible for the federal free lunch program. Although not all nonwhite
students are educationally disadvantaged, historical patterns of discrimination,
limited family wealth or income, and the fact that some are recent immigrants
put many in that category. Most of the nonwhite students in North Carolina
are African American, although some are American Indian, and an increasing
number are Hispanic. Eligibility for free lunch, which is limited to families
with income below 130 percent of the poverty level, serves as a common,
although admittedly imperfect, measure of family income.11

We focus on these racial and economic proxies for educational disadvan-
tage largely because data for them are available at the school level. The reader
should bear in mind, 然而, that they are at best proxy measures for a
broader set of measures of student disadvantage that could well inﬂuence
teachers’ perceptions about a school’s working conditions. 基于数据
这 2000 人口普查, 桌子 1 reports correlations between various family or student
characteristics across North Carolina’s one hundred counties (whose bound-
aries are coterminous with school districts in most cases). The poverty rate for
children is positively correlated with the nonwhite percent of the population,
but the correlation is only 0.67 because the counties in the mountain region
tend to be poor but white, while rural coastal and Piedmont areas are poor
but nonwhite. Most striking are the high correlations between poverty and the
nonwhite share on the one hand and the fraction of single parent households

10. See the appendix for detailed explanations of how we constructed the variables.
11. Because low-income high school students tend to be less willing to participate in the subsidized
lunch program than their younger counterparts, the percentages are not directly comparable across
levels of schooling (Gleason 1995). For that reason, whenever we use a single free lunch measure for
all schools, we normalize it based on the means and standard deviations for each level of schooling.
因此, A 1 standard deviation difference at the elementary level represents a somewhat larger
difference in the actual percentage of students on free lunch than at either the middle or the high
school levels. 为了 2004, the means (standard deviations) in percents across schools were: elementary
45.7 (21.5); 中间 41.1 (18.1); high school 29.1 (29.8).

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
/

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

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

405

TEACHER MOBILITY, SEGREGATION, AND PAY

桌子 1. Correlations between Measures of Population Characteristics (百分比), North Carolina Counties

Nonwhite
(包括
Hispanic)

Poor
(Percent of
Children in
Poor Households)

低的
教育
(Less Than
中学)

高的
教育
(More Than
大学)

Single Parent
with Children
under Age 18

Nonwhite

Poor

0.67

0.28

0.55

−0.16

−0.47

0.89

0.64

来源: Calculated by authors based on data from the 2000 我们. 人口普查.

在另一. With only one parent at home, and one that may well be working,
children in such families are likely to receive less educational support at home
than are those in more advantaged families. Thus an apparent aversion of
teachers to either the racial or economic (或两者) characteristics of students
could in fact represent an aversion to other characteristics of the students that
are correlated with the ones we measure.12

Qualiﬁcations of Teachers
We use four measurable qualiﬁcations of teachers as proxies for teacher quality.
The ﬁrst two are preservice qualiﬁcations: teachers’ average licensure test
scores and whether they graduated from a very competitive undergraduate
机构. The other two are their years of teaching experience and whether
they are certiﬁed by the National Board. All four have been shown to be
predictive of student achievement in North Carolina and elsewhere.

Studies conﬁrm, 例如, that teachers’ own ability or achievement,
as measured by some form of test score, whether an SAT, ACT, or teacher
licensure score, is predictive of student achievement. 的确, teachers’ test
scores are the credential that most consistently emerges as predictive of student
achievement across studies of various types (see summary in Goldhaber 2008).
The research is somewhat less clear about the predictive power of the quality of
a teacher’s undergraduate institution, as typically measured by Barron’s college
ratings. Our own research using North Carolina data conﬁrms its predictive
power at the high school level but not at the elementary level (Clotfelter, Ladd,
and Vigdor 2006, 2007, 2010).

12. Because disadvantaged students tend to achieve at below-average levels, it could also represent an
aversion to teaching low-performing students. We have chosen not to include student achievement
as a separate measure of disadvantage because it is partly endogenously determined by the quality
of teachers in the schools. 此外, the racial and economic mix of students in a school may also
be correlated with other working conditions valued by teachers. Research has shown, 然而, 那
the racial characteristics of schools still emerge as signiﬁcant predictors of teacher movement, 甚至
in models that control for a variety of other working conditions, such as the quality of leadership,
school safety, and resources as perceived by teachers (Ladd forthcoming).

406

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
/

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

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

Charles T. Clotfelter, Helen F. Ladd, and Jacob L. Vigdor

With respect to years of experience, studies from a number of states con-
sistently show that, regardless of how effective they may eventually become,
teachers with no or very limited experience are far less effective at raising
student achievement than teachers with more experience (see summary in
Goldhaber 2008). Although the studies differ on the patterns beyond the ﬁrst
few years of experience, it seems safe to conclude that on average, 教师
with three or more years of experience are more effective than those with less
经验. 此外, most careful studies, including several based on North
Carolina data, show that National Board–certiﬁed teachers are more effective
at raising student achievement than are those who are not certiﬁed (Goldhaber
and Anthony 2007; Clotfelter, Ladd, and Vigdor 2006, 2007, 2010). Only at
the high school level, 然而, does it appear that the process of certiﬁcation
itself makes teachers more effective (Clotfelter, Ladd, and Vigdor 2010).

How we use these measures will become clear in the analysis below. Sufﬁce
it to say at this point that for some of the analysis we treat the preservice
qualiﬁcations differently than those related to subsequent employment so as
not to confound decisions made by teachers after they enter the profession
with more exogenous measures of teacher qualiﬁcations. Not included in this
list of qualiﬁcations is whether a teacher has a master’s or other advanced
程度, because such credentials do not generally emerge as predictive of
student achievement.13

One might legitimately ask why we use teacher qualiﬁcations as proxies for
teacher effectiveness, 而不是 (as is the case in some other recent studies) A
more direct value-added measure based on gains in their students’ test scores
(例如, Hanushek et al. 2005; Goldhaber, 总的, and Player 2009). The question
is valid because variation in these four teacher qualiﬁcations explains at best
only a portion of the total variation in teacher quality as measured by gains in
student test scores (Goldhaber 2008). One answer is that the instability of the
value-added measures for individual teachers from year to year raises questions
about their reliability (Koedel and Betts 2007; Lockwood, McCaffrey, and Sass
2008). Another is that such measures can be estimated only for teachers
in grades or courses that are tested annually, which in the North Carolina
context would typically restrict the analysis to elementary schoolteachers of
math and reading. 最后, they are not well suited for teachers with little or

13. Our own research from North Carolina shows, 例如, that elementary schoolteachers who
obtain a master’s degree between one and ﬁve years into teaching are less effective on average than
other teachers (Clotfelter, Ladd, and Vigdor 2007). Presumably this pattern says more about who
decides to get a master’s degree once they start teaching than about the value of the degree itself.
As might be expected, at the high school level where subject knowledge matters more, 硕士
degrees are somewhat more positively predictive of student achievement than at the elementary
等级, but even here the effects are very small (Clotfelter, Ladd, and Vigdor 2010).

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
/

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

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

407

TEACHER MOBILITY, SEGREGATION, AND PAY

no experience. Given our interest in the movements of all teachers who start
teaching stints within a nine-year period at all levels of schooling, 我们有
chosen to rely on teacher qualiﬁcations that we can observe for most of them.

Teacher Pay
In contrast to many other states, North Carolina is quite centralized, 和
teacher associations have no collective bargaining powers. The state govern-
ment provides more than 60 percent of the operating funding for the state’s
学校, and there is a statewide salary schedule for teachers. Variation across
districts comes from the fact that districts can, and typically do, supplement
teacher salaries with local tax revenues. Using information on the total amount
of supplements in each district and, whenever available, more detailed infor-
mation on how they were distributed among teachers, we constructed salaries
for different types of teachers in each district for each year of our analysis.14
In 2004–5, the average local supplement across the state was about $2,500, with supplements in the two biggest districts of Charlotte and Wake close to $5,000 and the average in the rural areas of the state about $1,500.15 In our models of teacher movement, the salary for a particular teacher, speciﬁed in logarithmic form, represents her salary in her current district. As we highlight below in the context of the results, our use of observational data for salaries generates downward-biased estimates of their effects on teacher mobility.16 Although some districts, particularly the state’s large and fast-growing ur- ban districts, operate in a state and national labor market for teachers, the relevant labor market for many districts is quite local. 尤其, when existing teachers are making their decisions about remaining in or leaving their current school, the most relevant alternative salaries are those for jobs, both teaching and nonteaching, within commuting distance. We measure the alternative teaching salary available to a teacher in each district as a weighted average of the salaries of teachers with similar characteristics in the districts within a thirty-mile radius, with the weights being student enrollment in each l D o w n o a d e d f r o m h t t p : / / 直接的 . 米特 . / F / e d u e d p a r t i c e – p d l f / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 df . 来宾来访 0 7 九月 2 0 2 3 14. In general information is available only on the total supplemental payments and the number of recipients. For some districts, more detail is available on the Internet about how the supplements are distributed among teachers. 在其他情况下, we had to make reasonable assumptions about its distribution. Details are provided in the appendix. 15. The evidence suggests that some of this variation is attributable to differences in the cost of living and to salary supplements that are higher in districts with higher proportions of novice teachers, presumably used to recruit more teachers. This statement is based on Walden and Sogutlu 2001 and on our own unpublished estimates for a more recent year. We control for some of these compensating differentials in our various models by including regional ﬁxed effects. 16. Compared with many other states, the cross-district variation in teacher salaries is small. 在 2004, 例如, the standard deviation in salaries for teachers with a bachelor of arts degree and two years of experience across the teachers in our sample was about $800, or about 3 的百分比
average salary of $27,000. 408 Charles T. Clotfelter, Helen F. Ladd, and Jacob L. Vigdor district. We measure nonteaching salaries as the employment-weighted aver- age of salaries in all counties within a thirty-mile radius of the school district of the teacher in question. We also include a measure of the unemployment rate within the same area to capture information about the availability of jobs. 最后, we incorporate indicator variables into our models to represent salary-related incentive programs speciﬁcally designed to attract and retain teachers in hard-to-staff schools. These programs include the Equity Plus pro- grams used in two of the state’s largest districts (Charlotte-Mecklenburg and Winston-Salem) as well as a statewide bonus program that operated between 2002 和 2004. The Equity Plus programs give certain schools a variety of beneﬁts, including additional pay for some or all of their teachers.17 The statewide bonus program, which we have described and analyzed in more de- tail elsewhere (Clotfelter, Ladd, and Vigdor 2008), 假如 $1,800 bonuses
for certiﬁed math, 科学, and special education teachers to teach in disadvan-
taged middle schools and low-performing high schools throughout the state.

Basic Patterns
The basic empirical patterns in North Carolina are consistent with the view
that teachers are inequitably distributed across schools. 数字 1 文件,
例如, the relationship between one measure of teacher quality—the
fraction of teachers whose certiﬁcation test scores fall in the top quartile of
the test score distribution—and each of the two measures of disadvantage.
Although the top graph is based on all three levels of schooling, the bottom one
refers to elementary schools alone because of the limitations of that measure
for older students. Schools that have higher proportions of nonwhite students
or higher proportions of students receiving free lunches tend to have smaller
proportions of teachers with strong qualiﬁcations by this measure, a pattern
that we also document in prior research (Clotfelter, Ladd, and Vigdor 2005;
Clotfelter et al. 2007).

Also consistent with the literature that attributes some of this inequity to
differential attrition rates by school are the patterns in ﬁgure 2. The ﬁgure
depicts, by the percentage of nonwhite students in the school, the propor-
tions of teachers who had started a job in the 1994–95 school year and had
left the school by 2002–3. The two lines distinguish between teachers with

17. The program in Charlotte-Mecklenburg began in 1997–98 and has undergone several name changes
since then. The program gives signing bonuses to teachers going to a designated school, 和
experienced “master teachers” can receive up to $2,500 for teaching in such a school. The program in Winston-Salem/Forsyth began in 1999–2000. Equity Plus schools in that district are determined by the percentage of students receiving free and reduced price lunches. All teachers in the designated schools receive bonus pay equal to 20 percent of the local supplement. This bonus increases with degree and experience, typically ranges from $500 到 $1,500 per teacher, and is paid annually. l 从http下载 : / / 直接的 . 米特 . / F / e d u e d p a r t i c e – p d l f / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 df . 来宾来访 0 7 九月 2 0 2 3 409 TEACHER MOBILITY, SEGREGATION, AND PAY l D o w n o a d e d f r o m h t t p : / / 直接的 . 米特 . F / / e d u e d p a r t i c e – p d l f / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 df . 来宾来访 0 7 九月 2 0 2 3 数字 1. Share of Teachers in Top Quartile of Test Scores by Percentages of Disadvantaged Students at the School Level, 1994–95 high test scores and those with average test scores.18 For schools serving pre- dominantly white populations (low percentages of nonwhite students), the nine-year attrition rates for teachers hover around 60 percent and appear to be nearly independent of a teacher’s licensure test score. In schools serving 18. To enhance the readability of the ﬁgure, we have collapsed schools into bins of width 0.01 (based on the school percent nonwhite). Thus the points represent groups of schools with very similar percentages of nonwhite students, not individual schools. 410 Charles T. Clotfelter, Helen F. Ladd, and Jacob L. Vigdor Figure 2. Teacher Attrition Rates and Percent Nonwhite Students by All Schools for Top Test Quartile Teachers and Other Teachers, 1994–95 to 2002–3. (Schools are grouped into bins of width 0.01 in terms of the percent nonwhite students in the school.) overwhelmingly nonwhite student populations, attrition rates are higher over- all and show a more distinct discrepancy between teachers with test scores in the highest quartile and all others. At schools where the nonwhite proportion of students exceeds 80 百分, the attrition rates over the period for top quar- tile teachers are usually above 80 百分, while those for teachers with lower certiﬁcation test scores generally fall in the 60–80 percent range. 4. THE REPLACEMENT PROCESS: FILLING VACANCIES Starting from an uneven distribution of teachers across schools, the quickest way to level the playing ﬁeld would be for the schools serving disadvantaged students to ﬁll vacancies with applicants having strong qualiﬁcations. The question of interest here is the extent to which salary might be used as a draw for such teachers. To that end, we report in table 2 four probit models of the propensity that a school will ﬁll an open position with an applicant having each of the four qualiﬁcations predictive of higher achievement discussed above: high test scores (in the top quartile), a degree from a very competitive college, more than two years of teaching experience, and National Board certiﬁcation. We note that teachers with the ﬁrst two types of qualiﬁcations include both those with and without experience. 相比之下, those in the other two categories include no novice teachers, with virtually all the National Board−certiﬁed teachers (NBCTs) having several years of experience, as is consistent with the l D o w n o a d e d f r o m h t t p : / / 直接的 . 米特 . F / / e d u e d p a r t i c e – p d l f / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 压力 . f f b y g u e s t t o n 0 7 九月 2 0 2 3 411 TEACHER MOBILITY, SEGREGATION, AND PAY Table 2. Filling Vacancies, Probit Models, 1995–96 to 2003–4 (1) High Test Score (2) Very Competitive Undergraduate (3) Experienced (4) National Board Certiﬁcation School Characteristics Middle school High school Free lunch, elementary (%) Free lunch, 中间 (%) Free lunch, high school (%) Nonwhite, K–12 (%) Building age District Characteristics Free lunch (%) Nonwhite (%) Salary: Bachelor’s degree plus 2 (ln) Alt. salary (ln) Unemployment rate 0.0270 (0.0222) 0.0492∗ (0.0198) −0.0005 (0.0004) −0.0016∗∗ (0.0006) −0.0006 (0.0006) −0.0021∗∗∗ (0.0004) −0.0001 (0.0002) 0.0012 (0.0008) −0.0040∗∗∗ (0.0007) 2.3013∗∗∗ (0.3125) 0.6725∗∗ (0.2254) −0.0208∗∗∗ (0.0045) Equity Plus (EP) Incentive Programs CM EP school −0.0344 (0.0360) CM ever EP WS EP school WS ever EP −0.0450 (0.0304) −0.0527 (0.0766) −0.0369 (0.0484) 0.0859∗∗∗ (0.0220) 0.2265∗∗∗ (0.0195) −0.0005 (0.0004) −0.0010 (0.0006) −0.0025∗∗∗ (0.0006) −0.0015∗∗∗ (0.0004) −0.0006∗ (0.0002) 0.0025∗∗ (0.0008) 0.0001 (0.0007) 4.6181∗∗∗ (0.3052) 2.2526∗∗∗ (0.2311) −0.0325∗∗∗ (0.0047) 0.0040 (0.0348) −0.1118∗∗∗ (0.0295) −0.1231 (0.0753) −0.0475 (0.0472) −0.0252 (0.0188) −0.0007 (0.0165) 0.0000 (0.0003) −0.0013∗∗ (0.0005) −0.0002 (0.0005) −0.0019∗∗∗ (0.0003) −0.0011∗∗∗ (0.0002) 0.0004 (0.0006) −0.0014∗∗ (0.0005) 1.7777∗∗∗ (0.2571) 0.2701 (0.1869) 0.0043 (0.0037) −0.0032 (0.0311) −0.1058∗∗∗ (0.0258) −0.0788 (0.0626) −0.0946∗ (0.0413) −0.1020 (0.0553) −0.1053∗ (0.0467) −0.0017 (0.0010) −0.0016 (0.0015) −0.0053∗∗ (0.0019) −0.0028∗∗ (0.0009) −0.0024∗∗∗ (0.0006) 0.0047∗∗ (0.0016) −0.0055∗∗ (0.0017) 2.7870∗∗∗ (0.7485) 0.6497 (0.5531) 0.0099 (0.0116) 0.0526 (0.0913) 0.0757 (0.0774) −0.4276 (0.2408) −0.0886 (0.1649) 氮 129,388 129,388 129,388 129,388 Notes: Controls: Log enroll (dist), rural, coastal, mountain, beach, 年, indicator for no district within in 30 miles. Union schools in 2003 were excluded due to missing data. Standard errors are in parentheses. CM = Charlotte-Mecklenburg; WS = Winston-Salem. ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001. 412 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d . f f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Charles T. Clotfelter, Helen F. Ladd, and Jacob L. Vigdor requirements of that qualiﬁcation. All the models were estimated based on data from 1995–96 to 2003–4 and include the variables listed in table 2 as well as others listed in the table footnote. (See appendix table A.1 for means and standard deviations.) The unit of observation is a school vacancy that is ﬁlled with a replacement teacher by year.19 Because probit coefﬁcients are not straightforward to interpret, we begin by focusing on their signs and statistical signiﬁcance, and leave to a subsequent table the policy implications of the estimated magnitudes. Emerging most clearly and consistently across all ﬁve equations are the statistically signiﬁcant negative coefﬁcients on the nonwhite percentage of the school’s students, the relatively consistent negative coefﬁcients on the nonwhite percentage in the district, and the statistically signiﬁcant positive coefﬁcients on district salary (expressed in logarithmic form).20 These patterns imply, as expected, that the higher the nonwhite share of students in the school or the district, the harder it is for the school to ﬁll a vacancy with a teacher having any of the speciﬁed qualiﬁcations. However, these patterns also imply that higher salaries make it easier to do so. Results for many of the other variables are more mixed across the equa- tions. Compared with elementary schools, for example, middle schools and high schools ﬁnd it easier to ﬁll vacancies with teachers having degrees from a very competitive college and possibly with high test scores. Consistent with the racial patterns for those two levels of schooling, the coefﬁcients of the proportion of students eligible for free lunch are negative in all cases but are not always statistically signiﬁcant. We also included a variable denoting the age of the building with the expectation that older schools might ﬁnd it more difﬁcult to attract teachers than newer ones with more modern amenities. As expected, the evidence is generally consistent with that expectation. Among the district-level variables, most intriguing are the results for the salaries in nearby districts and the local unemployment rate. The somewhat unexpected positive sign in the ﬁrst two columns for the nearby teacher salaries (alt. salary) suggests that, holding constant the salary in the speciﬁc district, teachers with strong preservice qualiﬁcations are more attracted to schools in districts located near high-paying districts than to those in districts in geo- graphic areas with lower salaries. Our interpretation is that for some teachers 19. All the models presented here pool the three levels of schooling. We have also estimated separate models by level of schooling. The patterns do not differ much by level of schooling. Where differences emerge, we highlight them in a later footnote. In addition, we have estimated a model that examines the probability of hiring a teacher with a combination of all the strong qualiﬁcations or three of the four qualiﬁcations, but the probabilities of those combinations are too low in many cases to make the results meaningful. 20. The district salary is for a teacher with a bachelor’s degree and two years of experience. 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d . f f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 413 TEACHER MOBILITY, SEGREGATION, AND PAY the initial district may be the ﬁrst step toward obtaining a job in the higher- paying district. The negative coefﬁcient on the unemployment rate in those same two columns is consistent with that general story: a higher overall unem- ployment rate in the local area makes it more difﬁcult for a school to attract a teacher with either of the strong preservice qualiﬁcations. These patterns differ from those for the chances of hiring the more experienced teachers, as shown in the ﬁnal two columns. Presumably because many teachers with experience or who are board certiﬁed are already in the area, neither the alternative salary nor the unemployment rate has a statistically signiﬁcant effect on the chances that such teachers will be hired.21 To infer the impact of the Equity Plus programs in the Charlotte- Mecklenburg (CM) and Winston-Salem (WS) districts, we introduce two sets of indicator variables. Each set includes a variable designating whether the school was eligible for the program in the current year (EP school) and whether the school ever participated in the program (ever EP), which controls for any per- manent unobserved difference between participating and other schools. Thus the coefﬁcient of the “EP school” variable generates our best estimate of the treatment effect of the program. The results in table 2 show no evidence that participation in either district’s Equity Plus program in a particular year in- creased the probability that schools would recruit more teachers with strong qualiﬁcations. Although the generally negative coefﬁcients on the “ever EP” indicators show, as expected, the greater challenges such schools face in hiring teachers with strong qualiﬁcations, none of the program-speciﬁc coefﬁcients are statistically signiﬁcant. We do not ﬁnd the absence of a recruiting effect surprising, given that many prospective teachers may not expect the bonus program to be sustained over time. Table 3 spells out the salary implications of the key coefﬁcients related to the school and district characteristics from the top two panels of table 2. The entries in each cell are the salary differences required to level the playing ﬁeld between schools with the speciﬁed differences in nonwhite shares of students in the school (panel A) and in the district (panel B). For example, panel A indicates that to neutralize a difference in the nonwhite school percentage of 50 percentage points (e.g., between schools with 25 and 75 percent nonwhite shares) on the probability of hiring a teacher with a high test score, the salary would have to be 4.7 percent higher in the more nonwhite school, controlling for the district-level share. Further, the required salary needed to offset a 50 21. We have not included nonteaching salaries in this equation because the variable was missing for some of our districts in some years. In samples restricted to the districts for which we had complete data, its coefﬁcient is only marginally signiﬁcant in one model, that for hiring a teacher from a selective college, where it is negative. 414 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d f . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Charles T. Clotfelter, Helen F. Ladd, and Jacob L. Vigdor Table 3. Salary Differences Required to Level the Playing Field (Percent) Panel A Percentage Point Difference in Nonwhite Share in School High test score Undergrad very competitive Experienced NBCT teacher Panel B High test score Undergrad very competitive Experienced NBCT teacher 10 0.9 0.3 1.1 1.0 20 1.8 0.7 2.2 2.0 30 2.8 1.0 3.3 3.1 40 3.7 1.3 4.4 4.1 Percentage Point Difference in Nonwhite Share in District 10 1.8 0.0 0.8 2.0 20 3.5 0.0 1.6 4.0 30 5.4 0.1 2.4 6.1 40 7.2 0.1 3.2 8.2 50 4.7 1.6 5.5 5.2 50 9.1 0.1 4.0 10.4 percent difference in both the school and the district percentage of nonwhite students on the probability of hiring a teacher with a high test score would be the sum of the relevant entries in the two panels, or 13.8 percent. On a base salary of, say, $30,000, the required differential would amount to $4,100.22 Four observations are worth making about these entries. First, given that the required salary differentials are estimated from existing salary differentials across districts, they should be interpreted as persistent salary differences rather than as differences in the form of bonuses that apply to either a single year or a short period, as might be the case for the Equity Plus programs. The second is that we are less conﬁdent about the exact numbers for required salary differentials that are large relative to the actual variation in our sample than we are for relatively small implied salary differentials, such as those in the range of 6–8 percent.23 Nonetheless, we can be quite conﬁdent that they are big. The third is that the entries should be interpreted as upper bound estimates of the required salary differentials, because the estimated coefﬁcients of the salary variables most likely underestimate the effect of salary on hiring decisions. That is the case because some of the interdistrict salary differences undoubtedly reﬂect compensating differentials for district characteristics not fully controlled for in the models, including, for example, differences in the 22. Separate estimates by level of school imply that the salary differentials would have to be highest at the high school level and lowest at the elementary level. Note in addition that it would not be correct simply to add the relevant coefﬁcients to determine the required salary differential needed to hire a teacher with more than one of the speciﬁed qualiﬁcations because of the potential for positive correlation among them. 23. Within our sample, a 1 standard deviation in salary in any one year is about 3 percent. 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d f . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 415 TEACHER MOBILITY, SEGREGATION, AND PAY cost of living. If the downward bias were as large as 25 percent, for example, unbiased estimates of the required differentials would be 25 percent smaller than those reported in table 3.24 Third, even with the downward adjustments of this magnitude, the entries suggest that salary differentials are potentially powerful tools for equalizing the ability of schools to ﬁll vacancies with strong teachers. At the same time, the salary differentials required to neutralize the effects of large differences in concentrations of student disadvantage are likely to be far higher than the $1,800 bonus that the state legislature embedded in its short-lived Bonus Program. 5. ATTRITION AND MOVEMENT Once hired to work in a school, a teacher can leave the school by one of three routes: she can leave the teaching profession, transfer to another school within the same district, or move to another district. Because of the nature of our data, we cannot separate those teachers who leave the profession from those who move to another state, into the private sector, or into a charter school. Hence, for this analysis, the option labeled “leaving teaching” in fact means leaving the traditional North Carolina public school system. As in the previous section, our goal is to identify the role of student de- mographics and speciﬁc policy levers, most notably salary, in predicting these three types of teacher moves so that we can assess the viability of compen- satory policies to equalize rates of departure, with particular attention to the moves of teachers with strong qualiﬁcations. Our models include a number of teacher-level, school-level, district-level, and local area-level covariates, as well as indicator variables by year. These covariates capture both the personal predictors of departure, such as being a female of childbearing age, and the local amenities or opportunity costs that might inﬂuence teacher decisions. We model teacher choices using a discrete time, competing risk hazard model, where the hazard rate λi (tj) is the probability that a teaching spell will end at the close of year tj by way of exit mode i, conditional on the teacher not having left his or her school before this period. In order to avoid the compli- cations of dealing with left-censored observations, we restrict our analysis to teachers who began teaching spells during the period under study—that is, be- tween 1994–95 and 2003–4. We separately analyzed two sets of teachers, those who had never taught before the new spell began (initial teaching spell) and those who had previously taught before the spell began (second or later teach- ing spell). We adopt the Cox proportional hazard model, a semi-parametric 24. We use the 25 percent ﬁgure because that is the degree of estimated bias reported in Hanushek, Kain, and Rivkin 2004. 416 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d . f f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Charles T. Clotfelter, Helen F. Ladd, and Jacob L. Vigdor speciﬁcation that is agnostic with respect to the form of the baseline hazard function,25 and estimate a system of equations of the form: λi(tj) = λ0i(tj) exp(X βi + μi), where λi (tj) is the hazard rate applying to exit mode i, λ0i (tj) is the baseline hazard, X is a matrix of teacher, school, district, and region characteristics relevant to movement or attrition for teachers in the sample, μi is an error term, and βi is a vector of coefﬁcients. For 0–1 dichotomous variables, the hazard ratio relevant to exit mode i—the estimated multiplicative impact of a unit change on the conditional probability of a spell ending at the close of year tj, given that the teacher has remained in the original school up to that point— is calculated as exp(βi). Consistent with the proportional hazard model, the impact of these covariates is assumed to be independent of the duration of the teaching spell. Over the period of study, we observed 48,753 teachers in their ﬁrst spell and 27,928 teachers in a later spell, some of whom overlap with the ﬁrst group. For the former teachers, we allow their ﬁrst spells of teaching in the same school to last for up to nine years, which generates a sample of 121,547 teacher-year observations. Based on all our observations of these teachers over the sample period, the probability that a teacher in her ﬁrst spell of teaching would leave the state’s public schools in any given year is 9.4 percent. The corresponding probability of leaving the district for another district in the state is 6.9 percent, and the probability of switching schools within a district is 14.2 percent. For the 27,928 veterans, we follow each new spell that started during the period 1995–96 to 2003–4, which generates a sample of 99,754 teacher- year observations.26 The corresponding departure probabilities for this group are 13.5 percent for leaving the profession, 4.8 percent for leaving the district, and 10.6 percent for leaving the school for another school within the district. Full results for our panel of teachers in their ﬁrst spells are reported in table 4 and in subsequent spells in table 5. The entries are reported as hazard ratios, with ratios greater than one indicating that a factor makes departure via a particular route more likely and below one less likely. In addition to the variables of interest, we have also included indicators for the gender and race of each teacher and a set of indicator variables for the age of the teachers, 25. In particular, the proportional hazard speciﬁcation accommodates either positive or negative du- ration dependence. That is, the period a teacher has been in her current position could be either positively or negatively associated with the probability that she would leave the school in the next period. 26. There is one less year for the veteran teachers because we had to use the 1994–95 information to determine whether the teacher was starting a new spell in the subsequent year. 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d f . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 417 TEACHER MOBILITY, SEGREGATION, AND PAY Table 4. Teacher Departures, by Exit Route, for Initial Teaching Spells (Hazard Ratios) Variable Teacher Characteristics Female Black Other nonwhite Teacher age 25–29 Teacher age 30–34 Teacher age 35–39 Teacher age 40–44 Teacher age 45–49 Teacher age 50 and over Teacher age 25–29 ∗ female (teacher) Teacher age 30–34 ∗ female (teacher) Teacher age 35–39 ∗ female (teacher) Teacher age 40–44 ∗ female (teacher) Teacher age 45–49 ∗ female (teacher) Teacher age 50 and over ∗ female (teacher) Teacher Qualiﬁcations Teacher test score—highest quartile Undergrad college very competitive National Board certiﬁed Has advanced degree Graduated NC college Graduated bordering state college Teacher test score—missing School Characteristics Middle school High school Free lunch eligible (%) (normalized) Nonwhite students (%) (demeaned) Age of school building Building age missing District Characteristics Free lunch eligible (%) (demeaned) Nonwhite students (%) (demeaned) Enrollment (ln) Growth rate Rural Coastal Exit Route All Exit Routes Leave Teaching Switch Districts Change Schools 0.957∗∗ 1.070∗∗∗ 1.008 0.936∗∗∗ 0.947 0.948 0.946 0.898 1.057 1.097∗∗∗ 1.034 0.970 0.902 0.957 0.991 1.025∗ 1.037∗∗∗ 0.872∗∗ 1.052∗∗ 0.841∗∗∗ 0.966 1.376∗∗∗ 1.066∗∗∗ 0.998 1.030∗∗ 1.002∗∗∗ 1.000 0.985 1.000 1.001 0.987 1.460 1.009 0.961∗ 0.884∗∗∗ 1.228∗∗∗ 1.069 0.973 0.952 0.967 0.996 0.792∗∗ 1.013 1.194∗∗∗ 1.106 0.960 0.818∗ 0.990 1.017 1.152∗∗∗ 1.092∗∗∗ 0.522∗∗∗ 1.053 0.599∗∗∗ 0.931∗∗ 2.061∗∗∗ 1.106∗∗∗ 1.224∗∗∗ 1.023 1.002∗∗∗ 1.000 0.989 1.001 1.002∗∗ 0.995 1.844 0.968 0.952 1.061∗ 0.795∗∗∗ 0.720∗∗∗ 0.837∗∗∗ 0.838∗ 0.839 0.784 0.872 1.018 1.041 0.887 0.854 0.651∗∗ 0.793 0.780 0.978 1.043 1.172 1.265∗∗∗ 1.269∗∗∗ 0.993 0.744∗∗∗ 1.205∗∗∗ 1.153∗∗∗ 0.994 1.006∗∗∗ 1.000 0.991 0.999 1.000 0.784∗∗∗ 0.149∗ 1.002 1.084∗∗ 0.968 1.066∗∗ 1.125∗ 0.926 0.991 0.965 0.936 1.126 1.093 1.038 1.060 1.154 1.324∗ 1.056 1.242 0.897∗∗∗ 0.935∗∗∗ 1.089 0.968 1.104∗∗∗ 1.014 0.948∗ 0.955∗∗ 0.620∗∗∗ 1.064∗∗∗ 1.000 1.002∗∗∗ 0.967 0.999 0.998 1.188∗∗∗ 2.066 1.108∗∗∗ 0.889∗∗∗ 418 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d . f f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Charles T. Clotfelter, Helen F. Ladd, and Jacob L. Vigdor Table 4. Continued. Variable Mountain Research university or college in county Beach county Labor Market Characteristics Salary (ln) (demeaned) Teacher salary (weighted avg.) in surrounding districts (ln) Local area nonteaching salary (ln) Area unemployment rate Equity Plus (EP) Incentive Programs† CM EP school CM ever EP WS EP school WS ever EP Interactions with Teacher Test Score in Highest Quartile Salary (ln) (demeaned) Nonwhite students (school) (%) (demeaned) Nonwhite students (district) (%) (demeaned) Free lunch eligible (school) (%) (normalized) Free lunch eligible (district) (%) (demeaned) 1.920∗∗∗ 1.003∗∗ 0.997∗∗ 0.972 1.004∗∗ Interactions with Very Competitive Undergraduate College Salary (ln) (demeaned) Nonwhite students (school) (%) (demeaned) Nonwhite students (district) (%) (demeaned) Free lunch eligible (school) (%) (normalized) Free lunch eligible (district) (%) (demeaned) 1.309 1.001 0.998 0.966 1.003 Exit Route All Exit Routes Leave Teaching Switch Districts Change Schools 0.983 1.017 1.066∗∗ 0.957 1.026 1.180∗∗∗ 0.224∗∗∗ 0.065∗∗∗ 1.371 1.035 1.113∗∗ 0.989∗∗ 1.012 0.982∗∗ 1.035 0.970 0.844∗ 0.964 0.966 1.006 0.828 0.870 7.338∗∗∗ 1.005∗∗∗ 0.998 0.889∗∗∗ 1.005∗ 0.965 0.999 1.002 0.986 0.998 1.023 0.956 0.848∗∗∗ 0.158∗∗∗ 2.869∗∗ 1.836∗∗∗ 1.009 0.648∗∗∗ 1.185 0.963 0.776 0.752 0.999 1.002 1.036 1.001 1.076 1.004 0.989∗∗∗ 0.929 1.012∗∗∗ 0.973 1.069∗∗ 1.120∗∗ 0.860 1.675 0.840∗ 0.986 1.053 1.074 1.081 0.899 0.737 1.002 0.994∗∗ 1.047 1.002 2.267∗∗∗ 1.004∗∗ 0.995∗ 0.976 1.002 Number of teacher-years 121,547 121,547 121,547 121,547 †CM = Charlotte-Mecklenburg; WS = Winston-Salem. ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p< 0.001. both by themselves and interacted with whether the teacher is female. These variables control for different departure propensities by age and also among women of childbearing age. As one might expect, female teachers in their initial teaching spells in the 25–29 age range exhibit higher probabilities of leaving the profession than male teachers and teachers of other ages in any speciﬁc year, given that they have been in the school to that point. Some of them, however, may reappear as veteran teachers starting a new spell in a subsequent year. For both samples, females are less likely than males 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d . f f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 419 TEACHER MOBILITY, SEGREGATION, AND PAY Table 5. Teacher Departures, by Exit Route, for Subsequent Teaching Spells (Hazard Ratios) Variable Teacher Characteristics Female Black Other nonwhite Teacher age 25–29 Teacher age 30–34 Teacher age 35–39 Teacher age 40–44 Teacher age 45–49 Teacher age 50 and over Teacher age 25–29 ∗ female (teacher) Teacher age 30–34 ∗ female (teacher) Teacher age 35–39 ∗ female (teacher) Teacher age 40–44 ∗ female (teacher) Teacher age 45–49 ∗ female (teacher) Teacher age 50 and over ∗ female (teacher) Teacher Qualiﬁcations Teacher test score—highest quartile Undergrad college very competitive National Board certiﬁed Has advanced degree Graduated NC college Graduated bordering state college Teacher test score—missing School Characteristics Middle school High school Free lunch eligible (%) (normalized) Nonwhite students (%) (demeaned) Age of school building Building age missing District Characteristics Free lunch eligible (%) (demeaned) Nonwhite students (%) (demeaned) Enrollment (ln) Growth rate Rural Coastal Exit Route All Exit Routes Leave Teaching Switch Districts Change Schools 0.894∗ 1.085∗∗∗ 1.072∗ 1.031 0.972 0.839∗∗∗ 0.880∗ 0.896∗ 0.999 1.067 1.112 1.161∗∗ 1.044 0.984 0.973 0.994 1.018 0.863∗∗∗ 1.078∗∗∗ 0.874∗∗∗ 0.981 1.060∗∗∗ 1.058∗∗∗ 0.946∗∗∗ 1.043∗∗∗ 1.002∗∗∗ 1.001∗∗ 1.031 1.000 1.001 0.989 2.003 0.773∗∗∗ 1.175∗∗∗ 0.893∗∗ 0.973 0.842∗∗ 0.746∗∗∗ 0.732∗∗∗ 0.800∗∗ 1.084 1.292∗∗∗ 1.366∗∗∗ 1.320∗∗∗ 1.118 0.945 0.928 1.082∗∗∗ 1.089∗∗∗ 0.451∗∗∗ 1.097∗∗∗ 0.737∗∗∗ 1.007 1.228∗∗∗ 1.090∗∗∗ 1.096∗∗∗ 1.002 1.002∗∗ 1.000 0.983 1.002 1.003∗∗ 0.983 14.693∗∗∗ 0.993 0.911∗∗∗ 0.983 0.944 0.883 0.885∗∗∗ 0.819∗∗ 0.835 0.996 0.804 0.929 0.906 0.858 1.170 0.940 0.954 0.936 0.904 0.919 0.829∗∗∗ 0.938∗ 0.925 1.125∗∗∗ 0.942∗ 0.860∗∗∗ 0.802∗∗∗ 1.279∗∗∗ 1.218∗∗∗ 1.099∗∗∗ 1.004∗∗∗ 1.000 1.035 0.996∗ 1.002 0.727∗∗∗ 2.067 1.021 0.968 1.150 1.074∗∗ 1.453∗∗∗ 1.280∗∗ 1.216∗ 1.045 1.144 1.079 0.977 0.751∗∗ 0.883 1.012 0.924 0.976 1.050 0.974 0.945∗∗ 1.315∗∗∗ 1.031 1.048∗∗ 1.005 0.995 0.945∗∗ 0.681∗∗∗ 1.071∗∗∗ 1.001 1.002∗∗∗ 1.095∗∗∗ 1.001 0.997∗∗ 1.189∗∗∗ 0.020∗∗∗ 1.023 0.840∗∗∗ 420 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d . f f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Charles T. Clotfelter, Helen F. Ladd, and Jacob L. Vigdor Table 5. Continued. Variable Mountain Research university or college in county Beach county Labor Market Characteristics Salary (ln) (demeaned) Teacher salary (weighted avg.) in surrounding districts (ln) Local area nonteaching salary (ln) Area unemployment rate Equity Plus (EP) Incentive Programs† CM EP school CM ever EP WS EP school WS ever EP 0.976 1.017 1.080∗∗ 1.171∗ 0.731∗∗∗ 0.971 0.992 0.945 1.080∗ 0.876 0.853∗ Interactions with Teacher Test Score in Highest Quartile Salary (ln) (demeaned) Nonwhite students (school) (%) (demeaned) Nonwhite students (district) (%) (demeaned) Free lunch eligible (school) (%) (normalized) Free lunch eligible (district) (%) (demeaned) 0.975 1.001 1.000 0.982 1.000 Interactions with Very Competitive Undergraduate College Salary (ln) (demeaned) Nonwhite students (school) (%) (demeaned) Nonwhite students (district) (%) (demeaned) Free lunch eligible (school) (%) (normalized) Free lunch eligible (district) (%) (demeaned) 0.933 1.001 0.998 1.001 1.000 Exit Route All Exit Routes Leave Teaching Switch Districts Change Schools 1.001 1.065∗∗ 1.116∗∗ 1.172 0.446∗∗∗ 0.687∗∗∗ 0.980∗∗ 0.905 1.099 1.186 0.793∗ 1.266∗ 1.000 1.002 0.985 0.997 0.857 0.999 1.002 0.995 0.996 1.001 0.983 0.833∗∗ 0.737 1.551∗∗ 1.605∗∗∗ 1.023∗ 0.900 1.068 0.853 0.870 0.734 1.001 1.001 0.899 1.007 1.168 1.002 1.001 1.035 1.002 0.919∗∗ 0.991 1.175∗∗∗ 1.199 1.155 1.124 0.990 1.095 0.963 0.918 0.611∗∗∗ 0.844 1.002 0.998 1.016 1.001 1.087 1.004∗∗ 0.991∗∗∗ 1.004 1.003 Number of teacher-years 99,754 99,754 99,754 99,754 †CM = Charlotte-Mecklenburg; WS = Winston-Salem. ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001. leave their current school, and black teachers are not only more likely to leave their current school but are also more likely than white teachers to leave the profession altogether. Effects of Strong Qualiﬁcations The various measures of teacher qualiﬁcations enter the models in different ways. One measure of quality, having some experience, is included implicitly by the estimation of separate models for initial and subsequent spells. By 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d f . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 421 TEACHER MOBILITY, SEGREGATION, AND PAY deﬁnition, any teacher observed in a second or subsequent spell has at least one year of experience and could well have many more. Consequently, consistent with the evidence cited earlier, such teachers are likely to be more effective than typical teachers in their ﬁrst teaching spells. The two preservice qualiﬁcations variables—having a highest quartile test score and having attended a very competitive undergraduate institution—enter the model both directly as well as interacted with the salary and demographic variables, so we can test for differential responses of teachers with those qualiﬁcations. NBCT status is included in the equations as a control variable but is not in- teracted with other variables because teachers can select into this qualiﬁcation after they enter the teaching profession. Only those teachers who are commit- ted to the profession are likely to undertake the rigorous process to become cer- tiﬁed, given that the certiﬁcation is not transportable to nonteaching jobs. Sim- ilarly, many teachers select into the category of having a master’s degree after they enter the profession with the goal of obtaining a higher salary as a teacher. For that reason, we have included whether a teacher has an advanced degree as an additional control variable that could affect teacher mobility. Consistent with the selection processes just described, table 4 shows that teachers who are board certiﬁed exhibit lower hazards of leaving their current schools than teachers who are not board certiﬁed and that the lower hazard is attributable primarily to their greatly reduced hazard of leaving the profession. In contrast, teachers with advanced degrees emerge as more willing than comparable teachers without an advanced degree to leave their current school, but they do so primarily by changing districts. Thus their advanced degrees increase their mobility within, but not necessarily outside, the teaching profession. The models also include as control variables whether the teacher graduated from a North Carolina college or a college from a bordering state. Graduation from a North Carolina college reduces the exit hazard for a teacher both for all types of exit and for leaving the profession, which, recall, could refer to leaving the state. Table 6 reports the hazard ratios of the two preservice teacher qualiﬁca- tions separately for the two samples of teaching spells. In each case the ratios are relative to a “regular” teacher, deﬁned as one who neither has a licensure test score in the top quartile nor attended a very competitive college. For the teachers in their initial spells, we ﬁnd that the two measures of strong qual- iﬁcations, both separately and combined, are associated with higher overall exit hazards, despite the fact that these teachers are less likely than regular teachers to change schools within the district. Their higher overall exit rates arise because of their far greater likelihood of leaving the profession. In partic- ular, a teacher with both high test scores and a degree from a very competitive college is more than 25 percent more likely to end a teaching spell by leaving 422 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d f . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Charles T. Clotfelter, Helen F. Ladd, and Jacob L. Vigdor Table 6. Effects of Teacher Qualiﬁcations on Teacher Exits (Hazard Ratios) All Exit Routes Leave Teaching Exit Route Switch Districts Change Schools Panel A: Initial Teaching Spells Has a high quartile test score Teacher qualiﬁcations relative to regular teacher in average school 1.152∗∗ 1.092∗∗ 1.258∗∗ High test + very competitive college 1.025∗ 1.037∗∗ 1.063∗∗ Undergraduate very competitive Panel B: Subsequent Teaching Spells Has a high quartile test score Teacher qualiﬁcations relative to regular teacher in average school 1.082∗∗ 1.089∗∗ 1.179∗∗ High test + very competitive college Undergraduate very competitive 1.012 0.994 1.018 0.978 1.043 1.020 0.829∗∗ 0.938∗ 0.778∗∗ 0.897∗∗ 0.935∗∗ 0.839∗∗ 0.974 0.945∗∗ 0.920∗∗ 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 . ∗statistically signiﬁcant at the .10 level; ∗∗statistically signiﬁcant at the .05 level. the profession than is a regular teacher. For the teachers in their second or later teaching spells, in contrast, the overall exit hazards for teachers with this combination of strong qualiﬁcations of leaving their current school are not statistically different from those of a regular teacher. These overall exit ratios, however, mask differential exit routes of the teachers with strong qual- iﬁcations. Although such teachers are about 8 percent less likely than other teachers to end a teaching spell by moving to a different school within the same district and about 22 percent less likely to change districts, they are about 18 percent more likely to end it by leaving the profession. Responses to Own Salary and School Demographics The entries in table 7 illustrate how exit hazards are affected by salaries and by the demographic characteristics of the teachers’ schools, with particular attention to the different response rates of regular teachers and teachers with strong qualiﬁcations. In all cases, the salary change refers to the teacher’s salary in her current district, assuming no change in teaching or nonteaching salaries in the local area. Of most interest are the ﬁndings for the teachers in their initial spells as reported in the ﬁrst panel. Consider ﬁrst the simulated hazard ratios for a regular teacher of leaving her current school by any of the three exit routes. A 10 percent increase in salary reduces the probability that she will leave her current school in any given year by about 14 percent (one minus the hazard ratio of 0.861). In contrast, a 10 percentage point increase in the nonwhite students in the school increases the probability by about 2 percent, and a 1 standard deviation / / f e d u e d p a r t i c e - p d l f / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d . f f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 423 TEACHER MOBILITY, SEGREGATION, AND PAY Table 7. Responses to Salary and School Demographics, by Type of Teacher (Hazard Ratios) Exit Route All Exit Routes Leave Teaching Switch Districts Change Schools Panel A: Initial Spells Predicted response to a 10 percent increase in district salary (in average school) Regular teacher Regular teacher plus 2 qualiﬁcations 0.861∗∗ 0.944∗,++ 0.761∗∗ 0.925∗,++ 0.831∗∗ 0.814∗∗ 0.985 1.037 Predicted response to a 10 percentage point increase in percentage of nonwhite students in the school Regular teacher Regular teacher plus 2 qualiﬁcations 1.023∗∗ 1.069∗∗,++ 1.022∗∗ 1.067∗∗,++ 1.060∗∗ 1.096∗∗ 1.005 1.067∗∗,++ Predicted response to a 1 standard deviation increase in the percentage of free lunch students in the school Regular teacher Regular teacher plus 2 qualiﬁcations 1.030∗∗ 0.967,+ 1.023 0.897∗∗,++ 0.994 0.957 1.064∗∗ 1.088 Predicted response to a 10 percentage point increase in percentage of nonwhite students in the district Regular teacher Regular teacher plus 2 qualiﬁcations 1.010 0.960∗∗,++ 1.023∗∗ 1.024 1.005 0.914∗∗,++ 0.983 0.880∗∗,++ Panel B: Second or Later Spells Predicted response to a 10 percent increase in district salary (in average school) Regular teacher Regular teacher plus 2 qualiﬁcations 1.016∗ 1.007 1.016 1.025 0.970 0.955 1.019 1.010 Predicted response to a 10 percentage point increase in percentage of nonwhite students in the school Regular teacher Regular teacher plus 2 qualiﬁcations 1.022∗∗ 1.045∗∗ 1.024∗∗ 1.015 1.042∗∗ 1.065 1.011 1.081∗∗,++ Predicted response to a 1 standard deviation increase in the percentage of free lunch students in the school Regular teacher Regular teacher plus 2 qualiﬁcations 1.043∗∗ 1.026 1.002 0.983 1.099∗∗ 1.022 1.071∗∗ 1.091 Predicted response to a 10 percentage point increase in percentage of nonwhite students in the district 0.968∗∗ 0.869∗∗,++ Regular teacher plus 2 qualiﬁcations 1.029∗∗ 1.070∗∗ 1.008 0.996 1.018 1.037 Regular teacher Notes: ++ and + denote whether the underlying hazard coefﬁcient for a teacher with strong qualiﬁ- cations differs from the underlying hazard coefﬁcient for a teacher with regular qualiﬁcations at the .05 level and .10 levels, respectively. ∗statistically signiﬁcant at the .10 level; ∗∗statistically signiﬁcant at the .05 level. increase in the school’s free lunch percentage increases it by about 3 percent. Because the models separately control for student demographics at the district level, the simulated changes in school demographics should be interpreted as responses to changes in one school’s characteristics relative to those of other schools. With the exception of a change in the school’s free lunch percentage, which increases the probability that the teacher will switch to another school within the same district, the other demographic changes highlighted in the 424 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d . f f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Charles T. Clotfelter, Helen F. Ladd, and Jacob L. Vigdor Table 8. Salary Differentials (as a Percent of Salary) Required to Offset the Fraction Nonwhite at the School Level on the Overall Departure Hazard Difference in Nonwhite Share (Percentage Points) Regular teacher Strong teacher 10 1.5 10.2 20 3.0 23.1 30 4.5 34.7 40 6.3 46.3 50 7.5 58.3 table for a regular teacher all affect the odds that the teacher leaves the district or the profession, and not the odds of moving to another school within the district. For teachers with strong qualiﬁcations, the general patterns of hazard ratios are similar to those for regular teachers. Importantly, however, those with strong qualiﬁcations exhibit a much more muted response to salary differentials (0.944 hazard ratio vs. 0.861) and a more pronounced response to the nonwhite share of students (1.069 hazard ratio vs. 1.023). These patterns indicate that salary differentials are a relatively powerful motivator for keeping regular teachers in their initial teaching spells in their original schools but a far less powerful motivator for teachers with strong qualiﬁcations. In particular, table 8 shows the simulated salary increases that would be required to counter the effects of differing percentages of nonwhite students in a particular school for the two types of teachers. For a regular teacher, the table indicates that a small salary increase of 3 percent would sufﬁce to offset a 20 percentage point difference in the nonwhite share of students in the teacher’s current school, and a 7.5 percent salary increase would offset a 50 percentage point difference. We note that these estimates, which under plausible assumptions once again are upper bounds, represent salary differentials that are within the range of observed salary differences in our data and are also within the realm of political feasibility. To retain teachers with strong qualiﬁcations in schools with high nonwhite shares, in contrast, the required salary increases are far higher, ranging from over 10 percent to offset a small percentage point difference in the nonwhite share to 58.3 percent to offset a 50 percentage point difference. We note that these projected salary differentials are large relative to the observed variation in actual salaries in our data and hence represent out-of-sample predictions. While that means we cannot be very conﬁdent about the precise numbers, we can conclude that the salary differences required to retain teachers with strong qualiﬁcations are high and likely beyond the realm of political feasibility.27 27. This statement is true even with the additional qualiﬁcation that the coefﬁcients on the salary estimates may be biased downward, which would lead to an overstatement of the required salary differentials. 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d f . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 425 TEACHER MOBILITY, SEGREGATION, AND PAY Consistent with this ﬁnding, salary also emerges as a relatively ineffective policy tool for changing the behavior of the more experienced teachers in later teaching spells. That conclusion emerges from panel B of table 7, which shows that salary differentials have little or no effect on the exit decisions of such teachers.28 In particular, the entries in the top line indicate that a 10 percent increase in district salary could possibly even induce an increase in the hazard of exiting, although the coefﬁcient is only marginally signiﬁcant. Further, such a salary increase would have essentially no effect on the hazard ratio for teachers with strong qualiﬁcations. At the same time, however, these experienced teachers respond just as strongly to the shares of nonwhite students or low-income students in their schools as do the teachers in initial spells, with teachers having strong qualiﬁ- cations being more responsive than those with regular qualiﬁcations. Because they tend to be established in a community, they are more likely than inexpe- rienced teachers to respond to high percentages of nonwhite students in their current school by transferring to another school within the same district rather than moving to another district or dropping out of teaching. With respect to district-level differences in the share of nonwhite students in the district, both regular and strong teachers exhibit higher probabilities of leaving the profes- sion. We have not replicated table 8 for teachers in these later teaching spells since the lack of responsiveness of such teachers to salary differentials rules out salary differentials as a tool for offsetting the demographic characteristics of the schools. Responses to Other Labor Market Conditions, Including Local Bonus Programs Tables 4 and 5 also provide results related to other labor market conditions. First, the coefﬁcients on teacher salaries in surrounding districts indicate the expected ﬁnding that, controlling for the salary in her current district, a teacher is more likely to switch districts the higher the teacher salaries are elsewhere. Second, but somewhat harder to explain, is that a teacher is also more likely to change districts but not to leave the profession in response to higher nonteaching salaries in the local area. Given that we have also controlled for the local unemployment rate, however, those higher salaries in nonteaching jobs need not be associated with job openings. As expected, teachers respond to higher local unemployment rates by reducing the rate at which they leave the profession in the current year. 28. Based on a different model speciﬁcation and with data from Texas, Hanushek, Kain, and Rivkin (2004) conclude that the power of salary differentials reaches a peak for teachers with 3–5 years of experience but falls off for more experienced teachers. Their ﬁndings seem roughly consistent with ours but are hard to compare, given that we focus on teaching spells, not years of experience. 426 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d . f f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Charles T. Clotfelter, Helen F. Ladd, and Jacob L. Vigdor Some positive, but at best limited, evidence emerges related to the Equity Plus programs in Charlotte-Mecklenburg and Winston-Salem. Recall that the relevant coefﬁcients are those associated with designation as an Equity Plus program in the current year, controlling for whether a school ever met the criteria for the program. Consistent with the goals of these programs, three out of four of the relevant hazard ratios in the ﬁrst column for teachers in either initial or subsequent spells are less than one. Although these coefﬁcients suggest that the program was associated with a lower overall departure hazard, only one of them (for initial spells in Winston-Salem) is even marginally signiﬁcant. The most signiﬁcant of all the relevant coefﬁcients related to the two programs is the 0.648 coefﬁcient for the probability that a teacher in an initial teaching spell in Charlotte-Mecklenburg changes districts. It is a bit puzzling why the program in that city would reduce the odds that a teacher would leave the district but not the odds that she would leave the designated school in which she received the bonus. The Statewide Bonus Program Because of the complicated eligibility conditions surrounding the state’s short- lived $1,800 bonus program, we augmented the basic models of the form shown in tables 4 and 5 with a set of six additional variables and limited the analysis to middle and high schools. Although the speciﬁcations are similar in spirit to those we present in Clotfelter et al. 2008, they differ in that these models are estimated separately for teachers in their initial or subsequent teaching spells, they differentiate exit routes, and they include only the teachers starting teaching spells during the relevant period. Of most interest are the coefﬁcients of the variable indicating that a teacher was eligible to receive a $1,800 bonus.29 The other coefﬁcients refer to whether the teacher was eligible for the bonus (deﬁned as a certiﬁed teacher of math, science, or special education) in an eligible school (deﬁned as a high-poverty middle school or a low-performing high school), both directly and interacted with the years the program was in existence. Table 9 reports results for all teachers in their initial spells (panel A) and for all teachers in subsequent spells (panel B). No distinction is made in this table between teachers with regular or strong qualiﬁcations.30 The most compelling ﬁndings emerge in panel B. The relevant coefﬁcients imply that, consistent 29. We used an indicator variable rather than adding the bonus amount to a teacher’s salary because of the temporary nature of the bonus, which differentiates it from a permanent increase in salary (Clotfelter et al. 2008). The coefﬁcient of the indicator represents a difference-in-difference estimate of the effect of receiving the bonus. 30. Models that interact all the bonus variables with a measure of strong preservice qualiﬁcations provide some hints that the effects may be more pronounced for regular than for strong teachers, but the differences are not statistically signiﬁcant. 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d . f f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 427 TEACHER MOBILITY, SEGREGATION, AND PAY Table 9. Effects of the $1,800 Bonus Program Panel A: Initial Teaching Spells Received $1,800 bonus Eligible teacher Eligible school Eligible teacher 2002–4 Eligible school 2002–4 Eligible teacher∗Eligible school All Exit Routes Leave Teaching Exit Route Switch Districts 1.065 0.969 0.978 0.991 1.018 0.992 0.943 0.979 0.985 0.807∗∗∗ 1.063 0.981 0.921 1.007 0.891∗ 1.102 0.937 1.135 Change Schools 1.460∗∗ 0.890∗∗∗ 1.013 1.328∗∗∗ 1.002 0.962 Number of teacher-years 61,713 61,713 61,713 61,713 Panel B: Subsequent Teaching Spells Received $1,800 bonus Eligible teacher Eligible school Eligible teacher 2002–4 Eligible school 2002–4 Eligible teacher∗Eligible school 0.860∗ 1.083∗∗∗ 0.719∗∗ 1.210∗∗∗ 0.970 0.996 1.061 1.024 1.013 0.941 1.028 1.000 0.979 1.035 0.880∗ 1.141 1.139 0.956 1.022 0.898∗∗ 0.951 1.052 1.074 1.113 Number of teacher-years 50,904 50,904 50,904 50,904 ∗statistically signiﬁcant at the .10 level; ∗∗statistically signiﬁcant at the .05 level; ∗∗∗statistically signiﬁcant at the .01 level. with the goals of the program, experienced teachers who received the bonus were 14 percent less likely to leave their school in the following year than were other teachers, all other factors held constant. In addition, they were 28 percent less likely to leave the profession. In contrast, the program did not have its intended effect on teachers in their ﬁrst teaching spells and, in fact, appears to have increased the probability that they would move to another school within the district. These patterns are generally consistent with those from our earlier study, where we found that the statewide bonus program had a larger effect on experienced than on inexperienced teachers (Clotfelter, Ladd, and Vigdor 2008). At the same time, the greater responsiveness to the bonus among the more experienced teachers may at ﬁrst appear inconsistent with the results reported in table 8 showing larger responsiveness to salary differentials among teachers in their ﬁrst spells. The two ﬁndings can be reconciled by the recognition that many teachers expected the bonus program to be short lived, which in fact it turned out to be. It is quite plausible that teachers in their ﬁrst teaching spells would be far more responsive to permanent salary differentials than 428 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d f . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Charles T. Clotfelter, Helen F. Ladd, and Jacob L. Vigdor to short-term bonuses in making decisions about whether to remain in their current position. More experienced teachers, in contrast, may well perceive the bonus as a short-term incentive to remain in teaching or in their current school slightly longer than they otherwise would have. 6. CONCLUSION Among the possible policies for ensuring equal access of racial and socioe- conomic groups of students to quality educational resources at the school level, two are of particular interest. One is to distribute students of all groups evenly across schools, in which case the pattern by which resources (such as teachers with strong qualiﬁcations) are distributed across schools would have no adverse equity consequences. The racial desegregation that federal courts pushed in the 1960s and 1970s could be interpreted as a version of this ﬁrst policy strategy. This approach has always been problematic, however, because existing patterns of residential segregation across states, counties, or other geographic regions make it difﬁcult, if not impossible, to attain an even distribution of students across schools within districts, let alone metropolitan areas or states. Moreover, recent court decisions have all but precluded the possibility of using this strategy to promote educational equity within school districts. A second method of promoting equity in the presence of school segrega- tion is to use policy levers to equalize resources across schools. Research has consistently shown that the most signiﬁcant educational resource—teacher quality—is distributed very unevenly among schools to the clear disadvantage of minority students and those from low-income families. The results of this study highlight the difﬁculties inherent in overcoming this historical pattern using the most obvious available ﬁnancial policy lever—differential teacher salaries. On the positive side, our analysis indicates that salary differentials do have some role to play in easing the challenge that hard-to-staff schools face in attracting teachers with strong qualiﬁcations. The more schools are segregated (particularly by race), the greater the required salary differentials needed to level the playing ﬁeld. In addition, our results indicate that salary differentials of feasible magnitudes can counter the repelling effects of concentrations of disadvantaged students for at least one group of teachers—those with average qualiﬁcations who are in their ﬁrst teaching spell. On the negative side, salary differentials emerge as a far less effective tool for changing the school departure behavior of teachers with strong preservice qualiﬁcations or those who are no longer in their ﬁrst teaching spells. This conclusion reﬂects our ﬁndings that such teachers are both more responsive 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d . f f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 429 TEACHER MOBILITY, SEGREGATION, AND PAY to the racial and socioeconomic mix of a school’s students and less responsive to salary than are their less well qualiﬁed counterparts when making decisions about remaining in their current school, moving to another school or district, or leaving the teaching profession. Thus, as we reported in table 8, for teach- ers with strong preservice qualiﬁcations in their initial teaching spells, the simulated salary differentials required to neutralize the effect of large concen- trations of disadvantaged students are large and on the order of 40–50 percent of salary. In addition, taken literally, our estimates for teachers in subsequent spells imply that no salary differentials would be large enough to compensate them for being in schools with concentrations of disadvantaged students. Given these ﬁndings, we are not optimistic about the power of salary differentials alone to promote educational equity in the context of schools that are highly segregated by educational disadvantage, and particularly by race. Although such differentials, whether in the form of permanent salary differences or short-run bonus programs, may reduce turnover in hard-to- staff schools, which in itself is a desirable outcome, and may keep some teachers from leaving the profession, they are far less effective in equalizing the quality of teachers across schools. Thus even with a judicious use of salary differentials speciﬁcally designed to promote a more equitable distribution of teachers across schools, the more segregated the schools are, the more unequal is likely to be the quality of teachers across schools. This research was supported by a grant from the U.S. Department of Education through the Center for the Analysis of Longitudinal Data (CALDER) in Washington, DC. Re- search support was also provided by the Russell Sage Foundation and the Spencer Foundation. The authors are grateful to the North Carolina Education Research Data Center for making the data available and to research assistants Janeil Belle, Reid Chisholm, Sarah Gordon, Robert Malme, Patten Priestley, Valorie Rawlston, Shirley Richards, and Russ Triplett. REFERENCES Baugh, William H., and Joe A. Stone. 1982. Mobility and wage equilibration in the education labor market. Economics of Education Review 2(3): 253–74. Becker, Howard. 1952. The career of the Chicago public school teacher. American Journal of Sociology 57: 470–77. Betts, Julian R., Kim S. Rueben, and Anne Danenberg. 2000. Equal resources, equal outcomes? The distribution of school resources and student achievement in California. San Francisco: Public Policy Institute of California. Boyd, Donald, Pam Grossman, Hamilton Lankford, Susanna Loeb, and James Wyckoff. 2008. Who leaves? Teacher attrition and student achievement. NBER Working Paper No. 14022. 430 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d . f f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Charles T. Clotfelter, Helen F. Ladd, and Jacob L. Vigdor Boyd, Donald, Hamilton Lankford, Susanna Loeb, and James Wyckoff. 2005. Explain- ing the short careers of high-achieving teachers in schools with low-performing stu- dents. American Economic Review 95: 166–71. Clotfelter, Charles T., Elizabeth Glennie, Helen F. Ladd, and Jacob L. Vigdor. 2008. Would higher salaries keep teachers in high-poverty schools? Evidence from a policy intervention in North Carolina. Journal of Public Economics 92: 1352– 70. Clotfelter, Charles T., Helen F. Ladd, and Jacob L. Vigdor. 2003. Segregation and resegregation in North Carolina’s public school classrooms. North Carolina Law Review 81: 1463–1511. Clotfelter, Charles T., Helen F. Ladd, and Jacob L. Vigdor. 2005. Who teaches whom? Race and the distribution of novice teachers. Economics of Education Review 24(4): 377–92. Clotfelter, Charles T., Helen F. Ladd, and Jacob L. Vigdor. 2006. Teacher-student matching and the assessment of teacher effectiveness. Journal of Human Resources 41(4): 778–820. 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. Clotfelter, Charles T., Helen F. Ladd, and Jacob L. Vigdor. 2008. School segregation under color-blind jurisprudence: The case of North Carolina. Virginia Journal of Social Policy and the Law 16: 46–86. Clotfelter, Charles T., Helen F. Ladd, and Jacob L. Vigdor. 2010. Teacher credentials and student achievement in high school: A cross-subject analysis with ﬁxed effects. Journal of Human Resources 45(3): 655–81. Clotfelter, Charles T., Helen F. Ladd, Jacob L. Vigdor, and Roger A. Diaz. 2004. Do school accountability systems make it more difﬁcult for low-performing schools to attract and retain high-quality teachers? Journal of Policy Analysis and Management 23(2): 251–71. Clotfelter, Charles T., Helen F. Ladd, Jacob L. Vigdor, and Justin Wheeler. 2007. High poverty schools and the distribution of teachers and principals. North Carolina Law Review 85(5): 1345–79. Dolton, Peter, and Wilbert van der Klaauw. 1995. Leaving teaching in the UK: A duration analysis. Economic Journal 105: 431–44. Falch, Torberg, and Bjarne Strom. 2005. Teacher turnover and non-pecuniary factors. Economics of Education Review 24(6): 611–31. Feng, Li, and Tim R. Sass. 2008. Teacher quality and teacher mobility. Unpublished paper, SUNY Fredonia. Gleason, Philip M. 1995. Participation in the national school lunch program and the school breakfast program. American Journal of Clinical Nutrition 61(1): 213S– 20S. 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d f . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 431 TEACHER MOBILITY, SEGREGATION, AND PAY 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. Goldhaber, Dan, and Emily Anthony. 2007. Can teacher quality be effectively assessed? National Board certiﬁcation as a signal of effective teaching. Review of Economics and Statistics 89(1): 134–50. Goldhaber, Dan, Bethany Gross, and Daniel Player. 2009. Teacher career paths, teacher quality, and persistence in the classroom: Are schools keeping their best? CALDER Working Paper No. 29. Greenberg, David, and John McCall. 1974. Teacher mobility and allocation. Journal of Human Resources 9(4): 480–502. Gritz, R. Mark, and Neil D. Theobald. 1996. The effects of school district spend- ing priorities on length of stay in teaching. Journal of Human Resources 31(3): 477– 512. Hanushek, Eric, John Kain, Daniel O’Brien, and Steven Rivkin. 2005. The market for teacher quality. NBER Working Paper No. 11154. Hanushek, Eric, John Kain, and Steven Rivkin. 2004. Why public schools lose teachers. Journal of Human Resources 39(2): 326–54. Hollingshead, August B. 1949. Elmtown’s youth: The impact of social classes on adolescents. New York: John Wiley and Sons. Imazeki, Jennifer. 2005. Teacher salaries and teacher attrition. Economics of Education Review 24: 431–49. Jackson, C. Kirabo. 2009. Student demographics, teacher sorting and teacher quality: Evidence from the end of school desegregation. Journal of Labor Economics 27(2): 213– 56. Johnson, Susan Moore, and Sarah E. Birkeland. 2003. Pursuing a “sense of success”: New teachers explain their career decisions. American Educational Research Journal 40(3): 581–617. Koedel, Cory, and Julian R. Betts. 2007. Re-examining the role of teacher quality in the educational production function. Working Paper No. 708, University of Missouri. Krieg, John M. 2006. Teacher quality and attrition. Economics of Education Review 25(1): 13–27. Ladd, Helen F. Forthcoming. Teacher working conditions: How predictive of policy relevant outcomes? Education Evaluation and Policy Analysis. In press. Ladd, Helen F. 2008. Reﬂections on equity, adequacy, and weighted student funding. Education Finance and Policy 3(4): 402–23. Lankford, Hamilton, Susanna Loeb, and James Wyckoff. 2002. Teacher sorting and the plight of urban schools: A descriptive analysis. Educational Evaluation and Policy Analysis 24: 37–62. 432 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d . f f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Charles T. Clotfelter, Helen F. Ladd, and Jacob L. Vigdor Lockwood, J. R., Daniel F. McCaffrey, and Tim R. Sass. 2008. The intertemporal stability of teacher effect estimates. Paper prepared for the National Conference on Value-Added Modeling, University of Wisconsin, Madison, April. Milanowski, Anthony T., Hope Longwell-Grice, Felicia Saffold, Janice Jones, Kirsten Schomisch, and Allan Odden. 2009. Recruiting new teachers to urban school districts: What incentives will work? International Journal of Education Policy and Leadership 4(8): 1–13. Mont, Daniel, and Daniel I. Rees. 1996. The inﬂuence of classroom characteristics on high school teacher turnover. Economic Inquiry 34: 152–67. Murnane, Richard J., and Randall J. Olsen. 1989. Will there be enough teachers? American Economic Review 79: 242–46. Murnane, Richard J., and Randall J. Olsen. 1990. The effects of salaries and opportunity costs on length of stay in teaching: Evidence from North Carolina. Journal of Human Resources 25(1): 106–24. Podgursky, Michael, Ryan Monroe, and Donald Watson. 2004. The academic quality of public school teachers: An analysis of entry and exit behavior. Economics of Education Review 23: 507–18. Reed, Deborah, Kim S. Rueben, and Elisa Barbour. 2006. Retention of new teachers in California. San Francisco: Public Policy Institute of California. Rickman, Bill D., and Carl D. Parker. 1990. Alternative wages and teacher mobility: A human capital approach. Economics of Education Review 9(2): 127–36. Roemer, John E. 1998. Equality of opportunity. Cambridge, MA: Harvard University Press. Scaﬁdi, Benjamin, David Sjoquist, and Todd R. Stinebrickner. 2007. Race, poverty, and teacher mobility. Economics of Education Review 26(2): 145–59. Stinebrickner, Todd R. 1998. An empirical investigation of teacher attrition. Economics of Education Review 17(2): 127–36. Walden, Michael L., and Zulai Sogutlu. 2001. Determinants of intrastate variation in teacher salaries. Economics of Education Review 20(1): 63–70. APPENDIX DATA DEFINITIONS Local Teacher Salary Supplements Teacher salaries in North Carolina are the sum of (1) a mandated state portion taken from an annually revised schedule based on degree and years of expe- rience and (2) an optional local supplement added by most school districts. Some districts offer a ﬂat rate supplement for all teachers; others offer a uni- form percentage increase in the state salary; and some make the supplement a function of such things as certiﬁcation status, years of experience, or degree 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d . f f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 433 TEACHER MOBILITY, SEGREGATION, AND PAY level. Since these formulas are not stored centrally, the information available to calculate them is generally not available for any but the most recent years covered by our data set. The only information on local supplements that is available over all years in our data set is the total amount of supplements paid by a district for a school year. Since the form of the local supplement necessarily affects the salary sched- ule facing teachers in various districts, we sought to retain the differences in formulas. We therefore combined the aggregate amount paid in each district with information on the form of the supplement for the 2004–5 year, on the assumption that the form of the supplement did not change over time. Thus for districts whose local supplement was a ﬂat amount in 2004–5, we assumed that their local supplements were ﬂat amounts in other years. For these districts, therefore, we calculated each teacher’s local supplement to be the average supplement paid to teachers in that district in the correspond- ing year. For districts whose supplements were simple percentages of the state-determined salary, we determined the rate as the ratio of the total local supplement for teachers to the total state salary bill for teachers in the district in that year. For the districts that applied exact percentage rates to different classes of teachers, we computed the percentage rates that would yield a total district supplement equal to the reported total. For these districts, we ﬁrst applied the 2004–5 rates by category (most of these districts differentiated by expe- rience categories), calculated the implied total amount, noted the percentage error, then adjusted the rates for all categories proportionately so that the adjusted percentages yielded the correct district total. For a number of dis- tricts we were able to obtain the formulas used in 2001–2 and were able to compare the categories and percentage rates to those for 2004–5. If the earlier formulas differed, we used the 2001–2 formulas as the basis for the estimates for 2001–2 and earlier years, using the same approach described above. For the remaining districts, whose local supplement was determined by more involved formulas, we compared the supplements given in tabular form to the pattern of salaries given in the state’s salary tables to determine if the formula was closer to a ﬂat amount or to a ﬁxed proportion. Speciﬁcally, two parameters were calculated for the supplements given in tabular form: average experience progression (the average increase per year of experience calculated as an exponential growth rate) and the salary premium given to teachers with a master’s degree as compared to those with a bachelor’s. In the state’s salary scale for 2004–5, average experience progression was 0.020, and the master’s degree premium for teachers with ten years of experience was 10.0 percent. For districts that expressed their supplement in tabular format, the average 434 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d f . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Charles T. Clotfelter, Helen F. Ladd, and Jacob L. Vigdor experience progression was calculated by comparing supplements for bach- elor’s level teachers at three and twenty-nine years, and the master’s degree premium, if any, was calculated at ten years of experience. A district was as- sumed to have a proportionate form for its supplement if either one of these rates exceeded the state rate or if both were more than half the state rate for that year. Otherwise, a district’s supplement was assumed to be a ﬂat amount for all teachers. Proportional formulas could be of four types: (1) a simple percentage of the state salary, (2) a percentage of the state salary based on the teacher’s de- gree, (3) a percentage based on the teacher’s experience (number of years teach- ing), and (4) a percentage that used both degree and experience. The breakdown of each type for the 117 districts in 2003–4 was (1) 73, (2) 2, (3) 8, and (4) 2; the remaining districts had either an additive supplement (22) or no supplement at all (10). Arriving at a formula for each district that both retained the form that was used in 2004–5 and was consistent with the aggregate value of supplements paid in the district in a given year required a two-step estimation procedure. In the ﬁrst step, the applicable formula for each district was used to calculate the supplement amount for each teacher in each district. In the second step, the formulas used in the ﬁrst step were adjusted so that the total of all supplements calculated for each district would be equal to the total for the district given in the supplement data. Alternative Teacher Salaries To assess a teacher’s earnings alternatives, we sought to calculate the aver- age teacher salary available in nearby school districts. To keep to a reasonable level the amount of necessary calculations, we compared for one of six stan- dard teacher proﬁles the salary available in a teacher’s own district with the enrollment-weighted average salary available in districts within a thirty-mile radius (measured between centroids) of the teacher’s own district. These six proﬁles combined two certiﬁcate types (bachelor’s and master’s) and three experience categories: 0–4 years of experience (median 2); 5–11 years (median 8), and 12 or more years (median assumed to be 18). The measure of relative teaching salaries is the logarithm of the ratio of the own-district teacher salary to the average regional teacher salary, both deﬁned for the category into which each teacher falls. To illustrate the importance of the functional form used in calculating local supplements, the salary ratio of Wake to neighboring Franklin County in 2003–4 ranged from 1.07 for inexperienced teachers with a bachelor’s degree to 1.12 for teachers holding a master’s having twelve or more years teaching experience. 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d f . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 435 TEACHER MOBILITY, SEGREGATION, AND PAY Nonteaching Salaries The measure of nonteaching salaries is the employment-weighted average of salaries in all counties within a thirty-mile radius of the (centroid of the) school district of the teacher in question. The deﬁnition of nonteaching earn- ings we use is private employment (equal to total employment minus farm employment and government employment), a deﬁnition that includes pro- prietors (from BEA Regional Economic Accounts Regional Economic Infor- mation System data 1995–2003; available www.bea.gov/regional/reis, Table CA-6, Compensation by Industry, and CA-25, Total Employment by Industry). An alternative deﬁnition, nonfarm wage and salary employment (nonfarm em- ployment minus nonfarm proprietor employment), a deﬁnition that includes government, yielded similar results. Classiﬁcation of Schools by Level In North Carolina, the most common grade ranges corresponding to each of the three school levels are elementary (K−5 or PK−5), middle school (6–8), and high school (9–12). In 1999–2000, over 70 percent of the state’s public schools conformed to one of these grade ranges. To classify the remaining schools, each was assigned to the level in which most of its grades fell, or to the lower level in the case of equal numbers in two levels. Thus, for example, schools covering grades 3–5 are classiﬁed as elementary and those with 6–12 are classiﬁed as high schools. College Selectivity The categories were derived from information from Barron’s College Admis- sions Selector for 1988, based on information for ﬁrst-year students in each university in 1986–87. Our category “very competitive” includes universities rated as most competitive, highly competitive, or very competitive; “compet- itive” includes those rated as competitive; “less competitive” includes those rated as less competitive or noncompetitive; and “unranked” includes spe- cial programs such as art schools, international universities, or universities for which we were not able to ﬁnd a rating. Barron’s uses criteria such as the median entrance examination scores, percentages of students scoring ﬁve hundred and above and six hundred and above on both the math and verbal parts of the SAT or comparable scores for the ACT, the percentage of students who ranked in the upper ﬁfth or two-ﬁfths of their high school class, and the percentage of applicants who were accepted. If information for a university was missing for 1988, we substituted the ranking for the 1979 or 1999 selector, with the choice varying with the era in which the teacher attended college. 436 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d f . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Charles T. Clotfelter, Helen F. Ladd, and Jacob L. Vigdor Teacher Test Scores Teachers took a variety of standardized tests as part of the state’s licensure requirements. We used results from nineteen of the most frequently taken tests. We formed a standardized licensure test score variable for each teacher by converting test scores from different test administrations in North Carolina to standardized scores using the means and standard deviations for tests taken in each year by all teachers in our data set. We normalized test scores on each of these tests separately for each year the test was administered based on means and standard deviations from test scores for all teachers in our data set and then assigned to each teacher the average standardized score on the tests taken by that teacher. Classiﬁcation of Districts All districts in counties that were 45 percent or more urban in 1990 were classiﬁed as urban, as were all city districts in any county with enrollments of at least two thousand in 2001–2, not counting charter school enrollments. The boundaries between coastal, Piedmont, and mountain counties were taken from North Carolina Division of Travel and Tourism, Yours to Discover: North Carolina State Parks and Recreation Areas (1998). For the classiﬁcation of spe- ciﬁc districts, see Clotfelter, Ladd, and Vigdor (2003, Appendix A). 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d f . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 437 TEACHER MOBILITY, SEGREGATION, AND PAY Table A.1. Descriptive Statistics for the Probit Sample Mean SD Min Max School Characteristics Middle High school Nonwhite, K–12 (%) Building age Free lunch, elementary (%)∗ Free lunch, middle (%)∗ Free lunch, high school (%)∗ District Characteristics Free lunch (%) Nonwhite (%) Enrollment (ln) Growth Rural Coastal Mountain Beach Salary (BA + 2) (ln) Labor Market Characteristics Alt. salary (ln) Unemployment rate Years 1997 1998 1999 2000 2001 2002 2003 2004 0.24 0.28 43.36 35.53 40.66 33.22 20.25 30.56 40.81 10.00 0.02 0.37 0.16 0.17 0.05 10.16 10.39 4.78 0.12 0.11 0.12 0.13 0.12 0.12 0.12 0.04 0.43 0.45 25.82 18.48 21.42 18.47 15.40 11.81 18.57 1.08 0.01 0.48 0.36 0.38 0.23 0.08 0.64 2.05 0.33 0.32 0.33 0.33 0.33 0.33 0.32 0.20 0 0 0 0 0 0 0 0 0.82 6.50 −0.03 0 0 0 0 9.97 0 1.20 0 0 0 0 0 0 0 0 1 1 100.00 93.00 99.14 94.87 95.00 75.87 97.42 11.66 0.06 1 1 1 1 10.31 10.63 18.20 1 1 1 1 1 1 1 1 N = 129,388 ∗Calculated separately by school level. 438 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 / / / / / 6 3 3 9 9 1 6 8 9 2 7 9 e d p _ a _ 0 0 0 4 0 p d . f f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 TEACHER MOBILITY, SCHOOL image

下载pdf