Competition, Wages and Teacher Sorting: Lessons Learned from a Voucher Reform

Competition, Wages and Teacher Sorting:
                        Lessons Learned from a Voucher Reform
                                                      Lena Hensvik∗
                                                      April 15, 2012


                                                          Abstract
            This paper examines how the entry of private independent high schools in Sweden affects the mobility
        and wages of teachers in a market with individual wage bargaining. Using matched employer-employee
        panel data covering all high school teachers over 16 years, I show that the entry of private schools
        is associated with higher teacher salaries, also for teachers in public schools. The wage returns from
        competition are highest for teachers entering the profession and for teachers in math and science. Private
        school entry also seems to have increased wage dispersion between high- and low-skilled teachers within
        the same field.


1       Introduction

The teacher labour market has received growing attention among social scientists and policy
markers as evidence suggests that teacher quality is one of the key inputs in improving
school performance (see e.g. Rockoff, 2004, Hanushek et al., 2005). Despite this, teacher pay
remains low and compressed in many countries compared to other occupations with similar
qualification requirements. In addition, the factors that determine teacher salaries often bear
little relation to student achievement.
     The main concern with this pay structure is that it may limit the supply of potential
teachers and push the most highly skilled teachers into other segments of the labour market
(Ballou and Podgursky, 1997; Hoxby and Leigh, 2004). To achieve a school system of high
quality it is therefore important to understand how the particular features of the teacher
labour market affect teacher compensation.
    ∗ Institute
              for Labor Market Policy Evaluation, Box 513, 75120 Uppsala, Sweden. Email: lena.hensvik@ifau.uu.se. I thank
the editor and two anonymous referees for their constructive suggestions. I am also grateful to Olof Aslund, Erling Barth, David
Figlio, Erik Grönqvist, Caroline Hoxby, Francis Kramarz, Mikael Lindahl, Matti Sarvimäki, Peter Nilsson, Oskar Nordström
Skans, Jonas Vlachos and audiences at IFAU, VATT, the ”Labor Development Reading Group” lunch at Stanford, the 2010
ELE Summer Institute in Reykjavik and the 2010 All California Labor Conference in Santa Barbara for helpful discussions and
comments. I am also particularly grateful to Björn Öckert, Olle Folke, Mikael Lindahl and Anders Böhlmark for kindly sharing
the data. Part of this work was completed while visiting Stanford Institute for Economic Policy Research. I thank FAS and
the Berch and Borgström foundations for their financial support.
This paper provides evidence on how private school competition affects teacher mobility
and wages. I investigate the consequences of a Swedish policy reform implemented in the
mid 1990s, which allows publicly funded private schools to operate in the market for high
school education.1 The reform initiated a rapid expansion in the number of private schools
(Figure 1) and large temporal and regional variation in private school entry. Because almost
all Swedish schools were run by local governments prior to the reform the new sector of
private employers dramatically increased competition in the teacher labour market.
    Reforms that increase local school competition can have a significant impact on teacher
compensation through reductions in the monopsonistic power of incumbent schools and in-
creased competition over students (Boal and Ransom, 1997, Hoxby, 2002, Manning, 2003).
The teachers market has long been subject to monopsony concerns due to the limited geo-
graphic and occupational mobility options for teachers, which is likely to generate significant
market power for schools when setting wages. While recent estimates of teacher mobility
provide indirect evidence of monopsony power in the teacher labour market (Falch, 2010,
Ransom and Sims, 2010), few empirical studies have convincingly demonstrated its actual
impact on teacher wages.2
    Many countries have recently adopted or considered reforms aimed at increasing competi-
tion between schools.3 The Swedish experience is particularly interesting given the substan-
tial increase of private alternatives in the education market. The variation in private high
school openings over time and across regions, together with the rich data at hand moreover
enables me to address many of the limitations in existing work, which has mainly relied
on cross-sectional data (Vedder and Hall, 2000 and Medcalfe and Thornton, 2006). Impor-
    1 The reform was implemented in 1992 at the compulsory level and in 1994 at the high school level. The reason for focusing on

high school teachers is first of all that the expansion of private schools is larger and second, that the teachers have well-defined
fields of specialization, which enables a more detailed analysis of the differential effects with respect to teacher characteristics.
    2 Several studies document large and systematic wage differences between observably identical workers, both across industries

(Krueger and Summers, 1988, Katz and Summers, 1989 and Murphy and Topel, 1990) and across local labour markets (Moretti,
2011). One theoretical explanation behind such differences is that they reflect variations in the competitiveness of markets that
arise from, for example, search frictions or entry barriers (cf. Manning, 2003). However, few studies have empirically investigated
the consequences of increased competition in any labour market on workers’ wages. Black and Strahan (2001) and Bertrand
and Kramarz (2002) provide evidence from the private sector in France and the US.
    3 These reforms have spurred considerable debate on whether market reforms are effective in raising student achievement.

However, there is no consensus in the literature on the effects of competition on school performance.A number of papers have
investigated the relationship between private school competition on student test scores, grades and university attendance,
finding only weak and inconsistent evidence of such student achievement gains both in Sweden and elsewhere (cf. Ahlin (2003)
Sandström and Bergström (2005) and Böhlmark and Linddahl (2009) for evidence from Sweden, Gibbons et al., (2008) and
Clark (2009)from the UK, and Hoxby (2003) and Figlio and Hart (2010) from the US).


                                                                 2
tantly, the local governments who run the public schools have little influence over the inflow
of private schools, since school entry is decided at the national level.
   The most closely related paper is Kirabo Jackson and Cowan (2009), who study the
effects of charter school entry in North Carolina on the teacher labour market. They provide
evidence that private charter school entry is associated with higher public school teacher
salaries, in particular in difficult-to-staff schools. Though this evidence is compelling, one
limitation is that fixed teacher credentials determine teacher pay to a large extent, which
limits schools’ ability to respond to local competition. An important contribution of this
paper is that it examines the wage effects of competition in a context where the rigid wage
scales have been abandoned in favor of individual wage setting. Hence there is substantial
room for wage adjustments to the local market conditions and the needs of recruiting and
retaining employees in a competitive environment.
   In addition, detailed data on teacher’s field of specialization along with measures of their
cognitive and social skills allow for an examination of whether competition has created
winners and losers in the teaching pool. Such distributional effects are clearly important
from an education policy perspective, as a more differentiated wage setting of teachers may
affect the characteristics of the teacher’s pool and the quality of the teachers remaining in
public schools. By looking at outcomes related to both teacher salaries and teacher mobility,
this paper aims to provide a comprehensive evidence on how competition affects teacher
outcomes.
   The results suggest that private schools hire different teachers than the traditional public
schools; they hire from a broader array of occupations and recruit more from the private
sector than public schools. The entry of private schools also seem to affect the composition
of the teachers remaining in public schools, as private schools systematically hire public
school teachers with less formal qualifications but high cognitive ability. However, I also find
that competition leads to higher teacher salaries. The effects are particularly pronounced for
teachers with high mobility; teachers entering the profession in the most competitive areas
receive 2 to 3 % higher wages than comparable teachers in areas without competition from


                                               3
Fig. 1: Private share in Swedish Local Labour Markets 1991-2006


private schools. The effects persist once individual heterogeneity is controlled for and when
restricting the sample to public school teachers. This suggests that public schools attempt
to reduce teacher turnover by paying higher salaries to the most mobile teachers.
   The empirical strategy exploits changes in private high school entry within and across
local labour markets, hence it accounts for many of the potential confounders that could gen-
erate a spurious relationship between school competition and wages, such as time-invariant
differences between local labour markets and local linear trends in unobserved determinants
of wages. Still however, the main concern is that the effects capture changes in time-varying
unobserved characteristics of local labour markets rather than competition. I provide a range
of robustness tests addressing this concern; all contradict that the results are simply driven
by spurious correlations.
   In the final part of the paper, I examine whether the effects of competition vary with
teachers’ field as well as their cognitive and social skills. Notably, the results suggest that the
magnitude of the competition effect varies substantially depending on teacher characteristics.
The effect is first of all concentrated to teachers in difficult-to-staff areas, such as math and
science and vocational subjects. Second, the competition effect is most pronounced for math
and science teachers with high cognitive skills and among social science teachers with high


                                                  4
social skills. In contrast, there is no effect on teachers below the median in the cohort-specific
skill distribution. Together these results suggest that school competition and flexible wages
lead to a more market-based wage setting, with a stronger association between local market
conditions and teacher aptitude on the one hand and teacher compensation on the other.
    The rest of the paper outlines as follows: Section 2 provides a background and related
literature, Section 3 describes the data, and Section 4 describes the reform generating the
variation exploited in the paper and the setting of teacher wages in Sweden. Section 5
provides the results and Section 6 concludes.



2     Background and Related Literature

Private school entry could change the wages and mobility of teachers through several chan-
nels. On the supply side, a recent strand of the literature emphasize the relevance of monop-
sony effects in any labour market where employers have some market power over workers, for
example due to imperfect information, search frictions or firm specific skills (see Boal and
Ransom, 1997 and Ashenfelter et al., 2010 for literature reviews on monopsony). Monopsony
effects have long been a concern in the labour market for teachers, as the limited number
of schools in a geographic area as well as the occupation-specific skills may allow schools to
pay lower wages than in a fully competitive market.4
    Recent estimates of teachers’ labour supply elasticity support that teachers have low
mobility, although these studies do not provide evidence of whether or to what extent schools
actually exploit their market power to lower wages (Falch, 2010, Ransom and Sims, 2010).
Another strand of the literature based on cross-sectional evidence shows that areas with more
private schools have higher public school teacher salaries (Vedder and Hall, 2000, Medcalfe
and Thornton, 2006). However, given the inherent difficulties of isolating the impact of
competition from other sources of regional wage differentials mentioned earlier, it is unclear
whether these studies render the true association between school competition and wages.
   4 In addition, teachers are often secondary wage earners in the household, which may further limit their mobility. For

example, Boyd et al., (2005) examine teacher mobility in the US and find that teachers delineate their job search to relatively
small geographic areas close to where they grew up.



                                                              5
Market-based reforms implemented in several countries improve the scope for credible
identification of competition effects in the teacher labour market as they allow researchers
to focus on changes in market competition over time. In the dynamic monopsony models
of Burdett and Mortensen (1998), the extent of employer’s market power depend on the
rate at which job opportunities arrive to workers. The lower the arrival rate of job offers,
the more market power employers will have. As the number of employers increase through
private school entry, the expanded job opportunities for teachers is thus expected to lessen
the monopsonistic power of incumbent schools as competition forces them to raise wages to
avoid teacher turnover.
   Because the threat of separation is more powerful for teachers with low job switching
costs, the wage effects from competition are expected to be more pronounced among high
mobility teachers. Wage competition is also assumed to be stronger for teachers in difficult-
to-staff areas, as these are likely to receive more job offers from entering private schools.
Manning (2003) proposes to use the fraction of new hires from non-employment to proxy
for labour market tightness in different segments of the labour market. If the proportion
of new hires from non-employment is low, workers are assumed to be more mobile. If new
hires come primarily from non-employment, this indicates little mobility across employers.
Calculating the fraction hired from non-employment in my data reveals that while there is
an ample supply of teachers in social science, there is a greater need for teachers in math
and science and vocational subjects.
   In addition, school competition may change the overall demand for certain teacher char-
acteristics. As suggested by Hoxby (2002), the increased competition over students may
raise the incentives for all schools to retain and attract the most effective teachers. Since
the Swedish independent private schools compete with the traditional public schools over
funding on a per-student basis, schools should face strong incentives to hire and retain teach-
ers that are valued by students. Even if existing work has found it difficult to pinpoint the
characteristics associated with teacher quality more competition has been associated with
schools valuing teachers’ effort, independence, math and science skills and the quality of their


                                               6
college education (Hoxby, 2002). Previous work also suggests that private schools demand
different teacher characteristics than public schools, such as more inexperienced teachers
and teachers with less formal qualifications than traditional public schools (Podgursky and
Ballau, 2001).
    Private schools’ demand for certain teachers as well as public school teacher’s willingness
to teach in private schools determine how private school entry affects teacher turnover in
public schools. This will however also depend on public schools ability to compete with wages.
Kirabo Jackson and Cowan (2009) show that charter school entry in the US led to slight
declines in teacher quality and higher salaries in difficult-to-staff traditional public schools.
There was, however, no significant impact on the distribution of wages with respect to teacher
value added, certification status or subject area skills. Though this evidence is compelling,
one limitation of their setting is that in the US, fixed teacher credentials determine teacher
pay to a large extent, which limits schools’ ability to respond to local market conditions. An
important contribution of this paper is that I examine the consequences of school competition
in a labour market where local administrators have substantial room to adjust wages at the
individual level tailored to the needs of recruiting and retaining certain employees.



3    Data

This section provides the key features of the data and describes the variables used to measure
competition and local labour markets. A more detailed description of the different data
sources and the construction of the data is provided in Appendix A.
    The data used to examine the relationship between private school entry and wages contain
all high school teachers in Sweden in the years 1991 to 2006. Individuals with non-teaching
appointments, such as study counselors, are excluded from the sample. Apart from standard
background and regional characteristics, the data include information on whether the indi-
vidual is certified to be a teacher, his/her individual field of specialization, annual income
and monthly wages adjusted to full time. From 1995 onward, it also contains unique school



                                               7
identifiers for the school in which the teacher is employed.
    While most of the variables are available for the universe of teachers, wages are available
for all teachers employed in public schools and for a sample of privately employed teachers.
Because part of the analysis will rely on within-teacher variation in competition from private
schools, only private school teachers who appear in the sample twice or more will help to
identify the coefficient of interest. To handle this, I impute the log monthly wage for private
school teachers who are not sampled in a given year. This is possible because the data
contain annual income for all workers, which can be used to recover information on wages
for teachers in the private sector. I will check the sensitivity of the results using the weights
contained in the data. However, it should be emphasized that for public school teachers, who
constitute the great majority of the teaching pool, wages are available for the full working
population.


3.1      School Competition and Local Labour Markets

I use Statistics Sweden’s definition of local labour market regions (LLMs) to define the
market in which schools compete for labour. These are based on commuting distance and
seem to capture the teacher’s true labour market quite well; 88 % of all teachers in the
sample work in their residential local labour market. As a sensitivity check, I also consider
alternative geographical boundaries of the local labour market.5
    The main competition measure will simply be the share of private high school teach-
ers in a given local labour market and year.6 To focus on the impact of the expansion of
the government funded private schools introduced by the voucher reform, I disregard vari-
ation in competition due to openings/closings of other public schools.7 I will however also
show experiments with a Herfindahl index, where I exploit changes in market concentration
generated by the entry of private independent high schools.
  5 There  are 109 local labour markets in Sweden, which have 2.6 municipalities on average. Figures A1 and A2 provide a map.
  6 An  alternative available measure would be to use the share of private high schools in the local area. However, because
private schools are systematically smaller than public schools (see Figure 1), this definition would lead me to understate the
impact of competition. Another common measure is to use the share of high school students attending private schools in the
local labour market. Because these data are not available for the study period of interest in this paper I will focus on the
measure described in the main text.
   7 While competition effects between public employers are interesting, they are not the primary focus of this paper.




                                                              8
3.2     Teacher’s Cognitive and Social Skills

Apart from standard demographic characteristics, the teacher register can also be linked to
information on the cognitive and non-cognitive skills for a large part of the male population.
The measures are obtained from the military enlistment, where comparable data are available
for cohorts born between 1951 and 1980. In these cohorts almost all males went through the
draft procedure at age 18 or 19.
    The cognitive tests provide an evaluation of cognitive ability based on several subtests of
logical, verbal and spatial abilities and are similar to the AFQT in the US. Individuals are
graded on a 1-9 scale, which I use to construct a percentile ranking within each cohort of
teachers.
    The non-cognitive test scores are based on a standardized interview with a certified psy-
chologist, with the objective to evaluate the conscript’s ability to succeed in the military.
The personality traits evaluated in the draft procedure are psychological endurance, emo-
tional stability, the ability to take initiative, social outgoingness, sense of responsibility and
ease of adjusting to a military environment. The motivation for doing the military service
is not a factor to be evaluated. Just as for the cognitive tests, the individuals are scored on
a scale from 1-9 and ranked by percentile within each cohort.
    An advantage with these measures compared to, for example, value-added measures of
teacher quality is that the tests are taken before individuals select into the teaching profession
and thus do not rely on assumptions about the matching process of students to teachers.8
Previous research shows that both the cognitive and non-cognitive ability measures are
strongly related to labour market outcomes, such as future wages and earnings (Lindqvist
and Vestman, 2011). In the population of high school teachers used in this paper, the
estimated wage-test score relationship appears to be approximately linear (not in paper).
Teachers’ results on the military tests have also been associated with student outcomes at
the compulsory level (Grönqvist and Vlachos, 2008). Together, these findings suggest that
   8 See Rothstein, 2010, for a critical evaluation of value-added models, and Lindqvist and Vestman, 2011, for a more detailed

description of the ability test scores. It should be noted that during this time period, it was not possible to avoid the military
service by scoring low on the enlistment test. In contrast, there were strong incentives to obtain a high score, as the decision
about the type of military service was based on the conscripts performance.


                                                                9
the test scores capture teaching skills that parents and students care about.
     The average cognitive and social ability test scores separately by field suggest that there is
variation in the average skills across subjects; math and science teachers have higher cognitive
and non-cognitive test scores than the rest of the teachers (Table B1). Moreover, comparing
teachers to the college-educated population in non-teaching professions supports the notion
that teachers are disproportionally drawn from the lower parts of the skill distribution.9
     It is not clear a priori whether cognitive or social teaching skills are relatively better pre-
dictors of student achievement. Grönqvist and Vlachos (2008) show that high-performing
students benefit from having teachers with high cognitive ability, whereas low-aptitude stu-
dents are better off with teachers with non-cognitive skills. Moreover, being a good teacher
in one field may not require the same skills as being a good teacher in another. In fact, the
results will later suggest that the returns to cognitive and social skills differ depending on a
teacher’s field of specialization.



4         Institutional Framework

The voucher reform, implemented in 1994, requires local governments, who run the public
schools, to provide private schools with funding on a per-student basis. Importantly, local
governments have limited possibilities to influence the entry of private schools in their mu-
nicipality, as entry is approved at the national level by the National Agency of Education
(NAE). To qualify for public funding, private schools must follow the same rules for en-
rollment as public schools, which means that they must be tuition-free and admit students
based on grades. Besides this, the requirements to receive funding are fairly lax. There are,
for example, no regulations on ownership structure and schools are operated by religious,
non-profit cooperatives and for-profit corporations.10
     Municipalities receive block grants from the central government to be spent on schooling
    9 The college-educated population outside teaching score on average 6.65 (5.78) on the cognitive (non-cognitive) tests.
  10 In the immediate aftermath of the reform, private schools were mainly run by non-profit organizations offering special
profiles. Schools could charge tuition up to an amount considered reasonable by the NAE until 1997, although few schools did
in reality do so. After this initial stage, the growth in private schooling has mainly been driven by independent schools with a
general profile, often run by for-profit companies (Skolverket).




                                                              10
but there are no ear-marked money for schools. Consequently, there is scope for differences in
expenditures on public schools across municipalities (Björklund et al., 2005). A key feature
of the Swedish labour market for teachers is that wages are determined at the local level,
through negotiations between the teacher and the principal. The individualized pay regime
came into place in 1996 through an agreement between the employer’s organization and
the teacher labour unions. Prior to this, salaries were largely determined by fixed teacher
credentials based on the type of work and experience, although local deviations were common
when faced with, for instance, teacher shortages.
   The intention of the reform was to give employers more discretion over wages to reward
teacher quality and effort. However, the conditions of teacher pay and employment are
governed by central agreements between the teacher unions and the employer’s organization.
During the period under study, two agreements have been in place. The first was a five
year agreement of a 10 % increase in teacher pay collectively over the five-year period with
guaranteed minimum salaries after one and five years of employment. The second agreement,
which came in place in 2000, entailed a minimum 20 % increase in total for teachers in the
country as a whole over the five year period but the five year guarantee was relaxed.
   Since the agreement applies to all teachers in the country, it allows for variation in local
wage policies. In the second agreement it was in fact explicitly stated that salaries should
be linked to local governments objectives and needs to recruit and retain effective teachers,
with consideration to budget constraints (Strath, 2004). The pay compression induced by
the agreed minimum salaries could lessen schools monopsony power over workers in the lower
part of the wage distribution and hence mute the importance of market competition in these
segments of the teacher labour market.
   Quantitative evidence suggests that the move to individualized pay had limited impact on
the overall wage dispersion among high school teachers (Söderström, 2006). There are several
possible explanations for this. First, there were already deviations from the wage scales before
1996; the labour union of the majority of high school teachers (Lärarnas Riksförbund) had,
in fact, already accepted individualized wage setting in 1992 (Söderström, 2006). Because


                                              11
the wage scales had a steep age-earnings profile in the old regime, wage increases in the
lower parts of the age distribution could produce a more compressed wage structure than
before. Interviews with single principals indeed highlight that teachers entering the profession
have benefitted most from the market-based wages (Skolledningsnytt 06/2004). It is also
possible that schools’ incentives to introduce individualized pay were too weak in a non-
competitive environment. To understand the full impact of individualized wage setting, it
thus seems important to examine how the enforcement of localized wages interacts with the
competitiveness of the labour market.
    It should be mentioned that the market based reforms in the Swedish education sector
where implemented during a severe economic downturn in the first half of the 1990s, which
had a significant impact on the financial resources in local municipalities. The high unem-
ployment rates and the need to cut expenditures are likely to have increased the incentives
and ability for local governments to possess monopsony power over teachers, which may
increase the economic relevance of competitive forces in the teachers’ labour market studied
in this paper.
    The regional variation in private school openings is illustrated in Figure 2, which shows
the local labour market specific changes in privatization between 1991 and 2006. It is clear
that local labour markets had very different levels of private school penetration during the
study period. Whereas some labour markets experienced increases in the share of private
school teachers with up to 30 percentage points; in some locations, there had still been no
entry of private schools in 2006. The empirical strategy uses the within- and cross-regional
expansion of private schools along with several robustness checks to identify the effect of
                                                       11
school competition on teachers’ wages.
  11 The private schools operating in the market prior to 1994 were boarding schools, schools for students with special needs or

religious schools. A few of these received state funding, although not on a per-student basis.




                                                              12
Fig. 2: Local Changes in Private Entry 1991-2006


5    Results on Hiring Patterns

The mobility of teachers between public and private schools depends on several factors. On
the supply side, certain teachers may find it particularly attractive to teach in private schools,
where working conditions could differ from public schools. On the demand side, previous
work suggests that private schools prefer teachers with different characteristics than those
preferred by public schools, such as less experience and formal qualifications. How private
school entry influences the hiring and mobility patterns of teachers in addition depends on
how public schools’ respond to the wage competition in order to attract and retain teachers
in the public sector.
    Table 1 shows the characteristics of teachers hired by public and private schools respec-
tively. The data are longitudinal with unique identification numbers for workers and firms
from 1995 and onwards, hence, all teachers can be followed over time as well as across schools
and alternative employers. New hires are defined as those not observed in the same school in
the preceding three years, which restricts the sample period for this analysis is 1998 to 2006.
In line with previous work, I find that private schools hire younger and fewer certified teach-
ers than public schools. Private schools also hire more frequently from other private schools,
other industries and from non-employment. When recruiting individuals from non-teaching


                                                 13
professions, private schools mainly attract workers from the business and retail industries.
Public schools in contrast, mainly hire workers from other public sector industries.12

                         Table 1: Teachers Hired by Public and Private Schools 1998-2006
Hiring school is:                                                                   Private                                Public
Hire characteristics:
Age                                                                                   38.6                                   41.7
Certified                                                                             0.43                                   0.58
Female                                                                                0.49                                   0.51
Cognitive Ability (males)                                                             0.42                                   0.39
Social Ability (males)                                                                0.42                                   0.40
Fraction hired from:
Public high schools                                                                   0.14                                   0.24
Private high schools                                                                  0.08                                   0.01
Other education levels                                                                0.36                                   0.39
Other industries                                                                      0.24                                   0.20
Non-employment                                                                        0.18                                   0.16
Fraction from other industries:
Manufacturing                                                                         0.10                                   0.09
Construction                                                                          0.02                                   0.05
Wholesale and retail sale                                                             0.13                                   0.11
Hotels and restaurants                                                                0.05                                   0.06
Transport, storage and communication                                                  0.05                                   0.05
Financial intermediation                                                              0.01                                   0.01
Real estate, renting and business activity                                            0.23                                   0.15
Public administration and defense                                                     0.06                                   0.09
Health and social services                                                            0.15                                   0.20
Other community, social and personal services                                         0.18                                   0.17
Notes. New hires are defined as workers not receiving compensation from their current school in any of the three
preceding years. This restricts the sample period to 1998-2006. Industries that employ less than one % of the total
hires (”Agriculture, hunting and forestry”, ”Mining and quarrying” and ”Electricity, gas and water supply”) are not
shown in the table.


    Björklund et al., (2005) show that the probability to leave a public school for a private
increased proportionally with the expansion of the private schools during the 1990s. (Table
1 suggested that, on average, 14 % of the private school teachers come from public high
schools.) Next, I look at the characteristics of the teachers who leave public for private
schools. An advantage of the data used for this study is that they contain all teachers
employed in each school, which allows me to compare teachers who leave for a private high
school in comparison to all his/her co-workers who remain in the traditional public school.
In practice, I estimate linear probability models of the following type:
  12 Public and private schools are subject to the same rules regarding recruitments in that all schools are obliged to hire teachers

who have a degree (i.e. teacher certification) in the teaching they will be undertaking according to the Swedish Education Act
(1985:1100). Exceptions can be made if people with the required training are not available.




                                                                 14
Hipt = α + Xipt β + θpt + ipt .                  (1)



where Hipt is a dummy taking a value of one if teacher i in public school p in year t switched
to a private school, Xipt is a vector of teacher characteristics (age, gender, certification status
and field of education) and θpt a vector of school year dummies (i.e., a fixed effect for each
set of co-workers for public teacher i). Because the model includes school×year fixed effects,
it accounts for all school characteristics that could influence the decision to leave a public
school in a given year, such as teacher and student composition and the regional location of
the public school.
   The estimated β’s presented in column (1) in Table 2 suggest that the probability for a
teacher to leave a public school for a private, decreases with age and is significantly lower
for certified teachers.13 Among certified teachers, there is a tendency for private schools
to hire public school teachers in math and social science than in other subjects, although
the differences are not statistically significant. Column (2) includes the teachers’ cognitive
and social skills. Notably, these results indicate that teachers moving from public to private
schools have above average cognitive skills than those who remain in public schools. Part
of this effect seems to be explained by systematic skill differences related to teacher’s field,
although the pattern remains the same also when including school×year×field fixed effects
(column 3).
   For comparison, columns (4)-(9) display the same results for teachers who leave a public
school for another public school. I distinguish between destination schools located in the
same municipality (columns 4-6) and those located in a different municipality (columns 7-9),
as schools located in different municipalities are more likely to compete over teachers than
public schools operated by the same decision-making unit.
   The results suggest that within a municipality, mobility is substantially higher for certified
teachers, which may partly be explained by the involuntary reshuffling of teachers between
 13 The   fraction of teachers switching schools is 8 %.



                                                           15
public schools. Cross-municipality mobility is more similar to teacher mobility between
public and private schools. However, one distinct difference can be noted; whereas private
schools hire teachers from the upper part of public school teachers’ skill distribution, public
schools appear to recruit teachers with below average cognitive and social skills. These
differences in mobility patterns suggests that competition may have increased the overall
demand for teacher aptitude in the labour market.



6     Results on Wages

This section examines the relationship between private school competition and teacher wages.
Before turning to the econometric specification and the estimation results, I provide a brief
description of the variation in school competition exploited in the empirical analysis and
illustrate the raw wage differences between teachers in more/less competitive labour markets.


6.1    Empirical Strategy

To estimate the impact of school competition on teacher wages, I exploit the local variation in
private school expansion induced by the voucher reform using individual data. The empirical
specification is given by:


                       log(wilt ) = β1 Plt + µl + µt + ψl · t + β2 Xilt + ilt .           (2)



where wilt is the wage for teacher i in local labour market l in time period t; Plt is the
continuous measure of the degree of competition in the local labour market, Xilt is a vector
containing observable teacher characteristics (gender, age, educational attainment and certi-
fication status) as well as the number of pupils of high school age in the local labour market
where the teacher is employed, µt and µl are year and local labour market dummies, ψl ·t are
local labour market-specific time trends and ilt is the error term.
    This baseline specification takes into account many of the confounding factors that could


                                                  16
Table 2: Teacher Mobility from Public Schools 1998-2006
                                                       Private                               Public   (same employer)                   Public (different employer)
                                            (1)           (2)             (3)            (4)              (5)         (6)            (7)             (8)            (9)
                                            All         Males           Males            All            Males       Males            All           Males          Males
     Age                               -0.021***      -0.030***       -0.023***      -0.027***          -0.005      -0.011       -0.061***       -0.076***      -0.079***
                                         (0.002)       (0.005)         (0.006)        (0.003)          (0.009)     (0.014)        (0.003)         (0.009)        (0.014)
     Female                               -0.006                                       -0.077                     -0.150***
                                         (0.026)                                      (0.071)                      (0.044)
     Certified                          -0.100**         -0.113                       0.190**            0.188                    -0.114*         0.123
                                         (0.049)        (0.099)                       (0.095)          (0.165)                    (0.070)        (0.156)
     ×Math and science                     0.078          0.061                        -0.139            0.058                    0.170*          0.255
                                         (0.053)        (0.659)                       (0.118)          (0.257)                    (0.094)        (0.233)
     ×Social science                       0.089          0.124                        -0.029            0.120                    0.228**         0.241
                                         (0.470)        (0.110)                       (0.100)          (0.231)                    (0.073)        (0.206)
     ×Vocational subjects                 -0.074         -0.057                        -0.063           -0.041                     0.083          0.072




17
                                         (0.329)        (0.066)                       (0.090)          (0.177)                    (0.052)        (0.145)
     Ability
     Cognitive                                         0.221**           0.071                           0.008     –0.286                         -0.230          -0.289
                                                       (0.109)         (0.154)                         (0.220)     (0.336)                       (0.187)         (0.320)
     Social                                             -0.087          -0.077                          -0.231      -0.193                       -0.312*           -241
                                                       (0.117)         (0.169)                         (0.216)     (0.335)                       (0.176)         (0.280)
     Mean of dependent variable          0.333          0.322            0.164          2.30             2.91        2.225         1.029           1.087           0.963
     Observations                       217,887        49,345          31,324         217,887          49,345      34,174         257,894        59,791          40,010
     R2                                  0.097          0.097            0.256         0.537            0.519        0.603         0.018           0.03             0.09
     School×year dummies                  yes             yes             yes           yes               yes         yes           yes             yes             yes
     School×year×field dummies            no              no              yes            no               no          yes           no              no              yes
     Notes. *,** and *** denote statistical significance at the 10, 5 and 1 % levels, respectively. Standard errors are robust for clustering at the school level are shown
     in parentheses. The model is a linear probability model where the dependent variable is an indicator variable taking the value one if the teacher left the public
     school for a private/public destination school. The dependent variable has been scaled by 100, hence the mean probability for a public school teacher to leave for
     a private school is approximately 0.3 %. The sample in columns (3), (6) and (9) is restricted to males born between 1951 and 1981.
generate a spurious relationship between competition and wages; the covariates in Xilt ac-
count for compositional changes in the observed characteristics of the teaching pool and for
changes in the local demand for schooling due to cohort size fluctuations; the year dummies
control for smoothly evolving factors such as business-cycle effects and long-term national
trends and the local labour market dummies account for permanent spatial differences in
economic outcomes. Importantly, the long time period allows me to eliminate local linear
labour market-specific trends, which implies that the parameter of interest is identified from
the residual variation in each labour market around its own linear time trend.
    One potential concern is that teachers may sort into labour markets with more or less
competition based on unobserved characteristics. If this is the case, β1 may capture both
direct effects of competition for incumbent teachers as well as compositional changes in
the teaching pool. An advantage of using longitudinal data is that I am able to control
for such compositional changes by including teacher fixed effects. To this end, I augment
equation (2) with a vector of teacher-specific indicators, µi . Since this specification relies
on variation in teachers’ exposure to local school competition it will identify the impact of
school competition for incumbent teachers only.
    The parameter of interest is β1 , which captures the full impact of competition in the local
labour market averaged across all teachers, both public and private. I will however also
estimate (2) for public school teachers separately, in order to assess how private school entry
affect salaries of those who remain in the public schools.
    Apart from being associated with teacher mobility, privatization could also affect the
composition of students remaining in public schools. For example, if private schools cream-
skim, the estimated effects could capture wage compensation for increased segregation in
public schools rather than changes in market power (Epple and Romano, 1998). Because
the private schools cannot charge tuition and must follow the same admission rules as pub-
lic schools, there is probably less room for such selectivity in the Swedish system than in
other settings.14 Moreover, student mobility would only affect average wages in the local
  14 MacLeod and Urquiola (2009) show that competition via non-selective, for-profit schools leads to less stratification compared

to a system with selective schools.



                                                               18
labour market if teacher wages increase disproportionally to the share of low-ability students.
Although it is impossible to fully rule this explanation, it is difficult to reconcile with the
distinct heterogeneous patterns w.r.t teacher characteristics shown later in the study.
   The assumption maintained for identification is always that the regional private school
expansion is uncorrelated with the error term once I have conditioned on all covariates
included in (2). The main source of heterogeneity that is not controlled for and that may
generate a spurious relationship between school competition and wages is the presence of
local and non-linear trends in unobserved determinants of wages that are correlated with
the degree of private school competition. Higher economic growth in a region could, for
example, attract parents with a higher demand for private schooling, in which case regions
with private school openings may even in its absence have experienced increasing wages. I
discuss this and similar threats to identification in greater detail in Section 6.4, where I also
present a number of robustness checks to validate my findings.


6.2   Descriptive Patterns: Local Private Entry and Wages

Figure 2 illustrated the local changes in private high school entry during the study period.
Geographically, private schools opened in all parts of Sweden, although most of the expansion
took place in the population dense areas in the southern parts (Figure B1 and B2). Table
3 reports the partial correlations between local (municipal) characteristics and private high
school entry. The first column shows that localities with higher income, population density
and a right wing local majority in the pre-reform period were particularly attractive targets
for private school entry. The second column displays which local characteristics in the post-
reform period that predict private high school entry in the subsequent (t+1) year. The only
variable that is significant is population size. However, the signs on the other coefficients
indicate that private schools seem to locate in municipalities with a higher fraction of college
educated, native and high income individuals and higher teacher salaries. The current level
of private schools is moreover negatively related to the location decision, which suggests that
private schools choose to enter markets with little competition from other private schools.


                                               19
Table 3: Private Entry and Local Characteristics
Dependent variable                                   Private school share                           Private entryt+1
                                                             2006                                      1994-2006
Fraction college educated                                    0.017                                         0.152
                                                           (0.486)                                       (0.215)
Fraction immigrants                                         -0.087                                        -0.062
                                                           (0.220)                                       (0.134)
Mean log income                                            0.475**                                        -0.066
                                                           (0.217)                                       (0.090)
Right wing majority                                        0.101**                                         0.037
                                                           (0.037)                                       (0.023)
Log Population                                             0.082**                                      0.169***
                                                           (0.030)                                       (0.020)
Private school share                                                                                      -0.031
                                                                                                         (0.061)
Log teacher wages                                                                                          0.139
                                                                                                         (0.161)
Observations                                                 546                                           2,359
Mean dependent variable                                     0.076                                          0.062
Notes. The dependent variable in column (1) is the fraction of private high school teachers in 2006 and the explanatory
variables are measured in 1993 (year prior to the voucher reform). The model in the second column is a linear
probability model where the dependent variable is a dummy taking the value of one if a private high school opened
in the municipality in the following year. The estimates in both columns are weighted according to the number of
residents in the municipality used to calculate the means. *,** and *** denote statistical significance at the 10, 5
and 1 % levels, respectively. Standard errors are robust for clustering at the school level are shown in parentheses.


   For my empirical strategy to be valid, it is crucial that private school entry is not system-
atically related to trends in other factors that determine local wages. As a first assessment of
the validity of this assumption, I look at the unrestricted time series of wages in local labour
markets separated at the median amount of private school expansion during the study pe-
riod. As illustrated by Figures 3 and 4, there is no clear trend prior to 1994, whereas wages
start to diverge after the reform in favor of teachers in more competitive labour markets.
Unless this pattern is explained by unobserved time-varying differences between more or less
competitive markets the figure clearly suggests that private competition has a positive effect
on teachers’ wages. For non-teaching occupations with similar qualification requirements as
teachers, such as pre-school teaches and nurses, there seems however to be no difference in
wages between regions with more or less private high school entry. This suggests that the
pattern below is probably not an outcome of general trends common to these workers in the
same local labour market (Figure B.3 and B.4).




                                                          20
Fig. 3: Median Wages for Teachers in Treated and Comparison Local Labour Markets




 Fig. 4: Relative Median Wages for Teachers Employed in Treated and Comparison Local Labour Markets


6.3   Estimation Results

The estimate in the first column in Table 4 shows the baseline effect from estimation of equa-
tion (2), which relates teacher wages to the private school share in the assigned local labour
market. The dependent variable is the individual log monthly wage and all specifications
include individual wage controls, the log of the number of individuals in high school age,


                                                 21
year dummies, local labour market fixed effects and local labour market linear trends.15
    The estimated effect indicates that private entry has a positive impact on teacher wages,
although the effect is not statistically different from zero and it is rather small. Columns (2)
and (3) continue to show the differential impact between entering and incumbent teachers,
where entering teachers are defined as those who are not observed in the teacher register in
any of the five preceding years. As previously discussed, the impact of competition is likely
to be higher among teachers who are entering the profession than among incumbents, due to
the higher mobility in this group. Consistent with this, I find that the effect is significantly
larger among new teachers; those who enter the most competitive areas (where the private
share is one third) receive 3 % higher wages than those who enter labour markets without any
competition from private schools. This difference is roughly comparable to one additional
year of college education, and somewhat smaller than the difference in wages between a
certified and non-certified teacher who enter the teaching profession (5 %). It moreover
corresponds to a monthly wage difference of roughly 550 SEK/Euro 50/USD 85.
    Columns (4) and (5) present the results from teacher fixed effects models. As argued
above, it is possible that the main effects capture both the direct impact of competition
and compositional changes in the teachers’ labour pool. The estimates increase and are
more precisely estimated when teacher heterogeneity is controlled for, suggesting a negative
sorting of teachers into more competitive markets.16 The estimated effect of 5 % implies that
teachers receive on average 1.5 % higher monthly salary in areas with the highest realized
levels of private school entry.
    Finally, the last column shows that the estimated effect remains approximately the same
when the sample is restricted to public school teachers only, suggesting that public schools
respond to private school competition by raising the wages for incumbent teachers.
  15 To conserve space I do not report the estimates of the control variables but it should be noted that all of these enter with

expected signs; wages are higher for males than for females, increases with age and level of education and are higher for certified
teachers. Weighting the sample instead of using imputed wages does not alter any of the results although the estimates are less
precise. These results are available upon request.
  16 Note that identification in the teacher fixed effects specification comes from both within and between local labour market

variation in school competition. However, including teacher×LLM fixed effects produces a similar estimate which suggests that
changes in the competition measure is not driven by teacher mobility from low to high private school districts.The estimate
using the weights contained in the data instead of the imputed wages yields the estimate (0.046 [0.022]).




                                                                22
Table 4: Baseline Estimates
                                                     Dependent variable: log(monthly wage)
Sample:                     All teachers,      Entering teachers,                Incumbent teachers
                            All schools           All schools        All schools    All schools   Public schools
                                 (1)                   (2)               (3)            (4)            (5)
Private share                   0.038               0.111**             0.029         0.050**       0.049**
                               (0.023)              (0.050)            (0.023)        (0.022)        (0.021)
Observations                  351,875                31,389            320,486        320,389       308,455
R2                              0.731                0.661              0.735          0.903          0.903
LLM fixed effects                yes                  yes                yes            yes            yes
LLM linear trends                yes                  yes                yes            yes            yes
Teacher fixed effects            no                    no                 no            yes            yes
Notes. Each column represents a separate regression. *,** and *** denote statistical significance at the 10, 5 and 1 %
levels, respectively. Standard errors are robust for clustering at the local labour market level are shown in parentheses.
The sample covers 1991 to 2006. In addition to the fixed effects indicated by the table, all regressions control for
year fixed effects, a dummy indicating whether the individual is a certified teacher and a dummy indicating whether
the wage is imputed or not, and the log of the number of student in high school age in the given labour market and
year. The individual controls include gender, age, age2 and education dummies (6 bins).


6.4     Robustness Specifications

An alternative measure of competition to the share of private high school teachers would
be to use a measure that captures the overall competitiveness of the local labour market.
Table 5 shows results using a Herfindahl index score, which is a common measure of market
concentration. The index is calculated as the sum of the squared market shares held by each
competitor in the market and it takes values from 0 to 1, where a higher value corresponds
to market characterized by lower concentration of employers, and hence less competition.
Because it is not clear whether/how public schools within the same municipality compete
over teachers, I computed two alternative index scores. The first measures competition by the
concentration of teachers into all schools in the local labour market and the second measures
competition by the concentration of teachers into decision-making units (local districts and
private schools).17
    To continue focusing on competition from private school entry, I use the private share as
an instrument for the Herfindahl index score, which allows me to compare the estimates to
those in Table 4. Th first stage regression displayed in the above panel of Table 5, suggests
that labour markets with the highest levels of private entry, has a 0.24 lower concentration
  17 Hanushek and Rivkin (2003) use a similar strategy. Note that the index treating all schools as independent competitors

naturally generates lower concentration (higher competition) since labour markets without private schools and only one districts
but many schools within the district would will have low concentration according the first measure, but high according to the
second.



                                                              23
Table 5: Alternative Competition Measures
                                                                    (1)                                              (2)
                                                                                        First stage
Dependent variable                                          H-index schools                                  H-index districts
Privateshare (teachers)                                        -0.620***                                        -0.797***
                                                                (0.005)                                          (0.003)
Mean dependent variable                                          0.255                                            0.512
                                                                                       Second stage
H-index schools                                                 -0.021**
                                                                 (0.009)
H-index employers                                                                                       -0.064***
                                                                                                          (0.012)
Observations                                                351,875                                      351,875
R2                                                           0.730                                         0.730
LLM fixed effects                                             yes                                           yes
LLM linear trends                                             yes                                           yes
Teacher fixed effects                                          no                                           no
Notes. The table report estimates from a two stage least squares model where the share of private high school
teachers is used as an instrument for the overall competitiveness of the local labour market. Overall competition
is measured by the Herfindahl index score, which takes the sum of the squares of the market shares held by each
competitor. In the first column, a ”competitor” is defined as any school in the local labour market. In the second
column, a ”competitor” is defined as an independent employer, which can be either a public school district or a
private independent school. Higher values of the Herfindahl index implies less competition. Except for the controls
indicated in the table, the model includes the same controls as Table 4, the sample covers 1991 to 2006, *,** and ***
denote statistical significance at the 10, 5 and 1 % levels, respectively. Standard errors are robust for clustering on
the local labour market.


index. According to the second column, which displays the estimates based on the Herfindahl
decision-unit index, this translates into a 1.5 % higher wage. The magnitude is comparable
to that reported in column (3) of Table 4, although more precisely estimated. Using the
index based on all schools produces smaller estimates, which indicates that the decision-unit
concentration index is a more relevant measure in capturing the competitiveness of the local
labour market.18
    A causal interpretation of the estimates in Table 4 relies on the assumption that the
expansion of private schools is uncorrelated with trends in unobserved determinants of wages
not captured by the local linear trends. This would be violated if for example private schools
choose to locate in areas with higher demand for private schooling. Interviews with private
secondary schools indicate that attitudes toward privatization are an important factor for
  18 I also estimated the model using the private school share as instrument instead of the share of private high school teachers.

As indicated by Figure 1 in the introduction, this generated a much lower estimate on the association between private entry and
market concentration, which is explained by the fact that private schools are typically smaller than public schools. The estimate
from the second stage however, was very similar to column (2) of Table 5. In addition to the exercise with the Herfindahl index,
I also tried interacting the private share in eq. (4) with the absolute number of private high schools in the area but found no
difference depending on the number of schools entering the market.




                                                               24
the location decision (Böhlmark and Lindahl, 2008). In addition, if the fixed salary schemes
were binding prior to the wage bargaining legislation in 1996, the results may simply capture
a spurious relationship between private school competition and wages showing only after the
removal of the wage scales.
    To address this main concern I examine whether wages of other occupational groups are
also affected when the share of private high schools increases. The results provided earlier
showed that the inflow of private high schools did not impose any increased competition
for preschool teachers (only 0.8 % of the total hires come from preschools). These teachers
should therefore be unaffected by the variation generated by the voucher reform, unless high
school entry is correlated with trends in other factors that influence public sector wages,
such as public spending, demands for private schooling, area amenities or labour quality.
    The point estimate of the private high school share on pre-school teacher salaries is 0.007
(0.024); hence private school expansion is unrelated to the wage growth of preschool teachers,
which supports the conclusion that the main effect is not driven by local trends in omitted
factors common to teachers at the preschool and high school levels (reported in Table C2).19
    Several additional robustness checks strengthen the interpretation of the main results.
The estimates are somewhat sensitive to omitting the linear trends, suggesting that these
capture unobserved determinants of wages. However, I find no association between future
privatization (t+2) and current wages (Table C1). Unless the baseline model is picking
up spurious effects, future privatization should not affect todays wages, conditional on the
current level of privatization. I also tried restricting the sample period to the post wage
bargaining legislation years (Post 1996), which reduced the coefficient of interest in my
preferred linear trends specification. However, this is not surprising since the local time
trends are likely to absorb much of the variation in private school expansion, which seems to
have evolved gradually in many local labour markets, and in particular those employing many
teachers (the standard errors increase quite a lot when the pre-reform years are discarded).
  19 Pre-school teachers also experienced an increase in private alternatives although much less dramatic. Hanspers and Hensvik
(2010) show that this increase did not affect wages among pre-school teachers. In addition, because the funding of schools is
based on the number of pupils enrolled, any potential negative spill-over effects of wage increases among high-school teachers
are likely to be small.



                                                              25
You can also read
Next part ... Cancel