Abstract

There is growing concern among policy makers over the quality of the teacher workforce in general, and the distribution of effective teachers across schools. The impact of teacher attrition on overall teacher quality will depend on the effectiveness of teachers who leave the profession. Likewise, teacher turnover may alleviate or worsen inequities in the distribution of teachers, depending on which teachers change schools or leave teaching and who replaces them. Using matched student–teacher panel data from the state of Florida, we examine teacher mobility across the distribution of effectiveness (as measured by teacher value added). We find that top-quartile and bottom-quartile teachers exit at a higher rate than do average-quality teachers. Additionally, as the share of peer teachers with more experience, advanced degrees, or professional certification increases, the likelihood of moving within-district decreases. We also find some evidence of assortative matching among teachers—more productive reading/language arts teachers are more likely to stay in teaching if they have more productive peer teachers.

1.  Introduction

Given the central role of teacher quality in determining student achievement,1 there is growing concern over the impact of teacher job change on both the overall level of teacher quality and the distribution of teacher quality across schools. In particular, do the best teachers leave teaching and does teacher mobility within the profession exacerbate differences in educational quality across schools? The answers to these questions have important implications for designing policies to promote student achievement and reduce achievement gaps across students from different racial, ethnic, and economic backgrounds.

The effects of teacher labor market decisions on teacher quality and student achievement are ambiguous, a priori. If high quality teachers possess transferable skills that are valued in other occupations, attrition will tend to erode average teacher quality. On the other hand, attrition may have a positive effect on the average quality of teachers, if relatively less-effective teachers receive little job satisfaction, voluntarily leave the profession, and are replaced by more able teachers. Likewise, the effect of teacher movement between schools on the distribution of teacher quality across schools is not clear ex ante. Inter-school mobility of teachers could exacerbate the divergence in education quality across schools if schools serving disadvantaged populations lose their best teachers to schools serving more advantaged students. It is also possible, however, that switching of schools by teachers has no effect on the distribution of teacher quality across schools and simply enhances the utility of the teachers who move.

A number of previous studies have explored the relationship between various observable teacher qualifications on teacher attrition, including college entrance exam scores, performance on teacher certification exams, and possession of advanced degrees (Podgursky, Monroe, and Watson 2004; Boyd et al. 2005; Imazeki 2005; Feng 2009, 2014). Studies of student achievement find little correlation between these credentials and the impact of teachers on student test scores, however, particularly in elementary and middle school (Betts, Zau, and Rice 2003; Rivkin, Hanushek, and Kain 2005; Clotfelter, Ladd, and Vigdor 2007, 2010; Kane and Staiger 2008; Harris and Sass 2011).

Previous research has highlighted the disparity in qualifications of teachers in schools serving primarily disadvantaged and minority students versus teachers in schools with more advantaged student bodies (Lankford, Loeb, and Wyckoff 2002; Clotfelter, Ladd, and Vigdor 2005; Goldhaber, Choi, and Cramer 2007; Sass et al. 2012; Isenberg et al. 2013). There is also circumstantial evidence that within-profession teacher mobility is contributing to these differences in teacher credentials. Teachers in schools serving primarily disadvantaged students are more likely to transfer to a new school district (Ingersoll 2001; Hanushek, Kain, and Rivkin 2004; Imazeki 2005) and teachers in urban inner-city schools are more likely to migrate away from their schools than teachers in other areas (Ingersoll 2001; Lankford, Loeb, and Wyckoff 2002). Similarly, teachers, particularly white teachers, tend to move away from schools with high percentages of minority students (Hanushek, Kain, and Rivkin 2004; Boyd et al. 2005; Imazeki 2005; Scafidi, Sjoquist, and Stinebrickner 2007; Feng 2009, 2010, 2014).

Given the generally weak relationship between observable teacher characteristics and student achievement, a handful of recent studies have attempted to directly investigate the relationship between teacher job choice and a teacher's contribution to student achievement, or teacher value added. Krieg (2006) analyzes the relationship between teacher attrition and teacher quality for fourth grade teachers in Washington State and finds that increases in teacher value added in math, writing, and listening (but not reading) are associated with reductions in attrition for female teachers. Boyd, Grossman, and Lankford (2008) study the association between teacher value added and teacher mobility for new fourth through eighth grade teachers in New York City. They find that higher value added in math reduces the likelihood of transfers within New York City for first-year elementary school teachers and lowers the probability of exit from public school teaching in New York State for both elementary and middle school first-year teachers. In contrast, no significant relationship between teacher mobility and teacher quality is found for English language arts (ELA) teachers in either elementary or middle school. There are no significant differences in value added by move status for math or ELA teachers after their second or third years of teaching. Hanushek and Rivkin (2010) compare mobility patterns in a single large urban Texas school district for math teachers in grades 4 through 8. Like Boyd, Grossman, and Lankford (2008), they find that first-year teachers who exit the Texas public schools have lower value added than those who stay. They also find, however, that first-year teachers who change schools within the district actually have higher value added than those who stay in their initial school placement. Also, similar to Boyd, Grossman, and Lankford (2008), they generally find no relationship between teacher quality and mobility for second- and third-year teachers. For teachers at the end of their fourth or greater years of teaching, within-district movers, cross-district movers, and those leaving the profession all have lower value-added than do their colleagues who stay in their current school. Boyd et al. (2011) are able to distinguish teacher preferences from school preferences by utilizing data on requests for transfers from teachers in New York City. They find that, holding teacher demographics and pre-service characteristics constant, increases in teacher value added are associated with reductions in the likelihood of requesting a transfer and of actually transferring between schools. Finally, Goldhaber, Gross, and Player (2011) analyze both attrition and inter-school mobility of early-to-mid-career teachers in North Carolina. Unlike previous work, they allow for differential mobility patterns across the value-added distribution.2 Controlling for school context and teacher demographics, teachers from the lowest quintile of the value-added distribution are more likely to change schools within a district, more likely to change school districts, and more likely to leave the public school system, than those from the middle quintile of value added. Similarly, teachers in the highest quintile are less likely to make a within-district move, less likely to make a cross-district move, and less likely to leave public school teaching.

We extend the literature and provide a number of new insights. First, our analysis includes high school teachers as well as elementary and middle school teachers. Second, we uncover a nonlinear relationship between teacher quality and attrition, whereby both the highest- and lowest-quality teachers are more likely to leave public school teaching than teachers in the middle of the quality distribution. Third, we find that campus peer teacher composition plays an important role in influencing the mobility decisions of a teacher. Finally, building on recent work which highlights the importance of teacher peer effects (Jackson and Bruegmann 2009; Jackson 2012), we explore how the average quality of faculty colleagues and the productivity of a teacher relative to her peer teachers affect teacher job choice.

2.  Institutional Background

Public schools in Florida are organized into 67 countywide school districts, ranging in size from Miami-Dade—with 350,000 students—to Jefferson County—with just over 1,000 students—in pre-K through twelfth grade (Florida Department of Education 2011). Although new teachers tend to obtain their first job near where they went to school, initial placements are spread throughout the state, with many working far from the location of their preparation program (Mihaly et al. 2013). New teachers exhibit a high degree of mobility, with 60 percent of teachers leaving their initial school placement within four years and only one-fourth of new teachers remaining in their initial school placement after six years (Feng 2009). Perhaps because of the relatively large geographic size of districts in Florida, inter-district transfers make up a relatively small proportion of teacher transfers—only about one in ten new teachers leaves his first placement to teach in a different district in the first four years of his career. In contrast, about 30 percent transfer to another school in the same district within four years, and about one-third leave the public school system altogether.

Among the 67 countywide school districts in Florida, all but one have collective bargaining agreements that govern personnel matters, including transfer rights. District collective bargaining agreements vary in the restrictiveness of the contract with respect to voluntary transfers, involuntary transfers, and reduction-in-force provisions. They also differ in the seniority provisions regarding each type of job action. Voluntary transfer provisions in collective bargaining agreements establish the criteria for selecting among candidates for open positions. In some districts transfer provisions do not mention seniority, in others it is one among many criteria, and in still others seniority takes priority over all other criteria (Cohen-Vogel, Feng, and Osborne-Lampkin 2013). Frequently, current teachers have preference over new hires. The preference may be as weak as simply giving advance notification of open positions or as strong as requiring district employees be placed into open positions before any new teacher is hired.

3.  Methods

Measuring Teacher Quality

In order to gauge teacher quality, we estimate value-added measures of teacher productivity derived from the following student achievement model:
formula
1

Ait is the achievement level of student i in year t, where achievement is measured by the student's scale score on the FCAT-NRT (Stanford Achievement Test), normalized by grade and year. The vector Xit represents time varying student/family inputs, which include free and reduced-price lunch status, gifted student status, limited English proficiency status, disability status (six indicators for broad groupings of disability types), student mobility across schools within a given year, and both structural moves (e.g., elementary to middle school) and nonstructural moves between school years. Classroom peer characteristics are represented by the vector P–ijmt where the subscript –i denotes students other than individual i in classroom j in school m. These student peer characteristics include class size, the fraction of classroom peers who are female, fraction of classroom peers who are black, average age (in months) of classroom peers, and the fraction of classroom peers who changed schools. The school-level input vector, Smt, includes the administrative experience of the principal, the principal's administrative experience squared, whether the principal is new to the school, and whether the school is in its first year of operation. Time-invariant student and family characteristics are represented by . Observable time-invariant characteristics include a student's gender and race/ethnicity. Teacher quality is captured by a year-specific teacher effect, . is a mean zero random error. The teacher effect represents a teacher's contribution to student achievement, net of prior educational inputs and contemporaneous student, peer, and school influences. Given that student test scores are normalized, the teacher effects are calibrated in standard deviation units.

Numerous variants of equation 1 have been estimated in the literature on teacher quality. Specifications can differ in the assumed persistence of educational inputs (), the method for capturing time-invariant student heterogeneity (), and whether or not teacher effects are measured separately by year or assumed to be constant across all years (i.e., = for all t). Although there is currently no consensus on which value-added model specification is most appropriate for measuring teacher quality, there is growing evidence that models that use student covariates to control for student heterogeneity, and include lagged test scores on the right-hand side to allow for incomplete persistence, are likely to produce relatively unbiased estimates of teacher effects under a range of conditions. Using simulated data, Guarino, Reckase, and Wooldridge (2015) find that such a specification, which they call Dynamic Ordinary Least Squares, produced the most accurate estimates across a wide variety of student grouping and teacher assignment scenarios. In addition, Kane and Staiger (2008) compare differences in pre-experimental value-added estimates of teacher quality with differences in student achievement across teachers under conditions of random assignment. They find that value-added estimates derived from a model with student covariates and partial persistence performed best in predicting student achievement under random assignment, and that one could not reject the null hypothesis that the estimates were unbiased. Further, McCaffrey et al. (2009) document considerable variability in single-year value-added estimates of teacher quality due to sampling error in student test scores.

Given the existing evidence, we focus our analysis on a teacher-quality measure estimated from a value-added model specification that does not constrain the value of to one (partial persistence), uses observable student characteristics (rather than student fixed effects) to control for student heterogeneity, and assumes teacher quality is time-invariant ( = for all t).3 This time-constant measure will mask any changes over time in true teacher performance but provides a less “noisy” signal of teacher quality than single-year estimates. The teacher effect estimates are recentered to have a mean value of zero in each school level (elementary, middle, high). Because there are no school fixed effects in the achievement model, the estimates represent the teacher's effect on student achievement relative to the average teacher in the state at the same school level.4

Estimating the Determinants of Teacher Job Choice

We model a teacher's decision about job change as an individual utility maximizing problem over a number of job choices.5 A teacher will select among a group of jobs based on her individual preferences and the characteristics of the job, including both pecuniary aspects and nonpecuniary components. A teacher will compare the available options and select the job that yields the highest present value of expected utility.

The decision facing a teacher during each time period t is represented by
formula
2
where the subscript kmrt indicates teacher k in school m and district r at time t. The first term, Dkmrt, represents a vector of control variables that have been shown in the literature to be important in influencing a teacher's mobility decision. These include teacher demographics and professional credentials as well as classroom, school, and district characteristics. A teacher's race, gender, and age are included to account for teacher preferences. Interactions between age and gender are included to account for women's reproductive decisions.6 A teacher's experience and education level, professional certification status, and subject-specific certification are all included to reflect human capital investment. A set of subject-area indicators allows for differences in teaching difficulty and outside opportunities. Teachers’ salaries are included to account for the monetary rewards from teaching.7 The effects of nonmonetary working conditions are captured by a variety of classroom, school, and district characteristics such as class size, average student math scores, disciplinary incidents, student racial/ethnic composition, and percent of students receiving free or reduced-price lunch (a proxy for family income). As is standard in discrete time hazard models, the natural log of time (the number of years a teacher has been teaching at her current school) is also included as an explanatory variable.

Qkmrt is an indicator of teacher quality or effectiveness, captured by the value-added measure described above. The value-added teacher quality measure captures a variety of unobserved teacher characteristics that impact teacher productivity and hence labor market decisions, including innate ability, noncognitive skills and pre-service (undergraduate) training. The teacher quality parameter may also proxy for intrinsic psychological rewards from teaching. In order to allow for heterogeneous effects across the distribution of teacher quality we divide teacher value added into four quartiles and include indicators for the top three quartiles in our regression (the bottom quartile is the omitted category).

Campus peer faculty effectiveness and other characteristics are represented by Fkmrt, where the subscript –k denotes teachers other than teacher k in school m. These teacher peer characteristics include value-added peer-teacher quality measures and peer-teacher credentials, such as teaching experience, advanced degree attainment, National Board certification, and professional (nontemporary) state certification. We also include measures of the percent of faculty peers with certification in specific subject areas to allow for different types of peer-human capital to have unequal impacts on the mobility of individual teachers.

A teacher maximizes his utility by selecting the option that provides the highest utility out of four possibilities: stay at the present school (S), move to a different school within the school district (W), move between districts to a new school in a different school district (B), and leave public school teaching in Florida (L). It is thus assumed that all moves are the results of utility-maximizing choices. Although this assumption may not be correct in cases of involuntary separation due to poor performance or workforce reduction, such cases are relatively rare in the time period we analyze. According to teacher exit interviews conducted by the Florida Department of Education, 85 to 90 percent of teachers exit voluntarily. Including involuntary separations in the estimation would tend to bias against finding significant results because involuntary separations are primarily unrelated to pay and working conditions.8

For teachers, most moves and exits occur at the end of the school year. In addition, information on schools and districts is typically only available at yearly intervals. Given this discreteness in the data, we use a discrete multinomial logit hazard model with both time-varying and time-invariant coefficients. The discrete-time hazard function models the probability that any of the four events—staying, moving within the district, moving between districts, or leaving—happens to teacher k during period t + 1, which is conditional on the event not occurring until that time. The discrete-time hazard function can be interpreted as the probability of transition at discrete time t + 1 given survival up to time t + 1:
formula
3
Assuming independence of irrelevant alternatives and error terms that are independently and identically extreme-value distributed, a multinomial logit hazard model specifies the probability of choosing each alternative as a function of teacher, school, and district characteristics.9 The cumulative probability of leaving a particular school is a summation of the transition probability of exiting teaching, the probability of moving within a district, and the probability of moving across districts:
formula
4A
formula
4B
formula
4C

Estimates are reported as exponentiated coefficients or relative risk ratios. A relative risk ratio greater than one indicates that a one unit increase in the predictor is associated with an increase in the odds of moving or leaving compared with the default of staying at the initial school. Likewise, a relative risk ratio of less than one implies a one unit increase in the explanatory variable is associated with decreased odds of leaving or moving relative to the default outcome.

The effect of a teacher's own productivity on mobility decisions (, , and ) will depend in part on the degree to which human capital traits that are associated with productivity in teaching are transferable across schools and occupations. For example, Harris and Sass (2014) find that teacher value added in math is positively correlated with a teacher's intelligence and subject knowledge (as rated by their principal). They also find the enthusiasm and motivation of reading teachers is positively associated with their value added. If such cognitive and noncognitive traits also have value in occupations other than teaching, then will be statistically significant and positive and its associated relative risk ratio will have a value greater than one, indicating that teachers from a given quartile are more likely to exit public school teaching than are bottom-quartile teachers. Presumably, productive traits of teachers would be more likely to be transferable across schools within the public education sector, making the associated relative risk ratios for and greater than . If, however, the skills that make a teacher productive are not easily observable by persons outside of the teacher's current school, the relative risk ratios for and may still be close to one, indicating teaching quality has little or no effect on mobility between public schools.

In addition to teacher own quality, the quality of a teacher's faculty peers may also impact mobility decisions (, and ). Greater peer quality could reduce teacher mobility if positive spillover effects exist, whereby more capable peers stimulate productivity via informal learning channels (Jackson and Bruegmann 2009). Alternatively, higher peer quality could generate social pressure to conform to a higher productivity norm (Mas and Moretti 2009). Some teachers may choose to switch schools or leave teaching altogether in order to avoid such pressure.

The effects of peer teacher quality may also vary with a teacher's own productivity. Podgursky, Monroe, and Watson (2004) find that the probability of female teachers exiting the teaching profession increases the greater the difference between their own ACT college entrance-exam score and the average of the ACT scores of other teachers at their school. This may reflect positive assortative matching whereby teachers seek out positions in which their productivity matches the productivity of their colleagues.10

If one's self perception is a function of the productivity of their colleagues, relatively less productive teachers might opt to move to schools where they would be surrounded by less able colleagues. In this case the relative risk ratio for will be less than one, that is, low quality teachers with higher quality peers and high quality teachers with lower quality peers are more likely to switch schools.

4.  Data

We utilize data from the Florida Education Data Warehouse (FL-EDW), an integrated longitudinal database that covers all public school students and teachers in the state of Florida.11 Like statewide administrative databases in North Carolina and Texas, the FL-EDW contains a rich set of information that is linked through time on both individual students and their teachers. Because students may have more than one instructor in a given subject at a point in time (e.g., a regular education teacher and a special education reading teacher in elementary school or two math classes taught by different instructors in high school), we limit the analysis to students with a single “solely responsible” teacher in a subject and year.12

Statewide data, as opposed to data from an individual school district, are particularly useful for studying teacher labor markets because we can follow teachers who move from one district to another within Florida. We cannot, however, track teachers who move to another state. Fortunately, because Florida is a peninsula surrounded on three sides by water, interstate emigration of teachers from Florida is relatively infrequent. Using the nationally representative School and Staffing Survey that tracks teachers across state borders and follows them into other occupations, Feng (2010) concludes that misclassifications of the three types of movers (i.e., inter-state movers, movers to private school within same state, and movers to private schools in other states) are not a major concern in geographically large states like Florida. Our analysis of teacher quality and mobility covers school years 2000–01 through 2003–04. Testing of math and reading achievement in consecutive grades did not begin until the 1999–2000 school year and only includes grades 3–10. Construction of our teacher quality measures is based on current and prior-year student achievement, thereby limiting our sample to math and reading teachers in grades 4–10.

5.  Results

Descriptive Evidence

Table 1 provides means and standard errors of our teacher quality measure in math and reading for each of the teacher mobility categories: stayers, intra-district movers, inter-district movers, and leavers. We also present pairwise t-tests of mean teacher quality between stayers and all others, between stayers and within-district movers, between stayers and inter-district movers, and between stayers and leavers. For both math and reading, the average quality of teachers who stay at the initial school is statistically higher than that of teachers who switch schools within their initial district, move to another district, or leave public school teaching altogether.

Table 1. 
Means of Teacher Quality by Mobility Status and t-Tests of Differences in Means
MathReading
Mobility TypeMean Teacher Value Addedt-Test of Difference from Stayer CategoryMean Teacher Value Addedt-Test of Difference from Stayer Category
Stayers 0.0306  0.0285  
 (0.208)  (0.193)  
All movers 0.0067 0.0239** 0.0057 0.0228** 
 (0.236) [11.98] (0.229) [12.00] 
Intra-district movers 0.0076 0.0230** 0.0038 0.0247** 
 (0.226) [8.32] (0.227) [9.42] 
Inter-district movers −0.0144 0.0450** −0.0001 0.0285** 
 (0.244) [8.31] (0.226) [5.57] 
Exit FL public schools 0.0114 0.0192** 0.0093 0.0192** 
 (0.244) [6.65] (0.231) [7.03] 
MathReading
Mobility TypeMean Teacher Value Addedt-Test of Difference from Stayer CategoryMean Teacher Value Addedt-Test of Difference from Stayer Category
Stayers 0.0306  0.0285  
 (0.208)  (0.193)  
All movers 0.0067 0.0239** 0.0057 0.0228** 
 (0.236) [11.98] (0.229) [12.00] 
Intra-district movers 0.0076 0.0230** 0.0038 0.0247** 
 (0.226) [8.32] (0.227) [9.42] 
Inter-district movers −0.0144 0.0450** −0.0001 0.0285** 
 (0.244) [8.31] (0.226) [5.57] 
Exit FL public schools 0.0114 0.0192** 0.0093 0.0192** 
 (0.244) [6.65] (0.231) [7.03] 

Notes: Standard errors are in parentheses and t-statistics are in brackets.

**p < 0.01.

The comparisons of average teacher quality by mobility status presented in table 1 assume a monotonic relationship between teacher mobility patterns and teacher quality. There are a number of reasons to expect teacher mobility to vary across the teacher quality distribution in a nonlinear fashion, however. For example, assortative matching could lead both relatively high- and low-quality teachers to change schools. Further, both higher- and lower-quality teachers may be more likely to leave teaching than the average-quality teacher, but for different reasons. High-quality teachers may be more likely to exit because they have particularly good outside opportunities, whereas low-quality teachers may exit because they recognize they are not particularly effective teachers or do not enjoy teaching.13

In figure 1 we investigate the relationship between teacher quality in reading and math and the probability of turnover at a school. Figure 1 presents local linear regression plots and 95 percent confidence intervals of the probability a teacher departs her current school against estimated teacher quality in reading and math using kernel-weighted local polynomial smoothing. There is a U-shaped relationship between teacher quality and turnover in both math and reading—that is, better-than-average and worse-than-average teachers both have a higher turnover rate compared with middling teachers. For example, in reading, about 25 percent of teachers that are one half of a standard deviation below the average teacher quality are leaving their schools for other schools or professions compared with 15 percent of average-quality teachers. Given the U-shaped relationship between turnover and teacher quality, we allow for differential mobility patterns by quartile of teacher quality in our quantitative analyses.

Figure 1.

Probability of Leaving Current School by Teacher Value Added in Math and Reading.

Figure 1.

Probability of Leaving Current School by Teacher Value Added in Math and Reading.

Multivariate Estimates of the Relationship of Teacher Quality to the Incidence of Teacher Mobility

Estimates of the multinomial hazard models of teacher mobility choices, allowing for differential effects by quartile of teacher quality, are presented as model 1 in table 2 (Math) and table 3 (Reading). Quartile one is the lowest quality and is used as the reference category.

Table 2. 
Multinomial Logit Hazard Estimates of the Effects of Teacher Quality in Math, Peer Teacher Quality, and Peer Teacher Characteristics on the Odds of Teacher Mobility
Model 1Model 2Model 3Model 4
Intra-District MoveInter-District MoveExit FL Public SchoolsIntra-District MoveInter-District MoveExit FL Public SchoolsIntra-District MoveInter-District MoveExit FL Public SchoolsIntra-District MoveInter-District MoveExit FL Public Schools
Teacher quality quartile 2 0.940 0.896 0.781** 0.934 0.887 0.776** 0.932 0.885 0.772** 0.965 1.087 0.744** 
Teacher quality quartile 3 0.997 0.912 0.827** 0.984 0.905 0.819** 0.974 0.900 0.813** 0.916 0.832 0.832 
Teacher quality quartile 4 1.078 0.816 0.953 1.063 0.815 0.943 1.055 0.808 0.937 1.262* 0.829 0.965 
Peer teacher quality quartile 2    1.119 1.193 1.098 1.117 1.179 1.095 1.119 1.144 1.015 
Peer teacher quality quartile 3    1.119 1.155 1.070 1.110 1.135 1.060 1.221* 1.353 1.070 
Peer teacher quality quartile 4    1.068 1.019 1.032 1.050 0.988 1.019 1.054 1.088 1.165 
Peer percent with master's degree       0.554** 1.009 1.398* 0.551** 1.017 1.397* 
Peer percent National Board certified       0.960 4.414 0.601 0.977 4.364 0.602 
Peer percent prof. certification       0.486 1.987 0.638 0.477* 1.961 0.636 
Peer percent with 1–2 years of experience       0.369** 0.126** 0.504* 0.372** 0.125** 0.498* 
Peer percent with 3–4 years of experience       2.476* 0.827 0.614 2.524* 0.831 0.611 
Peer percent of teachers with 5–9 years of experience       1.915 0.196* 0.844 1.944 0.198* 0.833 
Peer percent of teachers with 10–14 years of experience       0.405* 0.046** 0.458 0.409* 0.045** 0.455 
Peer percent of teachers with 15–24 years of experience       0.312** 0.276 0.499 0.316** 0.275 0.494 
Peer percent of teachers with 25 years of experience and more       0.428 0.331 0.473 0.432 0.331 0.469 
Own quality quartile #2 x Peer quality quartile #2          0.919 0.783 1.140 
Own quality quartile #2 x Peer quality quartile #3          0.845 0.630 1.096 
Own quality quartile #2 x Peer quality quartile #4          1.161 0.881 0.832 
Own quality quartile #3 x Peer quality quartile #2          1.062 1.400 1.093 
Own quality quartile #3 x Peer quality quartile #3          1.010 0.885 0.880 
Own quality quartile #3 x Peer quality quartile #4          1.136 0.978 0.887 
Own quality quartile #4 x Peer quality quartile #2          0.976 1.093 1.085 
Own quality quartile #4 x Peer quality quartile #3          0.747* 0.917 0.980 
Own quality quartile #4 x Peer quality quartile #4          0.748 0.863 0.808 
Model 1Model 2Model 3Model 4
Intra-District MoveInter-District MoveExit FL Public SchoolsIntra-District MoveInter-District MoveExit FL Public SchoolsIntra-District MoveInter-District MoveExit FL Public SchoolsIntra-District MoveInter-District MoveExit FL Public Schools
Teacher quality quartile 2 0.940 0.896 0.781** 0.934 0.887 0.776** 0.932 0.885 0.772** 0.965 1.087 0.744** 
Teacher quality quartile 3 0.997 0.912 0.827** 0.984 0.905 0.819** 0.974 0.900 0.813** 0.916 0.832 0.832 
Teacher quality quartile 4 1.078 0.816 0.953 1.063 0.815 0.943 1.055 0.808 0.937 1.262* 0.829 0.965 
Peer teacher quality quartile 2    1.119 1.193 1.098 1.117 1.179 1.095 1.119 1.144 1.015 
Peer teacher quality quartile 3    1.119 1.155 1.070 1.110 1.135 1.060 1.221* 1.353 1.070 
Peer teacher quality quartile 4    1.068 1.019 1.032 1.050 0.988 1.019 1.054 1.088 1.165 
Peer percent with master's degree       0.554** 1.009 1.398* 0.551** 1.017 1.397* 
Peer percent National Board certified       0.960 4.414 0.601 0.977 4.364 0.602 
Peer percent prof. certification       0.486 1.987 0.638 0.477* 1.961 0.636 
Peer percent with 1–2 years of experience       0.369** 0.126** 0.504* 0.372** 0.125** 0.498* 
Peer percent with 3–4 years of experience       2.476* 0.827 0.614 2.524* 0.831 0.611 
Peer percent of teachers with 5–9 years of experience       1.915 0.196* 0.844 1.944 0.198* 0.833 
Peer percent of teachers with 10–14 years of experience       0.405* 0.046** 0.458 0.409* 0.045** 0.455 
Peer percent of teachers with 15–24 years of experience       0.312** 0.276 0.499 0.316** 0.275 0.494 
Peer percent of teachers with 25 years of experience and more       0.428 0.331 0.473 0.432 0.331 0.469 
Own quality quartile #2 x Peer quality quartile #2          0.919 0.783 1.140 
Own quality quartile #2 x Peer quality quartile #3          0.845 0.630 1.096 
Own quality quartile #2 x Peer quality quartile #4          1.161 0.881 0.832 
Own quality quartile #3 x Peer quality quartile #2          1.062 1.400 1.093 
Own quality quartile #3 x Peer quality quartile #3          1.010 0.885 0.880 
Own quality quartile #3 x Peer quality quartile #4          1.136 0.978 0.887 
Own quality quartile #4 x Peer quality quartile #2          0.976 1.093 1.085 
Own quality quartile #4 x Peer quality quartile #3          0.747* 0.917 0.980 
Own quality quartile #4 x Peer quality quartile #4          0.748 0.863 0.808 

Notes: Reported numbers are the relative risk ratios. Explanatory variables included teacher's age, age squared, female, female and age interaction term, teacher's race, teacher's education level, professional certification, reading certification, middle school math certification, high school math certification, indicator variables for special education teachers, middle school education teachers, high school teachers, English teachers, math or science teachers, self-contained class teachers, social studies teachers, indicator variable for regular and full time teachers, teacher's experience and experience squared term, dummy variables indicating the cohort year and teacher's own salaries, and classroom, school, and district characteristics such as class size, average math score on the FCAT, disciplinary incidents, percent of minority students (black, Hispanic), percent of students on free or reduced-price lunch program, indicator variables for the school grade in 1999 and log(years of teaching). Significance levels are based on the underlying logistic regression coefficients and the associated robust standard errors clustered at the school level.

*p < 0.05; **p < 0.01.

Table 3. 
Multinomial Logit Hazard Estimates of the Effects of Teacher Quality in Reading, Peer Teacher Quality, and Peer Teacher Characteristics on the Odds of Teacher Mobility
Model 1Model 2Model 3Model 4
Intra-District MoveInter-District MoveExit FL Public SchoolsIntra-District MoveInter-District MoveExit FL Public SchoolsIntra-District MoveInter-District MoveExit FL Public SchoolsIntra-District MoveInter-District MoveExit FL Public Schools
Teacher quality quartile 2 0.926 0.941 0.794** 0.929 0.940 0.798** 0.924 0.935 0.798** 1.105 1.115 0.721** 
Teacher quality quartile 3 0.937 1.019 0.761** 0.946 1.006 0.768** 0.937 1.002 0.767** 1.006 1.159 0.738** 
Teacher quality quartile 4 0.993 0.988 0.935 1.005 0.969 0.950 0.997 0.967 0.950 1.071 0.926 1.237* 
Peer teacher quality quartile 2    0.996 1.252* 1.008 1.016 1.272* 1.012 1.155 1.481* 0.918 
Peer teacher quality quartile 3    0.892 1.085 0.895 0.913 1.093 0.898 0.985 1.035 0.941 
Peer teacher quality quartile 4    0.964 1.238 0.888 0.984 1.221 0.890 1.146 1.563 1.117 
Peer percent with master's degree       0.537** 1.075 1.158 0.536** 1.062 1.145 
Peer percent national board certified       0.837 8.749 0.715 0.833 8.923 0.734 
Peer percent prof. certification       0.470* 1.809 0.795 0.464* 1.768 0.803 
Peer percent with 1–2 years of experience       0.395** 0.089** 0.476* 0.397* 0.090** 0.469* 
Peer percent with 3–4 years of experience       2.040 0.739 0.492 2.039 0.745 0.486 
Peer percent of teachers with 5–9 years of experience       1.423 0.212* 0.646 1.433 0.211* 0.641 
Peer percent of teachers with 10–14 years of experience       0.489 0.028** 0.473 0.492 0.028** 0.460 
Peer percent of teachers with 15–24 years of experience       0.333** 0.315 0.536 0.335** 0.311 0.529 
Peer percent of teachers with 25 years of experience and more       0.328** 0.124** 0.341** 0.329** 0.124** 0.339** 
Own quality quartile #2 x Peer quality quartile #2          0.712** 0.678 1.315* 
Own quality quartile #2 x Peer quality quartile #3          0.815 0.885 1.050 
Own quality quartile #2 x Peer quality quartile #4          0.798 0.845 0.935 
Own quality quartile #3 x Peer quality quartile #2          0.921 0.791 1.133 
Own quality quartile #3 x Peer quality quartile #3          0.926 1.033 1.067 
Own quality quartile #3 x Peer quality quartile #4          0.823 0.664 0.801 
Own quality quartile #4 x Peer quality quartile #2          0.883 1.055 0.925 
Own quality quartile #4 x Peer quality quartile #3          0.954 1.375 0.667** 
Own quality quartile #4 x Peer quality quartile #4          0.847 0.769 0.555** 
Model 1Model 2Model 3Model 4
Intra-District MoveInter-District MoveExit FL Public SchoolsIntra-District MoveInter-District MoveExit FL Public SchoolsIntra-District MoveInter-District MoveExit FL Public SchoolsIntra-District MoveInter-District MoveExit FL Public Schools
Teacher quality quartile 2 0.926 0.941 0.794** 0.929 0.940 0.798** 0.924 0.935 0.798** 1.105 1.115 0.721** 
Teacher quality quartile 3 0.937 1.019 0.761** 0.946 1.006 0.768** 0.937 1.002 0.767** 1.006 1.159 0.738** 
Teacher quality quartile 4 0.993 0.988 0.935 1.005 0.969 0.950 0.997 0.967 0.950 1.071 0.926 1.237* 
Peer teacher quality quartile 2    0.996 1.252* 1.008 1.016 1.272* 1.012 1.155 1.481* 0.918 
Peer teacher quality quartile 3    0.892 1.085 0.895 0.913 1.093 0.898 0.985 1.035 0.941 
Peer teacher quality quartile 4    0.964 1.238 0.888 0.984 1.221 0.890 1.146 1.563 1.117 
Peer percent with master's degree       0.537** 1.075 1.158 0.536** 1.062 1.145 
Peer percent national board certified       0.837 8.749 0.715 0.833 8.923 0.734 
Peer percent prof. certification       0.470* 1.809 0.795 0.464* 1.768 0.803 
Peer percent with 1–2 years of experience       0.395** 0.089** 0.476* 0.397* 0.090** 0.469* 
Peer percent with 3–4 years of experience       2.040 0.739 0.492 2.039 0.745 0.486 
Peer percent of teachers with 5–9 years of experience       1.423 0.212* 0.646 1.433 0.211* 0.641 
Peer percent of teachers with 10–14 years of experience       0.489 0.028** 0.473 0.492 0.028** 0.460 
Peer percent of teachers with 15–24 years of experience       0.333** 0.315 0.536 0.335** 0.311 0.529 
Peer percent of teachers with 25 years of experience and more       0.328** 0.124** 0.341** 0.329** 0.124** 0.339** 
Own quality quartile #2 x Peer quality quartile #2          0.712** 0.678 1.315* 
Own quality quartile #2 x Peer quality quartile #3          0.815 0.885 1.050 
Own quality quartile #2 x Peer quality quartile #4          0.798 0.845 0.935 
Own quality quartile #3 x Peer quality quartile #2          0.921 0.791 1.133 
Own quality quartile #3 x Peer quality quartile #3          0.926 1.033 1.067 
Own quality quartile #3 x Peer quality quartile #4          0.823 0.664 0.801 
Own quality quartile #4 x Peer quality quartile #2          0.883 1.055 0.925 
Own quality quartile #4 x Peer quality quartile #3          0.954 1.375 0.667** 
Own quality quartile #4 x Peer quality quartile #4          0.847 0.769 0.555** 

Notes: Reported numbers are the relative risk ratios. Explanatory variables included teacher's age, age squared, female, female and age interaction term, teacher's race, teacher's education level, professional certification, reading certification, middle school math certification, high school math certification, indicator variables for special education teachers, middle school education teachers, high school teachers, English teachers, math or science teachers, self-contained class teachers, social studies teachers, indicator variable for regular and full time teachers, teacher's experience and experience squared term, dummy variables indicating the cohort year and teacher's own salaries, and classroom, school, and district characteristics such as class size, average math score on the FCAT, disciplinary incidents, percent of minority students (black, Hispanic), percent of students on free or reduced-price lunch program, indicator variables for the school grade in 1999 and log(years of teaching). Significance levels are based on the underlying logistic regression coefficients and the associated robust standard errors clustered at the school level.

*p < 0.05; **p < 0.01.

For both intra-district and inter-districts moves, we do not find a statistically significant relationship at the 95 percent confidence level between teacher quality and the likelihood a teacher changes schools. If the underlying coefficient for a variable is statistically significant and positive, the relative risk ratio will also be statistically significant and greater than one. On the other hand, if the underlying coefficient for a variable is statistically significant and negative, the relative risk ratio will be statistically significant and less than one. The point estimates of the relative risk ratios are frequently near one, indicating that the lack of a statistically significant relationship between teacher effectiveness and the probability of switching schools is not simply due to a lack of precision in the estimates.14 The finding that between-school moves are uncorrelated with teacher quality suggests that the initial match between teachers and schools is no different for more-effective and less-effective teachers. The lack of a significant relationship between teacher quality and inter-district moves is not surprising given the structure of school districts in Florida. Considering that Florida has countywide school districts, taking a job in a different district typically involves changing residential locations. Most likely residential change is affected by external factors, like spousal job change or the desire to move closer to other family members.

In contrast to between-school mobility, a clear bimodal pattern emerges for the relationship between teacher quality and exit. For both subjects, teachers in the middle two quartiles have much lower exit probabilities than do teachers in the bottom and top quartiles. The relative risk ratios for the second and third quartiles are around 0.8 and statistically significant at a 99 percent confidence level, indicating that teachers in the middle 50 percent of the quality distribution have 20 percent (1 – 0.8) lower odds of leaving teaching than do teachers in the lowest quartile of the quality distribution. Relative risk ratios for the highest quartile teachers are not statistically significant at the 95 percent confidence level, indicating that the highest performing teachers are just as likely to exit as are bottom-quartile teachers and are much more likely to leave teaching than teachers in the middle of the quality distribution after controlling for teacher characteristics. These results are consistent with both schools losing their best teachers to more attractive outside options and losing their worst teachers who may be better suited to other occupations.

Peer Teacher Characteristics, Peer Teacher Quality, and Match Quality

In addition to estimates from model 1, tables 2 and 3 display estimates from three additional multinomial logit hazard models which allow for nonlinear peer teacher quality effects (model 2), nonlinear peer teacher quality effects plus peer teacher characteristic effects (model 3), and nonlinear peer teacher quality effects plus peer teacher characteristic effects plus own and peer teacher quality interaction effects (model 4).

Estimates from model 2 yield similar estimates with respect to the effects of own teacher quality on exit as those from model 1, but do not reveal a consistent pattern in the effect of peer teacher quality on the likelihood of exit. In all but one instance (peer teacher quality quartile 3 in reading) the estimated effects of peer teacher quality on individual teacher mobility decisions are insignificantly different from zero at a 95 percent confidence level.

In model 3 we examine whether observable peer teacher characteristics are associated with an individual teacher's mobility decision. We find that having more experienced peers and higher percentages of colleagues with a master's degree in reading and math or professional (non-temporary) certification in reading translates into a lower likelihood of transferring within-district. This suggests that peer teachers with better qualifications may be more likely to provide positive spillovers or otherwise enhance the work environment. Curiously, increases in the shares of colleagues who hold master's degrees are associated with a greater likelihood of exiting the public school system for math teachers. No such effect is found for reading teachers.

Our final model of teacher job choice combines all variables of interest into a single specification and adds interaction terms between teachers’ own quality quartiles and peer teacher quality quartiles. These interaction terms will capture the match quality between a teacher's own productivity and their colleagues’ productivity. Controlling for teacher own quality, peer teacher quality, and peer observable qualifications, fourth-quartile math teachers are less likely to make a within-district transfer if their peers are at least above average (third or fourth quartiles). Similarly, fourth quartile Reading/ELA teachers are less likely to exit the Florida public school system if they have above-average peer teachers.

Mobility and the Distribution of Teacher Quality

To understand the implications of teacher mobility on the distribution of teacher quality across schools, we first compare the mean characteristics of the origin and destination schools for teachers who switch schools. The averages and t-statistics for the differences in means are presented in table 4. Consistent with prior research, teachers tend to move to schools where students have higher achievement, a smaller fraction of students are black, and a smaller proportion of students are from low-income families (as measured by free/reduced-price lunch status). Teachers also tend to move to schools with higher school grades.15 For within-district moves, both math and reading average teacher quality tends to be slightly higher for receiving schools than for sending schools. Though counterintuitive, teachers making intra-district moves tend to move to schools with more disciplinary incidents per student. This may be because of a higher likelihood of reporting of disciplinary incidents in high-performing schools.

Table 4. 
Comparison of Student and Teacher Characteristics at Origin and Destination Schools
Intra-District MoversInter-District Movers
Destination SchoolOrigin SchoolDifference in MeansDestination SchoolOrigin SchoolDifference in Means
School Average CharacteristicsMean (SE)Mean (SE)Mean Difference [t-statistic]Mean (SE)Mean (SE)Mean Difference [t-statistic]
Percent black students 0.2345 0.2850 −0.0505** 0.2306 0.2676 −0.0370** 
 (0.0031) (0.0036) [–14.37] (0.0063) (0.0070) [–4.25] 
Percent Hispanic students 0.1995 0.1901 0.0094** 0.1588 0.1878 −0.0289** 
 (0.0027) (0.0027) [3.91] (0.0046) (0.0059) [–4.53]** 
Percent free/reduced price lunch students 0.4448 0.5262 −0.0814** 0.4439 0.5048 −0.0609** 
 (0.0034) (0.0036) [–22.43] (0.0074) (0.0073) [–6.50] 
Disciplinary incidents 0.3901 0.3553 0.0348** 0.4291 0.4054 0.0237 
 (0.0075) (0.0078) [3.96] (0.0186) (0.0160) [1.10] 
Math performance 308.5647 300.4491 8.1156** 309.2028 301.7787 7.4241** 
 (0.3380) (0.3342) [21.75] (0.6802) (0.6408) [8.73] 
Math teacher quality 0.0367 0.0214 0.0153** 0.0225 0.0147 0.0078 
 (0.0019) (0.0014) [7.23] (0.0041) (0.0037) [1.47] 
Reading teacher quality 0.0297 0.0127 0.0170** 0.0239 0.0211 0.0028 
 (0.0018) (0.0014) [8.08] (0.0040) (0.0031) [0.58] 
School grade 2.9328 2.6224 0.3104** 2.9425 2.7184 0.2241** 
 (0.0173) (0.0177) [14.42] (0.0319) (0.0328) [5.35] 
Intra-District MoversInter-District Movers
Destination SchoolOrigin SchoolDifference in MeansDestination SchoolOrigin SchoolDifference in Means
School Average CharacteristicsMean (SE)Mean (SE)Mean Difference [t-statistic]Mean (SE)Mean (SE)Mean Difference [t-statistic]
Percent black students 0.2345 0.2850 −0.0505** 0.2306 0.2676 −0.0370** 
 (0.0031) (0.0036) [–14.37] (0.0063) (0.0070) [–4.25] 
Percent Hispanic students 0.1995 0.1901 0.0094** 0.1588 0.1878 −0.0289** 
 (0.0027) (0.0027) [3.91] (0.0046) (0.0059) [–4.53]** 
Percent free/reduced price lunch students 0.4448 0.5262 −0.0814** 0.4439 0.5048 −0.0609** 
 (0.0034) (0.0036) [–22.43] (0.0074) (0.0073) [–6.50] 
Disciplinary incidents 0.3901 0.3553 0.0348** 0.4291 0.4054 0.0237 
 (0.0075) (0.0078) [3.96] (0.0186) (0.0160) [1.10] 
Math performance 308.5647 300.4491 8.1156** 309.2028 301.7787 7.4241** 
 (0.3380) (0.3342) [21.75] (0.6802) (0.6408) [8.73] 
Math teacher quality 0.0367 0.0214 0.0153** 0.0225 0.0147 0.0078 
 (0.0019) (0.0014) [7.23] (0.0041) (0.0037) [1.47] 
Reading teacher quality 0.0297 0.0127 0.0170** 0.0239 0.0211 0.0028 
 (0.0018) (0.0014) [8.08] (0.0040) (0.0031) [0.58] 
School grade 2.9328 2.6224 0.3104** 2.9425 2.7184 0.2241** 
 (0.0173) (0.0177) [14.42] (0.0319) (0.0328) [5.35] 

**p < 0.01.

The effect of teacher mobility on the distribution of teacher quality across schools not only depends on the types of schools that teachers move to but also on the effectiveness of the teachers making those moves. In table 5 we compare the quality of a teacher making a move and the quality of faculty at the receiving school. Because we examine the average faculty quality prior to the move, we are avoiding any confounding caused by incoming teachers changing the average quality of teachers on campus. Looking down the diagonal at the row percentages, it is apparent that the pluralities of movers into each of the four receiving-school quality categories are of comparable quality to the average quality of faculty at the receiving school. For example, 43 percent of fourth-quartile math teachers who change schools move to schools where average math teacher quality is in the fourth quartile of the math teacher quality distribution. Just under 40 percent of reading teachers who switch schools also land in schools where the average reading teacher quality is in the top quartile. Looking across rows at the column percentages, it is also evident that the fraction of movers goes down with the difference in the quality of the moving teacher and the average quality of teachers at the receiving school. Schools whose average math teacher quality is in the top quartile draw many more top-quartile teachers (36 percent) than do schools where average teacher quality is in the bottom quartile (18 percent). Likewise, bottom-quartile schools are disproportionally attracting bottom-quartile teachers (43 percent) compared to top quartile schools (13 percent). The pattern is similar in reading, where schools whose average teacher quality draws 35 percent of transferring teachers who are in the top quartile, and for schools whose average teacher quality is in the bottom quartile, 37 percent of teachers transferring are in the bottom quartile of the teacher quality distribution.

Table 5. 
Percent Distribution of Entering Math and Reading Teacher Quality for Each Quartile of Average Receiving School Teacher Quality
Receiving School Average Teacher Quality
Entering Teacher Quality1st Quartile Row Percent/Column Percent2nd Quartile Row Percent/Column Percent3rd quartile Row Percent/Column Percent4th Quartile Row Percent/Column PercentTotal Row Percent/Column Percent
Math 
1st quartile 33.53 26.43 21.47 18.58 100.00 
 43.01 29.08 22.74 18.22 27.38 
2nd quartile 22.05 28.33 28.46 21.15 100.00 
 25.46 28.06 27.14 18.68 24.65 
3rd quartile 15.64 25.17 28.42 30.77 100.00 
 18.21 25.14 27.32 27.39 24.85 
4th quartile 12.30 19.07 25.50 43.13 100.00 
 13.32 17.71 22.80 35.71 23.12 
Total 21.35 24.89 25.85 27.92 100.00 
 100.00 100.00 100.00 100.00 100.00 
Reading 
1st quartile 31.69 28.41 22.13 17.77 100.00 
 36.77 33.29 25.64 20.64 29.08 
2nd quartile 28.09 27.62 25.43 18.86 100.00 
 27.77 27.59 25.11 18.67 24.78 
3rd quartile 22.20 22.61 28.04 27.15 100.00 
 21.21 21.82 26.75 25.97 23.94 
4th quartile 16.09 19.35 25.43 39.14 100.00 
 14.25 17.31 22.49 34.71 22.20 
Total 25.06 24.81 25.09 25.03 100.00 
 100.00 100.00 100.00 100.00 100.00 
Receiving School Average Teacher Quality
Entering Teacher Quality1st Quartile Row Percent/Column Percent2nd Quartile Row Percent/Column Percent3rd quartile Row Percent/Column Percent4th Quartile Row Percent/Column PercentTotal Row Percent/Column Percent
Math 
1st quartile 33.53 26.43 21.47 18.58 100.00 
 43.01 29.08 22.74 18.22 27.38 
2nd quartile 22.05 28.33 28.46 21.15 100.00 
 25.46 28.06 27.14 18.68 24.65 
3rd quartile 15.64 25.17 28.42 30.77 100.00 
 18.21 25.14 27.32 27.39 24.85 
4th quartile 12.30 19.07 25.50 43.13 100.00 
 13.32 17.71 22.80 35.71 23.12 
Total 21.35 24.89 25.85 27.92 100.00 
 100.00 100.00 100.00 100.00 100.00 
Reading 
1st quartile 31.69 28.41 22.13 17.77 100.00 
 36.77 33.29 25.64 20.64 29.08 
2nd quartile 28.09 27.62 25.43 18.86 100.00 
 27.77 27.59 25.11 18.67 24.78 
3rd quartile 22.20 22.61 28.04 27.15 100.00 
 21.21 21.82 26.75 25.97 23.94 
4th quartile 16.09 19.35 25.43 39.14 100.00 
 14.25 17.31 22.49 34.71 22.20 
Total 25.06 24.81 25.09 25.03 100.00 
 100.00 100.00 100.00 100.00 100.00 

6.  Summary and Conclusions

In this paper we provide new evidence on the relationship between teacher quality and teacher job change, and on the corresponding impacts of teacher mobility on the distribution of teacher quality across schools. We do not find a strong relationship between teacher quality and the likelihood of either intra-district or inter-district teacher mobility. In the case of exit from the public school sector, however, we uncover a bimodal quality distribution. The most effective teachers are more likely to exit than middling quality teachers, but teachers at the low end of the quality distribution are also more likely than middling teachers to leave. Although attrition of less effective teachers from the profession is desirable, the higher departure rate of top-quartile teachers should be of concern to policy makers. Not only does the exit of high-quality teachers reduce student achievement, reductions in teacher quality may also have negative effects on long-term outcomes for students, including a higher incidence of teen pregnancy, a lower probability of college attendance, and reduced earnings in adulthood (Chetty, Friedman, and Rockoff 2014).

Additionally, we find the fit between teacher own-quality and peer-faculty quality is an important factor in teacher mobility decisions. Top quartile math teachers tend to stay at a school when they are surrounded by highly productive peers. Likewise, top quartile reading teachers tend to stay within the public school sector when they are surrounded by relatively productive peers. We also find evidence of human capital spillovers. Peers with job-specific teaching experience, professional certification, and advanced degrees dampen intra-district mobility. These peers may provide school-specific job skills and knowledge to their colleagues, resulting in lower turnover.

Teachers who move tend to go to a school where the average teacher quality is like their own. The fraction of top quartile movers hired by schools whose faculty is in the top quartile of the quality distribution is much higher than that of schools whose faculty is in the bottom quartile of the quality distribution. The net result is that the movement of teachers across schools tends to exacerbate differences in teacher quality.

Given the strong link between teacher quality and student performance, our results suggest that teacher mobility tends to increase the achievement gaps between white and minority students and between poor and more affluent students. This suggests that mechanisms that reduce the natural flow of teachers to schools with superior faculties could help reduce student achievement gaps. One potential policy option that has been shown to affect teacher mobility choices are programs that offer financial inducements for teachers to teach in schools serving high proportions of low-achieving or disadvantaged students (Clotfelter et al. 2008; Glazerman et al. 2013). For differential salary schemes to alter the distribution of teacher quality, however, any monetary inducements must be tied to teacher quality. The ultimate impact on student achievement of inducing high quality teachers to switch schools will depend on the responsiveness of teachers to incentives and the effectiveness of teachers who transfer. Our results suggest that the positive effects of such transfers could be muted somewhat by attrition of high-quality teachers if transfers significantly alter the composition of faculty at their former schools.

Notes

1. 

Rockoff (2004), Hanushek (2011), Rivkin, Hanushek, and Kain (2005), Aaronson, Barrow, and Sander (2007), and Kane, Rockoff, and Staiger (2008) demonstrate that teacher quality is the most important schooling input in the determination of student achievement.

2. 

Boyd, Grossman, and Lankford (2008) and West and Chingos (2009) provide graphical presentations of teacher mobility broken down by quartiles and terciles, respectively. Nevertheless, neither study provides confidence intervals or tests of equality across levels of the value-added distribution.

3. 

In Appendix table A.1 we provide correlations between our selected value-added measure and a variety of alternative specifications. The within-subject correlation is fairly strong for both math and reading. For example, the correlation between teacher-by-year estimates, assuming complete persistence and those with partial persistence, is 0.81 for both math and reading. When student fixed effects are used in place of observed student characteristics to control for time-invariant student heterogeneity, however, the correlations are lower (0.61 in math and 0.54 in reading). When we compare teacher-by-year estimates with the corresponding multi-year teacher effects the correlations are relatively strong, ranging from 0.64 to 0.73 in math and 0.61 to 0.73 in reading. For a more detailed analysis of alternative value-added models and the degree to which they produce similar estimates of teacher value added, see Sass, Semykina, and Harris (2014).

4. 

The use of school fixed effects in the present context is problematic for two reasons. First, the inclusion of school fixed effects means that teacher quality would be measured relative to other teachers at the same school, making it impossible to compare the quality of teachers across schools. Second, when we consider the effects of teacher peer quality, there is a mechanical (negative) correlation between own- and peer-teacher quality when within-school measures of teacher quality are used.

5. 

This is the traditional approach that has been used in the analysis of teacher labor markets. However, a recent paper by Boyd et al. (2013) uses an alternative approach based on a two-sided matching model. We intend to explore this two-sided approach in future work.

6. 

Commuting time to school is possibly an important aspect of teachers’ employment choice. Unfortunately, the available data do not provide such information.

7. 

In particular, we use individual-specific annual salary (excluding supplements for coaching or extended-day services like after-school tutoring and performance-based bonuses), as our measure of teacher compensation. For evidence on the effects of salaries on teacher mobility and retention, see Murnane and Olsen (1990); Mont and Rees (1996); Gritz and Theobald (1996); Reed, Rueben, and Barbour (2006); Feng (2010); and Clotfelter, Ladd, and Vigdor (2011). We also estimated some specifications with district fixed effects to account for differences in the cost of living across counties. Given our interest in examining teacher mobility both within and across districts, however, we focus on estimates from models without district fixed effects. Results from models with district fixed effects are available upon request.

8. 

In order to gauge whether the assumption of teacher-initiated job change affects our results, we separately plotted the distribution of teacher quality for involuntary turnover (e.g., termination, contract nonrenewal, and reduction in force) and voluntary turnover (e.g., retirement and resignation). Appendix figure A.1 presents the kernel density distribution of math teacher quality for four groups of teachers—stayers, voluntary turnover, involuntary turnover, and unknown reasons. The graphs must be interpreted with some caution because for many teachers the separation reason is unknown and for those whose separation reason is known, the vast majority are voluntary departures. Nonetheless, the distributions of teacher quality for teachers involved in involuntary and voluntary turnover nearly overlap with each other. This confirms earlier findings in New York (Boyd et al. 2011) that the relationship between teacher quality and teacher mobility is similar whether one considers only teacher-initiated transfers or all transfers.

9. 

Small-Hsiao tests do not reject the null hypothesis that elimination of any one outcome (staying, moving within districts, moving across districts, and exiting Florida public school teaching) does not alter the coefficients for our variables of interest. In addition, we estimated multinomial probit models using the same specifications as our main models. Unlike the multinomial logit model, the multinomial probit model does not assume the independence of irrelevant alternatives. Results from the multinomial probit model are nearly identical to the results from the multinomial logit models.

10. 

Note that our teacher-quality measures are not conditional on experience, so an inexperienced teacher could have experienced colleagues who are unconditionally more productive than the new teacher, but who would be less productive conditional on experience.

11. 

Detailed descriptions of the Florida data are provided in Sass (2006) and Harris and Sass (2011).

12. 

This exclusion reduces the sample of students by 17 percent for math and 26 percent for reading/ELA.

13. 

Conditional on exit, Chingos and West (2012) find that higher value-added teachers in Florida tend to earn more outside of teaching than do teachers who are less effective in promoting student achievement.

14. 

To the extent that teacher-characteristic controls in the multinomial hazard model, such as experience, are correlated with estimated teacher quality, this could induce some multicollinearity which would tend to reduce the statistical significance of the teacher-quality measures in the estimation of the incidence of teacher mobility.

15. 

In Florida, schools are assigned letter grades based on student performance on standardized exams (see Florida Department of Education 2010 and Feng, Figlio, and Sass 2010). School grades were converted to numeric scores of 4, 3, 2, 1, and 0 for A, B, C, D, and F schools, respectively.

Acknowledgments

We wish to thank the staff of the Florida Department of Education's K–20 Education Data Warehouse for their assistance in obtaining and interpreting the data used in this study. We also gratefully acknowledge the National Center for the Analysis of Longitudinal Data in Education Research (CALDER), funded through grant R305A060018 from the Institute of Education Sciences, U.S. Department of Education, for supporting this research. This paper has benefited from comments from Thomas Dee, seminar participants at New York University, and conference participants at the American Education Finance Association meetings, the American Educational Research Association meetings, and the 4th Annual CALDER Conference. We also thank Bryce Cashell and Elizabeth Blackburn for valuable research assistance. The views expressed in this paper are solely our own and do not necessarily reflect the opinions of the Florida Department of Education or our funders.

REFERENCES

Aaronson
,
Daniel
,
Lisa
Barrow
, and
William
Sander
.
2007
.
Teachers and student achievement in the Chicago public high schools
.
Journal of Labor Economics
25
(
1
):
95
135
. doi:10.1086/508733.
Betts
,
Julian R.
,
Andrew C.
Zau
, and
Lorien A.
Rice
.
2003
.
Determinants of student achievement: New evidence from San Diego
.
San Francisco
:
Public Policy Institute of California
.
Boyd
,
Donald
,
Pam
Grossman
, and
Hamilton
Lankford
.
2008
.
Who leaves? Teacher attrition and student achievement
.
NBER Working Paper No. 14022
.
Boyd
,
Donald J.
,
Hamilton
Lankford
,
Susanna
Loeb
, and
James
Wyckoff
.
2013
.
Analyzing the determinants of the matching of public school teachers to jobs: Disentangling the preferences of teachers and employers
.
Journal of Labor Economics
31
(
1
):
83
117
. doi:10.1086/666725.
Boyd
,
Donald
,
Hamilton
Lankford
,
Susanna
Loeb
, and
James
Wyckoff
.
2005
.
Explaining the short careers of high-achieving teachers in schools with low-performing students
.
American Economic Review
95
(
2
):
166
171
. doi:10.1257/000282805774669628.
Boyd
,
Donald
,
Hamilton
Lankford
,
Susanna
Loeb
, and
James
Wyckoff
.
2011
.
Teacher layoffs: An empirical illustration of seniority versus measures of effectiveness
.
Education Finance and Policy
6
(
3
):
439
454
. doi:10.1162/EDFP_a_00041.
Chetty
,
Raj
,
John N.
Friedman
, and
Jonah E.
Rockoff
.
2014
.
Measuring the impacts of teachers II: Teacher value-added and student outcomes in adulthood
.
American Economic Review
104
(
9
):
2633
2679
. doi:10.1257/aer.104.9.2633.
Chingos
,
Matthew M.
, and
Martin R.
West
.
2012
.
Do more effective teachers earn more outside the classroom?
Education Finance and Policy
7
(
1
):
8
43
. doi:10.1162/EDFP_a_00052.
Clotfelter
,
Charles
,
Elizabeth
Glennie
,
Helen
Ladd
, and
Jacob
Vigdor
.
2008
.
Would higher salaries keep teachers in high-poverty schools? Evidence from a policy intervention in North Carolina
.
Journal of Public Economics
92
(
5–6
):
1352
1370
. doi:10.1016/j.jpubeco.2007.07.003.
Clotfelter
,
Charles T.
,
Helen F.
Ladd
, and
Jacob
Vigdor
.
2005
.
Who teaches whom? Race and the distribution of novice teachers
.
Economics of Education Review
24
(
4
):
377
392
. doi:10.1016/j.econedurev.2004.06.008.
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
682
. doi:10.1016/j.econedurev.2007.10.002.
Clotfelter
,
Charles T.
,
Helen F.
Ladd
, and
Jacob L.
Vigdor
.
2010
.
Teacher credentials and student achievement in high school: A cross-subject analysis with student fixed effects
.
Journal of Human Resources
45
(
3
):
655
681
. doi:10.1353/jhr.2010.0023.
Clotfelter
,
Charles T.
,
Helen F.
Ladd
, and
Jacob L.
Vigdor
.
2011
.
Teacher mobility, school segregation, and pay-based policies to level the playing field
.
Education Finance and Policy
6
(
3
):
399
438
. doi:10.1162/EDFP_a_00040.
Cohen-Vogel
,
Lora
,
Li
Feng
, and
La’Tara
Osborne-Lampkin
.
2013
.
Seniority provisions in collective bargaining agreements and the ‘teacher quality gap’
.
Educational Evaluation and Policy Analysis
35
(
3
):
324
343
. doi:10.3102/0162373713482765.
Feng
,
Li
.
2009
.
Opportunity wages, classroom characteristics, and teacher mobility
.
Southern Economic Journal
75
(
4
):
1165
1190
.
Feng
,
Li
.
2010
.
Hire today, gone tomorrow: New teacher classroom assignments and teacher mobility
.
Education Finance and Policy
5
(
3
):
278
316
. doi:10.1162/EDFP_a_00002.
Feng
,
Li
.
2014
.
Teacher placement, mobility, and occupational choices after teaching
.
Education Economics
22
(
1
):
24
27
. doi:10.1080/09645292.2010.511841.
Feng
,
Li
,
David N.
Figlio
, and
Tim
Sass
.
2010
.
School accountability and teacher mobility
.
NBER Working Paper No. 16070
.
Florida Department of Education
.
2010
.
2010 guide to calculating school grades: Technical Assistance Paper
.
Tallahassee, FL
:
Bureau of Research and Evaluation, Florida Department of Education
.
Florida Department of Education
.
2011
.
Membership in Florida's public schools, fall 2011
.
Tallahassee, FL
:
Education Information and Accountability Services, Florida Department of Education
.
Glazerman
,
Steven M.
,
Ali
Protik
,
Teh
Bing-ru
,
Julie
Bruch
, and
Jeffrey
Max
.
2013
.
Transfer incentives for high-performing teachers: Final results from a multisite randomized experiment
. Available https://ies.ed.gov/ncee/pubs/20144003/pdf/20144003.pdf.
Accessed 11 August 2016
.
Goldhaber
,
Dan
,
Hyung-Jai
Choi
, and
Lauren
Cramer
.
2007
.
A descriptive analysis of the distribution of NBPTS-certified teachers in North Carolina
.
Economics of Education Review
26
(
2
):
160
172
. doi:10.1016/j.econedurev.2005.09.003.
Goldhaber
,
Dan
,
Bethany
Gross
, and
Daniel
Player
.
2011
.
Teacher career paths, teacher quality, and persistence in the classroom: Are public schools keeping their best?
Journal of Policy Analysis and Management
30
(
1
):
57
87
. doi:10.1002/pam.20549.
Gritz
,
R.
Mark
, and
Neil D.
Theobald
.
1996
.
The effects of school district spending priorities on length of stay in teaching
.
Journal of Human Resources
31
(
3
):
477
512
. doi:10.2307/146262.
Guarino
,
Cassandra M.
,
Mark D.
Reckase
, and
Jeffrey M.
Wooldridge
.
2015
.
Can value-added measures of teacher performance be trusted?
Education Finance and Policy
10
(
1
):
117
156
. doi:10.1162/EDFP_a_00153.
Hanushek
,
Eric A.
2011
.
The economic value of higher teacher quality
.
Economics of Education Review
30
(
3
):
466
479
. doi:10.1016/j.econedurev.2010.12.006.
Hanushek
,
Eric A.
,
John F.
Kain
, and
Steven G.
Rivkin
.
2004
.
Why public schools lose teachers
.
Journal of Human Resources
39
(
2
):
326
354
. doi:10.2307/3559017.
Hanushek
,
Eric A.
, and
Steven G.
Rivkin
.
2010
.
Constrained job matching: Does teacher job search harm disadvantaged urban schools
?
NBER Working Paper No. 15816
.
Harris
,
Douglas N.
, and
Tim R.
Sass
.
2011
.
Teacher training, teacher quality and student achievement
.
Journal of Public Economics
95
(
7–8
):
798
812
. doi:10.1016/j.jpubeco.2010.11.009.
Harris
,
Douglas N.
, and
Tim R.
Sass
.
2014
.
Skills, productivity and the evaluation of teacher performance
.
Economics of Education Review
40
:
183
204
. doi:10.1016/j.econedurev.2014.03.002.
Imazeki
,
Jennifer
.
2005
.
Teacher salaries and teacher attrition
.
Economics of Education Review
24
(
4
):
431
449
. doi:10.1016/j.econedurev.2004.07.014.
Ingersoll
,
R. M.
2001
.
Teacher turnover and teacher shortages: An organizational analysis
.
American Educational Research Journal
38
(
3
):
499
534
. doi:10.3102/00028312038003499.
Isenberg
,
Eric
,
Jeffrey
Max
,
Philip
Gleason
,
Liz
Potamites
,
Robert
Santillano
,
Heinrich
Hock
, and
Michael
Hansen
.
2013
.
Access to effective teaching for disadvantaged students (NCEE 2014–4001)
.
Washington, DC
:
U.S. Department of Education
.
Jackson
,
C. Kirabo
.
2012
.
Match quality, worker productivity, and worker mobility: Direct evidence from teachers
.
Review of Economics and Statistics
95
(
4
):
1096
1116
. doi:10.1162/REST_a_00339.
Jackson
,
C. Kirabo
, and
Elias
Bruegmann
.
2009
.
Teaching students and teaching each other: The importance of peer learning for teachers
.
American Economic Journal. Applied Economics
1
(
4
):
85
108
. doi:10.1257/app.1.4.85.
Kane
,
Thomas J.
,
Jonah E.
Rockoff
, and
Douglas O.
Staiger
.
2008
.
What does certification tell us about teacher effectiveness? Evidence from New York City
.
Economics of Education Review
27
(
6
):
615
631
. doi:10.1016/j.econedurev.2007.05.005.
Kane
,
Thomas J.
, and
Douglas O.
Staiger
.
2008
.
Estimating teacher impacts on student achievement: An experimental evaluation
.
NBER Working Paper No. 14607
.
Krieg
,
John M.
2006
.
Teacher quality and attrition
.
Economics of Education Review
25
(
1
):
13
27
. doi:10.1016/j.econedurev.2004.09.004.
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
(
1
):
37
62
. doi:10.3102/01623737024001037.
Mas
,
Alexandre
, and
Enrico
Moretti
.
2009
.
Peers at work
.
American Economic Review
99
(
1
):
112
145
. doi:10.1257/aer.99.1.112.
McCaffrey
,
Daniel F.
,
Tim R.
Sass
,
J. R.
Lockwood
, and
Kata
Mihaly
.
2009
.
The intertemporal variability of teacher effect estimates
.
Education Finance and Policy
4
(
4
):
572
606
. doi:10.1162/edfp.2009.4.4.572.
Mihaly
,
Kata
,
Daniel
McCaffrey
,
Tim R.
Sass
, and
J. R.
Lockwood
.
2013
.
Where you come from or where you go? Distinguishing between school quality and the effectiveness of teacher preparation program graduates
.
Education Finance and Policy
8
(
4
):
459
493
. doi:10.1162/EDFP_a_00110.
Mont
,
Daniel
, and
Daniel
Rees
.
1996
.
The influence of classroom characteristics on high school teacher turnover
.
Economic Inquiry
34
(
1
):
152
167
. doi:10.1111/j.1465-7295.1996.tb01369.x.
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
124
. doi:10.2307/145729.
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
(
5
):
507
518
. doi:10.1016/j.econedurev.2004.01.005.
Reed
,
Deborah
,
Kim S.
Rueben
, and
Elisa
Barbour
.
2006
.
Retention of new teachers in California
.
San Francisco
:
Public Policy Institute of California
.
Rivkin
,
Steven G.
,
Eric A.
Hanushek
, and
John F.
Kain
.
2005
.
Teachers, schools, and academic achievement
.
Econometrica
73
(
2
):
417
458
. doi:10.1111/j.1468-0262.2005.00584.x.
Rockoff
,
Jonah E.
2004
.
The impact of individual teachers on student achievement: Evidence from panel data
.
American Economic Review
94
(
2
):
247
252
. doi:10.1257/0002828041302244.
Sass
,
Tim R.
2006
.
Charter schools and student achievement in Florida
.
Education Finance and Policy
1
(
1
):
91
122
. doi:10.1162/edfp.2006.1.1.91.
Sass
,
Tim R.
,
Jane
Hannaway
,
Zeyu
Xu
,
David N.
Figlio
, and
Li
Feng
.
2012
.
Value added of teachers in high-poverty schools and lower poverty schools
.
Journal of Urban Economics
72
(
2–3
):
104
122
. doi:10.1016/j.jue.2012.04.004.
Sass
,
Tim R.
,
Anastasia
Semykina
, and
Douglas N.
Harris
.
2014
.
Value-added models and the measurement of teacher productivity
.
Economics of Education Review
38
:
9
23
. doi:10.1016/j.econedurev.2013.10.003.
Scafidi
,
Benjamin
,
David L.
Sjoquist
, and
Todd R.
Stinebrickner
.
2007
.
Race, poverty, and teacher mobility
.
Economics of Education Review
26
(
2
):
145
159
. doi:10.1016/j.econedurev.2005.08.006.
West
,
Martin R.
, and
Matthew M.
Chingos
.
2009
.
Teacher effectiveness, mobility, and attrition in Florida
. In
Performance incentives: Their growing impact on American k--12 education
, edited by
Matthew G.
Springer
, pp.
251
272
.
Washington, DC
:
Brookings Institution Press
.

Appendix A

Table A.1. 
Summary Statistics and Correlation of Teacher Quality Indicators in Math and Reading
Teacher Quality
Summary StatisticsCorrelations: MathCorrelations: Reading
Indicator and SubjectMeanS.D.MinMaxQ1-MathQ2-MathQ3-MathQ4-MathQ5-MathQ6-MathQ1-ReadingQ2-ReadingQ3-ReadingQ4-ReadingQ5-Reading
Q1-Math 0.000 0.262 −2.860 3.319 —           
Q2- Math 0.000 0.279 −2.702 3.201 0.806 —          
Q3- Math 0.000 0.427 −5.084 4.689 0.746 0.607 —         
Q4- Math 0.009 0.194 −2.846 3.344 0.707 0.598 0.524 —        
Q5- Math 0.026 0.214 −2.624 3.185 0.569 0.743 0.440 0.801 —       
Q6- Math 0.006 0.282 −3.773 3.861 0.585 0.500 0.637 0.814 0.672 —      
Q1-Reading 0.000 0.253 −3.508 2.568 0.279 0.217 0.205 0.221 0.198 0.186 —     
Q2-Reading 0.000 0.258 −3.022 2.111 0.242 0.431 0.187 0.207 0.381 0.190 0.806 —    
Q3-Reading 0.000 0.460 −5.361 5.457 0.173 0.123 0.205 0.138 0.127 0.159 0.682 0.540 —   
Q4-Reading 0.008 0.187 −3.524 2.539 0.244 0.212 0.186 0.335 0.290 0.276 0.681 0.559 0.472 —  
Q5-Reading 0.024 0.201 −3.065 2.118 0.208 0.349 0.177 0.281 0.499 0.262 0.535 0.729 0.365 0.770 — 
Q6-Reading 0.004 0.291 −4.277 4.597 0.181 0.138 0.189 0.249 0.214 0.287 0.537 0.425 0.611 0.778 0.589 
Teacher Quality
Summary StatisticsCorrelations: MathCorrelations: Reading
Indicator and SubjectMeanS.D.MinMaxQ1-MathQ2-MathQ3-MathQ4-MathQ5-MathQ6-MathQ1-ReadingQ2-ReadingQ3-ReadingQ4-ReadingQ5-Reading
Q1-Math 0.000 0.262 −2.860 3.319 —           
Q2- Math 0.000 0.279 −2.702 3.201 0.806 —          
Q3- Math 0.000 0.427 −5.084 4.689 0.746 0.607 —         
Q4- Math 0.009 0.194 −2.846 3.344 0.707 0.598 0.524 —        
Q5- Math 0.026 0.214 −2.624 3.185 0.569 0.743 0.440 0.801 —       
Q6- Math 0.006 0.282 −3.773 3.861 0.585 0.500 0.637 0.814 0.672 —      
Q1-Reading 0.000 0.253 −3.508 2.568 0.279 0.217 0.205 0.221 0.198 0.186 —     
Q2-Reading 0.000 0.258 −3.022 2.111 0.242 0.431 0.187 0.207 0.381 0.190 0.806 —    
Q3-Reading 0.000 0.460 −5.361 5.457 0.173 0.123 0.205 0.138 0.127 0.159 0.682 0.540 —   
Q4-Reading 0.008 0.187 −3.524 2.539 0.244 0.212 0.186 0.335 0.290 0.276 0.681 0.559 0.472 —  
Q5-Reading 0.024 0.201 −3.065 2.118 0.208 0.349 0.177 0.281 0.499 0.262 0.535 0.729 0.365 0.770 — 
Q6-Reading 0.004 0.291 −4.277 4.597 0.181 0.138 0.189 0.249 0.214 0.287 0.537 0.425 0.611 0.778 0.589 

Notes: Q1 = Value-added model with complete persistence, observable student characteristics, time-varying teacher quality; Q2 = Value-added model with partial persistence, observable student characteristics, time-varying teacher quality; Q3 = Value-added model with complete persistence, student fixed effects, time-varying teacher quality; Q4 = Value-added model with complete persistence, observable student characteristics, time-invariant teacher quality; Q5 = Value-added model with partial persistence, observable student characteristics, time-invariant teacher quality; Q6 = Value-added model with complete persistence, student fixed effects, time-invariant teacher quality.

Figure A.1.

Math Teacher Quality by Job Separation Reasons.

Figure A.1.

Math Teacher Quality by Job Separation Reasons.