Teacher Mobility and Financial Incentives

Abstract

Extensive teacher mobility can undermine policy efforts to develop a high-quality workforce. In response, policymakers have increasingly championed financial incentives to retain teachers. In 2006, the Denver Public Schools adopted an alternative teacher compensation reform, the Professional Compensation System for Teachers (“ProComp”). Using longitudinal teacher-level data from 2001–2002 to 2010–2011, I estimate hazard models that identify the relationship between ProComp and teacher mobility. Specifically, I compare mobility patterns of teachers who received a ProComp incentive with those who did not, with special attention to teacher mobility in high-poverty schools. Results suggest receiving a ProComp incentive is associated with a significant decrease in the odds of departure. This appears to be driven by a decrease in a teacher’s odds of leaving the district rather than moving to a new school within the district, by voluntary ProComp participants and by teachers who receive incentives that total more than $5,000.

Keywords

financial incentives teacher retention

There is currently much interest in increasing access to high-quality teachers (Clotfelter, Ladd, & Vigdor, 2010; Hanushek, 2007) through improved recruitment and retention. This study focuses on teacher mobility and describes the relationship between mobility and financial incentives. Prior studies have shown that it is difficult to retain teachers, particularly in high-poverty schools (Boyd et al., 2011; Ingersoll, 2004). Although there is no definitive method for how best to retain teachers, research suggests one reason for teacher turnover is dissatisfaction with low salaries (Ingersoll, 2001). Thus, it is possible that by offering teachers financial incentives and increasing their total compensation, they would become more satisfied with their jobs and turnover would decrease, ceteris paribus.

Financial incentives for teachers, either as part of an alternative compensation program or as a stand-alone bonus, have been championed by policymakers as one way to make the teaching profession and specific schools or subjects more attractive to more people. Although improved retention is not a silver bullet for providing all children with access to high-quality teachers and a better education, investigating whether financial incentives are associated with a change in teacher mobility is an essential first step in exploring this question. Moreover, mobility is an important outcome to consider because district and school administrators must often fill vacancies with new teachers, who have been shown to be less effective in their first years on the job than their more experienced counterparts (Clotfelter, Ladd, & Vigdor, 2006; Rivkin, Hanushek & Kain, 2005).

Denver’s Professional Compensation System for Teachers (“ProComp”) is one of the most prominent alternative teacher compensation reforms in the nation.¹ Through a combination of 10 financial incentives, ProComp seeks to increase student achievement by motivating teachers to improve their instructional practices and by attracting and retaining high-quality teachers to work in the district. This study examines the extent to which ProComp is associated with changes in teacher turnover by studying teacher mobility in Denver from 2001–2002 to 2010–2011. This panel allows me to explore the mobility of teachers, including moves from one school to another within the district, as well as exits out of the Denver Public Schools workforce. I pay special attention to the differential effects of receiving a ProComp financial incentive on teacher mobility at high-poverty schools.

In this paper, I provide an empirical foundation on which to build models of teacher mobility under the availability of a financial incentive program. The analytic tasks carried out in this study are driven by two key research questions:

Research Question 1: To what extent is ProComp associated with changes in teacher turnover from all schools in the district and from high-poverty schools specifically?

Research Question 2: How do the effects of ProComp vary for teachers’ within-district moves and teachers’ exits from the district?

Study Context

Denver Public Schools (DPS) is the second-largest school district in Colorado, serving approximately 80,000 students in 152 schools, with roughly 4,500 teachers in the 2010–2011 school year. The majority of students who attend DPS are Latino/a, and over 72% of DPS students qualify for the federally sponsored free or reduced-price lunch (FRL) program, a commonly used indicator of poverty.² Like many large, urban districts, DPS is beset with low student test-score achievement and a shortage of high-quality teachers (Torres, 2013). In terms of demographics, DPS is comparable with Seattle Public Schools (Washington) and San Diego Unified School District (California). Accordingly, results from this study may be generalizable to those and other large, urban districts, particularly in the West and Southwest.

ProComp has gained national attention as one of the only teacher compensation reforms jointly conceived of and implemented by the district and local teachers’ union (Gonring, Teske, & Jupp, 2007), and the comprehensive design of ProComp was championed by President Obama as a model of teacher compensation reform. The program received a $22.6 million Teacher Incentive Fund grant and was backed by Denver voters through a 2005 referendum to levy an additional $25 million in annual taxes to pay for the program.

Starting January 1, 2006,³ incumbent teachers could opt into ProComp or continue to be paid based on the DPS single-salary schedule. Teachers who were hired on or after this date were required to participate in ProComp. ProComp incentives range from a low of $376 for meeting annual student growth objectives, to a high of $3,379 for earning an advanced degree or specialty license (see online Appendix for ProComp incentive details).⁴ On average, teachers who participate in ProComp received $4,409 in financial incentives during the 2009–2010 school year. These incentives are received as a monthly supplement to teachers’ base salaries, which start at $37,551, for a teacher with a bachelor’s degree and no prior teaching experience, and average $52,845 (Denver Public Schools, 2011). There is no limit to the number of incentives a teacher can receive in a single year; however some are specific to subject/grade or school.

Under ProComp, teachers who work in high-poverty schools (called “hard-to-serve” [HTS] schools) are eligible to receive a targeted retention incentive. Schools are identified as high poverty based on percentages of FRL eligibility as follows: 87% or more for elementary and K–8 schools; 85% or more for middle schools; and 75% or more for high schools. These varying cut-offs are set such that roughly half of all DPS schools are identified as high poverty in a given year.

Mobility and Financial Incentives

For the better part of the past decade, policymakers have been searching for ways to reduce teacher turnover. The extensive body of prior research on teacher mobility has shown that, although the overall turnover of teachers may be comparable with other professions (Henke, Zahn, & Carroll, 2001), certain schools have a very difficult time retaining teachers. The most recent report of teacher turnover patterns published by the National Center for Education Statistics (NCES) estimates that 17% of elementary and secondary teachers left public and private schools at the end of the 2003–2004 school year (Panty et al., 2008). Of those exits, approximately half (8%) were due to transfers to a different school; the other half (9%) were because teachers left the profession. However, on average, high-poverty public schools have higher rates of turnover than low-poverty public schools (21% vs. 14%, respectively).⁵

Such patterns are well-established in the research literature: Schools with higher teacher turnover are likely to serve large populations of low-income, non-White, and low-performing students—the very students most in need of a steady and supportive education (Boyd et al., 2011; Clotfelter et al., 2006; Hanushek, Kain, & Rivkin, 2004; Wyckoff, Boyd, Lankford, & Loeb, 2003). Although extant research has characterized the problem of high teacher turnover, it has done little to identify successful avenues of reform to reduce turnover.

Evidence on Financial Incentives and Mobility

Extant research offers evidence to suggest overall teacher compensation is positively associated with retention (Hanushek et al., 2004; Imazeki, 2005; Ingersoll, 2001; Kirby, Berends, & Naftel, 1999; Lankford, Loeb, & Wyckoff, 2002). For example, Kirby et al. (1999) found that a $1,000 increase in salary was associated with a 3% to 6% decrease in the odds of departure. Similarly, Ingersoll (2001) found—after controlling for teacher characteristics—that a $1,000 difference in compensation was associated with a decrease of 3% in the odds of voluntary teacher departure. Finally, Hanushek et al., (2004) reported that a 10% increase in teacher pay reduces the probability of departure from a school by one to four percentage points.

On the basis of this evidence—that teachers who earn more are more likely to stay at their current schools—it is tempting to assume that the availability of financial incentives will yield similar results. All else being equal, one might expect financial incentives to mitigate turnover. Yet there is reason to question this assumption because there are many differences between uniform salary increases and the potential to earn additional pay through financial incentive programs (Fulbeck, 2012).

Empirical research that examines the effect of alternative teacher compensation programs on teacher mobility and retention is inconclusive. For example, researchers found no difference in teacher turnover at the district or school level as a result of Chicago’s Teacher Advancement Program (TAP) (Glazerman & Seifullah, 2010). However, results from the evaluations of alternative teacher compensation programs in Texas have been varied. These programs allow Texas districts and schools to set their own incentive award amounts. Perhaps, not surprisingly, researchers found that there were fewer turnovers among teachers who received large bonuses. However, they also found sharp increases in turnover associated with smaller incentives and for teachers who did not receive an incentive (Springer, Lewis, Podgursky, Ehlert, Gronberg, et al., 2009; Springer, Lewis, Podgursky, Ehlert, Taylor, et al., 2009).

Researchers studying a bonus program in North Carolina that specifically targeted math, science, and special education teachers who worked in high-poverty or low-achieving secondary schools, found a $1,800 annual incentive was associated with a 17% decrease in mean turnover rates (Clotfelter, Glennie, Ladd, & Vigdor, 2008). The largest effects were observed for experienced teachers. Researchers speculate that the effect may also have been negatively biased because many teachers indicated they did not understand the eligibility criteria of the program.

Conversely, Steele, Murnane, and Willett (2009) found no effect on turnover for new teachers who received the Governor’s Teaching Fellowship, a fairly large incentive ($5,000 each year for 4 years) available to academically talented teachers, who worked in low-achieving, high-poverty schools in California. More recently, Hough, Loeb, and Plank (2011) examined survey and interview data from teachers and principals about San Francisco’s Quality Teacher and Education Act (QTEA) alternative compensation program. Under QTEA, teachers earn a variety of financial incentives, including a $2,000 annual bonus for working in hard-to-staff schools. The authors found a majority of all teachers (not just those eligible to receive the hard-to-staff bonus) reported QTEA did not affect their decision to stay in their current school.

Like Hough et al. (2011), Fulbeck (2011) found that most DPS teachers do not consider ProComp incentives to be important when making their career decisions. Rather, teachers indicate that other nonpecuniary working conditions were more important than ProComp when considering their career options. This has been well documented in the research literature (Boyd et al., 2011; Hanushek et al., 2004; Milanowski et al., 2009; Wyckoff et al., 2003) and highlights the importance of accounting for nonpecuniary teacher and school characteristics in models of teacher mobility.

To sum, although overall compensation appears to minimize teacher turnover, there is mixed evidence about whether financial incentives are associated with reduced teacher mobility.

Method

Data

This study draws on teacher-level data from the 2001–2002 to 2010–2011 school years. Thus, there are nine transition periods between school years, in which a teacher could either remain in the same school or not: Four of these transition periods occur before the implementation of ProComp, whereas five occur after implementation.

Data contain longitudinally linked teacher professional assignment records and information on teacher and school characteristics. At the teacher-level, these include information on ProComp participation status, ProComp participation type (i.e., voluntary or compulsory), total incentives received, base salary, race/ethnicity, gender, years of teaching experience, probationary status, degree attainment, and age.⁶ At the school-level, records contain information on grade-level, the percentage of FRL-eligible students, and average student growth percentiles (SGPs)⁷ in reading and mathematics on the state summative assessment.

Teachers included in this study were limited to those eligible to participate in ProComp (regardless of whether they chose to do so) and those who were at least minimally competent to choose whether to remain teaching at their school from 1 year to the next. ProComp eligibility is restricted to teachers who are members of the teacher’s union (roughly 90% of DPS teachers were members in 2010–2011) and who do not work at charter schools. Due to the participation requirements for new hires to join ProComp, and the veteran teachers who chose to voluntarily opt in, the percentage of DPS teachers who participated in ProComp increased steadily from 42% to 77% during the period of study. Teachers who were terminated, retired, or left their schools involuntarily at the end of a given year because of a reduction in available teaching positions were excluded from analysis.

Constraining the population of teachers according to these criteria yielded a sample of roughly 3,900 to 4,200 teachers (from 4,555 total teachers in 2010–2011) each year. The total sample included 29,234 person-year observations from 7,333 different teachers. Over the study period, there were 2,502 cases of teachers transferring to a new school within DPS and 6,101 cases of teachers exiting DPS. In total, 70% of teachers in the study sample never moved schools, 9% moved schools but never exited DPS, and 21% exited DPS.⁸

Table 1 reports sample statistics for the 2009–2010 school year. On average, teachers received $4,409 in ProComp incentives or roughly 12% of the average base salary.

Table 1

Descriptive Statistics of Sample, 2009–2010

Teacher characteristics	M
Turnover
Overall departures	26.6%
Within-district moves	6.0%
Exits from district	20.6%
Compensation
ProComp participants	70.1%
Voluntary participants	27.9%
Compulsory participants	42.1%
All teachers: Received incentive	28.7%
Participants: Received incentive	96.6%
ProComp incentives	4,408.8
Base salary	52,867.9
Total compensation	57,276.7
Demographics
Female	73.6%
Minority	21.3%
Age years	42.6
Some graduate degree	52.3%
Probationary	28.2%
Years of teaching experience	10.0
School characteristics
High-poverty school	48.4%
Secondary school	41.2%
FRL eligible students	72.6%
Prior year reading SGPs	50.2
Prior year math SGPs	51.6
2009–2010 N	3,146
Total N in analytic sample	29,234

Note. FRL = free or reduced-price lunch; SGP = student growth percentile.

The percentage of ProComp participants who received at least one incentive in any given year was over 95%. Thus, almost all ProComp participants received some supplementary compensation through ProComp. However, there was considerable variation in the total amount of incentives ProComp participants received. Table 2 shows descriptive statistics of total ProComp incentive amounts received for ProComp participants and for the full sample of teachers. Although the modal amount received by all teachers was just under $4,000, some teachers received considerably more, resulting in a positively skewed distribution.

Table 2

Incentive Amount Received Over Time

Year	Minimum	Maximum	ProComp participants		Full sample (N = 7,333)
			M (SD)	Median	M (SD)	Median
2005–2006, n = 662	0	6,328	1,647 (1,165)	1,332	374 (885)	0
2006–2007, n = 1,093	0	7,866	2,429 (1,402)	2,394	885 (1,442)	0
2007–2008, n = 1,622	0	9,603	2,904 (1,648)	2,846	1,551 (1,884)	711
2008–2009, n = 1,987	0	16,340	6,635 (2,893)	6,887	4,351 (3,927)	4,542
2009–2010, n = 2,204	0	20,153	6,293 (3,140)	5,934	4,408 (3,901)	4,658

To serve as a basis for understanding, Figure 1 shows teachers’ departure rates plotted over time as a function of receiving an incentive and school poverty. This yields four subgroups: Vertical lines are superimposed on to the plot to indicate the implementation of ProComp (2005–2006) and the start of the Great Recession (2008–2009), respectively.

Figure 1.

Teachers’ overall departure rates, 2000–2001 to 2009–2010.

The plot suggests that, although there is considerable variability year-to-year, departure rates that tend to trend upward during most of the study fall sharply after the start of the Great Recession. As we would expect, this also shows that departure rates are higher for teachers who work in high-poverty schools. Importantly, veteran teachers who participate in ProComp, especially in the first year after implementation (i.e., the voluntary participants), have much lower rates of departure. This raises questions about selection to ProComp and is further examined in analyses presented below.

Mobility Outcomes

Teacher mobility may refer to one of three possible outcomes: (a) overall departures, (b) moves to another school within the district (within-district moves), and (c) moves out of DPS altogether (exits). Examining overall departures as an outcome is appealing because it is parsimonious and because—for a school—the impact of within-district moves and exits are the same: the teacher leaves the school. But as others have noted (e.g., Goldhaber, Gross, & Player, 2011), there are at least two reasons for disaggregating overall departures and considering within-district moves and exits separately. First, each type of move may be motivated by different factors. Second, different types of moves have different district-level consequences.

For the teacher, within-district moves may result in substantial changes in working conditions, including the subject/grade taught, principal and colleagues, and student demographic characteristics that change from one school to another. Indeed, research suggests teachers are sensitive to working conditions (Ingersoll & Smith, 2003). Specifically, Boyd and colleagues (2011) found that principals have a significant impact on teacher mobility and other studies have found an association between student characteristics and turnover (Guarino, Santibanez, & Daley, 2006; Hanushek et al., 2004; Lankford et al., 2002).

Traditionally, these within-district moves do not result in a change in salary because salaries are set using a common district-wide schedule, based on years of teaching experience and postbaccalaureate educational attainment. However, after the implementation of ProComp, teachers may be more likely to receive additional pay at certain schools in DPS⁹ and, moreover, not all teachers participate in ProComp. Thus, ProComp creates both within- and between-school variation in compensation. Accordingly, within-district moves for ProComp participants may result in changes in working conditions and salary.

Teachers exiting the system may also be seeking a change in working conditions and salary or may have decided they no longer wish to pursue the profession.¹⁰ Unlike within-district moves, exits from the system are more likely to involve greater costs for teachers associated with learning a new curriculum, district culture, and human resources protocols and policies (for those who stay in teaching), and in many cases, relocation (for those who stay and those who leave the teaching profession).

In addition to differences described above, within-district moves and exits are likely to have different impacts on DPS. Exits from DPS represent a loss to the teacher workforce, whereas within-district moves may indicate a possible redistribution of teachers within the system. Accordingly, within-district moves and exits have different costs associated with recruitment, hiring, and training new teachers (for details of the costs of teacher turnover, see Milanowski & Odden, 2007). Given the important differences in motivation, costs, and impacts of each type of mobility, the marginal effect of ProComp may differ for within-district moves and exits, respectively.

Models of Teacher Mobility

I have used competing hazard models to estimate the risk that individual teachers depart from their current teaching position given their receipt of a ProComp incentive, and other individual characteristics, and the characteristics of the school in which they teach. Hazard models are conceptually appealing for studies of teacher mobility. They predict the probability that a teacher departs from their school after year t, conditional on the length of time the teacher has been at a school, which I count in 1-year increments. Each model represents time as discrete because the school year provides the field with an annual hiring cycle during which most new hires, transfers, or exits occur. Accounting for time is important because new teachers are significantly more likely to move or leave than those who have more teaching experience. These models provide some flexibility with the data by allowing me to analyze the movement of teachers without necessarily viewing the entire career of all teachers. That is, these models can describe teacher mobility with censored data.¹¹ The models also allow the effects of explanatory variables to differ depending on the type of move the teacher makes. All models are estimated by a conditional maximum likelihood logit.

My analytic strategy has two basic components: I compare hazard rates (a) between teachers who receive a ProComp incentive and those who do not, and (b) across teachers who work in high-poverty schools and those who work in schools that are not high-poverty. A representative estimating equation is:

\log i t [λ (t_{i s})] = α + β_{1} T i m e + β_{2} \Pr o C o m p_{i} + β_{3} H i g h P o v_{s} + β_{4} \Pr o C o m p_{i} \times H i g h P o v_{s} + X_{i s} + ε_{i s}

where i indexes individuals and s indexes schools. The dependent variable λ(t_is) is the probability that a teaching spell at a given school ends at the close of time t as a function of the spell having lasted through time t. The variable Time is a continuous measure that ranges from 2001–2002 to 2009–2010.¹² The variable ProComp_i indicates whether the teacher in question received a ProComp incentive. The variable High Pov_s indicates whether that teaching spell took place in a high-poverty school. Finally, the matrix X_is contains assorted teacher and school characteristics.

The coefficient β₂ represents the estimate of ProComp’s association with the departure hazard for teachers from all schools in the district and addresses the first part of the first research question. The two-way interaction term in Equation (1) represents the differential association of receiving a ProComp incentive while teaching in a high-poverty school. Thus, the coefficient β_4 represents the estimate of ProComp’s association with the departure hazard for teachers in high-poverty schools and addresses the second part of the first research question. To address the second research question, I compute these competing risk models separately for teachers’ within-district moves and teachers’ exits from the district. I estimate these competing risk models with separate logit regressions, instead of using a multinomial logit,¹³ and report models that include school fixed effects.¹⁴ Because teachers can at times stay and at other times move or leave, individual teachers can be identified as stayers, movers, or leavers at different points in their career.

In analyzing the effects of ProComp on teachers’ overall departures, within-district moves, and exits from the district, I estimated the models incrementally as follows: (a) including only the main effect for ProComp and working in a high-poverty school; (b) adding an interaction term to account for the possibility that ProComp may have a differential effect on teacher mobility in high-poverty schools; and (c) adding individual teacher characteristics and a set of variables that describe school context. Adding teacher and school characteristics to the model adds explanatory power to the model (not presented) and is statistically significant.¹⁵ For the sake of parsimony, only the full models—which are most similar to those typically estimated in the existing empirical literature—are presented in the results below.

This analytic approach provides evidence of the association between ProComp and teacher mobility; it does not allow me to estimate the causal effect of ProComp because the comparison group of teachers—those who did not earn incentives—differs from the treatment group in key ways that may influence their turnover behavior (e.g., performance, pursuit of professional development opportunities) independent of the financial incentives they earn under ProComp. Nonetheless, these findings are important because they provide valuable evidence about the association between incentives of varying amounts and mobility of teachers who work at schools with varying levels of poverty. These findings help to establish an empirical foundation on which to build causal models of teacher mobility under the availability of financial incentives.

Findings

The estimated coefficients, standard errors, and odds ratios from the competing hazard model are presented in Table 3. I begin my analysis of teacher mobility by focusing on a teacher’s overall odds of departure from a school in any given year (Columns 1–3 of Table 3). On average, I find that receipt of a ProComp incentive is associated with a 30% decrease in a teacher’s odds of departure. This estimate is both practically and statistically significant. As one would expect, estimates suggest that teachers who work in high-poverty schools have greater odds of departure (a 27% increase) than teachers who work in schools that serve relatively wealthier students. This is true even if teachers receive a ProComp incentive, although the difference in the odds of departure is smaller (a 20% increase) for ProComp teachers in high-poverty schools relative to non-ProComp teachers who also work in these schools.

Table 3

Hazard Ratio Estimates From Models of Teachers’ Overall Departures, Moves, and Exits

Parameter	Overall departures			Within-district moves			Exits from district
	B	SE	Exp(B)	B	SE	Exp(B)	B	SE	Exp(B)
Time	0.000	0.015	1.000	−0.077	0.027	0.926**	0.018	0.017	1.019
ProComp incentive	−0.360	0.057	0.698***	0.154	0.099	1.167	−0.501	0.065	0.606***
High poverty	0.241	0.059	1.273***	0.442	0.111	1.555***	0.032	0.066	1.033*
ProComp incentive × High poverty	0.202	0.084	1.224*	−0.119	0.136	0.888	0.351	0.094	1.420***
Teacher characteristics
Salary/$1,000	−0.011	0.003	0.989**	−0.004	0.005	0.996	−0.015	0.004	0.986***
Female	0.157	0.045	1.170***	0.028	0.074	1.028	0.137	0.051	1.147**
Minority	0.089	0.048	1.093	0.097	0.077	1.101	0.048	0.053	1.049
Age years	0.013	0.002	1.013***	0.008	0.004	1.008*	0.010	0.003	1.010***
Master’s degree +	−0.028	0.047	0.973	0.041	0.079	1.042	−0.048	0.052	0.953
Probationary	0.372	0.074	1.450***	−0.048	0.130	0.953	0.379	0.081	1.461***
Years of experience	−0.017	0.013	0.983	0.049	0.023	1.050*	−0.046	0.014	0.955**
Years of experience2	0.000	0.000	1.000	−0.002	0.001	0.998**	0.002	0.000	1.002***
School characteristics
Secondary	−0.207	0.113	0.813	0.454	0.189	1.575*	−0.039	0.125	0.962
Average reading SGPs in t–1	−0.010	0.003	0.990***	−0.023	0.005	0.977***	−0.007	0.003	0.994*
Average math SGPs in t–1	−0.008	0.002	0.992***	−0.014	0.004	0.986***	−0.005	0.002	0.995*
–2 log likelihood	15,133.623			6,929.783			12,918.091

Note. All models include school fixed effects. N = 12,952. SGP = student growth percentile.

***

A hazards ratio significantly different from 1 at the 0.1% level; **at the 1% level; and *at the 5% level.

Thus, receiving a ProComp incentive is associated with lower departure odds for teachers in schools that are not high poverty than for teachers who work in high-poverty schools (a 30% decrease vs. a 7% decrease). The magnitude of the estimates on teacher’s hazard departure rates is in line with what others have found (e.g., Ingersoll, 2001; Kirby et al., 1999) for teachers who work in high-poverty schools but substantially larger than extant estimates for teachers who work in schools that are not high poverty.

Moves within DPS may have lower potential costs for teachers than exits from DPS. In addition, ProComp teachers may seek to maximize the amount of financial incentives they can receive by moving to schools designated to receive one or more school-wide incentives. However, results from models of within-district moves (Columns 4–6 of Table 3) do not suggest that teachers who receive a ProComp incentive have greater odds of making a within-district move relative to teachers who do not receive an incentive. As expected, teachers are much more likely (a 50% increase) to make a within-district move if they work in a high-poverty school but there does not appear to be a differential association with ProComp. However, it may still be the case that ProComp teachers are making more strategic moves to schools that qualify for more school-wide incentives. Further exploration of ProComp teachers’ within-district moves is beyond the scope of this paper, although such questions are explored in subsequent work (Fulbeck & Richards, 2013).

Teachers leaving DPS reflect a net loss of public school teacher resources to the district. These specifications (Columns 7–9 of Table 3) suggest the negative association between ProComp and overall departures is driven by a significant decrease in the odds that teachers who receive a ProComp incentive leave the district. Estimates suggest receipt of a ProComp incentive is associated with a 39% decrease in a teacher’s odds of leaving the district. Teachers appear slightly more likely (3% increase) to leave if they work in a high-poverty school and there is not a differential association for ProComp on these odds.

Tests of Model Sensitivity

The results described above reflect several decisions about the methods for estimating ProComp effects. To test the sensitivity of the results to these decisions, I explore several alternative specifications. First, I test the robustness of the results to a more nuanced parameterization of ProComp that varies according to the total dollar amount of the incentives received. In these alternative specifications, I first test a continuous measure of the total incentive amount per $1,000 and then test a categorical measure where the total incentive amount is identified as either small ($1–$5,000), moderate ($5,001–$10,000), or large ($10,001 or greater). Second, I explore whether results vary by ProComp participation type (i.e., voluntary vs. compulsory).

Table 4 presents results from the first set of sensitivity checks. For parsimony, only models of teachers’ overall departures are presented. The first alternative specification (Columns 1–3 of Table 4) suggests there is a negative association (roughly 7%) between the odds of departure and every $1,000 in incentives received. This association does not significantly differ by school poverty. The second alternative specification (Columns 4–6 of Table 4) provides greater nuance about the association between ProComp and teacher mobility. Results suggest that only ProComp incentives over $5,000 are associated with a significant decrease in teachers’ odds of departure. Receipt of a moderate incentive is associated with a 36% decrease in the odds of departure, whereas receipt of a large incentive is associated with a 50% decrease in the odds of departure. Again, this relationship does not significantly differ by school poverty.

Table 4

Sensitivity Tests of Incentive Amount on Hazard Ratio Estimates From Models of Teachers’ Overall Departures

Parameter	ProComp incentive/$1,000			ProComp incentive categories
	B	SE	Exp(B)	B	SE	Exp(B)
Time	0.025	0.018	1.025	0.017	0.018	1.018
ProComp incentive/$1000	−0.074	0.011	0.929***
Small incentive ($1,000–$5,000)				−0.029	0.063	0.971
Moderate incentive ($5,001–$10,000)				−0.453	0.089	0.636***
Large incentive ($10,001+)				−0.694	0.220	0.500**
High-poverty	0.317	0.056	1.373***	0.265	0.064	1.304***
ProComp incentive × High-poverty
ProComp incentive/$1,000 × High-poverty	0.014	0.014	1.014
Small incentive × High-poverty				0.168	0.100	1.183
Medium incentive × High-poverty				0.165	0.123	1.179
Large incentive × High-poverty				−0.091	0.269	0.913
Teacher characteristics
Salary/$1,000	−0.013	0.003	0.988***	−0.015	0.004	0.985***
Female	0.149	0.049	1.160**	0.144	0.049	1.155**
Minority	0.089	0.050	1.093	0.095	0.050	1.100
Age years	0.013	0.002	1.013***	0.014	0.002	1.014***
Master’s degree +	0.037	0.050	1.038	0.061	0.051	1.063
Probationary	0.338	0.079	1.402***	0.281	0.080	1.325***
Years of experience	−0.022	0.014	0.978	−0.016	0.014	0.984
Years of experience²	0.001	0.000	1.001	0.000	0.000	1.008
School characteristics
Secondary	−0.078	0.120	0.925	−0.104	0.120	0.901
Average reading SGPs in t–1	−0.009	0.003	0.991**	−0.008	0.003	0.992**
Average math SGPs in t–1	−0.007	0.002	0.993**	−0.007	0.002	0.993**
–2 log likelihood	13,436.778			13,435.212

Note. All models include school fixed effects. N = 11,552. SGP = student growth percentile.

***

A hazards ratio significantly different from 1 at the 0.1% level; **at the 1% level; and *at the 5% level.

The second set of sensitivity checks on teacher mobility explores the extent to which there is a differential association between ProComp and mobility for voluntary and compulsory ProComp participants. As previously explained, teachers incumbent to DPS prior to the implementation of ProComp were given a choice about whether to participate in the program: They could either join voluntarily or continue to be paid under the single-salary schedule. This presents a selection problem as we might expect mobility to differ between teachers who voluntarily opted into ProComp, those who were required to join, and those who decided not to join. For example, voluntary participants may hold more favorable opinions of DPS, and thus, may be less likely to leave their schools or the district, regardless of whether or not they receive additional compensation under ProComp.

To explore whether there is a differential association between incentives and mobility for the two types of ProComp participants, variables accounting for the two-way interactions between receiving an incentive and voluntary or compulsory participation were computed. Doing so resulted in four categories, each of which are compared with nonparticipants (who necessarily do not receive a ProComp incentive): (a) voluntary participants who receive an incentive, (b) voluntary participants who do not receive an incentive, (c) compulsory participants who receive an incentive, and (d) compulsory participants who do not receive an incentive.

Results from sensitivity tests of participation type are presented in Table 5. The first specification (Columns 1–3) includes the two-way interactions between receiving an incentive and voluntary and compulsory participation type. Coefficients that represent the unique relationship between mobility and receiving an incentive for each type of participant are Voluntary & Incentive and Compulsory & Incentive. Estimates suggest receiving a ProComp incentive is associated with a 17% decrease in the odds of departure for voluntary participants. However, receiving a ProComp incentive has no significant association with the odds of departure for compulsory teachers. It is interesting to note the large positive estimates for both voluntary and compulsory teachers who do not receive an incentive. Although only a small number of ProComp participants fail to receive any incentive in a given year, estimates suggest that these teachers have much greater odds of departing from their schools and from the district.

Table 5

Sensitivity Tests of Participation Type on Hazard Ratio Estimates From Models of Teachers’ Overall Departures

Parameter	Two-way interactions			Three-way interactions
	B	SE	Exp(B)	B	SE	Exp(B)
Time	−0.032	0.016	0.968*	−0.032	0.016	0.969
Voluntary and Incentive	−0.180	0.060	0.835**	−0.270	0.073	0.763***
Voluntary and No incentive	2.615	0.292	13.666***	2.828	0.335	16.912***
Compulsory and Incentive	−0.049	0.069	0.953	−0.057	0.082	0.945
Compulsory and No incentive	4.634	0.588	102.909***	4.901	0.717	134.388***
High-poverty	0.380	0.045	1.463***	0.325	0.061	1.384***
Voluntary and Incentive × High-poverty				0.236	0.108	1.266*
Voluntary and No incentive × High-poverty				−1.212	0.683	0.298
Compulsory and Incentive × High-poverty				0.035	0.102	1.035
Compulsory and No incentive × High-poverty				−1.280	1.251	0.278
Teacher characteristics
Salary/$1,000	−0.012	0.004	0.988**	−0.012	0.004	0.988**
Female	0.173	0.046	1.188***	0.174	0.046	1.190***
Minority	0.099	0.048	1.104*	0.098	0.048	1.103*
Age years	0.012	0.002	1.012***	0.012	0.002	1.012***
Master’s degree +	0.013	0.048	1.013	0.013	0.048	1.013
Probationary	0.276	0.078	1.318***	0.277	0.078	1.319***
Years of experience	−0.007	0.013	0.993	−0.006	0.013	0.994
Years of experience²	0.000	0.000	1.000	0.000	0.000	1.000
School characteristics
Secondary	−0.295	0.115	0.744*	−0.299	0.115	0.742*
Average reading SGPs in t–1	−0.011	0.003	0.989***	−0.011	0.003	0.989***
Average math SGPs in t–1	−0.009	0.002	0.991***	−0.009	0.002	0.992***
–2 log likelihood	14,778.093			14,769.151

Note. All models include school fixed effects. N = 12,952. SGP = student growth percentile.

***

A hazards ratio significantly different from 1 at the 0.1% level; **at the 1% level; and *at the 5% level.

The second specification (Columns 4–6) includes the two-way interactions between receiving an incentive and participation type and the three-way interactions between receiving an incentive, participation type, and school poverty. Here again, we observe a significant negative association between incentives and mobility for voluntary participants but not for compulsory participants.¹⁶ In addition, there appears to be a differential association between incentives and mobility for voluntary participants depending on school poverty. Estimates suggest receipt of an incentive is associated with a 24% decrease in a teacher’s odds of departure for voluntary participants who do not work in a high-poverty school. However, there does not appear to be an association between incentives and mobility for voluntary participants who work in high-poverty schools. Thus, the association between ProComp incentives and decreased mobility appears to be limited to voluntary participants who work in schools that are not high poverty. Given that these teachers voluntarily joined ProComp, it is not clear how much of this association is attributable to the financial incentives they received through ProComp or to any other unobserved factors (e.g., performance, motivation, satisfaction with the district, etc.).

Concluding Discussion

This study considers the association between ProComp financial incentives and teacher mobility within and from DPS. Results suggest receiving a ProComp incentive is associated with a significant decrease in the odds a teacher will leave DPS but is not associated with a teacher’s odds of making a within-district move. Results also suggest that the association between ProComp and exiting the district is much weaker for teachers who work in high-poverty schools. Importantly, the observed association between ProComp and mobility is driven by teachers who receive an incentive of over $5,000 and appears to be limited to teachers who voluntarily participate in ProComp.

Although we might expect voluntary participants to have higher retention than compulsory or nonparticipants, results suggest this pattern does not hold for volunteers who do not receive an incentive. Rather, voluntary teachers who do not earn an incentive tend to have much higher odds of departure than even nonparticipants, particularly those who work in lower-poverty schools. This suggests that the receipt of a moderate or large financial incentive—as opposed to simply the potential to receive an incentive—is essential and corroborates findings from previous studies (e.g., Springer, Lewis, Podgursky, Ehlert, Gronberg, et al., 2009; Springer, Lewis, Podgursky, Ehlert, Taylor, et al., 2009).

Given that ProComp incentives include performance and knowledge/skills-based incentives, this differential attrition may be desirable, suggesting that less effective and less committed teachers are leaving the district. Although findings do not establish that ProComp is causing changes in teacher mobility patterns, they provide evidence of an association between incentives and mobility, and raise questions about the potential of ProComp to alter the composition of the teacher workforce in DPS. Although an analysis of teacher effectiveness, ProComp participation, and teacher behavior is beyond the scope of this paper, these important questions are being explored by others (see e.g., Goldhaber, Bignell, Farley, Walch, & Cowan, 2013).

This research comes at a time when there is much enthusiasm for and action aimed at improving retention, and ultimately, quality through financial incentives. Yet, there also exists only scant evidence about the intended and unintended effects of such a policy approach. My results suggest ProComp is associated with a decrease in teacher mobility, particularly with respect to teachers’ exits from the district. This is encouraging as exits represent a net loss in the teacher workforce for DPS and create vacancies that must be filled with new, generally less-effective teachers. However, this association does not generally hold for teachers who work in high-poverty schools. As such, ProComp incentives do not appear to offset the more substantial issues of school leadership, climate, and working conditions. In moving forward, the policy community will be well served by research that focuses the relationship between effectiveness and incentives, particularly as it relates to improving retention and teacher quality in high-poverty schools.

Footnotes

Acknowledgements

I wish to thank Denver Public Schools (DPS) for permission to use the data. Thanks also to Drs. Ed Wiley, Kevin Welner, Derek Briggs, Dan Goldhaber, Rebecca Maynard, Meredith Richards, and Ben Domingue for their comments on earlier drafts. I am most grateful to Dr. Susana Loeb and all anonymous reviewers for their very thoughtful comments and suggestions. All errors in and shortcomings of this article are my responsibility.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was supported by a postdoctoral fellowship in education policy at the Graduate School of Education at the University of Pennsylvania (Penn).

Notes

Author

ELEANOR S. FULBECK is a research analyst in education at the American Institutes for Research (AIR). Her research focuses on exploring the effects of educational policies on educator human capital and equity.

References

Allison

P. D.

(1995). Survival analysis using SAS: A practical guide. Cary, NC: SAS Institute.

Boyd

Grossman

Ing

Lankford

Loeb

Wyckoff

(2011). The influence of school administrators on teacher retention decisions. American Educational Research Journal, 48, 303–333.

Clotfelter

C. T.

Glennie

E. J.

Ladd

H. F.

Vigdor

J. L.

(2008). Teacher bonuses and teacher retention in low-performing schools: Evidence from the North Carolina $1800 teacher bonus program. Public Finance Review, 36, 63–87.

Clotfelter

C. T.

Ladd

H. F.

Vigdor

J. L.

(2006). Teacher-student matching and the assessment of teacher effectiveness. Journal of Human Resources, XLI, 780–820.

Clotfelter

C. T.

Ladd

H. F.

Vigdor

J. L.

(2010). Teacher mobility, school segregation, and pay-based policies to level the playing field (Working Paper No. 44). Washington, DC: National Center for Analysis of Longitudinal Data in Education Research.

Darling-Hammond

(2001). The challenge of staffing our schools. Educational Leadership, 58, 12–17.

Denver Public Schools. (2011). Denver Public Schools: Facts. Retrieved from http://communications.dpsk12.org/newsroom/facts-and-figures/about-denver-public-schools/

Farley

A. N.

(2011). Exceeds expectations incentive: Teacher characteristics and student achievement, 2011 Report Brief. Retrieved from http://static.dpsk12.org/gems/newprocomp/FarleyEEFinalReportBriefFINAL.pdf

Fulbeck

E. S.

(2011). Teacher retention: Estimating and understanding thee of Financial incentives in Denver. retrieved from ERIC, ProQuest Digital Dissertations. (ED534347)

10.

Fulbeck

E. S.

(2012, January 12). Beyond Anecdotes: The evidence about financial incentives and teacher retention. Shanker Blog. Retrieved from http://shankerblog.org/?p=4729

11.

Fulbeck

E. S.

Richards

M. P.

(2013). The impact of financial incentives on teachers’ comings and goings. Manuscript submitted for publication.

12.

Glazerman

Seifullah

(2010). An evaluation of the Teacher Advancement Program (TAP) in Chicago: Year two impact report. Princeton, NJ: Mathematica Policy Research.

13.

Goldhaber

Bignell

Farley

Walch

Cowan

(2013). Who chooses pay for performance: Exploring the link between performance and preferences for compensation reform in the teacher labor market. Center for Education Data and Research working paper.

14.

Goldhaber

Gross

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, 57–87.

15.

Gonring

Teske

Jupp

(2007). Pay-for-performance teacher compensation: An inside view of Denver’s ProComp plan. Cambridge, MA: Harvard Education Press.

16.

Guarino

C. M.

Santibanez

Daley

G. A.

(2006). Teacher recruitment and retention: A review of the recent empirical literature. Review of Educational Research, 76, 173–208.

17.

Hanushek

E. A.

(2007). The single salary schedule and other issues of teacher pay. Peabody Journal of Education, 82, 574–586.

18.

Hanushek

E. A.

Kain

J. F.

Rivkin

S. G.

(2004). Why public schools lose teachers. Journal of Human Resources, XXXIX, 326–354.

19.

Henke

Zahn

Carroll

(2001). Attrition of new teachers among recent college graduates: Comparing occupational stability among 1992-1993 college graduates who taught and those who worked in other occupations. Washington, DC: National Center for Education Statistics.

20.

Hough

H. J.

Loeb

Plank

(2011). The quality teacher and education act: First year report. Palo Alto, CA: Center for Education Policy Analysis. Available from http://www.edpolicyinca.org/

21.

Imazeki

(2005). Teacher salaries and teacher attrition. Economics of Education Review, 24, 431–449.

22.

Ingersoll

R. M.

(2001). Teacher turnover and teacher shortages: An organizational analysis. American Educational Research Journal, 38, 499–534.

23.

Ingersoll

R. M.

(2004). Why do high-poverty schools have difficulty staffing their classrooms with qualified teachers? Washington, DC: Center for American Progress.

24.

Ingersoll

R. M.

Smith

T. M.

(2003). The wrong solution to the teacher shortage. Educational Leadership, 60, 30–33.

25.

Kirby

S. N.

Berends

Naftel

(1999). Supply and demand of minority teachers in Texas: Problems and prospects. Educational Evaluation and Policy Analysis, 21, 47–66.

26.

Lankford

Loeb

Wyckoff

(2002). Teacher sorting and the plight of urban schools. A descriptive analysis. Educational Evaluation and Policy Analysis, 24, 37–62.

27.

Milanowski

A. T.

Longwell-Grice

Saffold

Jones

Schomisch

Odden

(2009). Recruiting new teachers to urban school districts: What incentives will work? International Journal of Education Policy and Leadership, 4, 1–13.

28.

Milanowski

A. T.

Odden

A. R.

(2007). A new approach to the cost of teacher turnover (CRPE Working Paper No. 13). Seattle, WA: Center on Reinventing Public Education.

29.

National Center for Educational Statistics (2012). Common core of data, Denver Public Schools. Retrieved from http://nces.ed.gov/ccd/

30.

Panty

Hussar

Snyder

Provasnik

Kena

Dinkes

. . .Kemp

(2008). The Condition of Education 2008 (NCES 2008-031). Washington, DC: National Center for Education Statistics, Institute of Education Sciences, U.S. Department of Education.

31.

Rivkin

S. G.

Hanushek

E. A.

Kain

J. F.

(2005). Teachers, schools, and academic achievement. Econometrica, 79, 417–458.

32.

Slotnik

W. J.

Smith

M. D.

(2004). A catalyst for change: Pay for performance in Denver, Final Report. Boston, MA: Community Training and Assistance Center. Retrieved from http://static.dpsk12.org/gems/newprocomp/RptCatalystChangeFull2004.pdf

33.

Springer

M. G.

Lewis

J. L.

Podgursky

M. J.

Ehlert

M. W.

Gronberg

T. J.

Hamilton

L. S.

. . .Peng

(2009). Texas Educator Excellence Grant (TEEG) Program. Year three evaluation report. Nashville, TN: National Center on Performance Incentives.

34.

Springer

M. G.

Lewis

J. L.

Podgursky

M. J.

Ehlert

M. W.

Taylor

L. L.

Lopez

O. S.

Pen

A. X.

(2009). Governor’s educator excellence grant (GEEG) program: Year three evaluation report. Nashville, TN: National Center on Performance Incentives.

35.

Steele

J. L.

Murnane

R. J.

Willett

J. B.

(2009). Do financial incentives help low-performing schools attract and keep academically talented teachers? Evidence from California (Working Paper No. 14780). Cambridge, MA: National Bureau of Economic Research.

36.

Torres

(2013, May 7). Colorado 3rd-grade reading scores remain stagnant, test results show. The Denver Post. Retrieved from http://www.denverpost.com/ci_23190381/colorado-3rd-grade-reading-scores-remain-stagnant-test

37.

Wyckoff

Boyd

Lankford

Loeb

(2003). Understanding teacher labor markets: Implications for educational equity. In Monk

Plecki

(Eds.), School finance and teacher quality: Exploring the connection (pp. 55–84). Thousand Oaks, CA: Corwin Press.