TEACHER PENSION INCENTIVES AND TEACHER LABOR MARKET BEHAVIOR By Pin - En Annie Chou A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Economics Doctor of Philosophy 2018 ABSTRACT TEACHER PENSION INCENTIVES AND TEACHER LABOR MARKET BEHAVIOR By Pin - En Annie Chou My dissertation studies the effect of recent pension reforms on teacher labor market behavior . I I describe the across state and over time variation in the generosity of 41 defined benefit pension plans. To do so, I simulate the present value of the total pension compensation, net of contribution costs, for a representative teacher in each of the 41 defined benefit pension plans. I find large cross - state differences in the present value of n et pension wealth as well as net pension wealth cuts. When examining the effect of pension - , I find that states with more generous benefits pre - reform undertook larger cuts. However, a few st ates including Illinois, New Jersey, and New Hampshire, which were among the least generous pre - reform also experienced large benefit cuts, decreasing their net pension wealth by more than 50 percentage points. Last, I find that states that do not enroll t eachers in Social Security faced worse pension funding crises and harsher cuts compared to those with Social Security coverage. estimate the effect of pension genero sity on new teacher quality. Between 2007 and 2011, 19 states and four cities enacted salient pension reforms affecting all newly hired teachers. Using pooled cross - sectional teacher - level data from the 2003 - 2004, 2007 - 2008 and 2011 - 2012 Schools and Staffi ng Survey, I examine how new teacher quality changed differentially between the pre - benefit cuts and post - benefit cuts periods among states and cities that enacted pension reforms relative to those that did not have pension reforms. The hypothesis is that lower pension benefits, all else equal, decrease the attractiveness of teaching and discourage prospective teachers, particularly those with better alternatives, from entering the profession. Controlling for a complete set of time - varying factors that migh t affect teacher entry, I find that teachers entering the profession under pension reforms are from undergraduate institutions with 25.63 points lower 25 th percentile SAT scores. I also find that a one percent increase in the present value of net pension wealth results in schools hiring new teachers from undergraduate institutions with 32.15 points higher 25 th percentile SAT scores. sion Incentives on New I use recent pension reforms to identify the effect of pension incentives on the job change decisions of early career teachers. I test the hypotheses that lower expected compensation levels and higher vesting requi rements decrease the cost of changing jobs and increase teacher turnover, all else equal. Using pooled cross - sectional teacher level data from the 2003 - 2005, 2007 - 2009 and 2011 - 2013 school years, I examine how new teacher turnover changed differentially be tween the pre - wealth cut and post - wealth cut periods among states that enacted pension reforms, relative to those that did not. Controlling for teacher and school characteristics, I find no significant effect of benefit - reducing legislation or more stringe nt vesting requirements on teacher turnover among newly hired teachers . These estimates may suggest that new teachers only value short - term pension wealth accruals or that teachers lack full knowledge of pension structure. Alternatively, the composition of the new teacher pool may have changed during recent pension reforms, with less mobile teachers entering the profession. Copyright by PIN - EN ANNIE CHOU 2018 v ACKNOWLEDGEMENTS I thank my committee - Jeff Biddle , Paul Menchik , Peter Berg , and especially my chair Stacy Dickert - Conlin - for their counsel and support. I thank the Ministry of Education in Taiwan for supporting my research through a Government Scholarship to Study Abroad. I thank my parents for their inve stment of time and effort in my education. I also thank my boyfriend - Walter Melnik - for his love and supports . vi TABLE OF CONTENTS LIST OF TABLES ................................ ................................ ................................ ....................... vii i LIST OF FIGUR ES ................................ ................................ ................................ ....................... x i CHAPTER 1 HOW LARGE IS THE CUT? COMPARING NET PENSION WEALTH OF STATE DEFINED BENEFIT PENSION PLANS FOR TEACHERS ACROSS STATES AND OVER TIME ................................ ................................ ................... 1 1.1 Introduction ................................ ................................ ................................ ............................. 1 1.2 Related Literature ................................ ................................ ................................ .................... 3 1.3 I nstitutional D etails ................................ ................................ ................................ ................. 7 1.4 Pension Data ................................ ................................ ................................ ........................... 9 1.5 Simulations of Pension Wealth, Lifetime Contributions, and Net Pens ion Wealth ............. 12 1.5.1 Comparison of Starting Salary, Years of Service, and Final Average Salary ............... 15 1.5.2 Simulation of First - year Pension Benefits Across States ................................ .............. 17 1.5.3 Comparison of Simulated Pension Wealth and Lifetime Contributions ........................ 19 1.5.4 Comparison of Net Pension Wealth Across States and Over Time ............................... 20 1.6 Sensitivity to Assumptions ................................ ................................ ................................ ...... 23 1.7 Socia l Security and Teacher Pensions ................................ ................................ ..................... 24 1.8 Conclusion ................................ ................................ ................................ ............................... 27 CHAPTER 2 THE EFFECT OF RECENT TEACHER PENSION REFORMS ON NEW TEACHER QUALITY ................................ ................................ ......................... 2 9 2 .1 Introduction ................................ ................................ ................................ ........................... 2 9 2 .2 Related Literature ................................ ................................ ................................ .................. 32 2.2.1 Teache r Pensions and Teacher Turnover ................................ ................................ ....... 3 2 2.2.2 Teach er Pensions and Teacher Quality ................................ ................................ .......... 3 5 2.2.3 Other Factors that Attract Teachers ................................ ................................ ............... 3 7 2 .3 Institu tional Details and Pension Data ................................ ................................ .................. 38 2 .4 Teacher Data ................................ ................................ ................................ ......................... 4 4 2 .5 Methodology and Estimates ................................ ................................ ................................ .. 51 2.5 .1 Pension Reforms and Teacher Quality ................................ ................................ .......... 51 2 .5.2 Pension Wealth and Teacher Quality ................................ ................................ ............. 5 5 2 .6 Conclusion ................................ ................................ ................................ ............................ 56 CHAPTER 3 WHO CHOOSES TO STAY? THE EFFECT OF TEACHER PENSION INCENTIVES ON NEW TEACHER TURNOVER ................................ ............ 5 8 3.1 Introduction ................................ ................................ ................................ ........................... 58 3.2 Related Literature ................................ ................................ ................................ .................. 60 3.2.1 Teacher Pensions and Teacher Turnover ................................ ................................ ....... 60 3.2.2 Teacher Outside Options and Teacher Turnover ................................ ........................... 63 3.3 Institu tional Details and Pension Data ................................ ................................ .................. 64 3.4 Teacher Data ................................ ................................ ................................ ......................... 67 vii 3 .5 Methodology and Estimates ................................ ................................ ................................ .. 71 3.5.1 Pens ion Reforms and Teacher Turnover ................................ ................................ ........ 71 3.5.2 Pe nsion Wealth and Teacher Turnover ................................ ................................ .......... 75 3.6 Conclusion ................................ ................................ ................................ ............................ 76 APPENDICES ................................ ................................ ................................ .............................. 7 9 APPENDIX A PENSION WEALTH CALCULATION ................................ ......................... 80 APPENDIX B TABLES ................................ ................................ ................................ .......... 82 APPENDIX C FIGURES ................................ ................................ ................................ ....... 184 BIBLIOGRAPHY ................................ ................................ ................................ ....................... 1 9 5 viii LIST OF TABLES Table 1.1: Summary of Selected Pension Literature on Pension Wealth ................................ ...... 82 Table 1.2: Types of Pension Plans States offered Teachers ................................ .......................... 86 Table 1.3: State Teacher Pension Homepage, Handbooks, and Other Resources ......................... 88 Table 1.4: Comparison of the Number of Years Used to Calculate Final Average Salar y ........... 94 Table 1.5: Comparison of the Formula Multiplier ................................ ................................ ......... 96 Table 1.6: The Limitation on First Year Annual Be nefits as a Percentage of FAS ....................... 98 Table 1.7: Comparison of the Cost of Living Adjustments ................................ ........................... 99 Table 1.8: Comparison of Retirement Eligibility ................................ ................................ ........ 101 Table 1.9: Comparison of the Employee Contribution Rates ................................ ...................... 103 Table 1.10: Summary of Reforms Between 2007 and 2016 that Reduced Teacher Benefits ...... 105 Table 1.1 1 : Comparison of Starting Salary, Years of Service, and Final Average Salary .......... 107 Table 1.12: Simulation of First - year Pension Benefits Across States ................................ ......... 109 Table 1.13: Comparison of Simulated Pension Wealth ................................ ............................... 111 Table 1.14: Comparison of Simulated Lifetime Contributions ................................ .................. 113 Table 1.15: Comparison of Net Pension Wealth ................................ ................................ ......... 115 Table 1.16: Comparison of Net Pension Wealth Across States and Overtime ............................ 117 Table 1.17: Summary of Pension Reforms and S ala ry Changes Between 2007 and 2016 ......... 119 Table 1.18: Comparison of the Rankings of Net Pension Wealth, Measured by the Percentage of Final Average Salary ................................ ................................ ............................ 120 Table 1.19: Simulation of Net Pension Wealth for Teachers Hired in 2007 by Years to Live and State ................................ ................................ ................................ ................... 122 Table 1.20: Simulation of Net Pension Wealth for Teachers Hired in 2007 by Discount Rate and State ................................ ................................ ................................ ................... 124 ix Table 1 .21: Simulation of Net Pension Wealth for Teachers Hired in 2007 by COLAs and State ................................ ................................ ................................ ........................... 126 Table 2.1: Types of Pension Plans States and Cities Offered Teachers ................................ ...... 128 Table 2.2: State Teacher Pension Homepage, Handbooks, and Other Resources ...................... 130 Table 2.3: Comparison of the Number of Years Used to Calculate Final Average Salar y ......... 136 Table 2.4: Comparison of the Formula Multiplier ................................ ................................ ....... 138 Table 2.5: The Limitation on First Year Annual Benefits as a Percentage of FAS .................... 140 Table 2.6: Comparison of the Cost of Living Adjustments ................................ ......................... 1 42 Table 2.7: Comparison of Retirement Eligibility ................................ ................................ ....... 145 Table 2.8: Comparison of the Employee Contribution Rates ................................ ..................... 147 Table 2.9: Comparison of Vesting Requirements ................................ ................................ ....... 149 Table 2.10: Refund if Teac hers Leave Before Vested ................................ ................................ . 151 Table 2.11: Membership of Public Pension Plans that Enroll Teachers ................................ ...... 153 Table 2.12: Weighted Means of Variables ................................ ................................ ................. 159 Table 2.13: Weighted Means of Variables ................................ ................................ ................. 160 ................................ ................................ .. Table 2.15: Estimates of the Effects of Pension Reform on New Teacher Quality ..................... 16 2 Table 2.16: The Difference in New Teacher Quality between the Control and Treatment Groups before Recent Pension Reforms ................................ ................................ .. 163 Table 2.17: The Effect of Net Pension Wealth on Teacher Quality ................................ ............ 164 Table 3.1: Types of Pension Plans States and Cities Offered Teachers ................................ ...... 165 Table 3.2: State Teacher Pension Homepage, Handbooks, and Other Resources ...................... 167 Table 3.3: Pension Reforms Between 2007 and 2011 ................................ ................................ . 173 Table 3.4: Membership of Public Pension Plans that Enroll Teachers ................................ ........ 175 x Table 3.5: Weighted Means of Variables ................................ ................................ .................... 180 Table 3.6: The Effects of Pension Reform s on New Teacher Turnover ................................ ...... 181 Table 3.7: The Difference in New Teacher Turnover between the Control and Treatment Groups before the Pension Reforms ................................ ......................... 182 Table 3.8: The Effects of Net Pension Wealth on New Teacher Turnover ................................ . 183 xi LIST OF FIGURES Figure 1.1: Net Pension Wealth Change between 2007 and 2016 ................................ ............... 184 ................................ .......... Figure 1.3: Relationships between Net Pension Wealth Cut and Pension Funding Ratio ........... 186 ................................ ................................ ................................ ......... ................................ ................................ ................................ ......... ................... 1 CHAPTER 1 HOW LARGE IS THE CUT? COMPARING NET PENSION WEALTH OF STATE DEFINED BENEFIT PENSION PLANS FOR TEACHERS ACROSS STATES AND OVER TIME 1.1 Introduction Teacher compensation comes with a total package that includes salary, extra pay, in - kind benefits, and pensions ( U.S. Department of Education 2018) . Pension benefit s are different from other forms of compensation because there is a delay between the time teachers earn and receive their pensions. Moreover, most states require teachers to contribute a portion of their salary to fund their teacher pension plans, so emplo yee contributions need to be netted out from pension wealth because that is not part of the labor compensation ( Costrell and Podgursky 2009) . Today, most states still offer their public school teachers defined benefit pension plans. 1 Unlike defined contrib ution plans that tie retirement benefits to total contributions and are subject to market fluctuation, defined benefit plans guarantee retirees a specific level of annual benefits calculated as the product of a multiplier factor, within - system experience, and final average salary. While teachers can easily learn about their salary, extra pay, and in - kind benefits when they are hired, it can be difficult for teachers to understand how pension parameters could affect their deferred (and discounted) compensa tion. 2 For example, compared to Alabama, Massachusetts requires teachers to contribute a larger share of their salary and to start collecting their full retirement benefit at an older age. However, Massachusetts also provides a more generous 1 2 2 multiplier fac 3 Without calculating net pension wealth, it is hard to tell which state is more generous. Later in this paper, I show that Massachusetts is more generous for teachers who work in the same school from age 25 until they reach their normal retirement age. 4 The 2008 financial crisis left many pension systems underfunded and struggling. 5 Wilshire Consulting (2017) reports that the median funded level for state retirement systems was 87 percent in 2007 and fell to 67 percent by 2016. 6 7 To reduce pension debt, many states enacted pension reforms that provide less generous pension benefits. Pension plans reformed in numerous ways including decreasing multipliers, lengthening the years used to compute final average salary, raising retirement eligibility age or service, capping the annual benefit, increasing vesting rules, increasing teacher contribution rates, and lowering cost of living adjustments. Beyond this, many states required school districts to contribute increasingly higher shares of teacher salary into pension systems, which may discourage school districts from raising salaries to attract and retain teachers (Doherty et al. 2012). In fact, the average starting salaries in many states have fallen since the 2007 when adjusted for inflation. While teacher compensation became less attractive following the financial crisis, the Great Rec ession during the late 2000s and early 2010s reduced outside 3 4 5 6 7 3 options and may make teaching more attractive. 8 To characterize how recent pension reforms may affect new teachers, I simulate the present value of net pension wealth for a representative teache r in 41 defined benefit pension plans (40 states and the District of Columbia). These 41 defined benefit pension plans offered teachers a traditional defined benefit plan throughout the years 2007 to 2016 . I also present my simulations as the percentage of final average salary. I then rank each plan for its generosity before (2007) and after (2016) the pension reforms. In the following section, I review the related literature. Section 1.3 describes institutional details and Section 1.4 describe pension wealth, lifetime contributions, and net pension wealth. Section 1.6 tests the sensitivity of my findings, and Section 1.7 discuss es teacher pension interacting with Social Security. Sec tion 1.8 concludes. 1.2 Related Literature A recent set of papers explores the variation in pension wealth caused by the variation in pension parameters across plans or over time. Earlier pension literature introduced various measures of pension incentives (Coile and Gruber 2007; Stock and Wise 1990). Stock and Wise (1990) emphasize the importance of continuing to work for one more year and develop the option value measure, which captures the difference between current pension wealth and pension wealth afte r one additional year of work (one - year accrual). Coile and Gruber (2007) develop the peak 8 4 value measure, which captures the difference between current pension wealth and the maximum possible pension wealth. Table 1.1 summarizes pension literature on pensi on wealth. Two papers that focus on single - Costrell and Podgursky 2009; Toutkoushian et al. 2011 ) . Costrell and Podgursky six states using the option value measure. They slowly in the early years of her career, then accelerates in her mid - to late - fifties, and then drops off sharply over the next few years. This backloading of pension wealth accumulation creates a Toutkoushian et al. (2011) analyze the parameter differences among 49 state - run defined benefit pension plans for public school teachers. 9 For states that adopted hybrid plans, which combine defined benefit and defined contribution, the authors analyze the defined benefit part of the plans. They collect pension parameters in effect in 2008 from the Public Fund Survey and Schmidt (2010). The authors simulate the peak value net pension wealth a representative teacher would expect to rec eive when first hired in each of the 49 states. They assume that teachers spent their entire career in the teaching profession from the age of 22 to 65 and had a starting salary of $30,000 that grew 3 percent per year. They also present net pension wealth as a percentage of the representative of their lifetime salary. Overall, their simulations suggest that the number of years used to compute the final average salary does not affect net pension wealth by much. However, the caps imposed - year benefit in some states can greatly reduce the first - year pension benefit a 9 5 sion generosity and suggested that teachers should consider more than the multiplier factors when comparing pension generosity across states. On the other hand, Koedel, Ni and Podgursky (2014 ) focus on single - state over - time variation. They examine the ef - enhancement legislation enacted between 1995 and The authors find that the enhancements resulted in large gains in net pension wealth for teachers who were close to retirement. How ever, they find lower net pension wealth for younger teachers because of the associated contribution - rate increase. Much of the teacher pension literature focuses on teacher retirement and separation. Some researchers use the variation in pension paramete rs across plans to estimate the effect of pension incentives on teacher labor market. For example, Papke and Litwok (2013) estimated large cross - state differences in pension wealth upon vesting using the peak value measure in California, Florida, Michigan, and Wisconsin and found that pension characteristics such as vesting rules, availability of the Defined Contribution (DC) option, and Social Security coverage affect young t to use nationally representative data on teachers. They use the peak value measure to wealth in the 17 largest states. The authors use the variation in pension accrual profiles across states to identify the effect of pension in pension wealth reaches its maximum when teachers reach their normal retirement age. 10 Their results suggest that teachers delay retirement while pension wealth is still accumulating and then retire a bruptly when it reaches its maximum. 10 6 Some researchers study the effect of pension incentives on teacher retirement in specific states using administrative data that capture teacher exit behavior and earnings history (Furgeson, Strauss, and Vo gt 2006; Costr ell and McGee 2010 ; Koedel, Ni and Po dgursky 2014 ; Koedel and Xiang 2017 and their maximum possible pension wealth. They find that current pension wealth increases the probability of teacher retirement, while the expected future maximum pension wealth reduces the probability of retiring today. They also find a large increase in teacher retirement following the enactment in Pennsylvania of a policy that temporarily in creased early retirement benefits between the 1997 - 1998 and 1998 - 1999 school years. Costrell and McGee (2010 wealth using the peak value and option value measures for Arkansas teachers. They find that both peak value and one - yea Koedel and Xiang wealth when St. Louis enacted a policy change that increased teacher pension wealth for those who retired after the 1998 - 1999 school year. The authors show that newly hired teachers, who largely discounted future pension wealth, had a smaller increase in their pension wealth under the new policy, compared to those who were eligible for retirement. The authors use the heterogeneous fy the effect of pension enhancement on teacher retention. The authors find a temporary delay in retirement among teachers who were eligible for retirement in the 1997 - 1998 school year. However, they find no significant retention effect among teachers who were not eligible for retirement. They suggest that teachers may lack of full knowledge of their pensions. My paper is most closely related to Toutkoushian et al. (2011), who also provide simulations of net peak value pension wealth for a representative teacher across states and 7 pension in terms of its generosity . I expand their work by highlighting the cross - states differences in the magnitude of recent benefit cuts . To do so, I present the pension parameter changes between 2007 and 201 6 and provide simulations of the hypothetical pension wealth under old rules (before policy change) and new rules (after policy change) for a representative teacher in 41 defined benefit plans. Previous pension researchers either study variation among states in teacher pension wealth at a certain point of time ( Costrell and Podgursky 2009 ; Friedberg and Turner 2011 ; Papke and Litwok 2013; Toutkoushian et al. 2011 ) or study the effect of policy changes on pension wealth in specific states ( Costrell and McGee 2010 ; Furgeson, Strauss, and Vogt 200 6; Koedel, Ni and Podgursky 2014 ; Koedel and Xiang 2017) . While the studies that focus on single - state policy changes have the advantage of control ling for state - specific omitted variables, their results only apply to a single state. The results of my study demonstrate that the effects of recent pension - reducing legislation on pension wealth vary significantly across states. This variation of pension incentives could be used to explore the labor market effect. 1.3 Institutional Detail s In the U.S., all full - time public school teachers automatically participate in public pension plans. Although each state runs a pension system that covers teachers (state - run pension plans), some large cities like New York and Chicago operate their own mun icipality - specific pension plans . 11 I exclude these municipality - specific pension plans from my analysis for simplicity. T able 1.2 shows the types of pension plans that states and cities offer teachers during the years I study. 11 8 Historically, all states exc ept Indiana 12 have offered traditional defined benefit pension plans. In more recent years, some states adopted plan structures that shift investment performance risk to teachers (for example, defined contribution plan, hybrid plan, and cash balance plan), and some states allowed teachers to choose one plan or another, but most states still offer teachers traditional defined benefit plans that pay a specific amount upon retirement. Today, 37 states and the District of Columbia offer their public school teach ers only a mandatory defined benefit pension plan, and four states offer teachers a choice of defined benefits or other optional plans. In defined benefit plans, teachers contribute a portion of their salary toward the plan while employed. Teachers who re tire after becoming vested collect annual payments until their death. 13 The lifetime pension wealth of a teacher depends on the size of the initial annual payment, adjustments made for cost of living, and the length of time in retirement. All defined benefit plans use a similar formula to calculate annual payment: (1) First - year Annual payment (A) = FAS * Multiplier * Years of Service where FAS Because earnings tend to be highest in the final years of employment, FAS increases if pension plans consider fewer years of service when calculating the average. FAS also increases if states offer higher salary. The multiplier in the that she receives for each employment. In general, more years of service lead to higher annual payments. But some states cap the first - year annual paymen - year annual payment is: 12 13 9 (2) To counteract the impact of inflation on retirement income, public pension plans provide retirees with post - retirement cost - of - living adjustments (COLAs). Plans with higher cost - of - living a djustments are more beneficial to teachers. These pension parameters relate to the size of annual benefits. Other pension parameters such as retirement age and teacher contribution amounts can also affect the pension wealth, without directly changing the a nnual benefits. 14 Because teachers collect annual benefits from retirement until death, an earlier retirement age implies that teachers receive more from annual payments and enjoy higher lifetime pension wealth. Recall that most states require teachers to contribute a port tied to teacher contributions (see equation 1), an increasing contribution rate decreases net pension wealth. 1.4 Pension Data Table 1.3 presents a list of summaries of legislation. I obtain additional pension plan - level data from the National Education As sociation ( 2004, 2008, 2010, 2016) to confirm the years that pension parameters were in effect. I restrict my analysis to the 40 states and the District of Columbia that offered teachers a traditional defined benefit plan in 2007 and 2016. I exclude state s that offered teachers other types 14 10 difficult to predict. I choose 2007 as the initial year because 2007 pension parameters capture each - reducing legislation. The most recent available plan - level parameters were documented in 2016 by the National Education Association. In response to the 2008 financial crisis that lowered investment returns and exacerbated the pension funding crisis, many states enacted reforms after 2007 that reduced teacher benefits or increased required teacher contribution rates. In this section, I present how pension parameters changed between 2007 and 2016. Table 1.4 provides an overview of the number of years used to calculate FAS. In 2007, 29 pension plans used average salary for three or fewer of the highest years of salary. 12 pension plans used a similar definition with more than three years. Bet ween 2007 and 2016, 17 pension plans increased the number of years used to calculate FAS. By 2016, only 15 pension plans were using three or fewer years of salary to calculate FAS, and 26 pension plans were using more than the three highest years of salary to calculate FAS by 2016. Table 1.5 provides an overview of the multiplier used in each pension plan. In 2007, 31 pension plans used a constant multiplier that ranged from 1.6 percent (Wisconsin) to 2.67 percent (Nevada). 10 pension plans used multiplie rs that varied by retirement age or years of service. Between 2007 and 2016, 14 pension plans lowered their multipliers, and Vermont changed from a constant multiplier to multipliers that increase with years of service. Table 1.6 shows that in 2007, 19 pen sion plans limited the annual payment to a specific percent. By 2016, Pennsylvania had changed from no limitation on annual payment to 100 percent of the teac 11 Vermont became more generous by changing the cap from 53.34 percent to 60 percent. T he pension plans also vary by their cost - of - living adjustments (COLAs). Table 1.7 provides an overview of COLAs in each pension plan. In 2007, 28 states automatically adjusted the cost of living based on a specific percentage or the consumer price index (C PI). On the other hand, 11 pension plans relied on state legislation to decide COLAs for each year, and two states tied their COLAs to their funding level. By 2016, 14 pension plans changed COLAs in ways that were less generous compared to 2007. The change s include lowering the COLA rate, delaying the receipt of COLAs, or switching from automatic adjustments to adjustments that tied COLAs to funding levels or other indexes. Table 1.8 provides an overview of retirement eligibility in each state. Compared to 2007, by 2016, 28 pension plans had raised the retirement eligibility age or years of service, which shortens the length of years for receiving annual benefits for many teachers. Table 1.9 provides an overview of the employee contribution rate required b y each state. In 2007, Florida and Wyoming did not require their teachers to make any contribution to their pension plans. Other pension plans required their teachers to make contributions ranging from three percent (Delaware) to 12 percent (Missouri) of t heir salary. Thirty - five pension plans required teachers to contribute at least five percent of their salary. By 2016, every pension plan required teachers to make contributions that ranged from 2.3 percent (Wyoming) to 14.5 percent (Missouri and Nevada) o f their salary. Twenty - eight pension plans increased their employee contribution requirements by 2016. Table 1.10 summarizes the parameter changes in each of the 40 states and the District of Columbia. We can see that among the 37 states that enacted pens ion reforms between 2007 and 12 2016, the most frequent parametric pension reforms increased the eligibility age and employee contributions. Twenty - eight states increased retirement age, 28 states increased employee contribution rates, 17 states increased the number of years used as a base for computing final average compensation, 14 states lowered COLAs, 14 states decreased the multipliers, and two the parametric pension reforms enacted in each state. Thirty states changed multiple parameters and seven states focused on a single parameter (e.g., increasing retirement age or increasing employee contribution rates). Among states that changed multiple parameters, Ala bama, Florida, Hawaii, and New Jersey had the most changes. They all decreased the multipliers, increased the number of years used as a base for computing final average compensation, increased retirement age, and increased employee contribution rates. Note states made parameter changes that were more generous for teachers. Arizona, for example, increased the retirement age, increased employee contribution rates, increase d the number of years - year annual payment. Additionally, Vermont increased the retirement age and employee contribution rate, but also increased multipliers for people with more than 20 years of service and imposed fewer - year annual payments. In the next section, I discuss how those 1.5 Simulations of Pension Wealth, Lifetime Contributions , and Net Pension Wealth To more precisely evaluate pension plan generosity, I simulate and compare pension wealth, lifetime contributions, and net pension wealth that a hypothetical teacher would expect to receive 13 from each of the 41 defined benefit pensi on plans if she started teaching in 2007 and in 2016. 15 Pension wealth is the actuarial present value of the stream of annual payments teachers expect to receive upon retirement. 16 It not only measures the size of annual payments, but also how long teachers receive these payments. S uppose a teacher started teaching in year t=1 , retires in year t=T , and collects pension benefits for d years, from t=T+1 to t=T+ . Then her expected present discounted net pension wealth (PDNW) would be calculated using the following formula: (3) = - where is the capped annual payment in year t >T, is the cost - of - living adjustment in year t, is the probability of surviving at year t, r is the discount rate, and is the teacher contribution in year t T . In calculating net pension wealth, I address three possible uncertainties outlined by Friedberg (2011). 17 First, the length of retirement varies among teachers because we do not know the date of death. Second, we do not know the true discount rate a teacher faces because future inter est rates time preferences are unknown . Third, the future cost - of - living adjustments may change for current teachers. Unlike other pension parameter changes presented in Table 1.10, COLAs are not viewed by courts as core benefits protecte d under state laws, so COLAs can fluctuate over time ( Munne ll et al. 2016 ) . 15 16 17 14 I base the simulations in this paper on the following set of assumptions: First, lik e Koedel, Ni and Podgursky (2014 ), I assume the hypothetical teacher starts teaching at age 25. Second, I assume that the hypothetical teacher works in the same school until she reached her maximum pension wealth and then starts collecting retirement benefits right away. Much of the previous pension literature simulates pension wealth using the peak value measure. pension wealth reaches its maximum when they reach their normal retirement age ( Friedberg and Turner 2011) , I therefore identify the closest normal retirement option for the hypothetical teacher in each state year. For example, a hypothetical Alabama teacher who started teaching in 2007 could retire with full retirement benefits at age 60 with at least 10 years of service or at any age with at least 25 years of service. The closest normal retirement option for her w ould be to retire at age 50. retirement using her age. For example, the hypothetical Alabama teacher who started teaching in 2007 would have 25 (= 50 - 25) years of se rvice. Third, average starting teacher salary reported by the National Education Association. 18 I then calculate the annual salary growth rate in each state using the bac helor's degree salary schedule reported by the National Center for Education Statistics Schools and staffing Survey (SASS) in 2007. 19 I normalize all salary amount in this paper to 2016 dollars using the Consumer Price Index (CPI) for all items less food an d energy reported by the Bureau of Labor Statistics . 20 Fourth, I assume 18 19 20 15 that the cost - of - living adjustment for teachers is 3 percent per year throughout her retirement, which is also used in Costrell and Podgursky (2009) and Toutkoushian et al. (2011) , assuming the individual would live until age 80 (a female American can expect to live on average 80 years). 21 Last, like Koedel, Ni and Podgursky (201 4 ), I assume a 4 percent discount rate that allows a positive real interest rate and a tim e preference in earning . In the next section, I examine the sensitivity of my findings to changes in the stated assumptions including expected longevity, assumed discount rate, and COLAs. With the stated assumptions, I calculate FAS using the following for mula: (6) where n is the number of years used to calculate FAS, is the starting salary in year t , is the salary annual growth rate, and YOS is the years of service. 1.5.1 Comparison of Starting Salary, Years of Service, and Final Average Salary The first and fourth columns of Table 1.11 present the starting salaries for the hypothetical teacher hired in 2007 and in 2016, all in 2016 dollars. The starting salaries ranged from $29,027 (Mon tana) to $52,034 (New Jersey) for those hired in 2007 and from $30,036 (Montana) to $51,359 (Delaware) for those hired in 2016. Compared to 2007, 27 states had lower starting teacher salaries in 2016 and 14 states had higher real starting salaries in 2016. Overall, both the mean and median starting salaries in 40 states and the District of Columbia decreased by more than $1000 between 2007 and 2016. 21 16 The second and fifth columns of Table 1.11 contain the years of service it takes for a hypothetical teacher to reach her closest normal retirement option in each state. The years of service ranged from 20 years (Massachusetts) to 41 years (Minnesota) in 20 07 and from 27 years (Kentucky) to 42 years (Massachusetts) in 2016. On average, teachers hired in 2016 would have to wait four more years, relative to those hired in 2007, to start collecting their full pension benefits. The results show that the variatio n in years of service caused by the different retirement options available across states and over time were economically large. For example, Massachusetts allows teachers hired in 2007 to retire with full pension benefits at age 65 with at least 10 years o f service or at any age with at least 20 years of service, but teachers hired in 2016 can only retire with full benefits at the age of 67 with at least 10 years of service. 22 These changes in retirement options increased the hypothetical Massachusetts teach years. The third and sixth columns of Table 1.11 contain the final average salaries a hypothetical teacher hired in 2007 and in 2016 would expect to earn, which I calculate using equation 6. On average, teachers hired in 2016 would expect to earn $4,186 higher final average salary than those average salary across states and over time depends on the starting salar y, annual salary growth rate, the number of years used for calculation, and the years of service. For example, Alabama decreased relative to those hired in 200 7. These two changes decreased the final average salary. However, under my assumptions that teachers teach until they reach their normal retirement age, a 22 17 hypothetical Alabama teacher hired in 2016 would expect to work 12 more years compared to those hired in 2007. This change increased the final average salary. Overall, a hypothetical Alabama teacher hired in 2016 would expect to earn about $7,000 more in final average salary than one hired in 2007. If the hypothetical Alabama teachers chose not to work lo nger when the state raised the normal retirement eligibility, her expected final average salary would decrease. 1.5.2 Simulation of First - year Pension Benefits Across States Recall that all defined benefit plans use the same formulas, equations 1 and 2 , to calculate a - year capped annual payments. Table 1.12 presents the simulated first - year capped annual payments that a hypothetical teacher hired in 2007 and in 2016 would expect to receive across the pension plans in 40 states and the Di strict of Columbia. First - year annual benefits (first and fourth columns) are the product of the multiplier, years of service, and final average salary. The second and fifth columns are the capped first - year annual benefits after accounting for the caps th at were imposed on first - year benefits. The third and sixth columns present the capped first - year annual benefits as the share of final average salary. The results show that the variation in the capped first - year annual benefits across states and over time were economically large. The capped first - year annual benefits ranged from $23,876 (Montana) to $96,096 (Connecticut), with a median of $45,140 for those hired in 2007. On the other hand, for those hired in 2016, the capped first - year annual benefits rang ed from $25,816 (South Dakota) to $154,589 (Massachusetts), with a median of $45,694. On average, teachers hired in 2007 would expect to receive $48,834 (62 percent of FAS) in the first year of retirement, while those hired in 2016 would expect to receiv e 18 $ 54,086 (66 percent of FAS). Although more years of expected service increases annual benefits (see equation 1), the shorter length of retirement decreases lifetime pension wealth. When comparing the values in the first and second columns of Table 1.12, I find that the cap is only binding for teachers hired in 2007 in Illinois, who would receive $1,953 less first - year annual benefits with a cap (compared to not having a cap), and teachers hired in 2007 in Nevada, who would receive $4,100 less with a cap. Other states either do not have a cap or their teachers would not earn more than the capped amount. Recall from Table 1.6 that between 2007 and 2016 , Pennsylvania and Alabama annual payment, while Arizona and Vermont relaxed their caps . Because the hypothetical teacher in these four states never earned more than the capped amount, recent changes in annual pension payment caps would have no influence on s pension wealth. Although the caps in other states did not change between 2007 and 2016, the pension reforms in other pension parameters could affect how much caps - year annual benefits. I find that caps imposed on annual benefits could largely affect teachers in states that greatly increased retirement eligibility age without decreasing multipliers. For example, Illinois and Massachusetts raised the retirement eligibility age for rst - year annual benefits, because more years of service directly contribute to higher annual benefits using equation 1 and indirectly increase annual benefits by increasing final average salary. The hypothetical Illinois teacher hired in 2016 would receive $16,954 less in first - year annual benefits with a cap, and the hypothetical Massachusetts teacher hired in 2016 would receive $48,309 less with a cap, relative to no cap. 19 1.5.3 Comparison of Simulated Pension Wealth and Lifetime Contributions The firs t and third columns of Table 1.13 contain the sum of lifetime pension payments that a hypothetical teacher hired in 2007 and in 2016 would expect to receive in each state. Payments ranged from $178,696 (Oklahoma) to $519,830 (Massachusetts), with the media n equal to $296,308 for those hired in 2007. For those hired in 2016, the payments ranged from $152,560 (Oklahoma) to $432,229 (Connecticut), with the median equal to $ 242,970. On average, teachers hired in 2016 would expect to receive $43,685 less in pens ion wealth compared to those hired in 2016. The second and fourth columns of Table 1.13 present pension wealth as a percentage of final average salary. Payments ranged from 197 percent of FAS (Minnesota) to 655 percent of FAS (Massachusetts) for those hir ed in 2007 and from 170 percent of FAS (New Hampshire) to 555 percent of FAS (Kentucky) for those hired in 2016. On average, teachers hired in 2016 would expect to receive 331 percent of their FAS in pension wealth, and those hired in 2007 would expect to receive 397 percent of their FAS in pension wealth. Recall from Table 1.9 that it costs teachers a proportion of their salary to participate in their pension plans, and the cost varies across states and over time. The first and third columns of Table 1.14 contain the simulated value of the sum of contributions a hypothetical teacher hired in 2007 and in 2016 would expect to pay in each state. The sum of contributions ranged from zero dollars (Florida, New York, and Wyoming) to $114,022 (Illinois), with a me dian of $58,895 for those hired in 2007. On the other hand, teachers hired in 2016 would expect to contribute from a low of $22,613 (Wyoming) to a high of $199,459 (Massachusetts), with a median of $76,499 . On average, a teacher hired in 2016 would expect to contribute $18,198 more than one hired in 2007. 20 The second and fourth columns of Table 1.14 present lifetime contributions as a percentage of the final average salary. We can see that lifetime contributions ranged from 0 percent of FAS (Florida, New Yo rk, and Wyoming) to 169 percent of FAS (Missouri), with a median of 86 percent of FAS for those hired in 2007. For teachers hired in 2016, the sum of contributions ranged from 33 percent of FAS (Wyoming) to 204 percent of FAS (Missouri), with a median of 9 6 percent of FAS . On average, teachers hired in 2016 would expect to pay 19 percentage points more of their FAS compared to those hired in 2007. 1.5.4 Comparison of Net Pension Wealth Across States and Over Time In Table 1.15, the first and fourth colum ns contain the simulated values of net pension wealth (pension wealth minus lifetime contributions) that a hypothetical teacher would receive in each state. The simulated value of net pension wealth ranged from $124,191 (Oklahoma) to $432,364 (Massachusett s) for those hired in 2007 and from $50,968 (Illinois) to $345,233 (Connecticut) for those hired in 2016. I find large differences among states in the present value of net pension wealth. Prior to the pension reforms, the net pension wealth that a hypothetical Massachusetts teacher would expect to receive upon retirement was more than triple the amount a hypot hetical Oklahoma teacher would expect to receive, $432,364 compared to $ 124,191 . Recent pension parameters and salary changes enlarge the differences among states. For those hired in 2016, the net pension wealth that a hypothetical Connecticut teacher woul d expect to receive upon retirement is six times more than the amount a hypothetical Illinois teacher would expect to receive, $ 345,233 compared to $50,968. The second and fifth columns of Table 1.15 present net pension wealth as a percentage of final average salary. Net pension wealth ranged from 132 percent of FAS (Minnesota) to 545 21 percent of FAS (Massachusetts), with a median of 303 percent of FAS for thos e hired in 2007. For teachers hired in 2016, net pension wealth ranged from 52 percent of FAS (Illinois) to 389 percent of FAS (Kentucky), with a median of 232 percent of FAS. To illustrate the size of pension income relative to salary income, the third a nd sixth columns salary. The results suggest that the size of net pension wealth relative to lifetime salary income ranged from 11 percent (Minnesota) to 54 percent (Massachusetts), with a median of 25 percent for those hired in 2007. For teachers hired in 2016, the size of net pension wealth relative to lifetime salary income ranged from four percent (Illinois) to 30 percent (Kentucky), with a median of 17 pe rcent. Compared to teachers hired in 2007, net pension wealth represents a much smaller share of lifetime salary for those hired in 2016. Table 1.16 compares how net pension wealth changed between 2007 and 2016. The first three columns contain the differe nces of net pension wealth measured as 2016 dollars, the percentage of FAS, and the percentage of lifetime salary before and after recent pension reforms. On average, a hypothetical teacher hired in 2016 would expect to receive $62,179 less net pension wea lth, 85 percent less of her FAS, and seven percent less lifetime salary compared to those hired in 2007. To describe the percentage change in the net pension wealth before and after recent pension reforms, I calculate the rates of change in each state usi ng the following formula: (7) where is one of the measures of net pension wealth described above in state s . Column 4 in Table 1.16 shows that the net pension wealth rate s of change ranged from positive nine percent (District of Columbia) to negative 76 percent (Illinois). When net pension wealth is measured by 22 the percentage of FAS and lifetime salary, the largest rate of change is negative 85 percent, in Massachusetts (s ee column 5 and 6). All three measures of net pension wealth show that hypothetical teachers hired in 2016 would expect to receive about 25 percent less net pension wealth compared to those hired in 2007. Table 1.17 summarizes the changes in pension parameters and starting salary in each of the 40 states and the District of Columbia. Between 2007 and 2016, 37 states enacted less generous parametric pension changes. Twenty - seven states lowered the starting s alaries for their new teachers, and 14 states increased the starting salaries instead. Among the four states that did not enact pension changes between 2007 and 2016, Arkansas and North Carolina decreased teacher starting salary while the District of Colum bia and Connecticut increased it. Because changes in pension parameters and teacher salary both result in changes in pension wealth , to distinguish the effect of pension parameter changes on pension wealth, t he blue bars in Figure 1.1 reflect the changes teacher salary holds constant overtime. The green and the blue bars combine to reflect the effect of changes in pension parameters and teacher salary on pension wealth. Figure 1.1 shows that the increases in pension wealth were all driven by salary increases. Furthermore, it shows that teachers in states that undertook larger pension cuts, such as Massachusetts and Illinois, experienced salary declines. 23 Overall, I find that in most states, mo st of the pension wealth changes were driven by pension parameter changes. The second and fourth columns of Table 1.18 compare the rankings of net pension wealth, measured by the percentage of FAS, where 1 is the most generous and 41 is the least. Column s 5 23 23 and 7 compare how expected net pension wealth was affected for teachers hired before and after the pension changes in terms of the magnitude difference and percentage declined, respectively. Columns 6 and 8 compare the rankings of columns 5 and 7, wher e 1 is the largest cut and 41 is the smallest. Results show that states that ranked among the most generous pre - reform experienced larger net benefit cuts as measured by magnitude difference and percentage decline . However, a few states including Illinois, New Jersey, and New Hampshire that were among the least generous pre - reform also experienced large benefit cuts that decreased net pension wealth by more than 50 percentage points. The four states that did not enact pension changes between 2007 and 2016 b ecame relatively more generous as measured by percentages of FAS when other states reduced rd in 2007 to the 18 th in 2016. 1.6 Sensitivity to Assumptions All of these simulations are for a hypothetical teacher with the set of assumptions described. While this paper provides a basic comparison of pension generosity across states and over time, the actual benefits and costs of participating in pension plans vary dependin career path and expected longevity. To consider the sensitivity of my findings to changes in these assumptions, I recalculate net pension wealth from Table 1.15 using different assumptions. Table 1.19 presents the simulated pension wea lth for teachers hired in 2007 by expected years to live. Unsurprisingly, longer expected longevity increases expected net pension wealth. I find that net pension wealth, measured as the share of lifetime salary, increased by six percentage points for ever y five additional years in retirement. Therefore, teachers who live longer would experience larger cuts in their expected pension wealth compared to those with shorter lifespans. However, if 24 rs of service and annual benefits, longer lifespan would allow teachers to receive more years of high annual benefits and could decrease the magnitude of the cuts. Table 1.20 contains simulations of the effect of changing the assumed discount rate on net pension wealth for teachers hired in 2007. I find that a one percentage point increase in the discount rate would reduce net pension wealth as measured by the share of lifetime salary by about 10 percentage points. Therefore, teachers who face higher disc ount rates would experience smaller cuts in their expected net pension wealth. Table 1.21 contains simulations of the effect of changing assumed COLAs on net pension wealth for teachers hired in 2007. Results suggest that a one percentage increase in COLA rate would increase net pension wealth as measured by the share of lifetime salary by about 4 percentage points. Therefore, teachers who expected higher COLA rates would experience larger cuts in their expected net pension wealth. Recall from Table 1.12 t hat increasing retirement eligibility age or years of service increases shortens the number of years she receives annual benefits . Under conditions where teachers do not work longer but still wait longer to start collecting retirement benefits, their annual benefits would be lower and they would experience larger cuts in their expected net pension wealth. 1.7 Social Security and Teacher Pensions In addition to participating pension systems, many teacher s mandatorily participate in another defined benefit plan Social Security. Perhaps more surprisingly, about 40 percent of all U.S. public school teachers are not covered by Social Security (Kan and Aldeman 2014). When 25 Social Security was first created in 1 935, all state and local government employees, including public school teachers, were excluded from the coverage. The exclusion was based on constitutional concern of whether the federal government should be allowed to impose a tax on state government. In the 1950s, Congress enacted Section 218 along with an amendment of the Social Security Act, allowing states to voluntarily extend Social Security coverage to state and local government employees. 24 Starting in 1991, the Omnibus Budget Reconciliation Act of 1990 extended mandatory Social Security coverage to all state and local government employees who were not enrolled in Social Security or a Social Security equivalent pension system (The United Sta tes Social Security Administration 2018). In my sample of 40 states and the District of Columbia, 27 states automatically enroll their teachers in Social Security (last column of Table 1.13). In the remaining 13 states and the District of Columbia, teacher s are exempt (or partially exempt) from Social Security. 25 In states that enroll their teachers in Social Security, teachers and their employers each pay burden of 12.4% to fund the Social Security system. 26 To be eligible to collect full Social Security payments, a person must have worked and paid the payroll tax for at least 10 years and must be age 67 or older for those born in 1960 or later. Teachers collect So cial Security upon retirement until death . Earnings (AIME), calculated as the average monthly salary in the 35 years of employment where y Insurance Amount (PIA), the monthly benefit under normal retirement age, is then calculated as 90% of AIME up to the first bend point, 32% of AIME 24 25 26 26 exceeds the first bend point but less than the second bend point, plus 15% of AIME exceeds the second bend point (The United States Social Security Administration 2018). 27 Pension plans usually offer higher formula multipliers for teachers who do not participate in Social Security. For example, in 2007 four out of the five most generous pension plans as measu red by pension wealth as a percent of FAS (Massachusetts, Kentucky, Louisiana, and Colorado; see Table 1.13) do not allow their teachers to participate in Social Security. In addition, these states use relatively high multipliers (2.35 to 2.50; see Table 1 pension benefits. In contrast, during 2007, four of five least generous states (Minnesota, New Hampshire, New Jersey, and Wisconsin) enroll their teachers in Social Security. These states use below - average multipliers (1.6 to 2.0; see Table 1.5) when calculating annual pension benefits, and four of five of them. Recall that many states enacted pension reforms because the 2008 financial crisis lowered investment returns and exacerbated the pension funding crisis. Some sc holars argued that many states with no Social Security coverage failed to meet the fundamental requirement of not enrolling in Social Security (the Omnibus Budget Reconciliation Act of 1990) (Doherty et al. 2012) . To study whether this argument holds, I first examine how pension funding level in states with no Social Security coverage has changed over this period, compared to states that enroll teachers in atio ( actuarial assets divided by the actuarial liability) data from the Public Plans Database. 28 Figure 1.2 plots average pension funding ratio by year from 2007 through 2016 for states that enroll their teachers in Social Security versus 27 28 27 states that do no t. The figure shows that states that do not enroll their teachers in Social Security experienced a steeper drop in the funding ratio between 2008 and 2009. This suggests that the 2008 financial crisis hit the states that do not enroll their teachers in Soc ial Security harder. Second, I examine whether states that suffered from increasing underfunded pension liabilities enacted larger pension benefit cuts. In Figure 1.3, I show the relationship between net pension wealth cuts and pension funding ratio, agai n by whether states enroll teachers in Social Security. The negative relationship between pension funding ratio and the magnitude of pension wealth cuts suggests that states enacting the harshest cuts sustained more serious financial trouble. In addition, harsher cuts given the same level of underfunding. In fact, I find that the two states that experienced the largest percentage change in net pension wealth un der recent pension reforms do not enroll their teachers in Social Security (Massachusetts and Illinois; see Table 1.18). 1.8 Conclusion This paper demonstrates large differences across states in the present value of net pension wealth and net benefit cuts. Prior to recent pension changes, the simulated values of net pension wealth in some states were more than triple those of the least generous state, and recent pension changes enlarged these differences by making net pension wealth in some states more than six times as large as the least generous state. I find that states that ranked among the most generous pre - reform experienced larger net benefit cuts. However , a few states including Illinois, New Jersey, and New Hampshire that were among 28 the least generous pre - reform also experienced large benefit cuts that decreased net pension wealth by more than 50 percent. I also find that states that do not enroll their teachers in Social Security experienced worse pension funding crises and harsher cuts than other states. If teachers respond to these changes, this could have implication for both state budgets and teacher labor market. To conclude, this stud y provides several policy implications. First, my paper provides information for state policy makers who want to learn about how their state pension plan and salary scale compare to those of neighboring states, especially for those struggling to staff thei r classrooms. Second, teachers without strong geographic preferences might choose to teach in a state that is more generous. Last, some scholars suggest that all teachers should join Social Security, a more secure defined benefit pension plan (Doherty et a l. 2012). Given the dramatic declines in teacher pensions among states that do not enroll their teachers in SS, future research should consider more carefully whether states are meeting the fundamental requirement in order to exclude their teachers in SS. 29 CHAPTER 2 THE EFFECT OF RECENT TEACHER PENSION REFORMS ON NEW TEACHER QUALITY 2.1 Introduction Teacher pensions represent a large share of lifetime compensation for public school teachers those who qualify for full pensions receive total pension payments worth about 46 percent of their lifetime salary ( Toutkoushian et al. 2011). 29 The cost to mainta in teacher pensions is high states and school districts in the United States pay more than $50 billion toward teacher pensions every year (Aldeman and Robson 2017). The increasing burden of pension on school districts forces policy makers to consider whe ther current teacher pension systems succeed in attracting and retaining better teachers. Empirical evidence consistently shows that senior teachers respond to pension retirement - timing incentives (Furgeson, Strauss, and Vo gt 2006; Costrell and McGee 2010 ; Costrell and Podgursky 2009; Koedel and Xiang 2017 ). A few papers also find that young teachers respond to cross - state differences in pension generosity and pension characteristics such as vesting requirements. ( Munnell and Fraenkel 2013; Papke and Litwok 2013). Nevertheless, we know very little about the influence of teacher pension incentives on teacher recruitment. Today, many teacher pension systems report large underfunded liabilities. In the past, some states required zero or very low teacher contrib ution rates; some states promised pension benefits that they could not afford; and some states and local governments paid less than their legally required contributions (Doherty et al. 2012). The 2008 financial crisis lowered investment returns and exacerb ated the pension funding crisis. To reduce pension debt, a few states changed their 29 30 pension structure from traditional defined benefit (DB) pension plan to plans that shift investment performance risk from the state to teachers. 30 Most states, moreover, eit her reduced benefits, increased teacher contributions, or both. Because most states have laws that prevent benefit cuts for current employees, significant changes to pension plans often only apply to new teachers. These changes provide an identification st quality and discovering the pension factors to which teachers respond. The hypothesis is that lower pension benefits, all else equal, would discourage prospective teachers, particularly teachers with be tter alternatives, from entering the profession. However, if new teachers are not forward looking or know little about their pension plans, the effect of pension changes on teacher recruitment may be small ( Goldhaber et al. 2017) . As many states continue to struggle with pension issues, understanding whether and how much new teachers respond to pension benefits gives policy makers a more complete picture of benefits and costs associated with potential pension reforms. I use selec institutional selectivity could be a good predictor of her aptitude. In particular , I use the 25 th 31 I also use an indicator for whether the teacher has a postgraduate degree and an indicator for whether the teacher majored in the subject she teaches as o ther teacher quality measures. Using pooled cross - sectional teacher - level data from the 2003 - 2004, 2007 - 2008 and 2011 - 2012 Schools and Staffing Survey, I estimate the differential change in new teacher quality before and after the cuts between states and c ities that enacted pension reforms between 2007 and 2011 and those that did not. My 30 31 31 analysis focuses on 2,640 first - year teachers who participated in a mandatory, traditional defined benefit plan between 2003 and 2011. Moreover, I consider year - to - year va riation in other time - related factors that might affect teacher entry. Over the past three decades, every state has reported teacher shortages to the U.S. Department of Education (U.S. Department of Education Office of Postsecondary Education 2017 ). 32 Struggling to staff their classrooms, some states lowered testing requirements for entry into the teaching profession ( Partelow 2015). Some school districts, on the other hand, used financial incentives like signing bonuses and free training to recruit te achers, especially in fields where the shortages are greatest. Entry barrier changes and more attractive district recruitment policy can affect teacher quality (Hanushek and Pace 1995; Figlio 2002) . Existing literature also consistently finds that relative ( Nagler, Piopiunik and West 2015; Munnell and Fraenkel 2013; Bacolod 2007; Figlio 2002 ). Over the last century, the e to other professional fields. The relatively low salary could make teaching a traditionally female - dominated occupation less attractive. On the other hand, the Great Recession during the late 2000s and early 2010s reduced the outside job options for te achers and could make teaching more attractive. To better understand the effect of pension reforms on new teacher quality, my analysis accounts for a complete set of time - varying covariates for these contemporaneous changes in state and district policies a nd economic conditions. My estimates show that potential new teachers respond to pension incentives offered by school districts. In states and cities that enacted pension reforms, schools hire teachers from undergraduate 32 32 institutions 25.63 points lower (m ore than one fifth a standard deviation) 25 th percentile SAT scores. I also find that a one percent increase in the present value of net pension wealth results in schools hiring new teachers from undergraduate institutions with 32.15 points higher (more than one forth a standard deviation) 25 th percentile SAT scores. In the next section, I review the relevant literature. Section 2.3 describes the pension data and institutional details. Section 2.4 provides a description of the teacher data from the Schools and Staffing Survey and how I match these to district level pension data. Section 2.5 presents the empirical specifications and results, and Section 2.6 concludes. 2.2 Related Literature 2.2.1 Teacher Pensions and Teacher Turnover Much of the literature on teacher pensions focuses on teacher reti rement. Costrell and McGee (2010 ) find tha t both peak value (the difference between current pension wealth and the maximum possible pension wealth) and one year accrual (the difference between current pension wealth and the pension wealth after one additional year of work) have negative effects on retirement. Some researchers study the effect of pension incentives on teacher retirement in specific states using administrative data that capture teacher exit behavior and earnings history (Furgeson, Strauss, and Vo gt 2006; Costrell and McGee 2010 ; Brown and Laschever 2012; Brown 2013; Koedel and Xiang 2017 ). They find that senior teachers respond to pension retirement - timing incentives. Two other papers explore policy changes affecting pension wealth and employ difference - in - difference models to estimate the effect of pension incentives on the timing of tea cher exit decisions. Furgeson, Strauss, and Vogt (2006) find a large increase in teacher retirement when Pennsylvania 33 enacted a pension policy which temporarily increased early retirement benefits between 1997 - 1998 and 1998 - 1999. Koedel and Xiang (2017) in vestigate a policy change in St. Louis that increased teacher retirement benefits for those who retired after the 1998 - 1999 school year. Using administrative panel data from the Missouri Department of Elementary and Secondary Education, they find a tempora ry delay in retirement among teachers who were eligible for retirement in the previous school year. However, they find no significant retention effect among teachers who were not eligible for retirement. They offer a possible explanation that teachers may lack of full knowledge of their pensions. Some papers ask how DB pension parameters affect teacher turnover by comparing differences across pension plans. Costrell and Podgursky (2009) analyze the time pattern of pension incentives by simulating pension - to late fifties, and drop off sharply over the next few years. This backl until they reach the time when pension value spikes. Once they pass the pension value spike, Friedberg and Turner (2011) use a peak value model to estimate the effect of variation in pension parameters on teacher retirement across 17 pension plans. Using teacher - level data from the School and Staffing Survey (SASS), they find that teachers tend to delay retiring whil e pension wealth accumulates, then retire abruptly after reaching peak pension wealth. Two other papers explore policy changes affecting pension wealth and employ difference - in - difference models to estimate the effect of pension incentives on the timing of teacher exit decisions. Furgeson, Strauss, and Vogt (2006) find a large increase in teacher retirement when 34 Pennsylvania enacted a pension policy which temporarily increased early retirement benefits between 1997 - 1998 and 1998 - 1999. Koedel and Xiang (2017 ) investigate a policy change in St. Louis that increased teacher retirement benefits for those who retired after the 1998 - 1999 school year. Using administrative panel data from the Missouri Department of Elementary and Secondary Education, they find a tem porary delay in retirement among teachers who were eligible for retirement in the previous school year. However, they find no significant retention effect among teachers who were not eligible for retirement. They offer a possible explanation that teachers may lack of full knowledge of their pensions. Unlike defined contribution (DC) plans that tie retirement benefits to total contributions and are subject to market fluctuation, defined benefit plans guarantee retirees a specific level of annual benefits calculated as the product of a multiplier factor, within - system experience, and final average salary. Other papers compare DB plans and other pension plans. For example, Gustman and Steinmeier (1993) use the Survey of Income and Program Participation (SIPP) to compare worker turnover between DB plans and DC plans. While DC plans accrue benefits overtime, DB plans backload pension benefits the formula used to calculate guaranteed annual benefits relies on within - system years of service (YOS) and final average salary (FAS). DC plans are not backloaded: th ey do not guarantee retirees minimum or maximum pension benefits and allow vested workers to take their full retirement saving with them when moving from one job to another. Gustman and Steinmeier (1993) find that pension coverage was associated with lower one - year turnover rate regardless of pension types (DB or DC). Friedberg and Owyang (2005) study the link between DB coverage and job tenure. Using data from 1983 - 2001 Survey of Consumer Finance (SCF) and 1993 Current Population Survey (CPS) and controlli ng for job characteristics, they find that workers with DB pensions work in the same job longer than workers with no pensions or with 35 DC pensions. 33 They also find that workers with more generous DB pensions stay longer, controlling for levels of earning. G oldhaber et al. (2017) investigate whether moving away from a backloaded defined benefit plan increases levels of employee turnover. Using data from Washington Teacher Retirement system, the authors compare the turnover rate of teachers who enrolled in eit her a traditional defined benefit (DB) plan or a hybrid plan (combines DB and DC) during the same period of time. They find that turnover was lower among teachers who transferred out of the DB plan into the hybrid plan. Papke and Litwok (2013) ask whether cross - state pension presence of defined contribution alternatives is positively and significantly related to the hazard rate of first exit. They also find that more stringent vesting requirements are positively and significantly related to the hazard rate of first exit. 2.2.2 Teacher Pensions and Teacher Quality 33 36 34 34 37 2.2.3 Other Factors that Attract Teachers A large l iterature exists on other factors that influence the decision to enter teaching and the quality of new teachers. The key explanatory variable of most of these studies is the role of salary and relative pay in teaching ( Nagler, Piopiunik and West 2015; Munn ell and Fraenkel 2013; Bacolod 2007; Figlio 2002 ). For example, Figlio (2002) uses district - level data from the SASS and uses undergraduate institution and an indicator of whether a teacher majored in the subject she teaches. He - qualified teachers incr eases with higher teacher salaries, which implies that district recruitment policies (for example, signing bonuses) matter when attracting teachers. cycle conditions as a source of that teachers who entered the profession during recessions are significantly more effective in raising student test scores. Other research shows that teacher licensing requirements affect new te and teacher quality (Berger and Toma 1994; Hanushek and Pace 1995). For example, Berger and certification and student performance when measured (1995) use a panel data from the High School and Beyond to trace a group of students who aspire to be a teacher when they are in high school. They find that the state requirements of courses and 38 tests for teacher c ertification lower the probability that students complete their teacher preparation program, which results to a smaller pool of trained teachers. 35 2.3 Institutional Details and Pension Data In the U.S., all full - time public school teachers automatically participate in public pension plans. While each state runs a pension system that covers teachers (state - run pension plans), some large cities like New York and Chicago operate their own municipality - specific pension plans . 36 T able 2.1 shows the types of pension plans tha t states and cities offer teachers during the years I consider. In this paper, I restrict my analysis to the 36 state - run pension plans and 11 municipality - specific pension plans that automatically enroll teachers into a mandatory, traditional defined bene fit plan throughout the years 2003 to 2011. 37 In defined benefit plans, teachers contribute a portion of salary towards the plan while 35 36 37 39 employed. Teachers who retire after becoming vested collect annual payments until their death. 38 The lifetime pension wealt h of a teacher depends on the size of initial annual payments, adjustments made for cost of living, and the length of time in retirement. All defined benefit plans use a similar formula to calculate annual payments: (3) First - year Annual payments (A) = FAS * Multiplier * Years of Service where FAS average salary that she receives for each year of service. The years more years of service lead to higher annual payments. But some states cap the first - ye ar annual - year annual payments is: (4) In order to counteract the impact of inflation on retirement income, public pension plans provide retirees with post - retirement cost of living adjustments (COLAs). Plans with higher cost of living adjustments are more beneficial to teachers . The pension parameters discussed above relate to the size of annual benefits. Other pension parameters such as retirement age and teacher contribution amount may also affect pension wealth, 38 40 without directly changing annual benefits. 39 Because teachers col lect annual benefits from retirement until death, an earlier retirement age causes teachers to receive more annual payments and enjoy higher lifetime pension wealth. Recall that most states require teachers to contribute a portion of salary to fund their p teacher contributions (see equation (1)), increasing contribution rate decreases net pension wealth. To more precisely evaluate pension plan generosity, I calculate net pension wealth as the actuarial present value of the payment stream that teachers expect to receive upon retirement, net of employee contributions. 40 S uppose a teacher started teaching in year t=1 , retires in year t=T , and collect pension benefits for d years, from t=T+1 to t=T+ . Then, her expected present discounted net pension wealth (PDNW) is calculated using the following formula: (3) = - where is the capped annual payment in year t >T, is the cost of living adjustment in year t, is the probability of surviving at year t, r is the discount rate, and is the teacher contribution in year t T . and summaries of legislation. I obtained additional pension plan - level data on pension funding ratio and employer contribution rate from the Public Plans Database (PPD). 41 It is possible that some new teachers enter teaching with the intent to teach for a short period of time, are uncertain how long they will remain in the profession, or are likely to be geographically 39 40 41 41 mobile. These teachers would care more about pension vesting rules, the amount they can withdraw if they leave before vested, and whether they have Social Security coverage. Vesting rules determine how long a teacher must work under the same pension plan to eligible for pension benefits. Teachers who leave before being vested not only lose future rights for pension benefits but also lose accrued funding save d teachers who leave before being vested can only withdraw their own contributions, sometimes with interest, and rarely with employer contributions. Therefore, the longer the vesting rule is (or the less teachers can withdraw before being vested), the les s attractive it is to teachers who do not plan to stay long. In addition to participating in pension systems, teachers in 31 pension plans mandatorily participate in another defined benefit plan Social Security (Table 2.3). Teachers who do not participate in Social Security have more dependency on their pension plans. In general, pension plans that allow teachers to participate in Social Security require lower teacher contribution rate and offer a lower formula multiplier. While some potential teachers migh t treat Social Security as a substitute for a teacher pension plan and value Social Security coverage for its portability across state lines and jobs for low risk of retirement benefits, some might find Social Security coverage less attractive because they need to contribute an extra 6.20 percent of their salary to fund Social Security. In response to the financial crisis among pension plans, many pension plans enacted reforms between 2007 and 2011 that reduced teacher benefits or increased required teacher contribution rates. Although legal constraints prevent most states from cutting benefits for current employees, all states can change the benefit rules affecting employees not yet hired. Therefore, all the pension parameters changes discus sed below affect teachers hired in the 2011 - 2012 school year. Table 2.3 provides an overview of the numbers of years used to calculate FAS. In 2007, 32 pension plans used average salary for the three or fewer of the highest years of salary. 15 pension 42 plan s used a similar definition with more than three years. Compared to 2007, 8 pension plans increased the number of years used to calculate FAS by 2011. While 25 pension plans used three or fewer years of salary, 22 pension plans used more than three highest years of salary to calculate FAS by 2011. Table 2.4 provides an overview of the multiplier used in each pension plans. In 2007, 37 pension plans use a constant multiplier that ranged from 1.55 percent (South Dakota) to 2.67 percent (Nevada). 10 pension plans used multipliers that varied by retirement age or years of service. Compared to 2007, eight pension plans changed their multipliers in ways that reduced benefits by 2011. Table 2.5 shows that 23 pension plans limited the annual payment to a specific percentage of By 2011, Pennsylvania changed from no limitation of annual payments to 100 percent of the inating the 80 percent cap. The pension plans also vary by their cost of living adjustments. Table 2.6 provides an overview of COLAs in each pension plans. In 2007, 34 states automatically adjusted cost of living based on a specific percentage or the Cons umer Price Index (CPI). On the other hand, 12 pension plans relied on state legislation to decide COLAs for each year, and two states tied their COLAs to their funding level. By 2011, ten pension plans changed COLAs in ways that are less generous. The chan ges include lowering the COLA rate, delaying the receipt of COLA, or switching from automatic adjustments to adjustments that tied COLAs to funding levels or other indexes. Table 2.7 provides an overview of retirement eligibility in each state. Compared t o 2007, 18 pension plans raised retirement eligibility age or years of service by 2011, which would shorten the length of years receiving annual benefits for many teachers. 43 Table 2.8 provides an overview of the employee contribution rate in 2007 and 2011. In 2007, Wyoming did not require their teachers to make any contribution to their pension plans. Other pension plans required their teachers to make contributions ranging from three to 12 percent of their salary. 40 pension plans required teachers to cont ribute at least a five percent of their salary. By 2011, every pension plan required their teacher to make contributions that ranged from 1.43 to 14 percent of their salary. 17 pension plans increased their employee contribution requirements between 2007 a nd 2011. Table 2.9 provides an overview of vesting rules. Seven pension plans increased the time to vesting between 2007 and 2011. Once exception to the increase in time is Kansas, which moved from a 10 - year vesting rule to a 5 - years vesting rule, but len gthened years used to compute FAS and increased time to retirement eligibility, so vesting is not worth what it was previously. Table 2.10 provides an overview of how much pension plans permit teachers to withdraw from their pension plan if they leave before being fully vested. I use the indicators presented in Table 2.10 to measure pension portability. The majority of pension plan s (81 percent) allow teachers to withdraw their own contribution plus interest if they leave before being vested. While three pension plans allow teachers to withdraw more than their contribution plus interest, five pension plans only allow teachers to wit hdraw their contributions and no interest if they leave before being vested. In Colorado, a teacher hired before 2011 may withdraw her contributions, interest, and partial employer contributions. However, a teacher hired in 2011 or later who leaves before being vested can only withdraw the employee contribution plus interest. 44 2.4 Teacher Data The teacher data I use come from the Schools and Staffing Survey (SASS), the largest and most extensive survey of elementary and secondary teachers, administrators, s chools, and school districts in the U.S. today. The National Center for Education Statistics (NCES) initially conducted the survey in school year 1987 - 1988 and conducted the survey seven times since then. I use the 2003 - 2004, 2007 - 2008 and 2011 - 2012 waves, because the most recent available data are from the 2011 - 2012 wave, and the earliest available data that contain consistent questions on district - level hiring practices over time are from the 2003 - 2004 wave. To be consistent with the pension parameters te achers faced when they started teaching, I will index school years by the calendar year in which a school year starts hereafter. For example, I will refer to the 2011 - 2012 school year as 2011. The SASS data have a number of strengths for this study. First , the information on each measures of teacher quality. Third, the r estricted - use state and zip - code identifiers allow me to match individual teachers to their pension parameters and state variables. Fourth, the information earnin 42 Lastly, the variation of interest in this paper is at the state by year level, and with the proper weighting, each wave of SASS is designed to be representative at the state level. 42 45 One of the weaknesses of using SASS is that I only observe workers who entered teaching but not those who considered teaching and later decided not to. Therefore, I do not directly measure how pension incentives affect potential teachers. Instead, this paper focuses on the effect of pension incentives on the composition of newly hired teachers . Another weakness of using SASS is that I only observe teachers in limited years (2003, 2007, and 2011), which only leaves potential teachers a few years to respond to the chang es in pension incentives. It will take more years of post - treatment data to observe a larger change in new teacher composition if recent pension Because pension re forms should affect newly hired teachers the most, I restrict my sample to newly hired teachers who work full time. 43 I define newly hired teachers as those who have no teaching experience before the survey year. 44 Overall, there are 3,410 teacher - year obser vations in 2003, 2007, and 2011 in the 36 states of interest. Integrated Postsecondary Education Data System (NCES IPEDS). NCES provides the 25 th percentile entering SAT/ACT scores for schools that require test scores for admission and have at least 60 percent of enrolled students submitting a test score. For teachers who attended an undergraduate institution that provides only ACT scores, I conver t ACT scores to SAT scores teachers who attended undergraduate institutions that did not collect or report SAT/ACT scores, which eliminates 440 teacher - year observat ions. I also exclude teachers at charter schools, which eliminates 210 teacher - year observations, because pension mandates do not always apply to charter 43 44 46 schools. Finally, I exclude teachers who are above age 50 from the sample, which eliminates 70 teacher - year observations, as older entrants are too close to retirement to accumulate much pension wealth. The final teacher - level dataset I use for the analysis contains 2,640 teacher - year observations. The other measures of teacher quality come directly from S ASS teacher education background information, including an indicator (Advanced Degree ) for whether the teacher has a postgraduate degree, and an indicator ( Subject Major ) for whether the teacher has specific training in the subject she teaches. 45 Teachers w ith specific training in the subject they teach, all else equal, are expected to be better qualified compared to other teachers who teach that subject (Figlio 2002). Because school district administrators often claim to want math and science teachers who m ajored in their subject (Angrist and Guryan 2008), I code Subject Major as a dummy that equals one if the teacher either a B.A, M.A., Ph.D., Certificate of Advanced Graduate Studies, or Education Specialist assigned is Natural Science, and she completed either a B.A, M.A., Ph.D., Certificate of Advanced Graduate Studies, or Education Specialist degree with a major in either Biology, Chemistry, Earth Science, Physics, or another Natural Science. Using restricted SASS data, I identify the state in which schools are located and match individual teachers to their sta te variables. NCES provides data on state testing requirements for initial certification of elementary and secondary teachers. 46 45 46 47 I use the following procedure to match individual teachers to their pension plan and benefit group. First, I assume that school s are covered under a municipal - specific pension if their U.S. Postal Service zip codes match the zip code of the big cities that operate their own pension plans. I then assume the state - run pension plans that enroll teachers automatically cover all remain ing 47 With the school residence and information of the year the teacher answered the survey, I then match individual teachers to their pension plan and benefit group. To control for outside options when teachers decide whether to stay, I calculate Alternative T eacher Pay administration from the March Current Population Survey (CPS). I normalize all salaries in this paper to 2011 dollars using the CPI for all items less food and energy, reported by the Bureau of Labor Statistics . I also include state unem ployment rates from the Bureau of Labor Statistics as a measure of the outside labor market options. Table 2.12 and Table 2.13 report weighted descriptive statistics for subgroups of teachers in the treatment group (states and cities that reduced benefits) and the control group (states and cities that did not reduce benefits). The top panel of Table 2.12 shows the summary statistics of the quality measures. Between the pre - treatment (2003 and 2007) and post - treatment (2011) periods, the average 25 th percen tile SAT score in the treatment group decreased from 975 to 962 (on a scale of 1,600), while the average 25 th percentile SAT score in the control group increased from 940 to 47 48 961. The means of Advanced Degree and Subject Major show that both groups experienced a higher share of teachers who majored in the subject they taught and a higher share of teachers who ente red teaching with an advanced degree over this period, especially in the control group. The next panel characterizes state and city level pension variables. The mean of Vesting Rules shows that it takes on average two years longer for treatment group teach ers hired post - treatment to vest, compared to those hired pre - treatment. On the other hand, post - treatment control group teachers became vested 0.66 year earlier than pre - treatment control group teachers. The mean of pension funding ratio shows that pensio n fund problems worsened over this period, especially in the treatment group (a drop from 86 percent funded to 72 percent). 48 Higher Teacher Contribution Rate may discourage high quality individuals from entering the profession , because it decreases teache - treatment periods, treatment group teachers contributed 0.98 percentage point more of teacher salary to fund pension post - treatment, and control group teachers contributed 0.26 percentage point more of teacher salary. Hi gher Employer Contribution Rates could also affect the quality of teachers attracted into the profession if higher employer contributions reallocate school resources away from current school expenditures that attract teachers. On the other hand, new teache r quality may increase if school districts demand fewer new teachers because of the higher pension cost. Compared to pre - treatment periods, school districts in both groups contributed a higher potion of teacher salary to fund pension plans from 7.84 percent to 11.87 percent in the treatment group and from 10.98 percent to 14.62 percent in the control group. 48 49 As discussed in Section 3, pension plans reformed in numerous ways including decreasing multipliers, lengthening the years used to c ompute final average salary, raising retirement eligibility age or service, capping the annual benefit, increasing vesting rules, increasing teacher contribution rates, and lowering cost of living adjustments. These pension reforms changed n wealth differently across states, cities, as well as individuals (depending on 49 To more precisely measure the change in pension incentives facing newly of entry into teaching, age, gender, and starting salary using equations (3) - (5) . See appendix A for calculation details. The mean of Net pension wealth shows that among teachers hired pre - treatment, treatment group teachers receive o n average 47 thousand dollars more net pension wealth compared to those in the control group. However, among teachers hired post - treatment, treatment group teachers receive on average 58 thousand dollars less net pension wealth compared to control group te achers. The bottom panel of Table 2.12 reports means of matched outside options variables. Compared to pre - treatment periods, the treatment and control groups both experienced higher unemployment rates. Over this period, the alternative teacher pay in the treatment group increased from 136 percent to 140 percent, compared to a decrease from 140 percent to 137 percent in the control group. 49 For example, Mr. Ara started teaching at the age of 35 and Mr. Brown started teaching at the age of 30. Both of them teach in Louisiana , which changed its retirement eligibility from ( 60/5; 55/25; A/30) to 60/5 . Before the policy change, both Mr. Ara and Mr. Brown will work 25 years until they are eligible to start collecting benefits. However, after the policy change, Mr. Ara would still work 25 years while Mr. Brown would have to work 30 years to start collecting benefits. 50 The top panel of Table 2.13 compares state requirements for becoming a teacher. 50 State required testing for teacher certification creates entry barriers to teaching professional that may increase or decrease teacher quality. While testing requirements set a higher knowledge bar for individuals to enter teaching, the entry barriers als o increase labor cost which could discourage individuals with higher outside options to become teachers (Angrist and Guryan 2008). Between the pre - treatment and post - treatment periods, the share of treatment group teachers teaching in states that required a basic skill test increased from 34 percent to 37 percent, while the share of control group teachers decreased from 78 percent to 70 percent. On the other hand, the share of treatment group teachers teaching in states that required a knowledge of teaching test or a teaching performance test decreased over this period, while the share of control group teachers remained the same. Both groups experienced a smaller share of teachers teaching in states that required a subject matter test. The next panel report s whether the school district uses incentives to recruit teachers in the survey year. Between the pre - treatment and post - treatment periods, school districts in both groups used fewer financial incentives to recruit new teachers, especially in the control g roup. The bottom panel of Table 2.13 reports individual teacher characteristics. Compared to the pre - treatment period, the average age of newly hired treatment group teachers decreased from 29 round 28. Compared to treatment group teachers, the share of female control group teachers increased between the pre - and post - treatment periods. 50 51 2.5 Methodology and Estimates 2.5.1 Pension Reforms and Teacher Quality In this section, I examine whether new teacher quality changed differently between the pre - benefit cuts and post - benefit cuts periods among states and cities that enacted pension reforms relative to those that did not. Table 2.14 summarizes the parameter c hanges between 2007 and These pension reforms generate a decrease in incentives to become public school teachers among potential teachers. Teachers who taught in states and cities listed in Table 2.14 belong to the treatment group in the following analysis. Using teacher - level data from SASS, I compare measures of new teacher quality in the pre - treatment periods (2003 and 2007) and the post - treatment period (2011) . I separately control for the pension parameters that do not directly affect the calculation of pension benefits. I begin by estimating the following difference - in - differences model: = , where is one of the measures of quality described above, i indexes the teacher, k , and s indexes the scho ol district and state in which the teacher worked respectively, and t indexes the year teacher started teaching (2003, 2007, or 2011). is an indicator that equals one if teacher i taught in a state or city that enacted benefit - reducing reforms from 2007 to 2011 (listed in Table 2.14). is an indicator for year 2011. The coefficient on captures the effect of benefit - reducing reforms on teacher quality. The vector includes and separate indicators for whether the state had required testing (basic skills exam, knowledge of teaching 52 exam, teaching performance exam, and subject matter ex am) for certification of elementary and secondary teachers. The vector includes teacher contribution rate, employer contribution rate, pension funding ratio, and separate indicators for whether the school district uses some incentives (signing bo existing staff for new teacher referrals, free training for fields with teacher shortages, and pay incentives to teachers who teach in fields with shortages) t o recruit teachers. All specifications include a full set of state dummies and year dummies ( . To account for the stratified sampling frame used by SASS, I weigh each teacher by the d in teacher - level SASS data. This weight contains for nonresponse rate and other sampling considerations that arise after the sample has been drawn. Tabl e 2.15 reports estimates of from specification (6) separately for the effects of pension reforms on the 25 th two alternative measures of teacher quality whether the teacher has an advanced degree and whether the teacher majored in the subject she taught. In the first column, the esti mate of suggests that under pension reforms, schools are hiring new teachers from undergraduate institutions with 25.63 points lower (more than one fifth a standard deviation) 25 th percentile SAT scores, with a standard error of 14.77. This finding, co nsistent with the hypothesis, suggests that a reduction in pension benefits reduces the quality of teachers attracted to the public schools. In other regressions, t he estimates of suggest that pension reforms lower the probability of schools hiring a n ew teacher with an advanced degree by 4.90 percentage points with a standard error of 0.05 (column 2) and lower the probability of schools hiring a new teacher who majored in her 53 subject by 6.35 percentage points with a standard error of 0.04 (column 3) . A lthough I cannot reject the null hypothesis that the estimated coefficients in column 2 and column 3 are equal to zero, the negative relationships between pension reforms and these two quality measures are intuitive. Table 2.15 also suggests that outside options matter to new teacher quality. The es timates of Alternative Salary Rate in the first two columns suggest that a higher alternative teacher pay attracts a pool of lower quality new teachers, although I find no statistically significant evidence in the third column. In addition, the estimate of Unemployment Rate in the second c olumn suggests that fewer job opportunities outside of teaching attracts a larger pool of teachers with advanced degrees, although I find no statistically significant evidence in the other two columns. These findings, consistent with existing literature, s uggest that lower alternative teacher pay and higher unemployment rates attract higher quality teachers into the profession ( Munnell and Fraenkel 2013; Nagler et al. 2015 ) . The estimates of in the second column suggests that a one percentage point inc rease in contribution rate lowers the probability of schools hiring new teachers with an advanced degree 54 by 2.42 percentage points (standard error of 0.01). This finding, consistent with the hypothesis, suggests that a higher contribution rate is less attr active to teachers. I find no statistically significant relationship between teacher quality measure and other pension parameters in the first and third columns. District recruitment policies appear to attract teachers from better undergraduate institution s. Column 1 suggests that schools in districts that offer a loan forgiveness attract teachers from undergraduate institutions with 36.99 points higher (more than one fourth a standard deviation) 25 th percentile SAT scores, with a standard error of 19.44, a lthough I find no statistically significant evidence of this policy on new teacher quality when measure in two other ways. Column 2 suggests that schools in districts that offer a signing bonus are 7.72 percentage points more likely to hired teachers who e ntered teaching with an advanced degree (standard error of 0.04). Although, I find no statistically significant evidence of the relationship between signing bonus policy and new teacher quality when in the first the third columns, t he positive signs are in tuitive. Interestingly, the second and third columns shows that schools in districts that offer a incentive to attract more qualified teachers into the dis existing staff who referred the new teachers. Moreover, schools might be willing to hire a less qualified teacher referred by an existing staff. 55 Overall, estimates in Table 2.15 suggest that teachers value their pension benefits even when they are far from their retirement eligibility. Less generous pension plans discouraged high quality individuals from entering the profess ion and result to a pool of lower quality new teachers. 2.5.2 Pension Wealth and Teacher Quality Recall that as well as individuals. These heterogeneous effects of the wealth allow me to examine the relationship between the natural logarithm of an individual = , where is the Similar to equation (6), all specifications include a full set of state dummies and year dummies ( , and control for time - varying state - level variables as well as school district characteristics The coefficient of interest in equation (7) is , which is the estimate of a one per cent increase in net pension wealth on new teacher quality. Table 17 reports estimates of from specification 56 (7) separately for the effects of pension reforms on measures of teacher quality. The first column suggests that a one percent increase in net pension wealth results in schools hiring new teachers from undergraduate institutions with 32 .15 points higher (more than one fourth a standard deviation) 25 th percentile SAT scores, with a standard error of 14.47. This finding, consistent with the hypothesis, suggests that school with more generous pension systems attract a pool of higher quality new teachers. In column 2 and 3, t he estimates of suggest that a one percent increase in net pension wealth increases the probability of schools hiring a new teacher with an advanced degree by 2.52 percentage points with a standard error of 0.05 and increases the probability of schools hiring a new teacher who majored in her subject by 2.94 percentage points with a standard error of 0. Although I cannot reject the null hypothesis that the estimated coefficients in column 2 and column 3 are equal to ze ro, the positive relationships between pension reforms and these two quality measures are intuitive. 2.6 Conclusion Much of the existing pension literature focuses on how senior teachers respond to pension incentives. Along with Munnell and Fraenkel (2013) and Papke and Litwok (2013), I am among the first to study how younger teachers respond to pension incentives . To h ighlight the effect of pension incentives on teacher recruitment, this paper studies whether, and to what degree, pension incentives affect new teacher quality. Based on my findings, pension incentives affect the composition of new entrants into teaching. Controlling for outside options and other factors that could affect teacher quality, I find that states and cities that cut pension benefits hire new teachers from undergraduate institutions with 25.63 57 points lower 25 th percentile SAT scores. I also find that a one percent increase in the present value of net pension wealth results in schools hiring new teachers from undergraduate institutions with 32.15 points higher 25 th percentile SAT scores. Note that my results are based on one year of post - treatment data (2011), so my findings only capture immediate changes in new teacher composition after states enacted their pension reforms. Thus, the effect of recent pension reforms on new teacher quality may be larger if recent pension reforms also affec transfer out) from education programs. Overall, my findings suggest that using benefit cuts to reduce pension debt might impair school districts attempts to attract qualified teachers. Recall that pension reforms occurre d during the Great Recession where there were fewer particularly attractive during the Great Recession, I control for time - varying outside options when exami ning the effect of pension generosity on new teacher quality. Consistent with existing literature, I find that lower alternative teacher pay and higher unemployment rates attract higher quality teachers into the profession. In addition, my results suggest that district recruitment policies affect new teacher quality. I find that school districts that offer a signing bonus policy or a loan forgiveness attract more pool of lower quality teacher. Perhaps school districts are willing to hire a less qualified teacher when she is refereed by an existing employee. According to the National Council on Teacher Quality, 22 states made changes to their teacher pension system s in 2012 alone and more states are joining the teacher pension reform (Doherty et al. 2012). Policy makers should consider the unintended cost of benefit cuts on teacher recruitments associated with potential pension reforms. 58 CHAPTER 3 WHO CHOOSES TO STAY? THE EFFECT OF TEACHER PENSION INCENTIVES ON NEW TEACHER TURNOVER 3.1 Introduction Though defined - contribution (DC) pension plans predominate a mong private - sector professions, the majority of states still offer their public - sector employees defined - benefit (DB) pension plans. 51 While DC plans accrue benefits overtime, DB plans backload pension benefits the formula used to calculate guaranteed annual benefits relies on within - system years of service (YOS) and final average salary (FAS). 52 Therefore, teachers accrue pension wealth more rapidly towards the end of their careers. Because DB pensions are rarely portable, it is costly for teachers to leave their pension systems those who leave will not preserve their years of service for pensi on purposes. 53 Moreover, FAS is frozen at the time the worker exits the pension system. Because earnings tend to be highest in the final years of employment , teacher who exit the system earlier will earn pensions based on lower FAS value. This backloading s tructure creates an incentive for teachers to stay. Theoretically, a higher degree of backloading increases the opportunity cost of leaving the pension system and could decrease teacher turnover. Improving retention is particularly important for schools. H igh turnover can significantly Ronfeldt et al. 2013; Boyd et al. 2005; Guin 2004; Hanushek et al. 1999). In recent decades, however, researchers have estimated mixed effects of backloaded 51 52 53 59 pension incentives on teacher turnover . Goldhaber, Grout, and Holden (2017) find a positive effect, while some find no significant effect (Gustman and Steinmeier 1993; Koedel and Xiang 2017), and some find a negative effect (Papke and Litwok 2013; Friedberg and Owyang 2005 ). Vesting rules the time it takes teachers to qualify for guaranteed pensions upon retirement may also provide a financial incentive for new teachers to stay. Figure 3. 1 shows an example of net pension wealth accrual over the early career cycle for a rep resentative Arkansas teacher who started teaching at age 25. In the figure, each point represents the simulated present value of maximum net pension wealth if the Arkansas teacher separates from her pension system immediately at 25. The y - axis presents net pension wealth in 2011 dollars. The teacher accrues nothing towards her pension wealth until vested. Once she reaches 5 years of service, her pension wealth jumps to 133 percent of her starting salary. The jump creates a financial incentive for teachers t o stay until vested. The financial incentive strengthens as teachers approach vesting. Thus , a longer vesting period reduces financial incentives for teachers to stay. Papke and Litwok (2013) find that remaining years to vesting positively correlates with turnover rate. The 2008 financial crisis lowered investment returns and sharply reduced funded levels among state pension plans. 54 Between 2007 and 2011, many states enacted pension reforms to reduce pension debt. Some states either reduced benefits, incre ased teacher contributions, or both. In addition, some states increased the years required to vest. Because most states have laws that prevent benefit cuts for current employees, significant changes to pension plans usually apply to new teachers only. In S ection 3, I illustrate the effect of recent benefit - reducing legislation on over the career cycle . I find that recent benefit - reducing legislation decreases the degree of backloading. 54 60 Using pooled cross - sectional teac her - level data from the Schools and Staffing Survey (SASS) and the Teacher Follow - up Survey (TFS) , I estimate the differential change in new teacher turnover after 2007 - 2011 benefit - reducing legislation between states and cities that enacted pension reform s and those that did not. My analysis focuses on 1,380 first - year teachers who participated in a mandatory, traditional defined benefit plan between 2003 and 2011. Controlling for teacher and school characteristics , I find no significant effect of benefit - reducing legislation on teacher turnover among newly hired teachers . I also examine how new teacher turnover changed between the pre - and post - reform periods among states and cities that increased their vesting rules relative to those that did not. Economic theory indicates a forward - looking, unvested teacher would consider the present value of pension wealth associated with becoming vested and weigh the opport unity cost of quitting. However, if teachers only consider short - term pension wealth accruals, changing vesting rules will have little influence on teacher exit decisions. I find no evidence that more stringent vesting requirements affect the probability o f a new teacher exiting after one year of employment. In the next section, I review the related literature. Section 3.3 describes pension data and institutional details. Section 3.4 provides a description of the teacher data from the Schools and Staffing Survey and the Teacher Follow - up Survey. Section 3.5 presents the empirical specifications and results, and Section 3.6 concludes. 3.2 Related Literature 3.2.1 Teacher Pensions and Teacher Turnover Teacher turnover includes moving to another sch ool and exiting the teaching profession. Much of the teacher pension literature focuses on senior teacher exit be havior. Costrell and McGee (2010 ) 61 find that both peak value (the difference between current pension wealth and the maximum possible pension wealth) and one year accrual (the difference between current pension wealth and retirement rate. Some papers ask how DB pension parameters affect teache r turnover by comparing differences across pension plans. Costrell and Podgursky (2009) analyze the time pattern of accumulates slowly in the early years of her career, then accelerate in her mid - to late - fifties, and drops off sharply over the next few years. This until t hey reach the time when pension value spikes. Once they pass the pension value spike, Friedberg and Turner (2011) use a peak value model to estimate the effect of variation in pension parameters on teacher retirement across 17 pension plans. Using teacher - level data from the School and Staffing Survey (SASS), they find that teachers tend to delay retiring while pension wealth accumulates, then retire abruptly after reaching peak pension wealth. Two other pap ers explore policy changes affecting pension wealth and employ difference - in - difference models to estimate the effect of pension incentives on the timing of teacher exit decisions. Furgeson, Strauss, and Vogt (2006) find a large increase in teacher retirem ent when Pennsylvania enacted a pension policy which temporarily increased early retirement benefits between 1997 - 1998 and 1998 - 1999. Koedel and Xiang (2017) investigate a policy change in St. Louis that increased teacher retirement benefits for those who retired after the 1998 - 1999 school year. Using administrative panel data from the Missouri Department of Elementary and Secondary 62 Education, they examine retention effects throughout the workforce using younger teachers as their control group. The authors find a temporary delay in retirement among teachers who were eligible for retirement in the previous school year. However, they find no significant retention effect among teachers who were not eligible for retirement. They offer a possible explanation that teachers may lack of full knowledge of their pensions. Other papers compare DB plans and alternative plans with no (or less) backloading. For example, Gustman and Steinmeier (1993) use the Survey of Income and Program Participation (SIPP) to compare wor ker turnover (not restricted to teachers) between DB plans and DC plans. DC plans are not backloaded: they do not guarantee retirees minimum or maximum pension benefits and allow vested workers to take their full retirement saving with them when moving fro m one job to another. Gustman and Steinmeier (1993) find that pension coverage was associated with lower one - year turnover rate regardless of pension types (DB or DC). Friedberg and Owyang (2005) study the link between DB coverage and job tenure. Using dat a from 1983 - 2001 Survey of Consumer Finance (SCF) and 1993 Current Population Survey (CPS) and controlling for job characteristics, they find that workers (not restricted to teachers) with DB pensions work in the same job longer than workers with no pensio ns or with DC pensions. 55 They also find that workers with more generous DB pensions stay longer, controlling for levels of earning. Goldhaber et al. (2017) investigate whether moving away from a backloaded defined benefit plan increases levels of employee turnover. Using data from Washington Teacher Retirement system, the authors compare the turnover rate of teachers who enrolled in either a traditional defined benefit (DB) plan or a hybrid plan (combines DB and DC) during the same period of time. They find that turnover was lower among teachers who transferred out of the DB plan into the hybrid plan. 55 63 Papke and Litwok (2013) are the first to study the effect of pension incentives on younger - state pensio n differences in four states contribution alternatives is positively and significantly related to the hazard rate of first exit. They also find that more stringent ve sting requirements are positively and significantly related to the hazard rate of first exit. 3.2.2 Teacher Outside Options and Teacher Turnover A large literature studies how teacher pay and outside options influence the decision to leave. For example, Rees (1991) finds a negative relationship between teacher salary and teacher turnover. He also uses education level as a measure of outside options and finds that teachers with higher educational degrees are more likely to leave. 56 Gritz and Theobald (1996) find that higher salary increases the probability of a teacher staying in the teaching profession. However, these salary effects diminish with tea ching experience. Stockard and Lehman (2004) find that higher salary increases the probability of a teacher staying in the same school next year. My research advances the existing literature in several ways. Along with Koedel and Xiang (2017), I am among t he first to study how benefit - reducing teacher pension reforms affect retention. My paper focuses on several states and therefore provides more general results than previous research identifying the effect of a single state policy change. Moreover, because pension reforms changed over time, I also account for time - varying outside options covariates that previous researchers show to affect teacher turnover. 56 64 3.3 Institutional Details and Pension Data In the U.S., all full - time public school teachers automatically participate in public pension plans. While each state runs a pension system that covers teachers (state - run pension plans), some large cities like Ne w York and Chicago operate their own municipality - specific pension plans . 57 T able 3.1 shows the types of pension plans that states and cities offer teachers during the years I consider. In this paper, I restrict my analysis to the 36 state - run pension plans and 11 municipality - specific pension plans that automatically enroll teachers into a mandatory, traditional defined benefit plan throughout the years 2003 to 2011. 58 In defined benefit plans, teachers contribute a portion of salary towards the plan while e mployed. Teachers who retire after becoming vested collect annual payments until their death. 59 The lifetime pension wealth of a teacher depends on the size of initial annual payments, adjustments made for cost of living, and the length of time in retiremen t. All defined benefit plans use a similar formula to calculate annual payments: (5) First - year Annual payments (A) = FAS * Multiplier * Years of Service where FAS ch year of service. The years 57 58 59 65 more years of service lead to higher annual payments. But some states cap the first - year annual payment to not exceed a specif - year annual payments is: (6) In order to counteract t he impact of inflation on retirement income, public pension plans provide retirees with post - retirement cost of living adjustments (COLAs). Plans with higher cost of living adjustments are more beneficial to teachers. The pension parameters discussed above relate to the size of annual benefits. Other pension parameters such as retirement age and teacher contribution amount may also affect pension wealth, without directly changing annual benefits. 60 Because teachers collect annual benefits from retirement until death, an earlier retirement age causes teachers to receive more annual payments and enjoy higher lifetime pension wealth. Recall that most states require teachers to contribute a portion of salary to fund their pension plans. Because a t teacher contributions (see equation (1)), increasing contribution rate decreases net pension wealth. To more precisely evaluate pension plan generosity, I calculate net pension wealth as the actuarial present value o f the payment stream that teachers expect to receive upon retirement, net of employee contributions. 61 S uppose a teacher started teaching in year t=1 , retires in year t=T , and collect pension benefits for d years, from t=T+1 to t=T+ . Then, her expected present discounted 60 61 66 net pension wealth (PDNW) is calculated using the following formula: (3) = - where is the capped annual payment in y ear t >T, is the cost of living adjustment in year t, is the probability of surviving at year t, r is the discount rate, and is the teacher contribution in year t T . and summaries of legislation. I obtained additional pension plan - level data on pension funding ratio and employer contribution rate from the Public Plans Database (PPD). 62 Table 3.3 summarizes changes in pension parameters in each of 36 state - run pension plans and 11 municipality - specific pension plans between 2007 and 2011. 63 While legal constraints prevent most states from cutting benefits for current employees, all states can change the benefit rules affecting employees not yet hired. Therefore, all the pension parameters changes discussed below affect teachers hir ed in the 2011 - 2012 school year. 64 Between 2007 and 2011, 23 pension plans enacted legislation that reduced benefits. Figure 3.2 illustrate s net pension wealth accrual over the career cycle for a representative teacher, who started teaching at age 25, in each state under new rules (post - reform ) and old rules (pre - reform) . 65 66 62 63 64 65 66 Note that m ost state review and adj ust contribution rates annually and COLAs are not viewed by courts as c ore benefits protected under state laws (Munnell et al. 2016 ). Thus, CO LAs and contribution rates can fluctuate over time. For simplicity, I calculate 67 lth profiles are single peaked. The exception is Illinois, which has a cap that binds the peak pension wealth for teachers representative teacher. 67 Compared to teachers h ired pre - reforms , those hired post - reforms have flatter pension wealth accruals and lower pension wealth at peak. Lower expected pension wealth may increase teacher turnover because it lowers the opportunity cost of separating from the pension system. The right column of Table 3.3 shows that seven pension plans increased the time to vesting between 2007 and 2011. The exception is Kansas, which moved from a 10 - year vesting rule to a 5 - years vesting rule. Figure 3 .3 illustrates how changing vesting rules affect a representative wealth lowers pension wealth accruals and decreases the incentive to stay. However, if teacher s only consider short - term pension wealth accruals, changing vesting rules will have little influence on teacher exit decisions. For example, Illinois moved from a 5 - year vesting rule to a 10 - year n wealth accruals remain zero. If Illinois teachers only consider 4 - year or less pension wealth accruals, increasing vesting rule will not affect the incentive to stay. 3.4 Teacher Data Data on teachers come from the Schools and Staffing Survey (SASS), th e largest and most extensive survey of elementary and secondary teachers in the U.S. The Census Bureau conducted 67 68 the survey seven times between 1987 - 1988 and 2011 - 2012. 68 SASS covers a wide range of topics including teacher characteristics, teacher compensa tion, and general conditions in schools. To determine whether teachers still taught at the same school, had moved to a different school, or had left the teaching profession since the SASS administration, NCES conducted the Teacher Follow - up Survey (TFS) fo r a sample of teachers who had completed SASS during the previous school year . 69 The most recent available TFS data are from the 2012 - 2013 wave, and the earliest available SASS data that contain consistent questions are from the 2003 - 2004 wave. 70 For these reasons, I use the 2003 - 2004, 2007 - 2008 and 2011 - 2012 waves of the SASS, and the 2004 - 2005, 2008 - 2009 and 2012 - 2013 waves of the TFS. To remain consistent with the pension parameters teachers faced when they started teaching, I index school years by the ca lendar year when a school year starts. For example, I refer to the 2011 - 2012 school year as 2011. year of entry into teaching. This lets me target a sample of ne w entrants affected by recent pension gives me a measure of one - year mobility to study teacher turnover . Third, the restricted - use state and zip - code identifiers allow me to match individual teachers to their pension parameters and her alternative opportunity within - state relative to teaching salary, and to impute an earnings interest in this paper is at the state by year level. With the proper weighting, each wave of SASS is designed to be representative at the stat e level. One of the weaknesses of using SASS is that I 68 69 70 69 only observe one year of follow up for each teacher. Therefore, if a teacher left then returned, I still considered this as an exit. I restrict my sample to newly hired teachers who work full time. 71 I define newly hired teachers as those who have no teaching experience before the survey year. 72 Because pension mandates do not always apply to charter schools, I exclude teachers employed at charter schools, eliminating 130 teacher - year observations. To se parate exit decisions from involuntarily exit behavior, I exclude teachers whose contracts were not renewed or were laid off, eliminating 130 teacher - year observations. 73 The final teacher - level dataset I use for the analysis contains 1,380 teacher - year obs ervations in 2003, 2007, and 2011 in the 36 states of interest, representing 257,070 teachers in the population. I measure teacher turnover using an indicator for whether the teacher leaves her pension system after the first year of employment ( Leave ). I code Leave as a dummy that equals one if the teacher met any of the following criteria in the next school year: 1) the teacher left the teaching profession ; 2) the teacher was teaching in a different state. I use one - year turnover measure because scho is highest in the first year. In addition, financial incentives may have larger retention effects on new teachers ( Gritz and Theobald 1996). I use the following procedure to match each individual teacher to her pension plan and benefit group. First, I assume that a municipality - specific pension covers the school if its U.S. Postal Service zip code matches the zip code of a big city that operates its own pension p lan. I then assume state - run pension plans cover all remaining schools. Table 3.4 explains how each state and city 71 72 73 70 74 I then match each individual teacher to her pension plan based on where she teaches and when she started t eaching. To control for outside options when teachers decide whether to stay, I calculate Alternative teacher pay administration alternative salary, based on an earning from the March Current Population Survey (CPS). I normalize all salaries in this paper to 2011 dollars using the CPI fo r all items less food and energy, reported by the Bureau of Labor Statistics . I also include state unemployment rates from the Bureau of Labor Statistics as a measure of the outside labor market options. Table 3.5 reports weighted descriptive statistics fo r subgroups of teachers in the treatment group (states and cities that reduced benefits) and the control group (states and cities that did not reduce benefits). The top panel describes teacher mobility. Between the pre - treatment (2003 and 2007) and post - tr eatment (2011) periods, both groups experienced a higher share of teachers who left in the following year. The next panel characterizes state and city level pension variables. The mean of Net pension wealth shows that among teachers hired pre - treatment, t reatment group teachers have more generous pension plans treatment group teachers receive on average 35 thousand dollars more net pension wealth compared to those in the control group. However, after the benefit cuts, treatment group teachers receive on average 64 thousand dollars less net pension wealth compared to control group teachers. 74 71 For teachers hired after the reforms, it takes on average 1.79 more years for treatment group teachers to vest, while it takes on average 0.60 fewer year for control g roup teachers. Compared to pre - treatment periods, treatment group teachers contributed 1.27 percentage points more of teacher salary to fund pension post - treatment, and control group teachers contributed 0.36 percentage point less of teacher salary. Higher Teacher Contribution Rate may increase teacher turnover because it decrea Employer Contribution Rates could also indirectly affect teacher turnover if higher employer contributions reallocate school resources away from current school expenditures that attract teachers. Compared to pre - treatment perio ds, school districts in both groups contributed a higher potion of teacher salary to fund pension plans, especially in the treatment group (from 7.55 percent to 12.67 percent). The middle panel reports means of matched outside options variables. Compared t o pre - treatment periods, the treatment and control groups both experienced higher unemployment rates. The alternative teacher pay in the treatment group increased from 136 percent to 141 percent, compared to a decrease from 138 percent to 132 percent in th e control group. Given that a higher alternative teacher pay may make teaching less attractive while fewer outside options make stable teaching jobs more attractive, I control for both of the time - varying variables to more precisely estimate the effect of recent pension reforms on teacher turnover. 3.5 Methodology and Estimates 3.5.1 Pension Reforms and Teacher Turnover I compare the difference in one - year teacher turnover from the pre - treatment to post - treatment periods among states and cities that red uced benefits relative to the difference in states with no benefit - reducing legislation. I also estimate the effect of vesting rule changes on turnover and 72 separately control for the pension parameters that do not directly affect the calculation of pension benefits. I estimate the following difference - in - differences model: = , where is a dummy that equals one if the teacher left the teaching profession or was teaching in a different state after the first year of employment, i indexes the teacher, j and s index the school districts and state where the teacher worked, and t indexes the year the teacher started teaching (2003, 2007, or 2011). is an indicator that equals one if teacher i taught in a state or city that enacted benefit - reducing reforms from 2007 to 2011. is an indicator for year 2011. The coefficient on captures the effect of benefit - reducing reforms on teacher turnover. is an indicator that equals one if teacher i taught in a state or city that increased vesting rules from 2007 to 2011. The coefficient on captures the effect of more stringent vesting requirements on teacher turnover. The vector include vector includes teacher contribution rate, employer contribution rate, and pension funding ratio. All specifications include a full set of state dummies and state dummies ( . To account for the stratified sampling frame used by SASS, I weigh each teacher by the - level TFS data. This weight contains the 73 ing the TFS survey year, as well as adjustments for nonresponse rate and other sampling considerations that arise after the sample has been drawn. Table 3.6 reports the difference - in - difference coefficients controlling for potential turnover determinants that changed simultaneously with the benefit - reducing reforms . The estimate of suggests that among teachers hired pre - treatment, treatment group teachers are 34.6 percentage points less likely to leave her pension system after one year of employment, w ith a standard error of 0.14. Recall from Table 3.5 that states that later enacted pension reforms between 2007 and 2011 originally offered teachers more generous pension benefits compared to states that did not cut pension benefits. Thus, this finding sug gests that higher expected compensation levels are negatively correlated with teacher turnover. The estimate of suggests that under pension reforms, new teachers are 2.6 percentage points less likely to leave in the following year, with a standard error of 0.10. The point estimate is statistically insignificant. Thus, I have no sufficient evidence that pension bene fits cuts affect the probability of a new teacher exiting. This finding is inconsistent with the hypothesis that lower expected compensation levels decrease the cost of moving jobs and could increase teacher turnover. On the other hand, the estimate of suggests that among teachers hired pre - treatment, those hired in states that later increased vesting requirements are 42.8 percentage points more likely to leave her pension system after one year of employment, with a standard error of 0.18. Because stat es that later increased vesting requirements originally allowed teachers to vest sooner, the estimate of suggests that shorter vesting requirements are positively correlated with teacher turnover. The estimate of suggests that increasing the years until a teacher is fully vested is associated with a 18.8 percentage points increase in the probability of a teacher leaving her pension system 74 in the following year, with a standard error of 0.18. Although the point estimate is large (about one half a st andard deviation), it is statistically insignificant. Again, I cannot reject the null hypothesis that equal to zero. Thus, I have no sufficient evidence that more stringent vesting requirements change the probability of a new teacher exiting. The estimate of the coefficient on teacher contribution rate is also statistically insignificant, suggesting increasing the teacher contribution rate does not change the probability of a new teacher exiting. This finding is inconsistent with the hypothesis th at lower take home pay decreases the incentive to stay. The estimates in Table 3.6 also suggests that a one percentage point increase in the employer contribution rate is associated with a 0.005 percentage point increase in the probability of a teacher lea ving in the following year, with a standard error of 0.003. This finding may suggest that higher employer contribution rates relocate school resources from expenditures that help retain teachers. It may also suggest that higher pension costs discourage sch ool districts from retaining teachers. A one percentage point increase in the pension funding ratio is associated with a 0.002 percentage point increase in the probability of a teacher leaving in the following year, with a standard error of 0.002. The point estimate is small and statistically insignificant. When examining the effect of outside options on teacher turnover, I find that alternative salary matters to teacher exit decisions. The point estimate 0.0007 (0.0004) is statistically significant , suggesting that a one standard deviation increase in alternative salary is associated with a 0.02 percentage point increase in the probability of a teacher leaving in the following year. The estimate of the coefficient unemployment rate is negative, suggesting that fewer job opportunities outside of teaching decreases teacher turnover, although the point estimate - 0.03 (standard error=0.03) is statistically insignificant. 75 To identify the estimates, I assume that trends in new teacher turnover in states with no benefit - reducing legislation provide an accurate counterfactual for the trend in states with pension wealth cuts. To supp ort the identification assumption, I compare the difference in one - year teacher turnover among the treatment and control groups in the pre - treatment periods. Table 3.7 shows the difference - in - difference coefficients equation (4) using 2007 and 2011 as post - treatment periods and 2003 as pre - treatment period. I cannot reject the null hypothesis that and are equal to zero. I find no evidence that new teacher turnover in the treatment group was different from the control group before the pension reforms . 3.5.2 Pension Wealth and Teacher Turnover As discussed in Section 3, pension plans reformed in numerous ways. To more precisely age, gender, and starting salary. See appendix A for calculat ion details. These heterogeneous effects of the pension the relationship between the natural followi ng model: = , where is the Similar to equation (4), all specifications include a full set of state dummies and year dummies ( , and control for time - varying state - level variables school and pension characteristics , and individual characteristics . 76 The coefficients of interest in equation (5) are and , which are the estimates of a one percent increase in net pension wealth and a one additional year in vesting requirements on the probability of a teacher leaving after the first year of employment . Table 3.8 reports the estimates from equation (5). The estimate of suggests that a one percent increase in net pension wealth is associated with a 0.03 percentage point increase i n the probability of a teacher leaving in the following year, with a standard error of 0.03. Because the estimate is statistically insignificant, I cannot reject the null hypothesis that equal to zero. Thus, I have no sufficient evidence that a one percent increase in net pension wealth affect teacher turnover. The estimate of is positive, suggesting that more stringent vesting requirement increases teacher turnover, although the poin t estimate 0.0007 (002) is small and statistically insignificant. 3.6 Conclusion High teacher turnover presents an ongoing concern for many states and school districts. Turnover forces schools to spend more on recruitment, hiring, and training. Turnover i s highest among new teachers and those close to retirement. Because retirement is not avoidable, schools may want to focus more on retaining new teachers ( Raue and Gray 2015) . The purpose of this paper is to study whether, and to what degree, pension incen tives affect new teacher exit decisions. I use recent pension policy changes to identify the effect of pension incentives on teacher exit decisions. Based on my results, I find no sufficient evidence that more stringent vesting requirements affect the pro bability of a new teacher exiting. Recall from Table 3.5 that it takes pre - treatment treatment group teachers on average five years to be vested. Perhaps teachers only consider four - year or less pension wealth accruals. Thus, increasing vesting requirement s from 77 whether the back - loading structure of DB pension plan helps retain new teachers, I find no evidence that pension benefits cuts affect the probability of a new teacher exiting. Perhaps new teachers do not value their pension benefits or lack full knowledge of their pensions. Alternatively, the composition of the new teacher pool may have changed during recent pension reforms, with less mobile teachers ente ring the profession. For example, Chou (2018) finds that teachers entering the profession under pension reforms were from undergraduate institutions with lower average SAT scores. Given that teachers with lower test scores are less likely to leave (Podgurs ky et al. 2004), recent pension reforms may have a positive effect on teacher recruitment but was offset by the effect on teacher recruitment. I find no evidence that increasing the teacher contribution rate changes the probability of a new teacher exiting . Given the small variation in teacher contribution rates between 2007 and 2011, it is possible that teachers are not aware of the changes. My estimates suggest that the employer contribution rate affect teacher turnover . I find that a one percentage point increase in the employer contribution rate is associate with a 1.38 percentage points increase in the probability of a teacher leaving in the following year. While a change in the s pension wealth, it may reallocate school resources away from current school expenditures that attract teachers. Overall, my findings suggest that using increasing employer contributions to reduce pension debt might impair school districts attempts to ret ain new teachers. As a final note on my findings, the issue of using a one - year turnover measure is that some first - year teachers may exit teaching after finding teaching a bad match for them. Teachers may also decide to leave because of non - financial inc entives that are difficult to parameterize, such as 78 lack of support from school principle and student discipline problems. If these omitted variables changed between 2007 and 2011, then my findings will be biased. 79 APPENDICES 80 APPENDIX A PENSION WEALTH CALCULATION I assume all teachers work in the same school until they reach their eligibility for normal retirement and start collecting retirement benefits right away. 75 Since my sample only includes first - year teachers, their teaching from retirement using their age. SASS only reports current earnings but not earning histories of teachers, so I impute an earning history for each individual teacher and calculate her final average salary. With FAS calculated above, I impute teacher pension wealth using equations (3) (5) mentioned in section 2.3. I use the Actuarial Life Table provided by the Social Security their life cycle. 76 This paper assumes a 4 percent discount rate that allows a positive real interest rate and a time preference in earning. I assume that Consumer Price index increases 3 percent annually for pension plans that tied COLAs to changes in Consumer Price Index a nd assume a 3 percent ad hoc for teacher pension that tied COLAs to state legislation. Note that courts do not view COLAs as core benefits protected under the state laws ( Munnell et al. 2016 ) . At the same time, most state review and adjust contribution rates annually. Therefore, teachers may expect that COLAs and contribution rates would fluctuate over time. For simplicity, 75 76 81 ate and COLAs released in the year she was hired. 82 APPENDIX B TABLES Table 1.1: Summary of Selected Pension Literature on Pension Wealth 83 84 85 Table 1.1 86 Table 1.2: Types of Pension Plans States offered Teachers 87 88 Table 1.3: State Teacher Pension Homepage, Handbooks, and Other Resources 89 90 91 Table 1.3 92 93 94 Table 1.4: Comparison of the Number of Years Used to Calculate Final Average Salary 95 96 Table 1.5: Comparison of the Formula Multiplier 97 98 Table 1.6: The Limitation on First Year Annual Benefits as a Percentage of FAS 99 Table 1.7: Comparison of the Cost of Living Adjustments 100 101 Table 1.8: Comparison of Retirement Eligibility 102 103 Table 1.9: Comparison of the Employee Contribution Rates 104 105 Table 1.10: Summary of Reforms Between 2007 and 2016 that Reduced Teacher Benefits 106 107 Table 1.11: Comparison of Starting Salary, Years of Service, and Final Average Salary 108 109 Table 1.12: Simulation of First - year Pension Benefits Across States 110 111 Table 1.13: Comparison of Simulated Pension Wealth 112 113 Table 1.14: Comparison of Simulated Lifetime Contributions 114 115 Table 1.15: Comparison of Net Pension Wealth 116 117 Table 1.16: Comparison of Net Pension Wealth Across States and Overtime 118 119 Table 1.17: Summary of Pension Reforms and Salary Changes Between 2007 and 2016 120 Table 1.18: Comparison of t he Rankings of Net Pension Wealth, Measured by the Percentage of F inal Average Salary 121 122 Table 1.19: Simulation of Net Pension Wealth for Teachers Hired in 2007 by Years to Live and State 123 124 Table 1.20: Simulation of Net Pension Wealth for Teachers Hired in 2007 by Discount Rate and State 125 126 Table 1.21: Simulation of Net Pension Wealth for Teachers Hired in 2007 by COLAs and State 127 128 Table 2.1: Types of Pension Plans States and Cities Offered Teachers 129 130 Table 2.2: State Teacher Pension Homepage, Handbooks, and Other Resources 131 132 133 134 135 136 Table 2.3: Comparison of the Number of Years Used to Calculate Final Average Salar y 137 138 Table 2.4: Comparison of the Formula Multiplier 139 140 Table 2.5: The Limitation on First Year Annual Benefits as a Percentage of FAS 141 142 Table 2.6: Comparison of the Cost of Living Adjustments 143 144 145 Table 2.7: Comparison of Retirement Eligibility 146 147 Table 2.8: Comparison of the Employee Contribution Rates 148 149 Table 2.9: Comparison of Vesting Requirements 150 151 Table 2.10: Refund if Teachers Leave Before Vested 152 153 Table 2.11: Membership of Public Pension Plans that Enroll Teachers 154 155 156 157 158 159 Table 2.12: Weighted Means of Variables 160 Table 2.13: Weighted Means of Variables 161 Table 2.14: Reforms that Reduced Teacher Benefits 162 Table 2.15: Estimates of the Effects of Pension Reform on New Teacher Quality 163 Table 2.16: The Difference in New Teacher Quality between the Control and Treatment Groups before Recent Pension Reforms 164 165 Table 3.1: Types of Pension Plans States and Cities Offered Teachers 166 167 Table 3.2: State Teacher Pension Homepage, Handbooks, and Other Resources 168 169 Table 3.2 170 171 172 173 Table 3.3: Pension Reforms Between 2007 and 2011 174 175 Table 3.4: Membership of Public Pension Plans that Enroll Teachers 176 177 178 179 180 Table 3.5: Weighted Means of Variables 181 Table 3.6: The Effects of Pension Reforms on New Teacher Turnover 182 Table 3.7: The Difference in New Teacher Turnover between the Control and Treatment Groups before the Pension Reforms 183 Table 3.8: The Effects of Net Pension Wealth on New Teacher Turnover 184 APPENDIX C FIGURES Figure 1.1: Net Pension Wealth Change between 2007 and 2016 185 Figure 1.2: Pension Funding Ratio Trends from 2007 through 2016 186 Figure 1.3: Relationships between Net Pension Wealth Cut and Pension Funding Ratio 187 188 189 190 191 192 193 194 195 BIBLIOGRAPHY 196 BIBLIOGRAPHY ACT Report. 2015 . http://www.act.org/content/dam/act/unsecured/documents/Future - Educators - 2015.pdf Aldeman, Chad and Kelly Robson. 20 Education Next, May 16. http://educationnext.org/why - most - teachers - get - bad - deal - pensions - state - p lans - winners - losers/ Allen, Steven G., Robert L. Clark, and Ann A. McDermed. 1988. "The pension cost of changing jobs." Research on Aging 10.4: 459 - 471. Angrist, Joshua D., and Jonathan Guryan. 2004. "Teacher testing, teacher education, and teacher characteristics." The American Economic Review 94.2: 241 - 246. Angrist, Joshua D., and Jonathan Guryan. 2008. "Does teacher testing raise teacher quality? Evidence from state certification requirements." Economics of Education Review 27.5: 483 - 503. Bacolod, Marigee P. 2007. "Do alternative opportunities matter? The role of female labor markets in the decline of teacher quality." The Review of Economics and Statistics 89.4: 737 - 751. Bempah, E. Osei, Michael S. Kaylen, Donald D. Osburn, and Robert J . Birkenholz. 1994. "An econometric analysis of teacher mobility." Economics of education review 13.1: 69 - 77. Berger, Mark C., and Eugenia F. Toma. 491. Boivie, I. 2017. Revisiting the Three Rs of Teacher Retirement Systems: Recruitment, Retention, and Retirement. Boyd, D., Lankford, H., Loeb, S., and Wyckoff, J. 2005. The draw of home: How teachers' preferences for proximity disadvantage urban schools. Journal of Policy Analysis and Managem ent: The Journal of the Association for Public Policy Analysis and Management , 24 (1), 113 - 132. Brewer, Dominic J . 1996: "Career Paths and Quit Decisions: Evidence from Teaching." Journal of Labor Economics 14.2:313. ProQuest. Web. 28 Aug. 2018. Brown, Kristine M. Journal of Public Economics 98: 1 14. 197 Brown, Kristine M., and Ron A. Laschever. - Four: Peer Effects and American Economic Journal: Applied Economics 4 (3): 90 115. Bureau of Labor Statistics. 2003 - Department of Labor. Bureau of Labor Statistics. 2003 - 2011. - https://data.bls.gov/pdq/SurveyOutpu tServlet Center for Education Data & Research Working Paper. WP# 2016 - 5. Chingos Education Finance and Policy 10(2), 193 - 222. Chou, Pin - rms on New Teacher Review of Economics and Statistics (89:2): 234 - 246. Costrell Education Finance and Policy 5(4), 492 - 518. Costrell, Robert M., and Michael Podgursky. 2009. "Peaks, cliffs, and valleys: The peculiar incentives in teacher retirement systems and their consequences for school staffing." Education Finance and Policy 4.2: 175 - 211. Costrell, Robert M., and Michael Podgursky. 2010. "Distribution of benefits in teacher retirement systems and their i mplications for mobility." Education Finance and Policy 5.4: 519 - 557. Dal Bó, Ernesto, Frederico Finan, and Martín A. Rossi. 2013. "Strengthening state capabilities: The role of financial incentives in the call to public service." The Quarterly Journal of Economics 128.3: 1169 - 1218. Doherty, Kathryn M., Sandi Jacobs, and Martin F. Lueken. 2015. "Doing the Math on Teacher Pensions: How to Protect Teachers and Taxpayers." National Council on Teacher Quality. Doherty, Kathryn M., Sandi Jacobs, and Trisha M. Madden. 2012. "No One Benefits: How teacher pension systems are failing BOTH teachers and taxpayers." Washington, DC: National Council on Teacher Quality . 198 Dorans, Neil J. 1999. "Correspondences betwee " ETS Research Report Series 1999.1 . https://www.ets.org/Media/Research/pdf/RR - 99 - 02 - Dorans.pdf Eberts, R. W. 1987. Union - negotiated employment rules and teacher quits. Economics of Education Review , 6 (1), 15 - 25. Egalite, Anna J., Brian Kisida, and Marcus A. Winters. 2015. "Representation in the classroom: The effect of own - race teachers on student achievement." Economics of Education Review 45: 44 - 52. Feng, Li, and Tim R. Sass. The Impact of Incentives to Recruit and Retain Teachers in - to - Data in Education Research, 2015. Industrial and Labor Relations Review 55: 686 - 699. American Economic Journal: Economic Policy , 6, 120 - 154. Freeman, R. B. 1980. The exit - voice tradeoff in the labor market: Unionism, job tenure, quits, and separations. The Quarterly Journal of Economics , 94 (4), 643 - 673. Frölich, Markus. 2004. "Programme evaluation with multiple treatments." Journal of Economic Surveys 18.2: 181 - 224. Friedberg, Leora. 2011. "Labor market aspects of state and local retirement plans: a review of evidence and a blueprint for future research." Journal of Pension Economics & Finance 10.2: 337 - 361. Friedberg, Leora and Owyan 022D. for Teachers Education Finance and Policy 5(4) (Fall), 463 - 491. Friedberg, Leora, and Sarah Turner. 2011. "Pensions and public school teacher retirement: An analysis using national teacher data." Research Dialogue Issue 99. Furgeson Education Finance and Policy 1(3): 316 - 48. 199 Goldhaber - 8. Seattle, WA: University of Washington Bothell. Goldhaber, Dan, Cyrus Grout, and Kristian Holde n. 2015. "Teacher Pension Systems and Mobility Decisions: Evidence from Washington State." Washington, DC: American Institutes for Research . Goldhaber, Dan, Cyrus Grout, and Kristian L. Holden. 2017. "Pension structure and employee turnover: Evidence from a large public pension system." ILR Review 70.4: 976 - 1007. Gritz, R. M. and Theobald, N. D. 1996. The effects of school district spending priorities on length of stay in teaching. Journal of human resources , 477 - 512. Guin, K. (2004). Chronic teacher turnover in urban elementary schools. Educational Evaluation and Policy Analysis , Vol. 12, No. 42, 1 - 25. Gustman, Alan L., and Thomas L. Steinmeier. 1993. "Pension portability and labor mobility: Evidence from the survey of income and program participation." Journal of Public Economics 50, no. 3: 299 - 323. Hanushek, Erik A., and Finis Welch. Handbook of the Economics of Education. Volume 3 ISSN 0169 - 7218, DOI: 10.1016/S0169 - 7218(11)03005 - X Hanushek, Eric A., and Richard R. Pace.1995. "Who chooses to teach (and why)?" Economics of education review 14.2: 101 - 117. Hanushek, E.A., Kain, J.F. and Rivkin, S.G. 1999. Do higher salaries buy better teachers? Working paper No. 7082. Cambridge, MA: National Bureau of Economic Research. Hanushek, Eric A., and Steven G. Rivkin. 2010. "Generalizations about using value - added measures of teacher quality." The American Economic Review 100.2: 267 - 271. Hoxby, Caroline, M., and Andrew Leigh. 2004. "Pulled Away or Pushed Out? Explaining the Decline of Teacher Aptitude in the United States." American Economic Review , 94(2): 236 - 240 . Kan, Leslie, and Chad Aldeman. 2014."Uncovered: Social Security, Retirement Uncertainty, and 1 Million Teachers." Bellwether Education Partners. Kim, Dongwoo. 2016. "Worker Retirement Responses to Pension Incentives: Do They Respond to Pension Wealth?" http://www.economics.illinois.edu/people/candidates/documents/Kim_PensionKnowledge.p df 200 Kim, Dongwoo, Cory Koedel, Shawn Ni, Michael Podgursky, and Weiwei Wu. 2017. Pensions and Late - Career Teacher Retention . No. 2016 - 08. https://economics.missouri.edu/sites/default/files/wp - files/late_career_retention_jpam_revised_7_11_17.pdf Kimball, Steven M., Herbert G. Heneman, and Eileen M. Kellor. 2005. "Can pensions help attract teachers?" Journal of Education Finance 30.4: 399 - 411. All Changes Are Equal: Assessing How Dif ferent Pension Policies Can Impact Teacher https://www.rand.org/pubs/research_briefs/RB9898.html . Koedel , Cory, Shawn Ni, and Michael Podgursky. 2014. "Who benefits from pension enhancements?" Education Finance and Policy 9, no. 2 (2014): 165 - 192. Koedel, Cory, Michael Podgursky, and Shishan Shi. 2013. "Teacher pension systems, the composition of the teachi ng workforce, and teacher quality." Journal of Policy Analysis and Management 32.3: 574 - 596. Koedel, Cory, and P. Brett Xiang. 2017. "Pension Enhancements and the Retention of Public Employees." ILR Review 70.2: 519 - 551. Lankford, H., Loeb, S., & Wyckoff, J. 2002. Teacher sorting and the plight of urban schools: A descriptive analysis. Educational evaluation and policy analysis , 24 (1), 37 - 62. McCormick, Barry, and Gordon Hughes. 1984. "The influence of pensions on job mobility." Journal of Public Economics 23.1 - 2: 183 - 206. Munnell, Alicia H., Jean - Pierre Aubry, and Mark Cafarelli. 2016. "COLA cuts in state - local pensions." Journal of Pension Economics & Finance 15, no. 3: 311 - 332. Munnell , Alicia H., and Rebecca Cannon Fraenkel. 2013. "Compensation Matters: The Case of Teachers." State and Local Pension Plans Issue in Brief 28. Nagler, Markus, Marc Piopiunik, and Martin R. West. 2015. Weak markets, strong teachers: Recession at career start and teacher effectiveness . No. w21393. National Bureau of Economic Research. National Center for Education Statistics . 2003 - 2005, 2007 - 2009, 2011 - 2013 Schools and Staffing Survey Restricted - Institute of Education Science. National Education Association. 2004, 2008, 2010, 2016. Characteristics of Large Public Education Pension Plans. Washington, DC . 201 Ni, Shawn, and Michael Podgursky. 2010. "Teacher Retirement Behavior: Structural Estimates and Policy Simulations." Manuscript, University of Missouri, Columbia . Ni, Shawn, and Michael Podgursky. 2016. "How teachers respond to pension system incentives: New estimates and policy applications." Journal of Labor Economics 34.4 : 1075 - 110 4. Partelow World Report, October 14. https://www.usnews.com/opin ion/knowledge - bank/2015/10/14/address - americas - teacher - shortage - the - right - way 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. Podgursky, Michael, and Matthew Springer. 2011. "Teacher compensation systems in the United States K - 12 public school system." National Tax Journal 64.1: 165. Raue, K., and Gray, L. 2015. Career Paths of Beginning Public School Teachers: Results from the First through Fifth Waves of the 2007 - 08 Beginning Teacher Longitudinal Study. Stats in Brief. NCES 2015 - 196. National Center for Education Statistics . Rees, D. I. 1991. Grievance procedure strength and teacher quits. ILR Review , 45 (1), 31 - 43. Rockoff, Jonah E., Brian A. Jacob, Thomas J. Kane, and Douglas O. Staiger. 2011."Can you recognize an effective teacher when you recruit one?." Education finance and Policy 6, no. 1:43 - 74. Ronfeldt , Matthew, Susanna Loeb, and James Wyckoff. 2013 "How teacher turnover harms student achievement." American Educational Research Journal 50.1: 4 - 36. Rothstein, Jesse. 2011. "Unemployment Insurance and Job Search in the Great Recession." Brookings Papers o n Economic Activity 2011.2 : 143 - 213. Web. Rumberger, Russell W. 1987. "The impact of salary differentials on teacher shortages and turnover: The case of mathematics and science teachers." Economics of education review 6, no. 4 (1987): 389 - 399. Madison, WI: Wisconsin Legislative Council. Report retrieved 8/17/17 from https://docs.legis.wisconsin.gov/misc/lc/comparative_retirement_study/2015_retirement.pdf . 202 Madison, WI: Wisconsin Legislative Council. Report retrieved 9/16/18 from http://legis.wisconsin.gov/lc/publications/crs/2008_retirement.pdf . Shen, J. 1997. Teacher retention and attrition in public schools: Evidence from SASS91. The Journal of E ducational Research , 91 (2), 81 - 88. Steven Ruggles, Katie Genadek, Ronald Goeken, Josiah Grover, and Matthew Sobek . 2017. Integrated Public Use Microdata Series: Version 7.0 [dataset]. Minneapolis: University of Minnesota. https://doi.org/10.18128/D010.V7.0 . Stock, James and David Wise . 1990. Pensions, the Option Value of Work, and Retirement , Econometrica , 58 , (5), 1151 - 80 Stock, James, and Wise, David.1990b. The pension inducement to retire: an option value analysis. In Wise, D. (ed.), Issues in the Economics of Aging. Chicago, IL: University of Chicago Press, pp. 463 468. Stockard, Jean, and Michael Bryan Lehman. 2004, "Influences on the satisfaction and retention of 1st - year teachers: The importance of effective school management." Educational Administration Quarterly 40, no. 5:742 - 771. The United State Social Security Administration (SSA). https://www.ssa.gov Report retrieved 8/5/18 from: https://www.ssa.gov/section218training/basic_course_4.htm Theobald, N. D. (1990). An examination of the influence of personal, professional, and scho ol district characteristics on public school teacher retention. Economics of education review , 9 (3), 241 - 250. Toutkoushian, Robert K., Justin M. Bathon, and Martha M. McCarthy. 2011. "A National Study of the Net Benefits of State Pension Plans for Educators." Journal of Education Finance : 24 - 51. U.S. Census Bureau . 2003 - - 8. Median Household Income by State. https://www.census.gov/data/tables/time - series/demo/income - poverty/historical - income - households.html Report retrieved 4/5/18 from: https://www.ed.gov/oii - news/teacher - compensation U.S. Department of Education Office of Postsecondary Education Ar eas (TSA) Nationwide Listing for 1990 - 1991 through 2017 - Report retrieved 9/16/18 from: https://www2.ed.gov/about/offices/list/ope/pol/bteachershortag eareasreport201718.pdf