|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
1 THREE ESSAYS ON PUBLIC ECONOMICS: TEACHER TRAINING, ACCOUNTABI LITY, AND PUBLIC PENSIONS By NATALIYA PAKHOTINA A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2010
2 2010 Nataliya Pakhotina
3 To my family
4 ACKNOWLEDGMENTS I am grateful to David Denslow for his cont inuous support, encouragement, and instruction throughout all the stages of my graduate study and researc h. This work has also benefited tremendously from the guidance of David Fig lio and James Dewey. I thank Richard Romano, Paul Sindelar, and Steven Slutsky for helpfu l advice and comments. I especially thank my parents who always encouraged me to pursue an academic career and my husband whose support and help are invaluable.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS...............................................................................................................4 LIST OF TABLES................................................................................................................. ..........7 LIST OF FIGURES.......................................................................................................................11 LIST OF ABBREVIATIONS........................................................................................................ 12 ABSTRACT...................................................................................................................................14 CHAP TER 1 INTRODUCTION..................................................................................................................16 2 TEACHER ATTRITION: ALTERNATIVE AND TRADI TIONAL PATHWAYS TO TEACHING............................................................................................................................19 Introduction................................................................................................................... ..........19 Efficiency of Alternative Routes to Certification................................................................... 25 Previous Literature..................................................................................................................30 Description of Data............................................................................................................ .....34 Definition of Experience Cohorts...........................................................................................35 Variables Included in the Model............................................................................................. 46 Empirical Strategy............................................................................................................. .....51 Results.....................................................................................................................................54 Analysis of Possible Self-Selection........................................................................................59 Results with Controls for College Quality and Family Income.............................................66 Simulation of Efficiency of Alte rnative Certification Program s............................................75 Conclusions.............................................................................................................................78 2 THE IMPACT OF SCHOOL ACCO UNTABILITY ON TEACHER QUALITY ................ 92 Introduction................................................................................................................... ..........92 Previous Literature..................................................................................................................94 Types of Accountability Policies............................................................................................99 Empirical Strategy............................................................................................................. ...104 Description of Data............................................................................................................ ...105 Results...................................................................................................................................106 Results with Controls for District Size................................................................................. 111 Conclusions and Directions for Further Work...................................................................... 115 3 INVESTMENT STRATEGIES OF PUBLIC PENSION FUNDS ......................................124 Introduction................................................................................................................... ........124
6 Theoretical Background........................................................................................................130 Data Description...................................................................................................................134 Empirical Strategy............................................................................................................. ...136 Results...................................................................................................................................137 Conclusions...........................................................................................................................139 4 CONCLUSION..................................................................................................................... 150 APPENDIX A DETAILED RESULTS FOR TEAC HER ATTRITION ESTIMATION ............................ 151 B DETAILED RESULTS FOR ESTIMATION OF THE EFFECT OF SCHOOL ACCOUNTABILITY ...........................................................................................................163 LIST OF REFERENCES.............................................................................................................183 BIOGRAPHICAL SKETCH.......................................................................................................188
7 LIST OF TABLES Table page 2-1 Number of observations in experience cohorts form ed by three different approaches..... 80 2-2 Distribution of different types of exit by experience cohor ts and types of certificates ..... 81 2-3 Odds ratios for interactions of altern ative route to certi fication (ARC) indicator with experience dummies...........................................................................................................81 2-4 Marginal effects for results of the logit regression estim ation.......................................... 82 2-5 Odds ratios for interactions of ARC indicator with experience dumm ies with new variables added...................................................................................................................83 2-6 Marginal effects for results of the logit regression estimation with the new variables added ..................................................................................................................................84 2-7 Cumulative teacher attriti on rates and exit rates by path way for elementary, middle, and junior high school teachers in New York City (NYC), 2000-2004............................84 2-8 Simulation of ARC efficiency...........................................................................................85 2-9 Cost of ARC alternative t each er certification programs.................................................... 86 3-1 Teachers certification ratios by year and field................................................................ 118 3-2 Logit results for regressions estim ated separately for each field group ........................... 118 3-3 The effect of district-targeted accountabi lity on the like lihood of teacher certification.. 118 3-4 Logit results of regressions estimated for pairs of field groups (effect of districttargeted policies) ..............................................................................................................119 3-5 Logit results for regressions with controls for d istrict size estimated for each field group separately............................................................................................................... 120 3-6 The effect of school-targeted accountabil ity on th e likelihood of teacher certification conditional on the size of the district............................................................................... 120 3-7 The effect of district-targeted accountability on the like lihood of teacher certification conditional on the size of the dist rict in percentage points..............................................121 3-8 Logit results for regressions estimated for pairs of field groups for school-targeted policies, con ditional on district size.................................................................................121 3-9 Logit results for regressions estimated fo r pairs of field groups for district-targeted policies, con ditional on district size.................................................................................122
8 4-1 1926-1996 average annual rates of return........................................................................ 141 4-2 1926-2001 average annual rates of return........................................................................ 141 4-3 Wilshire asset class assumptions..................................................................................... 141 4-4 Assignment of survey items for investme nts and holdings to ge neral asset classes ........ 142 4-5 Assignment of survey items for investme nts and holdings to ge neral asset classes using W ilshire asset class assumptions............................................................................ 143 4-6 Summary statistics......................................................................................................... ..144 4-7 Ordinary least squares (OLS) regressions re sults for expected rate of return (ERR) based on the historical data and right to work law as a measure of labor union power.. 144 4-8 OLS regressions results for ERR based on the historical data and state union m embership as a measure of labor union power.............................................................. 145 4-9 OLS regressions results for ERR based on th e historical data a nd right to work law and state union m embership as a measure of labor union power.................................... 145 4-10 OLS regressions results for ERR based on W ilshire asset class assumptions and right to work law as a measure of labor union power.............................................................. 146 4-11 OLS regressions results for ERR based on W ilshire asset class assumptions and state union membership as a measure of labor union power.................................................... 146 4-12 OLS regressions results for ERR based on W ilshire asset class assumptions and right to work law and state union membership as a measure of labor union power................ 147 4-13 OLS regressions results for average risk based on W ilshire asset class assumptions and right to work law as a measure of labor union power............................................... 147 4-14 OLS regressions results for average risk based on W ilshire asset class assumptions and state union membership as a measure of labor union power.................................... 148 4-15 OLS regressions results for average risk based on W ilshire asset class assumptions and right to work law and state union memb ership as a measure of labor union power. 148 A-1 Summary statistics......................................................................................................... ..151 A-2 Results of logit regression estimation.............................................................................. 153 A-3 Wald test for joint significance........................................................................................ 154 A-4 Summary statistics for teachers with 1 year of ex perience (weighted)........................... 155
9 A-5 t-tests for difference in mean values of personal characteristics of ARC a nd traditional routes to certification (TRC) teachers with 1 year of experience (weighted)........................................................................................................................155 A-6 Summary statistics for ARC and TRC teach ers with less than 6 years of experience (weighted) ........................................................................................................................156 A-7 t-tests for difference in mean values of personal characteristics of ARC a nd TRC teachers with less than 6 years of experience (weighted)................................................ 156 A-8 t-tests for difference in mean values of personal characteris tics of alternatively certified teachers with 1 and 2 years of experience......................................................... 157 A-9 t-tests for difference in mean values of personal characteris tics of alternatively certified teachers with 2 and 3 years of experience......................................................... 157 A-10 t-tests for difference in mean values of personal characteris tics of alternatively certified teachers with 3 and 4 years of experience......................................................... 158 A-11 t-tests for difference in mean values of personal characteris tics of alternatively certified teachers with 4 and 5 years of experience......................................................... 158 A-12 Results of logit regression estimation.............................................................................. 159 A-13 Wald test for joint significance........................................................................................ 160 A-14 Results of logit regression estimation.............................................................................. 161 A-15 Wald test for joint significance........................................................................................ 162 B-1 Accountability policy...................................................................................................... .163 B-2 Summary statistics by fields and year..............................................................................165 B-3 Logit results for student-targeted accountability policy.................................................. 167 B-4 Logit results for school-targeted accountability policy................................................... 167 B-5 Logit results for district-targeted accountability policy................................................... 168 B-6 Logit results for aggregat ed accountability policy index ................................................. 168 B-7 Logit results for sample consisting of English language arts (base field) and m athematics teachers....................................................................................................... 169 B-8 Logit results for sample consisting of Eng lish language arts (base field) and social sciences teachers ..............................................................................................................170
10 B-9 Logit results for sample consisting of Eng lish language arts (base field) and sciences teachers ............................................................................................................................171 B-10 Logit results for sample consisting of math em atics (base field) and sciences teachers.. 172 B-11 Logit results for sample consisting of ma them atics (base field) and social sciences teachers............................................................................................................................173 B-12 Logit results for sample consisting of social sciences teachers (base field ) and sciences teachers..............................................................................................................174 B-13 Logit results for student-targeted accountability p olicy with controls for district size... 175 B-14 Logit results for school-targeted accountability po licy with controls for district size.... 175 B-15 Logit results for district-targeted accountabi lity po licy with controls for district size.... 176 B-16 Logit results for aggregated accountability pol icy index with controls for district size ..176 B-17 Logit results for sample consisting of English language arts (base field) and m athematics teachers with controls for district size........................................................ 177 B-18 Logit results for sample consisting of Eng lish language arts (base field) and social sciences teachers with c ontrols for district size ...............................................................178 B-19 Logit results for sample consisting of Eng lish language arts (base field) and sciences teachers with contro ls for district size............................................................................. 179 B-20 Logit results for sample consisting of ma them atics (base field) and sciences teachers with controls for district size............................................................................................ 180 B-21 Logit results for sample consisting of ma them atics (base field) and social sciences teachers with controls for district size............................................................................. 181 B-22 Logit results for sample consisting of so cial sciences (bas e field) and sciences teachers with controls for district size............................................................................. 182
11 LIST OF FIGURES Figure page 2-1 District labor mark et for novice teach ers........................................................................... 87 2-2 Additional supply of novice teach ers in shortages fields (SF) ..........................................87 2-3 Districts labor market for novice teachers........................................................................ 88 2-4 Share of teachers certified in main assignm ent field in 1999-00 and 2003-04 by field groups.................................................................................................................................88 2-5 Number of individuals issued certificates thr ough alternative routes to certification (ARC) by year ....................................................................................................................89 2-6 Experience matrix.......................................................................................................... ....89 2-7 Diagonal definition of experience cohorts......................................................................... 90 2-8 Definition 1............................................................................................................... .........90 2-9 Definition 2............................................................................................................... .........91 3-1 Teachers' certification ratio by field s in the 1993-94 and 1999-00 school years............. 123 4-1 Overestimation of expected rate of re turn (ERR) leads to hiring m ore public employees than optimal................................................................................................... 149
12 LIST OF ABBREVIATIONS AAEE American Association fo r Employment in Education AASCU American Association of State Colleges and Universities ARC Alternative Rout es to Certification BEBR Bureau of Economic and Business Research Calpers California Public Employees Retirement System Calsters California State Teachers Retirement System CCD Common Core of Data CPRE Consortium for Policy Research in Education CRR Center for Retirement Research CRRA Constant Relative Risk Aversion DB Defined Benefit DC Defined Contribution ERR Expected Rate of Return LPS Low-Performing Schools MSA Metropolitan Statistical Area NCAC National Center for Alternative Certification NCEI National Center fo r Education Information NCES National Center for Educational Statistics NCTAF National Commission on Teaching and Americas Future NYC New York City NYCTF New York City Teaching Fellows OLS Ordinary Least Squares RF Regular Fields SASS Schools and Staffing Survey
13 SF Shortages Fields TAAS Texas Assessment of Academic Skills TFA Teach for America TFS Teachers Follow-up Survey TRC Traditional Routes to Certification
14 Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy THREE ESSAYS ON PUBLIC ECONOMICS: TEACHER TRAINING, ACCOUNTABI LITY, AND PUBLIC PENSIONS By Nataliya Pakhotina April 2010 Chair: David Denslow Major: Economics This dissertation analyzes three different public economic policy issues. The second and third chapters investigate implications of econo mic policies affecting th e teacher labor market, and in particular their effect on teacher shortages in United St ates. In the second chapter, I analyze the relative efficiency of alternative path s to teacher certification compared to traditional programs. I estimate turnover rates for novice teach ers who came to teachi ng through alternative and traditional preparation routes. Differences in attrition rates for gradua tes of alternative and traditional programs appear only for novice teachers who have 1 year of experience or less. As soon as a teacher gains 2 years of experience, th e effect fades out. Using estimated attrition rates in a simulation model, I estimate the upper bound of alternative programs training costs per teacher that makes the efficiency of alternative programs at least equal to that of traditional ones. In the third chapter, I examine whether a nd how different types of school accountability affect the supply of teachers. I find no eviden ce that student-targeted accountability policies affect the likelihood that teachers will be certifie d to teach in their main assignment fields. School-level accountability policies do influence the teacher shortage, but only in large districts. District-level accountability policie s affect teachers both in small a nd large districts. The effect
15 of accountability policies is more pronounced for high-stakes fields whose testing results are used for school evaluation. However, the imp act of accountability policies varies across highstakes fields as well: accountability policies te nd to result in more certified teachers in mathematics while reducing their pr evalence in English language arts. In the fourth chapter, I anal yze rates of return that public pension funds use to project their future assets. Using a theoretical model for an optimal investment strategy I discuss deviations from optimality caused by the influenc e of labor unions on the pension funds. Then, I empirically test whether labor uni ons do affect the investment st rategy of public pension funds. The regression estimation results confirm that the pr essure of labor unions that are interested in high rates of return may cause deviations from the optimal investment strategy. Public pension funds in states with strong unions tend to use riskier investment strategies, compatible with higher rates of return but at the cost of greater risk.
16 CHAPTER 1 INTRODUCTION This dissertation focuses on three different public economic policy issues. In the next chapter I analyze the efficiency of alternative paths to teaching. For this purpose, I compare the turnover rates for novice teachers who came to teaching through alternative and traditional preparation routes. The analysis is based on a data set that pools nationally-representative teacher-level surveys from the 1999-00 and 2003-04 school years. Novice teachers who have less than 6 years of experience are assigned to 5 experience cohorts. Different assignment mechanisms for translating possibly ambiguous surv ey responses to cohorts are used to ensure the robustness of the results. The logistic and or dinary least squares (O LS) regressions results show that teachers who came to the profession through alternative routes are approximately 1418% points more likely to quit their main assignmen t field after their first year than teachers who completed traditional certification programs. The difference in attrition rates for the participants of alternative and traditional programs is re vealed only for teachers who have 1 year of experience or less. As soon as a te acher gains at least 2 years of experience, the effect fades out. After the third year, however, graduates of co mpetitive undergraduate colleges who came to teaching through alternative routes are also approximately 11-19% more likely to quit their main assignment field than graduates of less competitiv e colleges. Using estimated attrition rates in a simulation model, I estimate the relative efficiency of alternative routes to certification as a solution to the teacher shortage problem. The simulation shows the level of alternative programs training costs per teacher that equates the efficiency of alternative programs to traditional ones. In the third chapter, I exam ine the effects of school accoun tability on the teacher labor market. Evaluating schools on the basis of stude nt performance has been a popular education reform in the United States and abroad for over a decade. However, these policies are by no
17 means identical in nature or potential effect: they vary in both their mechanisms and their targets for improvement. While all focus on student impr ovement, some of them may also increase or decrease a teachers willingness to work as th ey may impose additional pressure on teachers, increase teachers salaries, or improve or ha rm working conditions. Because the accountability systems treat subject fields differently, they may also affect teacher s of various subjects differently. Using nationally-repr esentative teacher-level da ta for the 1993-94 and 1999-00 school years, I estimate the impact of differe nt forms of school accountability policies on the teachers labor market by comparing prea nd post-accountability teacher certification characteristics by type of schools and subject fi elds. I find no evidence that student-targeted policies affect the likelihood that a teacher will be certified to t each in his/her main assignment field. School-level accountability policies influe nce the teacher shortage, but only in large districts. District-level accountability policies affect teacher s in both small and large districts. The effect of accountability po licies is more pronounced for high-stake fields whose testing results are used for school evaluation. However, the impact of accountability policies varies across high-stake fields as well: an accountability policy tends to result in more certified teachers in mathematics while reducing their pr evalence in English language arts. In the forth chapter, I analyze the rates of return that public pension funds use to project their future assets. There are (at least) two poi nts of view on this question. Some financial economists argue that the projected returns of pu blic pension funds should be valued using rates of return on fixed income securities, while the act ual practice is to use higher rates of return. I find similarities between this discussion and de bates over the appropria te discount rate for evaluation of future costs caused by global warm ing. Using a tax-smoothing approach, I analyze deviations from optimal investment strategy a nd illustrate how labor unions may cause the
18 directors of pension funds to c hoose riskier investment strategies I test the hypothesis that labor unions influence investment strategies. The OL S estimation results confirm that strong labor unions appear to be associated with riskier investment stra tegies. Though not definitive, the results are compatible with the notion that public unions have pushed pension funds toward aggressive portfolios.
19 CHAPTER 2 TEACHER ATTRITION: ALTERNATIVE AND TRADITIONAL PATHWAYS TO TEACHING Introduction Over the last two decades, the school teachers shortage in certain subjects has been considered a problem of high prio rity in the United States. The creation and quick expansion of alternative routes to certificat ion (ARCs) was one of the numerous policies employed to address the shortage problems. The effectiveness of thes e programs is currently an issue that causes debate. This chapter focuses on th e analysis of the attrition of novice teachers who came to the profession through ARCs. Empirical results are used to estimate the relative efficiency of these programs as compared to traditional routes to certification (TRC). Before proceeding with the analysis of teachers a ttrition, it is important to understand the specific characteristics of the teacher shortages in United States. Though a near crisis of the teachers supply in United States has often been predicted since the 1980s, a severe shortage has not ye t come to pass. Between 1999 and 2001, the number of teachers in elementary and secondary public schools increased more than student enrollment. According to the American Association of State Colleges and Univ ersities (AASCU, 2005), currently there is no deficit of teachers in Unit ed States, though there exists a misalignment of supply and demand of teachers acros s geographic and subject areas. Murphy, DeArmond, and Guin (2003) estimated the late-fill ratio for the 1999-00 school year and found significant va riation across the country. Nine western, southwestern, and southeastern states experienced a significantly higher late-fill rate (2.3 %-5.9%) than the national average (1.5%). A nation-wide study of the demand and supply of teachers conducted by the American Association for Employment in Educ ation (AAEE, 2004) provides estimates of the relative demand of teachers by field. It reveal s a significant imbalance between the supply and
20 demand of teachers in several particular subject areas. Fields experiencing considerable shortages include special educa tion, physics and mathematics. There are also some shortages in other sciences, bilingual education, English as a second language, Spanish language, and technology education. Other subjec t fields have a relatively ba lanced supply and demand, and there even exists some surplus in elem entary education and social sciences. One cause of the deficit of teachers is the hi gh attrition rate of novi ce teachers, which in shortage fields interacts with the insufficient production of new teachers. Only 50-60% of teachers remain in the profession 5 years af ter entering (AASCU, 2005). According to the National Center for Education Information (NCEI, 2005) in 2005, K-12 teachers who were 50 years of age or older constituted 42% of the tota l number of teachers, and about 40% of current teachers were not expected to be working 5 years later. The National Commission on Teaching and Americas Future (NCTAF, 2008) reports that in 2008-09, in 19 states more than 50% of their teachers were older than 50. Hence while today the shortages problem is relatively moderate, the majority of their positions are filled with older teachers. The retirement peak will be reached during the 2010-2011 school year. In less than a decade more than 50% of all veteran teachers will leave the professi on. In addition attrition rates of beginning teachers have been increasing for more than a decade. According to the National Center for Educational Statistics (NCES, 2007) during the 2004-05 school year, the annual attrition rate for public school teachers with 1-5 years of full-time experience was approximately 8%, and for teachers without full-time experience it was as high as 19%. Another important aspect to consider is the quality of teachers. While schools might not be experiencing problems with filling positions, they still might face difficulties in finding experienced and qualified teachers. Baker and Smith (1997) found that the percentage of teachers
21 who are not certified in field they teach has been increasing over ti me. Ingersoll (1997) notes that many schools report difficulties in finding qualified teachers, rather than in just filling positions. However, it is difficult to estimate the shortage in terms of quality, since there is no way to measure it directly. Quality depends on many different factors such as experience, the teachers undergraduate college, correspondence of the teac hers main assignment field to his major in college, certification type etc. Hence the main characteristics of the current shortage of teachers in United States are: considerable shortage in special education, mathematics and sciences high imbalance between supply and demand of teachers in western, southwestern and southeastern states high attrition of novice teachers shortages of qualified teachers Why are these problems unable to be resolved by the market? Why is the teacher labor market unable to achieve equilibrium through a series of wage corrections? The inbalance between teachers supply and dema nd across fields and areas orig inates from a system of equal pay for all teachers of given se niority and degrees, independent of the field they work in. Moreover, there exist various entry barriers that reduce the mobility of teachers. The supply of teachers is a dire ct function of the relative wage If the reservation wage of the prospective teacher is lower than the ratio of a teachers wa ge to the alternative occupation wage, than he chooses to enter the teaching pr ofession. The opportunity costs of teaching vary across different subject fields: people specializing in mathematics and sciences are able to get higher compensation in non-educational jobs than those who specialize in the humanities. However the current system of teacher compensati on creates a uniform salary schedule that does not allow for differences in opportunity costs. Pr oposals to implement a di fferential wage system
22 with merit-based and field-based compensation face the opposition of strong professional unions. As a result, shortages of teachers occur in fields that have higher opportunity costs. The current certification and compensation syst em also creates barriers for teachers mobility across regions. Many states do not automatically confirm certificates that were issued in other states. Therefore, teachers cannot freely move into areas with shortages because they have to take additional courses and exams to get ce rtified in the new loca tion. Another barrier for relocation is the system by which salary rises with seniority. If teachers move to other districts, they usually lose credits for seniority and thus are paid lower salaries at the new jobs. These barriers exacerbate the problem of misalignment across regions. Since the 1980s, various policies have been implemented to combat teacher shortages, including offering ARCs. During th e last two decades, ARCs ha ve rapidly spread across the country. The important question is whether these programs represent an effective way to increase the supply of teachers, and do they specifically address teacher shortages? There are arguments both in favor of and against al ternative certification. Some feat ures of the ARCs clearly do address current shortage issues. These programs were originally designed to lessen the problem of misalignment of teachers supply by ge ographical areas and subject fields. Most alternate route programs are created specifically to meet the demand for teachers in the areas where they are established. Seventyone percent of providers of alternate route programs say their alternative programs serv e students in a high-needs area (e.g. low socioeconomic area, high poverty level, high mi nority) school. An additional 27% say that they serve some students in high needs areas Only 2% say their programs do not serve students in high needs areas (Feistrizer, 2005, p.63). While ARCs do increase the supply of novice te achers, particularly in high-need areas, the quality of the teachers coming to the professi on through these routes is a question that fosters a lot of discussion. On the one hand, according to the classification of ARC, the most common types of these routes have been designed fo r the explicit purpose of attracting talented
23 individuals who already ha ve at least a bachelor degree in a field other than education into elementary and secondary school teaching (F eistrizer, 2005, p.61). Ninety-eight percent of alternative programs have a bachelors degree as an entry requirement (Feistrizer, 2005). Hence, ARCs attract highly educated individuals into teaching. Since the average competitiveness of the graduates from colleges of educatio n is lower than that of graduates specializing in other major fields, ARCs are likely to attract more talent ed individuals into teaching than are TRCs. For those people who would like to change their profession in mid-career, ARCs provide an opportunity to enter teac hing with relatively low entry costs. In a survey, 47% percent of the respondents who participated in al ternative programs answered that they would not have become a teacher if an ARC had not been available. Only 22% answered that they would have completed a traditional program (Feistrizer, 2005). In their turn, the opponents of the ARCs ar gue that they provide under-prepared teachers. ARC programs have reduced requirements for course work and experience prior to becoming a teacher. Ninety percent of ARCs partic ipants work as full-time teachers before they complete the program (Feistrizer, 2005). Hence novice teachers coming to the profession through ARCs are less prepared to teach when they firs t time enter a class than are TRC teachers. However, this possible disparity in quality is not likely to persist as cohorts mature. Using a nonparametric investigation of experience Ri vkin, Hanushek, and Kain (2005) found that experience effects are mostly concentrated in the first few years of teaching. Specifically, teachers in their first and, to a somewhat le sser extent, their second year tend to perform significantly worse in the classroom. Hence, when a teacher gains at least 2-3 years of experience, the type of prepara tion is not likely to matter any mo re and the difference in quality dissipates over time. This leads to the question regarding the le ngth of teaching spells for the
24 graduates of ARCs. Do they stay in teachi ng long enough to become as qualified as their colleagues who completed TRCs or do they leav e the profession in few years? Opponents of ARCs argue that the costs of these programs ex ceed their benefits, sin ce most of the teachers coming into the profession through ARC leave teachi ng during their first 3 years. If this is the case, students get under-prepared teachers who la ck pedagogical skills and then these teachers leave the profession before they gain enough e xperience to become well-qualified for the job. The reasons for leaving may be different and co uld include being less devoted to teaching than those coming through TRCs or feeling under-prepared and ill-s uited for the job. Hence the relative efficiency of the ARC depe nds on the attrition rates of the graduates of these programs. Since the difference in preparat ion matters only for the novice teachers, it is important to know whether the retention rates for teachers coming through ARC are lower than for graduates of TRC. If graduates of altern ative programs mostly l eave teaching during their first years of working in elementary and secondary schools, then the relative efficiency of the ARC is low. To answer this question this chapter focuses on the analysis of the attrition patterns of teachers coming to the profession through ARC and TRC. Teacher-level national representative data is used to estimate the exit rates for t eachers belonging to the same experience cohort who obtained their teaching skill thr ough TRC and ARC. The main research question of my study is whether the exit rate of ARC novice teachers from teaching in their main assignment fields is different than the corresponding exit rate of their colleagues that completed TRCs. The understanding of the attrition patterns of the pa rticipants and completers of ARCs will provide background information for the analysis of the e ffectiveness of these programs, which I will use at the end of this chapter to estima te relative efficiency of the ARCs.
25 Efficiency of Alternative Routes to Certification As mentioned above, the shortage s are linked to particular su bject fields and geographical areas, not to the whole teachers labor market. A major source of the teacher deficit is a high turnover rate of novice teachers. The market is prevented from solving these problems by the uniform salary schedule. Let us look at two se gments of the market for novice teachers: the market for shortage field (SF) teachers and the market for regular field (RF) teachers. Teaching in the SF induces higher opportunity costs; hence teachers working there should be compensated with higher wages. For example, teachers working in math or sciences fields can find jobs with higher wages outside of education with greater ease than their colleagues specializing in humanities. Likewise, teachers working in schools with a high share of socially disadvantaged students might experience worse working c onditions and they shoul d be appropriately compensated. Therefore, the supply curve for SF teachers lies above the supply curve for RF teachers, since the higher wage is required in th e SF to provide the same supply of teachers. The most straight-forward way to solve the shortage problem is to increase wages for novice teachers. However, a predominant majority of teachers (70-80%) are working in the RF and the share of tenured teachers in the teachers labor force is higher than the share of novice teachers. Therefore, it is natural to assume that teacher labor unions first of all serve the interests of the RF tenured teachers. As a consequen ce, the current system of teachers pay favors experienced teachers. Salaries increases with years of experience more than in proportion to increases in productivity. The effect of a t eachers experience on st udent attainment was estimated by Hanushek, Kain, OBrien, and Rivk in (2005). They found that only the transition from 1 to 2 years of experience has a statistically significant positive effect on the teachers productivity. Hence, a teachers productivity incr eases in the first years of teaching and then remains stable until retirement. Nevertheless, labor unions support the system of backloaded pay
26 and are not interested in high wages for novice teachers. Therefor e, the uniform wage for novice teachers is defined by the equilibrium of th e RF segment of the teacher labor market. Panel A of Figure 2-1 shows the situation of the district labor market for RF teachers. The demand is assumed to be perfectly inelas tic, determined by two fixed parameters: the number of students and class size. The supply is elastic and the inte rsection of supply and demand determines the equilibrium number of teachers ( NRF) and equilibrium wage ( wRF). On panel B of Figure 2-1, one can see the corres ponding equilibrium on the labor market for SF teachers. The SF supply curve lies above the RF supply curve, and the intersection of demand and supply occurs when the wage is equal to wSF. However, because of the uniform salary schedule, the wage actually paid is the same fo r teachers working in SF and RF. Since the labor unions are dominated by the tenured RF teachers, the wage is fixed at wRF. The equilibrium wage is too low to provide the demanded number of SF teachers and results in a shortage that is equal to NSF= NSF NSF( wRF). Theoretically, there are two possible solutions for the SF shortage problems: to increase the wage from wRF to wSF or to shift the supply of SF teacher s to the right. Let us look at the results of the wage increase option. Due to the uniform salary schedule, it will lead to high marginal input costs of newly hired SF teachers because the salary will increase not only for SF teachers, but also for RF teachers. Assume that the wage has increased from wRF to wSF and now NSF teachers are hired in addition to NSF( wRF). Assume that the total number of teachers is N when the wage is fixed at wRF level ( N=NRF+NSF( wRF)). Then after the wage increases, the total number of teachers is N+ NSF and new wage is wSF. In this case, the marginal input cost of a newly hired SF teacher is ,RFSFSF SF SFN MICwwwPVww N (2-1)
27 where MIC ( w wSF) is the marginal input cost of the la st SF teacher hired thanks to the wage increase. w is the difference between wages wSF and wRF, and PV ( wSF) is the present discounted value of future payments to a newly hired SF te acher (starting from his second year in teaching and up to his last year of work), assuming wSF as the salary for all future periods. For simplicity, I assume that the present value of future payments to previously hired teachers ( N ) does not change. Hence, the marginal input costs can be di vided into two main cate gories: costs related to the payments of newly hired SF teachers and cost s related to the increase in pay to previously hired teachers. Now let us look on another solution for the SF shortages: increasing the total supply of SF teachers. ARCs are one of the possible ways to increase the supply of SF teachers. These routes allow individuals in mid-career to enter the teaching profession. Thus, they attract to teaching people who would not otherwise cons ider this option since they already hold a bachelors degree in another fi eld and do not want to enter a standard certification program. Since a majority of the ARC programs are restricted to shortage fields and areas, they typically increase the supply of SF teachers. Hence, creatio n of ARCs shifts the SF supply function to the right and thus decreases the s hortages in the SF without changing the wage. Figure 2-2 shows the new equilibrium at the SF segment of the teachers labor market after the introduction of ARCs. In this case the marginal input cost s of the last hired SF teacher is ,ARCRF RFARCMICwPVwC (2-2) where CARC is the average cost of training 1 ARC t eacher. Again marginal costs can be divided into two categories: costs related to the salary payments of newl y hired teachers and costs related to the production of newly hired teachers ( CARC). Since wRF < wSF, the first component of the MICARC is always lower than the first component of MIC ( w wSF)
28 .RF RFSF SFwPVwwPVw (2-3) Hence, MICARC
29 higher attrition of ARC teachers results in hiring more ARC novice teachers to provide NSF, than in case of higher wages. Attr ARCMIC accounts for these additional costs ,Attr RFARC ARCRF ARCARCARCwCM P L MIC PVw RRR (2-5) where is a coefficient of relative productivity loss induced by hiring a novice teacher (0<<1), and MPL is the marginal product of labor for an e xperienced teacher. Introducing the first-year attrition rate to the model also changes marginal costs in the case of th e higher wages approach SF RFSFSF TRCSFTRCwN M P L MICww PVww RN R (2-6) When corrections for productivity lo sses and different attrition rate s are added to the model, the gap between the red and blue lines becomes smalle r, because the share of inexperienced teachers is larger with ARCs. Now assume that the productivity of an ARC novice teacher is lower than the productivity of a TRC novice teacher, since ARC t eachers begin to teach before they complete a certification program. In this case loss of productivity will be even higher if the shortage problem is solved by an ARC rather than by higher wages: 1 ,Attr RFARC ARCRF ARCARCARC M PL wC MICPVw RRR (2-7) where is a coefficient of relative loss in the productivity induced by hiring a novice ARC teacher instead of a novice TRC teacher (0< <1). The additional lo ss in productivity also decreases the gap between red and blue lines, thus making ARCs relatively less efficient. In the subsequent sections I proceed with an empirical analysis of the attriti on rates of ARC and TRC teachers that allows me to estimate empirically RARC and RTRC. Then, at the end of this chapter I apply empirically estimated RARC and RTRC to a simple simulation model based on Equations 2-1,
30 2-2, 2-5, 2-6, and 2-7 to analy ze the relative efficiency of ARCs as a solution to the shortage problem. Previous Literature It is important to remember that ARCs b ecame an option in 1980, and until recently no statistically representative data sets describing characteristics a nd the length of teaching spells for graduates of these programs existed. Due to the lack of data only a few studies have addressed the question. Boe, Cook, and Sunderland (2007) analyze th e attrition rate s of full-time public school teachers using a na tionally representative data set from the 2003-04 school year. Teachers prepared by ARCs with 1-3 years of e xperience exit teaching at a 12% rate, while those who came to the specialty through TRC exit at a 7% rate. For the cohort with 4-6 years of experience, the exit rate for teachers prepared by ARCs decreases to 3%, but for TRC teachers it increases to 9%. However, the authors do not co nsider these disparitie s to be significant. Boe, Shin, and Cook (2008) explore how the intensity of different types of teacher preparation affects transferring between subjec t areas and exiting from teaching. Bivariate logistic regression is applied to the pooled sa mple of teachers with 1-5 years of experience. Extensive teacher preparation is defined as completing either 10 or more weeks of practice teaching or 5-9 weeks of practice teaching al ong with four common components of teacher preparation1. The results show that extensive pre-serv ice preparation reduces the probability of exiting teaching but does not affect the tran sits between different subject fields. Boyd, Grossman, Lankford, Wykoff, and Loeb (2006) re ported the 2004-05 cumulative attrition rates of New York City (NYC) t eachers by experience cohorts and by different pathways to teaching. They distinguish 6 pathwa ys: college recommended, individual evaluation, 1 Those components include: coursewo rk in selecting and adapting instructional materials, coursework in educational psychology, observation of other classroom teaching, and receiv ed feedback on their teaching.
31 New York City Teaching Fellows (NYCTF), Teac h for America (TFA), temporary license, and other. The first two groups represent TRC, while NYCTF and TFA are two big ARC programs that are implemented at the city and at a na tional level respectively. The other category includes all teachers that do not f it into the 5 categories defined above and presumably could be considered as other alternative routes to teaching. The turnover rate occurs to be substantially higher for the other category in the first year and for TFA in s ubsequent years. The main focus of their study is the effect of the program type on the student ach ievement. The results show that relatively small differences in student achievement can be attributed to preparation pathways, and these effects mainly exist only among firstyear teachers. Typically ARC teachers provide smaller gains in student achievements than T RC teachers at least duri ng the first years of teaching. However, these differences are small in magnitude and the variation in effectiveness within the program is greater than th e average difference between pathways. Darling-Hammond, Holtzman, Gatlin, and Heilig (2005) use data for teachers from the 1996-2001 school years in the Houston Independent School District to an alyze the relative efficiency of teachers by different types of preparation programs (including alternative certification group and TFA group). Teachers participation in TFA program appear to have a positive effect on students scores on the Texas Assessment of Academic Skills (TAAS) math test, but at the same time the negative effect of TFA on students scores in SAT-9 and Aprenda (the test given to Spanish-speak ing students) in mathematics and reading has been revealed. The effect varies across years and until 2001 it is mainly statistically insignificant. The negative effects on the student achievements were also found for the ARC teachers. In this study, experience was defined as a continuous variab le, hence teachers with different types of preparation were not compared w ithin experience cohorts. The resu lts suggest that standard or
32 regular certification is an importa nt factor that increases the teach ers efficiency and that negative effects of teachers participation in the TFA pr ogram dissipates as soon as TFA recruits obtain certificates. However the benefits of this improve ment in their teaching skills are relatively small since the majority of TFA partic ipants leave teaching af ter their second or third year (time when they usually ear n certificate). Kane, Rockoff, and Staiger (2008) use logi t regression to estimate the cumulative retention rates of NYC teachers from different preparation programs. The retention patterns for teachers coming trough TRC and for teachers co ming through NYCTF program appears to be equivalent, while the retention rates for TFA t eachers drop dramatically after the second year. This is not surprising since the recruits of TFA commit for only 2 years of teaching. Similarly to Boyd et al. (2006) only small impacts of the initial certification status of the teacher to student test performance is found, while large and pers istent differences in teacher effectiveness are revealed within the groups of teachers who have the same level of experience and identical initial certification status. Various studies of the effectiveness of the ARCs generally focus on the analysis of the impact of the type of preparation program on stude nt achievements. Attrition patterns are rarely the main interest in these studies. There are, however, a number of papers analyzing teacher attrition that examine the determinants of exits from teaching and transfers between schools and districts, but do not dist inguish different types of prepara tion programs. These empirical works mainly exploit either a discrete-choice or a duration empirical framework. The recent studies in this research area have revealed that though teacher salaries and opportunity costs do affect transfers between districts, they have little influence on the decision to exit teaching (e.g., Imazeki, 2005; Hanushek, Kain, & Rivkin, 2004). Positive effects of the
33 real teacher wages on the durat ion of teaching spells was also found by Murnane and Olson (1989), Dolton and van der Klaauw (1995), and Krieg (2004). Several studies found evidence of the impact of some teachers char acteristics (sex, age, marital stat us, interactions of sex and age, and interaction of sex and marriage status) on the attrition rates. For example, females are more likely to leave teaching than males (e.g., Imazeki, 2005; Stinebrickner, 1999; Murnane, & Olsen, 1989). Teacher attrition varies across subject area s and depends on the school environment. For example, teachers are more likely to leave schools with socially disadvantaged and lowachievement students (e.g., Krieg, 2004; Dolt on, & van der Kalauw, 1995; Boyd et al., 2005) while elementary school teachers are less likely to quit teaching or change district than high school teachers (Murnane, & Olson, 1989; Imazeki, 2005). My study focuses on the empirical analysis of the attrition rates of the novice teachers by preparation pathways and by experience cohorts. The advantage of using national-representative data allows me to analyze not only the effects of the big nation-wide ARC programs, like TFA or city-specific programs like NYCTF, but also to assess the average attrition rate for participants of various ARC across states that are pooled together. I assign all novice teachers to 1, 2, 3, 4 and 5 years experience cohorts. Using this appro ach, it is possible to co mpare attrition rates by the type of preparation program within the cohort s, as well as to compare attrition rates across cohorts for teachers coming through the same prepar ation route. Since previous studies of the effects of teachers preparation on student ach ievement found evidence of quality disparity between ARC and TRC teachers only for the teach ers with 1-2 years of experience, it is important to analyze the attrit ion of novice teachers by years of experience. Hence, instead of pooling teachers with 1-3 or 1-5 years of experien ce into 1 or 2 cohorts as previous works that used the same data set have done (Boe et al., 2007; Boe et. al., 2008), novice teachers are
34 assigned to 5 separate cohorts by years of experience. I also us e different criteria to assign teachers to alternative and traditio nal groups than Boe et al. (2007) and Boe et al. (2008). They focus analysis on the effect of the intensity of teacher preparation on teac her retention, defining three categories of preparation: extensive, some, and little. The definitions are based on the four common components of teacher prep aration: coursework in selec ting and adapting instructional materials, coursework in educational psychology, observation of other cl assroom teaching, and feedback on their own teaching. In my analysis I assign teachers to the al ternative and traditional groups according to their answers ab out the type of certificate they hold and the type of program that they completed to earn their certificate. I also take into a ccount the findings of the previous empirical studies of teacher attrition and control for the t eachers, schools, and district characteristics that have been proven to be im portant determinants of a teachers decision to leave the profession or to transfer to another school or district. Description of Data Data used in my study mainly comes from the Schools and Staffing Survey (SASS) and the Teachers Follow-up Survey (TFS) administer ed by NCES. Given the focus of my study, information on teachers employme nt status for at least two consecutive years and information about types of their pr eparation programs was required. SASS and TFS are two interrelated surveys. TFS is administered the year after SA SS, and tracks the career paths for about 10% of teachers who participated in the survey in the prev ious year. Thanks to this feature, it is possible to estimate turnover rates for participants of T FS. Both surveys are nati onally-representative and are administered every 4-6 years in public and private schools. However, in this study I utilize data only for public schools teachers. About 40, 000 public school teachers participate in SASS and respectively about 4,000 of them particip ate the following year in TFS. SASS includes several questionnaires: t eacher, school, district, and principal and library media center. For the
35 purposes of my study, I only used data coming fr om the teacher and school questionnaires which provide well-rounded information regarding the type of preparation program, educational attainment, experience, work assignments, compensation, economic and social background characteristics of teachers, and schools main attributes. Districts size data are extracted from the Co mmon Core of Data (CCD) also administered by NCES. This program annually collects fiscal and non-fi scal data for all public schools, public school districts and state educati on agencies in the United States2. Price indexes for Metropolitan Statistical Areas (MSAs) developed by the Bur eau of Economic and Business Research (BEBR) of the University of Florida are used to adjust teachers salaries across different regions and time periods (Dewey, 2005). Definition of Experience Cohorts Based on the answers of SASS participants to the questions concer ning their professional preparation, two main groups of teachers are defined: traditional and alternative. In 1999-00, SASS teachers were asked about the type of certifi cate they hold in their main assignment field. They had 5 possible options: re gular or standard, probationary provisional, temporary, and emergency. Teachers that reported they hold prob ationary, temporary and emergency certificates are considered neither alternative nor traditional and are dropped from the sample. Therefore, the sample includes only teachers who chose the re gular or standard or p rovisional categories. Those teachers who chose regular or standard were also asked about the type of program that they had completed to earn their certificate. He nce, the definition of alternative and traditional groups is based on the responses of teachers to the survey questions. The accuracy of the definition thus depends on how big the measur ement and response errors are. Ballou (1998) 2 This information is also provided in district questionnaires of SASS, however due to relatively large number of missing observations I decided to extract these data fr om CCD rather than SASS since it was more complete.
36 argues that respondents of the 199394 SASS were highly inaccurate in their answers concerning certification type. First, re-int erview studies conducted by th e NCES found a high degree of response errors in questions dealing with certif ication, thus indicati ng the ambiguity of the questions for the teachers. Second, a significant number of respondents from 8 states, that had not implemented ARC programs by 1993, answered that they hold an al ternative certificate. Third, the definition of the provisional certific ate might have been a source of confusion since in the survey it is described as being given to persons who are still pa rticipating in what the state calls an alternative certification program while in many states traditional teacher certification proceeds through two or more stages, with the provisi onal certificate given at the first stage. Hence, there are two main sources of potential response erro rs: teachers are unsure what type of certificate they hold and the definiti on of the provisional certificate in the survey is ambiguous. Fortunately, the structure of the 1999-00 a nd 2003-04 questionnaires improved and the first source of the problems was dealt with. Th e major difference is that in 1999-00 and 2003-04 alternative and advanced professi onal certificates are included in th e regular or standard state certificate category, and in an additional quest ion teachers are asked how the certificate or teaching skills were obtained. The respondent has to choose between 6 options: bachelors program, masters program, th year program, alternative progr am, professional development, and other. I assign teachers who earned a certificate as part of a bachelors, or masters, or th year program to the traditional group. Teachers who earned a degree as part of an alternative program, or through continuing professional devel opment, or through some other program are assigned to the alternative group. Thus, even if teachers are not su re about the type of certificate they hold, they clarify this issue through responses to addi tional questions, describing the
37 program they have completed. Also, in 1993-94 the ARCs were only in the early stages of development. In 1999-00, they were already widely developed across states and by 2003-04 ARCs contributed a significant sh are of novice teachers to the edu cation system. As the size and impact of these programs increases, misundersta nding of what is meant by alternative program in the survey is likely to fall compared to th e 1993-94 survey. Thus, the first type of possible response error is much less likely to occu r in 2003-04 and 1999-00 than in 1993-94. The second source of the problem, an ambiguity in the definition of the provisional category, is more difficult to deal with. For the 2003-2004 SASS the same technique to distinguish between alternative a nd traditional certificat es can be used since all teachers holding provisional certificates also an swer an additional question rega rding how they obtained their teaching skills. Hence, teachers wi th provisional certifi cates are assigned to the alternative group only if they obtained skills through alternative programs, or through continuing professional development, or through other program. However, in 1999-00, SASS teachers holding provisional certificates were not asked this question. To solve this problem for 1999-00 SASS data I exclude from the sample teachers who reported that they held a provisi onal certificate and who were work ing in states where in 1999-00 a provisional certificate was the first stage cert ificate in the TRC programs. I also exclude teachers who reported that they held a provisional certificate and who were from the states that had no ARCs at this time period3. In order to check whether response errors in the 1999-00 SASS drive our results, I also r un estimations for the 2003-04 sample separately. While all main results hold for 2003-04 separately, in this study I report results for the pooled sample since there 3 These states include: Colorado, Michig an, New York, Ohio, Virginia (provisiona l certificate as part of standard certification) and Alaska, California, Indiana, Kansas, Montana, Nebraska, Rhode Island, and Wisconsin (no ARC in 1999-00).
38 are few observations for teachers belonging to an alternative group. Pooling both periods together allowed me to increase the power of estimation. The 2003-04 SASS has the same pattern of questions concerning certification, however in this survey teachers are asked about any ce rtificates they have, while in 1999-00 SASS they are asked whether they hold a ce rtificate in their main assignm ent field. They also answer questions about additional certif icates they hold. To make th e definition of the certification groups consistent across years, I have constr ucted the same groups for 2003-04 SASS as in 199900 SASS by combining information on certifica tion with information about main teaching assignments. Participants in the survey report the code for the content area their certificate allows them to teach and the code of their main assignm ent field. I arrange all the codes in ten major field groups: elementary education, English an d language arts, mathematics, natural sciences, foreign languages, special educ ation, English as a second lang uage, social sciences, specific matter subject (health education, arts and musi c, philosophy and religious studies), and other (vocational, technical, other). The teacher is considered certified in the main assignment field if she holds a certificate within a cont ent area code that belongs to the same field group as her main assignment field. I also made additional correc tions to this definition: If the teacher holds a certificate allowing him to teach in elementary school he is considered certified if he is working only in grades 1-6. If the teacher holds a certificate allowing hi m to teach in elementary school and he has positively answered the question: You are an elementary school teacher who teaches only one subject. If the teacher holds a certificate that applies to elementary grades he is considered certified if he is working only in grades 1-6. Figure 2-4 shows the ratio of teachers certified in a main assignment field in 1999-00 (as reported by teachers themselves) and 2003-04 (constr ucted using content area codes of certificate
39 and main assignment field code). The self re ported data has a slightly higher ratio of correspondence of the cer tification area to the main assignment field, except for elementary education and special educati on field groups. Most probably for these two groups, the matching of content areas in certificates to main assignment fields was the most accurate, as for elementary education I applied additional co rrections mentioned above. Special education is a stand alone field which could hardly be counted as any other field. For othe r categories I might have missed some correspondences of certificatio n areas to main assignment fiel ds, but in further analysis I use only teachers who are certified in their main assignment field. Hence, those teachers who might have been wrongly counted as not cer tified are not included in that sample. I also tried to add data from the 1993-94 SA SS and the 1994-95 TFS to my analysis, but the structure of the questions concerning certific ation in the 1993-04 survey differs from 1999-00 and 2003-04. Instead of two questions, teachers answered only one in which they chose from seven different certificate defini tions: advanced professional cert ificate, regular or standard, certificate offered to persons who have comp leted an alternative certification program, provisional, probationary, temporar y and emergency. Hence, there is a risk that some teachers who reported that they had a regular certificate in 1993-94 actually held the same type of certificate as some teachers who reported in 1999-00 that they had a regular or standard certificate earned through an alternative program and were thus assigned to the alternative group. For my study, the 1993-94 SASS had another w eak pointduring that time ARCs were not widely developed. As Figure 2-5 shows, the number of teachers obtaining certification through ARCs has increased substantially sin ce the late 1990s. Therefore, adding the 1993-94 data would not significan tly increase the si ze of the alternative group, and since the system of
40 ARCs had just began to develop at that period it is not clear whether the definition for alternative programs in 1993-94 is consistent with the definition in 1999-00 and 2003-04. Hence, the sample includes data for participants in both the SASS and TFS surveys in 1999-00 and 2003-04. I restricted data to full-time and part-time teachers holding a certificate in a main assignment field. However, the sample which I used in the analysis is even smaller due to additional restrictions based on th e definition of the experience. I do not have time series data and cannot observe the duration of teachers empl oyment spells from the beginning to the end. I can only analyze the probability that teachers wi th different levels of experience leave the profession at one time point. Therefore, for th e purpose of my study it is imperative to precisely define experience cohorts. Since I focus on the comparison of turnover rates of novice teachers with different level of experien ce, I need to unambiguously distinguish teachers with 1 year of experience from those with 2 or more years of ex perience. That further restricts the sample size by eliminating some observations of teach ers with ambiguous experience levels. The analysis focuses on the tur nover rates of teachers with less than 6 years of experience and assigns them to 5 experience cohorts acco rding to years of expe rience: 1 year cohort (includes only teachers who have 1 year of experience), 2 years, 3 years, 4 years, and 5 years respectively4. The SASS questionnaire includes several questions concerning teacher experience. First, respondents are asked H ow many years have you worked as a full-time elementary or secondary teacher in public school? This que stion is followed by H ow many years have you worked as a part-time elementary or secondary teacher in public school? Finally, both questions are asked with respect to private schools. In a ll these questions the teach er is supposed to report only whole years and to in clude the current year. 4 I also did the same analysis with 10 experience cohorts for teachers with less than 11 years of experience. That yielded virtually the same results.
41 Part-time experience was considered as wort hy as full-time, and experience in private schools as valuable as experience in public schools. Hence, it is na tural to define total experience as the sum of full-time and part-time experience in public and private schools. p ub pub pr prTEFTEPTEFTEPTE, (2-8) where TE represents total experience and FTE and PTE full-time and part-time experience respectively. However, by using this definition of total experience I risk overlooking a couple of pitfalls. First, teachers may work both full-time and part-time in the same school year. It would be preferable to count it as 1 year of total experience, rather than 2 independent years of experience. Second, teachers may have had a brea k in service at some point and then recently returned to teaching. As my study focuses on the career paths of novice teachers with less than 6 years of experience, the primary interest is the comparis on of turnover rates of teachers who recently entered teaching and had no breaks in their servi ce. For example, it hardly makes sense to look for a difference in turnover rates for teachers with 1 year of experience and teachers with 2 years of experience who had a long break in the service a nd returned to teaching the same year as those entering the teaching profession. To avoid that sort of problem I used another piece of information concerning experience derived from the question In what year did you begin teaching, either full-time or part-time, at th e elementary or secondary level? Using this additional information I constructed a matrix for each SASS survey, where columns show the years teachers reported as the first year of teaching, and rows show the total experience calculated using Equation 2-7. Figure 2-6 represents a fragment of such a matrix for respondents to the 2003-04 SASS with 5 or less years of experi ence. Each cell shows the number of respondents who have reported their first year of teaching and TE that corresponds to this cell of the matrix. In this matrix I used
42 a sample that includes only teachers with regula r or alternative certification and for whom I have next year follow-up information (from 2004-05 TFS). It is natural to suppose that teachers whose answers lay on the diagonal of the matrix had no breaks in their servic e and correctly counted their years of experience. Teachers whose answer s lie on the upper corner of the matrix may have had a break in their service. However since there are relatively more re spondents whose answers lie at the cells adjoining to the diagonal, one assumes that some of these respondents could have miscalculated their experience by not including the current year. For example, there is a possi bility that some of the 33 respondents who reported their first year of teaching was in 2002 where TE is equal to 1 actually have not counted the current school ye ar (2003-04). Hence, it is ambiguous to what cohort these respondents belong. We also have some respondents in the lower corner of the matrix. While it is virtually impossible to enter teaching in the 2003-04 school year and to have 2 years of experience, 15 respondents reported such information. There are two possible explanations of these anomalies. First, as discussed above, some teachers may be working both as full-time and part-time teachers in the 2003-04 school year. In thei r case 1 year of experience was counted twice. This sort of mistake can be easily corrected by looking on the full-time experience of these teachers. p ubprFTEFTEFTE, (2-9) where FTE represents full-time experience of thos e working both in public and private schools. The fifteen teachers mentioned above that have 1 year of FTE and 2 years of TE, most likely belong to the cohort with 1 year of e xperience (assuming they worked both full-time and part-time in the 2003-04 school year). For the rest of those who have 2 years of FTE, the most probable explanation is that they entered teach ing in the Spring semester of the 2002-03 school year and literally their first year of teaching wa s 2003. Technically they have about 1.5 years of
43 experience. Hence, it is ambiguous to what cohort these respondents belong 1 or 2 years of experience. One should not exclude the possibility that some of these teachers actually entered teaching in the Fall semester of the 2003-04 scho ol year, but might have counted calendar years (2003 and 2004), rather than school years (2003-04) t hus causing discrepancies. Diagonal definition. Taking all these possibilities into account, the simplest way to define experience is to use only the diagonal cells of th e matrix. For example, define the 1 year experience cohort as teachers w hose first year of teaching was in 2003 and who had 1 year of TE. In this case, one is secure from counting teache rs with 1 year of experience as teacher with 2 years of experience etc. This definition will be referred to as diagonal. Figure 2-7 illustrates the mechanism used in the diagonal definition fo r assigning teachers to the experience cohorts. This definition discards observa tions lying outside the diagon al of the matrix. The main shortcoming of this approach is a sharp reduction of the sample size. While the whole sample is quite representative, there is a shortage of observations for ARC novice teachers in each experience cohort. That is why it is important to use as many observations as possible for teachers with less than 6 years of experience. The main sources for the additional observations are cells adjoining the diagonal. As discussed above, it is not always clear what cohort these observations belong to. That is why in this study 3 different definitions of the experience cohorts, including the diagonal, are used. Each definiti on treats the ambiguous observations on a case-bycase basis. In these definitions, I focus mostly on the ce lls adjoined to the diagonal. I discard all observations lying in the lower corner of the matrix on the cells not adjoining to diagonal. It constitutes just a few observations, and most likel y contains errors due to respondents answers. In all definitions, except one, I i gnore observations lying in the uppe r corner of the matrix which
44 are not adjoining the diagonal because the primary interest of my study is to analyze the career paths of teachers who had no breaks in service. However, in order to increase the sample size as much as possible I exploit these observations in one definition of the experience cohorts. The definitions of experience cohorts are given more fully below: Definition 1. This definition exploits all observati ons from the cells adjoining to the diagonal. The following example illustrates all th e steps required to defi ne the cohort with 2 years of experience. This cohort includes all observations lyi ng in the corresponding cell on the diagonal of the matrix: teachers who have 2 years of TE and reported 2002 as their first year of teaching. Next, I look at the cells adjoining the diagonal cell. The lower corner includes some respondents with 2 years of TE and 2003 as their first year of teach ing. For those respondents, I check their FTE and if it is equal to 1 I consider the respondent to have 1 year of experience. If the respondent has 2 years of FTE, I consider him to have 2 years of experience assuming that he entered teaching in the spring semester of the 2002-2003 school year. Figu re 2-8 illustrates the logic used in definition 1 for assigni ng teachers to experience cohorts. I use the same intuition for respondents who reported 3 years of TE and 2002 as their first year of teaching. They are considered to have 2 years of experience if their FTE is equal to 2 years, and they are counted in th e 3 years of experience cohort if their FTE is equal to 3. Then I proceed with cells in the upper corner. If the re spondents reported 1 year of TE and 2002 as their first year of teaching, and they are working as full-time or part time teachers in the 2003-04 school year, they are considered to have 2 years of experience assuming th ey did not include the current school year in thei r response. I use the same logic for all other cohorts. Definition 2. This definition attempts to increase the sample size as much as possible and exploit all observations in the upper corner of the matrix. This cohort includes all observations
45 lying in the corresponding cell on the diagonal of the matrix: teachers who have 2 years of TE and reported 2002 as their first year of teaching. For the cells from lower corner of the matrix that are adjoining the diagonal I use the same mech anism as in Definition 1. For all cells in the upper corner of the matrix, I assume that TE is reported correctly by resp ondents. Teachers with 2 years of TE whose first year of teaching was before 2002 had had a break in their service at some point, so they are considered to have 2 years of experience. Fi gure 2-9 illustrates the mechanism used in definition 2 for a ssigning teachers to experience cohorts. The same logic applies for all cohorts excep t the cohort with 1 year of experience, because it is not possible to have a break be tween the 2002-03 and 2003-04 school years. The number of observations in each experience cohort varies across different definitions. The diagonal definition provides the smallest sample and definition 2 the largest sample. Table 2-1 shows the number of observations in each experience cohort for ARC and TRC novice teachers with less than 6 years experience. The data incl udes observations for teachers from 48 states (Hawaii and Alaska are excluded) and the District of Columbia. Some explanatory variables (school and district attributes) have missing observations and this slightly decreases the sample size. The total number of observations for the pooled data set consists of 5224 to 6020 observations depending on the definition of expe rience cohorts. Though this sample is quite representative, for the purpose of this study I comp ared turnover rates for teachers with different types of certification and various levels of experience. For ex ample, I compared ARC teachers with 1 year of experience to regularly certified teachers with 1 year of experience. Looking at the distribution of teachers across years of experience I have a relatively small number of observations in each category. Hence I had to pool data from two surveys for my study. Separately, each survey does not provide enough observations.
46 Variables Included in the Model Generally, teachers attrition studies exploi t several different definitions regarding teacher exits including exiting from teaching, tran sferring to another district, and transferring to another school. The choice of the most appropri ate definition usually depends on the scope and the main research interest of th e analysis. For example, for the analysis of different state and district policies, at the state leve l the transfers between districts a nd schools are of interest. At the national level they do not significantly affect the main picture. Since the primary interest of my study is the effectiveness of ARCs in solving the shortage problem at the national level, teachers are said to exit if they do not continue to teach in the same main assignment field the next year. In other words I analyze whether or not novice teac hers continue to work in the main assignment field of their teaching certificate. Those who do not work in the same field the next year might represent several different types of exits including exiting from the education system, exiting from teaching to administrative work, or remaining a teacher but working in another subject field. I do not distinguish between these subcategor ies, since the shortage problem is an attribute of the particular subject field and I am interested in whether or not teachers who completed the ARC do in fact solve the shortage problem in the field they are certified to teach in. For example, if teachers completing the alternative program for special education usually work only a few years in the field, are they likely to change their main assignment field to elementary or vocational education? If this is the case the a lternative program is not effective in solving the problem of the shortage of special education teachers. Previous empirical studies of attrition suggest that several particular teacher characteristics should be included, such as race, gender, marital status, educational attainment, and subject field. Marital status might affect attr ition because married teachers are more likely to move to another place because of spouse relocation. The effects of marital status might vary by
47 gender since women are more likely to quit teaching due to personal or family reasons. For example, women may quit teaching for maternity leave or to care for children. To account for this effect, the interaction of gender and marital status is included in the regression. As a measure of educational attainment a mast ers degree is used. Some regressions also include a dichotomous measure of college quality, which is based on an index of college quality. The index varies from 1 to 5, where 5 corresponds to the most competitive colleges and 1 to the least competitive. Different definitions for the college quality variable were explored, such as an index varying from 1 to 5, a set of four du mmy variables, and different binary variable definitions. Finally, I settled on the dichotomous meas ure, because it provides the best fit for the model with scores 3 to 5 of th e college quality index corresponding to more competitive colleges and scores 1 and 2 to less competitive colleges. For subject fields a set of indicators show s whether a teacher works in mathematics, English language arts, natural sciences, social sciences, special educat ion, or other subject fields5, leaving as the base category those working in elementary education. These variables are defined according to the subject fi eld group classifications discussed in the previous sections. I also indicate whether a teacher is a member of a professional union, since it is more difficult to fire a member of a union. One of the main indicators of teacher quality is experience. The longer a person stays in the teaching profession the less likel y he is to pursue another occupation, because he already has mastered this occupation while in another it would take him severa l years to become a professional. On the other hand, the longer a teac her stays in teaching the more likely he is to retire, while the annual increase in the skill of a teacher decreases with years of working. Hence, 5 English as a Second Language is included into the other subject fields group since the turnover rates in this field group are closer to the other field groups th an to the English language arts field group.
48 experience is likely to have a positive influence on a teachers decision to stay in the profession, but this effect tends to dissipate over the years. To control for these factors, I include in the equations teachers total experience and experi ence squared. Murnane and Olsen (1989) have found that individuals who start teaching before age 30 are more likely to quit teaching. I tried two different definitions of this indicator including a continuous variable showing the age they began teaching, and a dichotomous variable fo r whether the teacher entered the profession by age 30 or more. The first defini tion provided a better fit, so I chose it for the final model. Another possible source of difference in tu rnover rates might be family income. A teacher who is the primary earner in the family is more likely to be sensitive to the wage and opportunity costs and may be looking for a job outsi de of teaching if it co uld provide him with a higher income. A secondary earner might be less interested in alternative occupations. Hence, I also include in some regressions a set of four dummy variables indicating what range of family income the teacher belongs to. As in case with the college quality variable, I tried different definitions of this variable and have chosen the one that provides the best fit for the model. Since I pool together data from two diffe rent time periods (1999-00 and 2003-04), I add an indicator of school year into the regression equation in order to control for possible changes in the general economic environment and the relative prestige of the teaching profession. The teachers individual data is supplemented by school and district characteristics. I control for the teachers working environment by including the r acial composition of students and teachers and the share of students el igible for free lunch as a proxy for the proportion of low-income students in the school. A few variables are included to de scribe the relative attr activeness and business activity of the area where the teacher works and th e variety of job choice s in a school education system in the vicinity. For these purposes, I indicate whether the school is in a rural area and
49 include a measure of the relative size of the schools in the distric t: district enrollment divided by number of schools. Several previous studies have investigated the effect of salaries and opportunity costs on teacher turnover. Imazeky (2005) found that a teach ers decision to quit the profession depends on the teachers salary and its ratio to the averag e salary of teachers in the neighboring districts. Murnane and Olsen (1989) found a positive relation between the probability of quitting teaching and the average starting salary paid by business and industry to college graduates. Stinebrickner (1998) found that the length of the first teaching spell is responsive to the teachers wage. To control for these factors I incl ude the teachers annual salary6 in the regression equation, but since I am not particularly interested in the eff ect of wages on teacher attr ition, I do not directly control for opportunity costs. However, I do account for these costs in two ways. First, I include state fixed effects into the model to catch regi onal differences. Second, I weight the salary of teachers by a comparable wage index. The idea behi nd this index is that in equilibrium the ratio of wages in all occupations between any two regi ons is the same and is defined by the difference in the prices and the differences in local amenities. 01 ,i c c i cw P wA (2-10) where wi c represents the wage in occupation i and city c, with c=0 denoting the reference city. The spatial cost of living index, based on the rela tive prices of the market-basket, is represented by Pc. Ac is a measure of the relative willingness to pay for the amenities of area c relative to the reference location 0, valued at area c prices, with higher values corresponding to preferred amenity bundles. Since in equilibrium this condi tion holds for all occupations, I am able to control for opportunity costs by weighting teacher salary with this comparable wage index. 6 2003-2004 school year salaries were deflated to the 1999 prices using CPI.
50 Weighted salaries are comparable across different locations. Hence, if teachers have a relatively high weighted salary they are overcompensated in their region compared to other occupations, and if they receive a relatively low weighted salary they ar e undercompensated and other occupations may be attractive to them. The compar able wage index is available for all MSAs. I used school zip codes to locate their latitude and longitude, and then each school was matched via geographical coordinates to th e closest MSA. For those schools that are located far from any MSA, I used the average index for the state, which was found as sum of indexes for MSAs located in the corresponding state, weighted by their population. My primary interest is in th e effect of participation in an ARC on the novice teachers decision to quit their main assignment field. I de fine two variables to control for certification. First, I include an indicator s howing whether a teacher obtained his professional skills in the main assignment field by completing an ARC. Th e second certification variable controls for whether a teacher has a traditiona l certificate in any other field (not main assignment). It might be relatively easy and more attractive to change the main assignment field for those teachers who hold certificates in other fields. B ecause of the specifics of working in special education, I also include an interaction indicator of whether a teachers main assi gnment field is special education and a variable showing whether the teacher holds a traditional certific ate in another field. Finally, a set of dummy variables for 5 expe rience cohorts of the novice teachers is included in the equation. Then, I interact cohor t dummies with type of certification indicators those are the variables of main research interest I investigate whether t eachers who came to the profession through ARCs are more likely to exit in their first years of working in the school education system compared to TRC teachers. Moreover, I am looking at the effects of ARC preparation separately for each experience cohor t. For policy purposes it is important to know
51 how quickly the effect of the pr ogram characteristics dissipates. Does experience eliminate the differences between ARC and TRC teachers, and how quickly do the differences disappear? In other words, I am interested in whether teach ers who completed ARC are more likely to quit their main field only after the first year, or if they still have higher turnover in consequent years. If a statistically significant difference in turnover rates is observed only after the first year of teaching and not later, that lead s to important policy implications. First, it would make it easier to predict the tur nover rates of graduates of the ARC for cost-benefit analysis purposes. Second, it would imply that ARCs provide teachers of qua lity comparable to the graduates of TRCs. It also implies that even if ARC teachers are le ss qualified and less committed to the profession during their first year in the school education system, these di screpancies disappear by the second year. Summary statistics for the variables used in estimati ons are presented in Table A-1 of the Appendix A. Empirical Strategy The background for empirical estimation of t eacher attrition is based on the utility model of individuals sorting across jobs. The preferences of teachers are reflected by utility functions: ,,ik kkiUUwZX (2-11) Where wk is the wage of the alternative job k Zk represents non-pecuniary job attributes (working conditions, geographical location, distri ct and school charact eristics etc.), and Xi is a scalar measure of teachers characteristics which may affect his or her preferences for wk and Zk. An individual chooses a job by maximizing utility su bject to costs of each alternative. Hence, an individual will stay in teaching if the net utility of the job is the highest when compared to alternatives. There are two main empirical strategies comm only used in teachers attrition analysis. Recent studies are mostly based on a durati on model framework. Duration models allow
52 estimating the conditional probability of a teach ers exit, given he had not left during the previous years. Imazeki (2005) analyzes teacher ex its and transfers to another district using the Cox proportional hazard duration model. Podgu rsky, Monroe, and Watson (2004) define a discrete Cox proportional hazard model to estimat e the probability of a te acher leaving the state school system conditional to the years spent prev iously teaching. Murnane, Stringer and Willet (1988) exploit a proportional hazard model to pred ict the lengths of teachers first and second spells in teaching. Stinebrickner (1998) applies two stage estimation methods. At the first stage, he estimates the joint distribution of variables with missing data, and then the results of the first stage are integrated into the duration model of the second st age. Dolton and van der Klaauw (1995) use a proportional hazard model to analyze teacher retention and turnover7. The structure of the data set considered does not allow for tracking the career paths of teachers for longer than 1 year In the two time periods I pool ed together, I observed only whether a teacher stays or exits th e next year. This feature of the data makes it impossible to use a duration model framework because this type of model requires information about length of teaching spells beginning from the moment an individual entered the profession and for some subsequent interval of time. This is not the case in this data set, so the estimation strategy was restricted to other types of models. Other commonly used empirical strategies to es timate teacher attrition include binary and multi-response discrete models, where the dependent variable takes two (in the case of binary models) or several (in the cas e of multi-response models) valu es, each of them representing possible outcomes. The choice of a multinomial model is preferable if some explanatory variables are likely to affect exit and transfer d ecisions differently. According to the results of 7 For other examples of duration models used in teachers attrition studies see: Grissmer and Kirby (1992), Kirby, Naftel, & Berens (1999), and Stinebrickner (2002).
53 several previous studies some factors do in fact affect the decisions to quit teaching and to transfer to another district diffe rently. This difference is especial ly pronounced for the effects of wages and opportunity costs. Imazeki (2005) and Hanushek et al. (2004) have found that transfers to other school district s are more sensitive to teachers salaries and opportunity costs than are exit decisions. Boyd, Loeb, Lankoff, and Wykoff (2005) also found that the decision to quit the school system is more related to the le vel of student achievemen t in the school than is the decision to transfer to another school or district. The focus of my research is on the national level; therefore I do not analyze transfers between schools and districts. My definition of a teacher exit (not teaching in the same main assignment field the next year) includes three different alternatives: continue to work as a teacher, but in a different main assignment field; still work in the schoo l educational system, but not as a fullor part-time teacher; and exit from the elementary and secondary educational system. To account for these differences I first tried to apply a multinomial logit model to the analysis by defining four differe nt alternative choices: stay in th e same field, change field, quit teaching, and quit secondary and elementary edu cational system. However, I found out that I have too few observations in each category when I subdivide them by types of certificates and assign them to different experience cohorts. Ta ble 2-2 shows the number of observations for each type of exit8. For some cohorts I have less than 10 exits of the 2nd and 3rd types for ARC teachers. Hence, I do not have enough observations to de fine a multi-response discrete model. However, since I am not particularly interest ed in the analysis of the effect s of the salaries and opportunity costs on the teachers decisions to quit or transf er, and the scope of my analysis is national, 8 Table 1-2 shows number of observations for each type of exit corresponding to the sample formed by definition 1 of the experience cohorts.
54 rather than states, it is not necessary to disti nguish between different types of exits. The main purpose of this study is to analyze the efficiency of ARCs at the federal level. For this purpose I am mainly interested in whether graduates of th ese programs continue to work in the field they were trained in after their first years of teaching. Thus, from the policy point of view it is not really important to distinguish between these diffe rent types of exits. Sinc e the teachers shortage is an attribute of the subject field, those who leave the subject field do not solve the problem of the shortage any more than do those who exit teaching. Hence, I proceed with a binary choice mode l. The probability an individual exits ( y=1 ) is defined as a function of the school, district and teachers characteristics 1,,iiiPyXFx (2-12) where,iFx is a logistic distribution function exp 1exp w Fw w (2-13) and Xi represents a vector of observable characteristics. Results Table A-2 of Appendix A reports the result s of the logit regression estimations. The results are stable across all definitions of the experience cohorts. The coefficients have close magnitudes and similar levels of significance. To make the results more interpretable Table A-2 shows hazard ratios calculated as an exponential function of the coefficient from the regression ( exp( ) ). For individual variables it shows the rela tive shift in the hazar d caused by a one-unit change in the variable. For example, if the haza rd ratio for female teachers is 2.04, it means women are 2.04 times more likely to exit than me n are. If the hazard ra tio for teachers working in the special education is 0.63, it means they are by 37% less likely to exit the main assignment field than elementary school teachers are.
55 For the interaction of two or more variables, however, the odds rati o has to be calculated by multiplying coefficients for individual variables and their interactions with other variables. For example, for teachers who completed ARCs and belong to the 1 year experience cohort, the odds ratio can be calculated as the interaction of hazard ratios for th e ARC indicator and its interaction term with th e experience cohort dummy: 11expexp.AltExp Alt AltExpOR (2-14) Hence, if the hazard ratio for ARC teachers is 0.83, and the hazard ratio for the interaction of the 1 year experience cohort dummy with ARC dummy is 3.24, then ARC teachers with 1 year of experience are 0.83.24=2.70 times more likely to exit than TRC teachers from the same experience cohort. The main research interest of this study is the analys is of differences in attrition rates of teachers with di fferent types of certificates c onditional on the experience level. Table 2-3 reports odds ratios for ARC teacher s with less than 6 years of experience9. Teachers who completed ARC programs are 2.7 times more likely to exit the main assignment field immediately after the first year of teaching than are TRC teachers. This result is statistically significant at the 5% level and st able across all specifications of the experience cohorts. However, this discrepancy between t eachers with different types of certificates disappears as they earn at least 2 years of e xperience. For 2-5 year cohorts no statistically significant difference in exit rate is revealed. The magnitudes of the odds ratios for cohorts with 2 and 3 years of experience are also stable across all specifications. For experience cohorts with 4 and 5 years of experience these magnitudes are relatively higher for the diagonal specification. The higher magnitudes for the diagonal model might be caused by the relatively small number of observations for these experience cohorts in th is particular definiti on, which provides the 9 I also did similar analysis with 10 cohorts (1-10 years of experience), and the result holds the same.
56 smallest sample for this analysis. The magnitude of the effect is also higher for the 5th year cohort for the 2nd definition of the experience cohorts compared to definition 1. The 2nd definition is the one which provides the largest sample since it might include teachers with breaks in their service. The like lihood for teachers to have breaks increases as the number of years of experience increases. Therefore, some teachers in the 5 years experience cohort in the 2nd definition might have had a break and return ed to teaching 1-3 years ago. Hence, their behavior might be more similar to teachers in the cohorts 1-4. Therefore, this difference in magnitudes produces no serious doubts in the cons istency of results, espe cially since in all specifications this effect is not statistically si gnificant for experience cohorts with 2-5 years of experience. Hence the results of the logit estimation suggest that the vari ation in the exit rate across teachers with different types of certificates ex ists only for novice teachers with no experience, and inclusion of indicators for experience cohorts 2-5 does not add anything to the model. To confirm these results I also test the joint signific ance of this set of explanatory variables. Table A-3 of the appendix presents the results of th e Wald test. The hypothesis of insignificance cannot be rejected for cohorts with 2-5, 3-5, or 4-5 years of experience. Therefore, the test results also confirm the conclusion that the teacher attrition rate is different for ARC and TRC teachers only immediately after the first year of working in the school educati on system and not for subsequent years. After the second year their exit rates do not differ from each other. The marginal effects of coming to the pr ofession through ARCs rather than through TRCs on the probability of exiting the main assignme nt field are presented in Table 2-4 for four different social groups of teachers: white and non-white males and white and non-white females. ARC male teachers are 14-16% points more likely to change their main assignment field after
57 the first year of teaching than male TRC teachers. The effect is quite large in magnitude, taking into account that the probability to exit after the first year for TRC teachers is about 10-12%. For ARC women the effect is even larger; they are mo re likely to exit by 17-1 8% points compared to female TRC teachers. These numbers are somewh at larger than usual estimates of teacher attrition rates. However, the common approach to the definition of attrition is not based on exit from the main assignment field, as is done in this study. Going back to the statistics presented in Table 2-2, it becomes clear that actually the number of exits from the field is comparable to the number of exits from the school education system That is why my estim ates of exit rates are relatively higher. For other explanatory variables, the results are reasonable and consistent with previous research on the determinants of teacher attrition. Women are more likely to exit than men, but married females are less likely to leave the fielda result similar to the finding of Steinbrickner (1999). The effect of experience has a U-shaped form, as the probability of exit first decreases with experience and then rises again as the teacher approaches retirement age, with the critical point approximately at 19 years of experience. Gi ven that average entry age is 26, the critical point corresponds to age 45. As one might exp ect, a high proportion of low-income students increases attrition. The same effect was found in a number of previous studies10. Contrary to many studies focused on the e ffect of opportunity costs and salaries on teacher attrition, I have detected no effect of real salaries on teachers decisions to exit. However, most of the previous findings on this topic sugges t that salary and opportunity costs mostly affect transits between school districts rather than exits from the school system11. Since I do not distinguish between cases when the teacher transfer s to another district but continues to work in 10 See for example Dolton and Klaauw (1995); Boyd et al (2005). 11 See, for example, Imazeki, 2005; Hanushek et al ., 2001.
58 the same assignment field and a teacher who con tinues to work in the same field and same district, the transits might be c ounted as stays as well as exits in my study. Hence, it is not surprising that the effect wa s not confirmed by my model. Finally, I found substantial varia tion in attrition rates of te achers working in different subject fields. As expected, teachers working in Other fields like vocational studies, technical studies, etc. are 2.08-2.27 times more likely to exit than elementary school teachers (omitted category). Surprisingly, English t eachers demonstrate a higher rate of exit from the field than those specializing in mathematics and sciences, the fields experiencing the most severe shortage. However, the problem of shortage in mathematic s and science primarily relates to the attraction of newly qualified teachers rather than to the retention of teache rs with experience. Moreover, for mathematical and science teachers it is mo re difficult to change fields, since their professional preparation is specific and cannot be applied in other subjects. Specialists in English language arts may change to humanities with relativ e ease. These results are generally consistent with the findings of Murnane and Olson (1989). They found that high school teachers are more likely to leave teaching after a few years on the j ob than elementary teachers, and that English teachers have relatively shorter working spells than mathematics teachers. My results suggest that attrition of novice t eachers is quite different for graduates of different types of certificati on programs. Teachers that completed an ARC are 14-18% points more likely to exit after the first year in the profession than their colleagues who completed a TRC. The difference however disappears for teachers with at least 2 years of experience. These results have important implications for the pol icy making. Since ARC began to develop widely only since 1999, there are no avai lable detailed statistics for the career paths of ARC teachers. This makes it difficult to analyze the efficiency of ARC programs. However, according to the
59 results of this study, only the first year matters, after the first year ARC teachers have the same attrition patterns as TRC teachers. Another interesting question that arises from the results conc erns the reasons behind this dramatic difference in attrition rate after the first year. Do graduates of ARC exit because they are less prepared for teaching, or are individuals who decide to enter ARC different from participants in TRCs? In other words, are ther e attributes of the ARC that cause the higher attrition of novice teachers, or are individuals choosing these types of programs according to their personal attributes, which a ffect not only their program choice, but also the probability they will leave teaching or change subject field? Are ARC teachers less devoted to the profession or are they less prepared to face a classroom? Is there a self-selection of teachers by the types of programs? The next section discusses this possibility. Analysis of Possible Self-Selection Assuming that an individuals choice of certification programs might be based on personal qualities, which might also affect the decision to leave teach ing or to change field, it is important to check whether teachers that completed ARC are, in general, similar to the graduates of TRC or if they are somehow different. For these purposes, I analyze the social and demographical characteristics of teachers for the subsamples of ARC and TRC teachers, and look for any substantial difference in ch aracteristics between the two groups. Since the results of logit estimation showed a substantial difference in exit rates between ARC and TRC teachers only immediat ely after the first year of wo rking in the elementary and secondary education system, the most important question is whether teac hers with 1 year of experience have similar characteristics in both gr oups. Therefore, I first address this cohort of teachers. Table A-4 of the Appendix A presents summary statistics for the observable personal characteristics of novice teachers wi th 1 year of experience. Thes e characteristics include: age,
60 race, sex, educational attainment, competitiveness of the teachers undergraduate college, marital status, and family income and size. The next year follow-up data is not needed for this analysis because the whole SASS data set was used for the comparison of these characteristics across certification groups, thus significantly increasing the sample si ze. However, marital status and family characterizes are available only in the T FS, hence for these variables the sample size was relatively small. There are no dramatic differences in averag e personal characteristics across certification groups; however, there are still some noticeable di sparities in mean values of a few personal attributes. In particular, it seems the share of male teachers is higher among graduates of the ARCs, and also representatives of this group are more likely to have graduated from competitive undergraduate colleges, less likely to be married and on average they have lower family incomes. Simple t-tests were performed to check for statistically significant differences. I run OLS regressions for all personal characteristics with type of program indicat or as an explanatory variable: 0,iA l tiiPC Alt (2-15) where PCi is a personal characte ristic of individual i and Alti is an indicator of the type of certification program. If the null hypothesis Alt=0 cannot be rejected, th en there is no reason to believe that participants of ARCs and TRCs ar e different from one another. However, if the hypothesis can be rejected, then there is some ground for the self-selection hypothesis. I performed t-tests for two time periods, 1999-00 a nd 2003-04, separately and then a difference-indifference approach was applied for the pooled sa mple to check whether the differences between groups vary over the time 0,iA l tiiA l t Y e a riiiPC AltyearAltyearu (2-16)
61 where yeari is the indicator of the time period and AltYear shows whether the disparity between groups varies across time periods. Table A-5 of the appendix presents the result s of t-tests for the teachers with 1 year of experience (the set of personal ch aracteristics is similar to the one analyzed in Table A-4). The first four columns show Alt and corresponding t-statistics for the 2003-04 and 1999-00 periods respectively, and the last two columns report resu lts of the t-tests for difference-in-difference estimations ( AltYear) The results indicate that ARC teachers are, on average, 2 years older than graduates of TRCs, they are approximately 8-12% points more likely to be male, and they more frequently represent graduates of competitive colleges. The most controversial result is observed for the indicator of a teacher holding an a ssociates degree. In 1999-00, ARC teachers were about 8% more likely to possess such a degree, while in 2003-04 they were 11% less likely to have it. There is nothing surprising in age differences since teachers ente ring ARCs are usually those who did not choose teaching as a specialty right after high school in contrast to the teachers who completed TRCs. The higher share of male teachers might be explained by the fact that women are more likely to choose teaching as their main specialty and to enter TRCs while men might be less likely to consider this opportuni ty just after the school. In the regression estimation, discussed in the prev ious section, I have already included controls for age and sex and the interaction of sex and ma rital status, so I have already accounted for possible differences between ARC and TRC categories that might be cons equences of gender structure rather than the type of program. However I did not control for the competitiveness of the undergraduate college, while t-tests show that in the 2003-04 school year teachers in ARC category, in average, had attended better colleges than teachers in the traditional category. That might be caused by the
62 development of the TFA program in the 2000s that recruits graduates from the best colleges with 2 years contracts to work in elementary and se condary schools. This di sparity raises certain concerns about whether ARC and TRC novice teachers are of the sa me quality and whether they have the same job opportunities outside of the sc hool system. Hence, while overall the analysis shows few major differences between the two groups of teachers, I cannot be certain there is no self-selection. The same analysis is repeated regarding the personal characteristics of teachers with 1 to 5 years of experience (representatives of all 5 experience cohort s pooled together). While the comparison of novice teachers with 1 year of experience is of the main interest, taking into account the results of the logit estimations, this study is focused on the analysis of the turnover rates of teachers with less than 6 years of experience. The test ing of possible self-selection is prolonged to teachers belonging to all 5 cohorts that are under the scope of the analysis. The summary statistics and t-tests for teachers with less than 6 years of experience are presented in Appendix A in Table A-6 and Table A-7 respectively. The results for the pooled sample reveal even more disparities acro ss certification groups. ARC teachers appear to be 2 years older. They are less likely to be white and the difference in the racial composition between groups increases over time. They are also less likely to have bachelors and masters degrees, and they more frequently represent graduates of competitive colleges. After I compared two groups of teachers w ith different types of certificates, I also compared personal characteristics of the ARC t eachers belonging to the different experience cohorts. Since I found a higher exit rate only for A RC teachers with just 1 year of experience, then, if it is in fact the effect of personal characteristics rather th an the effect of the program, it is
63 natural to suppose that those graduates of ARC that are different from TRC teachers sort out after the first year. Exit rates for cohorts with more than 1year of expe rience do not vary across certification groups. Therefore releva nt discrepancies in personal ch aracteristics are not likely to be present after the sor ting out occurring after the first year. In other words, if there is selfselection to the programs that caus es a higher attrition after the fi rst year of teaching, then ARC representatives of the 1 year e xperience cohort must be different from ARC representatives from cohort 2. Tables in Appendix A (from Table A-8 through Table A-11) illustrate t-tests performed for pair-wise comparisons of two adjoining c ohorts of ARC teachers. Table A-8 shows the results of a comparison of cohorts 1 and 2. Ta ble A-9 compares cohorts 2 and 3, Table A-10 compares cohorts 3 and 4, and Table A-11 comp ares cohorts 4 and 5 respectively. Assuming there is self-selection of teacher s to programs, I expect to find the most significant difference between cohorts 1 and 2, which then decreases with years of experience. However, with few exceptions, ARC teachers with 1 year of experience appear to be similar to those with 2 years of experience. In 2003-2004, first year teachers were less likely to be married than teachers with 2 years of experien ce, and first year teachers were also less likely to be black in 2003-04. In general, comparisons of other experience c ohorts show the same picture: differences in marital status and family characteristics (which may be caused by changes in personal life, rather than by sorting out af ter first year); a difference in sex composition between cohorts 2 and 3, and cohorts 4 and 5 in 1999-00; and a difference in the average quality of undergraduate colleges for cohorts 3 and 4 (t eachers with 3 years of experience are more likely to have graduated from a competitive coll ege). The last finding rais es concerns about the difference in the quality of teach ers and their possible choice of jobs outside of the education field.
64 Surprisingly, the difference between ARC and TRC teachers appears to increase as experience increases. I expected th e opposite effect in the case of self-selection to programs. The same tendency was also revealed comparing th e experience of cohorts that graduated from ARCs. The differences between the cohorts are more pronounced for cohorts 2 and 3 and 3 and 4, rather than for the cohorts 1 and 2. These results suggest that wh ile individuals entering different types of programs are not really different from each other, the at trition patterns for these two groups might not be the same over time. For example, white teachers with ARCs might be more likely to exit main assignment fields duri ng their second or third year than white TRC teachers, but they might also be less likely to exit during the forth or fifth year. It is of interest to analyze whether the patterns of teacher attrition over time are different for the ARC and TRC groups. For these purposes I ra n a set of OLS estimati ons that includes the set of explanatory variables used in the model discussed in the pr evious section. I also included one additional personal characte ristic, its interaction with the experience dummy, its interaction with the certification type indicator, and the in teraction of characterist ic, experience cohort, and certification type. For example, for the analysis of effects of college quality on the attrition of teachers with one year of experience, Equation 2-10 estimates: 012345 561 ,ii i i ki kii i ki ii ikiiiPyXXAltExpExpAltcollegeExp collegeAltcollegeExpAltu (2-17) where Xi represents the vector of observable char acteristics (teacher white, teacher black, experience, experience squared, entry age, nonwhite students, non-white teachers, poor, number of schools, district enrollment, union, urban, special education, regular s econd, interaction term of special education and regular second, year). Alti is equal to 1 if the teacher i completed an ARC. Expki indicates whether teacher i belongs to cohort k Collegei is an index of the relative competitiveness of the undergraduate college attended by teacher i The main interest is in
65 coefficients 5 and 6. If 5 is not equal to zero, than the behavior of teachers with the same characteristics might be different for representa tives of the two certificat ion groups. Likewise, if 6 is different from zero, than the differences might be pronounced only for this experience cohort. Equation 2-10 was estimated for 15 different personal characteristics and for 5 different experience cohorts (each regres sion includes the indicator for one cohort). Due to the small sample size in this analysis, a nother definition of cohorts was us ed. The first set of regressions include the indicator of the 1 y ear of experience cohor t, the second set incl udes the indicator for teachers with less than 3 years of experience, the third set for teachers with less than 4 years of experience, the fourth set for those with less than 6 years of experience, and the final set includes an indicator for teachers with 6 to 10 years of experience. Thus, this particular analysis constitutes 75 regressions and I do no t report these tables in this study. According to the results of this analysis, th ere is a difference in exit rates for teachers with various level of experien ce, however within experience c ohorts there are no statistically significant differences across certification groups This means white teach ers with 1 year of experience might have a different attrition rate compared to white teachers with 2 years of experience, however white ARC teachers with 1 year of experience behave similarly to white TRC teachers from the same experience c ohort. More precisely, for coefficient 2, 3, in many cases, the null hypothesis ( =0) cannot be rejected, while coefficients 7 and 7 are not statistically different from zero in the majority of regressions. However, there are a few exceptions. The most sustainable effect was found in the family income variable. ARC teachers with high income levels are less likely to exit teach ing than TRC teachers belonging to the same income category. The coefficient 6 is negative and is signifi cant with a 5-10% level for all
66 experience cohorts in the regressions where family income is used as the main personal characteristic variable. However, there is no re ason to believe that this effect varies across experience cohorts for ARC teachers ( 7 is not statistically different from zero). Second, teachers who graduated from competiti ve colleges are more likely to exit their main assignment field if they completed an ARC. This effect was only re vealed in 1 regression specification, however I have already found dispar ities in the mean values of college quality regarding ARC and TRC certified t eachers in a previous analysis discussed in this section. Finally, coefficients 6 and 7 were occasionally found to be statistically significant in 1 or 2 out of 5 specifications for a few personal characteristics. Those characteristics include bachelor degree, never married, widowed-separat ed-divorced, educational specialist or some graduate study. Unfortunately, I do not have enough variation in th ese variables for ARC teachers, as well as in the bachelor degree vari able for TRC teachers. Hence, these occasional results might be driven by just a few observations. Hence, the effects of family income and colle ge quality raise the most concerns regarding the results presented in the previous section. Th erefore, I included these two variables and their interaction with ARC indicator into the model to check whether the difference in exit rates of teachers is driven by the difference in these personal characteristics or if they are driven by some particular feature of the certification program. Results with Controls for College Quality and Family Income Both personal characteristics in question, colle ge quality and family income, used in the previous analysis, are defined as an index, which varies from 1 to 5. This is the only available information for them. Both indexes are organized in descending order, where 1 corresponds to the least competitive colleges and the lowest level of income respectively. Before proceeding with the main analysis, I first tried different de finitions for these variab les based on the available
67 indexes. Those included the following approaches: a set of dummy variables for each category of the index, the index itself, and a different grouping of categories. Fi nally, I decided to stay with the set of dummy variables for each category of income respectively, and a dichotomous indicator for the college quality. These definitions provide the best fit for the model and there are no controversial effects within the groups when categories with the opposite effects are pooled together. Hence, in my model college quality take s two values: 1 and 0. It is equal to 1 if a teacher graduated from a relatively competitive college (the original index is equal to 3, 4, or 5), and it is equal to 0 if a teacher graduated from a less competitive college (the original index is equal to 1 or 2). First, only these variables were added to the model discussed earlier. The results are reported in Table A-12 in the Appendix A. A ll explanatory variables retained the same magnitudes and significance levels. For the variable s of main interest, the previous results hold and ARC teachers are more likely to exit a main assignment field after the first year. After that, the difference in the exit rate across certification groups dissip ates. Hence, the inclusion of additional variables does not signif icantly affect the model. The odds ratios and marginal effects also are very close to those found in the first spec ification. The analysis of the joint significance of the variables for experience cohorts presented in Table A-13 in Appendix A also confirms the previous findings (the variab les describing cohorts 2 through 5 are jointly insignificant). However, the analysis provided in the previ ous section shows these two characteristics could affect ARC and TRC teacher s differently. Thus, the interac tion terms for family income and college quality variables were added to the model. Since earlier I found a persistent effect of family income on ARC teachers likelihood to exit the main assignment field, which does not vary across years of experience, I included only the inte raction of this variable with the ARC
68 indicator. However, for college quality, the most pronounced effects were revealed for 1 and for less than 4 years of experience c ohorts. There is reason to believe th at these effects are not stable and vary depending on years of expe rience. Hence, for college quality, the interaction terms with years of experience and ARC indicator were included in the model as well as its interactions with dummies for experience cohorts a nd its interactions w ith ARC variables. This allowed me to check not only whether the effects of college quality on the probability to exit varied across experience cohorts, but also whether these e ffects are the same for ARC and TRC teachers belonging to the same experience cohort. Table A-14 of the Appendix A shows the result s of the logit model with the newly added interaction terms. All explanatory variables, which are not of part icular interest to this study, have almost the same magnitudes of hazard ratios and the same level of significance. Most of the family income dummies are insignificant a nd are not quite stable across specifications. Interestingly, ARC teachers with a high level of family income are less likely to exit than TRC teachers. The possible explanation for this eff ect might be that for ARC teachers with high family income, the main reason for entering teac hing did not involve inco me, but rather having an occupation they like. On the other hand, T RC teachers, who chose this specialty when they were an undergraduate in college and do not really enjoy the prof ession, might feel free to quit teaching for some other occupation if they are no t constrained by the fear of losing their salary. The college quality dummy is also statistica lly insignificant in all specifications, except for a flag used for missing observations for this variable. I looked at these missing observations more precisely to understand what exact effect th is flag variable catche s. Most of the missing observations are actually caused by two reasons. The first category of missing college quality information is when a college is not included in the classification used for the construction of the
69 index, or if the college no longer exists. Th e second category of missing observations includes teachers who graduated from foreign college s. There are also other types of missing observations, like incomplete info rmation provided by the respondent but these cases constitute a small portion of the total number of missing obser vations compared to the first two categories. I tried to assign missing observations to these three categories and to r un a regression with 3 flags instead of 1. The results confirm that a negative effect is driven by teachers from the first two categories and these categories are likely to represent the low quality of a college. Colleges not listed in the index are probably of a lower qua lity than colleges with the grade 1. Teachers that graduated from foreign colleges might experience language difficulties. Since our main focus is on the effect of the certification progr ams on the exits, I have pooled all the missing cases together. While explanatory variables retain almost the same magnitudes of the coefficients and the same level of significance, one can observe that re sults for the variables of main interest have now changed. The ARC indicator an d its interaction with college quality are now significant at a level of 10% in two specifications out of three. The hazard ratio for the interaction term is higher than 1 and for the ARC dummy it is less than 1. Therefore, ARC teachers who graduated from a low-quality college are less likely to exit a main assignment field than graduates from competitive colleges holding the same type of certificate. One can also observe that these effects are not stable across different experience cohorts. All teachers are less likely to exit after their second year independent of the t ype of certificate. However, the attrition of the ARC graduates from competitive colleges varies significantly acro ss the first 3 years of experience. Particularly after their third year teachers are much more likely to quit teaching than those who graduated from less competitive colleges. These results make sense: teachers with a better undergraduate
70 preparation are less likely to quit because they do not feel they are incompetent for the job (the factor that causes attrition during first the year s of teaching), however with better education they have more chances to find a better job outside of secondary and elementary education or to move to administrative positions inside the education system (they are more likely to exit their main assignment field after their third year). For the graduates from less competitive colleges, the results have the same pattern as before: they are more likely to exit a main assignmen t field after the first year but as they gain at least 2 years of experience all differences in at trition rates between them and graduates of TRC belonging to the same cohort disappear. There is a statistically signi ficant effect for the experience cohort 4 in the diagona l specification of experience (that provides the smallest sample). However, this result does not hold in more preferable specifications. While previous results have shown no statistically significant effects for cohorts 2 and 3, their standard errors were relatively high; with newly added variables the standard errors for these cohorts decreased dramatically, clearly indicating that no significan t difference in attrition rates exists between teachers with different types of certificate belong ing to the same cohort (except for effects for ARC graduates from competitive colleges that I discussed above). Table 1-5 reports the odds ratios for ARC t eachers with less than 6 years of experience for the model with new variables added12. The ACR graduates of a competitive college are 1.133.39 times more likely to exit a main assignment field just after the first year than TRC graduates of competitive colleges. The relative chances of exiting for the graduates of less competitive colleges are almost the same: they vary from 1.08 to 3.41 for teachers with 1 year of experience. For all ARC teachers the lowest relative chances of exiting are associated with the highest level 12 I also did similar analysis with 10 cohorts (1-10 years of experience) and result holds the same.
71 of family income. The highest attrition probability is the attribute of the middle categories of family income which belong to the interval from 35 thousand to 100 thousands. For the graduates of competitive colleges, the probability of exiting a main assignment field is higher after the third year (0.98-4.04) than after the firs t year (1.13-3.39). This effect is statistically significant in all specifications. Table 1-6 presents the marginal effects for A RC teachers. The results are calculated using mean values of family income variables. Af ter the first year, male ARC teachers are 12-13% points more likely to exit a main assignment fi eld than male TRC teachers. The same category females are 14-16% points more lik ely to exit than their collea gues who came to the profession through TRC. The exit rates both for men and wo men who graduated from competitive colleges are higher for the 3 year expe rience cohort (11-16% and 13-19% re spectively). For the graduates of less competitive colleges, the exit rates are quite close to those from competitive colleges, except for the 3 year experience cohort. After the 3rd year ARC teachers from less competitive colleges are less likely to exit th e main assignment field than TRC teachers. However, this effect was not statistically significant in any specification. Then I proceed with the analysis of the jo int significance of the experience cohorts indicators as I did for the previous models. Ta ble A-15 in Appendix A shows the results of the Wald test for joint significance for all variables related to experience cohorts and ARC preparation. With college quality controls the hypothesis of joint insignificance cannot be rejected for cohorts with 3-5 y ears of experience tested together The value of the F-statistics increases dramatically for the cohort s 4-5 tested without cohorts 1-3. Generally, the new model provides the same results as the model discussed previously. With only one exception, the attrition rates of teachers with different types of certificates are not
72 statistically different from each other for the experience cohorts 2-5, while there is substantial disparity in exit rates of teachers with only 1 year of experien ce. This disparity is extremely high for graduates of the less competitive colleges. While it is smaller in magnitude for the graduates of the competitive colleg es, it is still strongly statistically significant. The only exception from this attrition pattern is the high exit rate of graduates from competitive colleges after the third year. While generally this result makes intuitiv e sense, some questions concerning that effect arise. It is not clear why it is present only fo r ARC teachers, as graduates from the competitive colleges might have more job opportunities outside of teaching independent of the type of the certificates they hold. One possible explanation for this difference could be that ARC teachers who had chosen another specia lization in the undergraduate school might have in their possession other diplomas. Therefore, they might ha ve higher chances to find jobs outside of the field of education and they also might have stronger motivation to look for another type of job. However, I offer another explanation of this effect. Since the beginning of the 2000s, an ARC program TFA was rapidly deve loping. This program recruits the graduates of the most competitive colleges for teaching in elementary and secondary schools. Participants in this program are committed to work 2 years as school t eachers. The high attriti on rates after the third year for the ARC teachers who graduated from highly competitive colleges could be caused by the high exit rates of the participants of this particular nation-wide program. To check this hypothesis I run two separate logit regressions for 1999-00 and for 2003-04 subsamples. Since the program was not well developed before 2000, it could not have affected attrition patterns in 1999-00. The effect was confirmed for 2003-04, while no significant results were found for the 1999-00 sample. Thus, the estimation results probably ha ve revealed the effect of this particular program. However, there is still one concern. Ac cording to the program terms participants sign
73 up only for 2 years of work in the school system. Therefore, one would exp ect high rates of exits after their second year, rather than after the third, while our results confirm it only for the teachers with 3 years of experience. No statis tically significant effect was found for the ARC teachers with 2 years of experience who gradua ted from competitive colleges. To understand why this could be the case I addressed recent st udies focused on the analysis of this program. Boyd et al. (2006) compare the attrition rate s of novice teachers who entered teaching in the NYC through ARC and TRC. They focus on three types of ARC pr ograms: TFA, NYCTF, and Other, which include all other teachers whos e preparation does not fit other pathways. Table 2-7 shows the descriptive statistics characterizing the attrition rates that come from Boyd et al. (2006). College recommended and Individual evaluati on categories represent TRC. TFA is an alternative program recruiting grad uates of the most competitive colleges to work 2 years as a teachers. NYCTF is a NYC program that target s mid-career professionals as well as recent college graduates. The other category includes teachers whose preparation program does not fit into the other five categories. Hence, TFA teach er are ARC teachers who are considered to be graduates of competitive colleges in this analysis NYCTF graduates are also likely to belong to this category. On the contrary, the Other category probably in cludes ARC teachers who graduated from less competitive colleges. The descriptive statistics for NYC teachers confirm our findings. ARC teachers from less competitive colleges have the highest exit rate after their first year, and the rate decreases with experience earned. ARC graduates of competitive colle ges are more likely to exit after their third year (both TFA and NYCTF have the highest exit ra tes after their third years). Their exit rate is also high after the second and four th years but the peak occurs after the third year. The difference
74 in exit rates between TRC and ARC teachers is less pronounced for cohorts with 2 years of experience than for cohorts with 3 years of experience. This explains why I found a significant effect after the third year and no significant eff ect after the second. The differences in exit rates for Other, presumably, the category for the ARC teachers who graduated from less competitive colleges, and TFA are also most pronounced after the third year. Hence, our findings confirm the same pattern of teachers attrition revealed in Boyd et al. (2006). Other studies of the TFA program confirm th at the majority of its participants do not exit after the second year. Decker, Mayer, a nd Glazerman (2004) analyzed the retention of program participants for several regions (Bal timore, Chicago, Los Angeles, Houston, New Orleans and Mississippi Delta). Th ey argue that about 34% of the program alumni do not quit teaching after the second year of the program a nd except for them 25% continue to work in the field of education. There are also a couple studies regarding TFA effectiveness in Houston based on data from 1996-2001. Raymond, Fletcher and Luque ( 2001) do not distinguish between exiting from teaching and transferring to another school. Howeve r, for both transfers an d exits they found that significant proportions of TFA participants (a bout 40-20%) have stayed in teaching beyond 2 years, with the exception of 1998. Darling-Hammond et al. (2005) use the same data set and also report exits and transfers pooled together. Their results show that between 57% and 90% of TFA recruits leave teaching in Houston after their s econd year and between 72% and 100% of recruits leave after their third year. Since these studies do not distinguish between transfers and exits, it is difficult to interpret their data relative to my findings. Moreover, these studies are based on a relatively small data set (about 20 teachers in each experience cohort) for the 1996-2000 period when the TFA program was not yet well developed. Hence, I consider the statistics presented in
75 Boyd et al. (2006) as more representative rega rding this program. Th ey analyze about 100-300 observations per cohort, cover the 2001-2004 peri od, and their definition of exiting is more comparable to mine. In the next section I apply estimated attrition rates for ARC and TRC teachers to estimate the relative efficiency of ARCs in solving the teacher shortage problem. Simulation of Efficiency of Al ternative Certification Programs Using empirical results from the previous sections, it is now possible to estimate the simple simulation model presented earlier in this chapter. This simulation uses Equations 2-1, 22, 2-5, 2-6, and 2-7. Assume that wRF=40,000, N =1000, and NSF=240. For simplicity, the retention rate after the first year is considered to be 100% both for TRC and ARC teachers. Since there is no difference in their attrition rates if th ey stay in teaching for at least 2 years, it is irrelevant for the comparison purposes to account fo r their attrition rates in consequent years. The present value of future wage payments is calculated for a 5 year period, assuming the discount rate is 5%. Also assume that the re lative loss in productivity due to hiring an inexperienced teacher is 20% of the MPL ( =0.2)13, the productivity of a novice ARC teacher is approximately 20% lower than the productivity of a novice TRC teacher ( =0.2), and the MPL is equal to the equilibrium wage. Table 2-8 shows results of th e model simulation applying diffe rent elasticities of supply ( ES). Manski (1987) estimated the elasticity of teacher supply using data from 1972. According to his estimation, the elasticity of teacher supply varies between 2.4 and 3.2. However, the elasticity of supply could have changed since 19 72, because the composition of the teacher labor force, as well as economic and social characte ristics of the society ha ve changed since that. Hence, I use different elasticities varying from 2 to 10. 13 The negative effect of lack of experi ence on the students outcomes was esti mated as roughly 50% of the standard deviation of teacher quality in Hanushek et al (2005), which corresponds to a 19% efficiency loss.
76 The results of the simulations show that depending on the elastic ity of the supply the upper bound of ARC training costs that makes ARCs competitive with TRCs varies from $10,000 per ARC teacher (10 S SFE ) to $99,000 (2 S SFE ). For this simulation, it was assumed that a district has 1000 teachers with less than 6 years of experience. The bigger is the district, the more efficient ARC programs would be becaus e the second component of marginal costs for the wage growth approach ( wN)/ NSF increases as the size of di strict increases. The second component of ARC marginal costs ( CARC) decreases as the district size increases due to economies of scale for ARC programs. The NCAC pr ovides some data for the cost of alternative teacher certification programs across United States. Table 2-9 briefly summarizes their information. Assuming the costs estimation of the NCAC is a rough proxy for CARC, the ARC programs is an efficient way to solve shortages problems despite productivity losses and even in the case of high elasticity of supply. This analysis is based on an important assump tion, that after a few years of teaching there in no difference in the quality of teachers who came to the profession through different pathways. This assumption is based on the findi ng of Hanushek, Kain, OBrien, and Rivkin (2005) that a teachers productivity increases in the first years of teach ing and then remains stable until retirement. However, it is possible th at the quality of an AR C teacher will differ from the quality of a TRC teacher even after first y ears of teaching. Why can their quality remain different and how would it affect the results of simulation? If the assumption does not hold my estimation of the maximum training cost that a llows an ARC to be efficient may be either overestimated or underestimated. The maximum effective CARC is overestimated if the quality of average ARC teacher is always below the quality of an average TRC teacher with the same level of experience. It can happen if experience gained over years of teaching does not compensate for
77 the lack of the initial ARC t eacher preparation in children ps ychology, methodology of teaching, and other courses that are studied more pr ofoundly by TRC teachers. Moreover, ARC teachers may be less devoted to the profession assuming th eir choice of the career of teacher was not their first best choice and they would ra ther prefer to work in some other area, but are not able to pursue it due to personal reasons. That may reduce the productivity of ARC teachers compared to TRC teachers with the same experience. In these cases Equation 2-7 must be corrected to include all future productivity losses caused by hi ring an ARC teacher instead of a TRC teacher. The opposite situation takes place if the quality of an average ARC teacher is higher than the quality of an average TRC teacher. If ARC t eachers are on average more talented or have better undergraduate educations (the descriptive statistics in Table A-4 a nd Table A-6 show that on average ARC teachers graduate from more co mpetitive colleges than TRC teachers) they may become more productive than TRC teachers afte r they gain experience. In this case the maximum effective CARC is underestimated, and Equation 2-7 should be corrected to include all future gains in productivity caused by hiring an ARC teacher instead of a TRC teacher. Let i denote the coefficient of relative loss or gain in the productivity induced by substitution of one TRC teacher by an ARC t eacher with the same level of experience i Then the marginal input costs of preparation of an ARC teacher are 1 21 ,...,,, Attr RFARC ARCRFN ARCARCARCMPL wC M IC PVw PVMPL RRR (2-18) where PV( 2,, N, MPL ) shows the present value of losses or gains in productivity for an ARC teacher who works in the school education system for N years, starting from his second year in teaching and up to his last year of work. Howe ver, since ARCs began to develop rapidly only during the last decade no research is availa ble concerning the relative productivity of ARC
78 teachers during their whole careers. Existing stud ies analyze ARC teachers efficiency in the beginning of their careers and use relatively sm all samples. Boyd et al. (2006) found small differences in student achievement that can be attributed to preparation pathways only among first-year teachers. Darling-Ha mmond et al. (2005) found that hiring a novice TFA teacher has both negative and positive effects on students sc ores. Therefore, until more study is done in this area I believe that assumption that quality of experienced ARC and TRC teachers is equal is reasonable and I use it for ARC efficiency simulation in my analysis. Conclusions The recent development of ARCs has raised a number of questions concerning their relative efficiency compared to TRCs. I address one of these questions, whether attrition patterns are different for TRC and ARC teachers and if th ese differences disappear as a cohort matures. My findings confirm that teachers who come into the profession via ARCs are in fact more likely to exit their main assignment field after the first year. The graduates of competitive colleges w ho entered teaching through ARC are 14-18% points more likely to exit a main assignment field after their first year than teachers who completed TRCs and also graduated from compe titive colleges. Hence, there is a significant difference in attrition rates immediately after the first year between TRC and ARC teachers. However, this disparity disappear s after the first year. As teachers gain at least 2 years of experience, the exit rates of ARC and TRC t eachers become the same. There is only one exception for this pattern. TRC graduates of competitive colleges are 11-19% points more likely to exit after the third year of teaching than TRC teachers belonging to the same category. I found some evidence that this effect is driven by the high exit rate of the participants of the nation-wide alternative program TFA that recruits the gradua tes of the best colleges to work 2 years as
79 teachers. The majority of the participants of th is program do not stay in teaching longer than 3 years. Usually, the analysis of the attrition of AR C teachers is complicated because there is a lack of data that follows the career paths of the ARC participants after they complete the program. Taking into the account my findings, it is enough to have statistics of ARC graduates attrition after the first year to predict their attri tion in later years. Since the main disparities in exit rates of ARC and TRC teachers occurs just after the first year, th e attrition of the ARC teachers for the subsequent years could be pr edicted based on the attrition pattern of TRC teachers with the same social and demographic characteristics. There is much more data available for TRC teachers that tracks their career paths over time. Hence, a researcher needs only to collect data for the ARC participants care er paths for 1 year after the completion of the program, to facilitate a cost-benefit analysis of such programs. My findings also raise another interesting question: what is the cause of the significant difference in exit rates after th e first year? Are ARC teachers less committed to the profession? Do they only view it as a temporary occupation until they find another job? Or are they less prepared to face a class than TRC teachers? T RC require more intensive in-class practice and courses on pedagogy and classroom management which help novice teachers adopt quickly to new obligations. The graduates of ARCs might be less prepared for this type of the work and might be stunned and confused during their first year of teaching. The second explanation seems reasonable, since the difference in attrition rate s disappears as teachers gain more experience. However, further research is needed to answer this question more accurately. If the lack of practice in class causes a higher at trition rate of novice teachers, than emphasizing this part of preparation should be considered fo r the improvement of ARC programs.
80 Table 2-1. Number of observations in e xperience cohorts formed by three different approaches 1999-00 2003-04 Experience cohorts Regular Alternative Regular Alternative Total Diagonal definition 1 year 155 29 97 34 315 2 years 146 32 95 35 308 3 years 106 32 101 40 279 4 years 67 20 77 20 184 5 years 67 13 80 31 191 Whole sample 1981 453 2126 664 5224 Definition 1 1 year 162 31 107 36 336 2 years 164 39 129 40 372 3 years 122 38 128 50 338 4 years 80 24 117 36 257 5 years 79 18 111 36 244 Whole sample 2089 494 2390 739 5712 Definition 2 1 year 162 31 107 36 392 2 years 183 49 166 50 508 3 years 133 40 140 49 389 4 years 89 30 121 34 292 5 years 90 18 128 47 294 Whole sample 2197 527 2527 769 6267
81 Table 2-2. Distribution of different types of exit by experience cohorts and types of certificates Stay in same field Change field Quit teaching Quit school system Cohort Reg. Alt. Reg. Alt. Reg. Alt. Reg. Alt. 1 year (weighted) 199 (89.49) 38 (70.30) 35 (8.78) 7 (18.70) 3 (0.18) 5 (5.20) 13 (1.55) 4 (5.80) 2 years (weighted) 212 (91.30) 54 (85.90) 30 (7.20) 12 (10.50) 6 (0.50) 0 (0.20) 16 (1.00) 7 (3.40) 3 years (weighted) 160 (83.00) 56 (82.00) 40 (11.00) 9 (14.00) 8 (0.50) 4 (1.20) 23 (5.50) 7 (2.80) 4 years (weighted) 125 (83.50) 37 (83.80) 18 (12.70) 8 (14.30) 9 (2.00) 2 (0.50) 17 (1.80) 7 (1.40) 5 years (weighted) 115 (78.00) 30 (82.00) 22 (11.00) 2 (0.30) 9 (4.00) 6 (1.20) 17 (7.00) 6 (16.50 ) Total (weighted) 811 (85.00) 215 (81.30) 145 (10.00) 38 (11.90) 35 (1.50) 17 (1.90) 86 (3.50) 31 (5.00) Weighted shares of cohorts with given type of certificate in parentheses. (The weighted shares in the same row and from same certification group add up to 100%. For example, for t eachers with 1 year of experience who hold regular certificates: 89.49% + 8.78 % + 0.18% + 1.55% = 100%.) Table 2-3. Odds ratios for interactions of ARC indicator with experience dummies Odds ratio Experience cohort Diagonal Def.1 Def.2 1 year 2.70** 2.73** 2.74** 2 years 0.86 1.11 1.02 3 years 1.49 1.33 1.25 4 years 2.41 1.07 0.83 5 years 1.37 0.85 1.31 *, **, *** statistically significant at 10, 5, and 1% level respectively.
82 Table 2-4. Marginal effects for results of the logit regression estimation Experience cohort Diagonal Def.1 Def.2 Male, White, Average state 1 year 14% 14% 16% 2 years -1% 1% 0% 3 years 5% 4% 3% 4 years 13% 1% -3% 5 years 4% -2% 4% Female, White, Average state 1 year 17% 17% 18% 2 years -1% 1% 0% 3 years 6% 5% 3% 4 years 16% 1% -3% 5 years 6% -3% 5% Male, Non-White, Average state 1 year 14% 14% 15% 2 years -1% 1% 0% 3 years 5% 4% 2% 4 years 14% 1% -3% 5 years 5% -2% 4% Female, Non-White, Average state 1 year 18% 17% 17% 2 years -1% 1% 0% 3 years 6% 5% 3% 4 years 16% 1% -3% 5 years 6% -3% 4%
83 Table 2-5. Odds ratios for interactions of ARC indicator with experience dummies with new variables added Odds ratio Diagonal Def.1 Def.2 Cohort Competitive college Less competitive college Competitive college Less competitive college Competitive college Less competitive college Family income < 35 thousand 1 year 2.63* 2.65* 2.90** 2.78* 2.70 2.44 2 years 0.99 0.68 1.37 1.07 1.32 0.59 3 years 3.13** 0.17 2.51* 0.19 2.67** 0.21 4 years 2.24 2.29 1.14 1.07 0.59 1.18 5 years 1.99 0.78 1.39 0.51 1.49 1.01 35 thousand Family income < 50 thousand 1 year 2.35* 2.36* 2.39** 2.29* 2.86 2.59 2 years 0.88 0.60 1.13 0.88 1.39 0.63 3 years 2.80** 0.15 2.07* 0.16 2.83** 0.22 4 years 2.00 2.04 0.94 0.88 0.62 1.25 5 years 1.78 0.69 1.14 0.42 1.58 1.07 50 thousand Family income < 75 thousand 1 year 3.39* 3.41* 3.12** 2.99* 3.10 2.80 2 years 1.27 0.87 1.47 1.15 1.51 0.68 3 years 4.04** 0.22 2.70* 0.21 3.07** 0.24 4 years 2.89 2.95 1.23 1.15 0.67 1.35 5 years 2.57 1.00 1.49 0.55 1.71 1.16 75 thousand Family income < 100 thousand 1 year 3.20* 3.22 2.87** 2.75* 2.83 2.56 2 years 1.20 0.82 1.36 1.06 1.38 0.62 3 years 3.81** 0.21 2.49* 0.19 2.80** 0.22 4 years 2.73 2.78 1.13 1.06 0.61 1.23 5 years 2.43 0.94 1.37 0.50 1.56 1.06 Family income 100 thousand 1 year 1.23* 1.24 1.13** 1.08* 1.18 1.06 2 years 0.46 0.32 0.53 0.42 0.57 0.26 3 years 1.47** 0.08 0.98* 0.08 1.16** 0.09 4 years 1.05 1.07 0.44 0.42 0.26 0.51 5 years 0.94 0.36 0.54 0.20 0.65 0.44 *, **, *** statistically significant at 10, 5, and 1% level respectively.
84 Table 2-6. Marginal effects for results of the lo git regression estimation with the new variables added Experience cohort Diagonal Def.1 Def.2 Competitive college, Male, Average state, Average income 1 year 13% 12% 13% 2 years 0% 1% 2% 3 years 16% 11% 12% 4 years 12% 0% -7% 5 years 12% 3% 5% Competitive college, Female, Average state, Average income 1 year 16% 14% 15% 2 years 0% 2% 3% 3 years 19% 13% 14% 4 years 15% 0% -8% 5 years 14% 4% 6% Less competitive college, Male, Average state, Average income 1 year 14% 12% 13% 2 years -1% 0% -4% 3 years -12% -13% -10% 4 years 11% -1% 2% 5 years -3% -10% 0% Less competitive college, Female, Average state, Average income 1 year 17% 15% 15% 2 years -2% -1% -5% 3 years -16% -16% -12% 4 years 14% -1% 2% 5 years -3% -12% 0% Table 2-7. Cumulative teacher attrition rates a nd exit rates by pathway for elementary, middle, and junior high school teachers in NYC, 2000-2004 Experience College recommended Individual evaluation NYCTF TFA Temporary license Other Cumulative Teacher Attrition Rates 1 year 0.115 0.139 0.105 0.107 0.184 0.264 2 years 0.212 0.256 0.278 0.477 0.300 0.402 3 years 0.290 0.322 0.434 0.727 0.413 0.500 4 years 0.368 0.391 0.544 0.850 0.501 0.573 Exit rates 1 year 12% 14% 11% 11% 18% 26% 2 years 11% 14% 19% 41% 14% 19% 3 years 10% 9% 22% 48% 16% 16% 4 years 11% 10% 19% 45% 15% 15% (Source: Boyd et al., 2006, p.23)
85 Table 2-8. Simulation of ARC efficiency Variable 10SE 5SE 3SE 2SE w (thousands) 40 40 40 40 N 1000 1000 1000 1000 ()SFRFNw 200 200 200 200 ()SFRFNw 240 240 240 240 SFN 40 40 40 40 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 ARC R 0.75 0.75 0.75 0.75 TRC R 0.90 0.90 0.90 0.90 discount rate 0.05 0.05 0.05 0.05 RFPVw (thousands) 142 142 142 142 SFPVw (thousands) 145 149 152 156 SFw 41 42 43 44 0ARCARCMICC 182 182 182 182 *SFMICww 211 241 270 300 *ARC ARCSFCMICMICww 29 59 88 118 0Attr ARCARCMICC 198 202 205 209 *Attr SFMICww 216 246 275 305 *AttrAttr Attr ARCARC SFCMICMICww 17 43 69 96 0Attr ARCARCMICC 215 215 215 215 *Attr SF M ICww 224 255 284 314 *AttrAttr Attr ARCARC SFCMICMICww 10 40 69 99
86 Table 2-9. Cost of ARC alternat ive teacher certification programs State How much doest it co st to participate in your states alternative certification program Arkansas $800 per year. Plus tr ansportation, lodging, etc. California $0-$13,000 Colorado $2,000-$7,500 Connecticut $3,200 plus books and supplies Delaware About $3,000 tuition District of Columbia $11,000-$13,000 Florida From 0 to $2,000+ Maryland $600-$3,000 Michigan $9,000-$12,000 New Hampshire $180-$300 Oklahoma $575 Texas $3,000-$5,000 Wyoming $645 (Source: http://www.teach-now.org/Table7.pdf Last acces sed September, 2009).
87 Figure 2-1. District labor market for novice teachers. A) RF equilibrium, B) SF equilibrium. Figure 2-2. Additional supply of novice teachers in SF throug h ARC shifts district SF supply to the right. wage teachers N S FNSF (wRF) DSF SSF B A wRF wSF wage teachers NRF DRF SRF wage teachers NSF (wRF) DSF SSF wRF wSF SSF+SARC
88 Figure 2-3. Districts labor market for novice teachers. 0% 20% 40% 60% 80% 100% 120%Element a ry Engl i sh M ath N atu r al Sci e n c es F oreig n l ang u ag es S p ec i a l Edu c a t i o n Social sciences E SL V oc a tio n a l a n d other A rt s an d o t h e r% 1999-00 2003-04Figure 2-4. Share of teachers certified in main assignment field in 1999-00 and 2003-04 by field groups (the fu ll SASS samples of public school teachers for the 1999-00 and 2003-04 school years are used). D ARCC wRFwSF wage teachers MIC ( wRF wSF ) MICARC ( CARC=0 ) S N+ NSF SARC N
89 0 10000 20000 30000 40000 50000 60000 7000019 8 58 6 1987-88 1 98 9 -9 0 19 9 19 2 19 9 39 4 1995-96 1 99 7 -9 8 1 99 9 -0 0 20 0 10 2 2003-04 2005-06Figure 2-5. Number of indi viduals issued certificates through ARC by year (Source: http://www.teach-now.org/intro.cfm. Last accessed Septem ber, 2009). Year first time teaching (2003-04 SASS) TE 2004 2003 2002 2001 2000 1999 1 1 131 33 3 1 0 2 0 15 131 38 4 0 3 0 0 8 143 41 2 4 0 0 1 13 98 31 5 0 0 0 2 15 111 Figure 2-6. Experience matrix. TE shows years of teachers experience. For example, 33 teachers in the 2003-04 SASS reported that they have 1 year of e xperience and that they began teaching in 2002.
90 Year first time teaching (2003-04 SASS) TE 2004 2003 2002 2001 2000 1999 1 1 2 2 3 3 4 4 5 5 Figure 2-7. Diagonal defin ition of experience cohorts. Year first time teaching (2003-04 SASS) Total experience 2004 2003 2002 2001 2000 1999 1 1 FTE=1 2 FTE=1 FTE=2 2 FTE=2 3 FTE=2 FTE=3 3 FTE=3 4 FTE=3 FTE=4 4 FTE=4 5 F T E = 4 FTE=5 5 Figure 2-8. Definition 1. The diagram shows that teachers whose answers to 2003-04 SASS belong to the cells adjoini ng to diagonal cells are redi stributed to the closest diagonal cells according to their answers on the questions regarding full-time and part-time experience.
91 Year first time teaching (2003-04 SASS) Total experience 2004 2003 2002 2001 2000 1999 1 1 FTE=1 2 FTE=1 FTE=2 2 2 2 2 3 FTE=2 FTE=3 3 3 3 4 FTE=3 FTE=4 4 4 5 F T E = 4 FTE=5 5 Figure 2-9. Definition 2. The diagram shows that teachers whose answers to 2003-04 SASS belong to the cells adjoining to diagonal cells or lay in th e upper-right corner of the matrix are redistributed to the diagonal cells according to t eachers answers on the questions regarding full-time and part-time experience. All teachers who reported that have the same experience as in the di agonal cell in the same row of the matrix and who also reported they began teaching earlier than teachers whose answers belong to diagonal cell at this row, are assumed to have the same experience as teachers whose answers bel ong to the diagonal cell.
92 CHAPTER 3 THE IMPACT OF SCHOOL ACCOUNTABILITY ON TEACHER QUALITY Introduction For over a decade, the main education reform in the United States has been focused on efforts to improve the efficiency of school level educati on by imposing assessment-based accountability systems. These systems are based on the ranks or grad es assigned to schools according to testing results of their students. Th e goal of the reform is to induce competition between schools and create incentives for them to increase efficiency. A majority of states developed some form of an accountability sy stem in the period 1993 to 2000. These systems vary significantly across states; th ey utilize different mechanisms and have different targets. By 1999, almost every state had impl emented state-mandated tests, as well as some form of assistance for low-performing schools (LPSs). Some states also introduced various systems of sanctions and rewards for teachers, schools, a nd districts based on a sc hools performance. While all the policies are focused primarily on student improvement, some may also affect, in both positive and negative ways, the supply of and demand for teachers. School rankings and systems of assistan ce and punishment for LPSs and school districts may encourage administrators to apply more effort to hire and attain qualified teacher s, thus affecting the demand for teachers. On the other hand, these po licies may change the work environment of teachers, their salaries, and pose additional re quirements. They may limit a teachers class organizing flexibility thus affecting their willin gness to work, or their supply. These possible, and controversial, effects raise a number of questions concerning the impact of accountability systems on the teacher labor market. Do teachers avoid the pressure of accountability and flee from LPSs to high-performing schools (HPSs) or to other occupations outs ide of education? Are teachers working in various subject fields aff ected differently? Do di stricts and schools apply
93 more effort? Have they been able to fill positi ons in LPSs with qualifie d teachers? What effects prevail? In order to estimate the direct and indirect effects of school acc ountability on teachers, one has to understand the mechanisms of accountabi lity policies. Do they affect all teachers or only particular subgroups? For ex ample, some policies like sanc tions and assistance for schools, which are assigned according to school performan ce, are more likely to affect teachers working in LPSs. Other policies mostly affect teachers working in particular subject fields. Almost all states have introduced testing in reading, writing and mathematics, while some of them also have tests in other fields. Certainly, teachers working in high-stakes fields whose testing results are used for school evaluation are subject to higher pressure from accountabili ty policies than those teaching in low-stakes fields. The determinants of teacher supply also va ry across subject fields, which may cause different responses from teachers working in di fferent fields to accountability policies. The private sector offers more j ob opportunities and higher wages to those who have degrees in mathematics or science. Consequently, the shortage of qualified teachers in these fields is more critical than in language arts, social sciences, and othe r fields. Hence, the d ecision to work as a teacher is driven by different factors in high-shorta ge and low-shortage fiel ds. It is likely that teachers working in high-shortage fields are less sensitive to sc hool and district policies than teachers working in low-shortage fields, since they chose teaching to alternative occupations with higher salaries. Taking into account the variet y of accountability policies and possible differences regarding their impact on teacher supply and de mand, it is important to analyze policies separately and to distinguish their effects on teachers working in different types of schools and
94 different subject fields. This chapter focuses on the analysis of the impact of different accountability policies on teacher quality, more pa rticularly I examine th e likelihood of a teacher to be certified in his or her main assignment fiel d; the indicator reflects whether a teacher is well prepared to perform his of her teaching assignments. Previous Literature A number of recent studies analyzed the e ffect of accountability reforms on student outcomes, while only a few have addressed the po tential impact of the accountability systems on the teacher labor market. Interview and survey research confirms that teachers do in fact feel the pressure of accountability (Barksdale-Ladd, & Karen, 2000). Teachers in states with stronger accountability policies feel they ha ve less influence on setting perf ormance standards. They also worry about the security of th eir job because of the performa nce of their students (Loeb, & Estrada, 2003). They view the high-stakes tests as an invasi on of their classrooms (Luna, & Turner, 2001). Three studies have addressed the impact of accountability on the teachers career decisions and turnover rates. Clotfelter, La dd, Vigdor, and Diaz (2004) and Boyd, Hamilton, Loeb, and Diaz (2008) utilize panel data sets fo r North Carolina and New York respectively, while Loeb and Estrada (2003) use a national-representative data se t. The main issues addressed in these three papers include co mparing of turnover rates between testing and non-testing grades in preand post-accountability periods, and estimating of th e likelihood teachers leave LPSs compared to other schools. The findings are some what controversial. Cl otfelter et al. (2004) found that in the post-accountability period teachers are more likely to transfer from LPSs to other schools or from testing grades to other grades. Boyd et al. (2008) and Loeb and Estrada (2003), on the contrary, found evidence of a positiv e effect of accountability for LPSs and highstakes grades.
95 Boyd et al. (2008) analyze the response of teachers to the implementation of statemandated testing in New York State using information on public school teachers from 1994 through 2002. The authors estimate the likelihood of teachers to quit the fourth grade (statewide assessment grade) using a logit model. They tr ack turnover rates through the eight-year period and focus attention on 2 years, 1994-95 and 2001-2002, before and after the introduction of an accountability system (the accountability system in New York State placed pressure on schools and districts during 1997-98). They find that the probability of a teacher leaving was lower in testing grades and lower for teachers working in LPSs. This reduction in turnover was mostly due to a reduced number of transfers between grades rather than to a redu ction of decisions to quit the teaching profession. This result is mos tly driven by teachers in suburban schools. The reduction of the turnover rate was greater for nov ice teachers than for experienced teachers; however, more experienced teachers were more likel y to leave the fourth grade in HPSs than in LPSs. The second question addressed by Boyd et al (2008) is whether th e characteristics of teachers entering fourth grade in different types of schools cha nged after the introduction of the accountability system. The author s found that teachers entering th e testing grad e in the postaccountability period were less likely to be novice and more likely to be from competitive undergraduate colleges. The analys is of the overall turnover of t eachers across all grades showed no statistically si gnificant results. Clotfelter et al. (2004) explore the effect of accountability sy stems on the ability of LPSs to attract and retain quality t eachers in North Carolina. They estimate the likelihood teachers quit LPSs using a discrete-time duration model with a sample of elementary school teachers who taught in the period 1994-2000. The North Caro lina school accountability program called ABC (where A is for accountability, B for basic skills, and C for local control) was implemented in
96 1996-97, and the analysis focuses on the probabil ity teachers quit LPSs in the post-accountability period. Results show that after the introduction of the accountability system the probability to leave the school increased for a typical teacher from 16-17% to 17-19%. For novice teachers, the effect is even more pronounced, increasing from 31-34% to 36-39%. The authors also address the question whether accountability has influenced the quality of teachers. As a measure of teacher quality, they use an indicator of wh ether the vacancy was filled by a teacher who transferred from within the district, which m eans that the teacher mo st likely has teaching experience. The results of a probit model show no statistica lly significant effect of the accountability system on the percen tage of within-district transfers. Finally, the authors found that the percentage of teachers from non-competitive undergraduate colleges and the percentage of teachers without experience dropped before re form; however, after refo rm it increased. These results are not statistically significant, but the aut hors argue that the effect was not revealed only because the time period after reform was too br ief to translate the changes in flows into significant changes in stocks. Susanna Loeb and Felicia Estrada (2003) us e two-period national-representative panel data to compare teachers answer s to survey questions in the pre-accountability school year (1993-94) and the post-accountability school year (1999-2000). In the post-accountability period some states had strong accountability systems wh ile other states had weak or no accountability systems. Since accountability systems are not uniform across states, in order to make a crossstate analysis one needs a univers al measure of accountability syst ems defined in the same terms for all states. The authors use an Accountability Index by States, constructed in Carnoy and Loeb (2003), which measures the relative strength of th e state accountability systems and takes values from 0 to 5. In the absence of detailed admi nistrative data, which tr acks the career paths of
97 teachers for all states during this period, their st udy focuses on teachers responses to the survey questions: If you could go back to your college days and start over again, would you become a teacher or not? and How long do you plan to re main in teaching?. The authors compared the answers to these questions in the 1999-00 and 1993-94 school year s for states with a different strength of accountability in 1999-00. Their results showed no statisti cally significant differences in the teachers an swers to the first question (desire to enter the teacher profession) for high and low-accountability states. However, the results indicate that several categor ies of teachers are more likely to express a desire to choose the same profession in the post-accountability period compared to the preaccountability period. These categories include teach ers with less than 5 y ears experience, those working in urban and rural schools, teachers w ith less than 5 years experience working in schools with more than half of the students from low-income households, and in schools where a majority of the students are of black or Hispanic origin. He nce, these findings support the hypothesis that accountability systems have a pos itive effect on the teacher supply in LPSs. The authors found no evidence of a relationship between accountability and the desire to quit the teacher profession. Finally, the authors also l ooked at the characteristics of new entrants to teaching. They found no statistically significant relationships between the competitiveness of a teachers undergraduate college and the measure of the accountability system. Summarizing the findings of th e previous studies, Loeb a nd Estrada (2003) and Boyd et al. (2005) found no evidence of change in overall teacher turnover; however, they found some positive accountability effects on teacher tur nover and quality in LPSs, which with accountability are likely to have more experi enced teachers and a smaller probability that teachers leave. On the contrary, Clotfelter et al. (2004) found that the probability of teachers
98 leaving LPSs has increased due to the introduction of an accountability system. The advantage of studies based on single state data is the possibility to track the career path of teachers in detail before and after the accountability reform in the state. However, looking at a single state makes it impossible to compare different accountability po licies and to analyze them separately. Both studies (Boyd et al., 2008; Clotfelter et al., 2004) ar e based on the analysis of the overall impact of complicated accountability systems consistin g of various mechanisms. The single policies constituting these accountability systems may ha ve opposite effects on the teacher labor market and counteract each other. Anothe r possible pitfall of studies ba sed on the analysis of a single state are other factors which might have aff ected the supply of and demand for teachers over time. Cross-state analysis allows a comparison of different accountability systems. However, due to the lack of detailed ad ministrative data, the Loeb and Estrada (2003) study focuses not on actual teacher turnover or quality, but on the teachers answers concerning choosing the same profession and their plans to remain in teaching. These responses may be subjective. Moreover, the survey does not take into account those teachers who have already quit teaching due to accountability pressure. Also, the measure of accountability strength used aggregates all state accountability policies together in one index, while some of these policies may have opposite effects on teachers and counteract ea ch other thus leading to a rela tively small net effect of the whole system. My study makes two main contributions to the area of study. Like Loeb and Estrada (2004), it utilizes a national-representative data set and compares different accountability environments. However, in order to catch the eff ect of different policies separately, it utilizes three dummy variables that measure the strength of different types of accountability policies
99 instead of one aggregated index. In the absence of data for teacher tur nover rates right after the implementation of an accountability system in each state, I analyze whether teacher quality has changed in the post-accountability period and whether these changes are more pronounced for states with strong accountability systems. As an indicator of teacher quality, I use the likelihood of a teacher being certified in his or her main assignment field. Previous studies mainly focus on the analysis of the difference of accountability effects on teachers working in testing grades and non-tes ting grades and teachers working in LPSs and HPSs. While this study also compares the effects fo r different types of schools, it also takes into account potential differences of the impact of accountability across subject fields. For this purpose, 4 subject fields are used for the analysis: a high-stakes and low-shortage field (English language arts), a high-stakes and high-shortage field (mathematics), a low-stakes and lowshortage field (social sciences), and a low-stakes and high-shortage field (s cience). Hence, this paper adds to the existing research by analyzi ng the effects of differe nt accountability policies separately and comparing the effect s across different subject fields. Types of Accountability Policies Like Carnoy and Loeb (2003), this study util izes a database deve loped by the Consortium for Policy Research in Education (CPRE) to defi ne the different types of accountability policies which were implemented in 1993 through 1999. This dataset provides a description of the state accountability systems which were in effect in 1999. These desc riptions are not uniform and, while they follow the same pattern of questions, they are not directly co mparable but vary both in terminology and in level of scrutiny. As a result, a variety of policies ar ise, each different from the other in some details. Thes e various policies might be arranged in groups according to the main targets to which they are applied. Thus, I define three main groups: student, school, and district-targeted policies.
100 Student-targeted. These policies directly affect student s. Examples of such policies are testing and a system of awards and requirements for students. By 1999, te sting had been widely introduced in United States. Almost all states adopted testing in reading, mathematics and writing by this time, and many states also had testing in other fiel ds. The testing grades and form of the tests significantly varied across states. Wh ile not affecting teacher s directly, testing may add additional pressure to teacher s working in the high-stakes fields especially when student test scores are used for school evalua tion and the assignment of assistance or punishment to a school or district. It implies a greater responsibility and may lead to less flexibility for teachers in arranging curriculum, teaching programs, and thei r class. They may have to concentrate their efforts on training students to the test and appl y more of their time to difficult-to-teach students. The different requirements and awards for students like minimal test scores, attendance rates, summer remedial programs, stipends, and certificates increase the pressure on students and this may affect teachers. However, it is not li kely to have a strong impact on teachers. Some states require students to pass the exam to graduate (exit test), which also primarily affects students, though it still might impose additional pressure on teachers. Overall, the student-targeted pol icies do not affect teachers di rectly. However, they might reduce the teacher supply in high-stake fields by contributing to additiona l pressure on them and thus decreasing the attractiveness of the job. The effect is not likel y to be strong, and is likely to be more pronounced in the states which also ha ve school-targeted or di strict-targeted policies based on the testing results of stude nts. While student-targeted policies are applied to all types of schools, they may affect teachers working in LPSs in a more negative way, since it is difficult to improve scores of low performing students.
101 School-targeted. These policies include th e assignments of school rankings, different types of sanctions and assistance for LPSs, and some form of aw ards for HPSs. School ranking based on the test scores of students and their dyn amics over the year is also one of the most widely used policies. It provides public informa tion about school performance, and may attach a stigma label to a school, which may have a negative psychological effect on teachers. The systems of ranking and the approaches used to evaluate student scores and school progress over time also vary significantly across states. In some states, the performa nce of particular subgroups of students (for example minority students or economically disadvantaged students) plays an important role in the ranking criteria. Rankings by themselves do not si gnificantly affect teachers. However, in many states, the ranking is used to impose sanctions on schools which do not satisfy state standards. These sanctions can take a harsh form, like closing a school, withhol ding school funds, or allowing vouchers for students in LPSs to transf er to better schools. These sorts of policies are likely to reduce teachers willingness to work in LPSs, since they may worry about the security of their job. In order to avoid th is uncertainty, they might try hard er to transfer to another school or to find a job outside of the school system. Another type of sanction for LPSs has a more administrative character and may positively affect the demand for teachers. Policies like school reconstitution, the reassignment of school administration, or interven tion into the management pro cess may force the administration to try harder to fill vacancies and to retain qua lified teachers. Assistance to LPSs usually takes the form of technical support, the organization of workshops assistance in planning, and additional financing. While it does not affect teachers directly, it may improve working conditions in LPSs and thus positively affect the supply of teachers. Finally, some states have
102 recognition or even financial rewa rds for teachers working in the best schools. This might affect the supply of teachers by increasing the attractiv eness of working in a HPS compared to other schools. Hence, some school-targete d policies are working in differe nt directions. Most of them are likely to reduce teachers s upply in LPSs. However, admini strative sanctions may boost the demand for good teachers. Hence, the overall eff ect of a strong school-targeted accountability system is ambiguous. However, most likely, th e negative effect for LPSs will prevail. District-targeted. These policies are less common than studentand school-targeted policies. Only a few states had some district level accountability policies, except district ranking or accreditation, in the 1999-00 sc hool year. Mostly, th ese policies are simila r to those of schooltargeted accountability, but they ar e applied at the district level, rather than the school level. They include the ranking or accreditation of sc hool districts and admi nistrative sanctions or assistance for low-performing districts. Thes e policies are mostly focused on improving the district management. Administrative sanctions, like the suspen sion of the school board, state takeover, limiting the authority of or firing th e superintendent, or publ ic hearings may force district authorities to put forth more effort to hire qualified teachers. District-targeted policies are likely to boos t teacher demand and may increase the quality of teachers, especially in LPSs Some accountability policies may have no effect on teachers at all, while others may affect teacher shortages in different ways. All accountability policies are more likely to be pronounced for high-stake fields (mathematics and English language arts) when compared to low-stake fields, because pe rformance in high-stake fields affect school evaluation, and therefore the assignment of awards, sanctions and assistance. Columns 2 through 15 of Table B-1 in Appendi x B illustrate the types of accountability policies discussed above that we re implemented in each state in 1999-00. Construction of this
103 table was based on information from a data ba se developed by the CPRE. Columns 16 through19 show three dummy variables which illustrate th e strength, by state, of student, school, and district-targeted accountab ility policies. These dummies take the value of one if this type of accountability is relatively strong in the given state. Student-targeted policies are cons idered strong if a state not only tests, but also has a relatively strong system of require ments for student promotion to the next grade or graduation. School-targeted policies are considered strong if there is a syst em of state-wide school evaluation and a relatively strong system of sanctions and assistance at the school le vel. District-targeted accountability policy is considered strong if there exists a district evaluation system and a system of administrative sanctions and assistance is well developed at th e district level. This study focuses on the effects of these three types of policies and, for comparison reasons, they are supplemented by an alternat ive aggregate measure of state accountability system. The aggregated index, constructed by Carnoy and Loeb (2003), is shown in the last column of Table B-1 in Appendix B. According to the analysis of the mechanisms of different policies presented in this section, one can try to predict the pos sible effect of each policy on the teacher labor market. Student-targeted policies are not likely to affect the teacher labor market. However, when schoolor district-targeted policies are based on th e results of student testing, teachers working in a high-stakes field may experience additi onal pressure which may reduce the supply of teachers in these fields. Consequently, this effect is likely to be revealed by an analysis of schooland district-targete d policies, rather than student-targeted policies. School-targeted policies may have both pos itive and negative impacts on the teacher labor market. However, the negative effect of acc ountability pressure is likely to prevail in LPSs.
104 District-targeted policies are focused on the improvement of district management, thus positively affecting teacher demand. All types of accountabili ty policies are likely to be more pronounced for high-stake fields than for low-stake fiel ds. The responsiveness of teachers to the accountability policies is likely to be higher in fields in which teacher shortage is not severe (i.e., English language arts and social sciences), rather than in high-shortage fields (i.e., mathematics and natural sciences), assuming teachers in high-s hortage fields are selected from those who are more dedicated to teaching, having demonstrated low sensibility to higher pay in alternative occupations. Empirical Strategy One of the possible measures of teacher qual ity is certification. I define the dependent variable as the probability that a teacher has a regular or alterna tive certificate in his or her main assignment field. The teacher quality indicator takes a value of 1 if a teacher is certified and 0 if not. Most states introduced stat e-wide accountability policies in the period from 1993 to 1999. In order to reveal the effects of the accountability policies, I use the pooled data set which combines the 1993-94 and 1999-00 school years. A logistic probability model is used to estimate the probability of a teachers being certified in a main assignment field. The regression equation includes year and state dummies, the subject field, and accountability policy dummies and their interactions with an LPS indicator. Since ther e is no available dataset providing information on school performance, it is not possible to create a precise indicator of LPSs. However, there is a correlation between the percent of economically disadvantaged students in the school and the likelihood the school performs badly. That is why I use the share of student s eligible for the free lunch program as a proxy for school performance. Explanatory variables include the teachers characteristics (sex, race, experience, age when the teacher first began fullor part-time teaching), school characteris tics (ratio of free lunch eligible st udents, ratio of minority students,
105 ratio of minority teachers, urban city dummy), and district characteristics (enrollment, contract with teachers union). The model is the following ,iiit it it m itjjittitit4it5tit 0123 j1 itit7ittittititit 689exp(x) Pr(y=1|x)=F(x)= 1+exp(x) x x year AIyear Field LPS yearLPS FieldLPS Fieldyear FieldyearLPS FieldAIyear 2,it kK ititititjit 1011 k kField AIyearLPS AIyearLPSstateu (3-1) where xjit are the school, district, an d teachers characteristics; yeart is a dummy variable equal to 1 for 1999-00 school year; AIt is an accountability policy dummy or index of accountability; Fieldit is a dummy indicating the subject field in two-way comparisons; LPSit is defined as the share of K-12 students elig ible for free lunch; and statek is dummy for the state k The coefficients of interest are 2, 9, 10, and 11. I define 4 subject fields (English language arts, mathematics, social sciences, and sciences) and run the logit regressions for each of these fields separately. For these regressions that use observations only for one field, I do not include a Fieldit dummy and its interactions with other variables in the regression equation. I also run regressions for each possible pair of fields. All analyses are weighted and clustered by states Since the model include s a large set of dummy variables, I also estimate a linear probability mode l with the same set of explanatory variables. Results of linear probability estimations are not reported in this study since they were consistent with the logistic probability model results. Description of Data The teachers, schools and district characte ristics come from SASS for the 1993-94 and the 1999-00 school years. The survey consists of teacher, school, district and principal questionnaires for public and privat e schools. In this study, only the public school data is used.
106 Table 3-1 provides statistics on th e certification ratios by subject field and time period. In both periods, the ratio of teachers cer tification was highest in English and lowest in social sciences. Since 1993-94, the proportion of certified teacher s has decreased in all 4 fields; however, the most dramatic change occurred in the high-shorta ge fields (mathematics and sciences). Figure 31 illustrates the difference in certification ratio by fields and years. Table B-2 of Appendix B shows summary statistics for va riables by fields and years. Results The results discussed in this section include the logistic regressions, estimated for each subject field group separately, and paired re gression estimations, where the sample for each regression includes observations representing two different groups The full set of results is included in Table B-3 through Tabl e B-22 of Appendix B. This s ection only addresses estimation results for variables pertaining to the main focus of my study. Table 3-2 presents the estimated effects of accountability policies on the likelihood of teacher certification in 4 differe nt subject field groups. The detailed results for these estimations are reported in Appendix B (Table B-3 through Table B-6). In genera l, the signs and coefficients of all explanatory variables are reasonable. Ma le teachers and experienced teachers are more likely to be certified, while t eachers working in schools with high proportions of economically disadvantaged students, or high ratios of minority t eachers, are less likely to be certified. The proportion of certified teachers is generally higher if a district has a contr act with a union. Each regression includes two variables estimating the effect of the pa rticular accountability policy on the likelihood of teacher certifi cation. The variables include a dummy indicating that in the 1999-00 school year the state had implemented this type of policy, and the interaction of the policy dummy with the share of students eligible for a free lunch program which is used as a proxy for LPSs.
107 The first two rows in the Table 3-2 show results for student-targeted policies. As expected, they have no statistically significan t impact on the teacher labor market. The only exception is a positive effect on social sciences teachers, which does not vary across different types of schools. Social sciences is a low-stake field, and teachers working in it are not subject to pressure from testing. However, a strong stude nt-targeted accountabilit y system, where students face test requirements for promotion to the next grade may make students more responsible thus creating a more comfortable environment for teachers. The last two columns show the results for aggregated measures of the whole accountability system of the state. None of them ar e statistically significan t. As noted earlier, the aggregate index is based on the strength of the whole system and thus treats all policies in the same way, but some of them may have quite opposite effects on teacher supply and demand. As a result, they may counteract each other. That ma y explain the absence of statistically significant results for the aggregated index of accountabil ity. Contrary to exp ectations, school-targeted policies do not affect teacher quality. This result is somewhat unexpected and will be discussed later in this section. The only consistently statistically significant effects of accountability can be observed for district-targeted policies. The results suggest th at strong accountability at the district level negatively affects the high-stakelow-shortage subject field (English language arts), but this negative impact is mitigated for LPSs. Such im pacts perfectly fit the hypothesis that teachers working in high-stake fields f eel the pressure of accountability which reduces their supply. However, in the presence of strong district accountability policies, the authorities may apply more effort to hire and attain qualified teachers in LPSs thus dimi nishing the negative effect of accountability pressure. It also explains the positive effect of district level polices on the social
108 sciences teachers. They do not suffer from the pressure of accountability as much as do highstake fields teachers, and at the same time they also experience the positive effect caused by improvement of the district management. The most striking result is that district accountability policies affect only low-shortage fields. The positive impact on science teachers is significant only at the 10% significance level and is not confirmed by the linear probability model. Since mathematics is a high-stake field, it is more likely to be affected by dist rict policies than social sciences. However, the labor market is not the same for high-shortage and low-shortage fields. The efforts of district administrators to fill vacancies with qualified teacher s are likely to be more effici ent in the case of low-shortage subject fields because there are more potential candidates for these jobs in the labor market. Moreover, the pressure of accountabili ty on teachers also varies across subject fields. First of all, there is evidence that it is easier to improve stud ent scores in mathematics compared to reading. Previous research on the effect of accountability on student outcomes has found that the gain in student scores in the post-accountabi lity periods is usually larger in mathematics compared to reading (e.g., Rouse, 1998; Hanus hek, & Raymond, 2004; Figlio, & Rouse, 2005; Reback, 2006). The possible explanation of this effect is that it is easier for teachers to affect mathematics scores because it is almost an en tirely school-based subjec t, while reading skills require more at-home learning a nd develop over longer periods of time. Therefore, mathematics teachers are likely to be less pr essed by testing standards than reading teachers. Second, highshortage field teachers may be less sensitive to th e accountability pressure because their supply is mostly driven by other labor market factors, like wages in alternative occupations and the availability of jobs outside of the school education system. Therefore, the difference in the effects of accountability policies on teachers working in low-shorta ge and high-shortage fields
109 can be explained by labor market factors and by specificities of the learning process in these fields. Table 3-3 shows the effect of district-targeted accountabil ity practices on the likelihood of a teacher to be certified, conditional on the type of school. The likelihood of certification is calculated using logit estimation results (Tab le B-5 of the Appendix B). Schools where the proportion of free lunch eligible st udents is at or below the 25th percentile of the whole distribution are considered highperforming, while schools wher e the proportion of free lunch eligible students is at or above the 75th percentile are viewed as low-performing. The results indicate that due to the pressure of accountability the quality of teachers diminishes in the English language arts fields. The ratio of certified teachers decreases by 7.5% points in high performing schools and by 4.4% points in LPSs. On the co ntrary, the effect of accountability is positive for social sciences, a low-stake, low-shorta ge field, and it increases the teacher certification ratio by 3. 04% points in HPSs and by 4.4% points in LPSs. Hence, the policy was more effective for LPSs in low-shortage subject fields. For high-shortage fields, no statistically significant results were found, hence we can make no conclusions about the effects of accountability policies in these fields. There was a weak significant positive effect for the sciences field group in the logistic probability m odel (Table 3-2), however it was not confirmed by linear model estimations and the level of significance was only 10%. It is also interesting to look at the comparative analysis of different field groups together. The paired regression estimations for all ty pes of accountability polic ies are presented in Appendix B (Table B-7 through Ta ble B-12). Since only district-t argeted policies demonstrate statistically significant effects on the teacher s labor market, I will only discuss paired regression estimation results for this type of accountability policy. The estimates regarding the
110 variables of main interest are presented in Tabl e 3-4. They include a dum my for district-targeted policy, an interaction term for the policy dum my and the subject field dummy (not base category), an interaction term for the policy dummy and the LPS indicator, and an interaction of the policy dummy, a field dumm y, and the LPSs indicator. Similar to the previous findings, the results show that district-targeted accountability policies mostly affect English langua ge arts teachers (high-stake, low-shortage field). The effect was negative when compared to low-stake fields and to the high-stake high-shortage field. The negative influence of accountability policies is less pronounced for LPSs. The intuition behind these effects is the same as discussed before; En glish language arts teache rs are exposed to more accountability pressure than teachers of mathema tics and teachers working in low-stake fields. This negative influence is decreased thanks to mo re efficient work of dist rict authorities, who, under the accountability system, are fo rced to operate more efficiently, especially with respect to LPSs. Generally, all results, except the lack of evidence for school-targeted accountability effects, are consistent with the hypotheses. Student-targeted accountab ility does not affect teachers. The aggregated measure of accountability system strength is not statistically significant in all regression estimations, possible due to a counteraction of positive and negative effects of some policies. Teachers working in high-stake, low-shortage fields are more affected by accountability than those working in high-shortage, lo w-stake field. Surprisingly, the estimates of the impact of school level accountability do not support the hypothesis that school-targeted policies should negatively affect teachers working in LPSs. This result is somewhat unexpected because at least some school-targeted polic ies, like sanctions and awards assigned according to school perfor mance, should have affected the working environment in different types of schools in different ways. Under these policies, teachers
111 working in LPSs are likely to f eel insecure about their jobs and be less flexible in their classrooms, so it is natural to a ssume that qualified teachers would be willing to transfer to other schools to avoid additional pres sure. However, we observe no such effect. One possible explanation is that in small school districts, teachers may not be ab le to transfer schools without changing their place of residence. Therefore, the effect of school-targete d policies may be strong only in large school districts, where teachers can switch from one school to another without moving to another residence. In order to check whether school-tar geted policies do in fact affect teacher quality in large districts, I include cont rols for the district size in the same set of regressions. The results are presented in the next section. Results with Controls for District Size I ran the same regressions for each subject field group separately, but I controlled for the possible differences in the impact of accountability policies in sma ll and large districts. For this purpose, I include in each regression equation two additional vari ables: an interaction term of a policy dummy and district enrollment, and an interaction term of a policy dummy, a LPS indicator and district enrollment. The detailed re sults of the regressions estimated for each field group are presented in the Appendix B (Tab le B-13 through Table B-16). The estimates regarding variables of main interest are summari zed in Table 3-5. As in previous findings, the effects of student-targeted accountability and ag gregated policies are, in general, mostly statistically insignificant. The only statistically significant result confir med by both logistic and linear probability models is a pos itive effect of the aggregated index of accountability on science teachers certification in large districts. School-targeted policies s till show no statisti cally significant results for English language arts or mathematics fields. A policy dummy and an interaction of the policy variable with an LPS indicator remain insignificant for all fields. However, school level accountability has a positive
112 effect, in large districts, on the likelihood that social sciences teach er are certified. This effect is smaller for teachers working in LPSs. It corresp onds to the hypothesis that due to the negative effect of accountability, qualifie d teachers tend to transfer from LPS to better performing schools. Also, in large districts, school-targe ted policies increase th e likelihood that science teachers will be certified if they work in LPSs. Si nce the sciences represent a high-shortage field, the labor market of teachers may be larger in large districts. Therefor e, while it may be not possible to hire qualified teachers for LPSs in small districts even with the additional effort by district authorities, it may be easier to do so in large districts where there is a relatively larger labor market for teachers. Table 3-6 shows the effect of school-targete d accountability on the likelihood that a teacher is certified, conditional on the type of school and size of district. The discrepancy between low-performing and HPSs is greater for social sciences teachers in large districts compared to small districts. In general, teachers are more likely to be certified in large districts (probably due to a larger local teacher labor mark et); however, teachers work ing in LPSs in large districts are less likely to be certified compared to teachers in HPSs. We do not observe such a difference in small districts. This result confir ms the hypothesis that certified teachers tend to flee from LPSs when there is a choice of bette r performing schools in th e vicinity. Due to the accountability policies in LPSs, th e probability that a teacher is certified is lower for those sciences teachers working in small district s compared to teachers in large districts. The third section of Table 3-5 presents the results regarding district-targeted policies. The results suggest that due to district level accoun tability, the likelihood low-shortage field teachers are certified is higher in large districts. This difference also can be explained by a larger local labor market of teachers in large districts. The mo st striking result is that district accountability
113 has opposite effects on English and mathematics teachers. While English teachers, on average, are negatively affected by district accountabi lity, there is a positive impact on mathematics teachers certification ratio. Following the same logic as in the previous section, both the sensitivity of these two groups of teachers to the pressure of accountability and their labor markets may differ. Since it is easier for a teache r, in terms of quantitative scores, to increase student test scores in mathema tics, math teachers do not feel th e accountability pressure as much as English language arts t eachers. At the same time, the district authorities apply more effort to hire and retain teachers in bot h high-stake fields. Therefore, si nce the administration operates more efficiently under a strong district level ac countability system, we can observe an overall positive effect on mathematics teachers. However, district efforts are not enough to fully offset the negative impact of accountabilit y pressure on English arts teacher s, which leads to a negative net effect for them. In the large districts, the negative impact of policy on English teachers is decreased thanks to a larger labor market of teachers. However, teachers working in low performing schools are more exposed to the pressure of account ability and tend to transfer to better schools. Therefore, in large districts, the positive effect of district-targeted policies is less for teachers working in LPSs. On the contrary, since in larg e districts mathematics teachers have more job opportunities outside of the educ ation system, the positive ef fect of district-targeted accountability policies on the mathematics teacher s is less in large dist ricts than in small districts. Also, in the case of mathematics teachers who are less sensitive to accountability pressure, district-targeted policie s are more efficient in LPSs. Table 3-7 shows the effect of district-targeted accountab ility on the likelihood that a teacher is certified, conditional to the type of school and size of the district. In HPSs in large
114 districts, the negative impact of accountability on the quality of Englis h language arts teachers fall from 9.6% points to 5.6% points. However, there is no noticeable difference between large and small districts in the certification ratio of LPS teachers. In large and small districts, accountability positively affects the likelihood that a teacher in mathem atics is certified. However, in large districts, the effect diminishes from 6.7% points to 3.4% points. For LPSs, the drop in the certification ratio is less dramatic in large districts compared to small ones; it decreases only by 0.5% points. To understand better the effects of school and district-targeted accountability I ran paired regression estimations for each type of accountability policy with controls for district size. The detailed results are pr esented in the Appendix B (Table B-17 through Table B-22). This section focuses only on the results for the schoo l and district-targeted policies, because the effects of the student policies and aggregated index are, in general, statistically insignificant. The estimates for school-level accountability effects are summarized in Table 3-8. In all subject fields school-level accountabi lity in general decreases the probability that a teacher will be certified. Because of a larger local teacher mark et, the negative effect of the school-targeted policies is lower in large districts when compar ed to small ones. However the difference between the certificati on ratio of small and large distri cts is lower for high-shortage fields than for low-shortage fields. These results can be explained by the fact that while large districts have a bigger local market of teachers, there are more alternative occupations outside of the education system. Hence, high -shortage teachers have, in ge neral, more opportunities to quit teaching and escape accountability pressure. In large districts, social science teachers working in LPSs are less likely to be certified compared to teachers in other schools. These results confirm the hypothesis that qualified teacher s are less willing to work in LPSs and tend to transfer to
115 better schools. Table 3-9 presents the results of di strict-targeted policies fo r pairs of fields. These results are consistent with previous findings that suggest district accountability reduces the certification ratio of English te achers and boosts the certification ratio of mathematics teachers. This difference is less pronounced in large districts and in LPSs. Conclusions and Directions for Further Work The goal of school accountability reform is to improve student outcomes. However, the mechanisms and nature of accountability policies a ffect not only students but also teachers. This study analyses the impact of account ability on teacher quality. In cont rast with previous research, the effects of different policies are measured se parately and not only for high-stake fields, but also for low-stake fields. The results suggest that school and district -targeted accountability do affect teachers in all subject field groups, while accountability at the student level affects teachers only when it is linked to school or district le vel policies. The aggregated meas ure of the strength of the whole accountability system reveals no statistically sign ificant effects of account ability on the teacher labor market. Apparently that aggregated measure is not the best way to estimate the impact of accountability on teacher certification. It includes all types of accountability policies, some of which do not affect teachers and some that may affect them in opposite directions. School-targeted policies a ffect the teacher labor market onl y in large districts and mostly in low-shortage fields. This effect decreases the quality of teachers in LPSs. In an effort to avoid the pressure of accountability, quali fied teachers prefer to work in HPSs wh ere the threat of school sanctions is lower. This effect takes place only in large districts because in small districts teachers have limited choices of schools and cannot easily transfer to another school. The effects of district-targeted policies vary significantly across diffe rent subject fields. In general, they affect high-stake fields; however, the nature of the effect is quite different. While
116 for English language arts teachers the negative effects of accountability pressure prevail, on the contrary, the quality of mathematics teachers increases in states with st rong district-targeted accountability policies. These oppos ite effects suggest a relative ly higher sensitivity of lowshortage field teachers to the pressure of acc ountability. Perhaps they may be explained by two factors. First, it is more difficu lt to improve student performance on tests in language arts than in mathematics. Second, the high-shortage fiel d teacher supply might be less sensitive to accountability policies than the lowshortage field. Therefore, the effect of accountability in lowshortage fields is driven prim arily by the negative impact of accountability pressure, which is lessened in LPSs and in large districts by th e positive effect of district management improvements. However, in high-shortage fields the negative effect of accountability pressure is weak, and overall the effect is positive due to district efforts to find qualified teachers. Another source of difference in teacher responses to distri ct and school level accountability arises from the difference of the la bor markets for low-shortage and high-shortage fields. Teachers with degrees in high-shortage fields have more alternative job opportunities outside of the school system compared to teac hers working in low-s hortage fields. This difference is especially apparent in large districts. Therefore, wh ile the school and district level policies are relatively more effective in large districts for teachers working in low-shortage fields, they are less effective for th ose working in high-shortage fields. Further research should more fully define the accountability policy variables for the 1993-94 school year. The database used in this study contains information only for the 1999-00 school year accountability system s. The assumption for the 1993-94 school year is that accountability policies had not been implemented. However, that is not true for all types of accountability policies. Some stat es implemented testing or some form of accreditation and LPS
117 assistance before the 1993-94 school year. Hence, it would be useful to more precisely define the policies variables. It is also of interest to try other definitions of teacher quality and to analyze the same models with other depe ndent variables. These possible measures of shortage may be include the correspondence of a teachers bachelo rs degree to his assignments in school, the competitiveness of a teachers undergraduate colleg e, a teachers decision to quit teaching or to move to another school. Another in teresting direction for further res earch is to analyze the effect of school-targeted accountability using more disaggregated accountability policy variables, such as sanctions, awards and assistance. These policies may have positive and negative effects on teacher willingness to work in LPSs, and it might be useful to look at thei r effects separately. Since school-level policies affect teacher quality only in large districts, this analysis should focus only on the large districts. One of the most interesting findi ngs of this study is the diffe rent effects of the school and district-targeted policies on Eng lish language arts teachers and mathematics teachers. To better understand the nature of teacher responses to th e policy, and to analyze whether the reason for the different reaction is in fact explained by th e difference in the teacher labor market for lowshortage and high-shortage fields it would be interesting to analyze teacher movements from one school to another or to an occupa tion outside of the school system in large districts, conditional on type of school and subject field.
118 Table 3-1. Teachers certificat ion ratios by year and field Field 1993 1999 Change English language arts 94.9% 90.5% -4.4% Mathematics 93.6% 87.7% -5.9% Social sciences 91.3% 86.9% -4.5% Sciences 93.7% 87.4% -6.3% Table 3-2. Logit results for regressions estimated separately for each field group English language arts Mathematics Social Sciences Sciences Variable odds ratio odds ratio odds ratio odds ratio Student policy 0.86 0.64 2.25* 1.48 Student policy LPS 1.86 2.75 0.63 0.91 School policy 0.76 0.53 0.88 0.59 School policy LPS 1.68 0.80 0.67 1.73 District policy 0.25*** 3.00 2.18*** 0.53* District policy LPS 5.10*** 0.64 3.00 1.27 Accountability Index 1.12 0.95 1.03 0.87 Accountability Index LPS 0.75 0.71 0.94 1.14 Number of observations 8747 6547 5593 5794 State fixes effects yes yes yes yes Weights yes yes yes yes *, **, *** statistically significant at 10, 5, and 1% level respectively. Table 3-3. The effect of district-targeted accountability on the likelihood of teacher certification Field HPS LPS English language arts -7.50% -4.40% Mathematics 4.93% 6.04% Social Sciences 3.04% 4.40% Sciences -4.54% -4.51%
119 Table 3-4. Logit results of regressions estimated for pairs of field groups (effect of districttargeted policies) English lang. arts (base) vs. Mathemati cs English lang. arts (base) vs. Social Sc. English lang. arts (base) vs Sciences Math (base) vs. Social Sc. Math (base) vs. Sciences Social Sc. (base) vs. Sciences Variable odds ratio odds ratio odds ratio odds ratio odds ratio odds ratio District 0.43* 0.37* 0.24*** 2.14 2.72 1.65** District Field 4.53*** 3.67* 2.29** 0.51 0.76 0.61 District LPS 3.56*** 3.94** 3.94*** 0.61 0.60 2.03 District LPS Field 0.30 1.40 0.61 2.14 5.31 0.47 *, **, *** statistically significant at 10, 5, and 1% level respectively.
120 Table 3-5. Logit results for regressions with c ontrols for district size estimated for each field group separately English language arts Mathematics Social Sciences Sciences Variable odds ratio odds ratio odds ratio odds ratio Student policy 0.90 0.63 1.99 1.28 Student policy LPS 1.48 1.46 0.66 0.75 Student policy bigD 1.28 2.23 1.45 1.69** Student policy bigD *LPS 0.87 0.76 0.72 1.03 School policy 0.55 0.52 0.67 0.63 School policy LPS 1.49 0.44 1.08 1.00 School policy bigD 3.24 1.71 1.96** 0.91 School policy *bigD *LPS 0.37 0.67 0.35*** 1.76*** District policy 0.20*** 6.55*** 1.49 0.43 District policy LPS 9.39*** 0.12* 1.79 1.52 District policy bigD 22.37*** 0.01*** 2.76* 7.72 District policy *bigD *LPS 0.01*** 4.49*** 0.01 0.28 Accountability Index 1.00 0.89 1.00 0.79 Accountability Index LPS 0.97 0.89 0.95 1.22 Accountability Index bigD 1.16* 1.01 1.06 1.20*** Accountability Index *bigD *LPS 0.78*** 0.95 0.96 0.79** Number of observations 8747 6547 5593 5794 State fixes effects yes yes yes yes Weights yes yes yes yes *, **, *** statistically significant at 10, 5, and 1% level respectively. Table 3-6. The effect of school-targeted accountability on the likelihood of teacher certification conditional on th e size of the district Small district Large district Field HPS LPS HPS LPS English language arts -2.6% -2.6% -1.7% -1.9% Mathematics -7.3% -11.7% -6.2% -10.9% Social sciences -2.0% -2.0% -1.5% -1.8% Sciences -2.6% -3.2% -2.7% -3.1%
121 Table 3-7. The effect of district-targeted accountability on the likelihood of teacher certification conditional on the size of the district in percentage points Small district Large district Field HPS LPS HPS LPS English -9.6 -3.6 -5.6 -3.5 Math 6.7 5.6 3.4 5.1 SS 2.4 3.2 4.5 5.0 Sciences -5.7 -5.4 -3.3 -3.3 Table 3-8. Logit results for regressions estimat ed for pairs of field groups for school-targeted policies, conditional on district size Variable English vs. Math English vs. Social Sc. English vs. Sciences Math (base) vs. Soc. Sc. Math (base) vs. Sciences Social Sc. vs. Sciences School policy 0.51*** 0.53** 0.59 0.59 0.61 0.88 School policy Field 1.07 1.42 1.02 1.22 0.91 0.67 School policy LPS 1.26 1.95 1.42 0.66 0.53 0.59 Sch. policy *LPS Field 0.45 0.31 0.62 0.80 1.54 1.99 Sch. policy *bigD 3.08* 3.43* 3.24** 1.68 1.77 1.78*** Sch. policy *bigD LPS 0.37 0.28 0.36 0.64 0.65 0.47*** Sch. policy bigD Field 0.69* 0.56 0.25*** 0.99 0.43*** 0.42*** Sch. policy *bigD*LPS* Field 1.44 1.58 5.71*** 0.83 3.28*** 3.89*** Number of observations 15379 14340 14571 12140 12341 11417 State fixes effects yes yes yes yes yes yes Weights yes yes yes yes yes yes *, **, *** statistically significant at 10, 5, and 1% level respectively.
122 Table 3-9. Logit results for regressions estimated for pairs of field groups for district-targeted policies, conditional on district size Variable English vs. Math English vs. Social Sc. English vs. Sciences Math vs. Social Sc. Math vs. Sciences Social Sc. vs. Sciences District policy 0.41** 0.34** 0.20*** 6.42*** 4.48*** 0.97 District policy Field 10.18*** 2.01 1.95 0.20* 0.21* 1.06 District policy LPS 4.10*** 5.10*** 6.69*** 0.09** 0.14** 3.09 District policy *LPS Field 0.07*** 2.25 0.55 42.52*** 8.58* 0.27 District policy *bigD 8.00*** 5.51 11.28*** 0.01*** 0.01*** 2.99 District policy *bigD LPS 0.03** 0.03 0.01*** 7.05*** 4.80*** 0.01* District policy bigD Field 0.01*** 3.51 5.64 8.81*** 2.91*** 0.01* District policy *bigD*LPS* Field 5.04*** 0.01** 0.70 0.01*** 0.01** 4.54*** Number of observations 15379 14340 14571 12140 12341 11417 State fixes effects yes yes yes yes yes Yes Weights yes yes yes yes yes yes *, **, *** statistically significant at 10, 5, and 1% level respectively.
123 80% 82% 84% 86% 88% 90% 92% 94% 96% 98% 100% EnglishMathematicsSocial SciencesSciences 1993-94 1999-00Figure 3-1. Teachers' certif ication ratio by fields in the 1993-94 and 1999-00 school years.
124 CHAPTER 4 INVESTMENT STRATEGIES OF PUBLIC PENSION FUNDS Introduction During the late 1990s and into the 2000s, many public pension funds shifted toward more aggressive investment strategies. Though some of them simply increased their holdings of stocks relative to bonds, others emulat ed the Swensen model, sometimes known as the Yale model, which became popular because of the consistent high returns obtained by David Swensen, who took charge of investing Yales endowment in 1985 and recently left. The idea was that funds that have limited need for liquidity should put su bstantial shares of thei r portfolios in illiquid assets, such as real estate a nd hedge funds, thereby gaining highe r expected returns with little increase in risk. The timing of this shift by public pension f unds was unfortunate, causing them to be heavily invested in illiquid asse ts just as the financial crisis of 2008-09 froze their markets, making them impossible to value. Many public pension funds suffered enormous losses. According to Bonafede, Foresti, and Browning (2009) 93% of 59 st ate retirement systems that reported their actuarial data for 2008 were underf unded. The average assets -to-liabilities ratio has decreased from 94% in 2007 to 77% in 2008. Many big state retirement systems experienced a decline in assets of about 20-30%. The State Board of Administration that manages many of Floridas public investments had a $62-billion loss in assets which constitut es about a third of their value (Freedberg, & Humburg, 2008). The sharp decrease in the assets was caused by higher-risk investments which had been project ed to provide high prof its, but generated big losses instead. The Colorado Public Employees Reti rement Association that had $41.4 billion in assets in December 2007 lost more than $10 billion last year and now can meet only 51.8% of its obligations (Hoover, 2009). Its 26% lo ss in assets is a result of decisions made in the 1990s to
125 lower contribution rates and to project an 8.5% return on investments. Ne w York States pension fund lost 26.3% of its assets value (Hakim, 2009). The Californ ia Public Employees Retirement System (Calpers) assets value fe ll 23.4% last year and assets of the California State Teachers Retirement System (Calsters) sh rank by 27% (Garrahan, 2009). Standard actuarial practices coupled with legal funding requirements will force public pension fundsunless there is a remarkable market recoveryto siphon off much larger shares of state and local revenues just when those reve nues have been driven down by the recession and by falling property values. In addition, public empl oyees are being required to accept pay cuts, increased pension contributions, and involuntary part time. The im pact of the losses by public pension funds on state and local governments w ould have been far worse had it not been for aggressive monetary and fiscal policies at the federal level. Th e Federal Reserve, by doubling its balance sheet, re-established liq uidity in frozen markets and cr eated arbitrary values for trouble assets. The Congress authorized a stimulus package of some $800 billion, includ ing $58 billion to state and local governments. Without such federal support layoffs by state and local jurisdictions would have been gr eater and the feedback into th e national recession larger. Even with the federal assistance, state and local governments are likely to experience budget difficulties for several years, in substantial part because of the need to rebuild their pension assets. The federal government was able to provide an effective stimul us package without escalating long-term interest rate s, which would have made the federal debt unsustainable and thus killed the stimulus, only because China a nd Japan were willing to continue to hold and indeed even to absorb more U.S. Treasury securities. No one can be certain that they will do the same again if in a few years there is another cr isis. Indeed, Chinas central bank has called for a
126 global reserve currency to repl ace the dollar, and French Presid ent Nicolas Sarkozy has stated the costs of the dollar as a reserve currency may exceed its benef its. If the next financial crisis is also a dollar crisis, instead of one in which the dollar benefits as a safe haven, the ability of the federal government to rescue state and local governments from stress caused by their pension funds will be severely limited. For this reason, it has become an urgent policy issue to understand why pension funds invest aggressively, putting local jurisdictions at great er risk of experien cing serious financial difficulties. Logically, the more fundamental ques tion is what is the optimal position for public pension funds to take on? I do not, however, addre ss that question directly in this chapter. If pension funds use rate of return that is higher than risk-free rate the ethical problem arises as current taxpayers who are consumers of product of labor of public employees may shift risks of the pension fund investments to future taxpayer s. Intergenerational equity requires that each generation of taxpayers pays for the public servi ces it consumes contemporaneously. Therefore, one school of thought contends that since publ ic pensions are a ve ry certain obligation historically they have always b een paidthe appropriate rate of return to project for pension portfolios is that on the safest asset, U.S. Treasury securities (Clark, Craig, & Ahmet, 2008; Waring, 2008). However, currently this is not the case. While the nominal market rate for fixed income securities is no greater than 5%, public pension plan actuaries us e nominal rates in the neighborhood of 8% (C lark et al., 2008). However, the question of intergenerationa l equity may be viewed from another perspective. New generations us ually consume more than the previous ones. Hence if tax payments of the current taxpayers cover all future be nefit payments to cu rrent public employees and the risks are not shifted to future taxpayers it may make the current generation relatively
127 worse off in the context of interg enerational equity and cause larg er tax distortions thus reducing welfare. Therefore, another sc hool of thought places optimal pub lic pension investing into the context of tax smoothing or, more generally, ta x-and-expenditure smoothing. According to this view, if future taxpayers could determine current investment strategy, they would balance the expected rate of return (ERR) against the risk of distortions caused by higher taxes to meet required pension obligations. Yet another view sa ys that though current beneficiaries share no risk, future beneficiaries who ar e currently public employees do sh are the risk of lower pay or higher contributions should investment returns sour before they retire. An appropriate investment strategy, by this view, should include their utility among its concerns. Actually the same debates are recently going regarding appropriate discount rates that should be used to evaluate future losses caused by global warming. Stern (2006) has used a very low discount rate of 1.4% to estimate the presen t value of damages of global warming. Nordhaus (2006) criticized this approach to discounting arguing that too low a discount rate would lead to average annual consumption losses over the i ndefinite future. Taking into account that Stern projects that future generations will be richer than the current generation (per capita consumption would grow at 1.3% per year and would increase from $7,800 t oday to $94,000 in 2200), a cut in consumption today in favor of future genera tions looks contradictory to the concept of intergenerational equity. Instead of the 0.1% ra te of social time pr eference used by Stern, Nordhaus suggests a 3% rate that is consistent wi th a logarithmic utility function, market interest rates and rates of private savings and invest ment. Other economists ha ve supported Nordhaus critiques of the low discount ra te. Mendelson (2008) claims that low discount rates are equitable only in the sense that they make every generation worse off (p.52). He notes that since market assets earn a higher rate of return than if investment dollars are spent on abatement of
128 greenhouse gas emissions the opportunity cost of lo sing future earnings is high (p.52). Weyant (2008) also supports using the ma rket rate of return rather than a low rate based on the intergenerational equity concept: The essential problem I have with Reviews use of a purely ethically based rate of time pref erence is that it leads to a rate of return on capital that is inconsistent with the currently observed rate of return on capital (p.90). Hence the use of a rate of return that corre sponds to risk-f ree assets for evaluation of public pension funds future earni ngs would lead to cuts in current consump tion, high opportunity costs, and higher tax distortions. Th erefore, the optimal rate of return in the general case would always be higher than the risk-free rate (except fo r special cases such as infinite risk-aversion of taxpayers). In a perfect world it would not be difficult to define the optimal rate of return. According to theory individuals make optimal choi ces and in a first-best context each individual just chooses the optimal investment strategy to provide appropriate sa vings for retirement. However, as Barr and Diamond ( 2008) note, in reality there are too many deviations from a simple theoretical world. Therefore due to li mitations and market imperfections, such as imperfect information, incomplete markets a nd progressive taxation, the policy should be formulated in a second-best context. Under the tax-smoothing approach it is not su rprising that the overwhelming majority of pension funds does not invest entirely or even mostly in Treasury s ecurities or base their projected returns only on such securities. As an empirical matter, in the United States at least, there is little to be le arned from trying to distinguish funds that follow such a strategy from those that do not, since virtually none do. That being the case, I will explore pension returns in the context of tax smoothing approach A number of prev ious studies have modeled the optimal investment strategies of public pension funds us ing a tax distortion appr oach. Epple and Schipper
129 (1981) provided such a model as an explana tion that underfunding of public pension funds may be optimal. DArcy, Dulebohn, and Oh (1999) pr esented a model to show that underfunding of public pension funds is optimal when the growth in pension costs over time is below the growth in the tax base. Recently Lucas and Zeldes ( 2009) provided a model of optimal investment strategy for public pension funds deriving the optimal condition for the share of investments in stocks that minimizes tax distortions. I use Lucas and Zeldes (2009) theoretical fr amework to analyze why pension funds may be using overestimated rates of return to discoun t their liabilities. Following this framework I consider the effect of a pension boa rds directing actuaries to use a pr ojected rate of return that is too high in the sense that it does no t accurately reflect th e actual tradeoff between risk and return available from the financial markets and or, altern atively, the true cost of tax distortions. The model itself has the very straightforward implication that unde restimating risk aversion or underestimating the risk associated with a given E RR, which of course leads to using a projected rate of return that is too high, results in excessi ve hiring of public employ ees. (It would also lead to setting their pension/cu rrent salary ratios too high, but I do not dwell on that.) The point is that in addition to the macroeconomic risks associated with projected rates of return that are set too high, there is also misallo cation of public resources. In the empirical portion of th is question, I address the issu es of whether union strength is associated with a high or a low ERR. I proceed with the assumpti on that many projected rates of return are too high, since there is a considerable variability in the projected rates of return used by public pension funds. My hypothesi s is that union strength leads to a higher projected rate of return, since unions will gain from the resulting underestimati on of the true cost of hiring more employees or giving existing ones larger benefits. It is also possible that st rong unions would
130 oppose higher projected rates of return, and the riskier portfolios associated with them, because they wish to avoid the risk of larger contributions or reduced be nefits if markets crash. I do not think that is the case: in most unions older workers, who face less risk of either higher contributions or reduced pensions, are in charge at least according to conventional wisdom. But the issue is an empirical one. I ne st the two hypotheses, allowing the data to reject either one or neither. What I find is that, as I expected, greater un ion strength is associat ed with public pension funds that allocate their portfolios toward ri skier assets. My results are merely indicative, rather than definitive, for various reasons. First, I am no t able to use projected rates of return directly but must construct proxies for them using por tfolio allocations. Sec ond, because of data limitations, I use ordinary least squares with many covariates rather than better ways of achieving identification, such as difference in difference, regression discontinuity, or instrumental variables. Third, one of my observation years extremely over-represents pension funds in a single state (Pennsylvania). Given the urgency of the issue, however, even a preliminary result associating ri sky strategies by public pension funds with union strength is an important finding. Theoretical Background Under the tax-smoothing approach Lucas and Zeldes (2009) suggest an optimal pension fund asset allocation model that minimizes the welfare cost of distortion taxes for two periods. The model is based on a trade-off between the hi gher average return on equities which lowers average taxes and the greater ri sk of equities which increase s expected tax distortions. The optimal share of stocks in the portfolio of the pension fund is found by minimizing tax distortion in two periods. Lucas and Zeldes (2009) formulate the following model:
131 22 1122 111 222 21min s.t. 22 11, fsfww ETTTT TC TC CLXrrr (4-1) where Ti is total taxes paid in period i and is equal to the sum of pension contributions Ci and other taxes i, is a subjective discount rate, and w is a curvature parameter (the marginal tax rate is assumed to be proportional to total tax collections). Contributions in the second period, C2, must be equal to the underfunding th at occurs in the second period. Li represents liabilities in period i future liabilities are stochastic with expected growth rate E( ), and standard deviation of growth ( ) X shows invested funds in the first period and is the sum of pension assets in the first period, A1, and contributions in the first period, C1, minus net payments in the first period, B1. The constant risk-f ree rate is given by rf, stock returns are stochastic with an expected return E(rs) and standard deviation (rs) The solution of the model 4-1 shows the optimal fraction of pension assets invested in stocks : 21 2 211( ) 11 sf sfsffsf sfErr ErrELrrXrErr w XErr (4-2) In the general case it is optimal to hold some stocks ( >0)14. So it is not surprising that on average the ERR used by pension funds exceeds the risk-free rate. However, there exists considerable variation in ERR across public pens ion finds. For example, data collected for 2006 for local pension funds by the Center for Retirement Research (CRR) at Boston College shows that ERR varies from 6% to 8.8% (the sample consists of data for 87 local pension funds). If 14 The share of stocks is zero only if there is strong negative correlation between other tax distortions and equity returns, or, if there is no correlation between other tax dist ortions and equity returns, if risk aversion is approaching infinity, future tax liabilities are zero and the plan is fully funded (Lucas, & Zeldes, 2009).
132 public pension funds use the optimal ERR why do es it vary across funds? It looks like some public pension funds may use an ERR that exceed s the optimal ERR and therefore causes larger tax distortions. Further in this chap ter I analyze why that may happen. The share of stocks is negatively relate d to the volatility of stock returns E(1+rs)2, which can be expressed as a function of the st andard deviation of the stock returns: 2 2211.sssErrEr (4-3) Let express the optimal share of stocks as a func tion of standard deviatio n of the stock returns: 2 21 2 2, 1 1(), 2. s sfsf sf fsf sfsfY rZ ErrELrr Err YrErr wXX ZErrErr (4-4) Assume the general case with >0. Then the share of stocks decr eases as standard deviation of the stock returns increases: 2 22 0. s s srY r rZ (4-5) Therefore, the optimal depends on the relationship between the risk and the rate of return ( ( rs)). However, in the case of a lack of transpar ency, the taxpayers in th e first period might not be aware of the true relationship between return and risk. In this case, if the pension fund uses a false ( rs), the chosen share of stocks would not be optimal. Assume ( rs) is a quadratic function of rs and some exogenous positive parameter a: 2.s sr r a (4-6) The higher is a the higher is the optimal share of stocks : 2 2 2 22 0. sss s srrYr ara arZ (4-7)
133 However, due to a lack of transparency, taxpa yers do not know the ex act form of the risk function. Assume the pension fund wishes to invest more in stocks than it is optimal. In this case it would pretend that the relationship between risk and stocks return is 2,,srawhile the true relationship is1,sra, where a2> a1. That would lead to a bigge r fraction of stocks in the pension fund portfolio than is optimal and cons equently to higher tax distortions and welfare loss. Why might pension funds tend to underestimate risk ? Assume a state or local government spends its budget on two main items: payments to public employees and financing another public good. The government income has only one sourcetax payments (T) The government chooses how to distribute its budget optima lly: how many employees to hire (n) and how much to spend on other public good (Y ). Assume that the other public good is the numeraire good and its price is equal 1. The price of a public employee cons ists of two itemswage and ERR on the pension fund investments. Hence, the govern ment faces the budget constraint ,, npwrYT (4-8) where p( w,r ) is the price of a public employ ee, that depends on the wage ( w ) and the ERR on investments in pension funds ( r ). The higher is share of stocks the higher is ERR and less investments per public employee is required in th e first period. Therefor e price of a public employee increases in w and decreases in r : ,, 0,0. pwrpwr wr (4-9) The optimal number of employees (n*) is determined by the tangency of the budget constraint and the indifference curve that occurs at point A in Figure 4-1. So, the optimal number of public employees is nA, and the optimal level of other public good production is YA. However, if instead
134 of using the true relationship be tween risk and rate of return 1,sra a false relationship 2,sra is used ( a2> a1), then the price of a public employ ee decreases and the budget constraint rotates to the right since it becomes possible to hire more public employees with the same tax base. Therefore, in the case of an overestimat ed ERR, the tangency point would be at point B on Figure 4-1, and it would provide a higher level of utility; more public employees will be hired (nB) and lower levels of other public goods ( YB) will be produced due to the substitution effect, assumed here to outweigh the income effect. Hence, with a low ERR employers would have to invest more in the pension fund to provide enough funds to finance the future retire ment benefit payments. That could be achieved either through increasing taxes or through decr easing the number of public employees or their wages. Both of these options could harm current public employees. On th e other hand, retirement benefits for the majority of public pension funds are guaranteed by state and local governments, so in the case of insu fficient funding of retirement benef its, public employees still get their benefits in full. Therefore, labo r unions representing public employ ees seek a high rate of return. When public pension funds are dominated by labor unions, they may tend to underestimate risk. As a result, investments in stocks will be hi gher than optimal and ERR will be overestimated. The next sections provide an empirical test of this hypothesis. Using a proxy measure of the union power, I estimate its effect on the ERR. If there is a positive correlation between labor union power and the ERR used by pension funds, then it is likely that public pension funds use an overestimated ERR because of pressure coming from labor unions. Data Description The data for the hypothes is testing comes from the United States Census of State and Local Government Employee-Retirement Systems 2007 Survey. The survey contains
135 information on the receipts and payments to th e system including employee contributions and the structure of holdings and investments. Unfortuna tely, there is no information about the ERR, which is required for testing the hypothesis. Ther efore, I constructed this variable using data on the holdings and investments structure of public pension funds and histor ical estimates of the long-term rates of return fo r different asset classes. Usually 4 different classes of assets are dis tinguished: money market or cash equivalents, bonds, stocks or equities, and real property. The risk and rate of return are positively correlated, and as the rate of return increases the risk grows. A summary of historical returns by class of asset is provided in Table 4-1 (Ibbotson, 1997). The historical statistics for government bonds and corporate stocks performance is also summa rized in Table 4-2 (Siegel, 2007). Apparently, stock market returns, long-term bonds, and shor t-term bonds in Siegel (2007) correspond to the DJI stocks, bonds, and T-Bills in Ibbotson (1997) respectively. To construct the ERR for the retirement systems participating in the Census survey, I assign different types of holdings and investments of the pension fund to the corres ponding asset class. Table 4-4 illustrates the assignments of assets to different categories. The ERR for the retirement systems is calculated using the following formula: ,iiccibbiss bs ippivvimmERRERR ERR ERR ERRERRERR (4-10) where ij shows share of asset j in the total holdings and investments of the retirement system i The classes of assets are denot ed by the following indexes: c cash, short-term investments, b bonds, s big and medium cup stocks, p real property, v small cap stocks, and m mortgage. ERR for mortgage is a little bit compli cated issue because usually a mortgage pool is divided into tranches with different rates of re turn and different risks. However, pension funds
136 are restricted from investing in highly risky asse ts, hence they are likely to hold mortgages with a risk level not exceeding high cap corporate stocks The ERR for mortgages was calculated as the average of the ERR for bonds and high and medium cap stocks. The data for the labor union power, politics and other state characteristics come s from the Statistical Abstract of the United States and Department of Labor statistics. Since I am constructing the ERR using historical estimates, there is a risk that the results of my estimations are driven by the particular design of the ERR constr uction. To confirm my results, I also run the same model using other definitions of the dependent variable that measures the relative risk of the pension fund portfolio. These two measures include ERR and average risk of the portfolio that are calculated using Wilshi res Asset Class Assumptions (Bonafede et al., 2009). Table 4-3 illustrates the rate of return and risk by differe nt classes of assets that are assumed in this report and Table 4-5 shows the assignments of assets to different categories according to Wilshires classificat ion. The average risk measure is constructed using the same logic as in Equation 4-18, but with exp ected risk estimates instead of ERR. Empirical Strategy In the empirical section of this chapter, I analyze the influence of labor unions on the ERR. Therefore, the empirical model for hypothesis testing is 0,iciuipiji ji jERRContrUnionPoliticsae (4-11) where Contri is the share of the employee contribution in total contributions of the retirement system i Unioni the measure of the labor union power in the state where the retirement system i is located, Politici is a measure of the politic preferences of the states population and state government, and aij a set of other state characteristics th at might affect the investment strategy of the retirement system. The variable of main interest is Union In the case when the dependent
137 variable is relative risk based on Wilshires Asset Class Assumptions, rather than ERR, the equation remains the same. The hypothesis is tested by the value of the coefficient Union. If pension funds located in the states with rela tively high union power have on average a higher ERR than funds located in states where unions are relatively weak, then is it is likely that the ERR is overestimated because of the pressure of labor unions. Results I use two proxies for labor union power: a du mmy variable for a right to work law and the percentage of union membership in the state. I ran regressions including each of these two variables separately and also a regression with both of these variables included. Those are the variables of main interest. The hypothesis of the labor union effect on the ERR is supported if the union membership and right to work laws are statistically signifi cant with positive and negative signs respectively. Union membership is likely to be higher in states where labor unions are relatively strong. Likewise, if a state has a right to work law, then unions are likely to be relatively weak in that state. As explanatory variables, I include two pension fund characteristics and several state characteristics that could affect the future ta x income of the state or that characterize which political party is dominant. Summary statistics for variables used in the analysis are shown in Table 4-6. I used data for the latest two available years (2006 and 2007) because the 2007 data has one strange feature: about 50% of all observations come from Pennsylvania state and local pension funds. To deal with this strange irregularity of the data, I also run the same model for the 2006 data set that has about 20% of its observa tions coming from Pennsylvania. I also ran regressions for both time periods without Pennsyl vania and weighted obse rvations by the total number of fund members to make sure that th e results are not driven by observations coming from a single state. I do not report results for th e regressions runs for the sample that excludes
138 Pennsylvania because when the weights are used, these results are similar to the regression results for the whole sample. Table 4-7 shows the results of the OLS regr essions estimations for the ERR constructed using historical data for the rates of return. As a proxy for labor union power in this regression I include the right to work variab le. For both years, the variable measuring labor union power is statistically significant and has the expected sign: negative. The eff ect of a right to work law is relatively large: it decreases ERR by approximately 24 percentage points of the standard deviation. Other explanatory vari ables also show the expected re sults. The increasing impact of the total number of fund members can be explained by the risk-pooling effect. Table 4-8 repeats the same analysis but with another measure of labor union power state union membership. Again for both years, the variable of main interest is statistically significant and has the expected negative sign. The union membership variable is significant at 5% level, and its effect is more pronounced than the effect of right to work laws. It increases ERR by approximately 39 percentage points of the st andard deviation. Other variables also have the expected signs. The positive effects of the total number of fund memb ers and state population reflect the risk-pooling effect. Table 4-9 shows results for th e OLS regression with both union membership and right to work law variables incl uded. In this setting union membership is not statistically significant, however the right to work law has a highe r magnitude in this regression compared to the results presented in Table 4-7. The lack of statistical significance of the union membership variable might be explained by the high negative correlation between union membership and right to work variables. Then, I run the same sets of regressions with alternative definitions of the dependent variable. As these alternative definitions I use ERR and average portfolio risk that are defined
139 using Wilshires Asset Class Assumptions. Tables 4-10, 4-11, and 4-12 show the results of these estimations for ERR and Tables 4-13, 4-14, and 4-15 for average risk. Both measures of labor union power retain the same signs and the same pattern of statisti cal significance in all regressions, with only one excepti on: the union membership is not statistically significant in estimations applied to 2007 data. In three out of six regressions I found a negative effect of the state net migration on the ERR and average risk for the 2007 data set. Th e states with high net migration might expect higher fu ture tax revenues and thus mi ght be less concerned by current underfunding of the public pension funds. However, this result is not stable across different specifications. Hence, empirical results suppor t the hypothesis of labor union influence on pension funds which encourages funds to use an overestimated ERR and make risky investments. However, the model is very rough since the data set does not provide the ERR. I had to construct the ERR according to the pension funds asset structure. Th erefore, the dependent variable in my study is not precisely defined. Also my measures of labor union power are not perfect. However, my results show that the union effect is quite possible and provides mo tivation for further research in this area. Conclusions In this chapter, I first reviewed current debate s over the discount rate that should be used for valuation of public pension funds future earn ings on investments. Some financial economists argue that future earnings of public pension funds should be valued using rates of return on fixed income securities, while the actuarial pension pr actice is to use market rates of return. This discussion is similar to the debate over the appr opriate discount rates for evaluation of future costs caused by global warming. Participants of these debates provide solid arguments in favor of using the market rate instead of the risk-free rate. Most importantly, additional contributions
140 into pension funds that would be required in the case of low discount rates would cause high opportunity cost because otherwise these assets could have been i nvested with a higher rate of return. Then, using a tax-smoothing model I illustrate how lack of transparency regarding the real relationship between rate of return and ri sk may cause deviations from optimality and why labor unions might want pension funds to use an overestimat ed rate of return. I look at the possible effect of the influence of the labor un ions on the pension funds investment strategies. Finally, I test the hypothesis of union influence empirically. Th e OLS estimation results confirm that the pressure of labor unions that desire high rates of retu rn may cause deviations from optimal investment strategy.
141 Table 4-1. 1926-1996 average annual rates of return Asset class Nominal rate of return (%) Real rate of return (%) Small Cap Stocks 12.5 9.4 Real Estate 11.1 8.0 DJI 10.0 6.9 Bonds 5.2 2.1 T-Bills 3.7 0.6 Inflation 3.1 (Source: Ibbotson (1997), p. 25) Table 4-2. 1926-2001 average annual rates of return Asset class Nominal rate of return (%) Real rate of return (%) Stock market 10.02 6.90 Long-term government bonds 5.20 2.20 Long-term government bonds 3.90 0.70 (Source: Siegel (2007), p.13., p.15.) Table 4-3. Wilshire asset class assumptions Asset class Rate of return (%) Risk (%) U.S. Equity 8.50 16.0 Non-U.S. Equity 8.50 17.0 Private Equity 11.55 26.0 Real Estate 7.00 15.0 U.S. Bonds 4.00 5.0 Non-U.S. Bonds 3.75 10.0 (Source: Bonafede et al. (2009), p.12.)
142 Table 4-4. Assignment of survey items for inve stments and holdings to general asset classes Variable description Cash, Short-term government bonds Bonds Big and medium cap Real Property Small cap Mortgage Expected rate of return 0.6 2.1 6.9 8.0 9.4 4.5 Cash, short-term investments X Federal securities X State and Local Gov. securities X Corporate bonds (bonds and mortgage-backed securities issued by FHLB, FHLMC, FNMA, Farm credit banks, and SLMA) X Other corporate bonds (include debentures, convertible bonds, and railroad equipment certificates) X Other securities (shares in mutual funds, conditional sales contracts, direct loans, loans to members, etc.) X Investments trusts (shares in funds administered by private agencies, governmental investment accounts) X Foreign and international securities (include corporate stocks and corporate equities) X Mortgages X Real property X Corporate stocks (include common and preferred stocks, warrants) X Other Investments (include venture capital, partnerships, real estate investment trusts, and leveraged buyouts) X
143 Table 4-5. Assignment of survey items for inve stments and holdings to general asset classes using Wilshire asset class assumptions Variable description U.S. Bonds U.S. Equity NonU.S. Equity Real Estate Private. Equity Mortgage ERR (%) Risk (%) 4.00 5.00 8.50 16.00 8.50 17.00 7.00 15.00 11.55 26.00 6.25 9.50 Federal securities X State and Local Gov. securities X Corporate bonds (bonds and mortgage-backed securities issued by FHLB, FHLMC, FNMA, Farm credit banks, and SLMA) X Other corporate bonds (include debentures, convertible bonds, and railroad equipment certificates) X Other securities (shares in mutual funds, conditional sales contracts, direct loans, loans to members, etc.) X Investments trusts (shares in funds administered by private agencies, governmental investment accounts) X Foreign and international securities (include corporate stocks and corporate equities) X Mortgages X Real property X Corporate stocks (include common and preferred stocks, warrants) X Other Investments (include venture capital, partnerships, real estate investment trusts, and leveraged buyouts) X
144 Table 4-6. Summary statistics 2007 2006 Variable Mean Std. Dev. Mean Std. Dev. ERR based on historical data 5.82 0.69 5.75 0.68 ERR based on Wilshire asset cl ass assumptions7.39 0.73 7.34 0.70 Average risk based on Wilshire asset class assumptions 13.85 1.82 13.99 1.78 Employee contribution dummy (0 if contribution=0) 0.992 0.087 0.916 0.278 Log of total members 12.183 1.460 6.215 2.270 Right to work law 0.372 0.483 0.575 0.495 Union membership (%) 13.48 6.44 13.44 6.48 Share of population with bachelor degree 0.278 0.046 0.280 0.035 Share of population liv ing in metropolitan areas 0.822 0.140 0.892 0.083 Birth to death ratio 1.790 0.455 1.409 0.341 Share of net migration in total population of state 0.023 0.045 0.062 0.052 Low House: Republicans to Democrats 1.289 1.038 1.027 1.441 Log of population 15.000 1.100 14.60 1.100 Number of observations 1720 782 Table 4-7. OLS regressions results for ERR based on the historical data and right to work law as a measure of labor union power 2007 2006 Variable std. dev. std. dev. Employee contribution dummy (0 if contribution=0) 0.18 0.24 -0.04 0.33 Log of total members 0.06* 0.03 0.08*** 0.03 Right to work law -0.46*** 0.14 -0.49*** 0.14 Share of population with bachelor degree 0.18 1.64 1.58 1.69 Share of population living in metropolitan areas -0.03 0.54 -0.33 0.54 Birth to death ratio 0.11 0.14 -0.05 0.16 Share of net migration in total popul ation of state -1.26 1.10 -0.17 1.33 Low House: Republicans to Democrats 0.09 0.05 0.06 0.06 Log of population 0.06 0.06 0.05 0.05 Constant 3.89 1.07 4.21 0.99 Number of observations 1720 782 Weights yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively
145 Table 4-8. OLS regressions results for ERR based on the historical data and state union membership as a measure of labor union power 2007 2006 Variable std. dev. std. dev. Employee contribution dummy (0 if contribution=0) 0.15 0.25 -0.09 0.35 Log of total members 0.06 0.03 0.07** 0.03 State union membership 0.02** 0.01 0.03** 0.01 Share of population with bachelor degree 0.77 1.61 2.16 1.63 Share of population living in metropolitan areas 0.00 0.60 -0.28 0.62 Birth to death ratio 0.10 0.13 -0.06 0.15 Share of net migration in total popul ation of state -1.36 1.05 -0.33 1.36 Low House: Republicans to Democrats 0.08 0.06 0.06 0.06 Log of population 0.11* 0.06 0.10* 0.06 Constant 2.57 1.07 2.68 1.04 Number of observations 1720 782 Weights yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively Table 4-9. OLS regressions results for ERR based on the historical data and right to work law and state union membership as a measure of labor union power 2007 2006 Variable std. dev. std. dev. Employee contribution dummy (0 if contribution=0) 0.20 0.24 -0.03 0.33 Log of total members 0.06* 0.03 0.08*** 0.03 State union membership -0.01 0.02 -0.01 0.01 Right to work law -0.55*** 0.20 -0.59*** 0.18 Share of population with bachelor degree 0.26 1.62 1.66 1.66 Share of population living in metropolitan areas 0.04 0.55 -0.25 0.55 Birth to death ratio 0.10 0.13 -0.06 0.16 Share of net migration in total popul ation of state -1.75 1.11 -0.72 1.44 Low House: Republicans to Democrats 0.08 0.05 0.05 0.05 Log of population 0.05 0.06 0.04 0.06 Constant 4.13 1.22 3.95 1.12 Number of observations 1720 782 Weights yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively
146 Table 4-10. OLS regressions results for ERR ba sed on Wilshire asset class assumptions and right to work law as a measure of labor union power 2007 2006 Variable std. dev. std. dev. Employee contribution dummy (0 if contribution=0) 0.23 0.28 0.00 0.40 Log of total members 0.07** 0.03 0.08*** 0.03 Right to work law -0.44*** 0.15 -0.49*** 0.15 Share of population with bachelor degree 0.97 1.69 2.02 1.72 Share of population living in metropolitan areas -0.34 0.56 -0.50 0.54 Birth to death ratio 0.15 0.14 -0.06 0.13 Share of net migration in total popul ation of state -1.55 1.06 0.03 1.24 Low House: Republicans to Democrats 0.09 0.05 0.05 0.05 Log of population 0.07 0.06 0.06 0.05 Constant 5.11 1.13 5.48 1.01 Number of observations 1720 782 Weights yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively Table 4-11. OLS regressions results for ERR ba sed on Wilshire asset class assumptions and state union membership as a m easure of labor union power 2007 2006 Variable std. dev. std. dev. Employee contribution dummy (0 if contribution=0) 0.20 0.29 -0.04 0.41 Log of total members 0.06* 0.04 0.08*** 0.03 State union membership 0.02 0.01 0.02** 0.01 Share of population with bachelor degree 1.66 1.69 2.64 1.69 Share of population living in metropolitan areas -0.25 0.62 -0.44 0.62 Birth to death ratio 0.13 0.14 -0.07 0.12 Share of net migration in total popul ation of state -2.02* 1.07 -0.27 1.33 Low House: Republicans to Democrats 0.08 0.06 0.05 0.06 Log of population 0.12** 0.06 0.11* 0.06 Constant 3.85 1.09 3.98 1.05 Number of observations 1720 782 Weights yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively
147 Table 4-12. OLS regressions results for ERR ba sed on Wilshire asset class assumptions and right to work law and state union member ship as a measure of labor union power 2007 2006 Variable std. dev. std. dev. Employee contribution dummy (0 if contribution=0) 0.26 0.28 0.02 0.39 Log of total members 0.07** 0.03 0.09*** 0.03 State union membership -0.02 0.02 -0.01 0.01 Right to work law -0.6*** 0.21 -0.61*** 0.18 Share of population with bachelor degree 1.11 1.67 2.12 1.69 Share of population living in metropolitan areas -0.21 0.57 -0.40 0.54 Birth to death ratio 0.13 0.13 -0.07 0.13 Share of net migration in total popul ation of state -2.45** 1.06 -0.68 1.39 Low House: Republicans to Democrats 0.08 0.05 0.05 0.05 Log of population 0.06 0.06 0.04 0.06 Constant 5.56 1.29 5.08 1.16 Number of observations 1720 782 Weights yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively Table 4-13. OLS regressions results for av erage risk based on Wilshire asset class assumptions and right to work law as a measure of labor union power 2007 2006 Variable std. dev. std. dev. Employee contribution dummy (0 if contribution=0) 0.23 0.28 0.00 0.40 Log of total members 0.07** 0.03 0.08*** 0.03 Right to work law -0.44*** 0.15 -0.49*** 0.15 Share of population with bachelor degree 0.97 1.69 2.02 1.72 Share of population living in metropolitan areas -0.34 0.56 -0.50 0.54 Birth to death ratio 0.15 0.14 -0.06 0.13 Share of net migration in total popul ation of state -1.55 1.06 0.03 1.24 Low House: Republicans to Democrats 0.09 0.05 0.05 0.05 Log of population 0.07 0.06 0.06 0.05 Constant 5.11 1.13 5.48 1.01 Number of observations 1720 782 Weights yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively
148 Table 4-14. OLS regressions results for average risk based on Wilshire asset class assumptions and state union membership as a measure of labor union power 2007 2006 Variable std. dev. std. dev. Employee contribution dummy (0 if contribution=0) 0.20 0.29 -0.04 0.41 Log of total members 0.06* 0.04 0.08*** 0.03 State union membership 0.02 0.01 0.02** 0.01 Share of population with bachelor degree 1.66 1.69 2.64 1.69 Share of population living in metropolitan areas -0.25 0.62 -0.44 0.62 Birth to death ratio 0.13 0.14 -0.07 0.12 Share of net migration in total popul ation of state -2.02* 1.07 -0.27 1.33 Low House: Republicans to Democrats 0.08 0.06 0.05 0.06 Log of population 0.12** 0.06 0.11* 0.06 Constant 3.85 1.09 3.98 1.05 Number of observations 1720 782 Weights yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively Table 4-15. OLS regressions results for average risk based on Wilshire asset class assumptions and right to work law and state union me mbership as a measure of labor union power 2007 2006 Variable std. dev. std. dev. Employee contribution dummy (0 if contribution=0) 0.26 0.28 0.02 0.39 Log of total members 0.07** 0.03 0.09*** 0.03 State union membership -0.02 0.02 -0.01 0.01 Right to work law -0.6*** 0.21 -0.61*** 0.18 Share of population with bachelor degree 1.11 1.67 2.12 1.69 Share of population living in metropolitan areas -0.21 0.57 -0.40 0.54 Birth to death ratio 0.13 0.13 -0.07 0.13 Share of net migration in total popul ation of state -2.45** 1.06 -0.68 1.39 Low House: Republicans to Democrats 0.08 0.05 0.05 0.05 Log of population 0.06 0.06 0.04 0.06 Constant 5.56 1.29 5.08 1.16 Number of observations 1720 782 Weights yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively
149 Figure 4-1. Overestimation of ERR leads to hiring more public employees than optimal. A B n Y T nBnA T T YA YB U2 U1
150 CHAPTER 5 CONCLUSION My dissertation investigated th ree different economic policy issues. I analyzed the effects of two recent educational reforms in US and its effects on the teachers labor market. I showed that attrition rates for teachers who came to the profession through alterna tive routes are different from attrition rates of teacher s who came to the profession thr ough traditional pathways only for novice teachers with less than 2 years of experi ence. I estimated the relative efficiency of alternative programs and using simulation model have illustrated how to calculate the upper bound of alternative programs training costs. Then I examined the effect of the school accoun tability on the supply and demand of teachers. I constructed the set of variables describing different t ypes if accountability policies. I distinguished three main types: school-targe ted accountability, school-targeted and districttargeted. I revealed no results of student-targeted policies on the teachers. School-targeted policies do affect teachers but only in the large districts and its e ffects vary across subject fields and across types of schools. Distri ct-targeted policies do affect teac hers supply in a positive way. In the fourth chapter I analyzed possible devi ations from optimal investment strategy for public pension funds due to influence of labor unions and illustrated how lack of transparency shifts the results from optimality. Then I te sted the hypothesis that because of lack of transparency labor unions encourage pension funds to make too risky investments. I found some evidence that labor unions do in f act affect pension funds invest ment strategies and the more powerful are labor unions in the state the more risky are the investments of the state and local public pension funds.
151 APPENDIX A DETAILED RESULTS FOR TEAC HER ATTRITION ESTIMATION Table A-1. Summary statistics Diagonal Definition 1 Definition 2 Variable Definition of variable Mean Std. Mean Std. Mean Std. Schools characteristics Non-white students Ratio of minority students 0.358 0.342 0.364 0.344 0.365 0.342 Non-white teachers Ratio of minority teach ers 0.143 0.21 6 0.145 0.216 0.145 0.216 Poor Ratio of students eligible for free lunch 0.381 0.279 0.383 0.279 0.385 0.280 Flag poor =1 if missing observation for poor 0.012 0.110 0.012 0.107 0.011 0.105 Urban =1 if school in bigor mid-size city 0.784 0.412 0.790 0.408 0.787 0.409 Districts characteristics Average size of the school in the district District enrollment/Number of schools in the district 584.4 264.7 587.7 264.2 589.2 268.0 Teachers characteristics Exit =1 if exit MA field next year 0.180 0.384 0.182 0.386 0.181 0.385 Year =1 if 2003-04 school year 0.512 0.500 0.523 0.500 0.524 0.499 Married =1 if married 0.732 0.443 0.725 0.447 0.723 0.447 Master =1 if has masters degree 0.501 0.500 0.494 0.500 0.488 0.500 Master flag =1 if missing observation for master 0.007 0.086 0.007 0.083 0.007 0.084 Female =1 if teacher is female 0.767 0.423 0.765 0.424 0.769 0.421 Non-white =1 if teacher is non-white 0.095 0.293 0.100 0.300 0.098 0.297 Experience Total experience (years) 16.9 10.0 16.1 10.1 15.6 10.0 Experience squared 385.5 371.0 361.1 368.7 343.7 365.0 Entry age At what age first time entere d teaching 26.0 3 6.60 26.13 6. 60 26.17 6.62 Union =1 if teacher is member of union 0.806 0.39 5 0.805 0.396 0.802 0.398 Special education =1 if special education is MA field 0.114 0.317 0.115 0.320 0.118 0.322 English =1 if MA field is English language arts 0.099 0.299 0.101 0.302 0.099 0.298 Mathematics =1 if MA field is mathematics 0.073 0.260 0.071 0.257 0.069 0.254 Sciences =1 if MA field is natural sciences 0.065 0.247 0.064 0.245 0.062 0.242 Social Sciences =1 if MA field is social sciences 0.058 0.234 0.060 0.237 0.061 0.239 Other fields =1 if MA field is elementary education 0.321 0.467 0.320 0.466 0.326 0.469 Regular second =1 if additional traditional certificate 0.292 0.455 0.284 0.451 0.285 0.452 Salary Real salary ($10000) 0.212 0.409 0.216 0.411 0.213 0.410 Alternative =1 if alternative certificate in MA field 0.032 0.177 0.031 0.174 0.030 0.169 Exp1 =1 if experience is 1 year 0.036 0.186 0.040 0.196 0.044 0.206 Exp2 =1 if experience is 2 years 0.032 0.175 0.036 0.187 0.039 0.193 Exp3 =1 if experience is 3 years 0.032 0.176 0.042 0.200 0.041 0.199
152 Table A-1. Continued Diagonal Definition 1 Definition 2 Variable Definition of variable Mean Std. Mean Std. Mean Std. Exp4 =1 if experience is 4 yeas 0.034 0.181 0.038 0.192 0.042 0.201 Exp5 =1 if experience is 5 years 0.659 0.474 0.661 0.474 0.655 0.476 College =1 if highly competitive college 0.049 0.217 0.047 0.212 0.055 0.228 Flag college =1 if missing observations for college 4.200 1.147 4.148 1.139 4.124 1.134 Income35 =1 if family income is less than $3500 0.060 0.237 0.064 0.244 0.066 0.247 Income50 =1 if $35000
153 Table A-2. Results of logit regression estimation Diagonal Definition Definition 1 Definition 2 Coefficient Std. error Hazard ratio Std. error Hazard ratio Std. error Hazard ratio Alt. certificate in MA -0.18 0.16 0.83 -0.15 0.15 0.86 -0.13 0.15 0.87 Alt. cert. X Exp.=1 1.18** 0.53 3.24** 1.16** 0.51 3.18** 1.14** 0.50 3.13** Alt. cert. X Exp.=2 0.03 0.56 1.03 0.26 0.51 1.29 0.15 0.47 1.17 Alt. cert. X Exp.=3 0.58 0.48 1.79 0.43 0.47 1.54 0.36 0.45 1.43 Alt. cert. X Exp.=4 1.06 0.67 2.89 0.22 0.62 1.24 -0.05 0.61 0.95 Alt. cert. X Exp.=5 0.49 0.98 1.64 -0.01 0.92 0.99 0.40 0.74 1.50 Exp.=1 year -0.42 0.40 0.65 -0.45 0.36 0.64 -0.50 0.35 0.61 Exp.=2 years -0.69* 0.38 0.5* -0.72** 0.34 0.49** -0.57* 0.32 0.57* Exp.=3 years -0.18 0.35 0.83 -0.09 0.31 0.91 -0.41 0.30 0.67 Exp.=4 years -0.1 0.40 0.9 0.08 0.37 1.09 0.30 0.34 1.34 Exp.=5 years 0.21 0.45 1.24 0.44 0.36 1.55 -0.00 0.37 0.99 Reg. cert. in other field -0.04 0.15 0.96 0.02 0.14 1.03 0.13 0.14 1.14 Reg. cert. in other field X MA field is sp. ed. 0.37 0.37 1.45 0.44 0.36 1.55 0.27 0.35 1.31 Special Education -0.49* 0.26 0.61* -0.61***0.26 0.54*** -0.51** 0.24 0.60** English 0.88*** 0.20 2.4*** 0.89*** 0.19 2.44*** 0.90*** 0.18 2.47*** Math 0.39 0.24 1.48 0.43* 0.24 1.53* 0.40* 0.23 1.49* Sciences -0.08 0.30 0.92 -0.02 0.28 0.98 -0.08 0.27 0.92 SocSc 0.57** 0.27 1.77** 0.60*** 0.25 1.82*** 0.53** 0.25 1.7** Other 0.82*** 0.17 2.27*** 0.80*** 0.16 2.22*** 0.73*** 0.16 2.08*** Year -0.19* 0.12 0.82* -0.16 0.11 0.85 -0.15 0.11 0.86 Married 0.33 0.27 1.39 0.20 0.25 1.22 0.00 0.25 1.00 Married female -0.46 0.29 0.63 -0.38 0.28 0.68 -0.10 0.28 0.91 Master 0.15 0.14 1.16 0.15 0.13 1.16 0.23* 0.13 1.26* Master flag 0.78 0.60 2.17 0.78 0.60 2.19 0.76 0.60 2.13 Female 0.69*** 0.27 1.99*** 0.62*** 0.26 1.86*** 0.34 0.25 1.40 Non-white 0.09 0.20 1.098 0.001 0.19 1.001 -0.06 0.19 0.95 Experience -0.07* 0.04 0.94* -0.07* 0.03 0.94* -0.08** 0.03 0.93** Experience sq. 0.01*** 0.00 1.01*** 0.01*** 0.00 1.01*** 0.01*** 0.00 1.01*** Age first time teaching -0.00 0. 01 1.00 0.00 0.01 1. 00 0.001 0.01 1.00 Union member -0.17 0.16 0.84 -0.19 0.15 0.83 -0.18 0.14 0.84 Real salary 0.01 0.07 1.01 -0.00 0.07 0.99 -0.01 0.07 0.99 Minority students 0.08 0.29 1.09 0.00 0.28 1.002 0.02 0.27 1.02 Minority teachers 0. 18 0.40 1.20 0.24 0.39 1.27 0. 28 0.36 1.32 Free lunch eligible 0.72*** 0.31 2.06*** 0.66** 0.29 1.94** 0.63** 0.28 1.88** Flag for free lunch 0.29 0.46 1.33 0.26 0.45 1.3 0.27 0.44 1.31 Av. school size -0.00 0.00 0.99 -0.00 0.00 0.99 -0.00 0.00 0.99 Urban 0.07 0.16 1.07 0.06 0.15 1.06 0.03 0.15 1.03 Constant -1.81*** 0.70 0.16*** -1.73 0.65 0.18 -1.58 0.64 0.21 State fixed effects yes yes yes # of observations 5224 5712 6020 *, **, *** statistically significant at 10, 5, and 1% level respectively.
154 Table A-3. Wald test for joint significance F-statistic Hypothesis Diagonal Def. 1 Def.2 Alt Exp. 2=0 Alt Exp. 3=0 Alt Exp. 4=0 Alt Exp. 5=0 P>F=0.42 P>F=0.89 P>F=0.92 Alt Exp. 3=0 Alt Exp. 4=0 Alt Exp. 5=0 P>F=0.28 P>F=0.81 P>F=0.83 Alt Exp. 4=0 Alt Exp. 5=0 P>F=0.26 P>F=0.94 P>F=0.86
155 Table A-4. Summary statistics for teacher s with 1 year of experience (weighted) 2003-04 1999-00 Alternative Traditional Alternative Traditional Personal characteristic Mean Std. dev.Mean Std. dev.Mean Std. dev. Mean Std. dev. Age 30.47 0.73 28.27 0.60 31.18 0.63 29.15 0.42 Teacher white 0.87 0.03 0. 91 0.02 0.86 0. 03 0.92 0.01 Teacher black 0.11 0.03 0. 07 0.02 0.07 0. 02 0.06 0.01 Entry age 29.47 0.73 27.27 0.60 30.17 0.63 28.14 0.42 Female 0.69 0.04 0.81 0.02 0.69 0.04 0.77 0.02 Bachelor degree 0.99 0.00 1.00 0.00 0.98 0.01 1.00 0.00 Master degree 0.14 0.02 0.15 0.02 0.12 0.02 0.16 0.02 Associates degree 0.08 0.02 0.18 0.03 0.14 0.03 0.07 0.01 College quality (index 1-5) 3.01 0.11 2.74 0.05 2.82 0.06 2.77 0.04 Education specialist or adv. grad. study 0.02 0.01 0.01 0.01 0.04 0.02 0.03 0.01 Married 0.33 0.07 0.42 0.06 0.36 0.09 0.51 0.06 Widowed, divorced, separated^ 0.02 0.02 0.03 0.02 0.09 0.05 0.06 0.02 Never married ^ 0.65 0.07 0.55 0.06 0.55 0.10 0.43 0.06 Family income (category: 1-5) ^ 2.57 0.20 2.59 0.15 2.08 0.18 2.27 0.13 Family size ^ 2.18 0.19 2.42 0.14 n/a n/a n/a n/a Kids under of age of 5^ 0.22 0.07 0.32 0.06 0.07 0.05 0.20 0.06 Number of obs. (SASS/TFS) 359/89 909/188 326/59 597/123 ^ data and sample weights from TFS Table A-5. t-tests for difference in mean valu es of personal characteristics of ARC and TRC teachers with 1 year of experience (weighted) 2003-04 1999-00 D-in-D Personal characteristic t-st. t-st. t-st. Age 2.20** 2.34 2.03*** 2.67 0.17 0.14 Teacher white -0.04 -1.3 3 -0.06 -1.5 5 0.02 0.32 Teacher black 0.04 1. 49 0.02 0.64 0.03 0.74 Entry age 2.20*** 2.33 2.04*** 2.69 0.16 0.13 Female -0.12*** 2.62 -0.08* 1.80 -0.05 0.71 Bachelor degree -0.01** -2.21 -0.02*** -2.74 0.01 1.61 Master degree -0.01 -0.41 -0.05 -1.46 0.03 0.69 Associates degree -0.11*** -3.40 0.08** 2.28 -0.18*** -3.99 College quality (index 1-5) 0.26** 2.09 0.05 0.64 0.21 1.47 Education specialist or advanced graduate study0.00 0.18 0.01 0.40 -0.01 -0.29 Married ^ -0.09 -0.96 -0.15 -1.36 0.06 0.44 Widowed, divorced, separated^ -0.01 -0.32 0.03 0.60 -0.04 -0.68 Never married ^ 0.10 1.03 0.12 1.00 -0.02 -0.14 Family income (category: 1-5) ^ -0.14 -0.61 -0.27 -1.31 0.12 0.39 Family size ^ -0.24 -1.03 n/a n/a n/a n/a Kids under of age of 5^ -0.10 -1.07 -0.13* -1.71 0.03 0.24 Number of obs. (SASS/TFS) 956/205 1235/247 1235/452 ^ data and sample weights from TFS; ***,**,* statistically significant at 1,5,10% respectively
156 ^ data and sample weights from TFS Table A-7. t-tests for difference in mean valu es of personal characteristics of ARC and TRC teachers with less than 6 years of experience (weighted) 2003-04 1999-00 D-in-D Personal characteristic t-st. t-st. t-st. Age 1.84*** 4.57 2.31*** 6.23 -0.47 -0.86 Teacher white -0.07*** -4.16 -0.03* -1.71 -0.05* -1.89 Teacher black 0.07*** 4. 04 0.02* 1.73 0.05** 2.06 Entry age 2.17*** 5.43 2.39*** 6.64 -0.23 -0.42 Female -0.05*** -2.70 -0.08*** -4.07 0.03 1.02 Bachelor degree -0.02*** -4.20 -0.02*** -4.66 0.00 -0.56 Master degree -0.05*** -2.39 -0.02 -1.06 -0.03 -1.10 Associates degree -0.02 -1.24 0.03** 2.03 -0.05** -2.26 College quality (index 1-5) 0.14*** 2.94 0.11*** 3.03 0.03 0.48 Education specialist or advanced graduate study0.00 0.23 0.00 -0.33 0.00 0.40 Married ^ 0.02 0.38 0.02 0.03 -0.002 -0.02 Widowed, divorced, separated^ 0.04 1.23 0.04 1.23 0.07* 1.80 Never married ^ -0.06 -1.14 -0.06 -1.14 -0.07 -0.85 Family income (category: 1-5) ^ 0.08 0.62 -0.04 -0.32 0.12 0.66 Family size ^ 0.16 1.13 n/a n/a n/a n/a Kids under of age of 5^ 0.00 -0.05 0.00 -0.05 0.01 0.13 Number of obs. (SASS/TFS) 5918/987 6525/971 12443/1958 ^ data and sample weights from TFS; ***,**,* statistically significant at 1,5,10% respectively Table A-6. Summary statistics fo r ARC and TRC teachers with less than 6 years of experience (weighted) 2003-04 1999-00 Alternative Traditional Alternative Traditional Personal characteristic Mean Std. dev.Mean Std. dev.Mean Std. dev. Mean Std. dev. Age 32.64 0.34 30.80 0.22 33.51 0.33 31.20 0.17 Teacher white 0.83 0.02 0. 90 0.01 0.86 0. 01 0.89 0.01 Teacher black 0.14 0.02 0. 07 0.01 0.09 0. 01 0.07 0.01 Entry age 29.67 0.34 27.50 0.22 30.58 0.32 28.19 0.17 Female 0.74 0.02 0.79 0.01 0.69 0.02 0.77 0.01 Bachelor degree 0.98 0.00 1.00 0.00 0.98 0.00 1.00 0.00 Master degree 0.23 0.02 0.28 0.01 0.20 0.02 0.22 0.01 Associates degree 0.12 0.01 0.14 0.01 0.11 0.01 0.08 0.01 College quality (index 1-5) 2.92 0.04 2.78 0.02 2.89 0.03 2.78 0.02 Education specialist or adv.grad.study 0.02 0.00 0.02 0.00 0.02 0.01 0.02 0.00 Married ^ 0.51 0.04 0.49 0.03 0.65 0.06 0.63 0.03 Widowed, divorced, separated^ 0.10 0.03 0.06 0.02 0.04 0.01 0.07 0.02 Never married ^ 0.39 0.04 0.45 0.03 0.32 0.05 0.31 0.03 Family income (category: 1-5) ^ 2.88 0.10 2.81 0.07 2.45 0.09 2.46 0.06 Family size ^ 2.56 0.12 2.40 0.08 n/a n/a n/a n/a Kids under of age of 5^ 0.40 0.06 0.40 0.04 0.28 0.08 0.30 0.04 Number of obs. (SASS/TFS) 1921/335 3997/652 1604/250 4921/721
157 Table A-8. t-tests for difference in mean values of personal characteristics of alternatively certified teachers with 1 and 2 years of experience 2003-04 1999-00 D-in-D Personal characteristic t-st. t-st. t-st. Age -1.52 -1.44 -0.66 -0.78 -0.86 -0.64 Teacher white 0.01 0.24 0.03 0.64 -0 .02 -0.35 Teacher black 0.02 0.48 -0.02 -0.4 3 0.03 0.65 Entry age -0.54 -0.51 0.32 0.39 -0.87 -0.64 Female 0.05 0.94 0.08 1.53 -0.03 -0.37 Bachelor degree 0.01 0.97 -0.01 -1.40 0.02* 1.70 Master degree -0.05 -1.11 -0.06 -1.44 0.01 0.23 Associates degree -0.10** -2.26 0.03 0.84 -0.13** -2.22 College quality (index 1-5) -0.07 -0.54 -0.04 -0 .42 -0.04 -0.23 Education specialist or advanced graduate study0.01 0.46 0.03 1.40 -0.03 -1.05 Married ^ -0.05 -0.47 -0.31** -2.31 0.26 1.56 Widowed, divorced, separated^ -0.12** -2.25 0.04 0.61 -0.16** -1.96 Never married ^ 0.17 1.63 0.27** 1.98 -0.10 -0.60 Family income (category: 1-5) ^ -0.14 -0.55 -0.39 -1.42 0.24 0.64 Family size ^ -0.07 -0.27 n/a n/a n/a n/a Kids under of age of 5^ -0.06 -0.53 -0.27 -1.44 0.21 1.00 Number of obs. (SASS/TFS) 775/157 677/134 1452/291 ^ data and sample weights from TFS; ***,**,* statistically significant at 1,5,10% respectively Table A-9. t-tests for difference in mean values of personal characteristics of alternatively certified teachers with 2 and 3 years of experience 2003-04 1999-00 D-in-D Personal characteristic t-st. t-st. t-st. Age -1.02 -0.94 -1.25 -1.30 0.24 0.16 Teacher white 0.08 1.49 -0.02 -0.3 0 0.09 1.27 Teacher black -0.09* -1.8 5 -0.04 -0.93 -0.05 -0.76 Entry age 0.00 0.00 -0.24 -0.25 0.24 0.16 Female 0.05 1.09 -0.13*** -2.34 0.18*** 2.49 Bachelor degree 0.00 0.49 0.00 0.22 0.00 0.28 Master degree -0.06 -1.13 0.01 0.28 -0.07 -1.04 Associates degree 0.05 1.13 -0.03 -0.61 0.08 1.25 College quality (index 1-5) 0.14 1.24 -0.12 -1.14 0.25* 1.68 Education specialist or advanced graduate study0.00 0.20 -0.02** -2.21 0.02 1.63 Married ^ -0.07 -0.67 0.02 0.16 -0.10 -0.52 Widowed, divorced, separated^ 0.10* 1.76 0.05 1.54 0.05 0.80 Never married ^ -0.03 -0.26 -0.07 -0.48 0.04 0.24 Family income (category: 1-5) ^ -0.14 -0.53 -0.20 -0.75 0.07 0.18 Family size ^ -0.22 -0.78 n/a n/a n/a n/a Kids under of age of 5^ -0.21 -1.35 0.07 0.32 -0.28 -1.08 Number of obs. (SASS/TFS) 780/146 673/133 1453/276 ^ data and sample weights from TFS; ***,**,* statistically significant at 1,5,10% respectively
158 Table A-10. t-tests for difference in mean values of personal characteri stics of alternatively certified teachers with 3 and 4 years of experience 2003-04 1999-00 D-in-D Personal characteristic t-st. t-st. t-st. Age -0.24 -0.21 -1.99* -1.83 1.75 1.12 Teacher white -0.04 -0.6 4 -0.06 -1.3 4 0.02 0.27 Teacher black 0.04 0.69 0.05 1.18 -0 .01 -0.14 Entry age 0.74 0.66 -0.99 -0.91 1.73 1.11 Female -0.04 -0.79 0.08 1.35 -0.12 -1.53 Bachelor degree 0.01 0.53 0.01 0.68 0.00 0.04 Master degree -0.01 -0.21 -0.08 -1.56 0.07 0.83 Associates degree -0.01 -0.15 0.02 0.46 -0.03 -0.43 College quality (index 1-5) 0.28** 1.96 0.05 0.42 0.24 1.30 Education specialist or advanced graduate study-0.02* -1.81 0.01 0.86 -0.03* -1.91 Married ^ -0.30** -2.28 -0.12 -0.78 -0.17 -0.85 Widowed, divorced, separated^ -0.06 -0.64 -0.01 -0.63 -0.05 -0.54 Never married ^ 0.36*** 3.48 0.13 0.84 0.23 1.20 Family income (category: 1-5) ^ -0.10 -0.27 0.10 0.34 -0.19 -0.43 Family size ^ -0.92** -2.22 n/a n/a n/a n/a Kids under of age of 5^ 0.06 0.24 -0.16 -0.79 0.22 0.69 Number of obs. (SASS/TFS) 735/122 651/95 1386/217 ^ data and sample weights from TFS; ***,**,* statistically significant at 1,5,10% respectively Table A-11. t-tests for difference in mean values of personal characteri stics of alternatively certified teachers with 4 and 5 years of experience 2003-04 1999-00 D-in-D Personal characteristic t-st. t-st. t-st. Age -1.28 -1.20 -1.94* -1.78 0.66 0.44 Teacher white 0.01 0. 27 0.01 0.26 0.01 0.08 Teacher black -0.01 -0.2 6 0.00 0.05 -0 .02 -0.24 Entry age -0.28 -0.27 -0.94 -0.86 0.66 0.43 Female -0.01 -0.28 -0.11** -2.00 0.10 1.28 Bachelor degree 0.01 0.52 0.02 1.18 -0.01 -0.22 Master degree -0.05 -0.82 -0.06 -0.97 0.01 0.10 Associates degree 0.00 0.09 0.05 1.33 -0.04 -0.75 College quality (index 1-5) -0.24 -1.78 0.02 0.22 -0.26 -1.53 Education specialist or advanced graduate study0.00 0.19 0.01* 1.80 -0.01 -0.59 Married ^ 0.06 0.41 0.11 0.50 -0.05 -0.19 Widowed, divorced, separated^ -0.09 -0.67 0.01 0.89 -0.10 -0.76 Never married ^ 0.02 0.21 -0.13 -0.56 0.15 0.60 Family income (category: 1-5) ^ -0.54 -1.36 0.09 0.22 -0.63 -1.09 Family size ^ 0.75* 1.72 n/a n/a n/a n/a Kids under of age of 5^ -0.18 -0.77 0.38** 2.07 -0.56 -1.88 Number of obs. (SASS/TFS) 782/107 600/58 ^ data and sample weights from TFS; ***,**,* statistically significant at 1,5,10% respectively
159 Table A-12. Results of logit regression estimation Diagonal Definition Definition 1 Definition 2 Variable Std. error Hazard ratio Std. error Hazard ratio Std. error Hazard ratio Alt. certificate in MA -0.15 0.15 0.86 -0.19 0.16 0.82 -0.16 0.15 0.85 Alt. cert. X Exp.=1 1.16** 0.51 3.18** 1.15** 0.52 3.16** 1.13** 0.50 3.08** Alt. cert. X Exp.=2 0.26 0.51 1.29 0.31 0.53 1.37 0.22 0.47 1.24 Alt. cert. X Exp.=3 0.43 0.47 1.54 0.48 0.48 1.61 0.43 0.46 1.53 Alt. cert. X Exp.=4 0.22 0.62 1.24 0.22 0.62 1.25 -0.06 0.61 0.94 Alt. cert. X Exp.=5 -0.01 0.92 0.99 0.09 0.95 1.10 0.48 0.76 1.61 Exp.=1 year -0.45 0.36 0.64 -0.54 0.37 0.58 -0.57 0.36 0.57 Exp.=2 years -0.72** 0.34 0.49** -0.81**0.35 0.45** -0.66** 0.32 0.52** Exp.=3 years -0.09 0.31 0.91 -0.16 0.32 0.86 -0.46 0.31 0.63 Exp.=4 years 0.08 0.37 1.09 0.04 0.37 1.04 0.26 0.34 1.30 Exp.=5 years 0.44 0.36 1.55 0.39 0.36 1.47 -0.06 0.37 0.94 College -0.00 0.07 0.99 0.12 0.13 1.12 0.10 0.13 1.10 Flag college -0.66 0.55 0.51 0.56* 0.27 1.75* 0.46* 0.24 1.58* Income35 -0.79* 0.44 0.46* 0.28* 0.21 1.33* 0.35* 0.21 1.42* Income50 0.07 0.53 1.07 0.08 0.16 1.08 0.11 0.16 1.12 Income100 -0.87 0.60 0.42 -0.23 0.16 0.79 -0.16 0.16 0.85 Income100plus -1.94*** 0.71 0.14*** 0.14 0.21 1.15 0.16 0.20 1.17 Reg. cert. in other field 0.02 0.14 1.03 0.03 0.14 1.03 0.13 0.14 1.14 Reg. cert. in other field X MA field is sp. ed. 0.44 0.36 1.55 0.42 0.36 1.53 0.26 0.35 1.30 Special Education -0.61*** 0.26 0.54*** -0.58**0.26 0.56** -0.49** 0.24 0.61** English 0.89*** 0.19 2.44*** 0.90***0.19 2.45*** 0.91*** 0.19 2.49*** Math 0.43* 0.24 1.53* 0.43* 0.24 1.54* 0.41* 0.23 1.51* Sciences -0.02 0.28 0.98 -0.04 0.28 0.96 -0.09 0.27 0.91 SocSc 0.60*** 0.25 1.82*** 0.58** 0.25 1.78** 0.52** 0.25 1.68** Other 0.80*** 0.16 2.22*** 0.78***0.16 2.18*** 0.73*** 0.16 2.07*** Year dummy -0.16 0.11 0.85 -0.18 0.13 0.83 -0.17 0.12 0.85 Married 0.20 0.25 1.22 0.26 0.26 1.29 0.06 0.26 1.06 Married female -0.38 0.28 0.68 -0.38 0.28 0.68 -0.09 0.28 0.91 Master 0.15 0.13 1.16 0.17** 0.13 1.19** 0.25** 0.13 1.29** Master flag 0.78 0.60 2.19 0.32 0.61 1.37 0.37 0.60 1.44 Female 0.62*** 0.26 1.86*** 0.59 0.26 1.81 0.31 0.25 1.36 Non-white 0.00 0.19 1.00 -0.00 0.19 0.99 -0.06 0.19 0.94 Experience -0.07* 0.04 0.94* -0.07**0.04 0.94** -0.07** 0.03 0.93** Experience sq. 0.01*** 0.00 1.01*** 0.01*** 0.00 1.01*** 0.01*** 0.00 1.01*** Age first time teaching 0.00 0. 01 1.00 0.00 0.01 1. 00 0.00 0.01 1.00 Union member -0.19 0.15 0.83 -0.18 0.15 0.84 -0.16 0.14 0.85 Real salary -0.99 0.67 0.37 0.00 0.07 1.00 -0.01 0.06 0.99 Minority students 0.00 0.28 1.00 -0.02 0.29 0.98 0.02 0.28 1.02 Minority teachers 0. 24 0.39 1.27 0. 25 0.40 1.29 0. 26 0.37 1.30 Free lunch eligible 0.66** 0.29 1.94** 0.69***0.29 2*** 0.66*** 0.28 1.94***
160 Table A-12. Continued Diagonal Definition Definition 1 Definition 2 Variable Std. error Hazard ratio Std. error Hazard ratio Std. error Hazard ratio Flag for free lunch 0.26 0.45 1.30 0.27 0.44 1.31 0.28 0.43 1.33 Av. school size -0.00 0.00 0.99 -0.00 0.00 0.99 -0.00 0.00 0.99 Urban 0.06 0.15 1.06 0.04 0.15 1.05 0.03 0.15 1.03 Constant -1.74 0.66 0.18 -1.85 0.68 0.16 -1.74 0.66 0.18 State fixed effects yes yes yes # of observations 5224 5712 6020 *, **, *** statistically significant at 10, 5, and 1% level respectively Table A-13. Wald test for joint significance F-statistic Hypothesis Diagonal Def. 1 Def.2 Alt Exp. 2=0 Alt Exp. 3=0 Alt Exp. 4=0 Alt Exp. 5=0 P>F=0.40 P>F=0.85 P>F=0.85 Alt Exp. 3=0 Alt Exp. 4=0 Alt Exp. 5=0 P>F=0.25 P>F=0.78 P>F=0.74 Alt Exp. 4=0 Alt Exp. 5=0 P>F=0.26 P>F=0.93 P>F=0.81
161 Table A-14. Results of logit regression estimation Diagonal Definition Definition 1 Definition 2 Variable Std. error Hazard ratio Std. error Hazard ratio Std. error Hazard ratio Alt. certificate in MA -0.43 0.32 0.65 -0.45 0.31 0.64 -0.33 0.30 0.72 Alt. cert. X Exp.=1 1.66* 0.92 5.25* 1.55* 0.87 4.69* 1.36 0.83 3.89 Alt. cert. X Exp.=2 0.29 1.06 1.34 0.59 0.94 1.81 -0.06 0.86 0.94 Alt. cert. X Exp.=3 -1.08 0.96 0.34 -1.11 0.91 0.33 -1.11 0.80 0.33 Alt. cert. X Exp.=4 1.51 1.02 4.53 0.59 0.91 1.81 0.63 0.96 1.88 Alt. cert. X Exp.=5 0.43 1.10 1.54 -0.15 1.06 0.86 0.47 0.96 1.61 Exp.=1 year -0.42 0.58 0.66 -0.41 0.54 0.66 -0.43 0.52 0.65 Exp.=2 years -1.48*** 0.48 0.23*** -0.95** 0.48 0.39** -0.58 0.42 0.56 Exp.=3 years 0.09 0.49 1.10 0.07 0.43 1.07 -0.30 0.42 0.74 Exp.=4 years -0.25 0.60 0.78 0.03 0.52 1.04 0.38 0.49 1.46 Exp.=5 years 0.08 0.58 1.08 0.48 0.49 1.62 -0.51 0.50 0.6 College 0.04 0.16 1.04 0.00 0.16 1.00 -0.01 0.16 0.99 Flag college 0.68*** 0.28 1.98*** 0.64*** 0.27 1.9*** 0.49** 0.24 1.63** College Exp1 -0.12 0.55 0.89 -0.15 0.53 0.86 -0.15 0.51 0.86 College Exp2 0.93* 0.50 2.53* 0.22 0.51 1.25 -0.09 0.45 0.91 College Exp3 -0.55 0.53 0.58 -0.33 0.48 0.72 -0.24 0.47 0.79 College Exp4 0.18 0.69 1.20 0.00 0.65 1.00 -0.18 0.59 0.83 College Exp5 0.10 0.73 1.10 -0.12 0.60 0.88 0.64 0.62 1.89 Alt College 0.63* 0.33 1.88* 0.68** 0.32 1.97** 0.52 0.32 1.68 Alt College Exp1 -0.64 1.11 0.53 -0.63 1.08 0.53 -0.42 1.04 0.66 Alt College Exp2 -0.25 1.24 0.78 -0.43 1.13 0.65 0.28 1.04 1.33 Alt College Exp3 2.28** 1.11 9.76** 1.88* 1.07 6.55* 2.04** 0.97 7.69** Alt College Exp4 -0.65 1.33 0.52 -0.62 1.24 0.54 -1.21 1.25 0.30 Alt College Exp5 0.31 1.57 1.37 0.32 1.49 1.38 -0.13 1.29 0.88 Income35 0.22 0.25 1.25 0.3 0.23 1.36 0.39* 0.23 1.47* Income50 0.06 0.19 1.07 0.13 0.18 1.13 0.12 0.18 1.13 Income100 -0.18 0.18 0.84 -0.22 0.18 0.80 -0.14 0.18 0.87 Income100plus 0.41* 0.24 1.51* 0.33 0.23 1.39 0.33 0.22 1.39 Alt Income35 -0.25 0.50 0.78 -0.07 0.44 0.93 -0.14 0.45 0.87 Alt Income50 -0.37 0.42 0.69 -0.27 0.39 0.77 -0.08 0.38 0.92 Alt Income100 -0.06 0.37 0.94 -0.08 0.36 0.92 -0.09 0.36 0.91 Alt Income100plus -1.01** 0.45 0.36** -1.01***0.43 0.36*** -0.97** 0.42 0.38** Reg. cert. in other field -0.02 0.15 0.98 0.04 0.14 1.04 0.13 0.14 1.14 Reg. cert. in other field X MA field is sp. ed. 0.35 0.38 1.42 0.42 0.37 1.52 0.24 0.35 1.28 Special Education -0.47* 0.27 0.63* -0.59** 0.26 0.55** -0.48* 0.25 0.62* English 0.86*** 0.20 2.37*** 0.90*** 0.19 2.45*** 0.91*** 0.19 2.47*** Math 0.43* 0.24 1.54* 0.47** 0.24 1.6** 0.44* 0.23 1.55* Sciences -0.12 0.30 0.89 -0.04 0.28 0.96 -0.10 0.27 0.90 SocSc 0.58** 0.27 1.79** 0.61*** 0.25 1.84*** 0.55** 0.25 1.73** Other 0.82*** 0.17 2.27*** 0.79*** 0.16 2.20*** 0.72*** 0.16 2.06***
162 Table A-15. Wald test for joint significance F-statistic Hypothesis Diagonal Def. 1 Def.2 Alt Exp2=0 Alt Exp3=0 Alt Exp4=0 Alt Exp5=0 College Alt Exp2=0 College Alt Exp3=0 College Alt Exp4=0 College Alt Exp5=0 P>F=0.07 P>F=0.17 P>F=0.03 Alt Exp3=0 Alt Exp4=0 Alt Exp5=0 College Alt Exp3=0 College Alt Exp4=0 College Alt Exp5=0 P>F=0.03 P>F=0.11 P>F=0.01 Alt Exp4=0 Alt Exp5=0 College Alt Exp4=0 College Alt Exp5=0 P>F=0.45 P>F=0.61 P>F=0.17 Table A-14. Continued Diagonal Definition Definition 1 Definition 2 Variable Std. error Hazard ratio Std. error Hazard ratio Std. error Hazard ratio Year -0.25* 0.13 0.78* -0.19 0.13 0.83 -0.18 0.12 0.84 Married 0.33 0.27 1.39 0.25 0.27 1.28 0.08 0.26 1.08 Married female -0.47 0.30 0.62 -0.38 0.29 0.69 -0.11 0.28 0.90 Master 0.16 0.14 1.18 0.18 0.13 1.19 0.26** 0.13 1.30** Master flag 0.49 0.63 1.63 0.48 0.61 1.61 0.53 0.60 1.70 Female 0.68*** 0.27 1.98*** 0.60** 0.26 1.82** 0.33 0.26 1.39 Non-white 0.11 0.20 1.12 0.00 0.19 1.00 -0.03 0.19 0.97 Experience -0.07* 0.04 0.94* -0.07* 0.04 0.94* -0.07** 0.03 0.93** Experience sq. 0.01*** 0.00 1.01*** 0.01*** 0.00 1.01*** 0.01*** 0.00 1.01*** Age first time teaching -0.00 0. 01 0.99 0.00 0.01 1. 00 0.00 0.01 1.00 Union member -0.16 0.16 0.85 -0.17 0.15 0.84 -0.16 0.14 0.85 Real salary 0.00 0.07 1.00 -0.00 0.07 0.99 -0.01 0.07 0.99 Minority students 0.04 0.30 1.04 -0.03 0.29 0.97 0.02 0.28 1.02 Minority teachers 0. 21 0.40 1.23 0. 24 0.40 1.28 0. 25 0.37 1.29 Free lunch eligible 0.77*** 0.31 2.16*** 0.71*** 0.29 2.03*** 0.66*** 0.28 1.93*** Flag for free lunch 0.30 0.45 1.36 0.29 0.44 1.34 0.30 0.44 1.34 Av. school size -0.00 0.00 0.99 -0.00 0.00 0.99 -0.00 0.00 0.99 Urban 0.06 0.16 1.06 0.05 0.15 1.05 0.02 0.15 1.02 Constant -1.85 0.73 0.16 -1.86 0.68 0.16 -1.73 0.66 0.18 State fixed effects yes yes yes # of observations 5224 5712 6020 *, **, *** statistically significant at 10, 5, and 1% level respectively.
163 APPENDIX B DETAILED RESULTS FOR ESTIMATION OF THE EFFECT OF SCHOOL ACCOUNTABILITY Table B-1. Accountability policy 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 State Test R,W,M Test other Student account. Exit test School evaluation Subgroup LPS Sanctions Sch. adm. sanctions HPS Award LPS assistance LPS fin. assistance District evaluation District sanctions District assistance Student School District CarnoyLoeb AL 1 1 0 1 1 0 0 0 0 1 1 0 0 0 1 1 0 4 AK 1 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 1 AZ 1 0 1 0 0 0 0 0 0 1 0 0 0 0 1 0 0 2 AR 1 0 0 0 0 0 0 0 0 1 1 0 1 0 0 0 0 1 CA 1 1 0 0 1 1 1 1 1 0 1 0 0 0 0 1 0 4 CO 1 1 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 CT 1 1 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 DE 1 1 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 1 FL 1 0 1 1 1 1 1 1 1 1 1 0 0 0 1 1 0 5 GA 1 1 0 1 0 0 0 0 1 1 1 0 0 0 1 0 0 2 HI 1 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 1 ID 1 1 0 0 1 0 1 0 0 1 1 0 0 0 0 1 0 1 IL 1 1 1 0 1 0 0 0 0 1 1 0 0 0 1 0 0 2.5 IN 1 0 0 1 1 0 0 0 1 1 1 1 0 0 1 1 0 3 IA 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 KS 1 1 0 0 1 0 0 0 0 1 1 0 0 0 0 1 0 1 KY 1 1 0 0 0 0 0 0 1 1 1 0 0 0 0 1 0 4 LA 1 1 1 1 1 0 1 1 1 1 1 0 0 0 1 1 0 3 ME 1 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 MD 1 1 0 1 1 0 0 1 1 1 1 0 0 0 1 1 0 4 MA 1 1 0 0 0 0 0 1 0 1 1 0 0 0 0 0 0 2 MI 1 1 1 0 1 0 0 0 0 1 0 0 0 0 0 0 0 1 MN 1 0 0 1 0 0 0 0 0 1 0 0 0 0 1 0 0 2 MS 1 0 0 1 0 0 0 0 0 0 0 1 1 1 1 0 1 3 MO 1 1 0 0 0 0 0 1 1 1 0 1 1 0 0 0 0 1.5 MT 1 1 0 0 1 0 0 0 0 1 1 0 0 0 0 0 0 1
164 Table B-1. Continued. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 State Test R,W,M Test other Student account. Exit test School evaluation Subgroup LPS Sanctions Sch. adm. sanctions HPS Award LPS assistance LPS fin. assistance District evaluation District sanctions District assistance Student School District CarnoyLoeb NE 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 NC 1 1 1 0 1 0 0 1 1 1 0 1 1 0 0 1 0 5 ND 1 1 0 0 1 0 1 0 0 1 1 0 0 0 0 1 0 1 OH 1 1 1 1 0 0 0 1 0 0 1 1 1 1 1 0 1 3 OK 1 1 1 0 1 0 0 0 0 1 1 0 0 0 0 1 0 1 OR 1 1 1 0 0 0 0 0 0 1 1 0 0 0 0 0 0 2.5 PA 1 0 1 0 0 0 0 0 1 0 0 0 0 0 0 1 0 1 RI 1 0 0 0 0 1 0 0 0 1 0 0 0 0 0 1 0 1 SC 1 1 1 1 0 1 0 0 0 0 0 1 1 1 1 0 1 3 SD 1 1 0 0 1 0 0 0 0 1 1 0 0 0 0 0 0 1 TN 1 1 0 1 0 0 0 0 0 1 1 0 0 0 1 0 0 1.5 TX 1 1 0 1 1 1 1 1 1 1 0 1 0 0 1 1 0 5 UT 1 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 1 VT 1 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 VA 1 1 1 1 1 0 0 0 0 0 0 0 0 0 1 0 0 2 WA 1 1 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 1 WV 1 1 1 0 1 0 1 0 0 1 1 1 1 1 0 1 1 3.5 WI 1 1 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 2 WY 1 0 0 0 0 0 0 0 0 1 1 1 1 0 0 0 0 1 Total 48 34 18 16 20 5 8 12 14 46 30 11 9 6 19 18 5
165 Table B-2. Summary statistics by fields and year English language arts Mathematics Social Sciences Sciences 1993-94 1999-00 1993-94 1999-00 1993-94 1999-00 1993-94 1999-00 Variable Mean Std Mean Std Mean St d Mean Std Mean Std Mean Std Mean Std Mean Std Certification 0.95 0.0058 0.91 0.011 0.94 0.01 0.88 0.013 0.91 0.010 0.87 0.014 0.94 0.01 0.87 0.016 LPS_field 0.32 0.0181 0.37 0.016 0.289 0.017 0.32 0.016 0.285 0.016 0.32 0.02 0.28 0.01 0.33 0.014 white teacher 0.92 0.0134 0.87 0.015 0.91 0.01 0.89 0.014 0.92 0.014 0.90 0.01 0.93 0.01 0.89 0.016 black teacher 0.07 0.013 0.10 0.016 0. 07 0.01 0.08 0.01 5 0.07 0.013 0. 08 0.01 0.05 0. 01 0.07 0.014 Sex 0.19 0.0142 0.17 0.018 0.44 0.02 0.39 0.024 0.63 0.022 0.61 0.02 0.54 0.03 0.47 0.020 Entry age 25.6 0.2 25.4 0.3 25.2 0.3 25.0 0.3 25.7 0.3 25.6 0.2 25.8 0.3 26.4 0.3 Experience 15.8 0.5 15.3 0.4 16.0 0.5 14.6 0.4 16.5 0.6 15.2 0.4 15.3 0.5 13.5 0.4 Union agreement 0.69 0.089 0.68 0.097 0.70 0.09 0.70 0.086 0.70 0.088 0.71 0.08 0.71 0.08 0.72 0.081 Urban 0.55 0.024 0.75 0.028 0.58 0.03 0.76 0.032 0.54 0.031 0.74 0.03 0.55 0.03 0.74 0.032 Free lunch 0.32 0.018 0.37 0.016 0.29 0.02 0.32 0.016 0.29 0.016 0.32 0.02 0.28 0.01 0.33 0.014 Minority students 0.29 0.035 0.36 0.037 0.28 0.03 0.32 0.030 0.28 0.037 0.32 0.03 0.27 0.03 0.32 0.033 Minority teachers 0.12 0.016 0.17 0.020 0. 12 0.02 0.16 0.01 4 0.12 0.016 0. 14 0.02 0.11 0. 01 0.15 0.016 District enrollment 0.256 0.543 0.569 2.091 0.269 0.592 0.597 2.384 0.264 0.623 0.599 2.351 0.264 0.661 0.451 11900 bigD LPS 0.115 0.312 0.311 1.319 0.114 0.347 0.313 1.436 0.104 0.331 0.281 1.132 0.101 0.297 0.237 7518 Student year 0.000 0.000 0.510 0.106 0.000 0.000 0.500 0.101 0.000 0.000 0.500 0.100 0.000 0.000 0.480 0.100 Student LPS 0.000 0.000 0.200 0.045 0.000 0.000 0.160 0.037 0.000 0.000 0.170 0.040 0.000 0.000 0.150 0.036 Student bigD 0.000 0.000 0.199 0.617 0.000 0.000 0.199 0.679 0.000 0.000 0.195 0.628 0.000 0.000 0.156 5030 Student bigD LPS 0.000 0.000 0.102 0.337 0.000 0.000 0.092 0.375 0.000 0.000 0.092 0.364 0.000 0.000 0.064 2157 School year 0.000 0.000 0.480 0.108 0.000 0.000 0.470 0.102 0.000 0.000 0.450 0.100 0.000 0.000 0.440 0.102 School LPS 0.000 0.000 0.190 0.046 0.000 0.000 0.170 0.038 0.000 0.000 0.160 0.040 0.000 0.000 0.160 0.038
166 Table B-2. Continued English language arts Mathematics Social Sciences Sciences 1993-94 1999-00 1993-94 1999-00 1993-94 1999-00 1993-94 1999-00 Variable Mean Std Mean Std Mean St d Mean Std Mean Std Mean Std Mean Std Mean Std School bigD 0.000 0.000 0.212 0.758 0.000 0.000 0.208 0.849 0.000 0.000 0.221 0.961 0.000 0.000 0.223 0.959 School bigD LPS 0.000 0.000 0.101 0.376 0.000 0.000 0.096 0.405 0.000 0.000 0.101 0.490 0.000 0.000 0.105 0.485 District year 0.000 0.000 0.110 0.055 0.000 0.000 0.110 0.054 0.000 0.000 0.100 0.050 0.000 0.000 0.100 0.051 District LPS 0.000 0.000 0.040 0.019 0.000 0.000 0.030 0.013 0.000 0.000 0.030 0.020 0.000 0.000 0.030 0.014 District bigD 0.000 0.000 0.014 0.080 0.000 0.000 0.010 0.052 0.000 0.000 0.009 0.040 0.000 0.000 0.009 0.047 District bigD LPS 0.000 0.000 0.008 0.046 0.000 0.000 0.004 0.020 0.000 0.000 0.003 0.016 0.000 0.000 0.003 0.015 Index year 0.000 0.000 3.010 0.356 0.000 0.000 2.970 0.326 0.000 0.000 2.900 0.330 0.000 0.000 2.800 0.331 Index LPS 0.000 0.000 1.170 0.170 0.000 0.000 1.005 0.142 0.000 0.000 0.960 0.140 0.000 0.000 0.940 0.135 Index bigD 0.000 0.000 2.309 10.726 0.000 0.000 2.438 12.187 0.000 0.000 2.484 11.946 0.000 0.000 1.675 5.984 Index bigD LPS 0 0 1.287 6.695 0.000 0.000 1.299 7.266 0.000 0.000 1.161 5.620 0.000 0.000 0.886 3.769 Number of observations 8747 6547 5593 5794 Population size 471404.76 313613.9 252977.01 272132.38
167 Table B-3. Logit results for stude nt-targeted accountability policy English Mathematics Social sciences Sciences Variable odds ratio t-st. odds ratio t-st. odds ratio t-st. odds ratio t-st. student year 0.86 -0.31 0.64 -0.81 2.25* 1.87 1.48 1.22 student LPS 1.86 0.81 2.75 1.31 0.63 -0.59 0.91 -0.19 year 0.81 -0.58 1.00 0.01 0.62** -2.07 0.39*** -3.50 LPS year 0.39* -1.72 0.24** -1.96 0.60 -0.70 1.55 0.96 white teacher 0.97 -0.05 1.11 0.52 0.49** -2.30 0.68** -2.25 black teacher 0.48 -1.22 0.83 -0.57 0.22*** -3.58 0.42** -2.20 Sex 1.36** 2.16 1.68* 2.91 1.88*** 3.19 1.35 1.15 Entry age 1.00 -0.27 0.99 -1.33 1.00 0.30 1.00 -0.02 Experience 1.06*** 4.65 1.04*** 3.28 1.13*** 6.94 1.06*** 2.96 Union agreement 0.85 -0.85 1.48 1.24 1.55* 1.90 1.55* 1.95 Urban 1.16 0.71 1.09 0.47 1.21 1.00 0.91 -0.36 Free lunch 0.48** -2.27 0.94 -0.09 1.15 0.28 0.37*** -3.14 Minority students 0.58 -1.27 0.80 -0.45 1.48 0.69 1.52 0.97 Minority teachers 1.90 0.98 0.59 -1.1 2 0.42* -1.8 4 0.79 -0.59 District enrollment 0.91 -0.74 0.96 -0.57 1.04 0.91 0.83*** -3.18 Constant 15.80 4.08 11.94 5.11 2.27 1.53 11.13 3.29 Number of obs. 8747 6547 5593 5794 State fixes effects yes yes yes yes Weights yes yes yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively. Table B-4. Logit results for sc hool-targeted accountability policy English Mathematics Social sciences Sciences Variable odds ratio t-st. odds ratio t-st. odds ratio t-st. odds ratio t-st. school year 0.76 -0.57 0.53 -1.04 0.88 -0.30 0.59 -1.53 school LPS 1.68 0.68 0.80 -0.25 0.67 -0.54 1.73 0.99 year 0.85 -0.48 1.04 0.09 0.93 -0.26 0.60** -2.16 LPS year 0.44 -1.57 0.50 -0.78 0.74 -0.44 1.21 0.41 white teacher 0.97 -0.06 1.17 0.87 0.49*** -2.33 0.68** -2.14 black teacher 0.48 -1.26 0.90 -0.32 0.22*** -3.46 0.43** -2.19 Sex 1.35** 2.10 1.63*** 2.72 1.88*** 3.15 1.35 1.14 Entry age 1.00 -0.38 0.99 -1.27 1.00 0.36 1.00 0.00 Experience 1.06*** 4.68 1.04*** 3.39 1.13*** 7.01 1.06*** 2.95 Union agreement 0.85 -0.83 1.48 1.27 1.58* 1.95 1.54** 1.98 Urban 1.16 0.73 1.08 0.42 1.20 0.91 0.92 -0.33 Free lunch 0.47*** -2.44 0.85 -0.24 1.02 0.05 0.36*** -3.26 Minority students 0.57 -1.31 0.80 -0.44 1.48 0.68 1.49 0.88 Minority teachers 1.99 1.11 0.66 -0.90 0.40** -2.2 6 0.79 -0.61 District enrollment 0.91 -0.76 0.94 -0.95 1.03 0.59 0.83*** -3.07 Constant 17.29 3.78 14.44 4.88 2.89 2.00 13.60 3.42 Number of obs. 8747 6547 5593 5794 State fixes effects yes yes yes yes Weights yes yes yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively.
168 Table B-5. Logit results for dist rict-targeted accountability policy English Mathematics Social sciences Sciences Variable odds ratio t-st. odds ratio t-st. odds ratio t-st. odds ratio t-st. District*year 0.25*** -3.62 3.00 1.46 2.18*** 3.10 0.53* -1.82 District* LPS 5.10*** 2.72 0.64 -0.52 3.00 0.80 1.27 0.36 year 0.89 -0.40 0.70 -0.84 0.81 -0.81 0.49*** -3.01 LPS year 0.45 -1.57 0.40 -1.11 0.55 -1.04 1.48 0.95 white teacher 1.01 0.02 1.13 0.64 0.47*** -2.35 0.68** -2.17 black teacher 0.50 -1.14 0.85 -0.45 0.21*** -3.63 0.43** -2.15 Sex 1.36** 2.12 1.67*** 2.77 1.90*** 3.13 1.36 1.18 Entry age 1.00 -0.40 0.99 -1.26 1.00 0.27 1.00 -0.01 Experience 1.06*** 4.75 1.04*** 3.35 1.13*** 7.04 1.06*** 2.99 Union agreement 0.85 -0.84 1.51 1.28 1.58** 1.96 1.55** 1.97 Urban 1.17 0.84 1.08 0.42 1.20 0.89 0.91 -0.36 Free lunch 0.48 -2.31 0.94 -0.09 1.04 0.08 0.37 -3.23 Minority students 0.56 -1.38 0.81 -0.42 1.49 0.72 1.54 0.97 Minority teachers 1.99 1.07 0.64 -0.95 0.40** -2.2 4 0.77 -0.63 District enrollment 0.91 -0.74 0.95 -0.74 1.04 0.95 0.83*** -2.94 Constant 15.64 3.72 11.94 4.72 2.97 2.01 11.94 3.47 Number of observations 8747 6547 5593 5794 State fixes effects yes yes yes yes Weights yes yes yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively. Table B-6. Logit results for aggr egated accountability policy index English Mathematics Social sciences Sciences Variable odds ratio t-st. odds ratio t-st. odds ratio t-st. odds ratio t-st. Index year 1.12 0.69 0.95 -0.22 1.03 0.28 0.87 -1.15 Index LPS 0.75 -0.98 0.71 -1.03 0.94 -0.27 1.14 0.80 year 0.54 -1.23 0.91 -0.16 0.80 -0.61 0.70 -1.05 LPS year 1.42 0.41 1.17 0.13 0.68 -0.41 1.06 0.10 white teacher 1.01 0.02 1.15 0.78 0.48** -2.31 0.69** -2.14 black teacher 0.49 -1.21 0.82 -0.60 0.22*** -3.44 0.43** -2.21 Sex 1.38** 2.18 1.67*** 2.82 1.88*** 3.18 1.36 1.17 Entry age 1.00 -0.41 0.99 -1.28 1.00 0.32 1.00 0.00 Experience 1.06*** 4.69 1.04*** 3.30 1.13*** 6.95 1.06*** 3.01 Union agreement 0.87 -0.72 1.46 1.23 1.58* 1.93 1.54* 1.95 Urban 1.14 0.66 1.09 0.45 1.20 0.91 0.92 -0.33 Free lunch 0.47*** -2.50 0.91 -0.15 1.06 0.12 0.36*** -3.27 Minority students 0.58 -1.32 0.82 -0.43 1.49 0.71 1.54 0.94 Minority teachers 1.99 1.06 0.64 -0.96 0.40** -2.1 5 0.78 -0.63 District enrollment 0.91 -0.73 0.97 -0.52 1.04 0.77 0.83*** -2.89 Constant 17.29 3.74 14.30 5.06 2.69 1.92 13.74 3.22 Number of obs. 8747 6547 5593 5794 State fixes effects yes yes yes yes Weights yes yes yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively.
169 Table B-7. Logit results for sample consisting of English language arts (base field) and mathematics teachers Student School District Aggregated index Variable odds ratio t-st. odds ratio t-st. odds ratio t-st. odds ratio t-st. Policy year 0.86 -0.35 0.67 -0.92 0.43* -1.85 1.08 0.51 Policy Math 0.76 -0.66 0.91 -0.20 4.53*** 3.39 0.92 -0.68 Policy LPS 1.80 0.87 1.46 0.54 3.56*** 2.33 0.74 -1.07 Policy LPS Math 1.38 0.38 0.58 -0.58 0.30 -1.43 0.96 -0.12 Math 0.53*** -2.39 0.52*** -2.46 0.52*** -2.39 0.52*** -2.41 year 0.79 -0.68 0.85 -0.53 0.81 -0.71 0.57 -1.24 Math year 1.27 0.64 1.17 0.34 0.93 -0.17 1.40 0.62 LPS year 0.44 -1.65 0.55 -1.24 0.52 -1.34 1.62 0.59 LPS Math 2.08 1.26 2.18 1.36 2.10 1.29 2.12 1.30 LPS Math year 0.52 -0.78 0.77 -0.27 0.70 -0.44 0.68 -0.27 white teacher 1.08 0.33 1.12 0.43 1.09 0.37 1.12 0.43 black teacher 0.64 -1.38 0.66 -1.27 0.64 -1.34 0.63 -1.44 Sex 1.54*** 3.45 1.52*** 3.34 1.54*** 3.37 1.55*** 3.44 Entry age 0.99 -1.20 0.99 -1.30 0.99 -1.30 0.99 -1.26 Experience 1.05*** 4.93 1.05*** 4.99 1.05*** 5.05 1.05*** 5.00 Union agreement 1.08 0.68 1.09 0.73 1.11 0.82 1.11 0.79 Urban 1.12 0.70 1.11 0.66 1.12 0.72 1.11 0.63 Free lunch 0.48*** -2.46 0.44*** -2.82 0.47*** -2.53 0.45*** -2.70 Minority students 0.68 -1.04 0.67 -1.05 0.68 -1.06 0.68 -1.08 Minority teachers 1.09 0.19 1.19 0.38 1.17 0.34 1.17 0.34 District enrollment 0.94 -0.78 0.92 -0.91 0.93 -0.83 0.94 -0.76 Constant 17.64 7.58 20.49 7.64 17.99 7.45 20.49 7.16 Number of obs. 15379 15379 15379 15379 State fixes effects yes yes yes yes Weights yes yes yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively.
170 Table B-8. Logit results for sample consisting of English language arts (base field) and social sciences teachers Student School District Aggregated index Variable odds ratio t-st. odds ratio t-st. odds ratio t-st. odds ratio t-st. Policy year 0.91 -0.18 0.76 -0.58 0.37* -1.66 1.14 0.86 Policy Soc. Sc. 2.53 1.55 1.26 0.41 3.67* 1.93 0.90 -0.50 Policy LPS 1.80 0.78 1.73 0.73 3.94** 2.15 0.77 -0.89 Policy LPS* Soc. Sc. 0.31 -1.31 0.30 -1.51 1.40 0.31 1.15 0.41 Soc. Sc. 0.30*** -4.88 0.30*** -5.01 0.30*** -4.92 0.30*** -4.89 year 0.82 -0.54 0.88 -0.40 0.89 -0.39 0.53 -1.38 Soc. Sc. year 0.63 -1.20 0.84 -0.40 0.81 -0.72 1.30 0.44 LPS year 0.39* -1.72 0.43 -1.57 0.46 -1.52 1.22 0.23 LPS Soc. Sc. 2.94*** 2.63 2.92*** 2.70 2.92*** 2.65 2.89*** 2.63 LPS Soc. Sc. year 2.39 1.07 2.75 1.51 1.55 0.73 0.95 -0.05 white teacher 0.79 -0.53 0.79 -0.53 0.80 -0.51 0.81 -0.48 black teacher 0.38* -1.92 0.38* -1.93 0.38* -1.95 0.39* -1.89 Sex 1.55*** 3.32 1.55*** 3.28 1.57*** 3.32 1.57*** 3.42 Entry age 1.00 -0.14 1.00 -0.09 1.00 -0.19 1.00 -0.17 Experience 1.08*** 7.76 1.08*** 7.80 1.08*** 7.72 1.08*** 7.76 Union agreement 1.06 0.39 1.08 0.47 1.08 0.50 1.08 0.50 Urban 1.15 0.92 1.15 0.94 1.15 0.97 1.14 0.88 Free lunch 0.41*** -2.93 0.39** -3.23 0.41*** -2.98 0.40*** -3.06 Minority students 0.81 -0.89 0.79 -0.95 0.80 -0.95 0.82 -0.83 Minority teachers 1.11 0.25 1.12 0.27 1.09 0. 23 1.09 0.23 District enrollment 0.99 -0.25 0.98 -0.29 0.99 -0.24 0.99 -0.19 Constant 12.06 4.76 13.87 4.80 13.33 4.81 13.20 4.52 Number of obs. 14340 14340 14340 14340 State fixes effects yes yes yes yes Weights yes yes yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively.
171 Table B-9. Logit results for sample consisting of English language arts (base field) and sciences teachers Student School District Aggregated index Variable odds ratio t-st. odds ratio t-st. odds ratio t-st. odds ratio t-st. Policy year 0.95 -0.11 0.82 -0.44 0.24*** -3.77 1.12 0.76 Policy Sciences 1.28 0.72 0.64 -1.30 2.29** 2.02 0.78*** -2.36 Policy LPS 1.77 0.80 1.52 0.59 3.94*** 2.37 0.76 -1.01 Policy LPS* Sciences 0.60 -0.74 1.12 0.15 0.61 -0.76 1.48 1.63 Sciences 0.75 -1.36 0.74 -1.43 0.76 -1.27 0.75 -1.39 year 0.80 -0.65 0.84 -0.55 0.92 -0.29 0.57 -1.32 Sciences year 0.51** -1.99 0.71 -0.92 0.51** -2.24 1.19 0.37 LPS year 0.39* -1.82 0.45 -1.60 0.44* -1.66 1.31 0.34 LPS Sciences 0.99 -0.03 1.00 0.00 0.96 -0.09 0.99 -0.03 LPS Sciences year 3.90* 1.85 2.89 1.49 3.35** 1.99 0.90 -0.10 white teacher 0.84 -0.49 0.84 -0.51 0.86 -0.45 0.86 -0.45 black teacher 0.47** -2.05 0.46** -2.15 0.47** -2.04 0.47** -2.09 Sex 1.36* 1.89 1.36* 1.86 1.36* 1.87 1.36* 1.87 Entry age 1.00 -0.33 1.00 -0.42 1.00 -0.45 1.00 -0.45 Experience 1.06*** 4.44 1.06*** 4.39 1.06*** 4.50 1.06*** 4.46 Union agreement 1.09 0.57 1.08 0.56 1.09 0.61 1.09 0.62 Urban 1.04 0.25 1.06 0.31 1.05 0.32 1.04 0.25 Free lunch 0.45*** -2.56 0.43*** -2.88 0.46*** -2.60 0.44*** -2.81 Minority students 0.84 -0.60 0.83 -0.66 0.84 -0.65 0.85 -0.59 Minority teachers 1.34 0.58 1.39 0.70 1.38 0. 64 1.36 0.63 District enrollment 0.88 -1.49 0.88 -1.50 0.88 -1.44 0.88 -1.47 Constant 15.64 5.36 18.36 5.01 16.12 5.12 18.54 4.81 Number of obs. 14571 14571 14571 14571 State fixes effects yes yes yes yes Weights yes yes yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively.
172 Table B-10. Logit results for sample consisting of mathematics (base field) and sciences teachers Student School District Aggregated index Variable odds ratio t-st. odds ratio t-st. odds ratio t-st. odds ratio t-st. Policy year 0.72 -0.74 0.66 -0.88 2.14 1.01 0.97 -0.15 Policy Sciences 1.73 1.58 0.72 -0.83 0.51 -0.98 0.87 -1.03 Policy LPS 2.39 1.17 0.79 -0.26 0.61 -0.60 0.75 -0.91 Policy LPS* Sciences 0.44 -1.02 1.95 0.77 2.14 0.75 1.49 1.57 Sciences 1.32 0.96 1.31 0.93 1.32 0.94 1.34 0.97 year 0.98 -0.04 0.99 -0.01 0.77 -0.67 0.89 -0.23 Sciences year 0.41*** -2.54 0.64 -0.99 0.58 -1.22 0.82 -0.38 LPS year 0.23* -1.94 0.45 -0.87 0.39 -1.17 0.96 -0.03 LPS Sciences 0.51 -1.00 0.51 -1.01 0.51 -1.01 0.50 -1.00 LPS Sciences year 6.96*** 2.56 3.29 1.13 4.22 1.60 1.31 0.23 white teacher 0.94 -0.40 0.97 -0.23 0.95 -0.35 0.96 -0.30 black teacher 0.66 -1.53 0.67 -1.42 0.66 -1.49 0.64 -1.65 Sex 1.52*** 2.40 1.52** 2.32 1.52*** 2.36 1.54*** 2.38 Entry age 0.99 -0.86 0.99 -0.84 0.99 -0.82 0.99 -0.83 Experience 1.05*** 4.06 1.05*** 4.09 1.05*** 4.13 1.05*** 4.11 Union agreement 1.52*** 2.51 1.54*** 2.58 1.55*** 2.59 1.52*** 2.56 Urban 1.01 0.05 1.01 0.05 1.00 0.03 1.01 0.04 Free lunch 0.84 -0.31 0.78 -0.43 0.81 -0.35 0.80 -0.38 Minority students 1.09 0.27 1.08 0.24 1.12 0.32 1.12 0.32 Minority teachers 0.68 -1.25 0.73 -1 .09 0.71 -1.1 2 0.70 -1.14 District enrollment 0.90* -1.92 0.89** -2.05 0.89* -1.92 0.90* -1.89 Constant 9.30 5.15 11.59 5.12 9.78 4.99 11.47 4.89 Number of obs. 12341 12341 12341 12341 State fixes effects yes yes yes yes Weights yes yes yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively.
173 Table B-11. Logit results for sample consisting of mathematics (base field) and social sciences teachers Student School District Aggregated index Variable odds ratio t-st. odds ratio t-st. odds ratio t-st. odds ratio t-st. Policy year 0.64 -0.83 0.64 -0.80 2.72 1.14 1.03 0.14 Policy Soc. Sc. 3.46*** 2.25 1.34 0.49 0.76 -0.30 0.97 -0.12 Policy LPS 2.64 1.10 0.91 -0.09 0.60 -0.55 0.73 -0.99 Policy *LPS*S.Sc. 0.23 -1.43 0.55 -0.54 5.31 1.16 1.23 0.56 Soc. Sc. 0.54** -2.17 0.54** -2.19 0.54** -2.16 0.54** -2.17 year 1.04 0.08 1.01 0.02 0.75 -0.68 0.76 -0.48 Soc. Sc. year 0.54 -1.23 0.83 -0.34 0.97 -0.06 1.02 0.03 LPS year 0.23* -1.90 0.43 -0.91 0.39 -1.16 1.08 0.07 LPS Soc. Sc. 1.57 0.94 1.51 0.91 1.49 0.86 1.51 0.88 LPS Soc.Sc.* year 3.63 1.38 2.66 0.91 1.82 0.69 1.01 0.01 white teacher 0.92 -0.41 0.93 -0.35 0.91 -0.46 0.92 -0.42 black teacher 0.54* -1.87 0.55* -1.80 0.54* -1.90 0.52** -2.05 Sex 1.68*** 3.32 1.67*** 3.20 1.68*** 3.22 1.68*** 3.34 Entry age 0.99 -0.66 0.99 -0.54 0.99 -0.61 0.99 -0.57 Experience 1.07*** 6.41 1.07*** 6.53 1.07*** 6.47 1.07*** 6.36 Union agreement 1.46* 1.85 1.51** 2.00 1.52** 2.00 1.51* 1.97 Urban 1.11 0.85 1.11 0.76 1.09 0.76 1.09 0.71 Free lunch 0.74 -0.52 0.69 -0.62 0.73 -0.53 0.70 -0.60 Minority students 1.15 0.39 1.14 0.35 1.16 0.42 1.16 0.42 Minority teachers 0.47** -2.24 0.49** -2 .29 0.48** -2.3 1 0.48** -2.23 District enrollment 0.99 -0.11 0.98 -0.52 0.99 -0.22 1.01 0.05 Constant 6.62 3.79 7.92 4.05 7.46 3.89 7.61 4.18 Number of obs. 12140 12140 12140 12140 State fixes effects yes yes yes yes Weights yes yes yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively.
174 Table B-12. Logit results for sample consisting of social sciences teachers (base field) and sciences teachers Student School District Aggregated index Variable odds ratio t-st. odds ratio t-st. odds ratio t-st. odds ratio t-st. Policy year 2.56*** 2.45 1.08 0.20 1.65** 2.06 1.02 0.11 Policy Sciences 0.45 -1.42 0.48 -1.34 0.61 -0.86 0.88 -0.60 Policy LPS 0.47 -0.97 0.46 -1.07 2.03 0.71 0.94 -0.21 Policy LPS Sc. 2.41 1.04 3.46 1.50 0.47 -0.65 1.22 0.65 Sciences 2.44*** 4.45 2.44*** 4.46 2.46*** 4.50 2.46*** 4.48 year 0.53*** -2.71 0.79 -0.93 0.77 -1.12 0.77 -0.70 Sciences year 0.80 -0.58 0.79 -0.64 0.59* -1.77 0.81 -0.38 LPS year 0.88 -0.19 1.06 0.11 0.68 -0.77 0.83 -0.22 LPS Sciences. 0.33*** -2.45 0.34*** -2.37 0.33*** -2.39 0.33*** -2.40 LPS Sc.* year 1.90 0.92 1.51 0.63 2.69* 1.92 1.49 0.41 white teacher 0.68*** -2.58 0.68*** -2.58 0.68*** -2.54 0.68*** -2.55 black teacher 0.38*** -3.01 0.39*** -3.10 0.38*** -3.01 0.39*** -3.04 Sex 1.54*** 3.71 1.55*** 3.57 1.55*** 3.62 1.55*** 3.67 Entry age 1.00 0.07 1.00 0.17 1.00 0.10 1.00 0.12 Experience 1.08*** 6.65 1.08*** 6.59 1.08*** 6.61 1.08*** 6.62 Union agreement 1.54*** 2.64 1.57*** 2.79 1.57*** 2.81 1.55*** 2.72 Urban 1.07 0.52 1.07 0.52 1.06 0.45 1.06 0.50 Free lunch 1.01 0.04 0.93 -0.20 0.97 -0.08 0.96 -0.12 Minority students 1.58 1.23 1.57 1.16 1.60 1.24 1.60 1.21 Minority teachers 0.56** -2.21 0.54*** -2 .38 0.54*** -2.4 1 0.54*** -2.39 District enrollment 0.94** -2.13 0.93** -2.15 0.94** -1.91 0.94* -1.80 Constant 2.94 2.20 3.63 2.58 3.39 2.57 3.42 2.42 Number of obs. 11417 11417 11417 11417 State fixes effects yes yes yes yes Weights yes yes yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively.
175 Table B-13. Logit results for stude nt-targeted accountability policy wi th controls for district size English Mathematics Social Sciences Sciences Variable odds ratio t-st. odds ratio t-st. odds ratio t-st. odds ratio t-st. Student year 0.90 -0.20 0.63 -0.83 1.99 1.50 1.28 0.76 Student LPS 1.48 0.57 1.46 0.55 0.66 -0.47 0.75 -0.47 Student bigD 1.28 0.26 2.23 1.21 1.45 0.95 1.69** 2.21 Student *bigD *LPS 0.87 -0.12 0.76 -0.36 0.72 -0.54 1.03 0.07 year 0.73 -0.93 0.87 -0.40 0.61** -2.01 0.39*** -3.75 LPS year 0.50 -1.52 0.35 -1.57 0.62 -0.63 1.58 0.97 white teacher 1.00 0.00 1.05 0.28 0.48** -2.32 0.71* -1.91 black teacher 0.47 -1.22 0.72 -0.99 0.21*** -3.68 0.44** -2.11 Sex 1.40*** 2.34 1.68*** 3.08 1.90* 3.21 1.35 1.14 Entry age 1.00 -0.33 0.99 -1.35 1.00 0.36 1.00 0.02 Experience 1.06*** 4.65 1.04*** 3.32 1.13*** 6.90 1.06*** 3.01 Union agreement 0.84 -0.86 1.42 1.10 1.55* 1.92 1.55** 1.97 Urban 1.14 0.65 1.08 0.43 1.21 0.99 0.89 -0.47 Free lunch 0.63 -1.41 1.25 0.34 1.25 0.45 0.36*** -2.79 Minority students 0.54 -1.41 0.68 -0.82 1.43 0.64 1.46 0.89 Minority teachers 1.82 0.95 0.57 -1.3 1 0.42* -1.9 2 0.74 -0.74 District enrollment 1.34 1.32 1.17 1.55 1.09 0.91 0.75** -2.29 bigD LPS 0.58*** -2.62 0.72*** -4.00 0.91 -0.74 0.91 0.48 Constant 14.30 4.02 12.81 5.29 2.23 1.54 11.59 3.33 Number of observations 8747 6547 5593 5794 State fixes effects yes yes yes yes Weights yes yes yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively. Table B-14. Logit results for school-targeted account ability policy with controls for district size English Mathematics Social Sciences Sciences Variable odds ratio t-st. odds ratio t-st. odds ratio t-st. odds ratio t-st. School year 0.55 -1.61 0.52 -1.10 0.67 -1.04 0.63 -1.19 School LPS 1.49 0.73 0.44 -0.88 1.08 0.10 1.00 0.00 School bigD 3.24 1.57 1.00 0.99 1.00** 2.19 1.00 -0.56 School *bigD *LPS 0.37 -1.19 1.00 -0.59 1.00*** -2.99 1.00*** 3.35 year 0.80 -0.70 0.89 -0.33 0.93 -0.24 0.60** -2.11 LPS year 0.56 -1.26 0.87 -0.17 0.73 -0.44 1.26 0.49 white teacher 1.00 0.01 1.12 0.66 0.48*** -2.36 0.65*** -2.59 black teacher 0.48 -1.23 0.83 -0.55 0.22*** -3.47 0.42** -2.27 Sex 1.41*** 2.35 1.69* 3.00 1.86* 3.09 1.34 1.10 Entry age 1.00 -0.51 0.98 -1.44 1.01 0.47 1.00 0.05 Experience 1.06*** 4.69 1.04*** 3.32 1.12*** 6.96 1.06*** 2.94 Union agreement 0.86 -0.81 1.45 1.24 1.59* 1.91 1.53** 1.96 Urban 1.16 0.74 1.10 0.51 1.21 0.96 0.93 -0.30 Free lunch 0.61 -1.50 1.14 0.18 1.06 0.13 0.36*** -3.26 Minority students 0.56 -1.34 0.69 -0.77 1.43 0.63 1.54 0.96 Minority teachers 1.90 1.07 0.70 -0.73 0.39** -2.2 2 0.81 -0.54 District enrollment 1.03 0.18 1.11 1.44 1.02 0.24 0.73** -1.97 bigD LPS 0.72 -1.47 0.71*** -4.59 1.01 0.13 1.01 0.34 Constant 17.10 3.84 15.54 4.97 2.87 2.01 15.02 3.40 Number of observations 8747 6547 5593 5794 State fixes effects yes yes yes yes Weights yes yes yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively.
176 Table B-15. Logit results for district-targeted acco untability policy with controls for district size English Mathematics Social Sciences Sciences Variable odds ratio t-st. odds ratio t-st. odds ratio t-st. odds ratio t-st. District year 0.20*** -3.81 6.55*** 3.09 1.49 1.12 0.43 -1.62 District LPS 9.39*** 4.55 0.12* -1.88 1.79 0.75 1.52 0.37 District bigD 22.37*** 3.56 0.01*** -4.97 2.76* 1.80 7.72 0.48 District *bigD *LPS 0.01*** -3.68 4.49*** 3.44 0.01 -1.56 0.28 -0.21 year 0.83 -0.67 0.64 -1.14 0.79 -0.85 0.50*** -3.05 LPS year 0.52 -1.39 0.52 -0.82 0.58 -0.92 1.40 0.83 white teacher 1.01 0.01 1.11 0.56 0.47*** -2.38 0.68** -2.21 black teacher 0.47 -1.33 0.84 -0.49 0.20*** -3.84 0.42** -2.22 Sex 1.42*** 2.42 1.70*** 3.04 1.92*** 3.14 1.35 1.17 Entry age 1.00 -0.51 0.99 -1.35 1.00 0.32 1.00 0.01 Experience 1.06*** 4.73 1.04*** 3.35 1.13*** 7.08 1.06*** 2.98 Union agreement 0.84 -0.87 1.48 1.22 1.60** 1.98 1.55** 1.99 Urban 1.17 0.82 1.11 0.54 1.20 0.90 0.90 -0.40 Free lunch 0.62 -1.46 1.15 0.21 1.13 0.25 0.35*** -3.01 Minority students 0.53 -1.49 0.68 -0.82 1.34 0.52 1.54 0.95 Minority teachers 2.10 1.12 0.71 -0.72 0.43** -2.0 0 0.79 -0.56 District enrollment 1.36 1.31 1.24 1.55 1.11 1.06 0.76** -2.00 bigD LPS 0.57** -2.24 0.68*** -3.51 0.90 -0.83 0.90 -0.52 Constant 14.01 3.73 11.70 4.67 2.89 2.02 12.30 3.44 Number of obs. 8747 6547 5593 5794 State fixes effects yes yes yes yes Weights yes yes yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively. Table B-16. Logit results for aggr egated accountability policy index w ith controls for district size English Mathematics Social Sciences Sciences Variable odds ratio t-st. odds ratio t-st. odds ratio t-st. odds ratio t-st. Index year 1.00 -0.04 0.89 -0.52 1.00 0.01 0.79 -1.61 Index LPS 0.97 -0.12 0.89 -0.36 0.95 -0.20 1.22 1.18 Index bigD 1.16* 1.73 1.01 0.04 1.06 0.87 1.20*** 2.38 Index *bigD *LPS 0.78*** -2.60 0.95 -0.53 0.96 -0.62 0.79*** -1.99 year 0.61 -1.01 1.03 0.04 0.83 -0.55 0.78 -0.70 LPS year 1.03 0.03 0.82 -0.17 0.66 -0.44 1.03 0.04 white teacher 1.01 0.01 1.11 0.55 0.46*** -2.40 0.69** -2.11 black teacher 0.46 -1.29 0.80 -0.66 0.21*** -3.39 0.44* -1.99 Sex 1.46*** 2.48 1.70*** 3.06 1.88*** 3.18 1.36 1.16 Entry age 1.00 -0.29 0.99 -1.36 1.00 0.29 1.00 0.10 Experience 1.06*** 4.73 1.04*** 3.33 1.13*** 6.94 1.06*** 3.01 Union agreement 0.85 -0.80 1.43 1.15 1.58* 1.95 1.54* 1.96 Urban 1.15 0.71 1.09 0.44 1.23 1.02 0.96 -0.17 Free lunch 0.42*** -2.59 0.94 -0.09 1.08 0.18 0.27*** -3.22 Minority students 0.56 -1.32 0.70 -0.77 1.46 0.65 1.57 0.98 Minority teachers 1.82 0.93 0.66 -0.9 0 0.43* -1.8 8 0.77 -0.62 District enrollment 0.84 -0.44 1.18 0.57 0.87 -0.40 0.41*** -2.47 bigD LPS 1.36 0.71 0.92 -0.20 1.08 0.23 1.01 1.59 Constant 16.78 3.90 14.44 4.65 2.94 2.09 17.46 3.23 Number of obs. 8747 6547 5593 5794 State fixes effects yes yes yes yes Weights yes yes yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively.
177 Table B-17. Logit results for sample consisting of English language arts (base field) and mathematics teachers with controls for district size Student School District Aggregated index Variable odds ratio t-st. odds ratio t-st. odds ratio t-st. odds ratio t-st. Policy year 0.88 -0.28 0.51*** -2.56 0.41** -2.11 0.99 -0.10 Policy*Math 0.74 -0.62 1.07 0.17 10.18*** 7.36 0.93 -0.58 Policy LPS 1.43 0.62 1.26 0.49 4.10*** 3.18 0.90 -0.39 Policy*LPS Math 0.99 -0.01 0.45 -0.89 0.07*** -2.69 1.01 0.04 Policy*bigD 1.42 0.38 3.08* 1.67 8.00*** 2.59 1.11 1.43 Policy *bigD LPS 0.77 -0.25 0.37 -1.30 0.03** -2.17 0.85** -2.10 Policy bigD Math 1.50 0.56 0.69* -1.75 0.01*** -12.10 0.94** -2.07 Policy*bigD*LPS* Math 1.02 0.03 1.44 1.26 5.04*** 5.10 1.07** 2.03 Math 0.53*** -2.38 0.52*** -2.46 0.53*** -2.39 0.52*** -2.42 year 0.70 -1.10 0.78 -0.88 0.76 -1.02 0.64 -1.04 Math year 1.23 0.59 1.14 0.29 0.91 -0.22 1.43 0.65 LPS year 0.57 -1.25 0.74 -0.66 0.61 -1.08 1.22 0.26 LPS Math 2.05 1.23 2.20 1.35 2.08 1.26 2.12 1.28 LPS Math year 0.57 -0.71 0.88 -0.14 0.76 -0.35 0.59 -0.36 white teacher 1.08 0.30 1.12 0.43 1.09 0.36 1.07 0.30 black teacher 0.60 -1.54 0.64 -1.38 0.63 -1.49 0.60 -1.60 Sex 1.57*** 3.71 1.58*** 3.84 1.58*** 3.88 1.60*** 4.04 Entry age 0.99 -1.23 0.99 -1.38 0.99 -1.38 0.99 -1.21 Experience 1.05*** 4.92 1.05*** 4.94 1.05*** 5.04 1.05*** 5.00 Union agreement 1.06 0.51 1.08 0.71 1.08 0.68 1.08 0.66 Urban 1.09 0.63 1.12 0.70 1.12 0.78 1.11 0.68 Free lunch 0.62 -1.62 0.56** -2.02 0.59* -1.77 0.45*** -2.45 Minority students 0.61 -1.41 0.62 -1.35 0.60 -1.48 0.62 -1.37 Minority teachers 1.07 0.14 1.19 0.38 1.27 0. 49 1.17 0.34 District enrollment 1.24 1.55 1.07 1.08 1.29 1.63 0.98 -0.07 bigD LPS 0.66*** -4.65 0.72*** -3.97 0.63*** -3.59 1.09 0.27 Const. 17.46 7.58 20.91 7.70 16.61 7.47 20.49 7.16 Number of obs. 15379 15379 15379 15379 State fixes effects yes yes yes yes Weights yes yes yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively.
178 Table B-18. Logit results for sample consisting of English language arts (base field) and social sciences teachers with controls for district size Student School District Aggregated index Variable odds ratio t-st. odds ratio t-st. odds ratio t-st. odds ratio t-st. Policy year 0.83 -0.40 0.53** -2.10 0.34** -2.08 1.03 0.25 Policy SS 2.86* 1.84 1.42 0.70 2.01 1.04 0.95 -0.25 Policy LPS 1.82 1.02 1.95 1.35 5.10*** 3.07 0.94 -0.24 Policy LPS SS 0.28 -1.46 0.31 -1.47 2.25 1.01 1.02 0.05 Policy bigD 1.71 0.57 3.43* 1.74 5.51 1.23 1.14 1.65 Policy bigD LPS 0.54 -0.59 0.28 -1.57 0.03 -1.29 0.82** -2.11 Policy bigD SS 0.61 -0.78 0.50 -1.03 3.51 1.55 0.93 -1.61 Policy bigD LPS SS 1.95 1.20 1.58 0.77 0.01** -2.01 1.10** 2.03 SS 0.30*** -4.86 0.30*** -4.94 0.30*** -4.86 0.30*** -4.91 year 0.76 -0.79 0.84 -0.57 0.84 -0.58 0.59 -1.20 SS. year 0.65 -1.16 0.87 -0.36 0.83 -0.68 1.20 0.33 LPS year 0.48 -1.60 0.52 -1.35 0.52 -1.36 0.95 -0.07 LPS SS 2.94*** 2.60 2.94*** 2.69 2.92*** 2.62 2.92*** 2.63 LPS SS year 2.08 0.98 2.51 1.49 1.48 0.67 1.15 0.13 white teacher 0.79 -0.52 0.79 -0.53 0.79 -0.57 0.80 -0.50 black teacher 0.37* -1.95 0.37** -1.97 0.36** -2.18 0.37* -1.91 Sex 1.57*** 3.50 1.57*** 3.40 1.58*** 3.49 1.58*** 3.72 Entry age 1.00 -0.18 1.00 0.01 1.00 -0.21 1.00 -0.08 Experience 1.08*** 7.81 1.08*** 7.85 1.08*** 7.74 1.08*** 7.78 Union agreement 1.06 0.34 1.08 0.48 1.08 0.45 1.08 0.47 Urban 1.14 0.89 1.15 0.92 1.15 0.94 1.16 0.97 Free lunch 0.49 -2.41 0.45 -2.85 0.49 -2.44 0.38 -2.86 Minority students 0.76 -1.25 0.79 -1.10 0.74 -1.37 0.78 -1.08 Minority teachers 1.12 0.29 1.05 0.13 1.17 0. 42 1.11 0.23 District enrollment 1.22 1.88 1.07 0.69 1.24 1.88 0.88 -0.42 bigD LPS 0.73 -3.36 0.83 -1.47 0.72 -2.98 1.24 0.67 Constant 12.43 4.77 14.59 4.90 13.07 4.95 13.20 4.55 Number of obs. 14340 14340 14340 14340 State fixes effects yes yes yes yes Weights yes yes yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively.
179 Table B-19. Logit results for sample consisting of English language arts (base field) and sciences teachers with controls for district size Student School District Aggregated index Variable odds ratio t-st. odds ratio t-st. odds ratio t-st. odds ratio t-st. Policy year 0.90 -0.23 0.59 -1.57 0.20*** -4.64 0.96 -0.28 Policy Sc. 1.31 0.72 1.02 0.06 1.95 1.06 0.85 -1.51 Policy LPS 1.57 0.71 1.42 0.70 6.69*** 3.74 0.96 -0.18 Policy LPS Sc. 0.48 -1.20 0.62 -0.75 0.55 -0.47 1.27 1.00 Policy bigD 1.59 0.52 3.24** 2.14 11.28*** 2.53 1.23*** 3.63 Policy bigD LPS 0.70 -0.33 0.36 -1.57 0.01*** -2.36 0.75*** -5.48 Policy bigD Sc. 0.93 -0.10 0.25*** -2.89 5.64 0.38 0.90* -1.68 Policy bigD LPS Sc. 1.70 0.51 5.71*** 3.32 0.70 -0.05 1.00* 1.72 Sc. 0.76 -1.32 0.76 -1.37 0.76 -1.27 0.75 -1.38 year 0.76 -0.89 0.81 -0.69 0.90 -0.42 0.66 -0.99 Sc. year 0.51** -2.02 0.72 -0.90 0.51** -2.22 1.05 0.12 LPS year 0.44* -1.86 0.53 -1.40 0.47* -1.66 1.02 0.03 LPS Sc. 0.95 -0.10 0.96 -0.08 0.94 -0.13 1.03 0.06 LPS Sc. year 3.90* 1.90 2.80 1.46 3.32* 1.98 1.07 0.07 white teacher 0.88 -0.39 0.84 -0.52 0.85 -0.49 0.86 -0.45 black teacher 0.48** -1.98 0.45** -2.17 0.46** -2.22 0.45** -2.14 Sex 1.38* 1.86 1.38* 1.83 1.38* 1.91 1.39* 1.90 Entry age 1.00 -0.37 1.00 -0.49 1.00 -0.55 1.00 -0.34 Experience 1.06*** 4.43 1.06*** 4.35 1.06*** 4.46 1.06*** 4.45 Union agreement 1.09 0.57 1.09 0.58 1.09 0.60 1.09 0.59 Urban 1.02 0.13 1.06 0.33 1.05 0.30 1.06 0.35 Free lunch 0.52** -2.14 0.50*** -2.34 0.51** -2.18 0.35*** -3.26 Minority students 0.82 -0.74 0.85 -0.55 0.82 -0.70 0.85 -0.55 Minority teachers 1.25 0.46 1.36 0.70 1.40 0. 66 1.30 0.53 District enrollment 0.98 -0.41 0.89 -1.41 0.89 -0.01 0.99*** -4.16 bigD LPS 0.99 -1.08 0.99 -0.92 0.99 -1.01 1.01*** 3.86 Constant 16.61 5.34 20.49 4.98 15.80 5.19 17.12 4.65 Number of obs. 14571 14571 14571 14571 State fixes effects yes yes yes yes Weights yes yes yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively
180 Table B-20. Logit results for sample consisting of mathematics (base field) and sciences teachers with controls for district size Student School District Aggregated index Variable odds ratio t-st. odds ratio t-st. odds ratio t-st. odds ratio t-st. Policy year 0.66 -1.00 0.61 -1.11 4.48*** 2.70 0.87 -0.85 Policy Sc. 1.80 1.63 0.91 -0.23 0.21* -1.92 0.93 -0.52 Policy LPS 1.48 0.60 0.53 -0.75 0.14** -2.02 0.94 -0.19 Policy LPS Sc. 0.49 -1.07 1.54 0.48 8.58* 1.73 1.25 0.85 Policy bigD 2.31 1.49 1.77 1.56 0.01*** -4.95 1.13* 1.75 Policy bigD LPS 0.73 -0.45 0.65 -0.88 4.80*** 4.03 0.83** -2.07 Policy bigD Sc. 0.66 -0.92 0.43*** -2.35 2.91*** 2.52 0.97 -0.74 Policy bigD LPS Sc. 1.59 0.61 3.28*** 2.54 0.01** -2.15 1.05 0.91 Sc. 1.35 1.01 1.32 0.96 1.32 0.94 1.35 0.99 year 0.91 -0.28 0.91 -0.26 0.73 -0.84 1.05 0.09 Sc. year 0.42*** -2.49 0.67 -0.90 0.59 -1.16 0.73 -0.60 LPS year 0.30* -1.74 0.64 -0.51 0.44 -1.04 0.68 -0.33 LPS Sc. 0.49 -1.04 0.49 -1.04 0.50 -1.02 0.51 -0.97 LPS Sc. year 6.55*** 2.55 2.75 0.98 4.01 1.54 1.79 0.49 white teacher 0.95 -0.37 0.92 -0.55 0.94 -0.39 0.92 -0.53 black teacher 0.62* -1.73 0.64* -1.66 0.66 -1.48 0.63* -1.72 Sex 1.52*** 2.41 1.52*** 2.32 1.54*** 2.41 1.54*** 2.44 Entry age 0.99 -0.89 0.99 -0.96 0.99 -0.88 0.99 -0.83 Experience 1.05*** 4.12 1.05*** 3.98 1.05*** 4.08 1.05*** 4.09 Union agreement 1.49*** 2.34 1.52*** 2.58 1.55*** 2.60 1.52*** 2.53 Urban 0.99 -0.09 1.01 0.08 1.01 0.05 1.03 0.21 Free lunch 0.97 -0.04 0.91 -0.15 0.90 -0.19 0.68 -0.63 Minority students 1.03 0.09 1.04 0.13 1.05 0.17 1.06 0.17 Minority teachers 0.63 -1.76 0.76 -0 .96 0.73 -1.0 3 0.73 -1.11 District enrollment 0.97 -0.69 0.95 -0.66 0.99 -0.01 0.64** -2.17 bigD LPS 0.86 -1.35 0.82* -1.86 0.84 -1.33 1.68 1.53 Constant 9.78 5.44 12.55 5.31 9.68 4.93 12.94 5.06 Number of obs. 12341 12341 12341 12341 State fixes effects yes yes yes yes Weights yes yes yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively
181 Table B-21. Logit results for sample consisting of mathematics (base field) and social sciences teachers with controls for district size Student School District Aggregated index Variable odds ratio t-st. odds ratio t-st. odds ratio t-st. odds ratio t-st. Policy year 0.58 -1.03 0.59 -0.98 6.42*** 2.69 0.95 -0.26 Policy SS 3.90*** 2.42 1.22 0.34 0.20* -1.88 1.02 0.09 Policy LPS 1.70 0.68 0.66 -0.44 0.09** -2.06 0.88 -0.43 Policy LPS SS 0.29 -1.17 0.80 -0.20 42.52*** 2.76 1.04 0.13 Policy bigD 2.50 1.22 1.68 0.84 0.01*** -5.95 1.05 0.83 Policy bigD LPS 0.62 -0.56 0.64 -0.65 7.05*** 3.90 0.92 -1.15 Policy bigD SS 0.44 -1.27 0.99 -0.02 8.81*** 2.47 0.99 -0.43 Policy bigD LPS SS 1.71 0.72 0.83 -0.31 0.01*** -3.27 1.03 1.33 SS 0.54** -2.17 0.54** -2.20 0.53** -2.17 0.54** -2.19 year 0.92 -0.19 0.90 -0.26 0.69 -0.91 0.85 -0.30 SS year 0.57 -1.19 0.88 -0.25 1.01 0.01 0.94 -0.09 LPS year 0.32* -1.71 0.63 -0.54 0.48 -0.95 0.79 -0.21 LPS SS 1.60 0.98 1.54 0.94 1.52 0.89 1.51 0.89 LPS SS year 2.94 1.27 2.16 0.78 1.63 0.59 1.35 0.26 white teacher 0.90 -0.55 0.90 -0.51 0.90 -0.54 0.88 -0.72 black teacher 0.49*** -2.37 0.52** -2.06 0.52** -2.09 0.50** -2.27 Sex 1.70*** 3.44 1.68*** 3.31 1.70*** 3.34 1.70*** 3.52 Entry age 0.99 -0.63 0.99 -0.50 0.99 -0.60 0.99 -0.63 Experience 1.07*** 6.34 1.07*** 6.48 1.07*** 6.47 1.07*** 6.34 Union agreement 1.43* 1.76 1.49** 2.00 1.51** 1.97 1.49* 1.94 Urban 1.11 0.82 1.11 0.83 1.11 0.85 1.11 0.83 Free lunch 0.90 -0.19 0.81 -0.34 0.84 -0.30 0.71 -0.56 Minority students 1.04 0.11 1.05 0.15 1.02 0.06 1.06 0.17 Minority teachers 0.47*** -2.42 0.50** -2 .03 0.53* -1.95 0.51** -2.04 District enrollment 1.16*** 2.73 1.11** 2.28 1.19*** 2.51 0.96 -0.21 bigD LPS 0.77*** -5.30 0.78*** -3.96 0.75*** -5.12 1.10 0.33 Constant 6.69 3.88 7.92 4.04 7.17 3.87 8.08 4.21 Number of obs. 12140 12140 12140 12140 State fixes effects yes yes yes yes Weights yes yes yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively
182 Table B-22. Logit results for sample consisting of social sciences (base field) and sciences teachers with controls for district size Student School District Aggregated index Variable odds ratio t-st. odds ratio t-st. odds ratio t-st. odds ratio t-st. Policy year 2.42* 1.99 0.88 -0.31 0.97 -0.08 0.95 -0.34 Policy Sc. 0.42 -1.42 0.67 -0.72 1.06 0.09 0.90 -0.52 Policy LPS 0.45 -0.92 0.59 -0.7 3.09 1.47 0.95 -0.19 Policy LPS Sc. 2.10 0.84 1.99 0.79 0.27 -1.18 1.24 0.74 Policy bigD 1.18 0.48 1.78*** 4.14 2.95 1.42 1.14*** 3.51 Policy bigD LPS 0.96 -0.07 0.47*** -4.43 0.01* -1.92 0.87** -2.02 Policy bigD Sc. 1.35 0.67 0.42*** -4.37 0.01* -1.69 0.97 -1.05 Policy bigD LPS Sc. 1.15 0.13 3.89*** 6.51 4.54*** 2.47 1.03 0.64 Sc. Dummy 2.45*** 4.4 2.43*** 4.43 2.45*** 4.48 2.46*** 4.29 Year dummy 0.54*** -2.61 0.79 -0.87 0.78 -1.08 0.81 -0.58 Sc. year 0.79 -0.61 0.78 -0.65 0.59* -1.77 0.79 -0.41 LPS year 0.88 -0.19 1.07 0.13 0.67 -0.79 0.90 -0.13 LPS Sc. 0.32*** -2.49 0.34*** -2.37 0.33*** -2.39 0.34*** -2.34 LPS Sc. year 1.95 0.95 1.48 0.62 2.69* 1.92 1.38 0.34 white teacher 0.70*** -2.35 0.66*** -2.75 0.68*** -2.54 0.66*** -2.67 black teacher 0.39*** -2.95 0.37*** -3.24 0.38*** -3.06 0.38*** -3.11 Sex 1.56*** 3.69 1.53*** 3.44 1.56*** 3.56 1.54*** 3.55 Entry age 1.00 0.13 1.00 0.24 1.00 0.13 1.00 0.14 Experience 1.09*** 6.67 1.09*** 6.64 1.09*** 6.66 1.09*** 6.67 Union agreement 1.54*** 2.64 1.56*** 2.79 1.57*** 2.84 1.57*** 2.86 Urban 1.06 0.43 1.06 0.47 1.06 0.44 1.10 0.75 Free lunch 1.04 0.12 0.92 -0.24 0.96 -0.12 0.82 -0.59 Minority students 1.56 1.22 1.59 1.21 1.56 1.19 1.60 1.18 Minority teachers 0.52*** -2.67 0.55*** -2 .51 0.55*** -2.3 5 0.57** -2.04 District enrollment 0.91 -1.07 0.90 -1.00 0.91 -1.02 0.58*** -3.58 bigD LPS 1.04 0.28 1.03 0.24 1.05 0.38 1.72** 2.07 Constant 2.95 2.15 3.75 2.59 3.41 2.53 4.08 2.81 Number of obs. 11417 11417 11417 11417 State fixes effects yes yes Yes yes Weights yes yes Yes yes *, **, *** statistically significant at 10, 5, 1 percent level respectively
183 LIST OF REFERENCES American Association of State Colleg es and Universities (AASCU) (2005). The factsand fictionsabout teacher shortages. Washington, DC: American Association of State Colleges and Universities. Almeida, B., Kenneally, K., & Madland, D. (2008). The new intersection on the road to retirement: public pensions, economics, pe rceptions, politics, and interest groups. Wharton Pension Research Council Working Paper WP 2008-12, University of Pennsylvania American Association for Employ ment in Education (AAEE) (2005 ). Educator supply and demand in the United States: Full report Columbus, OH: American Association for Employment in Education. Baker, D.P., & Smith, T. (1997). Three trends in the condition of educati on in the United States. Teachers College Record 99(1), 12. Ballou, D. (1998). Alternativ e certification: A comment. Educational Evaluation and Policy Analysis 20(4), 331. Ballou, D., & Podgursky, M. (1997). Teacher pay and teacher quality. Kalamazoo, Michigan: W.E. Upjohn Institute for Employment Research. Barksdale-Ladd, M. A., & Karen, F. T. (2000). Wh ats at stake in high-stakes testing: Teachers and parents speak out. Journal of Teacher Education 51(5), 384. Barr, N., & Diamond, P. (2006). Reforming pensions. Oxford Review of Economic Policy 22(1), 15. Boe, E. E., Cook, L. H., & Sunderland, R. J. (2007). The supply, qualifications, and attrition of teachers from traditional and alternative routes of preparation Paper presented at the 2007 OSEP Project Directors C onference, Washington, DC. Boe, E. E., Cook, L. H., & Sunderland, R. J. (2006). Attrition of beginn ing teachers: Does preparation matter ? Center for Research and Evaluati on in Social Policy Research Report No. 2006-TSDQ2, University of Pennsylvania. Boe, E. E., Shin, S., & Cook, L.H. (2007). Does teacher preparation matter for beginning teachers in either specia l or general education? The Journal of Special Education 40(3), 158. Bonafede., J. K., Foresti, S. J., & Browning, A. (2009). 2009 Wilshire report on state retirement systems: funding levels and asset allocation Santa Monica, CA:Wilshire Associates Incorporated.
184 Boyd, D., Grossman, P., Lankford, H., Wyckoff, J., & Loeb, S. (2006). How changes in entry requirements alter the teacher workforce and affect student achievement. Education Finance and Policy, 1(2), 176. Boyd, D., Lankford, H., Loeb, S., & Wyckoff, J. (2008). The impact of assessment and accountability on teacher recruitment and rete ntion: Are there unintended consequences? Public Finance Review 36(1), 88. Boyd, D., Loeb, S., Lankford, H., & Wykoff, J. (2005). Explaining the short careers of highachieving teachers in schools w ith low-performing students. American Economic Association Proceedings 95(2), 166. Carnoy, M., & Loeb, S. (2003). Does external a ccountability affect stude nt outcomes? A crossstate analysis. Education Evaluation and Policy Analysis 24(4), 305. Clark, R. L., Craig, L. A., & Ahmet, N. (2008). The evolution of public sector plans in the United States. Wharton Pension Research Council Wo rking Paper WP2008-16, University of Pennsylvania. Clotfelter, C. T., Ladd, H. F., Vigdor, J. L., & Diaz, R.A. (2004) Do school accountability systems make it more difficult for LPSs to attract and retain high-quality teachers? Journal of Policy Analysis and Management, 23 (2), 251. Clotfelter, C. T., Ladd, H. F., & Vigdor, J. L. (2005). Who teaches whom? Race and the distribution of novice teachers. Economics of Education Review 24(4), 377. DArcy, S. P., Dulebohn, J.H., & Oh, P. (1999) Optimal funding of state employee pension systems. Journal of Risk and Insurance 66(3), 345. Darling-Hammond, L., Holtzman, D. J., Gatlin, S. J., & Heilig, J. V. (2005) Does teacher preparation matter? Evidence about teacher cer tification, Teach for America, and teacher effectiveness. Education Policy Analysis Archives 13(42), 57. Decker, P. T., Mayer, D. P., & Glazerman, S. (2004). The effects of Teach for America on students: Findings from a national evaluation. Washington, DC: Mathematica. Dewey, J. (2005). Improvements to the 2003 Florida price level index: background, theory, tests based on the current population survey, the relationship between excess teacher turnover in Floridas schools and the market ba sket approach, and robustness checks. Gainesville, FL: Bureau of Economic and Business Research Technical Report, University of Florida. Dolton, P. J., & van der Klaauw, W. (1995). Le aving teaching in UK: A duration analysis. The Economic Journal 105(429), 432. Epple, D., & Schipper, K. (1981). Municipal pe nsion funding: A theory and some evidence. Public Choice, 37(1), 179.
185 Feistrizer, C.E. (2005). Profile of alternative route teachers. Washington, DC: National Center for Educational Information. Figlio, D. N., & Rouse, C. (2005). Do acc ountability and voucher threats improve low performing schools? Journal of Public Economics 90(1-2), 239. Freedberg P. S., & Humburg, C. ( 2008). Florida pension fund plummets. St. Petersburg Times, November 17. Garrahan, M. (2009). Biggest pens ion funds record steep losses. Financial Times July 22, p.3. Grissmer, D.W., & Kirby, S.N. (1992). Patterns of attrition am ong Indiana teachers, 1965-1987 Santa Monica, CA: RAND. R-4076-LE. Hakim, D. (2009). Pension costs fo r local governments may triple. The New York Times July 7. Hanushek, E. A., & Raymond, M.E. (2004). The e ffect of school accountability systems on the level and distribution of student achievement. Journal of the European Economic Association 2(2-3), 406. Hanushek, E.A., Kain, J.F., & Rivkin S.G. (2004). Why public schools lose teachers. Journal of Human Resources, 39(2), 326. Hanushek, E.A., Kain, J.F., OBrien, D.M., & Rivkin, S.G. (2005). The market for teacher quality. NBER Working Paper 11154. Hoffman, J. V., Assaf, L. C., & Paris, S. G. (2001). High-stakes testing in reading: Today in Texas, tomorrow? Reading Teacher 54(5), 482. Hoover, T. (2009). Colorado PERA benefits in peril. The Denver Post July 14. Ibbotson SBBI valuation yearbook: Market re sults for stocks, bonds, bills, and inflation 19262007 (2008). Chicago, IL: Morningstar. Imazeky, J. (2005). Teacher sala ries and teacher attrition. Economics of Education Review, 24, 431. Ingersoll, R.M. (1997). Teacher turnover and te acher quality: The recurring myth of teacher shortages. Teachers College Record 99(1), 41. Kane, T. J., Rockoff, J.E., & Staiger, D.O. ( 2008). What does certification tell us about teacher effectiveness? Evidence from New York City. Economics of Education Review, 27(6), 615. Kirby, S. N., Naftel, S., & Berens, M. (1999). Staffing at-risk school dist ricts in Texas: problems and prospects Santa Monica, CA: RAND, MR-1083-EDU. Krieg, J. (2004). Teacher quality and attrition Bellingham, WA: College of Business Economics.
186 Lankford, H., Loeb, S., & Wyckoff, J. (2002). Teach er sorting and the plight of urban schools: A descriptive analysis. Educational evaluati on and policy analysis 24(1), 38. Loeb, S., & Estrada, F. (2003). Have assessmen t-based school accountability reforms affected the career decisions of teachers? In C. A. Dwyer (Ed.), Measurement and research in the accountability era (pp.225). Mahwah, NJ: Lawr ens Erlbaum Associates. Lucas, D. J., & Zeldes, S. P. (2009). How should public pension funds invest? American Economic Review: Papers & Proceedings 99(2), 527. Luna, C., & Turner, C. L. (2001). The impact of the MCAS: Teachers talk about high-stakes testing. English Journal 91(1), 79. Manski, C. F. (1987). Academic ability, earn ings, and the decision to become a teacher: Evidence from the national longitudinal study of the high school class of 1972. In D. Wise (Ed.), Public sector payrolls (pp. 291). Chicago, IL: Un iversity of Chicago Press. Mendelson, R. (2008). Is the Ster n Review an economic analysis? Review of Environmental Economics and Policy 2(1), 45. Murnane, R.J., & Olsen, R.J. (1989). The effect of salaries and opportunity costs on duration in teaching: evidence from Michigan. Review of Economics and Statistics 71(2), 347. Murnane, R.J., Stringer, J. D., & Willet, J.B. (1 988). The career paths of teachers: Implications for teachers supply and methodological lessons for research. Educational Researcher, 17(6), 22. Murphy, P., DeArmond, M., & Guin, K. (2003). A national crisis or localized problems? Getting perspective on the scope and s cale of the teacher shortage. Education Policy Analysis Archives 11(23), 1. National Center for Educati on Information (NCEI) (2005). Profile of teachers in the U.S., 2005. Washington, DC. National Commission on Teaching and Am ericas Future (NCTAF) (2008). Learning teams: Creating whats next. Washington, DC: National Comm ission on Teaching and Americas Future (NCTAF). Nordhaus, W. D. (2006). The Stern Review on the economics of climate change. NBER Working Paper 12741. Podgursky, M., Monroe, R., & Watson, D. (2004). The academic quality of public school teachers: An analysis of entry and exit behavior. Economics of Education Review 23, 507 518. Raymond, M., Fletcher, S.H ., and Luque, J. (2001). Teach for America: An evaluation of teacher differences and student outco mes in Houston, Texas. Stanford, CA: Center for Research on Education Outcomes.
187 Reback, R. (2006). Teaching to the rating: School accountab ility and the distri bution of student achievement. Barnard College Working Paper 0602, Columbia University Rivkin, S.G., Hanushek, E.A., Kain, J.F. (2005) Teachers, schools, and academic achievement. Econometrica 73 (2), 417. Rouse, C. E., (1998). Private school vouchers an d student achievement: an evaluation of the Milwaukee parental choice program. The Quarterly Journal of Economics 113(2), 553 602. Siegel, J.J. (2007). Stocks for the long run: The definiti ve guide to financial market returns and long term investment strategies, 3rd edition. New York, NY: The McGraw Hill. Stern, N. (2006). Stern Review on the econom ics of climate change. London, UK: Her Majesty Treasury. Stinebrickner, T. R. (1998). An empirical investigation of teacher attrition. Economics of Education Review 17(2), 127. Stinebrickner, T. R. (2002). An analysis of o ccupational change and departure from the labor force: evidence of the reasons that teacher leave. The Journal of Human Resources 37(1), 192. Stinebrickner, T.R. (1999). Using latent variable s in dynamic, discrete choice models the effect of school characteristic s on teacher decisions. Research in Labor Economics, 18, 141. US Department of Education, National Cent er for Educational Statistics (NCES) (2007). Teacher attrition and mobility: Results from the 2004-05 teacher follow-up survey. Washington, DC: US Government Prin ting Office NCES 2007-307. Waring, M.B., (2008). Between Scylla and Charybdis: Improvi ng the cost effec tiveness of public pension funds. Wharton Pension Research Council Working Paper WP2008-25, University of Pennsylvania. Weyant, J. P. (2008). A critique of the Stern Re views mitigation cost an alyses and integrated assessment. Review of Environmental Economics and Policy 2(1), 77.
188 BIOGRAPHICAL SKETCH Nataliya Pakhotina was born in N ovosibirsk, Russia. She received a bachelors degree in Operations Research in Economics and a maste rs degree in Economics from Novosibirsk State University. She entered the PhD program at the University of Florida in 2003. Her research interests include public economics, economics of education, labor economics, and public finance.