Essays on Economic Aspects of Abortion in the United States BY SARA BORELL1 B.A., University of Verona, 2001 M.Sc, Collegio Carlo Alberto, 2002 M.A., University of Cergy Pontoise, 2003 M.A., University of Illinois at Chicago, 2006 THESIS Submitted as partial fulfillment of the requirements for the degree of Doctor of Philosophy in Economics in the Graduate College of the University of Illinois at Chicago, 2011 Chicago, Illinois UMI Number: 3484945 All rights reserved INFORMATION TO ALL USERS The quality of this reproduction is dependent upon the quality of the copy submitted. In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if material had to be removed, a note will indicate the deletion. Dissertation Publishing UMI 3484945 Copyright 2011 by ProQuest LLC. All rights reserved. This edition of the work is protected against unauthorized copying under Title 17, United States Code. uest ProQuest LLC 789 East Eisenhower Parkway P.O. Box 1346 Ann Arbor, Ml 48106-1346 This dissertation is dedicated to Davide Furceri, and to my parents Franca and Gigi Borelli in ACKNOWLEDGMENTS I am especially grateful to my supervisor, Robert Kaestner, for his support and supervision over the past years. 1 could have not accomplished this dissertation without him. I am also grateful to the other members of my dissertation committee Nathan Anderson, Barry Chiswick, Theodore Joyce and Anthony Lo Sasso for their feedbacks on my research and encouragement. Many people contributed in important ways to the completition of my graduate studies. My parents, Franca and Gigi Borelli, and the rest of my family have been an essential source of support over the years. They have been always present with their love and encouragement giving me the strength to go on. Special thanks go to Rosa Berardi, a dear friend, for hosting me at her house. I really felt I have found another family on the other side of the world. My thanks extend to my Italian friends who always supported me, especially Lidia Monaco, and to my UIC classmates for sharing very difficult times on the way. Above all, heartfelt gratitude goes to Davide Furceri for always being beside me with his love, understanding, patience and encouragement, sharing this experience with me from the beginning to the end. SB IV TABLE OF CONTENTS CHAPTER PAGE 1. INTRODUCTION 1.1 Background 1.1.1 Review of Abortion Legislation 1.2 Purpose of the Study and Contributions 2. EFFECT OF PROVIDER SUPPLY ON THE DEMAND OF ABORTION 2.1 Introduction 2.2 Abortion Legislation and Structure of the Abortion Market 2.3 Previous Research on Provider Availability 2.4 Conceptual Framework 2.5 Data 2.6 Empirical Specification 2.7 Results 2.7.1 Baseline Results 2.7.2 Alternative Specifications 2.7.3 Abortion and Marital Status 2.7.4 Provider Size 2.7.5 Endogeneity of Provider Availability 2.8 Conclusions 3. LONG 3.1 3.2 3.3 3.4 3.4.1 3.4.2 3.5 3.6 3.7 3.7.1 3.7.2 3.7.3 3.7.4 3.8 3.9 4. CONCLUSIONS 4.1 Overview 4.2 Summary of Contributions 4.3 Discussion 1 1 2 6 14 14 17 21 25 30 37 40 40 41 46 51 54 64 TERM EFFECTS OF ABORTION PARENTAL INVOLVEMENT LAWS 66 Introduction 66 Theoretical Impact of Parental Involvement Laws on Adult Fertility 69 Brief History of Abortion Parental Involvement Laws 71 Previous Research 76 Studies of Current Impact of Parental Involvement Laws 76 Previous Research on the Long Term Effects of Changes in Costs of Fertility Control.. 84 Empirical Specification 88 Data 89 Results 93 Fertility 93 Interpretation of the Results 96 Educational Attainment 98 Labor Market Outcomes 102 Robustness Checks 106 Conclusions 113 116 116 117 122 v CHAPTER PAGE CITED LITERATURE 125 APPENDICES APPENDIX A APPENDIX B APPENDIX C 129 129 130 136 VITA 138 VI LIST OF TABLES PAGE TABLE I. ABORTIONS BY STATE OF OCCURRENCE, NCHS AND AGI 32 II. VARIATION IN DISTANCE BY POPULATION DENSITY 37 III. DESCRIPTIVE STATISTICS 40 IV. EFFECT OF DISTANCE ON ABORTION RATES 42 V. EFFECT OF DISTANCE -DUMMIES-ON ABORTION RATES 43 VI. EFFECT OF PROVIDER RATE ON ABORTION RATES 44 VII. EFFECT OF DISTANCE ON ABORTION RATES BY MARITAL STATUS-WHITES 48 EFFECT OF DISTANCE ON ABORTION RATES BY MARITAL STATUS-NON-WHITES 49 VIII. IX. EFFECT OF DISTANCE ON ABORTION RATES BY MARITAL STATUS AND AGE 50 X. EFFECT OF DISTANCE TO A LARGE PROVIDER 53 XL EFFECT OF DISTANCE ON ABORTION RATES-COUNTIES WITH DISTANT PROVIDERS 55 XII. EFFECT OF DISTANCE ON ABORTION RATES-OLDER WOMEN 60 XIII. XIV. FIRST STAGE REGRESSION EFFECT OF DISTANCE ON ABORTION RATES-OLDER WOMEN-COUNTIES WITH POPULATION DENSITY < 300 RESIDENTS PER SQ. MILE FIRST STAGE REGRESSION COUNTIES WITH POPULATION DENSITY < 300 RESIDENTS PER SQ. MILE 61 63 XVI. ABORTION RESTRICTIONS-PERIODS ENFORCED 75 XVII. EFFECT OF EXPOSURE ON NUMBER OF CHILDREN PER WOMAN -WHITES .. 95 XVIII. EFFECT OF EXPOSURE ON NUMBER OF CHILDREN PER WOMAN -BLACKS . 96 XIX. EFFECT OF EXPOSURE ON COMPLETED HIGH SCHOOL - WHITES XV. vii 62 99 LIST OF TABLES (continued) PAGE TABLE XX. EFFECT OF EXPOSURE ON COMPLETED HIGH SCHOOL-BLACKS 100 XXI. EFFECT OF EXPOSURE ON SOME COLLEGE - WHITES 101 XXII. EFFECT OF EXPOSURE ON SOME COLLEGE-BLACKS 102 XXIII. EFFECT OF EXPOSURE ON WHETHER WORKED LAST YEAR-WHITES 103 XXIV. EFFECT OF EXPOSURE ON WHETHER WORKED LAST YEAR-BLACKS 104 XXV. EFFECT OF EXPOSURE ON EMPLOYED - WHITES 105 XXVI. EFFECT OF EXPOSURE ON EMPLOYED-BLACKS 106 XXVII. EFFECT OF EXPOSURE ON NUMBER OF CHILDREN PER WOMAN WHITES-LOW MIGRATION STATES 109 EFFECT OF EXPOSURE ON COMPLETED HIGH SCHOOL WHITES-LOW MIGRATION STATES 110 EFFECT OF EXPOSURE ON SOME COLLEGE WHITES-LOW MIGRATION STATES 111 EFFECT OF EXPOSURE ON WHETHER WORKED LAST YEAR WHITES-LOW MIGRATION STATES 112 XXVIII. XXIX. XXX. XXXI. EFFECT OF EXPOSURE ON EMPLOYED -WHITES LOW MIGRATION STATES 113 XXXII. DATA SOURCES 129 XXXIII. XXXIV. EFFECT OF DISTANCE ON ABORTION RATES, ALL ARA COUNTIES EFFECT OF PROVIDER RATES ON ABORTION RATES BY COUNTY OF OCCURRENCE 130 132 XXXV. EFFECT OF DISTANCE ON ABORTION RATES - OLDER WOMEN COUNTIES WITH POPULATION DENSITY < 100 RESIDENTS PER SQ. MILE... 133 XXXVI. FIRST STAGE REGRESSION COUNTIES WITH POPULATION DENSITY < 100 PER SQ. MILE VIM 135 LIST OF TABLES (continued) TABLE XXXVII. XXXVIII. PAGE EFFECT OF EXPOSURE ON NUMBER OF CHILDREN PER WOMAN WHITES-ALL CELLS 136 EFFECT OF EXPOSURE ON NUMBER OF CHILDREN PER WOMAN BLACKS-ALL CELLS 137 IX LIST OF FIGURES PAGE National abortion rate 1973-2005 7 Number of abortion providers 1973-2005 8 States with Parental Involvement laws in effect as of February 2011 11 Probability of provider in county and county population 20 Probability of provider in county and county population density 21 Abortion costs and probability of abortion given pregnancy 29 Abortion costs and probability of pregnancy 29 Abortion costs and probability of abortion 30 Trends in abortion rates, thirteen states area and national trends 33 Total, hospital and non-hospital providers 1973-2005, 50 states-AGI data 35 Counties experiencing a change in distance to the nearest provider, 1979-1988 36 Counties experiencing a change in distance to the nearest large provider, 1979-1988.... 36 Impact of Parental Involvement laws on Fertility 71 Number of states with minors' abortion restrictions, 1973-2011 74 Own children in the household and children ever born (CEB) per woman 91 x LIST OF ABBREVIATIONS AFDC Aid to Families with Dependent Children AGI Alan Guttmacher Institute ARA Abortion Reporting Area BEA Bureau of Economic Analysis CDC Center of Disease Control CEB Children Ever Born IV Instrumental Variables MPC Model Penal Code NCHS National Center of Health Statistics NCI National Cancer Institute OLS Ordinary Least Square PI laws Parental Involvement laws XI SUMMARY After nationwide abortion legalization in 1973, government policies continued to change becoming, in general, less favorable to abortion and accessibility of providers started to decline in the early 1980s. This research analyses the impact of some of these changes on abortion demand. The first chapter provides some background about abortion in the U.S., in particular the legal issues and framework that serve as reference to place the results of subsequent chapters into context. It also describes the purpose of the study and its contributions. Chapter two examines how changes in accessibility to abortion providers have affected abortion rates in the U.S. in the 1980s. The analysis of accessibility to abortion providers is important because the supply of services in this market has been traditionally uneven and restricted compared to other medical services. State level regulations, violence against providers and stigma impose burdensome constraints on the provision of abortions and may help to understand both the change in the number of providers over time and their geographical distribution. While some previous research has examined the association between abortion availability and abortion rates, these studies have been limited in their ability to provide estimates of a causal relationship. This is the first study to use a large panel of counties and a fixed effects approach to obtain robust estimates of the impact of changes in provider availability on abortion demand. Furthermore, I analyze the effect of changes in abortion services on different groups of women distinguished on the basis of demographic characteristics which can signal different behavioral responses. The chapter also addresses the simultaneity problem in the demand and supply of abortion services by means of an instrumental variable approach. The results show that availability of abortion services is strongly correlated with abortion rates over the sample period considered. The estimates are robust to a wide range of specifications and the instrumental variable results suggest that the observed correlation between abortion rates and provider availability can be plausibly interpreted as causal. The second part of the dissertation (chapter three) analyses the long term impact of abortion Parental Involvement laws for minors. Starting in the early 1980s, states instituted laws such that minors xii SUMMARY (continued) were required to either notify a parent(s) or obtain their consent before receiving an abortion. Together, these laws are referred to as Parental Involvement laws. Today, the majority of states require some form of parental involvement in minors' abortions. Previous research on parental laws has focused on the shortrun impact on abortion behavior of minors, and to a less extent births. But the effect of parental involvement laws may go beyond the short-run and affect a woman's in the long-run. In fact, the timing of first birth (a teen birth) may influence subsequent education choices, marriage, and other factors that influence fertility. Thus, the purpose of this project is to investigate whether parental laws have impact on fertility and other socioeconomic outcomes in the long-run. The research design exploits the fact that states enacted Parental Involvement laws at different times to identify cohorts that have been more or less exposed to abortion restrictions as minors. The results show that women who were more exposed to abortion restrictions as minors experience higher fertility later in life and have lower educational attainment. Some estimates also suggest non-zero effects on labor market outcomes. The results indicate that Parental Involvement laws may have permanent effects on women's outcomes. Finally, chapter four summarizes the main contributions of this dissertation, discusses its limitations and some venues for future research. xin 1. INTRODUCTION 1.1 Background In recent decades, significant global changes have occurred in both developed and developing countries towards legalization and regulation of abortion. This issue has received considerable attention, and its legality and availability have often generated controversy. In the Unites States, the movements towards legalization began in the mid-1960s, and legal abortion was suddenly extended to the entire country in 1973 with the landmark ruling of the U.S. Supreme Court in Roe v. Wade. Since then, abortion policy has been one of the most contentious issues in the U.S. domestic political agenda. In fact, over the past four decades, states have enacted laws and administrative rulings mostly aimed at regulating and limiting whether, when and under what circumstances a woman may terminate her pregnancy, and also at directly regulating the availability of abortion services. While the public debate about abortion is usually focused on ideological extremes, the contribution of economics is potentially important because it is focused on measuring consequences, which is central to the practical argument surrounding abortion laws and regulations. But why is abortion an economic issue? From an economic point of view, abortion is an individual choice, which can be affected by constraints (including economic constraints). These constraints can be altered by policies which, in turn, affect individual choices. Economic analysis can reveal the importance of these policies and of economic constraints, and the evidence gathered can be used to assess the consequences and to design policies targeted at abortion. The debate about abortion has fueled multiple lines of investigation in economic research: from the analysis of various types of restrictions and determinants of abortion, to the study of its effect on rliffprv^nt c n p i n - p p n t i n m i f 1 A i i t p n m p c v ^~~.~ ~~w.*v^**.~ w*.~^...w.J. A Inner th^cp* l i n p c t h i c H i c c ^ r t Q t i r m - f r ^ n c ^ c r\n tUp* c t n r K / n f u ; n m p n ' c ; < ».^.. ^.^^ & k .,~^~ *.*.~^, „ . ^ ~ . behavioral responses to changes in the abortion environment. 1 t «.-.v» A ww«^*-o v.. m ^ es yes no yes 13 92 8704 yes yes no yes 41 22 8704 County fixed effect) Year fixed effects State by year fixed effects County linear trend > Mean abortion rate Observations a -18 5764*** (2 0353) Marital status (l=marned) Standard errors clustered at the county k *** 1%, ** 5%, * 10% significance level -18 5772*** (2 0226) weight by the population of women in each cell 50 TABLE IX EFFECT OF DISTANCE ON ABORTION RATES BY MARITAL STATUS AND AGE Married Unmarried Married Unmarried Married Unmarried WHITES AGE 18-30 Miles to nearest provider Effect of 1 S.D. in distance relative to mean abt. rate Mean abortion rate -2.7256** (1.2536)a [-0.0801] -10.0134*** (3.6724) [-0.0684] -2.4896** (1.2291) [-0.0731] -9.0854*** (3.3349) [-0.0621] -1.5083* (0.8054) [-0.0443] -7.8210*** (2.3338) [-0.05346] 9.1902 39.5032 9.1902 39.5032 9.1902 39.5032 WHITES AGE 30-34 Miles to nearest provider -0.8823 (0.5731) -7.6925*** (2.4655) -0.8346 (0.5563) -8.4162*** (2.2028) -0.7062 (0.6230) -5.3962* (3.2514) Effect of 1 S.D. in distance relative to mean abt. rate Mean abortion rate [-0.0539] [-0.0947] [-0.0509] [-0.1034] [-0.0427] [-0.0665] 4.4189 21.9384 4.4189 21.9384 4.4189 21.9384 WHITES AGE 35-44 Miles to nearest provider Effect of 1 S.D. in distance relative to mean abt. rate Mean abortion rate -0.5704*** (0.2098) -3.5530*** (0.8601) -0.5423** (0.2189) -3 7173*** (0.8860) -1.0461*** (0.3084) -1.1962 (1.0133) [-0.0784] [-0.1145] [-0.0745] [-0.1200] [-0.1436] [-0.0386] 1.9635 8.3663 1.9635 8.3663 1.9635 8.3663 County fixed effects yes yes yes yes Yes Year fixed effects yes yes no no Yes State by year fixed effects no yes yes no No County linear trends no no no no Yes 1 Standard errors clustered at the county level; regressions weight by the population of women in each cell. *** 1%, ** 5%, * 10% significance level. yes yes no yes 51 2.7.4 Provider Size In evaluating the impact of changes in availability of abortion providers, it is important to consider that the presence of a given facility in an area does not necessarily represent meaningful access to services for women living there and in surrounding areas. In fact, the majority of abortions occur in large facilities and in many cases only large providers advertise their existence and services and are well known. Large providers are usually specialized abortion clinics and the percentage of non-hospital providers among large providers increased steadily over the sample period (from 70 to 90 percent). Instead, small providers more frequently coincide with a hospital or non-specialized clinic.24 Large and small providers charge different prices (with large specialized clinics providing lower priced abortions) and usually maintain different gestational limits. This is an important aspect of access because even if a woman has located a nearby abortion provider, services may not be available to her from that provider if her pregnancy has passed the earliest stages. For example, while the majority of providers offer services until eight weeks' gestation, the proportion of providers offering services drops after eight weeks and declines steeply after twelve weeks. Non-specialized clinics usually stop providing services at twelve weeks while specialized clinics are most likely to perform abortions until seventeen weeks; after seventeen weeks the majority of abortions are performed in hospitals (Henshaw and Finer, 2003). This suggests that women choose abortion services in a differentiated market and that changes in availability of large and small providers may have a differential impact on abortion rates, an issue that has been not investigated in previous research. In this section, I re-estimate the baseline model using as access measure distance to the nearest county with at least one large provider (where a large provider is defined as one performing more than Specialized clinics are usually defined as those where at least half of patient visits are for abortion services. Nonspecialized clinics are usually those in which the majority of patients receive services other than abortion (this include physician offices). 52 400 abortions in a given year). 25 To test the robustness of the estimates, I also include distance to the nearest small provider since the latter may play a role in areas without large facilities available within easily travelling distances (and interact distance to a small provider with a dummy indicating whether the small provider is also the nearest). The results (TABLE X) show that distance to a small provider is always insignificant while distance to a large provider is negative and significant, albeit its magnitude is sensitive to the inclusion of county linear trends (this may be due to the smaller within variation in distance to large provider over the sample period). Overall the estimates reflect the structure of a market where the large majority of abortions take place in a relatively small number of large facilities and where large providers represent the most effective supply of abortion services. 25 In distinguishing between large and small providers I follow previous classifications (Henshaw et al., 1982, 1984). Following the strategy described in the previous section, to minimize measurement error in abortion rates by residence, I run estimates on a sample of counties whose nearest large provider is always in ARA (496 counties). Standard deviation of distance to large provider is approximately 35 miles. TABLE X EFFECT OF DISTANCE TO A LARGE PROVIDER mean abortion rate=16 67, obs = 17488 Miles to nearest large provider (100) Effect of 1 S D in distance to large provider relative to mean 1A IB 1C ID 2A 2B 2C 3A 3B 3C -2 1129*** (1 0106)a [-0 0450] -2 6095** (1 1630) [-0 0555] -2 5808** (1 1689) [-0 0549] -2 2441** (1 0473) [-0 0477] -3 1851*** (0 9418) [-0 0677] .3 0844*** (0 9622) [-0 0656] -2 9269*** (0 9204) [-0 0623] -1 1484*** (0 4136) [-0 0244] -1 1400*** (0 4247) [-0 0243] -1 0478*** (0 4018) [-0 0223] 0 1697 (0 2876) 0 1071 (0 3015) 0 6028 (0 4172) 0 5759 (0 4271) 0 0424 (0 3825) 0 0257 (0 3972) Miles to nearest small provider(lOOs) -1 2903* (0 7173) Miles to nearest small provider * (Nearest provider small) -0 3677 (0 8980) -0 5251 (0 5913) 2 4456 (2 6257) 2 2795 (2 6371) 2 4145 (2 6246) -2 2 (2 3605) -2 7144 (2 3791) -2 5851 (2 3727) 6 2650** (2 4830) 6 2004** (2 4435) 6 1777** (2 4291) Percent of income from unemployment insurance -1 0174*** (0 3824) -0 9736** (0 3875) -0 9846** (0 3879) -0 203 (0 5108) -0 1392 (0 4805) -0 1388 (0 4818) -0 4132 (0 3058) -0 4113 (0 3029) -0 4127 (0 3038) Medicaid funding restricted -1 5442** (0 6786) -1 5565** (0 6779) -1 5617** (0 6696) -2 2891** (0 9736) -2 2979** (0 9547) -2 2798** (0 9569) -0 3316 (0 6978) -0 3191 (0 6987) -0 3423 (0 6951) -1 2876** (0 5020) -1 2925*** (0 4865) -I 2843*** (0 4849) Race (l=white) -5 4970*** (0 9221) -5 4968*** (0 9219) -5 4975*** (0 9219) -5 4987*** (0 9226) -5 4981*** (0 9223) -5 5025*** (0 9281) -5 5025*** (0 9281) -5 5025*** (0 9282) Marital status (l=marned) -21 1013*** (1 1705) -21 1012*** (1 1705) -21 1013*** (1 1705) -21 1005*** -21 1004*** -21 1004*** (1 1713) (1 1713) (1 1713) -21 1008*** (1 1766) -21 1008**" (1 1766) -21 1008*** (1 1766) Age 25-29 -9 3494*** (0 5478) .93494*** (0 5478) -9 3493*** (0 5478) -9 3500*** (0 5474) -9 3502*** (0 5474) -9 3521*** (0 5498) -9 3521*** (0 5497) -9 3521*** (0 5497) Age 30-34 -15 4661*** (0 8172) -15 4662*** (0 8173) -15 4660*** (0 8173) -15 4648*** -15 4653*** -15 4652*** (0 8176) (0 8176) (0 8176) -15 4612*** (0 8217) -154611*-* (0 8217) -154611*** (0 8217) Age 35-44 -20 8186**" (1 0966) -20 8186*** (1 0967) -20 8183*** (1 0967) -20 8153*** -20 8157*** -20 8156*** (1 0984) (10983) (1 0984) -20 8091*** (1 1031) -20 8090*** (1 1031) -20 8090*** (1 1031) Log per capita income Medicaid laws en|oined Distance small provider jointly sign -5 4978*** (0 9223) -9 3503*** (0 5473) 01411 _ yes yes yes yes no no yes yes yes yes yes no no no no yes yes yes 0 118 County fixed effects yes yes yes yes yes Year fixed effects yes yes yes yes no State by year fixed effects no no no no yes no no County linear trends no no no no 1 Standard errors clustered at the county level in parentheses, regressions weight by population of women in each cell *** 1%, ** 5% * 10% significance level yes 0 8965 ui UJ 54 2.7.5 Endogeneitv of Provider Availability Ideally, I would like to interpret estimates of the association between distance and abortion rates as causal, and toward this goal I have included controls for potentially confounding factors like county fixed effects, state-by-year fixed effects and county linear trends. However, another threat to causal interpretation is that it is difficult to evaluate whether changes in provider availability are driven by abortion demand itself. As discussed above, state level regulations and harassment against providers may have contributed to reduce the supply of abortion services forcing some providers to leave the abortion market and restricting the entry to others and this may have determined a decline in abortion rates. On the other end, the supply of abortion services is, at least in part, determined by the patient's demand for abortions so that changes in the number of providers may reflect a reduction in the demand of abortion itself. This means that any observed correlation between abortion access and abortions does not allow one to distinguish between these two effects. This section presents an analysis to evaluate whether and to which extent our results are biased by endogeneity. Some suggestive evidence comes from the analysis of those counties which never had an abortion provider and for which the nearest provider is at more than 50 miles away.26 Provider availability for this sample is more plausibly exogenous, as location decisions of these distant providers are more likely to be driven by the abortion demand of other (far away) counties (see Kane and Staiger, 1996, for a similar approach). Estimates on this restricted sub-sample are presented in TABLE XI. They show that an increase in distance is still associated with lower abortion rates (a 4.5 percent to 6 percent decline relative to the mean), suggesting that the impact of availability on abortion rates can be more plausibly interpreted as causal. This sample includes 280 counties. Standard deviation of distance to provider is approximately 32 miles. 55 TABLE XI EFFECT OF DISTANCE ON ABORTION RATES-COUNTIES WITH DISTANT PROVIDERS mean abortion rate=9.4746 1A IB 2 3 Miles to nearest provider (100s) Effect of 1 S.D. in distance relative to mean abortion rate Log per capita income Percent of income from unemployment insurance Medicaid funding restricted 1.9885*** (0.5145)a -1.7036*** (0.5198) -1.5348** (0.6652) -1.3196** (0.5139) [-0.0671] [-0.0575] [-0.0518] [-0.0446] 0.8273 (1.5746) -0.5716* (0.3216) 0.491 (1.6464) -0.3724 (0.2681) 4.5958*** (1.7491) -0.6848** (0.3150) 0.6322 -3.7111 (1.1154) (2.4259) Medicaid laws enjoined 1.0568 -2.0833** (1.0113) (1.0322) Race (l=white) -2.7913*** -2.7918*** -2.8048*** (0.5624) (0.5627) (0.5649) Marital status (l=married) -11.2726*** -11.2720*** -11.2720*** (0.5411) (0.5419) (0.5445) Age 25-29 -5.6882*** -5.6902*** -5.6875*** (0.3138) (0.3142) (0.3160) Age 30-34 -9.3039*** -9.3096*** -9.3103*** (0.4102) (0.4109) (0.4135) Age 35-44 -12.0091*** -12.0145*** -12.0224*** (0.4920) (0.4936) (0.4959) County fixed effects yes yes Yes yes Year fixed effects yes yes No yes State by year fixed effects no no Yes no County linear trends no no No yes Observations 23040 23040 23040 23040 a Standard errors clustered at the county level; regressions weight by the population of women in each cell. *** 1%, ** 5%, * 10% significance level. 56 Next, to explicitly address the issue of endogeneity in distance, I use an instrumental variable approach. Appropriate instruments to identify the demand for abortions are hard to come by, as they need to be correlated with the supply of providers but be also plausibly exogenous. Previous instrumentation attempts (Blanck et al, 1996; Brown et Jewell 1996; Brown et al., 2001) based on the number of nonOb/Gyn physicians and/or the total number of hospitals, or doctors per capita are unconvincing as these variables may not be plausibly exogenous neither strong predictor of abortion providers location decisions. In the following, I implement an instrumentation strategy that overcomes the limitation of previous studies and is based on information on college enrollment. The intuition starts from the observation that a concentration of young women who are sexually active would generate demand for an abortion provider. Providers tend to locate where the demand is strongest and young women in universities have the highest demand for abortion. For example, Shelton et al. (1976) observe that abortion frequency in Georgia is especially high not only in urban areas but also in counties with branches of the university system. Thus, assuming that the demand for abortion of college-age women is uncorrelated with the abortion demand of older women, college enrollment could be a good instrument for the measure of provider availability in the abortion demand equation of older women. In other words, a large college enrollment in a given area represents a kind of population inflow that should attract abortion providers and is plausibly exogenous in the demand of older women. College enrollment does not need to affect all types of providers in the same way. Large providers, which perform abortions on high volumes and on a routine basis, are more likely to take advantage of economies of scale deriving from locating near large colleges and universities. The data in fact suggest that there is a strong positive correlation between availability of a large abortion provider and college enrolIment. Instead, availability of a small provider does not appeai to be conelated with college enrollment. Together with the results in TABLE X, this suggests that distance to a large provider is the more appropriate measure for access to abortion. 57 In this framework, distance to the nearest large abortion provider is instrumented by using distance to the nearest county with large college enrollment (defined as at least 4,000 students). 27 College enrollment by county and year has been constructed using data from the Higher Education General Information Survey (HEG1S) and Integrated Postsecondary Education Data System (IPEDS) series from the National Center of Education Statistics. As the model is exactly identified, the exogeneity of the instrument cannot be explicitly tested. However, as described above, I argue that it is plausibly exogenous: once counties observable covariates and county fixed effects are controlled for, distance to a large college should have an effect on abortion demand of older women only through its impact on the availability of abortion services. As almost all large urban counties have an abortion provider, college enrollment may not constitute a relevant piece of information in the "information set" of providers locating in densely populated areas. Instead, for counties with low population density, college enrollment should be an important determinant of availability of abortion services as in absence of large colleges these counties would be unlikely to have a provider. Thus, the robustness of estimates is tested by limiting the analysis to smaller, less dense counties. To identify these counties, I analyzed the distribution of providers by population density and draw a line where we seem to see a large drop-off in the probability that a large provider locates in a county given population density and test how sensitive the estimates are to the definition of a given threshold. The OLS and instrumental variable estimates of abortion demand of older women (age 25 and above) are reported in TABLE XII. 28 The OLS results confirm the estimates of previous section: only distance to a large provider significantly affects abortion rates and the magnitude of the coefficient is largely unaffected by the introduction of a measure of small-provider availability. Furthermore, the first stage regression (TABLE XIII) shows that there is a strong positive correlation between distance to the 27 1 initially tried using several measures characterizing distances to the nearest county enrollment of various sizes, but this clearly over-killed and thus focused on the single threshold that seemed the key. 28 This table includes ARA counties whose nearest large provider is always in ARA (as in section 2.7.4, i.e. 496 counties). Standard deviation of distance to large provider is approximately 35 miles. 58 nearest large abortion provider and nearest county with large college enrollment: as distance to a large college increases so does distance to a large provider. 29 A similar first stage regression was run using as dependent variable distance to the nearest provider (regardless of size); the coefficient indicates that in this case college enrollment is not a strong predictor of abortion availability (results not shown). Overall, this suggests that distance to a large abortion provider is the key access measure and the instrumental variable estimates are focused on this variable. The second stage instrumental variable estimates are still negative and statistically significant and close in magnitude to the OLS estimates - the relative effect is about 1.5-2.5 percentage points larger than OLS (estimates, not shown, are not robust to the inclusion of county linear trends. The inclusion of trends absorbs most of the variation in enrollment data so that there is insufficient independent variation to estimate the model by instrumental variables). Regressions in TABLES XIV-XV replicate the previous estimates using a sample of counties identified on the base of a threshold of population density (300 residents per square mile). 31 The OLS results show an even larger negative effect of distance to a large provider on abortion rates of older women. This is not surprising since these estimates are based on a sample which excludes most large urban counties and where the availability matters most. The first stage regression shows, as above, a strong positive correlation between distance to the nearest abortion provider and nearest county with large college enrollment. The second stage instrumental variables estimates are negative and statistically significant. Also, albeit the IV point estimates are somewhat larger with respect to OLS, the magnitude of the standard errors leads us to conclude that IV and OLS results are not statistically different. The first stage F- statistics on the excluded instrument is above the Staiger and Stock (1997) reference value to deem an instrument as non-weak. 30 Treating distance to a small provider as endogenous together with distance to a large provider would require an additional identifying instrument which is not available. Together with the evidence in text that only distance to a large provider affects abortion rates, this suggests that distance to a small provider can be plausibly excluded from the model. I did also run estimates leaving distance to small provider in the equation and treating it as exogenous. Results are very similar to those presented in the text. jl The sample used includes 464 counties (std. dev. of distance approximately 42 miles). 59 Overall, these results provide evidence that the instrumental variables procedure is valid and together with the results in Table XI suggest that a reduction in the availability of abortion services over the 1980s has contributed to the decline in abortion rates over the period considered. In particular, women who reside in counties with longer travel distances to the nearest provider (and in particular large provider) have lower probabilities of aborting their pregnancies. TABLE XII EFFECT OF DISTANCE ON ABORTION RATES-OLDER WOMEN mean abortion rats = 9 6759 OLS OLS IV IV Miles to nearest large provider (100s) -1 5379** (0 6543)a -1 5471** (0 6558) -1 3246** (0 5825) -1 9711** (1 0101) -I 6436*** (0 5929) -1 6091*** (0 6016) -1 4480"* (0 5649) -2 3438*" (1 0980) Effect of 1 S D lr distance to large provider relative to mean abt rate [-0 0536] [-0 0567] [-0 0485] [-0 0722] [-0 0602] [-0 0589] -0 0531] [-0 0847] -0 0548 (0 1707) -0 0955 (0 1770) 02117 (0 2348) 0 1839 (0 2399) - - -0 8325"* (0 4067) - - - -0 5280 (0 3654) - Log per capita income 2 0954 (1 7474) 2 1509 (1 7717) 2 2349 (1 7689) 2 2658 (1 7389) -0 8759 (1 3277) -1 0655 (1 3383) -0 9320 (1 3353) -0 5592 (1 2949) Percent of income from unemployment insurance -0 3737 (0 3091) -0 3881 (0 2963) -0 3941 (0 2962) -0 3876 (0 3088) 0 0130 (0 3061) 0 0348 (0 2943) 0 0364 (0 2953) -0 0080 (0 3067) -1 1692*"* (0 4228) -1 1652*** (0 4220) -1 1691"** (0 4182) -1 1746*** (0 4193) -0 5223 (0 4484) -0 5263 (0 4507) -0 5409 (0 4492) -0 5167 (0 4484) Race (l=white) -5 0637*** (0 6510) -5 0637*** (0 6510) -5 0641*** (0 6510) -5 0638*** (0 6510) -5 0612*** (0 6520) -5 0610*** (0 6519) -5 0612*** (0 6519) -5 0615*** (0 6519) Marital status (l=marned) •16 7324*** (1 0827) -16 7324*** (1 0827) -16 7324"** (10827) -16 7324*** (1 0827) -16 7323*** (1 0826) -16 7323*** (1 0826) -16 7323*** (1 0826) -16 7323*** (I 0826) Age 30-34 -6 4798*** (0 3809) -6 4797*** (0 3809) -6 4796*** (0 3809) -6 4799*** (0 3809) -6 4786*** (0 3810) -6 4787*** (0 3810) -6 4787*** (0 3810) -6 4787*** (0 3809) Age 35-44 -12 0216*** (0 7245) -12 0216*** (0 7245) -12 0216*** (0 7245) -12 0219*** (0 7244) -12 0224*** (0 7245) -12 0225*** (0 7245) -12 0225*** (0 7245) -12 0226"** (0 7245) Miles to nearest small provider (100s) Miles to nearest small provider (100s) •"(Nearest provider small) Medicaid funding restricted Medicaid laws enjoined Distance small provider jointly significant (p-valus) County fixed effects Year fixed effects State by year fixed effects County linear trends F-test identifying instrument Observations 1 - - 0 1224 - - - 0 1556 - yes yes no no - yes yes no no - yes yes no no - yes yes no no yes no yes no - yes no yes no - Yes No Yes No 12 39 yes no yes no - 20 28 35616 35616 35616 35616 35616 35616 35616 35616 Standard errors clustered at the county level in parentheses, regressions weight by the population of women in each cell *** 1%, ** 5%, * 10% significance level © 61 TABLE XIII FIRST STAGE REGRESSION Miles to nearest county with large college enrollment Log per capita income Percent of income from unemployment insurance 0.8505*** (0.2416) 0.7851*** (0.1743) 0.3326* (0.1815) -0.0245 (0.0198) 0.3373 (0.2233) -0.0175 (0.0222) Medicaid funding restricted 0.0001 (0.0299) Medicaid laws enjoined 0.0083 (0.0319) Marital status dummy Race dummy Age dummies County fixed effects Year fixed effects State by year fixed effects County linear trends F-test identifying instrument Observations yes yes yes yes yes no no yes yes yes yes no yes no 12.39 35616 20.28 35616 TABEL XIV EFFECT OF DISTANCE ON ABORTION RATES-OLDER WOMEN-COUNTIES WITH POPULATION DENSITY < 300 RESIDENTS PER SQ MILE mean abortion rate = 6 7911 OLS OLS IV IV Miles to nearest large provider(lOOs) -1 5483** (0 6533) -1 5634** (0 6585) -1 3492** (0 5864) -2 1896** (1 0314) -1 8178*** (0 6175) -1 8320*** (0 6203) -1 6558*** (0 5920) -2 7209** (1 1719) Effect of 1 S D in distance to large provider relative to mean abt rate [-0 0967] [-0 0977] [-0 0843] -0 1367] [-0 1136] [-0 1144] [-0 1034] [-0 1699] Miles to nearest small provider(lOOs) - -0 1090 (0 1433) -0 2021 (0 1636) - - -0 116 (0 1666) -0 1777 (0 1829) - Miles to nearest small provider(lOOs) *(Nearest provider small) - - -0 8489** (0 4079) - - - -0 6053* (0 3528) - Log per capita income 0 6245 (0 9707) 0 6809 (0 9770) 0 7499 (0 9655) 0 9185 (09114) -1 2078 (0 9648) -1 1936 (0 9524) -1 0485 (0 9136) -0 6827 (0 9264) Percent of income from unemployment in* urance -0 2263 (0 1574) -0 2499 (0 1545) -0 2732* (0 1569) -0 2525* (0 1531) 0 1509 (0 1782) 0 1393 (0 1768) 0 1296 (0 1774) 0 1025 (0 1783) Medicaid funding restricted 1 5648*** (0 3842) -1 5591*** (0 3849) -1 5713*** (0 3707) -1 5964*** (0 3635) - - - - Medicaid laws enjoined 1 3860*** (0 3676) -1 3901*** (0 3670) -14163*** (0 3589) -1 3853*** (0 3642) - - - - Race (1 =white) 3 9901*** (0 3895) -3 9901*** (0 3895) -3 9911*** (0 3895) -3 9905*** (0 3895) -3 9873*** (0 3898) -3 9874*** (0 3898) -3 9878*** (0 3898) -3 9878"** (0 3898) Marital status (l=marned) 11 3539*** (0 5401) -11 3539*** (0 5401) -11 3539*** (0 5401) -11 3539*** (0 5401) -11 3535*** (0 5402) -11 3535*** (0 5402) -11 3535*** (0 5402) -11 3535*** (0 5402) Age 30-34 4 7314*** (0 1955) -4 7315*** (0 1955) -4 7314*** (0 1955) -4 7320*** (0 1955) -4 7273**" (0 1956) -4 7273*** (0 1955) -4 7273*** (0 1955) -4 7276*** (0 1955) Age 35-44 8 5286*** (0 3258) -8 5287*** (0 3258) -8 5285*** (0 3258) -8 5295*** (0 3257) -8 5251*** (0 3259) -8 5251*** (0 3259) -8 5252*** (0 3259) -8 5257*** (0 3258) - - 0 1143 - - - 0 228 - yes yes no no _ yes yes no no _ yes yes no no yes yes no no yes no yes no _ yes no yes no _ yes no yes no _ yes no yes no 19 63 32640 32640 Distance small provider |ointly significant (p-value) County fixed effects Year fixed effects State by year fixed effects County linear trends F-test identifying instrument „ 12 43 Observations 32640 32640 32640 32640 32640 32640 Standard error* clustered at the county level in parentheses, regressions weight by the population of women in each cell *** 1%, ** 5%, * 10% significance level ON 63 TABLE XV FIRST STAGE REGRESSION COUNTIES WITH POPULATION. DENSITY < 300 RESIDENTS PER SQ. MILE 1 2 Miles to nearest county with 0.8609*** 0.8069*** large college enrollment (0.2442) (0.1821) Log per capita income 0.3986* (0.2310) -0.0338 (0.0296) 0.4520* (0.2682) -0.0407 (0.0305) Medicaid funding restricted -0.0229 (0.0524) - Medicaid laws enjoined -0.0012 (0.0574) yes yes yes yes yes yes yes yes no no 12.43 32640 yes no yes no 19.63 32640 Percent of income from unemployment insurance Marital status dummy Race dummy Age dummies County fixed effects Year fixed effects State by year fixed effects County linear trends F-test identifying instruments Observations 64 2.8 Conclusions Since the early 1980s, the United States experienced a steady decline in the number and incidence of abortions, reaching today levels not seen since the early 1970s. To explain these patterns, a large part of the literature has investigated the effect of government restrictions as government policies continued to change subsequent to Roe and becoming less favorable to abortion. In contrast, relatively little attention has been given to explanations focused on supply side factors, despite the data showing trends in abortion rates and abortion providers being strongly interrelated. The issue of a woman access to abortion is important because starting in the early 1980s the market for abortion services has become increasingly concentrated and regulated often making it hard for providers to offer the service. This suggests that the constitutional right of a woman to terminate her pregnancy as established in Roe may be undermined to the extent that the lack or limited availability of services is not related to demand factors. This paper investigates whether the number and geographic distribution of abortion providers contributed to the decline in abortion rates over the 1980s. While some previous studies have examined the association between abortion availability and abortion rates, the presence of confounding factors and the simultaneity of demand and supply of abortions make it difficult to interpret results of previous research as causal. In this paper I made several contributions to the literature. First, I addressed the endogeneity of the number of abortion providers (supply) by means of an instrumental variable approach that uses changes in college enrollment to predict changes in availability of abortion services. Second, I analyzed the relationship between abortion rates and different measures of availability of abortion services using a panel of U.S. counties and a fixed effects approach that allowed me to obtain more robust estimates of the impact of changes in providers' availability on abortion demand. Furthermore, the impact of changes in abortion services is estimated on different groups of women that can be distinguished on the basis of demographic characteristics signaling different behavioral responses. Economic theory suggests that 65 abortion may act as a form of insurance against the risk of unwanted pregnancy creating ex-ante and expost moral hazard. In this paper I use the intuition that these forms of moral hazards (and thus the impact of an increase in abortion costs) may have different effects depending on the probability that the birth is wanted and thus depending on demographic characteristics that are associated with this probability. This is an aspect that has not been empirically investigated in previous research on the demand of abortion. The results show that reduced availability of abortion services is associated with a decline in abortion rates over the sample period considered and estimates are robust across a wide range of specifications. The effect varies across demographic groups, and there is some evidence that changes in distance have larger impact on unmarried women, a finding which is broadly consistent with predictions of economic theory. Finally, the results are robust to the attempt to address the possible endogenous location decision of abortion providers by an instrumental variables approach, suggesting that estimates of the correlation between abortion rates and reduced access to abortion services can be plausibly interpreted as causal. 3. LONG TERM EFFECTS OF ABORTION PARENTAL INVOLVEMENT LAWS 3.1 Introduction After nationwide abortion legalization in 1973, the main sources of change in abortion policy have been attempts, at the state level, to impose restrictions on abortion access. The court ruling itself, while legalizing abortion, left states some discretion to regulate the procedure. Among such restrictions, legal requirements for minors were established to either notify a parent(s) or obtain their consent before receiving an abortion. Together, these laws are referred to as Parental Involvement laws (PI laws), and constitute one of the few abortion restrictions left open to interpretation in Roe. Today, thirty-five states require some type of parental involvement in a minor's decision to have an abortion. 32 Another seven have laws on the books which are not enforced (Alan Guttmacher Institute, 2011). 33 Furthermore, states enforcing these statutes are geographically situated in a way that eliminates interstate travel as an option for many pregnant minors seeking an abortion (the state of Illinois is currently litigating enforcement of its parental notification law, which would remove the only remaining travel option for minors living in the Midwest). Over the last decades nearly every state has, at some point, enacted some form of abortion restriction for minors. These laws have been subject to great controversy in the courts and legislatures, and the number of states with PI laws in effect has more than tripled in the past twenty years. Supporters of these laws argue that excluding parents from a minor's abortion decision violates parental rights. They claim that the laws aid in family communication, reduce abortions and decrease births as young women become more cautious in terms of sexual activity and improve the use of contraception. j2 All enforcing states allow for a judicial bypass. Most also waive the requirement in case of medical emergency, abuse, assault, incest or neglect. •" Maryland and Connecticut have PI laws which are technically in effect, but minors in these states can de-facto access abortion services without involving a parent. Therefore, these states are treated as not having PI laws. 66 67 Opponents argue that young women would be deterred from having an abortion and forced to carry the pregnancy to term, delay the procedure (thus increasing risks and expenses) or travel to another state to get an abortion. Thus, given the inherent political, social and ethical dimensions related to legal abortion access by minors, empirical evidence (particularly with respect to births) is necessary to asses the validity of competing claims and to inform policymakers in the evaluation of such laws. Previous economic research, reviewed in detail later, has focused on the (current) impact of these laws on abortions and (to a less extent) births. Parental involvement increases the full cost of abortion for young women. Given a downward sloping demand curve, the short run effect would be a reduction in the number of pregnancy terminations, but minors may also travel to another state without law (or with a less restrictive law) to terminate their pregnancy. For this reason, the use of abortion data by place of residence would be especially important in this type of research. Because this information is generally not available, most studies used data by place of occurrence. In this case, however, the effect of the laws is likely to be overstated as changes in abortions may reflect both a true decrease in availability and in abortion location decisions. Most of previous research is especially vulnerable to this type of limitation because it analyzed laws enacted in the 1980s and early 1990s, when there was greater opportunity to travel-out-of-state, as few states had passed laws. In addition, the choice of appropriate comparison groups is crucial for the evaluation and interpretation of the results. Several studies used older teens and/or older women as comparison group without taking into account that both level and trends in abortions and births may greatly differ across those groups. Despite the described limitations, the results of the studies using more credible research designs suggest that PI laws reduce abortions among minors (for example Joyce et al., 2006; Levine, 2003). Whether this, in turn, translates into a rise in births depends on underlying assumptions about minors' behavior. Minors could respond to the laws after becoming pregnant only, but ihey could also change their ex-ante sexual and/or contraceptive behavior, thereby reducing pregnancy rates (Section 3.2). Thus, 68 the potential impact of the laws on births is ambiguous, and this is a key issue dividing opponents and proponents. However, there are few studies of, and no consensus on, whether PI laws affect births. Overall, the reading of the existing literature suggests that it is somewhat difficult to evaluate competing claims about these laws based only on short-run studies of abortions and births. In particular, with respect to births, an important point to consider is that previous research may have had limited statistical power to detect short-term. In fact, the group of minors actually affected by the laws is likely to be small because relatively few of them become pregnant each year, decide to get an abortion and not to involve their parents. However, the prevalence of the outcome under consideration (fertility) increases as women get older, and thus there may be more power to detect the impact of the laws. This suggests that analyzing short-run effects may give only a partial understanding of the full impact of these statutes. The issue is also a critical one, from a public policy point of view. A full evaluation of these statutes should take into account not only the current (short-term) impact, but also the effect on minors' lives as they become adults. This is especially important in the case of PI laws because they target women during ages that are critical for subsequent fertility and human capital investment decisions. Thus, the purpose of this paper is to investigate this issue. In particular, the research design aims at exploiting the fact that states enacted PI laws at different times to identify cohorts that have been more or less exposed to abortion restrictions as minors. I use PI laws in a woman's state of birth to construct a measure of exposure to abortion restrictions between the ages 15 to 17. Changes in the timing of introduction of the laws provide a source of variation which allows me to identify the effect of parental involvement on adult outcomes. This chapter is organized as follows: section 3.2 describes the theoretical impact of early abortion restrictions on adult fertility; section 3.3 provides a general overview of the legal framework surrounding the laws. Section 3.4 reviews previous research. Sections 3.5-3.6 describe the research design and data. Section 3.7-3.8 presents the results and Section 3.9 concludes. 69 3.2 Theoretical Impact of Parental Involvement Laws on Adult Fertility Parental involvement could affect a woman's behavior at several points along the chain of decisions leading to a pregnancy resolution. For example, when a state enforces a PI law, a minor, who would have otherwise aborted an unwanted pregnancy, may instead give birth. Then, this short-run change may represent a change in the timing of births or a permanent change in fertility. In other words, unless this additional birth is compensated by lower subsequent births, we would observe a gap in fertility across women who have been more or less exposed to these types of restrictions during their middle-teen years. Furthermore, the timing of first birth (a teen birth) may affect subsequent education choices and other factors that influence fertility later in life. If a woman becomes pregnant as a teenager, she may drop out of school or not going to college. These decisions are likely to have strong consequences on current and future human capital investments and subsequent fertility, for example through lower opportunity cost of time (Klepinger et al., 1999, is a good review of research relating adolescent childbearing to women's education, experience and labor-market outcomes. Most of the reviewed studies find that early fertility has a negative effect on each these outcomes). 34 Long-run effects could appear even in the case parental involvement affects pregnancy behavior. According to some authors these laws, by increasing the costs of terminating a pregnancy, would induce minors to be more cautious in terms of sexual behavior and/or contraceptive use, thereby reducing pregnancy rates (Kane and Staiger, 1996; Levine, 2003). If these changes are to be perpetuated over time as women age, they may end up on a different lifetime trajectory of fertility control effort, which impacts their fertility throughout their adult life. For this channel to work, we would need to assume that teenagers are aware of the parental involvement requirements and are forward-looking to incorporate this information in their decision making process. The authors also developed a life-cycle model of adolescent decisions regarding fertility and human capital investments, which recognized that these decisions are endogenous. The model suggested an identification strategy: variables affecting the cots of fertility control affect human capital investments only through fertility. The results confirmed the main findings of previous research. Early childbearing had negative consequences on education and work experience, and (for whites) early adult work experience. In turn, early childbearing (through lower human capital accumulation) significantly reduced wages at age 25. 70 These different hypotheses are summarized in Figure 13. The blue line represents the fertility pattern over a woman's lifecycle without the laws. Now, consider the implementation of a parental law affecting women between ages 15 to 17. The law increases the full cost of abortion and, if pregnancy behavior does not change, this will increase their fertility at those ages. This is shown by an increase in fertility prior to age 18 (dotted line). Then, three possible patterns are shown. If the law affects only the timing of births, fertility would slowly return at its original level (purple line). In this case we would not observe any difference in the long-run number of children born to women more exposed to the law. However, the law could permanently push women on a higher fertility trajectory. In other words, there might be a substitution of unwanted births for abortions that translates into a permanent difference in the number of children born (black bold line). Furthermore, there may also be a multiplier effect. As mentioned above, the timing of a first birth can affect other choices, which in turn affect fertility later in life. In this case the gap in the number of children ever born to women with different exposure to the laws may even be larger than the one implied by a "mechanical" substitution of unwanted births for abortions (black line). Finally, suppose, with time, the requirements of the law get well established so that more minors become aware of it. If they incorporate this information in their ex-ante decision making process, they may change their sexual and contraceptive behavior accordingly. In this case, there would be no effect (or reduction) in birth rates in the short-run, and there might be also an effect in the long run if the change in contraceptive behavior is strong enough to be perpetuated over time as women age. In the long-run, then, we could observe a reduction in the number of children ever born to women more exposed to the laws as minors. This hypothesis is summarized in Figure 13 by the maroon line. Overall, this discussion suggests that there may be several potential causal pathways of influence H u m p c t i v m a i i c i w o I U o u u u vv\^in\^ii 3 i w i L i n i y . m o p u i p v j o w u i u i o papv^i i s LKJ i i i v o ^ L i g a i t w n t i i i t i what extent PI laws affect women in the long run. a i m LU Fertility 14 18 23 30 Age Figure 13. Impact of Parental Involvement laws on Fertility 3.3 Brief History of Abortion Parental Involvement Laws Throughout the first half of the 20th century, abortion was an illegal procedure in the U.S. allowed only to preserve a woman's life. In 1962, the American Law Institute promulgated the new Model Penal Code (MPC), which liberalized abortion under limited circumstances including rape, statutory rape, incest and severe physical-mental defects of fetus or mother. In the late 1960s, 13 states began to allow abortions under the MPC provisions starting with Colorado in 1967 (Merz et al. 1995; 1996).35 Eight of these states expressly included in their statutes some type of parental involvement before a minor could obtain an abortion, otherwise physicians would have faced criminal penalties.36 The U.S. Supreme Court decision in Roe v. Wade (410 U.S. 113 January, 1973) legalized abortion nationally, but because Roe did not involve a minor, the question of parental involvement in i t The others were California and North Carolina (1967), Maryland (1968), Arkansas, Delaware, Georgia, New Mexico and Oregon (1969), South Carolina, Kansas and Virginia (1970) and Florida (1972). These states were Colorado, North Carolina, Arkansas, Delaware (not enforced), New Mexico, Oregon, South Carolina, Virginia and Florida. 72 minors' abortion decisions remained unanswered. The Court expressively left unaddressed the issue of the validity and constitutionality of parental involvement requirements in abortion statutes, which thereby remained one of the few abortion restrictions open to interpretation (Roe v. Wade, 410 U.S. at 165 n. 670). Since then, the history of abortion minors' restrictions is characterized by the passage of controversial laws, court battles, injunctions and Supreme Court decisions either invalidating or upholding states' provisions. Shortly after legalization, some states which previously adopted the MPC provisions, continued to enforce the parental involvement component of their statutes, for some period. Few other states with no pre-existing MPC abortion statutes enacted some form of parental involvement very shortly after Roe, as part of a comprehensive abortion legislation (Merz et al. 1995; 1996).37 At the same time, numerous legal attacks were launched upon these parental consent requirements, and a large volume of litigation arose from them, based on constitutional grounds. Only in the late 1970s the Supreme Court decided a line of cases that have defined more clearly the boundary of admissible parental involvement. The first major case directly involving a PI law was decided in Planned Parenthood of Central Missouri v. Danforth (July 1976). The case was about a Missouri law, which equired consent of the spouse for married individuals and parental consent for minors to perform an abortion. The court ruling held the law unconstitutional because it allowed an "absolute, and possibly arbitrary, veto" power on minors, thereby violating their privacy. Shortly after Danforth, most of the parental involvement requirements in effect before 1976 were either declared unenforceable because unconstitutional, or permanently enjoined. By the end of the decade, very few states had PI laws in effect. In the case of Bellotti v. Baird II (July 1979), the Supreme Court evaluated a Massachusetts parental consent law. Unlike Danforth, this law contained a judicial bypass mechanism allowing minors to waive parental consent by means of a court order. The possibility for the minor to circumvent parental involvement became a key element in subsequent court decisions. In fact, the Supreme Court has 37 Indiana, Idaho, Kentucky, Louisiana, Missouri, Montana, Nevada, North Dakota, Ohio, South Dakota (not enforced) and Utah. In most cases these laws were in addition to the common law rules which required parental consent for decisions regarding medical care for minors. 73 consistently upheld notification laws containing a bypass mechanism. Typically, to gain a judicial bypass, a pregnant minor must demonstrate either that she is mature enough to decide to have an abortion, or that the desired abortion would be in her best interest. Shortly after Bellotti, subsequent rulings further clarified the parameters of acceptable judicial bypass (HL. Planned versus Matheson, Parenthood Reproductive Health, 1980; City of Akron v. Akron center of Reproductive v. Ashcroft, 1983; Hodgson v. Minnesota, 1990). Finally, since Planned Parenthood Health, 1990; Ohio v. Akron of S.E. Pennsylvania 1983; center versus of Casey (1992), states are allowed to regulate abortion access as long as regulations do not impose an "undue burden" on women. The Court's adoption of the vague undue burden standard opened the door to further restrictions on minors' abortions, as PI laws were included among the policies that do not impose an undue burden (to the extent that judicial bypass is included). This series of court rulings shaped the pattern of introduction of explicit PI laws from the mid1980s through the mid-1990s (TABLE XVI and Figure 14).38 In the late 1970s-early 1980s few states enforced PI laws (they were nine in 1984), but that number increased to 15 by 1990 and to 26 by 1996. Today thirty-five states have PI laws in effect. More recently, states have been also amending existing statutes by placing further restrictions and/or requirements for minors to meet in order to access abortion services. For example, some have added notarization requirements, made the judicial bypass mechanism more difficult to access and made interstate travel illegal. This brief overview suggests that the legal environment defining abortion rights for minors was quite uncertain after legalization, during the 1970s and until the early 1980s, triggered on and off by a series of judicial decisions. This makes somewhat difficult to determine the actual ability of a minor under the age of eighteen to get an abortion. Starting with the court rulings from early 1980s ambiguity was progressively Ciiminatcu anu states tnat wanteu to restrict minors access to aoortion couiu uo so. r o r j8 As described in the text, it is important to note that several states tried to implement minors' restrictions before or right after legalization, but the status of these laws remained uncertain for some time. The specific timing of enactment of the laws in particular during the 1980s and 1990s was directly triggered by specific Supreme Court rulings that clarified the parameters of acceptable judicial bypass, and should therefore more plausibly unrelated to specific events going on in the states at that time. 74 this reason, the focus of the analysis is on explicit PI laws that were introduced mostly since the early-mid 1980s. This results in the sample restrictions employed in the empirical analysis. r~~ ON r- oo -H ON ON ON ON oo ON r- ^^ ON Os m OS ON ON ON >/-> (-- ON ON ON ON ON , — i o o tv^»nt i n r r p o c p i n K i f t h r a t p c a m r x n r r \ ; n i m f t o r Ax7/-*rv-i£*n i-£>lati\/f» +r\ tV»o 18-18.2 years old. Also the proportion of abortions performed past 12 weeks was higher in this subgroup. j9 The law applies only to minors younger than age 17 and a grandparent could satisfy the parental consent requirement. 81 This suggested that, in the case of Texas, the enactment of the law was associated with fewer abortions, induced some minors to continue their pregnancies and delay the procedure. Overall, the reviewed studies showed that the impact of the laws on abortions greatly varies according to the types of restrictions enforced, the appropriateness of the comparison groups and if the legal environment in the surrounding states is taken into account. The key point is that the effect of parental laws depends on access to "unrestricted" abortion services, which in turn is determined by how many neighboring states enforce similar statutes, and by the geographic distribution of abortion providers. For example, minors whose nearest abortion provider is in an unrestricted state are likely to be less affected by parental statutes than minors who do not have easy access to an out-of-state provider. Aggregate state analyses fail to take into account this source of heterogeneity. Also, as emphasized in the introduction, it is important to note that most of previous research is focused on laws enacted during the 1980s and early 1990s, when most states did not have these laws in effect. However, as more states enforced new statutes, the possibility for minors to travel to a less restrictive state dramatically reduced, making the results of early analyses not applicable to the current period. In the end, the more comprehensive and credible studies (Levine, 2003; Joyce et al., 2006; Colman et al., 2008; Colman and Joyce, 2009) do suggest that abortion rates decreased as result of enactment of the laws, and that the observed changes may not be entirely driven by interstate travel. From a policy point of view, the crucial point to understand, then, is whether and how parental laws affect births. In fact, competing claims about these statutes are based on the assumption that the laws either reduce or increase (unwanted) births among minors. Birth rates should rise if abortion rates fall, unless minors increase the use of contraception or decrease sexual activity in response to a parental notification , 40 or consent. Birth rates would not increase also in the case a large market of illegal abortion develops as consequence of abortion restrictions. However, this seems unlikely. No official numbers on illegal abortions are available, but deaths from abortion (stark indication of the prevalence of illegal abortion) declined dramatically after legalization and are now a rarity (AGI, 2003). 82 Relatively few of the reviewed studies analyzed the impact of parental laws on birth rates and generally lacked conclusive consensus. Some authors found no effect on (or a decrease in) birth rates, and concluded that this is due to the fact that PI laws persuade minors to abstain from sex or to use contraception more effectively. Among these, Kane and Staiger (1996) used NCHS data to construct county level birth rates between 1973 and 1988 for women ages 15-17, 18-19 and 20-29. They found that parental laws were associated with lower birth rates among minors, but also among the other age groups, suggesting that they might have not been able to control for all confounding factors (robustness of the results is not tested against the inclusion of trends in birth rates). Levine (2003) found that the laws have no significant impact on birth rates (but these actually increased and the coefficients were close to be statistically significant when state-specific trends were accounted for). Ellerston (1997), Rogers et. al (1991) also found no effect on birth rates. Some studies tried to directly gather evidence that minors change their sexual and/or contraceptive behavior in response to the PI laws. Levine (2003) used data from the National Survey of Family Growth (NSFG) to directly assess the impact of these statutes on the rates of sexual activity and contraception. The results showed that the laws were associated with greater use of contraception, albeit the very small sample size would cast some doubts about the credibility of the results. Klick and Stratmann (2008) used CDC state data from 1981 to 1988 to examine Gonorrhea rates by year, age and race among teens age 15-19 relative to older women age 20 and older. The results showed a reduction of 12 (21) percent for whites (Hispanics) and no effect on black females. The authors concluded that the laws induced women to avoid unprotected sex. However, this study has some weaknesses that cast doubts about the credibility of the results: teens ages 18-19 are included in the treatment group, women 20 and older are used as comparison, the results seem to be too large to be plausible and finally, the authors nrovide no explanation for wh.v no effects were found for male teenagers. Overall, the direct evidence that parental laws reduce sexual activity and increase the use of contraception among minors seems unconvincing. The results of studies showing a reduction or no effect 83 on birth rates are likely to suffer biases due to confounding factors, and in general declining trends in birth rates which are not appropriately accounted for. Furthermore, to justify the fact that birth rates do not increase after enactment of these laws, some strong assumptions have to be made regarding minors' behavior: that they are forward looking, aware of the requirements of the law and that these requirements are taken into account in their decisions regarding sexual activity and use of contraception. The credibility of these assumptions has to be carefully evaluated in light of the fact that teens tend to place excessive weight on the immediate benefits of sex while highly discounting the future costs. Perhaps the more convincing study showing that PI laws cause pregnant teenagers to choose birth over abortion is the Texas study (Joyce et al., 2006). In a sub-group analysis, using more credible treatment and comparison groups, birth rates are observed to rise by 4 percent among 17.5-17.75 yearolds at time of conception compared to 18-18.2 year-olds (10 percent among non-Hispanic whites). In a follow-up using Texas data, Colman et al. (2008) concluded that the lack of information on exact age at time of conception may have biased results of previous research of the effect of PI laws on birth rates toward a finding of no increase. As mentioned in the introduction, an important issue to consider in the evaluation of this literature is that short-run studies might have a limited statistical power to identify changes in births. Relatively few minors become pregnant each year, and not all of them will decide to have an abortion without telling their parents. It follows that relatively few minors are actually affected by the laws. Thus, even if the PI statutes determine a fall in abortions and an increase in births, the statistical power to detect such small changes is often lacking. In other words, the increase in births due to the laws might represent only a very small fraction of the births that would have been observed had the law not been enforced, an increase generally too low to detect given the sample sizes. It is also important to note that the two predicted effects on births (cither reduction/no change in births or rise in births) are not necessarily mutually exclusive. The initial enforcement of a PI law may cause an initial fall in abortion rates and rise in birth rates. However, over time enforcement of the law 84 will become clearer and well established, and minors may become more aware of the requirements and change their behavior accordingly. This overview suggests that looking at the short-run effects only may provide a partial understanding of the full impact of PI laws. Short run studies fail to report consistent results mostly because of intrinsic flaws in the research designs and limited statistical power to detect even small effects. However, it might still be possible to evaluate the overall impact of these laws in the long-run, as the prevalence of the outcome (fertility) increases when women get older (and thus there is more statistical power). Moreover, the potential long-run impact of the PI laws is important from a policy point of view because these affect women during years that are critical for many human capital investment choices (e.g. education). Changes during this period (e.g., fertility) may raise the costs and reduce the returns to time spent in human capital investments which, in turn, affect many outcomes later in life (Keplinger et al., 1999). This suggests that there are reasons to expect PI laws to have a long-run impact on women. In analyzing this issue, some insights come from previous research that has analyzed the long-term impact of changes in cost of fertility control. Some of these studies are reviewed below. 3.4.2 Previous Research on the Long Term Effects of Changes in Costs of Fertility Control Few papers analyzed the effect of access to the pill and broad abortion legalization in the early 1970s on women's outcomes when adults. Angrist and Evans (1996) estimated the impact of teen and out-of-wedlock childbearing on adult women's schooling and labor market outcomes, using variation in teen fertility associated with changing abortion laws in the late 1960s-early 1970s. As described in Chapter 1, some states began liberalizing their abortion laws before Roe. Thus, there was a short window where women who reached ages 15-19 during the late 1960s-early 1970s were exposed to more liberal abortion laws than others (states were classified into reform and non-reform states and state-of-birth was 85 used to impute laws).41 The authors focused on women born in the late 1940s and early 1950s, and observed in the 1980 and 1990 census micro-data. They first estimated reduced form models in which exposure to a liberalized abortion environment at ages 15-19 was linked to teen and out-of-wedlock childbearing, schooling and labor market outcomes. Data from the 1980 census showed that, for white women, greater exposure to abortion liberalization was associated with lower probability of teen childbearing and lower number of children born before age 20. The effects were small but marginally significant. Stronger negative effects were found for the probability of marrying before age 20. No evidence was found for schooling and labor market outcomes. Results for blacks were stronger than that for whites. Greater exposure was significantly associated with lower probability of teen childbearing, fewer children born before age 20, lower probability of teen out-of-wedlock childbearing, higher high school graduation rates and employment. Results obtained by pooling 1980 and 1990 samples showed similar results for schooling and labor market outcomes. 42 Given the results for blacks, the authors estimated also the impact of teen fertility (out-ofwedlock childbearing) on schooling and labor market outcome using exposure to liberalized abortion environment as instrument for black-teen fertility (out-of-wedlock childbearing). Instrumental variable estimates were marginally significant, and the authors concluded that reduced teen fertility (out-ofwedlock childbearing) brought about by abortion legalization caused an increase in high school graduation, college entrance and employment for black females. Goldin and Katz (2002) explored the relationship between the diffusion of birth control pill, age at first marriage and fraction of women entering professional occupations. The pill was approved in 1960 and diffused rapidly among married women. However, it did not diffuse among young unmarried women 41 Four of these states broadly legalized abortion in 1970 (repeal states). The others legalized it in specific circumstances in the late 1960s (Chapter 1). 42 With pooled data the authors could not estimate the impact of exposure on teen childbearing. They estimated the impact of exposure on number of children ever born. No effect was found for whites. For blacks, greater exposure to liberalized abortion reduced overall fertility, and the estimate were only close to being statistically significant. 86 until the late 1960s, after state level changes expanded the legal rights of individuals aged 18-21 by reducing the age of majority and extending mature minors decision. According to the authors, pill availability during college ages was a critical input to career decisions. Giving women a better control over their fertility, the diffusion of the pill lowered the cost of engaging in long-term career investments by reducing the penalty of abstinence and uncertainty of pregnancy. It also reduced the penalty of delaying marriage through a thickening of the marriage market for those pursuing long term career investments. Because the purpose was to investigate the impact of the pill on professional career choices, they focused on college graduate women. The analysis was based on 1970 to 1990 census data, and they examined how actual pill usage and pill access before age 21 (i.e. around the age of college entrance) was related to long-run career and marital status outcomes (between ages 30 to 49). They examined 20 age groups (ages 30-49) across the three census years, and covering the 1921-1960 birth cohorts of US-born college graduate women. Actual pill usage was proxied by the fraction of college-age women in a cohort talking the pill before age 21. Pill access was proxied by the fraction in a cohort born in a state with a nonrestrictive birth control access law at ages younger than 21. The results showed that greater pill usage at young ages was associated with greater representation in nontraditional professional occupations later in life. Both actual pill usage and pill access before age 21 had a particularly strong impact on the fraction of women employed as lawyers and doctors (professions requiring large, up-front, human capital investments). Pill access was significantly associated with an increase in the share never married and a reduction in the share divorced. The authors also controlled for whether abortion was legal at the time a cohort hit college entrance age (18 years) and the average abortion rate when each cohort was in college (between ages 18-21). The estimates showed a significant impact of abortion rate measures on the share of col!e°e women em^lo^ed in nontraditional professional occupations and on the share of wr*m^ri employed as lawyers and doctors. Legalization of abortion had a positive effect on the share of women never married and a negative (small) impact on the share currently divorced. 87 Bailey (2006) looked at the impact of the pill on the timing of first births and labor force participation. The pill should have reduced the cost of preventing unwanted fertility and improved the success in timing of child bearing. Following Goldin and Katz (2002), she used variation from state-level changes in the late 1960s and early 1970s that expanded legal rights of women ages 18 to 21. The data came from the 1964-2001 Current Population Surveys. For the analysis of age at first birth the attention was focused on women ages 36 to 44. For the analysis of labor force participation (hours worked in the reference week; weeks worked in the previous year) the sample included women between ages 18 and 44. The author calculated access to the pill before age 21 by year-of-birth and state (region) of current residence.43 The estimates suggested that access to the pill before age 21 reduced the likelihood of having a first birth before age 22 (by 14 to 18 percent), and increased the extent of 26-30 year-olds women's labor force participation (worked/looked for work the week prior to the survey) by approximately 8 percent. Ananat et. al (2007) investigated whether the short-run fertility impact of early abortion legalization documented in previous studies (Levine et al., 1999) translated into permanent (long-run) differences in fertility. In particular, the authors considered ages 16 to 26 to represent the peak ages at which the number of births is affected by abortion legalization (short-run effect). The largest impact on lifecycle fertility would then be observed for birth-cohorts entering ages 16-26 during the early 1970s (when legal status of abortion differed between repeal and non-repeal states). Using natality data from the NCHS and Census data, they constructed fertility histories for birth cohorts beginning in the 1930s and through the 1960s. They measured fertility at different ages, i.e. children ever born by age 25, 30, 35 and completed fertility. The results indicated that, by each of these ages, the number of children ever born to women in repeal states was significantly lower compared to women in non-repeal states, for cohorts that were between ages 16-26 at the time of early abortion legalization. In other words, early abortion legalization led to a permanent reduction in fertility which was 4j State of birth is not available in the CPS. Furthermore, in some years small states are aggregated into regions. 88 already evident at age 25 Because women might have traveled from non-repeal to repeal states, the former should be better control the greater the distance from the repeal states In some analyses, controls states were classified according to distance from repeal states and greater effects were found for the farther away set of control states The results showed that abortion legalization accounted for about 11 percent of the total decline in fertility across the different birth cohorts Overall, the findings of previous research suggest that improved fertility control does not simply lead to short-run changes in fertility, but it may permanently affect women's fertility and other socioeconomic outcomes in the long-run However, none of the previous studies has taken into account the long-run effect of exposure to PI laws Contrary to broader abortion legalization, these laws affect a well defined age group (teen minors), and may be especially important because they have an impact on women's decisions at young ages that are critical for human capital investments choices Thus, the purpose of this paper is to fill this gap in the literature and evaluate whether exposure to abortion restrictions when young has an effect on subsequent outcomes 3.5 Empirical Specification The purpose of the analysis is to obtain estimates of the association between women's outcome when adult and exposition to abortion restrictions as minors (between ages 15 to 17) that can be interpreted as causal To this purpose, the estimating equation is specified in the following way Ysat =a0+ PEXPSC + Ss + a + yt + 8C + ucas, (3 1) where s indicates state-of-birth, a indexes age, c year-of-birth cohort and / current (survey) year, Yast represents the outcome of women of age a observed in current year t (and thus born in year c since c = t — a) and born in state s I estimate the impact of exposure on fertility, educational attainment and labor market ontrnmes S is a set nf state-nf-hirth fixed effects /7)_are single vear of ace fixed effects v. are 89 current (survey) year fixed effects; 8C is a set of cohort dummies by five-year cohort groups; EXPCS measures exposure to restrictive abortion environment between ages 15 to 17 for women born in year c and state s. It is calculated as the fraction of time between ages 15 to 17 a given year-of-birth cohort born in state s is exposed to PI laws for abortion. This measure varies only by state-of-birth and year-of-birth. I focus on minors aged 15 to 17 since very few abortions occur among women younger than 15. Ideally I would like to know the state of residence of women at the time they were in their middle teens. However, no dataset provides this information. Instead, I use state-of-birth to impute abortion laws. It is, therefore, necessary to assume that changes in legal access to abortion in the observed state of birth were relevant to the individual's decisions during those ages. Since the majority of adolescents reside in the state where they are born, measures of the legal environment in the state-of-birth should be a reasonable proxy for the actual abortion laws affecting women between ages 15 to 17.45 Possible source of bias due to this classification are discussed later. I also include (linear and quadratic) state-of-birth specific trends by year-of-birth to allow for changing state-specific cohort characteristics, and measures of unemployment rate, labor force participation rate, per-capita income and the average number of abortion providers in state-of-birth (per 100,000 women of childbearing age) at the time a given cohort was between ages 15 to 17. These variables are introduced in an attempt to control for unmeasured confounding factors that vary at the state-of-birth and cohort level and that could confound the effect of exposure, thereby threatening the causal interpretation of the results. 3.6 Data The data used to estimate equation (3.1) come from the 2000 (1 percent sample) Census and the American Community Survey (2001-2008), which is a monthly survey of US households (Integrated 44 Some type of restriction is necessary because of the linear dependence between age, period and cohort effects. I experimented using also other cohort groupings (like ten years) and results are similar to those reported in the text. 45 Estimates from the 1990 and 2000 censuses show that between 77 and 80 percent of women aged 15 to 17 actually reside in the state where they were born in. 90 Public Use Microdata Series, IPUMS, Ruggles et al., 2010). The ACS sample design approximates the Census 2000 long form sample and it is intended to eventually replace it. The advantage of the ACS lies in its annual collection, that allows cohort behavior (defined by year-of-birth) to be tracked across ages. The primary outcome of interest is women's fertility. Ideally, I would need a measure of children ever born to women, but this information is not available in the 2000 Census and American Community Survey (since it is based on the Census long form questionnaire and this question was dropped starting in 2000). To circumvent this problem, I use the number of own in children in the household as an estimate of the number of children ever born (CEB). I take this to be a proxy for the number of CEB for relatively young women. However, the use of number of own children in the household in place of CEB is likely to bias estimates downward. Women with restricted early legal access to abortion might have had a child early in life. These women observed at, say age 40, will appear to have had fewer children, compared to women of the same age who were not exposed to the law. In fact, because the timing of birth was different, their child will be older and might already have left the house. Thus, greater exposure to abortion restrictions would be correlated to fewer own children in the household. But this would be just an arty-craft of measurement error. The focus on relatively younger women (below age 32) would mitigate this type of measurement error because at those ages their children will be more likely to still be in the household, regardless of whether women were exposed or not to the laws. By looking at the 1990 census (the last year in which the CEB variable is available), we can see that the number of own children in the household almost perfectly track the number of children ever born (CEB) for women about 32 years of age and younger (Figure 15). 91 Fertility 2 - 1.5 1 - yy ,/af .5 «*k^i> o - i 18 yy i 20 y y i 22 i 24 i 26 i 28 i i 30 32 i 34 i 36 i 38 i 40 i 42 i 44 Age Own children CEB Source: Author's calculations using 1990 Census of Population, 1% sample. Figure 15. Own children in the household and children ever born (CEB) per woman. A limitation of this approach is that we might be interested in looking at whether any gap emerges in fertility across women with different degree of exposure also at older ages and toward the end of childbearing age (completed fertility). Nevertheless, the analysis of fertility during early adulthood years is an interesting result per se, as this is a period of time that is crucial for women investments in human capital accumulation and labor market choices. Furthermore, previous research has found that abortion legalization determined long run changes in fertility which were already evident at age 25 and remained relatively constant over the rest of women's childbearing ages (Ananat et. al, 2007). Thus, the final sample includes adult women born between 1968 and 1987, and observed between ages 21 to 32.4b As described in the previous section, most PI laws were enacted between the late 1980s and late 1990s. Women born in 1968 were between ages 15 to 17 in the early-mid 80s when most states Foreign born women are excluded. 92 did not have parental involvement laws in effect. Women born in 1987 were in their middle teen years in the early 2000s when most states had parental involvement laws enforced. To the purpose of estimating equation (1), individual data are aggregated at the state-of-birth s, cohort c and age a (and thus current year) level. 47 In the baseline set of estimates exposure is measured as the fraction of time between ages 15 to 17 women born in cohort c and state s cannot abort without parental involvement. It is a measure that varies between zero and one, with greater values indicating greater exposure to a restrictive abortion environment. I also present results obtained using of a somewhat different measure of exposure. The construction of this measure starts from the observation that women exposed to the law in a given state may be able to travel to another state without laws to terminate their pregnancy. In fact, early research showed that traveling represents an important behavioral response by minors. Thus, I created an adjusted measure of exposure which takes into account the legal environment in surrounding states. For example, suppose a minor is exposed to PI laws in her own state. Consider then two scenarios. In the first, the surrounding states have no laws enforced. In the second, all bordering states also enforce parental requirements. The adjusted exposure would reflect a weaker form of restriction in the former case compared to latter.48 Albeit minors' abortions constitute a relatively small fraction of all abortions of women of childbearing age (about 7 percent in 1999) there are substantial differences across race, with black teens having abortion rates about four times larger compared to whites (Henshaw, 2008). Thus, all estimates are presented separately for non-Hispanic white and black women (Hispanics are excluded because they are 47 Given the large sample sizes, and since the key independent variable varies only at the state-of-birth and year-ofbirth level I run aggregate estimates for computational efficiency. The results are identical to those obtained using individual level data. 48 In more detail, I calculated the fraction of time between ages 15-17 each cohort can get an abortion without parental involvement in each of the bordering states (and took the average across them). Greater values of this variable indicate that a cohort has greater freedom to waive the parental requirement in own state by traveling to neighboring states. The baseline exposure measure is scaled down by the inverse of this variable, so that the exposure variable takes smaller values when a given cohort has some freedom to abort in bordering states. 93 mostly foreign born, and because of their greater likelihood of not living in the state between ages 15 and 17).49 Standard errors are clustered at the state-of-birth level to allow for non-independence within stateof-birth. 3.7 Results 3.7.1 Fertility Results in TABLES XVI1-XV1II report the estimates of the association between abortion restrictions when young and the number of own children per woman for whites and blacks, respectively. The format of the tables is as follows. For each sample (whites and blacks), several model specifications are estimated. Estimates in column one are from models that include state-of-birth fixed effects, year fixed effects and age fixed effects. Column two adds control for cohort fixed effects (grouped by 5 years). Column three adds state variables at the time the cohort was between ages 15 to 17. Columns four and five add state-specific linear and quadratic trends by year of birth, respectively. Finally, each model is estimated using both the baseline measure of exposure (top panel) and the adjusted measure to take into account the legal environment in surrounding states (middle panel). Using the baseline measure, the results for whites indicate that there is no significant association between exposure to restrictive PI laws between ages 15-17 and subsequent fertility. In fact, the estimates are both small and always insignificant. The use of the adjusted measure of exposure to take into account abortion options in surrounding states changes the estimates. The coefficients get larger in absolute value when trends are included, and are closer to be significant with quadratic trends (column five). Estimates in columns four and five suggest that moving from an environment of never being exposed to restrictive parental laws to always being exposed between ages 15 to 17 (both in own state and all surrounding states) is associated with an increase in whites' fertility of 0.024-0.034 children per woman (a 3 to 4.3 percent effect relative to the mean). Furthermore, the estimates exclude all cells with a sample size smaller than 25 observations. For blacks, this reduces the number of states available to 22. Results with the full sample of cells are reported in APPENDIX C. 94 Estimates for blacks tend to be larger. The coefficients in column one through five in the top panel indicate that always being exposed to restrictive abortion laws when young is associated with an increase in fertility later in life of 0.04 to 0.065 children per woman (approximately a 4 to 6 percent effect relative to the mean). The estimates obtained by using the adjusted measure indicate that facing a restrictive abortion environment when young (both in own state and in surrounding states) has an even larger impact, and the coefficients are significant when trends are included (7.2 to 9 percent effect relative to the mean, columns four and five). Overall, these estimates suggest that when (in particular black) women face a restrictive abortion environment between ages 15 to 17 they experience higher fertility as young adults. The estimates also underscore the importance of taking into account the abortion environment in neighboring states. In fact, associations are larger in states surrounded by other states that have PI laws in effect, as expected. 95 TABLE XVII EFFECT OF EXPOSURE ON NUMBER OF CHILDREN PER WOMAN-WHITES' (1) (2) (3) (4) (5) -0.0043 (0.0184)b [-0.0055]c -0.0082 (0.0181) [-0.0105] 0.0022 (0.0176) [0.0028] 0.0112 (0.0129) [0.0144] 0.0144 (0.0138) [0.0185] ADJU STED-EXPOSU RE -0.0130 (0.0238) [-0.0167] -0.0170 (0.0241) [-0.0218] -0.0040 (0.0238) [-0.0051] 0.0236 (0.0195) [0.0303] 0.0338 (0.0209) [0.0433] state of birth fixed effects age fixed effects year fixed effects yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes EXPOSURE yes yes yes yes cohort group fixed effects no state-cohort variables e no no yes yes yes linear trend no no no yes yes quadratic trend no no no no yes 0.78 0.78 0.78 0.78 0.78 mean of dependent variable 5019 5019 5019 5019 5019 observations a The dependent variable is the number of own children per woman in the cell defined by state-year-of-birth and age. b Standard errors are clustered at the state-of-birth level, regressions weight by the relevant cell population. c Coefficient divided by the mean in brackets. d Indicate dummies for the following year-of-birth cohort-groups: 1968-1972; 1973-1977; 1978-1982; 1983-1987. e Unemployment, labor force participation rate, per-capita income, abortion providers when cohort was ages 15-17. *** 1%, ** 5%, * 10% significance level. 96 TABLE XVIII EFFECT OF EXPOSURE ON NUMBER OF CHILDREN PER WOMAN-BLACKS : (2) (3) (4) (5) (1) 0.0419 0.0386 0.0397 0.0538* 0.0648* EXPOSURE (0.0294)b (0.0275) (0.0372) (0.0289) (0.0263) c [0.0384] [0.0354] [0.0364] [0.0494] [0.0594] ADJUSTED-EXPOSURE 0.0502 (0.0348) [0.0461] 0.0479 (0.0350) [0.0439] 0.0517 (0.0329) [0.0474] 0.0789*** (0.0277) [0.0724] 0.0978** (0.0377) [0.0897] yes yes state of birth fixed effects yes yes yes age fixed effects yes yes yes yes yes year fixed effects yes yes yes yes yes cohort group fixed effectsd no yes yes yes yes state-cohort variables e no no yes yes yes linear trend no no no yes yes quadratic trend no no no no yes mean of dependent variable 1.09 1.09 1.09 1.09 1.09 observations 1696 1696 1696 1696 1696 a The dependent variable is the number of own children per woman in the cell defined by state-year-of-birth and age. b Standard errors are clustered at the state-of-birth level; regressions weight by the relevant cell population. c Coefficient divided by the mean in brackets. d Indicate dummies for the following year-of-birth cohort-groups: 1968-1972; 1973-1977; 1978-1982; 1983-1987. e Unemployment, labor force participation rate, per-capita income, abortion providers when cohort was ages 15-17. *** 1%, ** 5%, * 10% significance level. 3.7.2 Interpretation of the Results The impact of the PI laws in the long run might seem relatively small given the numbers reported in the tables above. However, it is important to take into account that the estimated effects are relative to the average minor. In fact, not all women will be affected by the laws between ages 15 to 17. Here, I try to roughly asses the magnitude of the implied effect for those minors actually affected by the law. Consider, for example, the results for whites. The upper bound estimates in TABLE XVII (columns four and five) would suggest that always being exposed to parental laws when young increases the number of children per woman by 0.0236-0.0338 (a 3 to 4 percent effect, when the adjusted measure is used). But, if this is the estimated effect for all minors, what would be the effect for minors at risk? 97 Women in my sample are bom between 1968 and 1987, and they reached ages 15-17 mostly during the 1980s and 1990s. To identify the group of minors actually at risk we need to consider that, over this time period, about 4-6 percent of non-Hispanic white minors become pregnant each year. Thus, on average, about 15 percent of white females become pregnant between ages 15 to 17. Among these, about 50 percent have an abortion, but not all of them are at risk of not telling their parents. There are not widespread investigations of the extent to which minors involve parents in their abortion decisions. The limited available evidence (Henshaw and Kost, 1992) reports that approximately 60 percent of minors say their parents know about their pregnancy and desire to have an abortion. Calculations based on the data from this study also reveal some racial differences, with whites being more likely not to involve their parents in the decision to have an abortion with respect to non-whites (48 percent vs. 33 percent respectively). Based on these rough numbers, it follows that the proportion of whites at risk is approximately 0.15*0.50*0.48=0.036. Assume the law caused 1 of 2 minors at risk to have an abortion (prior evidence Joyce et al., 2006). Then, 1.8 percent will have a birth instead of an abortion. It follows that the 0.0236-0.0338 increase in fertility among whites documented in TABLE XVII (columns four and five) is driven by only 1.8 percent of minors. If we inflate these estimates, the resulting implied effect for minors actually affected by the law is 1.3-1.9 (i.e. 0.0236/0.018- 0.0338/0.018). In other words, every abortion averted is associated with about 1.3-1.9 additional births for white females between ages 21- 32. The upper bound estimates for blacks indicate that always being exposed to parental laws when young increases the number of children per woman by 0.0789-0.0978 (when adjusted measure of exposure is used; TABLE XVIII, columns four and five). Pregnancy rates for blacks ages 15-17 were between 12 and 16 percent over the 1980s and 1990s. Calculations similar to the ones reported above indicate that approximately 3.7 percent of black minors are affected, suggesting that the implied effect for U1 l,o . , , „ , , 1,* u „ 1 1 i c 5 0 uiaujv:* v v u u i u uw z - . i - z . . u . 50 About 45 percent become pregnant between ages 15 to 17. Half of them will have an abortion; 33 percent do not tell their parents. Assume the law causes 1 of 2 minors at risk to have an abortion. Then the fraction of minors affected is 0.45*0.5*0.33*0.5=0.037. Note that there are no studies assessing the impact of PI laws on minors by race except for Joyce et al. (2006). However the black sample size is very small to draw definitive conclusions. 98 It is important to keep in mind that purpose of this exercise is not to obtain precise numbers, but to evaluate the plausibility of the estimates. Overall, the calculations are very crude, but they seem reasonable given the magnitude of the standard errors. An implied effect greater than one would be consistent with the story that having an early birth may cause education to decrease and worsen labor market outcomes, making more children cheaper (Keplinger et al., 1999). In other words, as described in the introduction, women exposed to the laws when young will have higher fertility as adult not only because they "mechanically" substitute unwanted births for abortions, but also because the timing of their first birth actually affect other outcomes which, in turn, influence fertility later in life. Indeed, as shown in the next set of results, exposure to PI laws is also found to be associated with other socio-economic outcomes (in particular educational attainment). 3.7.3 Educational Attainment The results of the previous section suggest that if abortion PI laws have an impact on fertility of adult women, there may be impacts on other aspects of their lives as well. In particular, in this section I analyse the effect of exposure on the probability to complete high school and on the probability to have some college education. Women who have been exposed to a more restrictive abortion environment and ended up having more children when young may have also invested less in their education. Again, I investigate this aspect using exposure to a restrictive abortion environment between ages 15 to 17 as the key variable of interest. The outcome examined is the probability of completing high school (having some college) measured as the proportion of women in a given cell with completed high school (having at least some college education). 51 Consistently with the previous section, the sample includes women born between 1968 and 1987, and observed between ages 21 to 32. The results are presented separately for whites and blacks. 51 Both the 2000 census and ACS (until 2007) do not distinguish between regular high school diploma and GED. Thus, high school completition encompasses both measures. Tabulations using the 2008 data show that women with GED only constitute a relatively small fraction of women in my age range (about 3.5 percent for both whites and blacks). 99 The estimates for whites (TABLE XIX) show that exposure to a restrictive abortion environment when young has a small and insignificant effect on the probability of completing high school. The use of the adjusted measure of exposure affects the results. In particular, for whites estimates become statistically significant, albeit small in magnitude, and suggest that moving from never being exposed to restrictive parental laws to always being exposed between ages 15 to 17 (both in own state and all surrounding states) is associated with a lower probability of completing high school (a 0.3 to 1.2 percent effect relative to the mean). Results for blacks show that an increase in exposure to a restrictive abortion environment is associated to a significantly lower probability of completing high school (a 2.4 to 3 percent effect relative to the mean with the baseline measure of exposure, and a 3.5 to 4.4 percent effect with the adjusted measure). TABLE XIX EFFECT OF EXPOSURE ON COMPLETED HIGH SCHOOL-WHITES '' EXPOSURE ADJUSTED-EXPOSURE (1) (2) (3) (4) (5) -0.0035 (0.002 l) b -0.0019 (0.0021) -0.0042 [-0.0021] (0.0026) [-0.0045] -0.0045 (0.0041) [-0.0038]c -0.0019 (0.0021) [-0.0021] -0.0055** -0.0031 -0.0030 (0.0026) [-0.0059] (0.0027) [-0.0033] (0.0027) [-0.0033] -0.0080** (0.0038) [-0.0086] [-0.0048] -0.0108** (0.0049) [-0.0120] state of birth fixed effects yes yes yes yes yes age fixed effects yes yes yes yes yes year fixed effects yes yes yes yes yes cohort group fixed effectsd no yes yes yes yes state-cohort variablese no no yes yes yes linear trend no no no yes yes quadratic trend no no no no yes 0.93 0.93 0.93 mean of dependent variable 0.93 0.93 observations 5019 5019 5019 5019 5019 a Dependent variable is proportion of women in cell (defined by state-year-of-birth-age) with completed high school. b Standard errors are clustered at the state-of-birth level; regressions weight by the relevant cell population. c Coefficient divided by the mean in brackets. d Indicate dummies for the following year-of-birth cohort-groups: 1968-1972; 1973-1977; 1978-1982; 1983-1987. e Unemployment, labor force participation rate, per-capita income, abortion providers when cohort was ages 15-17. *** 1%, ** 5%, * 10% significance level. 100 TABLE XX EFFECT OF EXPOSURE ON COMPLETED HIGH SCHOOL-BLACKSa EXPOSURE ADJ-EXPOSURE (1) -0.0266** (0.0106)b [-0.0310]c (2) (3) (4) (5) -0.0246** (0.0099) [-0.0287] -0.0248** (0.0096) [-0.0289] -0.0248* (0.0123) [-0.0289] -0.0200* (0.0116) [-0.0233] -0.0380*** (0.0113) [-0.0443] -0.0359*** (0.0110) [-0.0412] -0.0349*** (0.0107) [-0.0407] -0.0365** (0.0159) [-0.0426] -0.0303* (0.0173) [-0.0354] yes yes yes state of birth fixed effects yes yes age fixed effects yes yes yes yes yes year fixed effects yes yes yes yes yes cohort group fixed effectsd no yes yes yes yes state-cohort variables e no no yes yes yes linear trend no no no yes yes quadratic trend no no no no yes 0.857 0.857 0.857 0.857 mean of dependent variable 0.857 1696 1696 1696 1696 observations 1696 a Dependent variable is proportion of women in cell (defined by state-year-of-birth-age) with completed high school. b Standard errors are clustered at the state-of-birth level; regressions are weighted by the relevant cell population. c Coefficient divided by the mean in brackets. d Indicate dummies for the following year-of-birth cohort-groups: 1968-1972; 1973-1977; 1978-1982; 1983-1987. e Unemployment, labor force participation rate, per-capita income, abortion providers when cohort was ages 15-17. *** 1%, ** 5%, * 10% significance level. The next two tables present the results of the effect of exposure on the probability of having at least some college education. Estimates for whites, TABLES XXI, show that always being exposed to PI laws when young is associated with lower probability of having at least some college. The magnitude of the effect is statistically significant when trends are included (a 1.3 to 1.9 percent effect relative to the mean when the baseline measure of exposure is used, and 2 to 3 percent effect when the adjusted measure of exposure is used-columns 4 and 5). Consistently with the estimates of high school completition, results for blacks (TABLE XXII) are larger in magnitude and significant when trends are accounted for, suggesting that an increase in exposure 101 is associated with a large reduction in the probability of completing some college (a 5 to 7 percent effect relative to the mean with the baseline measure of exposure, and a 8 to 11 percent effect when the adjusted measure is used-columns 4 and 5). EXPOSURE ADJUSTED-EXPOSURE TABLE XXI EFFECT OF EXPOSURE ON SOME COLLEGE - WHITES '' (1) (2) (3) (4) -0.0045 -0.0020 -0.0027 -0.0088* (0.0047) b (0.0046) (0.0046) (0.0047) c [-0.0064] [-0.0029] [-0.0039] [-0.0126] -0.0044 (0.0061) [-0.0063] -0.0011 (0.0059) [-0.0016] -0.0013 (0.0063) [-0.0019] -0.0141** (0.0069) [-0.0201] (5) -0.0131** (0.0064) [-0.0187] -0.0210** (0.0079) [-0.0300] state of birth fixed effects yes yes yes yes yes age fixed effects yes yes yes yes yes year fixed effects yes yes yes yes yes cohort group fixed effectsd no yes yes yes yes state-cohort variablese no no yes yes yes linear trend no no no yes yes quadratic trend no no no no yes 0.70 0.70 0.70 0.70 0.70 mean of dependent variable 5019 5019 5019 5019 5019 observations a The dependent variable is proportion of women in cell (defined by state-year-of-birth and age) with some college. b Standard errors are clustered at the state-of-birth level; regressions are weighted by the relevant cell population. c Coefficient divided by the mean in brackets. d Indicate dummies for the following year-of-birth cohort-groups: 1968-1972; 1973-1977; 1978-1982; 1983-1987. e Unemployment, labor force participation rate, per-capita income, abortion providers when cohort was ages 15-17. ***!%,** 5%, * 10% significance level. 102 EXPOSUREb TABLE XXII EFFECT OF EXPOSURE ON SOME COLLEGE - BLACKSa (1) (2) (3) (4) -0.0182 -0.0158 -0.0165 -0.0273** (0.0120) (0.0108) (0.0132)b (0.0118) c [-0.0496] [-0.0287] [-0.0300] [-0.0331] ADJ-EXPOSUREb -0.0278* (0.0142) [-0.0505] -0.0176 (0.0127) [-0.0320] -0.0262** (0.0120) [-0.0476] -0.0417** (0.0133) [-0.0758] (5) -0.0389** (0.0182) [-0.0707] -0.0593** (0.0235) [-0.1078] yes yes yes yes state of birth fixed effects yes age fixed effects yes yes yes yes yes year fixed effects yes yes yes yes yes cohort group fixed effectsd no yes yes yes yes linear trend no no no yes yes quadratic trend no no no no yes state-cohort variables e no no yes no no 0.55 0.55 0.55 0.55 mean of dependent variable 0.55 1696 1696 1696 1696 1696 observations "The dependent variable is proportion of women in cell (defined by state-year-of-birth and age) with some college. b Standard errors are clustered at the state-of-birth level; regressions are weighted by the relevant cell population. c Coefficient divided by the mean in brackets. d Indicate dummies for the following year-of-birth cohort-groups: 1968-1972; 1973-1977; 1978-1982; 1983-1987. e Unemployment, labor force participation rate, per-capita income, abortion providers when cohort was ages 15-17. *** 1%, ** 5%, * 10% significance level. 3.7.4 Labor Market Outcomes The next set of tables show the effect of exposure on the probability of working (measured as the proportion of women in a given cell reporting to have worked at least one week over the previous year), and on the probability of being currently employed (proportion of women in a given cell reporting being employed the week before the interview). The results for whites (TABLE XXIII) indicate that always being exposed to PI laws when young is associated with a lower probability of having worked in the previous year (a 0.5 to 0.7 percent effect relative to the mean with the baseline measure of exposure; and a significant or close to significant 0.9 to 1.4 percent effect relative to the mean when the adjusted measure is used). No significant effects are 103 found for blacks. However, for the latter group, exposure is associated with a lower probability of bein£ currently employed, a 3 to 7 percent effect relative to the mean (TABLE XXVI) . TABLE XXIIT EFFECT OF EXPOSURE ON WHETHER WORKED LAST YEAR (2) (3) (1) EXPOSURE ADJUSTED-EXPOSURE -0.0047 (0.0038)b [-0.0055]c -0.0048 (0.0039) [-0.0056] -0.0057 (0.0037) [-0.0067] -0.0074 -0.0076 (0.0048) [-0.0089] -0.0093* (0.0047) [-0.0087] (0.0049) [-0.0109] WHITES'' (4) -0.0039 (0.0031) [-0.0046] (5) -0.0061 (0.0043) [-0.0072] -0.0073 (0.0046) [-0.0086] -0.0122* (0.0062) [-0.0143] yes yes yes yes state of birth fixed effects yes age fixed effects yes yes yes yes yes year fixed effects yes yes yes yes yes d cohort group fixed effects no yes yes yes yes state-cohort variables e no no yes yes yes linear trend no no no yes yes quadratic trend no no no no yes mean of dependent variable 0.85 0.85 0.85 0.85 0.85 observations 5019 5019 5019 5019 5019 The dependent variable is proportion of women in cell (defined by state-year-of-birth-age) who worked last year. b Standard errors are clustered at the state-of-birth level; regressions are weighted by the relevant cell population. c Coefficient divided by the mean in brackets. d Indicate dummies for the following year-of-birth cohort-groups: 1968-1972; 1973-1977; 1978-1982; 1983-1987. e Unemployment, labor force participation rate, per-capita income, abortion providers when cohort was ages 15-17. *** 1%, ** 5%, * 10% significance level. 104 EXPOSURE TABLE XXIV EFFECT OF EXPOSURE ON WHETHER WORKED LAST YEAR (1) (2) (3) -0.0029 -0.0015 -0.0036 (0.0104) (0.0112) (0.0105)b [-0.0044] [-0.0035]c [-0.0018] BLACKS a (4) -0.0036 (0.0119) [-0.0044] (5) -0.0038 (0.0140) [-0.0046] -0.0004 (0.0128) [-0.0005] -0.0062 (0.0161) [-0.0076] -0.0037 (0.0208) [-0.0045] ADJ-EXPOSURE 0.0011 (0.0130) [0.0013] -0.0012 (0.0133) [-0.0015] state of birth fixed effects yes yes yes yes yes age fixed effects yes yes yes yes yes year fixed effects yes yes yes yes yes cohort group fixed effectsd no yes yes yes yes state-cohort variables e no no yes yes yes linear trend no no no yes yes quadratic trend no no no no yes mean of dependent variable 0.82 0.82 0.82 0.82 0.82 observations 1696 1696 1696 1696 1696 a The dependent variable is proportion of women in cell (defined by state-year-of-birth-age) who worked last year. Standard errors are clustered at the state-of-birth level; regressions weight by the relevant cell population. c Coefficient divided by the mean in brackets. d Indicate dummies for the following year-of-birth cohort-groups: 1968-1972; 1973-1977; 1978-1982; 1983-1987. e Unemployment, labor force participation rate, per-capita income, abortion providers when cohort was ages 15-17. *** 1%, ** 5%, * 10% significance level. 105 EXPOSURE ADJUSTED-EXPOSURE TABLE XXV EFFECT OF EXPOSURE ON EMPLOYED - WHITESa (2) (3) (4) (1) -0.0041 -0.0049 -0.0070 -0.0038 (0.0048)b (0.0048) (0.0045) (0.0033) [-0.0056]° [-0.0067] [-0.0096] [-0.0052] -0.0067 (0.0064) [-0.0092] -0.0077 (0.0064) [-0.0105] -0.0110* (0.0064) [-0.0151] -0.0068 (0.0047) [-0.0093] (5) -0.0051 (0.0052) [-0.0070] -0.0097 (0.0072) [-0.0133] yes yes yes yes yes state of birth fixed effects age fixed effects yes yes yes yes yes year fixed effects yes yes yes yes yes cohort group fixed effectsd no yes yes yes yes e state-cohort variables no no yes yes yes linear trend no no no yes yes quadratic trend no no no no yes mean of dependent variable 0.73 0.73 0.73 0.73 0.73 observations 5019 5019 5019 5019 5019 a The dependent variable is proportion of women in cell (defined by state-year-of-birth and age) currently employed. b Standard errors are clustered at the state-of-birth level, regressions weight by the relevant cell population. c Coefficient divided by the mean in brackets. d Indicate dummies for the following year-of-birth cohort-groups: 1968-1972; 1973-1977; 1978-1982; 1983-1987. e Unemployment, labor force participation rate, per-capita income, abortion providers when cohort was ages 15-17. *** 1%, ** 5%, * 10% significance level. 106 EXPOSURE TABLE XXVI EFFECT OF EXPOSURE ON EMPLOYED - BLACKS a 0) (2) 0) -0.0212* -0.0194* -0.0206* (0.0112)b (0.0107) (0.0112) [-0.0317]c [-0.0289] [-0.0307] ADJUSTED-EXPOSURE -0.0211 (0.0130) [-0.0315] -0.0191 (0.0127) [-0.0285] -0.0217 (0.0134) [-0.0324] (4) -0.0210 (0.0159) [-0.0313] (5) -0.0332** (0.0156) [-0.0495] -0.0357* (0.0193) [-0.0533] -0.0512** (0.0196) [-0.0764] state of birth fixed effects yes yes yes yes yes age fixed effects yes yes yes yes yes year fixed effects yes yes yes yes yes cohort group fixed effectsd no yes yes yes yes state-cohort variablese no no yes yes yes linear trend no no no yes yes quadratic trend no no no no yes mean of dependent variable 0.67 0.67 0.67 0.67 0.67 observations 1696 1696 1696 16% 1696 a The dependent variable is proportion of women in cell (defined by state-year-of-birth and age) currently employed. b Standard errors are clustered at the state-of-birth level, regressions weight by the relevant cell population. c Coefficient divided by the mean in brackets. d Indicate dummies for the following year-of-birth cohort-groups: 1968-1972; 1973-1977; 1978-1982; 1983-1987. e Unemployment, labor force participation rate, per-capita income, abortion providers when cohort was ages 15-17. *** 1%, ** 5%, * 10% significance level. 3.8 Robustness Checks A limitation of the present analysis is that the data have no information on state-of-residence between ages 15 and 17 that I can use to construct a measure of the legal abortion environment faced by women at those ages. Instead, I used the state-of-birth to assign the relevant laws. I argued that, while perfect information around ages 15 to 17 is not available, birth state may be a good proxy for the legal abortion environment during middle teen years. However, the potential misclassification of the legal Another advantage of using state-of-birth is that it is exogenous to any mobility decisions related to abortion availability. 107 abortion environment resulting from this lack of information is a concern. In this section, I try to address the extent of this problem. In particular, I exploit the fact that the data have information about both state-of-birth and current state-of-residence. While neither measure provides perfect information on place of residence around ages 15 to 17, I may be reasonably confident that the legal abortion environment has been assigned correctly for women currently residing in their state of birth. The data show that around 68 percent of women in my age range currently reside in the state they were born-in. For the remaining sample, the threat of potential misclassification is a concern. However, my own calculations also indicate that approximately 53 percent of women not residing in their state-of-birth currently reside in a state that would have had the same legal abortion environment. In other words, these women would have faced the same abortion restrictions between ages 15 to 17 regardless of whether state-of-birth or state of current residence are used as a proxy for their actual (unobserved) residence at those ages. This means that another 17 percent of the sample can be correctly assigned. Thus, I overall expect that about 85 percent of sample is assigned the correct legal abortion environment. 53 Nevertheless, I take the threat of potential misclassification seriously, and to reduce the amount of measurement error I re-estimate some specifications using a sub-set of states that had a relatively high proportion of current residents born in that state. In particular, I classify as low migration states those where more than 60 percent of current (non-foreign born) residents women in my age group were born in that state. I focus on this group of states because for them I can be more confident that a large fraction of the sample has been assigned the correct legal abortion environment. If measurement error is random, I would expect larger (absolute value) estimates because random measurement error biases estimates downward. However, heterogeneity driven by different states Data on migration is somewhat limited in the ACS survey. There is information on state-of-residence one year ago. About 95 percent of women in sample actually reside in the same state as the previous year. Performing calculations similar to those described in the test but using information of state-of-residence one year ago it follows that about 84 percent of women are correctly assigned. 108 groupings (low versus high migration states) may also play a role. Nevertheless, the estimates from the analysis using this sub-group of states will be less affected by measurement error. The new set of estimates is presented in the next tables. J report estimates for whites only, because for blacks the set of states does not change with the new restriction.54 The results show that for the subset of states the effect of exposure on the number of children (TABLE XXVII) is statistically significant (when quadratic trends are used) and somewhat larger in absolute value (about a 3 percent effect relative to the mean) compared to TABLE XVII (where the effect of exposure was always somewhat smaller and not significant). The same is true using the adjusted exposure measure. Estimates are significant when trends are added and larger than the corresponding values obtained using the full set of states. This would be consistent with measurement error slightly biasing estimates downward. The estimates for high school competition show no significant impact even on the restricted subset of states when the base measure of exposure is used. When the adjusted measure of exposure is used estimates become larger and statistically significant and are close in magnitude to the ones obtained using the full sample of states. Estimates for probability of having some college are significant or close to statistically significant when trends are included and about the same magnitude to the ones obtained using the full sample of states. The remaining results are qualitative and quantitatively similar to the ones obtained using the full sample of states. In particular, since the analysis excludes all cells with very few observations, the sub-sample of black states already excludes those identified as high migration state. For whites, the new restriction reduces the number of states to 30. 109 TABLE XXVII EFFECT OF EXPOSURE ON NUMBER OF CHILDREN PER WOMAN-WHITESLOW MIGRATION STATES a (1) -0.0094 (0.0229) b [-0.0122]c (2) (3) (4) (5) -0.0151 (0.0226) [-0.0196] 0.0061 (0.0224) [0.0078] 0.0202 (0.0148) [0.0263] 0.0228* (0.0137) [0.0296] ADJUSTED-EXPOSURE -0.0240 (0.0303) [-0.0312] -0.0305 (0.0304) [-.0396] 0.0025 (0.0313) [0.0033] 0.0417* (0.0208) [0.0542] 0.0499** (0.0198) [0.0648] state of birth fixed effects age fixed effects year fixed effects yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes cohort group fixed effectsd state-cohort variablese linear trend no no no yes no no yes yes no yes yes yes yes yes yes EXPOSURE no quadratic trend no no no yes mean of dependent variable 0.77 0.77 0.77 0.77 0.77 3236 3236 observations 3236 3236 3236 a The dependent variable is number of own children per woman in the cell defined by state-year-of-birth and age. b Standard errors are clustered at the state-of-birth level; regressions are weighted by the relevant cell population. c Coefficient divided by the mean in brackets. d Indicate dummies for the following year-of-birth cohort-groups: 1968-1972; 1973-1977; 1978-1982; 1983-1987. e Unemployment, labor force participation rate, per-capita income, abortion providers when cohort was ages 15-17. *** 1%, ** 5%, * 10% significance level. no EXPOSURE TABLE XXVI Fl EFFECT OF EXPOSURE ON COMPLETED HIGH SCHOOL-WHITES LOW MIGRATION STATES3 (4) (2) (3) (1) -0.0045* -0.0030 -0.0026 -0.0038 (0.0026)b (0.0025) (0.0026) (0.0031) c [-0.0048] [-0.0032] [-0.0041] [-0.0028] ADJUSTED-EXPOSURE -0.0076** (0.0031) [-0.0082] -0.0053 (0.0031) [-0.0057] -0.0051 (0.0033) [-0.0055] -0.0088* (0.0047) [-0.0095] (5) -0.0022 (0.0043) [-0.0024] -0.0087* (0.0050) [-0.0094] yes state of birth fixed effects yes yes yes yes yes age fixed effects yes yes yes yes yes year fixed effects yes yes yes yes cohort group fixed effectsd no yes yes yes yes state-cohort variables e no no yes yes yes linear trend no no no yes yes quadratic trend no no no no yes mean of dependent variable 0.93 0.93 0.93 0.93 0.93 observations 3236 3236 3236 3236 3236 "Dependent variable is proportion of women in cell (defined by state-year-of-birth-age) with completed high school. b Standard errors are clustered at the state-of-birth level, regressions are weighted by the relevant cell population. c Coefficient divided by the mean in brackets. d Indicate dummies for the following year-of-birth cohort-groups: 1968-1972; 1973-1977; 1978-1982; 1983-1987. e Unemployment, labor force participation rate, per-capita income, abortion providers when cohort was ages 15-17. *** 1%, ** 5%, * 10% significance level. Ill TABLE XXIX EFFECT OF EXPOSURE ON SOME COLLEGE-WHITES-LOW MIGRATION STATES a (!) (2) (3) (4) (5) -0.0074 -0.0050 -0.0072 -0.0091 -0.0140* EXPOSURE (0.0056) b (0.0053) (0.0049) (0.0059) (0.0074) c [-0.0103] [-0.0106] [-0.0071] [-0.0131] [-0.0200] ADJUSTED-EXPOSURE -0.0087 (0.0076) [-0.0124] -0.0016 (0.0071) [-0.0023] -0.0084 (0.0066) [-0.0120] -0.0142* (0.0085) [-0.0203] -0.0223** (0.0097) [-0.0319] state of birth fixed effects yes yes yes yes yes age fixed effects yes yes yes yes yes year fixed effects yes yes yes yes yes cohort group fixed effectsd no yes yes yes yes state-cohort variables e no no yes yes yes linear trend no no no yes yes quadratic trend no no no no yes 0.70 0.70 0.70 0.70 0.70 mean of dependent variable 3236 3236 observations 3236 3236 3236 a The dependent variable is proportion of women in cell (defined by state-year-of-birth-age) with some college. b Standard errors are clustered at the state-of-birth level; regressions are weighted by the relevant cell population. 0 Coefficient divided by the mean in brackets. d Indicate dummies for the following year-of-birth cohort-groups: 1968-1972; 1973-1977; 1978-1982; 1983-1987. e Unemployment, labor force participation rate, per-capita income, abortion providers when cohort was ages 15-17. *** 1%, ** 5%, * 10% significance level. 112 TABLE XXX EFFECT OF EXPOSURE ON WHETHER WORKED LAST YEAR-WHITES-LOW 0) (2) (3) EXPOSURE -0.0028 -0.0028 -0.0046 (0.0041 ) b (0.0042) (0.0044) [-0.0033]° [-0.0033] [-0.0054] ADJUSTED-EXPOSURE -0.0043 (0.0050) [-0.0051] -0.0045 (0.0053) [-0.0053] -0.0077 (0.0061) [-0.0091] MIGRATION STATES" (4) (5) -0.0042 -0.0036 (0.0031) (0.0049) [-0.0049] [-0.0042] -0.0081 (0.0050) [-0.0095] -0.0096 (0.0072) [-0.0113] state of birth fixed effects yes yes yes yes yes age fixed effects yes yes yes yes yes year fixed effects yes yes yes yes yes cohort group fixed effectsd no yes yes yes yes state-cohort variablese no no yes yes yes linear trend no no no yes yes quadratic trend no no no no yes mean of dependent variable 0.85 0.85 0.85 0.85 0.85 observations 3236 3236 3236 3236 3236 "The dependent variable is proportion of women in cell (defined by state-year-of-birth-age) who worked last year. b Standard errors are clustered at the state-of-birth level; regressions are weighted by the relevant cell population. 0 Coefficient divided by the mean in brackets. d Indicate dummies for the following year-of-birth cohort-groups: 1968-1972; 1973-1977; 1978-1982; 1983-1987. e Unemployment, labor force participation rate, per-capita income, abortion providers when cohort was ages 15-17. *** 1%, ** 5%, * 10% significance level. 113 TABLE XXXI EFFECT OF EXPOSURE ON EMPLOYED - WHITES-LOW MIGRATION STATES a (2) (3) (4) (5) (1) 0.0002 -0.0010 -0.0039 -0.0050 -0.0017 EXPOSURE (0.0050) (0.0046)b (0.0047) (0.0036) (0.0063) [0.0003]c [-0.0014] [-0.0053] [-0.0068] [-0.0023] ADJUSTED-EXPOSURE 0.0004 (0.0061) [0.0005] -0.0012 (0.0063) [-0.0016] -0.0053 (0.0072) [-0.0072] -0.0079 (0.0053) [-0.0107] -0.0045 (0.0087) [-0.0061] yes yes yes yes yes state of birth fixed effects age fixed effects yes yes yes yes yes year fixed effects yes yes yes yes yes cohort group fixed effects no yes yes yes yes state-cohort variablese no no yes yes yes linear trend no no no yes yes quadratic trend no no no no yes 0.74 0.74 0.74 0.74 0.74 mean of dependent variable 3236 3236 3236 3236 3236 observations a The dependent variable is proportion of women in cell (defined by state-year-of-birth and age) currently employed. b Standard errors are clustered at the state-of-birth level; regressions are weighted by the relevant cell population. c Coefficient divided by the mean in brackets. indicate dummies for the following year-of-birth cohort-groups: 1968-1972; 1973-1977; 1978-1982; 1983-1987. e Unemployment, labor force participation rate, per-capita income, abortion providers when cohort was ages 15-17. *** 1%, ** 5%, * 10% significance level. 3.9 Conclusions Abortion PI laws for minors constitute one of the few abortion restrictions left open to interpretation in Roe. After several years of battles in lower court rulings, the United States Supreme Court held that parental consent and notification do not infringe on a minor's constitutional right to terminate a pregnancy if judicial bypass options are available. As the constitutionality of the PI laws has been made clearer through several Supreme Court decisions over the years, states started to explicitly include them in then statutes. Cunently, the majOnty of States lequne nunOiS to involve then paients oi legal guardians (through either consent or notification) in their choice to obtain an abortion, and Illinois is litigating enforcement of its parental notification law - which would be the nation's 36th PI law in effect. 114 Given the controversial nature of these laws, it is important to understand whether and to what extent they affect minors in the U.S. Along these lines, PI laws have been object of study also in economic research. Economists have analyzed whether abortions and (to a less extent) births are affected by the implementation of these restrictions, but the evidence from previous research is not robust across papers. The problems of most previous studies are inherently linked to the lack of good data on minors' abortions, research design that fail to credibly take into account interstate travel, confounding factors or sample sizes that are too limited to detect even small effects. Overall, some of the studies using more credible research designs find that enactment of these laws actually reduces abortion rates. Then, the key issue dividing opponents and proponents is whether they also have an impact on births. In principle, PI laws could force minors to choose birth over abortion, but they could also induce them to practice contraception or a more responsible sexual behavior, thereby reducing pregnancy rates. However, the empirical evidence on these points is not strong, as there are few studies of, and no consensus on, whether PI laws affect births. Overall, this discussion suggests that studies focusing on the short-run effects only may give a partial understanding of the overall impact of this type of abortion restriction. To fully evaluate these laws and inform the public debate about appropriate abortion policies it is important to gather credible knowledge not only on their current impact, but also on their repercussions on women's lives as they become adult. Indeed, previous research on the long-run effects of changes in the cost of fertility control suggests that it might reasonable to investigate whether PI laws affect women later in life. Along these lines, in this paper I use PI laws in a woman's state-of-birth to construct a measure of exposure to abortion restrictions between the ages 15 to 17. In particular I analyze the impact of exposure on youngadult fertility, educational attainment and labor market outcomes. The estimates in this paper show that the legal abortion environment faced by women between ages 15 to 17 may have lasting effects, especially for black females. The effects are non-negligible and statistically significant in particular when I use a measure of exposure that takes into the abortion 115 environment in neighboring states. The latter might be interpreted as a more direct measure of the actual strength of restrictions faced by minors. In particular, the results show that females who grew up in a restrictive abortion environment are more likely to have more children as young adults. In preferred models, PI laws are associated with between a 2 to 4 percent (4 to 9 percent) increase in fertility of white (black) women ages 21 to 32. Furthermore, the laws are associated with lower educational attainment: a 0 to 1 percent (3 to 4 percent) decrease in probability of completing high school, and a 1 to 3 percent (5 to 11 percent) decrease in the probability of having some college for whites (blacks). Regarding labor market outcomes, in preferred models PI laws are associated with between a 0 to 1 percent (0 percent) decrease in probability of working last year for white (black) women ages 21 to 32, and with between 0 to 1 percent (3 to 5 percent) decrease in probability of being currently employed for white (black) women ages 21 to 32. Importantly, all estimates also show that associations are larger in states surrounded by other states that have PI laws. In light of these results, we might be tempted to conclude that PI laws have relatively modest long-run effects. However, we have to take into account that the estimates represent the average impact of the laws on all minors, including both those affected and unaffected by the laws. Rough calculations show that the implied impact on minors at risk of being affected by the law is actually larger: in this sub-group, being exposed to PI laws when young is associated with 1.3-1.9 (2.1-2.6) additional births among white (black) women. An implied effect greater than one would be consistent with the fact that an early birth may not only induce women to mechanically substitute unwanted births for abortions, but also affect human capital investments which, in turn, influence fertility later in life. Overall, the evidence presented here is not consistent with claims that PI laws decrease births, but shows that the laws actually increase women fertility and reduce human capital accumulation, suggesting uiai at icast some groups oi women may ue suustantiany anecteu uy minors auortion restrictions wen beyond adolescence. This constitutes important information from a public policy point of view to keep in mind in the evaluation of such laws. 4. CONCLUSIONS 4.1 Overview Abortion was legalized nationally in the Unites States through the Supreme Court decision in Roe v. Wade and today is one of the most common medical procedures undergone by women aged 15 to 44 in the United States. An estimated 1.2 million abortions were performed in 2005 and the abortion rate was 19.4 per 1,000 women aged 15 to 44. Accurate information about abortion incidence and services is necessary to monitor levels of unwanted pregnancy and women's ability to access abortion services, as well as inform research and policies affecting maternal and reproductive health. Since Roe, social scientists and public health researchers have explored the determinants of abortion and its effects on various socio-economic outcomes, and a great deal of debate and renovated interest in economic research into abortion has picked-up considerably in the past ten years. The economic approach to the study of abortion is important because it brings to the abortion debate a different perspective, which is based on economic reasoning and empirical analyses designed to better understand the behavioral implications that abortion policies generate. For example, from an economic point of view it is important to emphasizes and take into account the effect that policies may have on several decisions that are all interrelated and inherently social in nature: to use contraception, engage in sex, to terminate a pregnancy, to marry, to give birth, among others. Furthermore, economists bring to the empirical analysis a set of tools to evaluate the causal connections between the observed phenomena of interests. This is particularly important because the public debate about abortion is often not grounded on evidence of causal connections, which prove to be a key for the effective design of policies. In this context, this dissertation has been focused on the study of women's behavioral responses that are brought about by changes in the abortion environment. These behavioral responses represent an 116 117 important piece of information for researchers and policy makers in evaluating the consequences of abortion policies. 4.2 Summary of Contributions Economists have made several contributions to this field, but several issues still remain open or have not been addressed adequately in previous research. In particular, in this dissertation, 1 analyzed two themes: women behavioral responses to changes in the availability of abortion services and to changes in laws governing minors' abortion access, which are analyzed in chapters two and three, respectively. Chapter two is focused on the study of abortion availability as determinant of abortion. The motivation for this study comes from the observation that the abortion rate raised sharply after legalization reaching a peak in the early 1980s and then started to decline, until today. Reasons for these trends are unknown and actually there are no good explanations available. Notably, little attention has been given to the supply side of the abortion market. A potential explanation for the observed reduction in the prevalence of abortion is decreased access to abortion services, which has also been observed in the data at the same time that abortion rates were falling. A focus on access to abortion is becoming an increasingly important policy issue as the number and types of restrictions imposed on abortion providers have been rapidly growing, and as violence against providers also started to increase starting in the early 1980s. The present study attempted to determine whether provider availability has any significant impact on the demand for abortions. From an empirical point of view it is important to understand whether there is a causal connection beyond the observed correlation between abortion rates and abortion availability. Threats from casual interpretation come from unmeasured confounding factors affecting at the same time abortion rates and abortion providers, and simultaneity. Most of previous research has been limited in its ability to control for these confounding factors (like, for example, attitudes toward abortion, social and community norms regarding sexual activity and so one) due to flaws in the research designs which mostly exploit variation in abortion services across geographic areas. Furthermore, previous research usually 118 ignored the fact that abortion demand and supply are simultaneously determined. In fact, an observed correlation between the number of abortion providers and the number of abortions may be due to the fact that the demand for abortions is higher when abortion services are more easily available. But we need also to take into account that abortion providers may tend to locate where the demand is strongest. Observing a decrease in both the number of abortions and providers cannot tell us which explanation is true. In the first chapter 1 tried to overcome some of the discussed limitations in order to better assess whether and to what extent changes in availability of abortion services affect abortion rates. To this purpose, I rely on empirical strategies that allow me to control for the presence of unmeasured confounding factors and for the simultaneity problem. In particular, regarding the first point, I use a large panel of U.S. counties and a fixed effects design. To the best of my knowledge, this is the first study to look at the issue of availability of abortion services using a large panel of U.S. counties. This approach exploits variation in availability of abortion providers within counties over time to explain changes in abortion rates. The unit of observation (in this case the county) acts as its own control over time allowing to more credibly control for the presence of factors that greatly vary across geographic areas and may confound any estimated relationship between abortion availability and abortion rates. I use counties as unit of analysis because abortion availability greatly varies within states and counties are more homogenous with respect to states of metropolitan areas in terms of socio-demographic characteristics and medical resources. To deal with the potential endogeneity of any measure of supply of abortion services I use an instrumental variables approach. As long as demand and supply of abortion services are simultaneously determined, I need some variable that predicts provider availability but that is unrelated to the demand of abortion. I used information on college enrollment to construct my instrument. In particular, distance to a large concentration of college students is assumed to be a good predictor of abortion provider availability and is plausibly exogenous. The intuition is that young unmarried women have relatively high rates of abortion, and providers are likely to locate near concentrations of such persons. Older women living 119 closer to an area with a large concentration of young women (college enrollees) will face an increase the availability of abortion services, but the increased supply is not due to the demand for services of all women, but only young, college enrollees. Finally, another contribution of this research is that the data allows me to stratify the analysis according to women demographic characteristics which may signal different behavioral responses to changes in the availability of abortion providers. Standard economic models of fertility assume that fertility control is costliness and perfect and women have perfect information about the costs and benefits of an additional child at the time of pregnancy. However, women face substantial uncertainty regarding reliability of contraception and chance of pregnancy and uncertainty of wanting birth conditional on pregnancy. Some economic models thus emphasize the role of abortion as insurance against this uncertainty. These models give some insights about whether women may respond differently to changes in availability of services. The main intuition is that changes in abortion costs (and thus the ex- post and ex-ante moral hazard that may arise from them) have different effects depending on the probability that the birth is unwanted and thus depending on demographic characteristics that are associated with this probability. In the empirical analysis I used marital status as a proxy for the wantedness of pregnancy and tested whether changes in availability of providers differentially affect married and unmarried women. Overall, the results of the first chapter can be summarized as follows. Abortion availability is significantly correlated with abortion rates. The main measure of availability I use is distance from the woman county of residence to the nearest county with an abortion provider. As distance to the nearest provider increases abortion rates are observed to fall. The estimates are robust to a wide range of specifications and their sensitivity is tested using different types of fixed effects. I also find some evidence that an increase in distance has somewhat larger effects on unmarried women, a result that is broadly consistent with the theoretical framework described in the paper. Finally, the jesulis of my instrumentation strategy suggest that the instrument is valid and that estimates can be plausibly interpreted in a causal way. 120 The results obtained are important from a policy point of view because they would indicate that policies either expressly or indirectly limiting women's access to abortion decrease their use of the procedure. However, in evaluating these results, it is important to keep in mind that there are some limits to the data used that constrain my analysis. In particular, the results are based on data through 1988 and on a subset of states, which cannot be taken as fully representative for the entire U.S. population (especially non-white population). Furthermore, the political and policy landscape for abortion has changed substantially since 1988. The rise of antiabortion protests may have substantially increased the role of "climate" in determining both provider availability and the willingness of women to terminate their pregnancy. Finally, the landscape of availability of services itself has changed perhaps substantially following the introduction of medical abortion in early 2000s. A better defined measure of availability would therefore need to take into account this new dimension of services. Nevertheless, the estimates of this paper can be viewed as the first robust assessment of the importance of availability of abortion services as determinant of abortion. The result would suggest that as state policies have changed over the years making it more difficult for providers to offer their service so has been the women ability to terminate their pregnancies. The U.S. Supreme Court has generally struck down regulations that place an "undue burden" on women seeking an abortion but has usually applied this standard to one restriction at a time. The findings of independent effect of abortion availability in its broader dimension suggest that these standards should take into account a broader range of restrictions and factors affecting abortion services and use across women. In the second part of the dissertation I look at women behavioral responses related to a specific type of abortion restrictions: Parental Involvement laws for minors. According to these laws, minors are required to either notify a parent(s) or obtain their consent to terminate a pregnancy. The case of PI laws is particularly interesting because the choice to become a parent, to give a baby up for adoption, or to terminate a pregnancy represents a life-altering decision for a minor along several dimensions. Therefore, it is important to understand what the effect of such restrictions on minors is as states have been enforcing 121 these laws at an increasingly higher peace and also amending their original parental involvement statutes to include new requirements. As emphasized in chapter three, PI laws are one of the few restrictions left open to interpretation in Roe, and have been object of several debated court rulings in the past decades. Sound empirical evidence on the effects of these laws would therefore be important to correctly inform the debate about this intricate issue. However, previous research has been focused exclusively on the short run impact of these statutes and overall it does not allow drawing strong and definitive conclusions about their effects. Overall, the reading of previous literature suggests that women' responses to changes in these laws may vary widely according several dimensions. There may be heterogeneity in response to the laws along a geographic and/or demographic dimension, which the majority of previous research has ignored. This suggests that studies looking at the short rum impact of PI laws may give us only a partial understanding of their full impact. This motivates the analysis undertaken in chapter three. The analysis starts from the observation that, from a public policy point of view, it is important to know whether this type of restrictions have repercussions on women's lives as they become adult. This is especially true in the case of PI laws because the timing of a first birth may have a life-altering impact on minors. The research design exploits the fact that states enacted these statutes at different times to identify cohorts that have been more or less exposed to abortion restrictions while minors. I use PI laws in women's state of birth to construct a measure of exposure to abortion restrictions between the ages 15 to 17. In particular I analyze the impact of the legal abortion environment during middle teen years on young-adult fertility, educational attainment and labor market outcomes. The focus on young adult women (age 21 to 32) is dictated by restrictions on data available. The estimates suggest that the legal abortion environment when women are in their middle teen years may have lasting effects, in particular for black females. In particular, the results show that females who grew up in an environment that prohibited them to obtain an abortion without PI are more likely to have more children as young adults. Corresponding estimates for whites are smaller and statistically 122 significant only in some specifications (in particular when estimates are corrected for the potential measurement error in assigning laws based on state of birth). Abortion restrictions seem to have also an effect on educational attainment and labor market outcomes. Most importantly, rough calculations show that the implied impact on minors at risk of being affected by the law is actually larger: in this sub-group, being exposed to PI laws when young is associated with 1.3-1.9 (2.1-2.6) additional births among white (black) women. An implied effect greater than one would be consistent with the fact that an early birth may not only induce women to mechanically substitute unwanted births for abortions, but also affect human capital investments which, in turn, influence fertility later in life Again, in evaluating these results, it is important to keep in mind that there are some limits to the data used that constrain my analysis. In particular, the focus on women aged 21 to 32 may seem narrow or limited if we want to study the impact of these laws on lifecycle (completed) fertility. Nevertheless, even a different birth timing observed at these ages is an interesting outcome by itself since it might well affect other human capital investments choices which may have a life-long impact. Also, as mentioned in chapter three, the lack of availability of information on actual state of residence during ages 15 to 17 may be a source of concern. However, an advantage of my analysis is that I can exploit information available on both state of birth and state of residence to assess the extent of this misclassification bias and implement a strategy to better control for it. Overall, the results of this analysis would suggest that at least some groups of women may be substantially affected by minors' abortion restrictions later in life. Additional research would be needed to try to uncover the potential heterogeneity in the response to changes in abortion laws across groups which may be differentially affected by this type of restrictions. This would be important to appropriately inform the public debate about this issue and help to design appropriate abortion policies directed to minors. 4.3 Discussion This dissertation has investigated the behavioral response of women to changes in (broadly defined) access to abortion services. Two main types of access have been considered. In the first part of 123 the thesis 1 look at access in terms of physical availability of abortion providers. In the second part I look at a specific category of women (minors) and interpret access in terms of legal capacity of minors to get an abortion with or without parental involvement. Overall, the results of my study suggest that women are responsive to changes in the abortion environment that influence their private decisions through a change in the private costs they face. To some, this might seem a narrow way of looking at issues related to abortion access because the analysis has been performed from the woman's point of view as if she is the only relevant party, without addressing the broader (societal) impact of changes in abortion policy. In fact, there are reasons to believe that changes in policy may produce effects that go beyond the strict private costs and benefits to women themselves. For example, what would be the impact of abortion policies on the "balance of power" between men and women? Some economic models provide insights to look further into this issue. For example, Akerlof (1996) model suggests that more restrictive abortion policies would act as a credible threat for men, forcing them to be more engaged either financially or through marriage in case a pregnancy results. In turn, this would increase female bargaining power. Other models, however, would predict opposite results. Chiappori and Oreffice (2008) show that more efficient birth control technologies generally increase the "power," hence the welfare, of all women. They model a marriage market in which there are heterogeneous tastes for children, who reduce the mother's ability to earn income. A crucial ingredient is the legal and biological asymmetry between men and women regarding children: women do not need to be married to become mothers, while men cannot enjoy paternity without being in a couple. Therefore, those men for whom children are a public good are potentially willing to compensate their wife for childbearing. And this compensation is greater once women have better control over their fertility (through abortion for example). Thus, in this mode! less (more) restrictive abortion policies increase (decrease) female bargaining power. 124 Very little empirical research has been conducted on these issues (see, for example, Oreffice, 2007) but it would be an interesting aspect to study for a broader understanding of the effect of abortion policies. Further reasoning could go on and also encompass a broader range of externalities associated with abortion, but this is beyond the scope of this discussion. Also, in this dissertation I did not address other issues for which little is known but that have received considerable attention in the U.S. and abroad in the recent decade. For instance, the introduction of RU-486, or Mifepristone, continues to generate controversy among the health professionals and the public. From an economic point of view, it would be important to analyze what is the behavioral response of women to the availability of medical abortion, which has the potential to change the character of access to abortion services and thus of pregnancy termination in the United States. Moreover, as Mifepristone provides greater incentives to abort very early in pregnancy, it would be interesting to investigate whether its availability alters also women sexual and contraceptive choices. In turn, as other countries prepare to fully introduce the abortion pill in standard medical practices, additional evidence on this issue may be useful to policy makers for the evaluation of policies directly affecting women's choices in terms of pregnancy termination. To conclude, albeit several questions still remain unanswered, I believe this work has contributed to shed some light on a set of issues which where not addressed by previous research and that are important from a policy point of view. In this way, it contributed to a deeper understanding of the effect of some important changes that occurred in the U.S. abortion market in the past decades. Although the perspective of the analysis might be appear somewhat narrow given the complexity of an issue as intricate as abortion policy, the approach used offer a complementary and hopefully enriching way to think at it. CITED LITERATURE Akerlof, G., Yellen, J., and Katz, M.: An Analysis of out-of-wedlock childbearing in the United States. Quarterly Journal of Economics CXI : 277-318, 1996. Alan Guttmacher Institute: State Policies in Brief. An overview of abortion laws. New York: Guttmacher Institute, February 2011. Becker, G.S. A Treatise on the Family. Cambridge, MA: Harvard University Press, 1981. Ananat, E. O., Gruber, J., and Levine, P.B.: Abortion legalization and lifecycle fertility. Journal of Human Resources 42(2): 375-397, 2007 Angrist, D.J., and Evans, W.N.: Schooling and labor market consequences of the 1970 state abortion reform. NBER Working Paper Series 5406, Cambridge, 1996. Aspin, L.T., Bax, J., Bennett, K., Berman, D.R., Brown, R.W., Eisenstein, M. M., Hall, W.K, and Jewell, T.R.: An economic analysis of abortion: the effect of travel cost on teenagers. Social Science Journal 37(1): 113-124,2000. Bailey, M.J.: More power to the pill: the impact of contraceptive freedom on women's lifecycle labor supply. Quarterly Journal of Economics 121(1): 289-320. Beauchamp, A.: Regulation, imperfect competition and the U.S abortion market. Working Paper, Department of Economics, Boston College, 2010. Bellotti v. Baird, 443 U.S. 622,1979. Blanck, M.R., George, C.C., and London, R.A.: State abortion rates: the impact of policies, providers, politics, and economic environment. Journal of Health Economics 15:513-553, 1996. Brown, R.W., and Jewell, R.T.: The impact of provider availability on abortion demand. Contemporary Economic Policy 14: 95-106, 1996. Brown, R.W., Jewell, R.T., and Rous, J.J: Provider availability, race and abortion demand. Southern Economic Journal 67(3): 656-671, 2001. Cartoof, V.G. and Klerman, L.V.: Parental consent for abortion: impact of the Massachusetts law. American Journal of Public Health 76(4): 397-400, 1986. Chiappori, P.A. and Oreffice, S.: Birth control and female empowerment: an equilibrium analysis. Journal of Political Economy 116(1): 113-140,2008. City of Akron v. Akron Center for Reproductive Health, 462 U.S. 416, 1983. Colman S., Joyce, T. and Kaestner, R.: Misclassification bias and the estimated effect of parental involvement laws on adolescents' reproductive outcomes. American Journal of Public Health 98(10):1881-1885, 2008. 125 126 Colman, S. and Joyce, T.: Minors' behavioral responses to parental involvement laws: delaying abortion until age 18. Perspectives on Sexual and Reproductive Health 41(2): 119-126,2009. Dennis, A., Henshaw S.K., Joyce T.J., Finer, L. B., and Blanchard, K.: The impact of laws requiring parental involvement for abortion: a literature review. New York: Guttmacher Institute, 2009. Deyak, T.A., and Smith, K.V.: The economic value of statute reform: the case of liberalized abortion. The Journal of Political Economy 84(1): 83-100, 1976. Doe v. Bolton, 410 U.S. 179, 1973. Eisenstadt v. Baird, 405 U.S. 438, 1972. Ellerston, C : Mandatory parental involvement in minors' abortions: effects of the laws in Minnesota, Missouri, and Indiana. American Journal of Public Health 87(8): 1367-1374, 1997. Gohmann, S.F., and Ohsfeldt, R. L.: Effects of price and availability on abortion demand. Contemporary Policy Issues 11:42-55. 1993. Goldin, C, and Lawrence F. Katz, F.: The power of the pill: oral contraceptives and women's career and marriage decisions. Journal of Political Economy 110 (4): 730-770, 2002. Griswold v. Connecticut, 381 US 479, 1965. Gius, P.M: The impact of provider availability and legal restrictions on the demand for abortions by young women. Social Science Journal 44: 495-506, 2007. Haas-Wilson, D.: The economic impact of state restrictions on abortion. Parental consent and notification laws and Medicaid funding restrictions. Journal of Policy Analysis and Management 12:498-511., 1993. Haas-Wilson, D.: The impact of state abortion restrictions on minors' demand for abortions. Journal of Human Resources 31(1): 140-158, 1996. Harris v. McRae, 448 U.S. 297, 1980. Henshaw, K.S., Forrest, J.D., and Blaine, E.: Abortion services in the United States, 1981 and 1982. Family Planning Perspectives 16:119-127, 1984. Henshaw, K.S., and Kost, K.: Parental involvement in minors' abortion decisions. Family Planning Perspectives 24(5): 196-213, 1992. Henshaw, K.S.: Factors Hindering Access to Abortion Services. Family Planning Perspectives 27: 54-59, 1995. Henshaw, S.: Birth and abortion data, http://www.welfareacademy.org/pubs/dataneeds/dataneedshenshaw.pdf, 2000 Henshaw, K.S., and Finer, L.B.: The accessibility of abortion services in the United States, 2001. Perspectives on Sexual and Reproductive Health 35: 16-24,2003. 127 Henshaw, K.S., and Kost, K.: Trends in characteristics of women obtaining abortions, 1974 to 2004. New York: Guttmacher Institute, 2008. H. L. v. Matheson, 450 U.S. 398, 1981. Hodgson v. Minnesota, 497 U.S. 417, 1990. Jacobson, M., and Royer, H.: Aftershocks: The impact of clinic violence on abortion services. American Economic Journal: Applied forthcoming, 2011. Joyce, T., and Kaestner, R.: State reproductive policies and adolescent pregnancy resolution: the case of parental involvement laws. Journal of Health Economics 15: 579-607, 1996. Joyce, T., and Kaestner, R.: The impact of mandatory waiting periods and parental consent laws on the timing of abortion and state of occurrence among adolescents in Mississippi and South Carolina. Journal of Policy Analysis and Management 20(2): 263-282, 2001. Joyce, T., and Kaestner, R., and Colman, S.: Changes in abortions and births and the Texas parental notification law. New England Journal of Medicine 354(10): 1031-1038,2006. Kane, T.J., and Staiger, D.: Teen motherhood and abortion access. Quarterly Journal of Economics 111(2): 467-506, 1996. Klepinger, D., Lundberg, S., and Plotnick, R.: How does adolescent fertility affect human capital and wages of young women? Journal of Human Resources 24: 421-448, 1999. Klick. J. and Stratmann T.: Abortion access and risky sex among teens: parental involvement laws and sexually transmitted diseases. Journal of Law, Economics, and Organization 24(1):2-21, 2008 Levine, B.P., Staiger, D., Kane, J.K., and Zimmerman, D.J.: Roe v. Wade and American Fertility. American Journal of Public Health 89(2): 199-203, 1999. Levine, P.B., and Staiger, D.: Abortion as Insurance. National Bureau of Economic Research Working Paper No. 8813: 1-44,2002. Levine, P.B.: Parental involvement laws and fertility behavior. Journal of Health Economics 22: 861-878, 2003. Levine, P.B.: Sex and Consequences: Abortion, Public Policy and the Economics of Fertility. Princeton University Press, 2006. Matthews, S., Ribar, D., and Wilhelm, M.: The effects of economic conditions and access to reproductive health services on state abortion rates and birthrates. Family Planning Perspectives 29(2): 52-60, 1997. Merz, J.F., Jackson, C.A., and Klerman, J.A.: A review of abortion policy: legality, Medicaid funding, and parental involvement, 1967-1994. RAND-RP 565: 1-61, 1995. 128 Merz, J.F., Klerman, J.A., and Jackson, C.A.: A Chronicle of abortion legality, Medicaid funding and parental involvement laws, 1967-1994. Labour and Population Program Working Paper Series, 1996. Medoff, M.H.: The relationship between state abortion policies and abortion providers. Gender Issues 26: 224-237, 2009. Mocan, N.H.: Business cycle and fertility dynamics in the U.S.: a vector auto-regressive model. Journal of Population Economics 3: 125-146, 1990. Ohio v. Akron Center (88-805), 497 U.S. 502, 1990. Ohsfeldt, R.L., and Gohmann, S.F.: Do parental involvement laws reduce adolescence abortion rates? Contemporary Economic Policy 12(2): 65-76, 1994. Oreffice, S.: Did the legalization of abortion increase women's household bargaining power? Evidence from labor supply. Review of Economics of the Household 5: 181-207. Planned Parenthood v. Case, 505 U.S. 833, 1992. Planned Parenthood Assn. v. Ashcroft, 462 U.S. 476, 1983. Planned Parenthood of Central Missouri v. Danforth (482, U.S., 52). Roev.Wade, 410 U.S. 113, 1973. Rogers,J.L., Bonich,R.F., Stoms G.B., and DeMoya, D.: Impact of the Minnesota parental notification law on abortion and birth. American Journal of Public Health 81(3): 294-298, 1991. Ruggles, S., Trent, A.T., Genadek, K., Goeken, R., Schroeder, M.B. and Sobek, M.: Integrated Public Use Microdata Series: Version 5.0 [Machine-readable database]. Minneapolis: University of Minnesota, 2010. Shelton, J.D., Brann, E.A., and Schulz, K.F.: Abortion utilization: does travel distance matter? Family Planning Perspectives 8(6): 260-262. Staiger, D., and Stock,Y.: Instrumental variables regressions with weak instruments. Econometrica 65(3): 557-586. U.S. Dept. of Education, National Center for Education Statistics. Higher Education General Education Survey (HEGIS), various years: Fall Enrollment [Computer file]. ICPSR version. Washington, DC: U.S. Dept. of Education, National Center for Education Statistics. Ann Arbor, MI: ICPSR. Williams v. Zbaraz, 448 U.S. 358, 1980 APPENDICES APPENDIX A TABLE XXXII DATA SOURCES Abortion Data National Center of Health Statistics Abortion Provider Data Alan Guttmacher Institute County Level Economic Variables Bureau of Economic Analysis, Regional Economic Information System Population data National Cancer Institute AFDC caseload data Office of Family Assistant, Department of Health and Human Services Policy variables Blanck and London (1996) Distance Variable Center for Transportation Analysis, Oak Ridge National Laboratory Energy and Transportation Science Division College Enrollment data Higher Education General Information Survey (HEGIS) Series and Integrated Postsecondary Education Data System (IPEDS) Series, U.S. Dept. of Education, National Center for Education Statistics Unemployment data by state-year Bureau of Labor Statistics 129 APPENDIX B TABLE XXXIII EFFECT OF DISTANCE ON ABORTION RATES, ALL ARA COUNTIES 1A IB ID 2A 2B 3A 3B Miles to nearest provider (100s) •3 2428*** (0 9920)" -2 1634* (1 2970) -3 2480*** (1 1813) -1 8151 (1 4493) -3 0109*** (1 0864) -2 5144* (1 3864) -2 5980*** (0 6246) -1 8357 (1 4419) ARA(=\ if nearest provider in ARA) 1 2508*** (0 4625) 1 8454** (0 9127) 1 2188** (0 4854) 2 0064** (0 9104) 1 2447*** (0 4384) 1 5178* (0 8461) 1 0381* (0 5340) 1 4708 (12519) distance*/!/?^ -1 4664 (1 0592) -1 1055 (1 0282) Log per capita income -0 5071 (1 0021) -0 7604 (1 3885) 0 8812 (2 1877) 0 8913 (2 1865) -3 0967 (2 1580) -3 0787 (2 1582) 7 3634*** (2 1545) 73951*** (2 1548) Percent of income from unemployment insurance -0 7508** (0 3207) -0 7663** (0 3182) -0 0022 (0 4113) -0 0076 (0 4105) -0 2476 (0 2542) -0 257 (0 2537) Medicaid funding restricted -1 7516*** (0 6759) -1 7462** (0 6760) -2 5025*** (0 9151) -2 5016*** (0 9149) -0 621 (0 6522) -0 6218 (0 6522) -1 4915*** (0 4803) -1 4905*** (0 4802) Race (l=white) -4 8531*** (0 8524) -4 8531*** (0 8525) -4 8548*** (0 8528) -4 8548*** (0 8528) -4 8619*** (0 8575) -4 8619*** (0 8576) Marital status (l=marned) -198181*** (1 1262) -198181*** (1 1262) -19 8173*** (1 1268) -198173*** (1 1268) -198177*** (1 1321) -198177*** (1 1322) Age 25-29 -8 8339*** -8 8339*** -8 8324*** -8 8324*** -8 8354*** -8 8354*** (0 5080) (0 5080) (0 5078) (0 5078) (05101) (05101) -14 5793*** -14 5792*** -14 5768*** -14 5768*** -14 5757*** -14 5757*** (07811) (0 7811) (0 7814) (0 7814) (0 7854) (0 7854) -19 5896*** -19 5896*** -195871*** -195871*** -19 5832*** -19 5832*** (1 0543) (1 0543) (1 0556) (1 0556) (1 0602) (1 0602) Medicaid laws enjoined Age 30-34 Age 35-44 F-test distance variables jointly significant (p value) 0 0245 00141 0 0213 0 0002 APPENDIX B (continued) TABLE XXXIII (continued) EFFECT OF DISTANCE ON ABORTION RATES, ALL ARA COUNTIES 1C ID 2A County fixed effects Year fixed effects yes yes yes yes Yes Yes yes yes yes no State by year fixed effects County linear trends Mean abortion rate Observations no no no no yes 2B 3A 3B yes no yes yes Yes yes yes no no no no no no no no yes yes 15.6987 62528 15.6987 62528 15.6987 62528 15.6987 62528 15.6987 62528 15.6987 62528 15.6987 62528 15.6987 62528 132 APPENDIX B (continued) TABLE XXXIV EFFECT OF PROVIDER RATES ON ABORTION RATES BY COUNTY OF OCCURRENCE IB 1A 1C 2 3A In county provider rate Effect of 1 S D relative to mean abortion rate 4 4139*** (0 9161) [0 1465] Log per capita income Percent of income from unemployment insurance Medicaid funding restricted (Medicaid funding allowed) * (AFDC caseloads /fern pop) Medicaid laws enjoined - 3B 3 5901*** (0 8792) [0 1195] 3 6079*** (0 8776) [0 1201] 3 2255*** (0 8592) [0 1074] 1 8944** (0 8030) [0 0631] 1 8858** (0 8071) [0 0629] -3 3323 (7 1464) -0 9325 (1 0768) -2 9773 (7 1331) -1 0367 (1 0955) -8 2253 (6 5962) 0 8584 (0 9760) -1 2335 (4 2914) 0 0534 (0 8588) -1 2065 (4 3275) 0 1196 (0 8720) -1 5805 (2 1844) - - -2 7896 (2 5426) 0 0188 0 0473 -0 0453 (0 0372) 0 1166 0 2450 -1 5166* -1 7042* (2 1504) (2 5791) (0 8903) (1 0296) -0 2102 Race (l=white) -0 2100 -0 2099 -0 2284 -0 2283 (1 6985) (1 6985) (1 7005) (1 7079) (1 7079) Marital status (l=marned) -22 0227*** -22 0228*** -22 0222*** -22 0232*** -22 0232*** (2 8251) (2 8251) (2 8266) (2 8403) (2 8403) Age 25-29 -10 3308*** -10 3310*** -10 3220*** -10 3250*** -10 3249*** (1 4622) (1 4622) (1 4629) (1 4715) (1 4715) Age 30-34 -16 1493*** -16 1496*** -16 1437*** -16 1281*** -16 1273*** (2 1564) (2 1592) (2 1565) (2 1692) (2 1691) Age 35-44 -21 2930*** -21 2937*** -21 2874*** -21 2705*** -21 2704*** (2 8127) (2 8129) (2 8153) (2 8257) (2 8252) yes yes yes yes County fixed effects yes yes Year fixed effects yes yes yes no yes yes no no no State by year fixed effects yes no no no no no no County linear trends yes yes 174187 174187 174187 174187 174187 Mean abortion rate 174187 Observations 58560 58560 58560 58560 58560 58560 Standard errors clustered at the county level in parentheses, estimates weight by the numbei of women in each cell 1%, 5%, * 10% significance level APPENDIX B (continued) TABLE XXXV EFFECT OF DISTANCE ON ABORTION RATES-OLDER WOMEN-COUNTIES WITH POPULATION DENSITY < 100 RESIDENTS PER SQ MILE mean abortion rate=6 2521, obs 2744 Miles to nearest large provider (100s) Effect of 1 S D deA in distance OLS -1 4849** (0 6521) [-0 1161] Miles to nearest small provider (100s) Miles to nearest small provider (100s) *(Nearest provider small) IV -1 5022** (0 6544) [-0 1175] -1 2674** (0 5657) [-0 0992] -0 1826 (0 1807) -0 3295 (0 2147) -0 9498** (0 4508) -2 1709** (1 0317) [-0 1698] OLS -1 7372*** (0 6227) [-0 1359] IV -1 7612*** (0 6170) [-0 1377] -1 5704*** (0 5782) [-0 1228] -0 3317 (0 2334) -0 4309* (0 2585) -0 6905* (0 3895) -2 5394** (1 1264) [-0 1986] - - 0 8254 (1 0055) 1 1462 (0 9515) -1 2898 (1 0991) -1 2698 (1 0775) -1 1564 (1 0392) -0 9130 (1 0265) -0 1849 (0 1755) -0 2311 (0 1805) -0 1861 (0 1727) 0 0927 (0 2021) 0 0671 (0 2045) 0 0505 (0 2057) 0 0405 (0 1989) -1 3401* (0 6856) -1 3292* (0 6898) -1 3800** (0 6599) -1 4231** (0 6330) Medicaid laws enioined -1 2883** (0 6252) -1 2883** (0 6267) -1 3443** (0 6072) -1 3209** (0 6033) Race (l=white) -3 8984*** (0 3809) -3 8985*** (0 3809) -3 8998*** (0 3809) -3 8985*** (0 3809) -3 8978*** (0 3805) -3 8983*** (0 3805) -3 8984*** (0 3806) -3 8979*** (0 3805) Marital status (l=marned) 10 3642*** (0 5721) -10 3642*** (0 5721) -10 3643*** (0 5721) -10 3643*** (0 5721) -10 3642*** (0 5721) -10 3642*** (0 5721) -10 3642*** (0 5721) -10 3642*** (0 5721) Age 30-34 -4 2066*** (0 2163) -4 2068*** (0 2164) -4 2065*** (0 2163) -4 2075*** (0 2163) -4 2020*** (02161) -4 2020*** (02161) -4 2020*** (02161) -4 2019*** (0 2160) Age 35-44 -7 6967*** (0 3817) -7 6969*** (0 3818) -7 6966*** (0 3818) -7 6982*** (0 3816) -7 6915*** (0 3809) -7 6915*** (0 3809) -7 6916*** (0 3809) -7 6918*** (0 3808) - - Log per capita income 0 788 (1 0073) 0 853 (1 0185) Percent of income from unemployment insurance -0 1483 (0 1806) Medicaid funding lestncted - APPENDIX B (continued) TABLE XXXV (continued) EFFECT OF DISTANCE ON ABORTION RATES-OLDER WOMEN-COUNTIES WITH POPULATION DENSITY < 100 RESIDENTS PER SQ. MILE mean abortion rate=6.2521; obs.2744 OLS IV OLS IV Distance small provider jointly 0.1110 0.1645 significant (p-value) yes yes yes Yes yes yes yes yes County fixed effects yes yes yes yes No no no no Year fixed effects no no Yes yes yes no no yes State by year fixed effects no no no no no No no no County linear trends 12.11 19.49 F-test identifying instrument Standard errors clustered at the county level in parentheses; estimates weighted by the number of women in each cell. *** 1%, ** 5%, * 10% significance level. 135 APPENDIX B (continued) TABLE XXXVI FIRST STAGE REGRESSION COUNTIES WITH POPULATION DENSITY < 100 PER SQ. MILE Q) (2) Miles to nearest county with large college enrollment 0.9025*** (0.2592) 0.8828*** (0.1999) Log per capita income 0.5080** (0.2496) -0.0469 (0.0364) 0.3850 (0.2464) -0.0572 (0.0421) Percent of income from unemployment insurance Medicaid funding restricted -0.0545 (0.1045) Medicaid laws enjoined -0.0304 (0.1015) Marital status dummy Race dummy Age dummies County fixed effects Year fixed effects State by year fixed effects County linear trends F-test identifying instruments Observations yes yes yes yes yes no no 12.11 27744 yes yes yes yes no yes no 19.49 27744 136 APPENDIX C As describe in chapter three, the results in the main text have been performed excluding very small cells (those with less than 25 observations). Here, I test the robustness of some of the estimates (fertility) including all cells. The results for whites are virtually identical to those reported above. Estimates for blacks are somewhat noisier. This is not surprising since the greatest number of small cell is in the black sample. However, results are qualitatively similar to the main estimates reported in the text. TABLE XXXVII EFFECT OF EXPOSURE ON NUMBER OF CHILDREN PER WOMAN-WHITES-ALL CELLS ' (2) (3) (4) (5) (1) -0.0059 -0.0097 0.0013 0.0105 0.0138 EXPOSURE (0.0181)b (0.0175) (0.0137) (0.0178) (0.0128) [-0.0077]c [-0.0126] [0.0017] [0.0179] [0.0136] ADJUSTED-EXPOSURE -0.0152 (0.0233) [-0.0197] -0.0190 (0.0236) [-0.0247] -0.0052 (0.0236) [-0.0067] 0.0226 (0.0195) [0.0294] 0.0329 (0.0206) [0.0427] state of birth fixed effects age fixed effects year fixed effects yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes no yes cohort group fixed effects d yes yes yes state-cohort variables e no no yes yes yes linear trend no no no yes yes quadratic trend no no no no yes 0.77 0.77 0.77 0.77 0.77 mean of dependent variable 5508 5508 5508 5508 5508 observations a The dependent variable is the number of own children per woman in the cell defined by state-year-of-birth and age. b Standard errors are clustered at the state-of-birth level; regressions are weighted by the relevant cell population. c Coefficient divided by the mean in brackets. d Indicate dummies for the following year-of-birth cohort-groups: 1968-1972; 1973-1977; 1978-1982; 1983-1987. e Indicates the average unemployment rate, labor force participation rate, per capita income, number of abortion piovideis (pci 100,000 women of chiidbearing age) in state-of-birth when cohort was between ages 15 to 17. *** 1%, ** 5%, * 10% significance level. 137 APPENDIX C (continued) TABLE XXXV11I EFFECT OF EXPOSURE ON NUMBER OF CHILDREN PER (2) (1) 0.0337 0.0300 EXPOSURE (0.0284) b (0.0280) [0.0306]c [0.0273] ADJUSTED-EXPOSURE 0.0433 (0.0336) [0.0394] 0.0400 (0.0337) [0.0364] WOMAN-BLACKS-ALL CELLS a (5) (3) (4) 0.0347 0.0477* 0.0292 (0.0262) (0.0250) (0.0315) [0.0433] [0.0315] [0.0265] 0.0494 (0.0320) [0.0449] 0.0654** (0.0305) [0.0594] 0.0486 (0.0358) [0.0442] yes state of birth fixed effects yes yes yes yes age fixed effects yes yes yes yes yes year fixed effects yes yes yes yes yes cohort group fixed effectsd no yes yes yes yes state-cohort variables e no no yes yes yes linear trend no no no yes yes quadratic trend no no no no yes 1.1 mean of dependent variable 1.1 1.1 1.1 1.1 observations 4410 4410 4410 4410 4410 a The dependent variable is the number of own children per woman in the cell defined by state-year-of-birth and age. b Standard errors are clustered at the state-of-birth level; regressions are weighted by the relevant cell population. c Coefficient divided by the mean in brackets. indicate dummies for the following year-of-birth cohort-groups: 1968-1972; 1973-1977; 1978-1982; 1983-1987. e Indicates the average unemployment rate, labor force participation rate, per capita income, number of abortion providers (per 100,000 women of childbearing age) in state-of-birth when cohort was between ages 15 to 17. *** 1%, ** 5%, * 10% significance level. VITA NAME: Sara Borelli EDUCATION: Laurea (B.A.) Summa cum Laude, Economics and Business, University of Verona, Verona Italy, 2001 M.Sc, Economics, Collegio Carlo Alberto, Turin, Italy, 2002 D.E.A (Diplome d'Etudes Approfondies) Marche et Politiques Publiques, Cergy-Pontoise University, Cergy-Pontoise, France, 2003 M.A., Economics, University of Illinois at Chicago, Chicago, 2006 Ph.D. in Economics, University of Illinois at Chicago, Chicago, 2011 EXPERIENCE: Research Assistant to Professor Robert Kaestner through NBER, Fall 2010-Spring2011 Research Assistant to Professor Anthony Lo Sasso, School of Public Health, University of Illinois at Chicago, Chicago, 2009-2010 Research Assistant to Professor Anthony Lo Sasso and Lorens Helmchen, Institute of Government and Public Affairs-University of Illinois, Chicago, 2005-2007 Teaching Assistant, ECON 130, Department of Economics, University of Illinois at Chicago, Chicago, Spring 2005 Teaching Assistant, ECON 218, Department of Economics, University of Illinois at Chicago, Chicago, Fall 2004 HONORS/AWARDS: FMC Technologies, Inc., Fellowship Award, University of Illinois, 2007-2008 ; 2008-2009 Basset, Chiswick, Kosobud, Stokes Award for academic excellence, Univ. of Illinois at Chicago, Chicago, 2007 European Economics Association Fellowship, 2nd Lindau Meeting of the Nobel Prize Winners in Economics, Lindau-Germany , 08/2006 IZA, 8th IZA Summer School Fellowship, Munich-Germany (04/2005) Collegio Carlo Alberto graduate fellowship, Turin-Italy, 2001-2002 138 139 VITA(CONITNUED) Banca Popolare di Verona, Research Fellowship, Verona, Italy, 0408/2001 University of Verona scholarship to best undergraduate students for English course at the Language Institute University of Hull, UK, 08/2000 PRESENTATIONS: "Chewing the Flower of Paradise: Economic and Social Aspects of Qat Use in Djibouti" 20th International annual ESPE Conference, Verona, Italy (06/ 2006); Brown Bag Seminar Series University of Illinois at Chicago (04/2006); IZA Summer School in Labor Economics (poster presentation), Munich, Germany (05/ 2005) "Estimation of the Household Sharing Rule: the Case of Djibouti" Workshop on Equivalence Scales and Income Distribution, Accademia di Agricoltura Scienze e Lettere Verona, Italy (03/2001) SUMMER SCHOOLS: 1st Summer School on Inequality, Poverty and Development, Canazei, Italy (06/2006) 8th IZA Summer School in Labor Economics, Munich-Germany (05/2005) PUBLICATIONS: Borelli, S.: A Note on intergenerational mobility and intra-household balance of power. Bulletin of Economic Research,, doi 0.111 l/j.14678586.2009.00342.x, 2010 Borelli, S.: Social aspects of drug use in Djibouti: the case of the Leaf of Allah. Journal of African Economies. 18(4): 555-591, 2009 Furceri, D., and Borelli, S.: Foreign direct investments and exchange rate volatility in the EMU neighborhood countries. Journal of International and Global Economic Studies, 1(1): 42-59, 2008 Borelli, S. and Perali, F.: Drug consumption and intra-household distribution of the resources: the case of Qat in an African society. In: Household Behavior, Equivalence Scales, Welfare and Poverty, eds. Dagum, C. and Ferrari, G., pp. 163-194. Heidelberg, Physica-Verlag, 2004 PROFESSIONAL: MEMBERSHIP American Economic Association [...]... I consider the impact of changes in the availability of abortion services and of changes in restrictions to minors' access to abortion I concentrate on these two topics because they represent important aspects in the overall decision-making process about appropriate abortion policies In fact, over the past decades, the availability of abortion services has been facing increasing regulation and states. .. which, in turn, affect individual choices Economic analysis can reveal the importance of these policies and of economic constraints, and the evidence gathered can be used to assess the consequences and to design policies targeted at abortion The debate about abortion has fueled multiple lines of investigation in economic research: from the analysis of various types of restrictions and determinants of abortion, ... extremes, the contribution of economics is potentially important because it is focused on measuring consequences, which is central to the practical argument surrounding abortion laws and regulations But why is abortion an economic issue? From an economic point of view, abortion is an individual choice, which can be affected by constraints (including economic constraints) These constraints can be altered... that the current environment of abortion availability in the United States has been shaped by a long series of judicial and legislatives interventions that took place in the past decades Knowledge of these aspects will help in interpreting and put into context the results of the analysis of subsequent chapters The next section describes the purpose of the study and the contributions of this research, which... groups of women may be affected by minors' abortion restrictions well beyond adolescence This is an important aspect for policymakers to consider for the design of appropriate abortion policies directed to minors 2 2.1 EFFECT OF PROVIDER SUPPLY ON THE DEMAND OF ABORTION Introduction The incidence of abortion in the United States increased steadily after federal legislation that legalized abortion in 1973... groups of women? These issues can be examined in the framework of economic analysis of fertility The decision of terminating a pregnancy depends on the net benefit from 9 another child and abortion can be considered as a means of ex-post fertility control In this context, abortion decisions are function of cultural and religious values as well of all factors that affect the full cost of having children... Guttmacher Institute (AGI) the number of legal abortions (and abortion 4 Today five states have laws in effect restricting insurance coverage of abortion in all private insurance plans written in the state Twelve states restrict abortion coverage in insurance plans for public employees and other states have more than one of the above restrictions 5 The term abortion provider refers to any facility performing... decline after 1982, which is a period of increasing state regulations of abortions and increasing harassment of providers In particular, post abortion legalization, the number of hospital abortion providers declined, but the number of non-hospital providers increased sharply until the early 1980s, then remained relatively stable until mid 1990s, and then declined Also, the geographic distribution of abortion. .. few studies have investigated the effect of the regulatory environment on the supply of abortion services Medoff (2009) examined whether restrictive abortion laws have an impact on the number of providers in a state over the 1982-2005 period He found that Medicaid funding restrictions, PI Other states restrict the provision of abortion only to physicians licensed to practice medicine in the state or restrict... description of the judicial decisions and legislative environment related to abortion I then describe in more detail the purpose of this study and its contributions 1.1.1 Review of Abortion Legislation In the United States, at the time the Constitution was adopted, abortion before "quickening" (before the fetus first movements could be felt) was legally performed under common law Starting with New York in ... to consider for the design of appropriate abortion policies directed to minors 2 2.1 EFFECT OF PROVIDER SUPPLY ON THE DEMAND OF ABORTION Introduction The incidence of abortion in the United States. .. for the reporting of induced abortion - the U.S Standard Report of Induced Termination of Pregnancy These records include information on the date and place where abortion occurred, place of residence... explaining concentration of abortion provision in large clinics and the exit from the market of smaller clinics and hospitals Simulations show that the removal of all abortion providers' regulations