Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Welfare Reform and Abortion Jacob Alex Klerman The impetus for the current round of welfare reform derives from two complementary arguments. First, there is simply a concern that too many re- sources are being transferred from taxpayers to a dependent class, welfare recipi- ents.1 Second, there is a concern that the welfare system itself induces undesir- able behavior; in particular, the claim is that it induces women to have children when they cannot afford them and out of wedlock.2 Implicit in this second argument is that cutting welfare payments or radically restructuring the welfare system in the words of President Clinton, "ending welfare as we know it" will cut the number of children born to poor unmarried women. As we discuss in detail below, one way that this decline in the number of children could occur would be that women would not change their sexual behav- ior or their contraceptive behavior, but once they found themselves pregnant- and realizing that welfare would not support them and their child as it had previ- ously they would choose to abort the pregnancy. In net, fewer children would be born and fewer children would be on the welfare rolls. This possibility that welfare reform might induce more women to have abor- tions has led many in the pro-life community to oppose some welfare reform 1See Senator Phil Gramm's announcement speech in his unsuccessful bid for the Republican presidential nomination, College Station Texas, February 24, 1995: "I want to ask the able-bodied men and women riding in the wagon on welfare to get out of the wagon and help the rest of us pull. We've got to stop giving people more and more money to have more and more children on welfare." 20n the evidence for this claim, see Chapter 4; see also Moffitt (1992); and Jackson and Klerman (1996). 98
JACOB ALEX KLERMAN 99 proposals. For example, the Conference of Catholic Bishops issued a statement reading in part: Denying needed benefits for children born to mothers on welfare can hurt the children and pressure their mothers toward abortion and sterilization (Bishop John Richard, 1995:564~. The National Right to Life Committee argued the welfare reform-abortion link in more detail. The lead headline in its newsletter, National Right to Life News, for February 22, 1995 read "Welfare 'Reforms' Pose Threat to Unborn Babies." The article notes: National Right to Life has a long history of supporting alternatives to abor- tion since its formation in 1973. Consistent with this policy, National Right to Life is opposed to denying assistance to a newborn child who would otherwise qualify for assistance, based on the age or marital status of the mother or the fact that the mother is already receiving assistance for another child. NRLC is convinced that these proposals will result in the death of many additional chil- dren by abortion. NRLC President Wanda Franz, Ph.D., said "What is at stake is a signifi- cant portion of the very limited aid that is available for classes of children who are among the most vulnerable to abortion, children of unmarried teens and of the poor. If we were to stand silently by and not have to [sic, presumably it should be "the"] courage to speak out against these proposals, then we could never look anyone in the eye again and say we support alternatives to abor- tion" [emphasis in the original]. Proponents of the legislation seem to take two positions. Some grant that abortions will increase (if perhaps only in the short term) but argue that some increase in abortion is a worthwhile trade-off for a net decrease in children born onto welfare. For example, in congressional testimony, Charles Murray ex- plained the effects of ending Aid to Families with Dependent Children (AFDC) as follows: (T)he need to find support forces a self-selection process. ... It will lead others, watching what happens to their sisters, to take steps not to get pregnant. This is also good. Many others will get abortions. Whether this is good de- pends on what one thinks of abortion (testimony before the Subcommittee on Human Resources, House Committee on Ways and Means, July 29, 1994~. Or Marvin Olasky writing in the Wall Street Journal (Olasky, 1995~. Most pro-lifers understand that the current welfare system is fundamentally wrong-headed. But they are frozen because we cannot say for sure that welfare reform might not lead to a very sad, short-term abortion increase. My own tendency in such situations is to visit the cliche hall of fame: Two wrongs don't make a right. Such an answer will not satisfy a single-issue opponent of abortion (and there is no better issue to be single-issue about) who might argue that welfare is
100 WELFARE REFORM AND ABORTION the lesser of two evils and thus justifiable. To such a person we need to explain that passing out welfare dollars to avoid abortion is like succumbing to extor- tion, and extortionists will want more and more. Other proponents argue that the effect of the legislation will be to cut abor- tions because of changes in sexual behavior (Bauer and Gramm, 1995~: To reduce abortions, we must reduce the out-of-wedlock pregnancy that creates a perceived "need" for abortion .... If it is true that preserving the current benefits package will save lives, presumably we could save even more lives by increasing the entitlements tenfold or by expanding welfare without limit to create a pro-life socialism. If we are to reverse the trend in both illegitimacy and abortion, we must first acknowledge that illegitimacy and abortion are twin evils spawned by govern- ment policies that eliminate personal responsibility in favor of government responsibility. The twin concerns that the welfare system was encouraging nonmarital fer- tility and that changes to the system would increase the number of abortions strongly shaped the Personal Responsibility and Work Opportunity Reconcilia- tion Act of 1996 (hereafter, PRWORA). With respect to the first concern, ille- gitimacy, the official summary of the legislation (U.S. Congress, 1996: 6-8) explicitly names rising rates of "illegitimacy" and its negative consequences as a motivation for the reforms. To "reduce nonmarital births in general and teen births in particular," the legislation requires teen mothers to live at home and attend school, penalizes those who do not help to establish paternity, provides funding for abstinence education, requires reporting on state performance in reducing nonmarital birth ratios, and provides $1.4 billion in "performance bo- nuses" for states that reduce nonmarital births and illegitimacy ratios. With respect to the second concern abortion several other steps were taken. First, $0.4 billion of the performance bonus funds were specifically allo- cated "for the five States that are the most successful in reducing the number of out-of-wedlock births while decreasing abortion ratios" [emphasis added] (U.S. Congress, 1996:18~. In addition, the right-to-life community's strong lobbying forced the removal of a mandatory family cap from the final legislation. This chapter attempts to draw together what we know today about the likely effects of welfare reform on abortion and to outline promising strategies for evaluating the actual effects of the limited reforms to date and the wider reforms that are likely to follow. The chapter opens with two sections that briefly review the legal status and demographic importance of abortion today. The third section presents a simple rational choice model of a woman's choice of contraception, abortion, or fertility. The model focuses on the effect of welfare policy. The basic model implies that welfare reform will increase abor- tions. Several extensions to the model that might overturn the implication of the basic model are also considered. The section concludes with insights from the
JACOB ALEX KLERMAN 101 sociology and social psychology literature as to the likely magnitude of the effects. The following section reviews possible data sources for analyses of effects on abortion. It begins with a review of the evidence that abortion is seriously underreported in survey data to the point where such data are nearly useless. It then discusses several more promising non-survey-based sources of data on abortion. The fifth section provides a discussion of the methodological issues in evalu- ating the causal effects of welfare reform on the number of abortions. In the main, the issues raised here are likely to be common across most of the domains of possible demographic effects of welfare reform concerns that simple cross- sectional or time-series results might reflect spurious correlation rather than the true causal effect of the policies and statistical approaches to this concern. These considerations suggest that even if household survey data were reliable, it seems unlikely that the survey data would have large enough samples to have power to detect effects of the size that seem plausible. There is some prospect of detecting effects using other data sources. The next section combines the insights from our discussions of the theory, the data, and the methodological issues to try to draw some insights from the existing literature. The theoretical model suggests some analyses that might be insightful. The methods and data discussion suggests that many of the studies that might be insightful for the effects of welfare reform or abortions are unlikely to be empirically robust. Thus, this literature survey focuses on the papers that meet the data and method screens. Specifically, we explore the issues of whether welfare policy affects fertility and abortion and whether abortion policy affects contraceptive behavior, abortion, and fertility. Using the perspectives gained from our discussion of data and methodologi- cal issues, the final section tries to put these pieces together to sketch potential research strategies to explore the actual effects of welfare reform on the level of abortions. LEGAL STATUS OF ABORTION . We begin this section with a discussion of the legal status of abortion. Un- less otherwise noted, this discussion is based on the Merz et al. (1996) summary of the regulation of abortion. Complete citations to the statutes and case law can be found in Merz et al. (1995~. Abortion in the United States has been legal, essentially without restriction, since the Supreme Court's Roe v. Wade decision in 1973. That decision capped a movement towards the liberalization of abortion laws at the state level that began in Colorado in 1966 and included 19 states by the time of Roe v. Wade. Through the late-1980s, there were two major exceptions to the characteriza- tion of Roe v. Wade as guaranteeing abortion on demand: Medicaid funding and
102 WELFARE REFORM AND ABORTION parental consent. Both exceptions figure prominently in the polemics of the right-to-life movement with respect to welfare reform. Following Roe v. Wade, most state Medicaid programs nominally appear to have reimbursed abortion like any other medical procedure. As with other medi- cal procedures, this reimbursement was funded through a combination of state allocations and federal matching funds. In practice, the rates of funding of abortions varied widely across states (Alan Guttmacher Institute, 1979; Klerman, 1996~. Then in the late-1970s, even this de jure reimbursement policy changed. First in federal legislation known as the Hyde Amendment (first passed in 1976), the federal government stopped providing matching funds for state Medicaid expenditures for abortions. Then, after considerable controversy in the lower courts, a series of federal court decisions (Beal v. Doe, Maher v. Roe, and Poelker v. Doe, all in 1977, and Harris v. McRae in 1980) ruled that the states themselves did not have to fund abortions. Given the resulting sharply higher cost to the states of funding abortions without federal matching funds, the opportunity to lower the cost of their Medic- aid programs, and ideological/moral objection to abortion, most states promptly changed their policy from funding abortions under Medicaid to not funding abor- tions. Several states, however, have continued to fund abortions some by explicit legislation and some pursuant to the state supreme court's interpretation of either the federal or the state constitution. As of the mid-199Os, the following states were funding abortions under Medicaid: Alaska, District of Columbia, Hawaii, Maryland, New Mexico, New York, North Carolina, Washington, and West Vir- ginia by statute or administrative action; and California, Connecticut, Illinois, Massachusetts, New Jersey, Oregon, Pennsylvania, and Vermont by order of their state supreme courts. This state funding of abortions under Medicaid is a component of one of the pro-life community's antiwelfare reform arguments, which notes that "it is im- plausible that a change in welfare law will decrease teen sexuality while abortion on demand, often paid for by public funds, remains readily available for unin- tended pregnancies" (O' Steen, 1995~. The other major restriction on abortion allowed by the Supreme Court is a requirement for parental notification/consent. In a series of decisions in the early- 1980s, the Supreme Court ruled that, as long as an appropriate bypass procedure to establish maturity is provided, states may require that parents be notified or even give their consent before an immature minor has an abortion. As of the mid-199Os, the following states had consent statutes in force: Alabama, Indiana, Kentucky, Louisiana, Maine, Massachusetts, Michigan, Mississippi, Missouri, North Dakota, Rhode Island, South Carolina, Wisconsin, and Wyoming. In addition, the follow- ing states had notification statutes in force: Arkansas, Georgia, Kansas, Minnesota, Nebraska, Ohio, Utah, and West Virginia. The pro-life lobby claims that such parental notification laws are effective in lowering the number of births. Citing a drop in both abortions and teen births in
JACOB ALEX KLERMAN 103 Minnesota following the passage of a parental consent law, the pro-life lobby proposes such laws as an alternative to welfare reform: "Apparently the knowl- edge that should pregnancy occur, the girl's parents will become aware of it either through continuation of pregnancy or notification of an abortion does have an effect on teen sexuality" (O'Steen, 1995~. This legal landscape of abortion limited only by parental consent/notifica- tion and lack of Medicaid funding is now shifting. The Supreme Court, which had struck down essentially all other limitations, changed its position with the Casey decision of 1990. In that decision, it upheld additional state efforts to circumscribe the right to abortion. Most noteworthy are "informed consent" requirements. These laws require that women seeking abortion receive pre- scribed information specifically intended to deter the woman from having an abortion. DEMOGRAPHIC ROLE OF ABORTION Abortion in the United States is far from demographically insignificant. In 1992, the most recent year for which final data are available, there were about 1.5 million abortions performed (Henshaw and Van Vort,1994~. This number should be compared against approximately 4 million births. Therefore, more than a quarter of all pregnancies (27.5 percent) are aborted (see Table 5-1~.3 Put differ- ently, about 2.6 percent of women of childbearing age (15-44) have an abortion each year. The demographic importance of abortion varies widely across subgroups (see Table 5- 1~. Abortion rates (abortions per woman) are highest among women in their early 20s, but the abortion ratios (abortions per pregnancy4) are high for both young and older women (unless otherwise noted all of the figures in this section are from Ventura et al., 1995, and refer to 1991~. More than a third of pregnancies to women over 40 and under 20 are aborted. For women under 15, the figure is half. Both the abortion rate and the abortion ratio are nearly twice as high for nonwhites as for whites. These rates represent a decline of about 10 percent from their highs in the early 1980s (all the figures in this paragraph are from Henshaw and Van Vort, 1994~. The overall abortion rate (abortions per 1,000 women age 3Several other definitions of the abortion ratio are used in the literature. In some publications of the centers for Disease Control and Prevention, the abortion ratio is defined as abortions divided by births (i.e., abortions are not included in the denominator; see for example Koonin et al., 1995). In some publications of the National center for Health statistics, the denominator includes abortions and an estimate of the number of pregnancies ending in miscarriages (see, for example venture et aL, 1995). 4In this abortion ratio, pregnancies are defined as induced abortions plus births. Miscarriages and spontaneous abortions are excluded from the denominator. some sources use an abortion ratio that includes an estimate of miscarriages (e.g., venture et aL, 1995).
104 TABLE 5-1 Abortion Rates and Ratios by Subgroups WELFARE REFORM AND ABORTION Ratea Ratio (%)b 1976 1980 1991 1976 1980 1991 Age <15 1.6 1.7 1.4 57 61 50 15-17 24.2 30.1 24.3 42 48 39 18-19 49.3 60.6 55.9 38 42 37 20-24 39.6 51.6 56.6 26 31 33 25-29 24.1 31.0 33.7 18 22 22 30-34 15.0 17.2 19.1 22 22 19 35-39 9.3 9.4 10.4 33 32 25 40+ 3.7 3.5 3.0 45 46 34 Married 10.5 8.4 10 9 Unmarried 54.4 47.8 65 51 White 18.8 24.4 20.3 23 27 23 Married 8.6 6.6 8 7 Unmarried 47.4 39.1 72 53 Other Races 56.3 57.0 53.8 41 41 40 Married 24.7 20.6 21 19 Unmarried 82.7 75.8 52 49 Overall 29.4 26.3 25.9 30 27 27 aRate: abortions per 1,000 women. bRatio: abortions/(abortions + live births); note that this is not the definition used in the source. SOURCE: Computed from Ventura et al. (1995: Tables 1, 3, and 4). 15 to 44) increased nearly fivefold from first legalization in 1970 through 1980.5 Even from the first full year following national legalization, 1974, there was an increase of about a half; from 19.3 to 29.3. Then the abortion rate declined to 25.9 in 1992. The abortion ratio also declined from its peak of 30.1 in 1981 to 27.5 in 1992. Preliminary data from the Centers for Disease Control (CDC) suggest a further 7 percent drop in the number of abortions from 1992 to 1994. Furthermore, abortion ratios are much higher among unmarried women- those who might be eligible for AFDC if they did not abort. While in 1991 only 10 percent of pregnancies to married women ended in abortions, over half (51 percent) of pregnancies to unmarried women ended in abortion.6 Especially for 5It seems likely that not all of this increase in abortions in the early 1970s was real. Some of it is probably the reporting of what previously would have been illegal and unreported abortions. 6Marital status is defined at the time of the pregnancy outcome abortion or live birth. In the absence of abortion, some of these aborted pregnancies would probably result in postconception marriage and marital births. With the abortion option, such pregnancies instead result in nonmarital abortions. On the interaction between abortion availability and marriage, see Akerloff et al. (1996), Kane and Staiger (1996), and the theoretical discussion in the next section.
JACOB ALEX KLERMAN 105 whites, the abortion ratio for nonmarried women appears to have fallen over the last decade (from 72 percent in 1980 to 53 percent in 1991~. This decline in abortion among nonmarried women has been used to explain some of the in- crease over the 1980s in nonmarital birth rates (U.S. Department of Health and Human Services, 1995~.7 A THEORETICAL PERSPECTIVE In this section, we outline a simple rational choice model of fertility.8 The model is developed to emphasize that because welfare reform will make having a child less attractive, some women will chose to avoid a birth through abortion. Given that goal, the model deliberately excludes many of the other features of the standard proximate determinants approach to fertility.9 This basic model sug- gests that abortions will increase with welfare reform, but that the magnitude of the increase is crucially dependent on whether fertility effects are achieved through improved contraception or through abortion. The sociology literature on adolescent pregnancy is then surveyed for insights into this question. The section concludes with a consideration of alternative model features that overturn the strong result of the basic model. Model Structure Given the nature of the abortion decision as temporally subsequent to deci- sions about sexual activity and contraceptive practice, but temporally prior to receipt of welfare, it is useful to consider the abortion decision among the se- quence of decisions leading from sexual activity to welfare receipt. Figure 5-1 represents the sequence of decisions graphically. Time unfolds in four periods. At period 1 (labeled "Sex"), the woman chooses a level of sexual activity and contraceptive practice. Those decisions imply a probability of pregnancy. At period 2 (labeled "Conception"), nature "moves" and the woman does or does not become pregnant according to the probability implicit in her decisions about sexual activity and contraceptive practice. At period 3 (labeled "Abortion"), those women who conceive choose whether or not to abort. At period 4 (labeled "Marriage/Welfare"), those women who conceive, but choose not to abort, then 7While striking, this stratification based on marital status is problematic. The model presented in the next section treats marital status as a choice. In particular, it was once true that many nonmarital conceptions resulted in marital births. The likelihood of such marriages might itself vary with welfare reform. 8Similar models can be found in Plotnick (1993), Lundberg and Plotnick (1995), and Jackson and Klerman (1996). 9Among the features ignored are the wontedness of pregnancies and the availability of contracep- tion and abortion. For a more conventional proximate determinants approach, see U.S. Department of Health and Human Services (1995:39 If. and Chapter 1, in this volume).
106 WELFARE REFORM AND ABORTION - A: NEVER pregnant | 3. Abortion? | yes > B: ABORT \no C: MARRY ~ (no welfare) 4. Marriage/Welfare? > D: SINGLE -(no welfare) - E: Single (WELFARE) FIGURE 5-1: Model Schematic. NOTE: labels in capital letters are used in the text to refer to the final outcomes. choose among marriage (and no welfare), no marriage and no welfare, or no marriage and welfare. To simplify the following theoretical discussion, we refer to these final outcomes as NEVER, ABORT, MARRY, SINGLE (no welfare), and WELFARE (single, welfare). In addition, we often refer to the pre-PRWORA welfare system as "before" and to the post-PRWORA welfare system as "after." Implications for the Effects of Welfare Reform In this simple model, what would be the effect of a welfare reform that made WELFARE less attractive for every woman? We begin the analysis by consider- ing the standard rational choice assumptions when a woman makes decisions about sexual activity and contraception, she knows what she would do (ABORT, MARRY, SINGLE, WELFARE) if she became pregnant. Furthermore, she makes her decisions about sexual activity and contraception based on what she would do if she conceived. We refer to these assumptions as the "basic model." We consider the effect of relaxing these assumptions below. The insight of this sequential approach follows from dividing women into one of four groups based on the choice they would make if they found themselves pregnant before welfare reform: ABORT, MARRY, SINGLE, or WELFARE. Any woman who before welfare reform would have chosen MARRY, SINGLE, or ABORT, will still make that choice. Finding herself pregnant, she would not have chosen WELFARE before welfare reform. After welfare reform, WEL- FARE is less attractive; she certainly would not choose it. Furthermore, since the choice of such a woman if she became pregnant is unchanged, her contraceptive choices (at period 1, "Sex") are also unchanged. In particular, the group of women who finding themselves pregnant would have chosen ABORT (before welfare reform) will not change their contraceptive practices. Their probability of conceiving is thus unchanged. If they conceive,
JACOB ALEX KLERMAN 107 they will still abort. It follows therefore that the number of abortions will not decrease with welfare reform. The only way that the number of abortions could decrease would be if some women who had chosen to abort before reform, will choose not to abort after reform. We have, however, argued that any woman who would have aborted before will change neither her contraceptive practices nor her abortion decision if she becomes pregnant. Welfare reform will thus affect only women who finding themselves preg- nant would before welfare reform have chosen WELFARE. Finding that choice less attractive after welfare reform, some of them will change their behav- ior in one of three ways. First, some of them may "contracept" more effectively (or become less sexually active). Second, conditional on finding themselves pregnant, some of them may choose ABORT. Third, more of those who find themselves pregnant and choose not to abort will not choose SINGLE or MARRY rather than WELFARE. These changes are likely to lead to a larger number of abortions. Some women will choose to contracept more aggressively. Nevertheless, those meth- ods will not be perfectly effective. Having tried to contracept and failed, some of them will abort. Furthermore, knowing that abortion is available, some women will continue to choose not to contracept. Thus, the implication of the basic model is that welfare reform will not lower abortions. Any women who would have aborted before welfare reform will still abort. Some women who would have gone on welfare before welfare reform will now abort. This later result is true despite the fact that, given that welfare is less attractive after welfare reform, women who before would have gone on welfare if they became pregnant may now (after) contracept more aggressively, and such more aggressive contraception would lower the number of pregnancies. On Contraceptive Behavior The rational choice perspective of the previous section is strikingly at vari- ance with the tenor of much of the writing about adolescent fertility (e.g., Ellwood, 1987; Musick, 1993; Luker, 1996~. Ellwood (1987:16), for example, writes "There seems to be ample evidence to support almost any model of teenage behavior except a model of pure rational choice." There is much to be said against applying a model of rational choice to teenage fertility choices. Adolescence is the time when humans develop a sense of the long-term consequences of their actions (Petersen, 1988; Keating,1990~. Adolescents often view themselves as invincible, leading to risk-taking behavior. Adolescents who currently have children are disproportionately drawn from the bottom quartile of the reading and math skill distribution (Pittman and Govan,1986~. Each of these considerations points to the possibility that adolescents might not consciously choose to have a child. Furthermore, even given such a choice, they might have trouble implementing it.
108 WELFARE REFORM AND ABORTION The right-to-life community sometimes takes this position. O' Steen (1995), discussing the effects of parental involvement laws, quotes a Dr. Franz: Adolescent thinking and reasoning tends to be concrete rather than abstract and focused on the immediate rather than the long-term consequence of an action. Hence, adolescents have difficulties with rational analysis that requires them to determine cause and effect, choose delayed over immediate gratification, and realistically assess the likely consequences of their choices. This makes them very vulnerable to the incessant stimulation of our sex-drenched culture, which stresses immediate gratification. It also makes it very unlikely that they will forego sexual activity just because sometime in the future there won't be any welfare payments. That is, at least among adolescents, adjustments to PRWORA would be primarily through abortion. The right-to-life community appears to be of two minds about this question. Elsewhere in the same article, however, O' Steen argues that the Minnesota law had resulted in a drop in pregnancies, i.e., that changing abortion policy changed teenage sexual activity. The basic model suggests that welfare reform will cause nonmarital fertil- ity to decline. The strong form of this literature on adolescent decision making would imply that, in neither the proreform period nor the postreform period, have/will adolescents adjusted/adjust their fertility to the generosity of the wel- fare system. Chapter 4 reviews the empirical evidence on that question and its implications for the effects of welfare reform. Note, in particular, that a woman's decisions need not be "totally rational" in order for her to adjust her behavior with the changes in the generosity of the welfare system with welfare reform. In particular, a woman need not adjust her sexual practices and contraceptive strategies. For there to be a fertility effect, it is sufficient that, finding herself pregnant, she choose to abort the pregnancy. The logic of the literature on adolescent decision making suggests that behavioral changes are more likely after conception. At that point, the issue is no longer one of risk taking. Failure to act will result in the birth of a child. Thus, this line of argument suggests both that the magnitude of the changes in fertility is likely to be small and that any changes in fertility will be accomplished mostly through abortion. It is important to note, however, that there is an alternative theme in the literature. This theme emphasizes that for many poor adolescent girls, all options are unattractive. Motherhood may not be sufficiently worse than the other choices to make worthwhile the aggressive contraceptive strategies and high financial and emotional cost of abortion. For many girls, motherhood may actually look more attractive. Having a child gives her a clear role and certifies that she is an adult. The child is someone on whom the new mother can shower affection and from whom the new mother can expect unconditional love. The child is a chance to "start over," to make up for the errors of the previous generation (Luker, 1996; Musick, 1993:Chapter 5~.
JACOB ALEX KLERMAN 109 Furthermore, the child brings an entitlement to welfare. Welfare will pro- vide a way for the new mother to establish a household of her own, away from her parents. It will provide her with an independent source of income. Finally, the new child will free the woman from the pressures and expectations of the world of school and work. The preceding paragraph, of course, describes the proreform choice of moth- erhood and welfare. PRWORA was deliberately designed to make having a child out of wedlock less attractive. Under PRWORA, a minor mother must live with one of her parents. She will be expected to finish school. The payments will often be smaller. They will be of strictly limited duration. The payments are deliberately only "temporary assistance." Perhaps this worsening of the welfare option will be enough to cause women to change their preconception behaviors. If for women currently on welfare, having a child was a choice (or at least, they did not choose to take sufficiently aggressive efforts to avoid the birth of a child), then there is some hope that women might achieve changes in fertility through changes in coital frequency and contraceptive strategies. Certainly, so- cial disapproval of abortion would push some women in that direction. Even if women who would have gone onto welfare want to avoid births, it is not clear that they can succeed in avoiding pregnancy. Effective use of contra- ception is eased by a settled, regular life-style; regular sexual activity; and a supportive partner. Adolescent sexuality, however, tends to be irregular, partners change frequently, and men are not always supportive (Luker, 1996; Zabin and Clark, 1981; Zelnik and Shah, 1983; Alan Guttmacher Institute, 1994; Kost and Forrest, 1989~. Some of the sex is involuntary or close to involuntary (Moore et al., 1989~. Social norms about women looking too "ready" discourage women from contracepting outside of an ongoing relationship (Luker, 1996~. In sum, the sociology and social psychology literatures suggest that many women will have trouble contracepting effectively. Inasmuch as they react to welfare reform by desiring fewer children, these theoretical literatures suggest that much of the adjustment will be through abortion, rather than changed contra- ceptive usage. Nevertheless, some women do contracept successfully now. Some women who formerly would have gone on welfare will after reform also de- crease their coital frequency and contracept more effectively, leading to fewer pregnancies. Below we consider evidence on this question from reactions to other policy changes. By combining information on the response of abortions and fertility to policy changes, in principle we can infer something about the relative size of the changes in fertility due to changes in abortion and those due to changes in sexual behavior. For example, if a state raises the cost of an abortion (by stopping Medicaid funding, requiring parental notification/consent, etc.), the change in abortions /\a, can be decomposed into /\c, the change in conceptions due to better contracepting, and /\b, the change in births (where /\a = /\c + /\b). Since /\a and /\b are observable (more precisely, can in principle be estimated from the data),
0 WELFARE REFORM AND ABORTION we can compute /\c by subtraction. Similarly, if a state lowers its welfare pay- ments (or in some other way makes welfare less attractive), any change in fertil- ity, fib, can be decomposed into a change in abortions, /\a, and by subtraction a change due to contraceptive practice (/\c = fib - /`a). Generalizations of the Basic Model Logically, welfare reform could cause abortions to decline only if it induced some of the women, who before welfare reform would choose to abort, to choose some other outcome either to contracept more effectively or, conditional on pregnancy, to choose some option other than abortion. The analysis of our basic model showed that the standard rational choice assumptions rule out that possi- bility. Here we consider several modifications to the model that might overturn this strong theoretical result. The strong theoretical result that abortions must increase follows directly from the assumption that, when they choose a contraceptive strategy, women know exactly what they would do if they became pregnant. If new information will arrive after conception, then it is not appropriate to assume that a woman who currently chooses ABORT, knew with certainty that this would be her choice if she became pregnant. Such a woman would need to consider all of the possible types of information that would arrive. If some arriving information would make WELFARE the best option, then even a woman who currently chooses ABORT would need to consider the value of the welfare package in making her contracep- tive choices.l° In particular, she might choose to contracept better. Such better contraception would lower the number of abortions. Over the appropriate time span of under 3 months, a woman is unlikely to learn a lot about life in general her ability, her earnings prospects, etc. Preg- nancy, however, might directly cause the arrival of new information. The obvi- ous information is how the woman feels about being pregnant. Another learning has occupied a central place in the literature on abortion. In the model of Kane and Staiger (1996), women become pregnant partially to determine if the potential father will "support" the child.ll If, however, unlike in Kane and Staiger's paper, the woman chooses WELFARE (not MARRY) when the father indicates that he will support the child, and ABORT when he does not, Din the standard expected utility formulation, she would compute the probability that, given her knowledge at conception, she would choose each of the options ABORT, MARRY, SINGLE, and WELFARE. Then she would make her choice of contraceptive strategy considering the probability of each final decision conditional on pregnancy. In this formulation, even women who before reform did choose ABORT, need to consider the welfare payment when choosing a contraceptive strategy. 1lIn Kane and Staiger (1996), the choice is between MARRY and ABORT. Since these women would have married, welfare reform has no effect on their choices. Our discussion generalizes their model to the choice between WELFARE and ABORT. This model makes sense only if fathers provide support to their children even when the mother collects WELFARE.
JACOB ALEX KLERMAN 111 then abortions could increase with welfare reform. If welfare reform makes WELFARE less attractive, then fewer women would become pregnant to deter- mine if the father would support the child. By eliminating some of the pregnan- cies in which the man would have not "committed," this could lower the number of abortions. Presumably, finding themselves pregnant, some women who before would have chosen WELFARE, now find WELFARE less attractive and will choose ABORT. Again, the net effect is ambiguous. This analysis rests on the plausibility of the assumption that when the man commits, the woman chooses WELFARE and when he does not, she chooses ABORT. The alternative choices of MARRY/ABORT (for commit/don't commit) or MARRY/WELFARE seem more likely than WELFARE/ABORT.12 All of the preceding arguments assume that women know correctly what they would do if they became pregnant. Some women who had planned WEL- FARE if they became pregnant would when faced with the reality of pregnancy and the real possibility of parenthood choose to abort. We refer to any such discrepancy between the planned action and actual behavior as "time inconsis- tency." Welfare reform lowers the utility of having a child. This would induce some of these women to change their sexual behavior to make conceptions less likely. Fewer women would find themselves pregnant and choose to abort. Thus, this time-inconsistency scenario implies that welfare reform could decrease the number of abortions. Other cases imply that, as in the base case, the possibility of time inconsistency will cause welfare reform to increase the number of abor- tions. Table 5-2 enumerates the possible cases. The first row is the case just discussed. Before reform, women would have planned WELFARE and then actually chosen ABORT. After welfare reform, these women contracept better and never become pregnant. To understand the next four rows of the table, note that after welfare reform, some of the women who expected to choose WEL- FARE will now expect to choose the other choices (NEVER, ABORT, MARRY, and SINGLE). Finding themselves pregnant, they previously would have chosen ABORT. The other choices have not gotten any better, so they would still choose ABORT. The net effect in this first panel is to decrease abortions. Consider, however, the second panel (Plan = E, Action = E). It is composed of people who before welfare reform planned WELFARE and when pregnant actually chose WELFARE. After welfare reform, welfare looks less attractive, so they will be less likely to plan to choose WELFARE if they became pregnant. Recall that these are women who are likely to find ABORT more attractive when pregnant than they expected when making contraceptive decisions. Thus, some of those who planned MARRY, SINGLE, or WELFARE will choose ABORT. 12See Akerlof et al., 1996, who argue that a marriage MARRY/WELFARE pairing along with changes in contraceptive technology might explain the increase in black nonmarital fertility.
112 TABLE 5-2 Effects of Time Inconsistency WELFARE REFORM AND ABORTION Beforea Plan Action Afterb Change in Plan Action Abortionsb E B A A - B B = C B + D B + E B + E E A A = B B = C C = C B + D D = D B + E E = E B + B E B B + B C = B D = B E = NOTE: Letters refer to final states in Figure 5-1: A = NEVER pregnant; B = ABORT; C = MARRY; D = SINGLE (no welfare); E = single (WELFARE). + = increase; - = decrease; = is no change. a "Before" = before welfare reform. b "After" = after welfare reform. c "Change in Abortions" gives the change in abortions with welfare reform. All of these time inconsistencies increase the number of abortions. This increase is in addition to the time-consistent increase in abortions from the basic model. Similarly (the third panel of the table: Plan = B. Action = E), some women who plan to choose ABORT if they become pregnant, would in fact not be able to "go through with it" when they have to make the decision. Encouraging this outcome appears to be the intent of the "informed consent" statutes noted earlier. For those who would have chosen WELFARE before reform, there is an increase in abortions. If welfare reform lowers the utility of having the child, then more of these women would have the abortion. This reinforces the earlier results based on the basic model. Welfare reform should increase the number of abortions. An alternative pathway through which welfare reform could induce some of the women to change from ABORT to not ABORT would be for welfare reform to change the utility of the other options.l3 For example, if states respond to the 13Note that some relaxations of this form strengthen the basic result. For example, Jackson and Klerman (1996) note that welfare reform will also affect the utility of the other states. Some women
JACOB ALEX KLERMAN 113 performance bonuses by increasing funding for family-planning services (and such services are effective; see Meter and McFarlane, 1993), contraception would become cheaper, women would contracept more effectively, and fewer women would find themselves pregnant such that they needed to choose ABORT. Simi- larly, if the abstinence education programs funded by the legislation are success- ful in changing the behavior of women who would have aborted, then abortions would again decline. The earlier quotes from the right-to-life movement suggest another mecha- nism. Welfare receipt makes women categorically eligible for Medicaid. In many states, Medicaid funds abortions. If PRWORA makes few women eligible for Medicaid, abortions will become more expensive. After such a change in Medicaid eligibility, some women who, finding themselves pregnant, would have chosen abortion will instead choose one of the other three options and all women who would have chosen abortion will take more aggressive steps to avoid preg- nancy. The details of PRWORA make this outcome unlikely. PRWORA specifi- cally continues Medicaid eligibility for women who would have been eligible for AFDC under the proreform rules. Thus, PRWORA will only lower Medicaid eligibility if it induces a large number of women to marry at their first pregnancy or to go to work and have earned incomes so high that they would not have been eligible for AFDC. Finally, some proponents of welfare reform suggest that it will induce a broad change in social attitudes towards nonmarital sexual relations, abortion, and nonmarital childbearing. In this view, the current generous welfare system encourages a general climate of promiscuity and high levels of unprotected sexual activity. With welfare reform, women who would have chosen welfare will individually be more "careful." This will result in a lowering of the peer pressure for sexual relations (Musick, 1993; Luker, 1996), the pressures against contra- ceptive use, and the social acceptance of nonmarital fertility. Such a result would probably emerge from models of the marriage market such as that of Willis (1995) or Akerlof et al. (1996~. With lower coital frequency, more effective contraception, and more marriage, the number of abortions might fall. On the other hand, an increase in the stigma attached to nonmarital fertility might in- crease abortion. The net effect is ambiguous among those who become pregnant. Implications for Abortion Regulation The model also has predictions about the effects of abortion regulation. Most forms of abortion regulation in particular, making it illegal, requiring who have the child, but do not immediately go on welfare, may nevertheless treat welfare payments as "insurance," potential income if the marriage breaks up. If welfare reform makes welfare not as effective as insurance, then some women who would have chosen MARRY or SINGLE, will choose ABORT after welfare reform.
4 WELFARE REFORM AND ABORTION parental notification or consent, not funding it through Medicaid have the effect of making abortion more expensive (with "cost" viewed not merely in terms of out-of-pocket costs). This makes the choice of abortion more expensive. The analysis then proceeds as in the analysis of the effects of welfare reform. The only women who would be affected by welfare reform are those who would have chosen abortion had they found themselves pregnant. In general, the model predicts that these women will be less likely to choose abortion and more likely to choose each of the other options. In particular, they will contracept more aggressively (or lower coital frequency). Finding themselves pregnant, they will be more likely to choose the other three options: MARRY, WELFARE, and SINGLE. Thus, in net the total number of abortions performed will go down and the number of births overall, marital, nonmarital will go up. The abortion ratio the ratio of abortions to live births will also decrease. The generalizations continue to apply. In particular, under the Kane and Staiger (1996) model in which fathers reveal whether or not they will sup- port the child, births could go down. It is now more expensive to learn whether the father will support the child (because if he does not, the neces- sary abortion is more expensive). Thus, women will contracept more effec- tively. Fewer women will become pregnant to determine if the father would support the child. In net, there will be both fewer births and fewer abortions. Time inconsistency could also overturn these results. Consider the case of women who thought ex ante that they would abort, but ex post would not. Knowing that abortion was more expensive (and thinking that they would abort), they would contracept more aggressively. Since they would not abort, the smaller number of pregnancies yields a smaller number of births. The net effect is of course a function of the relative size of the two groups (the time-consistent group and the time-inconsistent group) and the relative magnitude of the effects. Note also that time inconsistency could plausibly have the opposite ef- fect. Women who thought they would not abort if they became pregnant, in fact do abort. Their preconception behavior is unaffected, but finding the cost of the abortion higher, they are less likely to abort. This possibility would reinforce the conclusion that with regulation of abortion, abortions would decrease and births would increase. ABORTION DATA More than most other subfields of demography, empirical research on abortion its levels, its determinants, and its effects is limited by the avail- able data. Abortion is close to the archetypical "sensitive topic" for survey research. Public opinion polling finds that a large fraction of the population
JACOB ALEX KLERMAN 115 is strongly opposed to abortion, so issues of interviewer/public approval are likely to lead to underreporting. Similarly, many Americans are themselves conflicted in their attitudes towards abortion and the decision is likely to have come at a difficult time, so suppressing unpleasant memories is also likely to result in underreporting. AGI Provider Surveys The best reports of the number of abortions in the United States appear to be collected by the Alan Guttmacher Institute (AGI).14 AGI does an approximately annual survey of abortion providers. Provider surveys were conducted annually from 1973 to the present, except for 1983,1986,1989,1990, and 1992. Henshaw and Van Vort (1994) provide a description of their methods. Given the origin of that organization as a semiautonomous division of Planned Parenthood Federation of America, more than anyone else, they are able to gain the confidence of abortion providers. Nevertheless, their survey undoubt- edly misses some small providers (e.g., gynecologists providing individual abor- tions to mature, long-term private patients). Also, AGI estimates abortions be- cause some of the larger commercial abortion providers do not report (at all or partially) for business reasons. The AGI survey of providers is subject to two important drawbacks. First, it collects no covariate information. AGI simply collects the number of abortions performed. An occasional AGI survey (the most recent one in 1994-1995; see Henshaw and Kost, 1996) of women having abortions collects some information, but this patient survey is far from annual. In addition, refusals are a problem in the patient survey. Many providers refuse to allow the patient survey; many patients refuse to participate. The other major problem with the AGI data is that, being a provider-based survey, it records state of occurrence. For most population-based analyses and in particular the effects of welfare reform we would like to know the number of abortions by the woman's state of residence. CDC (1997) reports that for 8.3 percent of abortions, the woman's state of residence is not the state in which the abortion occurs. These CDC estimates are likely to be a lower bound. See the discussion of this issue in Blank et al. (1996~. They note, for example, that interstate travel is the most plausible explanation for why the District of Columbia' s abortion rate is four times that of any other state. AGI also provides an estimate of abortions by state of residence. The algo 14See for example the statement of Jones and Forrest (1992a): "The AGI statistics nevertheless are widely accepted as the best available estimates of the incidence of abortion in the United States." This quotation is followed by a citation to the Statistical Abstract, which, in fact, reports the AGI numbers. It should be noted that both Jones and Forrest are senior staff members at AGI.
6 WELFARE REFORM AND ABORTION rithm for transforming AGI's state-of-occurrence data into state-of-residence es- timates is given in Henshaw and Van Vort (1992~. Particularly relevant for the methodological discussion below, the algorithm would not capture policy- induced changes in the demographic characteristics of women going out of state for an abortion. A consistent series is available for 1978-1988 (except for 1983 and 1986 when the underlying survey was not conducted). Earlier estimates using a different algorithm are available for 1974-1977. Blank et al. (1996) note that the series do not appear to splice well, and they do not use the earlier series. CDC Surveillance Data The Centers for Disease Control has an ongoing Abortion Surveillance Pro- gram. That program publishes annual data (1974-present, some data for earlier years) on abortions by state of occurrence, state of residence, race, and age. The data are compiled from reports of state central health agencies (for 47 reporting areas: 44 states, the District of Columbia, New York City, and the balance of New York State), supplemented with other sources including direct contacts with abortion providers (from five reporting areas; unless otherwise noted the infor- mation in this section is drawn from Koonin et al., 1995~. The data are clearly incomplete. Not all states provide reports to CDC. Among those states that do provide reports, the reported number of abortions is often well below the AGI numbers (overall, 12 percent lower). Also not all reporting states provided information on the characteristics of the woman (only 43 reporting areas provide information on the age of the woman; only 36 report- ing areas provide information on the race of the woman) or distinguish between state of occurrence and state of residence (10 reporting areas do not attempt to record out-of-state abortions). Most states report marital status of the woman and her ethnicity (Hispanic/non-Hispanic), but those data often have high missing data rates. Nevertheless, these are the only data that provide any demographic covariates. Standard practice at both AGI and CDC has been to use the AGI estimates for the total number of abortions and the CDC data to estimate the national distribution of abortions by age and race. Micro Data from Vital Statistics All states require individual-level data on births from their vital registration systems for births (i.e., doctors/hospitals are required to report all births on a "birth certificated. In contrast, not all states require individual-level reports of abortions. Only 14 states make similar individual-level data for abortions avail- able: Colorado, Indiana, Kansas, Maine, Missouri, Montana, New York, Ore- gon, Rhode Island, South Carolina, Tennessee, Utah, Vermont, and Virginia (Kochanek, 1989~.
JACOB ALEX KLERMAN 117 These vital statistics systems have two potential problems. First, despite a reporting requirement, compliance is not always complete. Joyce and Kaestner (1995) note severe underreporting (more than 15 percent below the comparable AGI figures) for Colorado, Kansas, New York, and Oregon. Second, many states do not collect data on out-of-state abortions to their own residents (Joyce and Kaestner, 1995, note that this is an issue for Maine, but not for Illinois). Even among those states that do collect data on out-of-state abortions to their own residents, the quality of the data is often suspect (Joyce and Kaestner, 1995, note severe problems with the data from Virginia and suspect that the problem is abortions performed in the District of Columbia). Survey Data Several major surveys of individuals (as opposed to the AGI survey of pro- viders, or the CDC survey of state health departments) among them the Na- tional Survey of Family Growth (NSFG) and the National Longitudinal Survey- Youth (NLS-Y) include questions on abortions. The effectiveness of these questions has been explored by Jones and Forest (1992a). They compare the survey reports with the age-race-time period appropriate national rates (estimated from AGI-CDC data). They conclude that in all of the surveys they examined, abortions are severely underreported. For most of the surveys and subsamples they consider, fewer than half of the abortions are recorded. The underreporting appears to be most severe among unmarried and nonwhite women. No clear pattern by age of the respondent is evident. Attempts by the NLS-Y to use a self- administered questionnaire to increase response rates appear to have been largely unsuccessful. This problem has been noted in most studies of fertility-related behaviors using survey data (Lundberg and Plotnick, 1990, 1995; Currie et al., 1996~. The magnitude of the underreporting and its differential nature with different sub- populations suggest that in evaluating the literature the published studies using survey data should be heavily downweighted.l5 Thus, survey data-based analy . c _ . . . . . . . . . . lilt should be noted that this problem of underreported abortions calls into question several other sets of results from survey data. In particular, the NSFG has been used to compute contraceptive termination, switching, and failure rates (e.g., Grady et al., 1983, 1989). Given the high abortion ratio, these findings should also be considered suspect. In particular, if there was a pregnancy that was terminated by an unreported abortion, then there must be some problem with the NSFG calen- dar. Among the possibilities are the following: (1) a contraceptive failure is reported as a period of continual usage, when in fact abortion terminated an unwanted pregnancy; (2) a period of contracep- tive use was reported as a period of nonuse to suppress the true failure (and the subsequent abortion); (3) the conception occurred during a period of nonuse (and the abortion was not reported). See Jones and Forrest (1989, 1992b) and Grady et al. (1986) for an attempt to address this problem. See also Currie et al. (1996), who propose and implement a strategy for estimating the effects of restrictions on abortions on pregnancy outcomes that is robust to this data problem.
8 WELFARE REFORM AND ABORTION ses do not seem promising for evaluations of the effects of welfare reform on abortion rates. METHODS In this section, we review two methodological issues in estimating the effects of welfare policy on abortion rates. First, we consider under what conditions comparing abortion rates in states with different policies recovers the true effect of the policy (i.e., the effect of changing the policy in a given state in a given year). Second, we consider the required sample sizes to detect substantively important effects. Estimating the Effects of Government Policies If good data are available, the obvious testing/estimation strategy is to re- gress a measure of abortions (or fertility) on policy variables and controls. As discussed in Chapter 4, the appropriateness of this approach depends crucially on how states choose policies. If policies are randomly assigned, such regressions would recover the causal effect of the policy. However, if states with otherwise high or low abortion rates choose a given policy, then such a regression would ascribe to the policy both its direct effect and the variation in the baseline level of abortion in the states that choose the policy. Similarly, if we estimate the effect of a policy change by the change in abortion rates for a state through time, we would ascribe to the policy both its true effect and the effect of other social changes occurring simultaneously. A crucial methodological issue is thus how to estimate the true effect of the law while controlling for persistent differences in the states adopting policies and other social changes. One approach to these issues is to exploit the available data on abortion and fertility rates for states through time. With such time series of cross- sectional data, we can include in the regression dummy regressors for each year and for each state. Such models can be interpreted as follows. The year dummies control for otherwise unmeasured national social changes. The state dummies control for persistent, otherwise unmeasured, differences across states. Some analyses (e.g., Matthews et al., 1995; Jackson and Klerman, 1996) also include state-specific linear time trends and find them to be signifi- cant. Most analyses also include detailed controls for state economic condi- tions. Considerable evidence is now emerging that empirical results are quite sensitive to the inclusion of such dummy variables (see, for example, Moffitt, 1994, on female headship). Nevertheless, these dummy variable estimators are essentially conventional linear regression in which we are using the dummy variables to control for the unobservables. Implicitly, the method assumes that the remaining variation in the crucial policy variable (i.e., after regressing it on the dummy variables and
JACOB ALEX KLERMAN 119 the other regression controls) is random. There is no reason to believe this, except that we have controlled, as well as we can, for the obvious observable . . vanat~on. As Ellwood and Bane (1985) noted in their seminal paper on these issues, state policies themselves do not arrive dens ex machine. They are determined by a political process. We could alternatively argue that the state policies and the outcomes of interest (abortion or fertility) are jointly caused by some unobserved effect. If so, then the estimated effects are spurious. There are at least three possible constructive approaches to this critique. First as Meyer (1995, among others) notes, we often have strong ex ante expec- tations about the relative size of the effect across subgroups. If so, we can stratify the analysis and check if our expectations are confirmed. In the context of abortion policy, such an approach would involve differentially exploring policy effects by age and education groups. We presume that welfare policy and Medicaid funding should primarily effect younger and less educated women. Their earnings prospects are worse, and therefore welfare is likely to be rela- tively more attractive. This will lead their decisions to be marginal with respect to Medicaid reimbursement for abortion or the details of welfare policy (see Klerman, 1996, for an empirical application of this idea). Application of this idea to abortion policy, however, is limited by the data. We noted that, due to the collection of abortion statistics from providers, abortion data stratified by covariates are even less available than high quality abortion data in general. The limited amount of covariate information makes it difficult to apply these stratification ideas to abortion rates. Second, for some cases, we can assert that a policy should have zero effect on a subgroup. When this is a good assumption, we can estimate the effect of the policy by comparing rates for the affected group to those for the unaffected group. The regression analogue is to include dummy variables for every state- year combination. In the context of abortion policy, the obvious example would be parental consent laws. They potentially have effects on minors (women who are 17 or below during the first trimester of their pregnancy) and no effect on women 18 and above at that time. Thus, while including age-year and age-state dummy variables (and perhaps age-state linear time trends), we can also differ- entiate younger women from older women. This would control for state-year effects, which are common across minor and adult women. Doing so is more subtle than it seems. The levels of fertility rates, abortion rates, and abortion ratios are quite different for minors and adults. Thus, it seems implausible that there is a common dummy variable in the level of abortions (Meyer, 1995, makes a similar point). A common dummy variable in the logs (i.e., multiplicative) seems more attractive. i6See Blank et aL (1996) and Haas-Wilson (1996). Both papers stratify by age.
20 WELFARE REFORM AND ABORTION Third, the nature of the legal battles on abortion policy suggests another approach. Presumably the sentiment effects operate most strongly at the level of the state legislatures, and much less so in the courts. The contentiousness of the abortion issue and the constitutional rights under Roe v. Wade have led to almost every piece of abortion legislation being litigated in the courts (see Merz et al., 1995, 1996 for details). The length of time from passage to implementation varies enormously. The correlation between the timing of passage of a restriction on abortion and the actual implementation is very weak. Thus, we can use periods in which a restriction is unenforced as a test for the importance of joint causation (rather than direct causation from policy to demographic rates). Blank et al. (1996), Currie et al. (1996), and Haas-Wilson (1996) implement this ap- proach for abortion. Sample Size Considerations Beyond these issues in estimating the true causal effect of the policy, there is a concern about the required sample sizes for the analysis. Several things con- spire to make the required sample sizes for analyses of the effects of public policies on abortion and fertility much larger than they are for other types of analyses. First, fertility and abortion are discrete events. Compared to continu- ous outcomes, discrete outcomes will in general require larger samples to identify regression effects. Second, the event in question is relatively rare. Overall, about 2.6 women in 100 (aged 15-44) have an abortion in a year. Thus, if the expected effect size is a 40 percentage point change in the age-specific birth/abortion rate, we are look- ing for a 1 percentage point change in the number of women having the event. Given a binary regressor (state abortion regulations, implementation of a particu- lar discrete element of welfare reform), simple normal approximations to the variance of a binomial outcome imply sample sizes of several thousand. Con- sider the example of a simple before-after comparison. Assume that a third of the population experiences a change in welfare/abortion policy and the abortion rate is 2.6 percent (the national average). Then, if the sample size is 50,000 both before and after the change in regime, one standard error of the estimate is 6.5 percent of the mean abortion rate (0.17 abortion per 100 women).~7 Even if the . ~ 1'The implied calculation is as follows. The variance of a binomial variable is simply pq/n. So the standard error of the difference-of-differences (before/after, experimental states/control states) is the variance of the sum of four binomial variables (two with population zN and two with population (1 - z)N where z is the fraction of the population experiencing the change in regime; we assume that the mean is p in each term): sD=.i2*iPq+ P AN (l-z)N J
JACOB ALEX KLERMAN 121 comparison yielded an estimate equal to a true effect of 12 percent, an analyst using 2 years of a survey including 50,000 women of childbearing age would be unable to reject the hypothesis that the deviation was due to chance! This is approximately the size of the largest ongoing survey in the United States the Current Population Survey. It does not record abortions. In many ways, these estimates are too optimistic. The change in the abortion rate corresponding to the above calculations is over all women of childbearing age. If we expect welfare reform to only affect abortion for a minority of the population, the equivalent effect size in that subpopulation would need to be even larger.l8 Most reforms are likely to be implemented in fewer than a third of the states or more than two-thirds of the states. As the fraction of states with the policy becomes more extreme, the required sample sizes increase. Stratifying on covariates (race, age, marital status) will further increase the required sample size. Third, this binomial formula applies only in the absence of covariates. The inclusion of dummy variables washes out much of the variation in the policies. When dummy variables for state are included in the regression model, the only sample that contributes towards the power computation of the previous paragraph are states that change policy during the period under study. States that do not change policy contribute nothing. When dummy variables for state and year are included, states in which no change in policy occurs help to estimate more pre- cisely the year effects but the general point remains. Fourth, the simple binomial formula calculations apply only in the case where up to sampling variability the regression model exactly describes the rates. Jackson and Klerman (1996) present a simple superpopulation formulation in which this statement is meaningful even for vital statistics. If, as seems likely, there are unmodeled components that affect everyone in the state, these simple sample size computations are too conservative. Even larger samples in particu- lar more states and years will be needed. Jackson and Klerman (1996) formulate the equivalent model for the log of the proportion (e.g., Madalla, 1983~. They compute that this state-year specific error is about 3 percent of fertility. Thus, despite the apparently large sample sizes in vital statistics calculations, small changes in fertility rates cannot be 18Some of the implied increase in required sample size is counteracted by the higher abortion rates among the current AFDC population. The estimates in Henshaw and Kost (1992) imply that women with family income below $15,000 have an abortion rate 1.86 times as high as the U.S. average. For those covered by Medicaid, the corresponding figure is 2.04. The pessimistic evaluation in the body of this chapter assumes the use of a general survey of both the affected population and the unaffected population. In that case, the smaller affected population will outweigh any increase in the rate in the affected population. However, if the sample is drawn only from the affected population, then the higher abortion rates for the affected subsample would lower the required sample size for the popula- tion of interest by about 30 percent.
22 WELFARE REFORM AND ABORTION detected in year-to-year changes. Instead, several years of pro- and postchange data will be needed to detect all but the largest effects. In practice, this means that several years of postreform data will be needed; thus, rapid evaluation will not be possible. These sample size considerations are not crucial for studies using AGI or CDC data on aggregate abortion rates. There are now about 60 million women of childbearing age in the United States. Most analyses analyze the full time series of cross sections. In doing so, they implicitly pool several years for "before" data and several years for "after" data. This pooling makes up for the fact that (at least until recently) few states had implemented parental consent laws and few states changed their Medicaid funding policies. These power considerations are, however, relevant for two other cases. First, even if individual-level survey data were of high quality (e.g., the NLS-Y or the NSFG), the sample sizes would not be large enough. We have argued that such dummy variable models are a minimal requirement for estimating causal effects of state policies. Similar arguments would apply to any ad hoc survey to monitor the effects of welfare reform. In addition for such an ad hoc survey, we would need to worry about both the size of the baseline sample (for "before" data) and the size of the postreform sample (for "after" data). Second, these sample size considerations imply that it will be difficult to do reliable "instant policy analysis." Consider a policy put into place in January of 1996, the year of the workshop for which this paper was written. Experience with the 1992 and 1994 data suggests that preliminary CDC data will be released in late 1998 or early 1999. Detailed CDC data, including the crucial disaggregation by age, will not be released until mid-2000. Furthermore, these power consider- ations suggest that reliable estimates will require several years of postreform data. Clearly these results hold to a lesser degree the larger the expected effect of a given reform. However, the effects of most reforms on the overall abortion rate will be moderate to small. Among the current welfare population, neither fertility nor abortions will fall to zero. Since only about a tenth of newborns are on welfare shortly after birth (Jackson and Klerman, 1996), the effect on overall abortion and birth rates will be much smaller. Thus, the largest plausible percent- age changes in overall fertility and abortion rates are well under 5 percent. EXISTING LITERATURE In this section, we review the existing empirical literature for insights into the likely effects of welfare reform on abortion rates. We begin with a review of the limited literature on the effects of AFDC payment levels on abortion rates. We then survey the literature on the effects of Medicaid funding on abortion. As is noted in the right-to-life literature, one possible pathway for the effect of welfare reform on abortions is through a change in the eligibility requirements for
JACOB ALEX KLERMAN 123 Medicaid. If welfare reform has the effect of making fewer women eligible for Medicaid and Medicaid funding of abortions increases abortion rates, then wel- fare reform would lower abortion rates. Finally, we discuss the relative magnitude of the birth and abortion effects of changes in abortion policy and economic conditions. We have noted in the previous section that this comparison of relative magnitudes might provide some insight into the extent to which any effect of welfare on births would come through changes in contraceptive behavior/coital frequency or through abortion. Welfare and Abortion The theoretical model outlined earlier suggests that higher AFDC payments should decrease abortions. Six studies have explored the effect of welfare pay- ments on abortion levels. Two early cross-sectional studies of teenagers (Moore and Caldwell, 1977; Singh, 1986) find little support for this hypothesis. Moore and Caldwell, using small survey samples, find no effect. Singh, using AGI data for 1 year, finds a negative effect. Lundberg and Plotnick (1995) estimate a nested logit model of adolescent childbearing using NLS-Y data (1,089 whites, 462 blacks). The model does not include dummy variables for state or year. They estimate that a 20 percent decrease in welfare payments lowers the probability of pregnancy for a white adolescent by 0.2 percentage point (a fifth of a percentage point based on a coefficient with a t-statistic of over 5~. It, however, lowers the conditional prob- ability of a live birth by 2.9 percentage points. The abortion rate therefore declines by 1.6 percentage points. There is no statistically significant effect on black adolescents. Thus, these results imply that most of the adjustment to variation in welfare payments is through abortion. Note, however, that these results are subject to the earlier caveats about sample sizes, lack of dummy variables, and the sizable underreporting of abortion in survey data. Blank et al. (1996) use both AGI and CDC data on abortions for 1974-1988 and state and year dummy variables to explore the determinants of state abortion rates including AFDC payment levels. Their AFDC results are not robust to different specifications reported in their paper. Their basic model with state and year dummy variables shows no effect of AFDC payment levels. This result is robust to whether they use the more reliable data on state of occurrence (of the abortion) or the conceptually more appropriate data on state of residence. They suggest that this is due to lack of variation in the welfare variable, but this suggestion is not consistent with equivalent models estimated on births (which do find a welfare effect). It would, however, be consistent with problems in the abortion data, particularly in the state-of-residence data. Consistent with Singh's (1986) results, models without state dummy variables show the "wrong" sign
24 WELFARE REFORM AND ABORTION (higher AFDC payments raise abortion rates) and are significant at the 1 percent level. Alternative results using the CDC data also find this "wrong" sign result. The result appears to be due to the subset of states. Regressions on the AGI sample (used in their basic model) for the CDC subset of states also show this positive effect. The advantage of the CDC data is that they make possible disaggregated analyses. Such disaggregated analyses do not appear to provide important additional insights. Using the CDC data, the effects of AFDC on teens versus nonteens and nonwhites versus whites are statistically indistinguishable. Matthews et al. (1997) explore the effects of AFDC (as well as other poli- cies, economic conditions, and health care availability) on both abortions and births. Like Blank et al. (1996), they use state and year dummy variables on aggregate rates (not disaggregated by age or race). Their results for abortions are similar to those of Blank et al. (1996~. AFDC payments significantly increase abortions in a specification without dummy variables, but with dummy variables (with or without state-specific linear time trends), there is no effect of AFDC on abortions. On births, they do find a significant positive effect of AFDC. Taken at face value, this result implies that the adjustment of births to AFDC does not occur through abortions. McKinnish and Sanders (1996) also use state and year dummy variables on aggregate data on births and abortions by state of occurrence. Unlike Blank et al. (1996) and Matthews et al. (1997), they use a nested logit model for aggregate data. Nevertheless, they also find no effect of AFDC on abortion and a highly significant effect on births. Finally, Joyce and Kaestner (1996) explore the effects of another component of the proreform welfare package health benefits through Medicaid on fertil- ity. They consider the effects of the Medicaid expansions of the late 1980s on abortion ratios using individual-level data from vital statistics reports on abor- tions in South Carolina, Tennessee, and Virginia. In addition to using cross-state differences in the timing of the expansion, they compare abortion ratios among women who were more or less likely to be eligible for Medicaid if they carried the child to term. Implicitly, they assume that any change in the relative abortion ratios of the more and less likely groups across the expansions (within a state) is due to the expansions. Joyce and Kaestner find large results. For white women 23 to 27 years old, they note that their estimates imply that the Medicaid expansions cut the abortion ratio by 2 to 5 percentage points (depending on which specification's estimates are used). This is equivalent to a 27 to 68 percent drop in the abortion ratio for Medicaid-eligible women. Even given the Long et al. (1994) estimate (cited by Joyce and Kaestner) that the value of the Medicaid expansion for prenatal care, delivery, and infant care is $6,850, these are large effects. This estimate of the cost to Medicaid of the expan- sions is likely to be an overestimate of the savings to mothers. Some of this care in
JACOB ALEX KLERMAN 125 particular, the large component due to the costs of childbirth would have been provided anyway, either in pubic hospitals or as charity care. Medicaid Funding of Abortions There is a moderate-sized literature exploring the effects of Medicaid fund- ing of abortions on abortion rates and birth rates. As discussed earlier, the extensive litigation and changing state policies provide a rich environment in which to explore policy effects. The classic study in the literature is Trussell et al. (1980~. They explored the time-series data on births and abortions around the end of Medicaid funding in Ohio and Georgia (using Michigan, which continued funding, as a control). They found that the number of abortions in Ohio and Georgia fell by more than a quarter (while abortions were approximately constant in Michigan), but that births were approximately constant in all three states. These results are consistent with the hypothesis that Medicaid funding does increase abortion rates. They are also consistent with the conclusion that women adjust contraceptive practice in re- sponse to changes in the price of abortion. A more recent study of the end of Medicaid funding in Michigan in 1988 comes to the opposite conclusion with respect to fertility (Evans et al., 1993~. Compared to the controls (Ohio and Indiana), births in Michigan rose, though by less than abortions fell. Again, funding affects abortion rates. This result sug- gests that changes in welfare policy would have effects on both contraceptive practice and abortion rates. Blank et al. (1996) also consider the effect of Medicaid funding. They find that Medicaid funding raises abortion rates in analyses of the state of occurrence data. This result washes out in analyses of state-of-residence. Medicaid funding should be a function of state of residence. Funding should not be available to women who have abortions in states with a funding policy if their state of resi- dence does not fund. Similarly, Medicaid should fund even if the abortion is performed out of state. Their strong negative effects of Medicaid funding in bordering states is, therefore, also anomalous. They conclude that the Medicaid funding results are spurious. Matthews et al. (1997) also find a positive Medicaid funding effect on abor- tions. Their effect, however, becomes statistically insignificant when state- specific time trends are included. Levine et al. (1996) perform similar computations on several of the Medicaid funding changes. With respect to the 1981 Supreme Court ruling allowing states not to fund abortions through Medicaid, they find a drop in abortions, but also an increase in births. More careful consideration of which states to include as controls leads them to conclude that the change in births is spurious (see their discussion of Ohio versus Pennsylvania, pp. 12-13; and Texas versus Colorado and Michigan discussed in footnote 20, p. 14~.
26 WELFARE REFORM AND ABORTION These results are robust in their multivariate analyses. Medicaid funding raises abortions by 1-1.5 per 1,000 women. The larger effect is in models without state-specific time trends; the smaller effect, in models with them. In models without state-specific time trends, Medicaid funding raises births (against the theory), but this effect washes out with the inclusion of state-specific time trends.~9 Haas-Wilson (1996) applies both ordinary least squares and state dummy variables (without year dummy variables) to explore the effect of abortion restric- tions on the abortion rate of minors using the CDC data. In both specifications, she finds a strong positive effect of Medicaid funding on abortions. Korenbrot et al. (1990) examine both births and abortions in three states that changed their funding policies between 1984 and 1985 (Colorado, South Caro- lina, and Pennsylvania). They conclude that ending funding lowers abortions and raises fertility, with the absolute value of the effect usually being larger on abor- tions. Disaggregated analyses for Colorado suggest that the birth effects are largest for blacks, but the North Carolina results do not support that suggestion. Changes in Abortion Versus Changes in Contraception We have noted earlier that welfare reform is likely to induce women who formerly would have had children and received AFDC to instead attempt to avoid births. For the concerns of this chapter, the crucial question is the extent to which they do so through lower coital frequency and more effective contraception or through abortion once they become pregnant. Papers that explore both births and abortions provide an opportunity to explore the relative importance of changes in abortion and changes in contraception for other policies that change birth rates. As noted in our theoretical discussion, there are two experiments. The first involves changes in the perceived value of children. The second involves changes in the perceived total cost of an abortion. The results for changes in the perceived value of children are striking. Ameri- can fertility has a pronounced (highly statistically significant) procyclical pattern (e.g., Silver, 1965; Jackson and Klerman, 1996; Black et al., 1996~. Matthews et al. (1997) also find this pattern in their regression coefficients for male and female wages (measured using CPS data on workers, with and without standard selection corrections). Nevertheless, their point estimates for the effect on abor- tion is positive, but statistically insignificant. In some of their alternative speci- fications, this positive effect of business cycle conditions is statistically signifi ~ r, 1>They also report results using the NLS-Y. Those results are consistent with their aggregate data results. Those data allow them to identify women who are more likely to be Medicaid eligible. We are inclined to downweight those results due to the small sample sizes, the misreporting of abortions, and the absence of any dummy variables.
JACOB ALEX KLERMAN 127 cant (i.e., per capita total personal income). This positive point estimate is consistent with the results in Blank et al. (1996) both for state of occurrence and for state of residence.20 Such a positive point estimate is not consistent with changes in births with the business cycle being achieved through abortions. The results with respect to abortion policy and access are also consistent with this conclusion. Matthews et al. (1997) find a negative effect of Medicaid abor- tion funding decisions in their models with dummy variables (it is only signifi- cant in their models without state-specific time trends). The effect on births is, however, also negative (again significant in models without state-specific time trends, insignificant in models with state-specific time trends). If there was no effect on contraception, we would expect abortions to rise. Similar results appear for parental notification/consent statutes. Consistent with the theory, abortions fall, but not consistent with the theory, births also fall. Again the results are only significant in the models without state-specific time trends. Their results for access to abortion providers also have a similar pattern. Access to abortion providers (by several measures) is strongly associated with more abortions (significantly both in models without and in models with state- specific time trends). The effect on births has the theoretically expected sign (negative) in the no-state-specific time-trends model, but it is much smaller in absolute value (in the log scale and also when transformed to rates). In the state- specific time-trends models the effect is positive, but insignificant. Finally, in empirical tests of their theory, Kane and Staiger (1996) also find evidence that adjustments are through contraception. Kane and Staiger' s theory suggested that if women use the availability of abortion to determine whether a father would support a child, then both abortions and fertility would rise as abortion became more available. More women would become pregnant, some of them would abort, some would deliver. In net, easier access to abortion could increase both births and abortions. They test this hypothesis using time series of cross-sectional data on fertility at the county level. Their fixed-effects results for 20This is not Blank et al.'s interpretation of their results. They discuss their economic results as follows: "Once state and year fixed effects are included, changes in economic or demographic variables over time within a state appear to have relatively small effects on state abortion rates. The strongest remaining effect is the positive relationship between unemployment and abortion rates. As the economy moves into recession, a 1-point rise in the unemployment rate leads to about a 3 percent increase in abortion rates. Estimates with slightly less precision but still significant at the 5 percent level, decreases in marriage rates and increases in per capita income lead to an increase in abortion rates." Note, however, that the unemployment rate result is not robust to the switch to state-of- residence data, but the positive per capita income result is robust and grows in statistical signifi- cance. Lundberg and Plotnick (1995) also explore Medicaid funding. They find that funding lowers abortion rates and raises birth rates. For the reasons discussed in the previous paragraph, we are inclined to downweight these results.
28 WELFARE REFORM AND ABORTION all three of their abortion access measures Medicaid funding, parental consent laws, and distance to an abortion clinic are all consistent with the hypothesis that births increase as abortions access improves. This could only happen if women are adjusting their contraceptive practices in response to abortion access. A simple adjustment of abortion conditional on fertility would imply that fertility would fall with improved abortion access. Consideration of the timing of the effects suggests that the Medicaid funding and parental consent results may be spurious, but the results based on distance to an abortion clinic are consistent with their theory. In all three cases, against the conventional wisdom (e.g., Hofferth, 1987), the implication is that rather than adjust abortions, women adjust contraceptive pat- terns. Note, however, that the effect on abortions itself is sensitive to specifica- tion. Blank et al. (1996) do not find any effect of parental notification/consent laws, and their effect of Medicaid funding washes out in the models using the preferred state-of-residence concept (which is what Matthews et al., 1996, use). Similarly, the preponderance of evidence in the Medicaid funding literature is that abortions change in the expected direction, but if there is any effect on births (at least in the theoretically expected direction), it is much smaller. Thus, these results should be interpreted with caution. More faith should be put in results based on the economic conditions. They appear to be robust (but see footnote 20~. CONCLUSIONS Congress has identified the effect of PRWORA on abortions as an important indicator of its success. Reducing nonmarital fertility is an explicit goal of the legislation. If there is a reduction in nonmarital fertility but it comes as a result of increases in the number of abortions, many of the legislation's proponents will judge the legislation to have been a mistake. Thus, evaluation of the net effect of the legislation and developing additional rounds of welfare reform will require information on the causal effect of the legislation on the number of nonmarital births and abortions. For some, the concern will be relative: in achieving any decline in nonmarital fertility, what was the relative importance of changes in abortion versus changes in coital frequency and contraceptive practice? This chapter has considered the likely effects of PRWORA on abortion and the research considerations in exploring that issue. While it is possible to con- struct theories with alternative implications, it seems reasonable to assume that, inasmuch as PRWORA decreases fertility, it will do so partially by increasing the number of abortions. Finding themselves pregnant and facing a less generous welfare option, more women will choose to abort rather than give birth. This is true despite the fact that the theory also suggests that a less attractive welfare option is likely to make women more aggressive contraceptors. We then also note that the data on abortion are poor. Survey data on abortion
JACOB ALEX KLERMAN 129 are unreliable and the sample sizes are too small to allow the estimation of policy effects. The AGI data, the best aggregate data available, lack covariate informa- tion. The AGI provider survey data are the most important data resource for the study of effects on abortion. The CDC abortion surveillance data allow partial corrections to those data for abortions to nonresidents and estimates of effects disaggregated by covariates. Continuation of both of these data collection efforts will be crucial for any attempt to evaluate the effects of welfare reform on abortion. An additional analysis strategy worthy of further consideration is analy- sis of those states that release individual-level records on abortions. For at least four states, those data appear to be nearly complete and to include out-of-state abortions (see Joyce and Kaestner, 1995, 1996~. The chapter has also reviewed the existing empirical literature and noted its sensitivity to the exact specification used. State dummy variables effects are crucial. The inclusion of state-specific time trends (which on a priori grounds are appropriate) makes estimated effects smaller. Also, there is some evidence for heterogeneity of effects by demographic group (which is difficult to explore using abortion data). Given these caveats, the existing literature finds no effect of AFDC pay- ments on abortion. This is true even when using the state and year dummy variables approaches that do find effects on births. This discrepancy between the fertility and abortion results might be real (welfare affects contraception, not abortion), or it could also be explained by the poorer quality of the abortion data (larger measurement error in the dependent variable leading to less precise esti- mates of regression coefficients). The literature does find significant effects of Medicaid funding on abortions (in the expected direction). There is some controversy about whether those effects are causal. At this point, a causal interpretation seems appropriate. Whether there is an effect on births remains an open question. Some studies find, against the theory, that births increase with Medicaid funding of abortion. Finally, the evidence is mixed on the extent to which adjustments to fertility occur through contraception or through abortions. The Medicaid funding litera- ture seems to suggest an effect on abortions. The welfare literature and the economic conditions literature seem to suggest an effect through contraception. The difference may partially be due to differential effects by subpopulation. ACKNOWLEDGMENTS This work was partially supported RAND and the National Institute of Child Health and Human Development (NICHD) (under Grants R01 HD31203 and P50 HD12639~. This chapter reflects on every page the input of my co-principal investigator on one of the NICHD-sponsored projects, Catherine A. Jackson. Most of the ideas discussed here were developed as part of our joint research. The comments of the discussant, Jacqueline Darroch, significantly improved this
30 WELFARE REFORM AND ABORTION chapter. As usual, the responsibility for any opinions or errors remains solely with the author. REFERENCES Akerlof, G.A., J.L. Yellen, and M.L. Katz 1996 An analysis of out-of-wedlock childbearing in the United States. Quarterly Journal of Economics 11(2):277-317. Alan Guttmacher Institute (AGI) 1979 Abortions and the Poor: Private Morality, Public Responsibility. New York: Alan Guttmacher Institute. 1994 Sex and America's Teenagers. New York: Alan Guttmacher Institute Bauer, G. and P. Gramm 1995 Why pro-lifers should support welfare reform. The Wall Street Journal (August 30): 10. Black, D., D. Kermit, and S. Sanders 1996 How much does economic growth help the poor: Evidence from the Appalachian Coal Boom and Bust. Heinz School Working Paper. Pittsburgh, Pa.: Carnegie Mellon University. Blank, R., C. George, and R. London 1996 State abortion rates: The impact of policy, provider availability, political climate, demog- raphy and economics. Journal of Health Economics 15(5):513-553. Centers for Disease Control (CDC) 1997 Abortion surveillance: Preliminary Analysis United States, 1995. Morbidity and Mor- tality Weekly Report 46(48). Currie, J., L. Nixon, and N. Cole 1996 Restrictions on Medicaid funding of abortion: effects on birth weight and pregnancy resolutions. Journal of Human Resources 31(1):159-188. Dash, L. 1989 When Children Want Children: The Urban Crisis of Teenage Childbearing. New York: William Morrow. Ellwood, D. 1987 Understanding Dependency: Choices, Confidence, or Culture. Report prepared for U.S. Department of Health and Human Services (Executive Summary), Center for Human Resources, Heller Graduate School, Brandeis University. Ellwood, D., and M. Bane 1985 The impact of AFDC on family structure and living arrangements. Research in Labor Economics 7: 165-195. Evans, M., E. Geicher, E. Feingold, M. Johnson, and R. Sokol 1993 The fiscal impact of the Medicaid funding ban in Michigan. Obstetrics and Gynecology (October) :555-560. Geronimus, A.T., and S. Korenman 1992 The socioeconomic consequences of teen childbearing reconsidered. Quarterly Journal of Economics 107(4): 1187-1214. Grady, W.R., M. Hayward, and J. Yagi 1986 Contraceptive failure in the United States: estimates from the 1982 National Survey of Family Growth. Family Planning Perspectives 18:200-209. Grady, W.R., M.B. Hirsch, N. Keen, and B. Vaughan 1983 Contraceptive failure and continuation among married women in the United States, 1970- 75. Family Planning Perspectives 14(1):9-19. Grady, W.R., M. Hayward, J. Billy, and F. Florey 1989 Contraceptive switching among currently married women in the United States. Journal of Biosocial Science 20:143-156.
JACOB ALEX KLERMAN 131 Gruber, J. 1994 The incidence of mandated maternity benefits. American Economic Review 84:622-641. Haas-Wilson, D. 1996 The impact of state abortion restrictions on minor's demand for abortions. Journal of Human Resources 31(1):140-158. Henshaw, S., and K. Kost 1992 Parental involvement in minors' abortion decisions. Family Planning Perspectives. 24:196-207, 213. 1996 Abortion patients in 1994-1995: Characteristics and contraceptive use. Family Planning Perspectives 28(4):140-147,158. Henshaw, S., and J. Silverman 1988 The characteristics and prior contraceptive use of U.S. abortion patients. Family Planning Perspectives 27:120-122. Henshaw, S., and J. Van Vort, eds. 1992 Abortion Factbook, 1992 Edition. New York: Alan Guttmacher Institute. Henshaw, S., and J. Van Vort 1994 Abortion services in the United States, 1991 and 1992. Family Planning Perspectives 26: 100- 106. Henshaw, S., L.M. Koonin, and J.C. Smith 1991 Characteristics of women having abortions, 1987. Family Planning Perspectives 23:75-81. Hofferth, S. 1987 Contraceptive decision-making among adolescents. In S. Hofferth and C. Hayes, eds., Risking the Future: Adolescent Sexuality, Pregnancy, and Childbearing, (Volume II), Washington, D.C.: National Academy Press. Jackson, C.A., and J.A. Klerman 1996 Welfare and American fertility. National Bureau of Economic Research, Summer Insti- tute, Cambridge, Mass., July, 1995. Revised October 1996. Jones, E.F., and J.D. Forrest 1989 Contraceptive failure in the United States: Revised estimates from the 1982 national survey of family growth. Family Planning Perspectives 21(3):103-109. 1992a Underreporting of abortion in surveys of U.S. women: 1976 to 1988. Demography 29:113-126. 1992b Contraceptive Failure Rates Based on the 1988 NSFG. Family Planning Perspectives 24(1): 12-19. Joyce, T., and R. Kaestner 1995 The Effect of Parental Involvement Laws on Pregnancy Resolution. Presented at the 1995 Population Association of America Annual Meetings, San Francisco, April 6. 1996 The effect of Medicaid income eligibility expansions on abortion. Demography 33(2): 181- 192. Kane, T., and D. Staiger 1996 Teen motherhood and abortion access. Quarterly Journal of Economics 111(2):467-506. Keating, D. 1990 Adolescent thinking. Pp. 54-92 in S. Feldman and G. Elliott, eds., At the Threshold: The Developing Adolescent. Cambridge, Mass.: Harvard University Press. Klerman, J.A. 1996 Fertility effects of Medicaid funding of abortions: A disaggregated analysis. Mimeo, RAND, Santa Monica, Calif. Kochanek, K.D. 1989 Induced terminations of pregnancy: Reporting states, 1987. Monthly Vital Statistics Re- port 38(9, suppl.), Hyattsville, Md.: U.S. Public Health Service.
32 WELFARE REFORM AND ABORTION Koonin, L.M., J.C. Smith, and M. Ramick 1995 Abortion surveillance United States, 1991. CDC Surveillance Summaries, Morbidity and Mortality Weekly Report 44:SS-2. Korenbrot, C.C., C. Brinis, and F. Priddy 1990 Trends in live births and abortions following state restrictions on public funding of abor- tions. Public Health Reports 105:6, 555-562. Kost, K., and J.D. Forrest 1989 American women's sexual behavior and exposure to risk of sexually transmitted disease. Family Planning Perspectives 6 (November-December):244-254. Levine, P., A. Trainor, and D. Zimmerman 1996 The effects of Medicaid abortion funding restrictions on abortions, pregnancies, and births. Journal of Health Economics 15:555-578. Long, S.H., M.S. Marquis, and E.R. Harrison 1994 The costs of financing of perinatal care in the United States. American Journal of Public Health 84:1473-1478. Luker, K. 1979 Taking Chances: Abortion and the Decision Not to Contracept. Berkeley, Calif.: Univer- sity of California Press. 1996 Dubious Conceptions: The Politics of Teenage Pregnancy. Berkeley, Calif.: University of California Press. Lundberg, S., and R. Plotnick 1990 Effects of state welfare, abortion, and family planning policies on premarital childbearing among white adolescents. Family Planning Perspectives 22:246-251. Adolescent premarital childbearing: Do economic incentives matter? Journal of Labor Economics 13:177-201. Maddala, G.S. 1983 Limited-dependent and Qualitative Dependent Variables in Econometrics. Cambridge, England: Cambridge University Press. Matthews, S., D. Ribar, and M. Wilhelm 1997 The effects of economic conditions and access to reproductive health services on state abortion and birth rates. Family Planning Perspectives 29(2):52-61. McKinnish, T., and S. Sanders 1996 Who is right: The Pope or Newt Gingrich? Mimeo, Carnegie Mellon University, Pitts- burgh, Pa. Meter, M.H., and D.R. McFarlane 1993 State Family Planning and Abortion Expenditures: Their Effects on Public Health. Ameri- can Journal of Public Health 84:1468-1472. Merz, J.F., J.A. Klerman, and C.A. Jackson 1995 A review of abortion legality, Medicaid funding, and parental involvement laws, 1967- 1994. Rutger's Women's Law Reporter. 1996 A chronicle of abortion legality, Medicaidfunding, and parental involvement laws, 1967- 1994. Labor and Population Program Working Paper 96-08. RAND, Santa Monica, Calif. Meyer, B. 1995 Natural and quasi-natural experiments. Journal of Business and Economic Statistics 13(2): 151-162. Moffitt, R. 1992 Incentive effects of the U.S. welfare system. Journal of Economic Literature 30(1): 1-61. 1994 Welfare effects on female headship with area effects. Journal of Human Resources 29(2):621-636. Moore, K., and S. Caldwell 1977 The effect of government policies in out-of-wedlock sex and pregnancy. Family Planning Perspectives 9:164-169.
JACOB ALEX KLERMAN 133 Moore, K., C. Winquist, and J.L. Peterson 1989 Nonvoluntary sexual activity among adolescents. Family Planning Perspectives 21(3): 110-114. Musick, J.S. 1993 Young, Poor, and Pregnant: The Psychology of Teenage Motherhood. New Haven, Conn.: Yale University Press. Olasky, M. 1995 Dismantle the poverty panopticon: Pro-life, anti-welfare. The Wall Street Journal (March) 22:16. O'Steen, D.N. 1995 Welfare 'reforms' pose threat to unborn babies. National Right to Life News (February 22):1 If. Petersen, A. 1988 Adolescent development. Annual Review of Psychology 39:583-607. Pittman, K., and C. Govan 1986 Model Programs: Preventing Adolescent Pregnancy and Building Youth Self-Sufficiency. Washington, D.C.: Children's Defense Fund. Plotnick, R. 1993 Welfare and Out-of-Wedlock Childbearing: Evidence from the 1980s. Journal of Mar- riage and the Family 52:735-746. Richard, Bishop John 1997 The factors of genuine welfare reform. Origins 24(34):564-566. Silver, N. 1965 Births, marriages, and business cycles in the United States. Journal of Political Economy 73:237-255. Singh, S. 1986 Adolescent pregnancy in the United States: An interstate analysis. Family Planning Perspectives 18(5):210-220. Trussell, J., J. Menken, B.L. Lin&eim, and B. Vaughan 1980 The impact of restricting medicaid financing for abortion. Family Planning Perspectives 12(3): 120- 130. U.S. Congress, Committee on Ways and Means 1996 Summary of the Welfare Reforms Made by Public Law 104-193: The Personal Responsi bility and Work Opportunity Act and Associated Legislation. WMCP 104-15, Washing- ton, D.C., November 6. U.S. Department of Health and Human Services 1995 Report to Congress on Out-of-Wedlock Childbearing. DHHS Pub. No. (PHS) 95-1257. Hyattsville, Md. Ventura, S., J. Taffel, W. Mosher, J. Wilson, and S. Henshaw Trends in pregnancies and pregnancy rates: Estimates for the United States, 1980-92. Monthly Vital Statistics Report 43(11, Suppl.) Hyattsville, Md.: National Center for Health Statistics. Willis, R.J. 1995 A Theory of Out-of-Wedlock Childbearing. Presented at The Symposium on the Eco- nomic Analysis of Social Behavior Convened by the Fraser Institute on the Occasion of Gary Becker's 65th Birthday, Chicago. Zabin, L. Schwab, and S. Clark, Jr. 1981 Why they delay: A study of teenage family planning clinic patients. Family Planning Perspectives 13(5):205-207, 211-217. Zelnik, M., and F. Shah 1983 First intercourse among young Americans. Family Planning Perspectives 15(2):64-70. Zelnik, M., and K. Ford 1981 Sex and Pregnancy in Adolescence. Beverly Hills, Calif.: Sage Publications.
