The effect of family planning programs on the fertility of welfare recipients: evidence from Medicaid claims.

I. Introduction

This paper examines the effectiveness of family planning services subsidized by the Medicaid program in lowering the fertility of women supported by public assistance. Family planning services have been a mandatory Medicaid benefit in all states for more than 20 years. Relatively inexpensive to administer, these services have the potential for large savings, depending on how much they reduce the fertility of the low-income women who are eligible to receive them. In past attempts to measure the effects of public family planning programs, a number of studies have been unable to find the expected negative effect on fertility, suggesting that the relationship between public expenditures and fertility may not be simple or obvious.

One of the most common shortcomings of the previous literature is the use of single-equation models to estimate the effects of family planning programs on fertility. Single-equation models are unable to account for the role of unmeasurable factors in both the decision to adopt family planning services and the likelihood of giving birth. A number of unobserved variables can affect both contraceptive acceptance and fertility, and, depending on the correlation of the unobservable factor with each outcome, may result in an estimate of contraceptive effectiveness that is biased either upward or downward. For example, unobserved preferences for children may act to increase other activities that make conception more likely and tend to decrease the use of family planning methods. When these preferences are not included in a model of fertility, the (negative) effect of contraceptive use will be biased upward in absolute value. Another factor that is frequently unmeasured in various data sets is the individual's level of sexual activity. Since both fertility and contraceptive use are likely to increase with sexual activity, this unobservable factor will result in an estimate of contraceptive effectiveness that is biased toward zero. An additional source of downward (toward zero) bias in the estimation of contraceptive effectiveness was noted by Rosenzweig and Schultz (1985) as arising from the inability to measure a couple's fecundity, or biological capacity to bear children.

The estimation of an unbiased effect of using family planning requires a methodology to control for possible correlation in the unobservables that affect both contraceptive acceptance and childbirth. In this paper, such a methodology is applied to a unique source of microdata that allows for the estimation of the effect of Medicaid-specific family planning services on childbirth within the Aid to Families with Dependent Children (AFDC) population. The sample is created from a census of all women on public assistance or Medicaid who gave birth or received Medicaid-provided family planning services in Maryland over a period of several years. The results of this paper suggest that a positive correlation between the unobservables affecting both contraceptive practice and fertility produces underestimates of contraceptive effectiveness in single-equation models. Controlling for that correlation shows that publicly funded contraceptives are much more effective in reducing births to women on welfare than single-equation models would suggest. Among women in the sample who accept contraceptives from a Medicaid provider, there is a subsequent 7.2 percentage point decrease in the probability of giving birth.

II. Past Estimates of the Effects of Publicly Funded Family Planning

Forrest and Samara (1996) suggest that every dollar spent by states and the federal government on Medicaid-subsidized contraceptives results in a savings of three dollars in Medicaid costs for health care for pregnant mothers and newborns. This estimate rests on various assumptions about contraceptive behavior and fertility in the absence of public programs. An alternative methodology for estimating program effects and imputing savings is multiple regression analysis, which has been used in numerous studies assessing impacts of public family planning programs on birth rates.

In Appendix 1, I contrast a number of U.S. studies by type of data used, years studied, measurement of family planning programs, and results. Many previous studies rely on cross-sectional data; most define the dependent variable as either the teen birth rate in a cross-section of states or counties or, in samples of individuals, the probability that a teen gives birth. Although teen births and birth rates are important to states funding family planning services, teens represent only a small subset of all women eligible for public family planning services, so that applying these findings to the population of Medicaid or AFDC recipients may not be appropriate. This is especially true in the case of the Cutright and Jaffe (1977) study, where the sample includes only married women, a group with very different characteristics than public assistance recipients.

The results of the studies summarized in Appendix I vary quite markedly. A number of studies find no significant effects of family planning programs on fertility (Winegarden 1974: Udry, Bauman, and Morris 1976; Meier and McFarlane 1994). Several other studies find significant program effects, which, in some cases, are quite large (Moore and Caldwell 1977; Brann 1979; Lundberg and Plotnick 1990; Davis, Olson, and Warner 1993). Additionally, a number of studies produce some rather surprising effects, which may signal potential weaknesses in the specification of these models. Among the puzzling results, Darney (1975) finds that for some groups higher acceptance rates of oral contraceptives and intrauterine devices (IUDs) are positively associated with fertility rates. Forrest, Hermalin, and Henshaw (1981) find a positive relationship between clinic enrollment and teen birth rates in a cross-section of counties in 1970.(1) In Singh (1986), the estimated effect of the percentage of teens served by family planning programs on the teen birth rate has the expected negative sign, yet in some models of pregnancy rates, Singh finds a positive and significant effect of the family planning variable.

Most of these previous studies are dated. With only one exception, they examine data for 1980 or earlier. Whether significant relationships for that time period hold for the decade of the 1980s is questionable. Sharp changes in attitudes regarding sexual behavior and out-of-wedlock births, as well as important advances in the technology and provision of family planning services, have occurred since 1980.

The often counterintuitive results that appear in these studies suggest a further limitation in the research. All of the family planning studies reviewed here used single-equation models of the birth rate or the probability of giving birth, which implicitly assume that the decision to obtain family planning services is exogenous to the probability of giving birth. Theoretical analysis of fertility has suggested, however, that family planning program variables are not exogenous when measured either at the individual level or at the aggregate level, and has shown that not correcting for endogeneity can lead to underestimates or overestimates of program effectiveness.

Borrowing from the model specification of Schultz (1990), one can represent the fertility (F) of individual i as:

(1) [F.sub.i] = [a.sub.o] + [a.sub.i][E.sub.i] + [a.sub.2][X.sub.r] + [a.sub.3][X.sub.f] + [a.sub.4][P.sub.i] + [a.sub.5][B.sub.i] + [e.sub.i],

where [E.sub.i] represents household characteristics such as income and age, [X.sub.r] is a set of variables defined over regions, [X.sub.f] represents family planning inputs, [P.sub.i] represents preferences for children, and [B.sub.i] represents the individual's supply of fertility, namely fecundity or the biological capacity to bear children. The role of family planning inputs is to reduce fertility, but the estimated effect of the family planning input may be biased when preferences and fecundity are unobserved by the researcher. Fecundity and preferences both may be correlated with family planning use, and when absorbed into the error term they can bias the estimated negative effect of family planning either downward toward zero or upward in absolute value. Rosenzweig and Schultz (1985) found that using an instrumental variables (i.v.) approach to account for the fact that more fecund couples are more likely to use contraceptives allows for the consistent unbiased estimation of the family planning effectiveness coefficient. Their work showed that the coefficient obtained from i.v. estimation is greater in absolute value than that obtained in single-equation models.

Suppose that the family planning input is instead a measure of resources allocated by the regional or local government for the administration of a family planning program, as with Meier and McFarlane (1994). The use of an aggregate measure allows the plausible assumption that the government allocation of family planning program resources is independent of the individual's fecundity. However, the aggregation of Equation I creates the potential for another source of bias in the …