Does Access to Family Planning Increase Children’s Opportunities?

II. A History of U.S. Family Planning Programs and Their Expected Effects on Children’s Outcomes

A. Background

The architects of President Lyndon B. Johnson’s War on Poverty viewed the price of reliable contraceptives as a barrier to reducing poverty and promoting children’s economic well-being. Of particular concern was that the Pill, the first oral contraceptive, was prohibitively costly in the 1960s. An annual supply of oral contraception sold for the equivalent of $812 in 2013 dollars (Bailey 2013)—roughly twice today’s annual cost and equivalent to more than three weeks of full-time work at the 1960 minimum wage (without factoring in the cost of visiting a physician). Numerous studies documented at the time that poor women were less likely to use effective contraceptives. We calculate in the 1965 National Fertility Study (NFS), for instance, that poor women were 44 percent (0.082 percentage points/0.185, SE 0.023) less likely to use the Pill than more affluent women. They also had roughly 0.60 (SE 0.16) more children on average than families with higher incomes.

To reduce disparities in access to contraception, the Office of Economic Opportunity (OEO) began funding family planning programs as early as 1964. OEO grants rose to roughly $427 million (2013 dollars) by 1970 and supported the opening of new clinics in disadvantaged areas and, to a lesser extent, the expansion of existing family planning programs. In November 1970, Congress passed Title X of the Public Health Service Act (also known as the Family Planning Services and Population Research Act, P.L. 91-572). By 1974, this legislation had increased federal support by 50 percent in real terms over 1970 levels. (See Bailey 2012, 2013 for a detailed history of these programs and federal funding.)

The aim of these programs was to bring education, counseling, and the provision of low-cost contraceptives and related medical services to disadvantaged women. Family planning grants did not fund abortion, which was illegal before 1970 except in special circumstances. Even after abortion was legalized in several states and nationally in 1973, Title X explicitly prohibited the use of federal funds “in programs where abortion is a method of family planning” (Section 1008). During these early years, little is known about these programs’ day-to-day operations. The federal government did not collect information on their services or patients, and officials talked very little about them—perhaps because topics like birth control were taboo.³ This dearth of evidence limits detailed exploration of the mechanisms of our estimated effects. To the extent that family planning programs varied in their implementation, the effects we estimate will represent the weighted combination of many services and types of programs, all of which provided reduced-cost contraceptives and related services.

Between 1964 and 1973, federally funded family planning programs were established across the United States, providing the identifying variation for this study. Figure 1 shows the rollout of these family planning programs at the county level. The counties with the earliest programs, established between 1964 and 1967, are shaded in light gray; the counties with programs established between 1968 and 1969, during the expansion of family planning as a national emphasis program, are medium gray; and the counties with programs established from 1970 to 1973, under Title X, are black. By 1973, federal funding had initiated or substantially expanded over 660 family planning programs across all of the lower 48 states. These programs funded services in locations where roughly 56 percent of the U.S. population of women ages 15–44 lived.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

The Rollout of Federally Funded Family Planning Programs, 1965–1973

Notes: Dates are the year that the county first received a federal family planning grant. Counties not receiving a family planning grant between 1965 and 1973 are not shaded. Source: Bailey (2012).

III. The Expected Effects of Family Planning Programs on the Economic Resources of Children

The increased availability of family planning services could affect the economic and living circumstances of children through two main channels. The first is what we call the resource effect, and the second is the selection effect.

The resource effect refers to the situation in which family planning directly raises the incomes of parents. This would be the case if soon-to-be parents could use family planning services to delay childbearing in order to get more education, pursue different career paths, or obtain different amounts of work experience and job training. In addition, women are more likely to make different investments in their careers or stay attached to a job if they expect to be able to control future childbearing. If family planning programs allow older women to time births better, they could raise the labor-force participation of mothers, which could have large effects on household incomes of children once they are born.¹³

Evidence regarding the existence of this resource effect comes from the literature on legal access to the birth control pill for younger women. Using variation in state laws regulating the age of consent for (mostly) unmarried teens (Bailey et al. 2011; Guldi 2011), this literature shows that access to the Pill affected marital and birth timing and had lasting effects on women’s and men’s career investments. With earlier legal access to the Pill, women and men were more likely to enroll in and complete college (Hock 2008; Bailey, Hershbein, and Miller 2012), and women were more likely to work for pay, invest in on-the-job training, and pursue nontraditional professional occupations (Goldin and Katz 2002; Bailey 2006; Bailey, Hershbein, and Miller 2012)—investments that resulted in higher wages for women later in adulthood (Bailey, Hershbein, and Miller 2012). Ananat and Hungerman (2012) additionally show that access to contraception at younger ages improved the economic resources of children born to these women, although they provide no evidence on whether these gains accrued due to the selection or the resource effect.¹⁴

Family planning’s resource effect may also act through partnership decisions. For instance, family planning programs could reduce the price of delaying marriage (Goldin and Katz 2002) and improve spouse matching and reduce marital stress, thereby reducing subsequent divorce rates (Christensen 2011; Rotz 2016). If family planning programs increase the likelihood that parents remain together after a child is born, this will also serve to increase the economic resources available to children.¹⁵

The alternative to the resource effect is the selection effect. Selection refers to the channel through which family planning programs impact children’s circumstances by altering who becomes a parent. This shift in the composition of parents could improve the living circumstances of the average child even in the absence of any direct effect on parental income. For instance, if fewer children are born to lower income parents, the resources of the average child could rise even though the incomes of any child that is born would not.

Selection may be a particularly important channel in explaining these programs’ effects on children because programs in the 1960s disproportionately served poorer households. Roughly 83 percent of program patients had incomes below 150 percent of the poverty line, and 13 percent were recipients of Aid to Families with Dependent Children (AFDC, the principal cash welfare program at the time) (Jaffe, Dryfoos, and Corey 1973; Torres and Forrest 1985, 284).

Separating the resource and selection effects in cross-sectional, retrospective data is impossible to do directly because the two effects appear identical. Notably, although studies quantifying the impact of legal abortion on children’s outcomes in the United States attribute all of the gains to selection, these findings could also partially reflect a resource effect. Analyzing the staggered legalization of first-trimester abortion in the United States, initially in five states around 1970 and then in the remainder of states after Roe v. Wade in 1973, these studies show that birth rates fell by 4–8 percent (Levine et al. 1996) and that the outcomes for cohorts born after abortion was legalized improved. Gruber, Levine, and Staiger (1999) show that children born after abortion legalization were less likely to die as infants, live with single parents or in families receiving welfare, and live in poverty. Ananat et al. (2009) find that, later in life, these children were more likely to graduate from college, less likely to rely on welfare, and less likely to be single parents. Donohue and Levitt (2001) show that cohorts born after abortion legalization were less likely to commit crime,¹⁶ and Charles and Stephens (2006) show that these cohorts were less likely to use controlled substances in their late teens. One study of Romania explicitly considers the role of selection. When Romania experienced the reverse policy change—the prohibition of abortion, Pop-Eleches (2006) shows that birth rates among the most educated mothers (who had been using abortion and family planning services) increased by around 30 percent. After accounting for this positive selection of mothers, however, the paper shows that these unintended children fared worse in terms of their later socioeconomic outcomes.

In summary, the availability of family planning programs may increase family resources of both younger and older parents. In addition, family planning programs may also affect children’s measured outcomes by altering selection into parenthood. Bailey’s (2013) findings that family planning programs improved adult outcomes are consistent with both channels: either increases in parents’ incomes or changes in selection (or a combination of both) could explain the higher college completion rates and the 2 percent increase in income for the average adult child. Understanding the quantitative importance of the resource and selection channels is, therefore, crucial to interpreting these estimates. The next section describes this study’s empirical strategy to estimate these combined effects and bound the role of selection to recover an estimate of the direct effect of family planning programs on parents’ incomes (for example, the resource effect).

IV. Research Design and Data

The restricted, long-form samples of the 1970 and 1980 Censuses provide information on children’s economic resources and living circumstances. In addition to their large sample sizes (20 percent and 16-percent samples of the U.S. population for 1970 and 1980, respectively), these samples identify county of residence (information that is suppressed in public samples).¹⁷ Our analysis aggregates the economic resources and living circumstances of children under age 18 into birth-year–county cohorts.¹⁸ These birth-year–county cohorts are then linked to information on when their county first established a federally funded family planning program.¹⁹

Our primary specification exploits variation in the timing of the first federal family planning grants to identify these programs’ effects. In practice, our event-study framework contrasts the evolution of outcomes for cohorts born after family planning programs were first established in the county to those born before, and allows adjustments for a variety of potentially confounding factors:

(1)

where Y is a measure of the outcomes for children residing in county j within state s and born in calendar year t. 1( ) indexes birth cohorts relative to the year of the first federal family planning grant, T_j*—the year the program started.²⁰ We run separate regressions using the 1970 and 1980 Censuses in order to recover the regression-adjusted evolution of children’s outcomes for cohorts born from six years before and up to six years after each county received its first federal family planning grant for a balanced set of counties. In our analysis, event time, c, runs from a years before up to b years after the date of the first family planning grant, where the 1970 Census covers a = −7 through b = 1, and the 1980 Census covers a = −3 through b = 7. We do not pool the two Censuses because the 1970 data identify τ using only the 461 counties funded prior to 1970, whereas the 1980 data identify τ in the full set of funded counties. Both Censuses, therefore, independently estimate event years −2 to 0, and using both Censuses allows us to estimate τ using a balanced set of counties for the greatest range of event time.²¹

Our baseline specification includes θ, a set of county fixed effects that capture time-invariant county-level differences, and γ, a set of birth-year fixed effects and a set of state by birth-year fixed effects that capture time-varying changes in state policies, including abortion legalization and the rollout of Medicaid. We present unweighted results (Solon, Haider, and Wooldridge 2015). Standard errors are robust to heteroskedasticity and corrected for serial correlation within state (Arellano 1987; Bertrand, Duflo, and Mullainathan 2004).

The robustness of these baseline results are examined by sequentially including covariates used in other studies of War on Poverty programs. Covariates include annual information on per capita measures of government transfers from the Bureau of Economic Analysis Regional Economic Information System (REIS) (cash public assistance benefits such as Aid to Families with Dependent Children, Supplemental Security Income, and General Assistance; medical spending such as Medicare and military health care; and cash retirement and disability payments) (Almond and Currie 2011). In some specifications, we also include the number of abortion providers, which accounts for within-state changes in the provision of abortion between 1970 and 1979 (Bailey 2012).²²

An important challenge to our analysis is that the Censuses contain information on a child’s residence in (or five years before) the Census year—not residence at the time of the child’s birth. If mothers move between the time of birth and observation in the Census, this could lead us to misclassify mothers’ access to federal family planning programs around the time of conception. We take several steps to limit this mobility-induced misclassification. First, we use county of residence five years before the Census, because 1975 is more temporally proximate to the year of birth for cohorts of children identifying our parameters of interest (and, therefore, more strongly correlated with mother’s exposure to federal family planning programs) than 1980; we use 1965 for the 1970 Census for consistency. Second, we exclude unfunded areas from our estimation sample because of their differential mobility relative to funded areas after family planning programs started. Finally, we follow Card and Krueger (1992) to adjust for mobility using a postestimation correction.²³ (See Online Appendix B for a discussion and additional results related to corrections for mobility bias.)

V. The Estimated Effects of Family Planning Programs on the Economic Resources of Children

We begin by examining the effects of family planning programs on the log household income of the average child born. Figure 2 presents results from our baseline model, which includes county and state by birth-year fixed effects. Estimates to the left of the vertical axis represent cohorts born in event years before family planning programs began (1970 Census estimates plotted in dashed lines), and estimates to the right of the vertical axis capture cohorts born after family planning programs began (1980 Census, solid lines with markers). Dashed lines plot 95 percent, pointwise confidence intervals for this model.²⁴

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

The Effect of Family Planning Programs on Log Household Income of the Average Child

Sources: 1970 (dashed lines with markers) and 1980 (solid lines with markers) restricted-use Censuses. Notes: The figure plots the percent change in household income (dependent variable) for children born up to six years before and up to six years after family planning programs began using the baseline model from estimating Equation 1 (see also Table 2 and Online Appendix Table A4). The x-axis plots children’s birth year relative to the year the county’s family planning program started. As covariates, the baseline model includes county, birth year, and state by birth-year fixed effects. Standard errors have been clustered by county and used to construct 95 percent, pointwise confidence intervals for the baseline model (dashed lines).

Consistent with the resource and the selection effects operating as expected, Figure 2 shows that cohorts born after federal family planning programs began had significantly higher household incomes. Although the household income of birth cohorts was stable before family planning programs began, the introduction of family planning programs corresponds to a notable trend break. Table 2 demonstrates the robustness of these estimates to different control variables and mobility bias adjustments for the 1980 Census. Column 1 includes only county and year fixed effects, Column 2 (our baseline model and preferred specification) adds 850 state by year fixed effects (50 states × 17 birth years), Column 3 adds county-level covariates, Column 4 adds controls for abortion providers, and Column 5 adjusts our baseline model for mobility bias using the Card and Krueger (1992) postestimation correction. Across specifications, cohorts born five years after the programs began had household incomes that were very similar—2.7 percent higher in our baseline specification (Column 2). Consistent with mobility attenuating our estimates, the Card and Krueger (1992) mobility correction raises the estimate modestly to 3.0 percent (Column 5). In contrast, the coefficients on household incomes among cohorts born up to six years before family planning programs began exhibit little trend and are not statistically different from the year the county received its first family planning program.²⁵ Our baseline specification implies that the income of the average child was around $1,840 higher ($68,000 pre-program mean family income × 0.027) among children born after family planning programs began.²⁶

Because the restricted Census data are not top-coded, we also investigate the importance of outliers. Trimming the top and bottom 1 percent of children’s household incomes, however, has little effect on these estimates.²⁷ Another check using per-capita household income as a dependent variable shows that the effects are even stronger after taking account of the number of people in the household in 1980: the average child in cohorts born five years after the programs began had a statistically significant 3.5 percent higher per-capita household income (Online Appendix Table A5, Column 2).

Increases in the income of the average child should be driven by households in the lower part of the income distribution. As noted previously, 83 percent of family planning patients had incomes below 150 percent of the poverty line. Consistent with this prediction, Figure 3 and Table 3 present evidence of sharp reductions in measures of children’s economic disadvantage. To simplify the interpretation of the coefficients, Figure 3 plots estimates of τ divided by the pretreatment mean dependent variable for both the 1970 and 1980 estimates. The series, therefore, denotes changes in percent for each birth cohort indexed relative to the year the family planning program began. Table 3 presents estimates in levels and also summarizes the relevant pretreatment means.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

The Effect of Family Planning Programs on the Economic Disadvantage of the Average Child

Notes: Figure plots estimates of τ from estimating Equation 1 divided by the pretreatment mean dependent variable for both the 1970 and 1980 estimates. The series, therefore, denotes changes in percent of children in a birth cohort with a given characteristic relative to the cohort born in the year the family planning program began. See also notes for Figure 2 and estimates in Table 3 and Online Appendix Tables A6, A9, and A10.

The results show that cohorts born five years after federal family planning programs began were 7.4 percent less likely to live in poverty (Figure 3A; Table 3, Column 1), 6.4 percent less likely to live below 1.5 times the poverty line (Table 3, Column 2), and 4.3 percent less likely to live below two times the poverty line (Table 3, Column 3). As was the case with household income, Figure 3A shows no evidence that these reductions reflect a pre-trend. These results are consistent with federal family planning programs having the largest effects on lower income women.

Administrative statistics also suggest that nonwhite women were overrepresented among family planning patients (30 percent of family planning patients versus only 17 percent of women of childbearing age). This suggests that the intention-to-treat effects of family planning might be larger among nonwhite children. Consistent with this prediction, Table 3 shows that the absolute reductions in poverty rates were two times larger for the average nonwhite child born five years after the program began (−1.6, SE 1.6) than for the average white child (−0.8, SE 0.6). Owing to smaller samples, the event study estimates for subgroups are imprecise. To increase precision, we summarize the event-study estimates in a standard differences-in-differences framework. Using a single dummy variable for the affected cohorts, we find that poverty rates among nonwhite children fell by a statistically significant 5.5 percent (−2.0/34.8, SE 0.81, not reported in tables for brevity) over the first six years after family planning programs began. For white children, poverty rates fell by an insignificant 2 percent (−0.28/14, SE 0.29). For nonwhite children, the effects of family planning programs are also generally larger for lower poverty thresholds, although imprecision limits strong conclusions about comparisons across thresholds.

Of course, reductions in child poverty could occur mechanically if family planning reduced the number of children in the household (which could lower the poverty threshold), even if household incomes did not change (that is, the resource effect is zero).²⁸ We, therefore, repeat our analysis by subtracting out the younger siblings that arrived after each child was born and then reconstruct poverty thresholds for each child at the time of birth. Notably, this lower threshold generates even larger reductions in poverty rates attributable to family planning, suggesting that subsequent births in the household reduced the estimated effects of family planning. This evidence foreshadows one of the paper’s main conclusions that the resource effect plays an important role in explaining the improvement in children’s living circumstances.

The introduction of family planning programs is also associated with a reduction in another measure of disadvantage: the share of children living in families receiving public assistance. Children born five years after family planning programs began were on average 12 percent less likely to live in households receiving public assistance (Figure 3B, Table 4, Column 1) than those born just before family planning programs began. While the magnitude of the coefficients suggests that the reductions mainly occurred among whites, the confidence intervals are too large to conclude that the effect of family planning on receiving public assistance differs by race. White cohorts born five years after family planning programs began were 10 percent less likely to live in households receiving public assistance (Table 4, Column 2). Although a greater share of nonwhite children lived in households receiving public assistance, the event-study estimates for nonwhite children are much less precise and noisier, exhibiting positive- to negative-signed coefficients (Column 3).

A final measure of children’s disadvantage is living in a household with a single head. Nonmarital childbearing and single headship rose dramatically in the 1960s and 1970s (McLanahan and Watson 2011; Bailey, Guldi, and Hershbein 2014), and both trends have served to increase child poverty over the longer term. If family planning programs encouraged these trends by reducing the cost of nonmarital sex, they could have increased nonmarital childbearing (Akerlof, Yellen, and Katz 1996). On the other hand, family planning programs could have dampened the longer term increase in nonmarital childbearing due to the reduced unwanted pregnancies among less committed couples.

Figure 3C shows evidence of the latter. Following the introduction of family planning programs, the share of children born to single heads decreased. Although the 95-percent confidence intervals in Figure 3C (Table 4, Column 4) include zero, a one-sided test rejects the hypothesis that the introduction of family planning programs increased the share of children living with single parents. As in the case of public assistance, these patterns are less noisy and more precisely estimated for white children relative to nonwhite children (Table 4, Columns 5 and 6). To investigate this result further, we conduct a complementary analysis of county-level Vital Statistics for marriage and divorce. Using marriages and divorces per woman ages 15–44 as the dependent variables in the same specification (and replacing cohort with year), we fail to reject that marriage and divorce rates remained the same after the introduction of family planning programs (Online Appendix Figure A4).

Two final analyses help explain how family planning programs affected these outcomes. Using both the age of mothers at the time of each birth and the number of each child’s older siblings (a measure of each child’s birth order²⁹) as dependent variables, we find that the age of the average child’s mother fell by around 0.3 years by Year 5, a reduction of 3.6 months (Table 5, Column 1). This finding is a weighted average of the effects of family planning programs on childbearing delay (which should increase the age of the average child’s mother) and women ending their childbearing careers sooner (often called stopping, which should decrease the age of the average child). Although births to older women contribute relatively less to the overall birth rate than births to younger women, the potential reduction in age by avoiding an unwanted pregnancy in one’s thirties is much larger (5–10 years on average) than the increase in age by delaying a pregnancy (1–3 years on average). On net, these opposing changes average out to a small negative number, suggesting that the share-weighted reduction in age was slightly larger than the share-weighted delay in birth.

The reduction in older siblings is also consistent with family planning programs helping older women avoid unplanned births. Children born five years after family planning programs began had significantly fewer older siblings, implying that they were more likely to be lower-order births (first or second children rather than third or higher) (Table 5, Column 4). Evidence using changes in parity-specific birth rates reinforces this impression. Bailey (2012, Table 4B) shows that family planning programs did not affect first birth rates in the longer term (Years 11–15), but they did reduce third-parity births for up to 15 years after they were established—a result consistent with family planning programs affecting stopping behavior. Changes in the living circumstances of the average child, therefore, appear to be driven largely by older mothers having fewer children. Because wages tend to increase with age, this makes the income and poverty results even more striking.

VI. A Framework for Bounding the Importance of Selection

To what extent do improved living circumstances for children born after family planning programs began reflect the selection effect (that is, reductions in childbearing among disadvantaged parents) or the resource effect (that is, increased investments by parents in their own human capital, careers, and partnerships)? If only the selection effect operates, then the resources of the average child could rise even though the resources available to any one child do not. This section develops a novel framework for bounding the importance of selection, which allows us to approximate the resource effect. First, we show that empirical estimates of the overall effect of family planning programs can be decomposed into the sum of the selection effect and resource effect. Second, we use this decomposition to quantify the role of selection in raising the income of the average child.

Let g(y) ≥ 0 represent the empirical distribution of the number of children across household income, y, before family planning programs began. We model income in discrete bins rather than as a continuous variable to match how it is reported in Census data and for simplicity of exposition. Summing over all households, is the total number of children, and is the household income of the average child. The introduction of a family planning program may change this average in two ways. First, selection may decrease the number of children that lower income families have. Let g^fp(y) ≥ 0 capture the empirical distribution of the number of children by household income in the context of a family planning program. Note, , where M < N and represents the number of children in all households after family planning programs. The inequality is strict because family planning programs strictly lower the number of children born. If family planning operated exclusively via the selection effect, then , as fewer children were born to lower income families. Second, family planning programs may directly raise the incomes of parents through the resource effect, which we capture as h(y) ≥ y. This resource effect would further increase the income of the average child, such that .

Our estimates, therefore, capture the treatment effects of family planning programs through both the resource and selection channels, which we can write as follows:

(2)

By adding and subtracting a cross-term, we can decompose this effect into these two components,³⁰

(3)

On one hand, if we assume family planning programs do not have a direct resource effect [that is, h(y) = y], the second part of the expression (the resource effect) becomes zero, and the treatment effect reflects solely selection. On the other hand, if we assume the selection effect is zero [that is, M = N and g^fp() = g( )], the treatment effect reflects only the resource effect.

We cannot estimate the selection effect directly. Instead, we use this equation to simulate the role of selection and then bound the resource effect. Our approach uses the 1960 Census data and family planning user characteristics from the 1970 NFS. We simulate two types of selection: (i) “lower truncation,” which assumes that missing births were at the bottom of the income distribution, and (ii) “likely selection,” which assumes missing births followed the observed empirical distribution of the users of family planning services.

Table 6 presents the results from the bounding exercise. Because, as shown in the first row of the table, the exercise assumes the selection effect is zero, by construction the resource effect explains the entirety of the estimated 2.75 percent increase in children’s household income (unweighted average of effects for cohorts born four to six years after family planning, Column 2 of Table 2). To simulate lower truncation, we drop the poorest 2 percent of children from the 1960 Census, which corresponds to Bailey’s (2012) estimates of the effect of family planning programs on fertility rates.³¹ This approach assumes that only the poorest parents use family planning programs to avert births.³² We simulate likely selection by assuming that missing births followed the empirical distribution of household income for the users of family planning services. We fit a lognormal distribution to the household incomes reported by family planning patients in the 1970 NFS, which results in a distribution centered at the poverty line for a family of four ($23,636 annually in 2013 dollars in the 1960 Census) and a standard deviation corresponding to one-quarter of the standard deviation in the household incomes. We then remove a random 2 percent of children from the 1960 Census sample of children ages five to seven from this distribution so that the household income of the missing children corresponds to the household income distribution of users of family planning programs. The result is that 81.4 percent of averted children come from households with incomes below 150 percent of the poverty line, which matches the poverty rate of family planning program users in administrative data (83 percent).

Table 6, Rows 2 and 3, simulate the resource effects under lower truncation and empirically likely selection. The 95 percent confidence intervals for the estimates are generated using a nonparametric bootstrap procedure (Johnston and DiNardo 1997).³³ Selection by truncation generates a 1.84 percent increase in household income, accounting for around 67 percent (1.84/2.75) of the increase in income of the average child, while the resource effect accounts for around 33 percent. The largest effect of selection (implied by its upper confidence interval of 2.0) suggests that selection can explain at most 72 percent of the gains. If we also account for the uncertainty in the estimate of the increase in income, however, we fail to reject that the resource effect could be zero and that the most extreme form of selection explains the entire increase in household income.

Likely selection can explain around 36 percent (one percentage point) of the estimated increase in income of the average child. Our best estimate of the magnitude of the resource effect is that it accounts for the remaining 64 percent (1.75/2.75) of the increase in child income. Taking account of the uncertainty in the estimate of the increase in income as well as the simulated effect of selection, the lower 95 percent confidence interval on the resource effect implies that it accounts for at least 16 percent of the income gains to the average child. In short, the simulation rejects that the resource effect is zero under likely selection. This finding underscores an important and understudied return to family planning programs: they may directly raise parents’ income by encouraging parents to invest in their human capital, careers, and partnerships.

VII. Funding Family Planning as an Antipoverty Program

More than 50 years after the first federally funded family planning programs began in the United States, this study provides new evidence that they increased children’s economic resources and reduced child poverty. Using a validated research design and large restricted-use Census samples, we find that children born after the introduction of federal family planning programs were 7 percent less likely to live in poverty and 12 percent less likely to live in households receiving public assistance. A bounding exercise suggests that these increases in the income of the average child were achieved through the sizable increases in the incomes of disadvantaged parents—net of changes in selection. Our bounding exercise using the distribution of incomes of family planning users shows that the direct effect of these programs on the incomes of parents accounts for around 64 percent of the income gains for the average child. Although our data do not allow us to establish the exact mechanisms for this effect, they are consistent with family planning programs reducing the constraints of unintended pregnancy and encouraging greater investments by parents in their human capital, careers, and partnerships. To the extent that there were spillovers in services to nearby counties, that older siblings (who are part of our comparison group) benefitted, or that parents’ geographic mobility obscures our ability to measure the true effects, our results may understate the resource effects of family planning programs on children’s economic resources.

The magnitudes of these estimated effects of family planning programs on income square well with others in the literature. For instance, our bounding exercise suggests that family planning programs directly increased the income of the average child by $1,200 (0.0275 × the average pre-family planning mean family income of $68,000 in 2013 dollars × share due to the resource effect of family planning net of selection, or 1.75/2.75). Using the Bastian and Michelmore (2018) estimate that college completion rates should rise by 4.2 percent for every $1,000 increase in EITC income, our findings imply that college completion of children born to parents with access to family planning should increase by around 5 percent. This estimate is similar to the findings in Bailey (2013), who shows that the combined income and selection effects of family planning programs raised children’s likelihood of college graduation by 2–7 percent.³⁴

The magnitudes of these estimates also imply that the returns to family planning programs compare favorably with other public policies aiming to increase the resources of disadvantaged children. In the 1960s the federal government spent an average of around $278 million per year (2013 dollars) on family planning, or $4.4 billion for 1964–1980 (the period considered in this analysis). Using our estimates of likely selection, family planning programs directly reduced child poverty by 1 percent for just under $1 billion [$4.4 billion/(7 × 0.64)], where 7 is the percent reduction in child poverty (Table 3) and 0.64 is the share of each point attributable to the resource effect (Table 6, Row 3, likely selection).

Comparing the parental income returns (resource effects) of family planning programs to those of other programs targeting child poverty places these estimates in perspective. According to the supplemental poverty measure (SPM) for 2012, TANF cost $10.24 billion in 2013 dollars and reduced child poverty rates by 2.7 percent (from 18.5 to 18.0 percent). Ignoring offsetting behavioral changes and deadweight loss, this implies that TANF reduced child poverty by 1 percent for every $3.8 billion spent. Another useful comparison is the EITC. In 2012, the EITC cost around $63 billion, and the 2012 SPM suggests that EITC and the refundable portion of the child tax credit reduced child poverty rates by 27 percent (from 24.7 to 18.0 percent). A similar calculation implies that EITC reduced child poverty by 1 percent for every $2.3 billion spent. In short, even when we ignore the effect of family planning programs on public assistance outlays and through the selection channel, the effects of family planning programs on parents’ incomes reduced child poverty at less than half of the cost of the EITC and roughly one-quarter of the cost of TANF.³⁵

These resource effects may translate into large and long-term gains for children. For instance, Dahl and Lochner (2012) use variation in EITC eligibility over time and find a 4–6 percent of a standard deviation improvement in children’s test scores for each $1,000 of additional income. The Milligan and Stabile (2011) study of Canada’s child benefit programs and the Chetty, Friedman, and Rockoff (2014) study of U.S. tax credits find comparable estimates. Aizer and Currie (2014) show that children receiving a 12–25 percent increase in household income through the mother’s pension program in the early 20th century went on to attain about 0.4 years more schooling, had healthier weights in adulthood, earned about 14 percent more as adults, and lived about one year longer. Consistent with this, Bailey (2013) provides suggestive evidence from public Census data that cohorts born after family planning programs began were 2 percent more likely to attain 16 or more years of education and as adults had family incomes 1 percent higher (see Schultz 2008 for a discussion of developing countries). Future work should investigate these longer run linkages as well as the intergenerational impact of family planning programs on the economy.