Alternative paths to parenthood: IVF or child adoption?

Gumus, Gulcin ; Lee, Jungmin

I. INTRODUCTION

Infertility, defined as the inability to conceive after 1 year of unprotected intercourse, is recognized as a medical condition. In the United States, infertility is a growing problem: over the last several decades, age-related infertility has become increasingly prevalent as a relatively larger portion of women has deferred childbearing due to effective birth

control methods, safe and legal abortions, better access to college education, and greater participation in the labor market (Angrist and Evans 1999; Buckles 2008; Caucutt, Guner, and Knowles 2002; Goldin and Katz 2002; Kane and Staiger 1996).

Individuals with fertility problems who want to have children have two main alternatives: adoption and infertility treatment. (1) Chandra et al. (2005) use data from the 2002 National Survey of Family Growth (NSFG) to show that about 57% of women who received infertility treatment also considered adoption or actively sought to adopt a child. Infertility is an important factor in determining the demand for adoption. Married childless couples are more likely to seek adoption, and adoption is more prevalent among women who have used infertility services (Bernal et al. 2008; Chandra et al. 2005). Our calculations using the 2002 NSFG show that, among the women who are seeking to adopt, a third have also sought medical assistance to get pregnant. Some individuals choose to pursue infertility treatments and child adoption concurrently. Among women who are currently pursuing medical help to get pregnant, about 7% are also concurrently seeking to adopt an unrelated child. Conversely, among those who are seeking to adopt an unrelated child, about 20% are also pursuing medical help to get pregnant. (2) The extent to which individuals view child adoption and infertility treatment as alternatives is an empirical question with important economic, social, and public health implications. In this article, we investigate the substitutability between assisted reproductive technology (ART) and child adoption. To the best of our knowledge, no prior study in the literature provides a formal empirical investigation of these two alternative paths to parenthood.

The main motivation for this study is the recently increasing trend in ART use in the United States. Since the beginning of the 1980s, diagnostic and treatment options for infertility have advanced dramatically in effectiveness and availability. As a result, Americans have been utilizing more and more ART, mostly in the form of in vitro fertilization (IVF). In 1995, an estimated 6.7 million women had impaired fecundity, of which 42% had received some form of infertility services (Stephen and Chandra 2000). In 2002, an estimated 7.3 million women had fecundity problems, and about 45% had received some form of medical assistance (Chandra et al. 2005). The most commonly used infertility services are noninvasive methods such as medical advice, infertility testing, and ovulation drugs. IVF is characterized as "a last resort" that is generally pursued only after other diagnostic problems are solved and less invasive approaches have failed (Staniec and Webb 2007). ART, including IVF, is used by less than 2. Van Den Akker (2001) analyzes whether individuals in the United Kingdom first pursue IVF, adoption, or both simultaneously, among other options. He concludes that first actions varied greatly depending on whether the individuals suffered from female or male subfertility, or both. 1% of women, but the number of children conceived through ART is quite high. The Centers for Disease Control and Prevention (CDC) reports indicate that the number of babies conceived using ART has more than doubled from 20,840 in 1996 to 54,656 in 2006 (CDC 2008).

In addition to the increased effectiveness and availability of ART, improved access due to infertility insurance coverage has contributed to the upward trend in ART utilization. In the late 1980s and early 1990s, some states enacted health insurance mandates for infertility treatment. Currently, 15 states require health insurance plans to provide some form of coverage for infertility treatment. ART use lends itself to controversies over who should have access to such technologies, who should pay for the treatments, and whether the practices and providers should be regulated. Most recently, a mother in California, who already had six young children conceived through IVF, gave birth to octuplets with the help of IVF. This incident sparked a discussion on the ethics of ART and whether the practice of ART should be regulated (Archibold 2009). Bioethicists who argue in favor of such rules point out that child adoption and foster care are more strictly regulated than infertility treatment (Caplan 2009).

Despite the technological advances, ART remains a risky, complex, and expensive endeavor. Such infertility treatments are costly both in terms of money and time. The total cost of a successful delivery using IVF ranged between $44,000 and $211,940 in 1992 (Neumann et al. 1994), and the average cost was estimated at more than $50,000 in 2001 (Collins 2001). ART may also have significant adverse health effects on both the mothers and babies, especially due to the increased prevalence of multiple gestations. In particular, triplet and higher order multiple pregnancies lead to a greater chance of complications such as prematurity and maternal morbidity (Van Voorhis 2007). Between 1980 and 1997, the number of live babies born annually from twin gestation rose by 52%, while the number of high-order multiple births increased by 404%; these growth rates are mostly attributed to ART (NCHS 1999). Bitler (2008) estimates that the additional hospital costs associated with a national mandate covering infertility treatment would be $334 million in 2000 dollars due to the expected increase in the number of twin births alone.

While ART use has increased, the number of adoptions has remained relatively flat. Recently, Bernal et al. (2008) provide descriptive evidence suggesting that these advances in ART may be linked to reductions in the demand for domestic infants. Calculating the ratio of women who delivered their own biological children with ART to the women who adopted unrelated children domestically, they show that this ratio has continuously increased from 15% in 1992 to 34% in 1996 and 60% in 2002. According to the U.S. Children's Bureau, about 51,000 children were adopted in 2006. However, every year thousands more children wait to be adopted. The estimated number of children in foster care waiting to be adopted was about 129,000 in 2006 (US DHHS 2008). The discrepancy between the number of children waiting to be adopted and the number actually adopted is both because the adoptive matching of parents and children requires considerable time and resources, and because the demand for adoption remains insufficient (Hansen and Hansen 2006). The shortage of adoptive parents is a social concern as childhood at foster care is associated with poor lifetime outcomes (Blome 1997; Doyle 2007).

In light of these recent trends in the adoption and ART markets, the research question that forms the focus of this study is whether these two markets are related to each other. Specifically, we attempt to estimate the causal effects of child adoptions on ART utilization. (3) This is a policy-relevant question given that some individuals at the margin might choose adoption over ART due to legislative changes regarding the adoption of domestic children from foster care. Bernal et al. (2008) acknowledged that the availability of ART might be partly responsible for the decline in adoption rates, although not until about 1985, since the first successful IVF was implemented in 1981. They argue that the effect of ART utilization on the number of children adopted could be sizeable, but they do not formally estimate this relationship. Although some individuals might consider adoption as a primary means to have children, many others might pursue adoption only if they fail to conceive a child biologically connected to themselves. Even then, changes in the cost of adoption might influence individuals' decisions regarding the number of ART cycles attempted before giving up on this option. We specifically focus on foster care adoptions as public policy tools are more often used to encourage adoptions from foster care.

We use multiple data sources to construct state-level panel for 1999-2006. Our empirical results indicate that changes in adoption markets significantly affect ART utilization. The estimates reveal that a 10% increase in child adoptions would lead to a 1.3%-1.5% decrease in the number of ART cycles performed. In our models, we control for state-specific fixed effects (FE) and also address the potential simultaneity bias by using an instrumental variables estimation method (FE-2SLS). As explained in further detail below, we argue that both estimation methods are likely to provide the lower bounds of the true effect of adoptions on ART utilization.

We carry out additional analyses according to the ages of the adopted children and adoptive mothers, and find that the substitutability is greater for women aged 35 and older and for adoptions of younger children. Finally, we perform a series of robustness checks. For example, we provide evidence that the responsiveness is much larger when we consider international adoptions, suggesting that ART use is more sensitive to the demand for private adoption. The results of a falsification test indicate that there is no relationship between ART markets and adoptions of related children from foster care.

The rest of the article is organized as follows: Section II provides a review of the previous literature, and Section III describes the data used in the analysis. Section IV outlines the empirical framework and presents the results. Section V concludes.

II. LITERATURE REVIEW

A. Literature on Child Adoption

Adoption might be less risky than ART as individuals typically end up with a child at the end of the process. In addition, most of the costs of adoption are incurred at the end, only when the actual adoption process has taken place (Neumann 1997). Still, child adoption remains fairly complex, time-consuming, and uncertain. The cost of child adoption, beyond the document preparation and legal fees, varies widely depending on the type of adoption. The U.S. Children's Bureau reports that foster care adoption costs can be as much as $2,500, while domestic infant adoption costs range from $5,000 to $40,000, and international adoption costs range from $7,000 to $30,000 (US DHHS 2004a). Adoptive parents are subject to extensive background investigations and may face long waiting periods that vary greatly depending on the circumstances of the child, the birth parents, and the adoptive parents. Moreover, adoptive parents face risks in terms of the child's fit within their family. In light of these factors, federal and state governments support adoptive parents by offering incentives in the form of tax credits and tax exclusions, as well as adoption subsidies and reimbursements. (4)

The impact of public policies on foster care, adoption, and child welfare outcomes have been extensively studied. (5) Such policies date back to the 1935 Title IV-B of the Social Security Act (SSA) which "provided federal funding to states for a broad range of preventive and protective child welfare services for abused and neglected children" (Allen and Bissell 2004). Next, Title IV-E of the SSA, introduced with the Adoption Assistance and Child Welfare Act of 1980, established the federal Foster Care and Adoption Assistance Program. Avery (1998) provides an overview of the Act, its subsequent amendments, and its implementations by the states. In the early 1980s, following the passage of this Act, there was indeed a decline in the number of children in foster care (Waldfogel 2004).

This trend, however, did not last long. Swann and Sylvester (2006) report that foster care caseloads more than doubled from 1985 to 2000. They attribute 31% of this growth to higher female incarcerations and another 15% to falling welfare benefits. Similarly, Bider et al. (2006) report that the welfare reforms of the 1990s were associated with more children living in foster care. In 1997, President Clinton signed the Adoption and Safe Families Act (ASFA) into law. In addition to other policy changes, this Act regulated the termination of parental rights (TPR) in an effort to reduce the likelihood that children stay in foster care for extended periods of time. A child in foster care cannot be adopted until the date of TPR. The ASFA seems to have reduced foster care caseloads by giving states incentives to move children out of foster care and into adoption or other permanent placements (Swann and Sylvester 2006). Others have investigated different factors, such as the increased availability of contraception methods and the legalization of abortion, that may have influenced the supply of adoptable children and, eventually, adoption rates (Bernal et al. 2008; Bitler and Zavodny 2002; Gennetian 1999).

State child welfare agencies offer financial compensation to adoptive parents in the form of adoption subsidies that are often financed by contributions from the federal government. Gibbs et al. (2006) claim subsidies to be "the single-most powerful tool used by the child welfare system to encourage adoption." Gibbs et al. (2006) and Hansen (2008a) discuss the wide variation in the design of states' subsidy programs. Special needs criteria are based on the child's characteristics and typically include age, race, ethnicity, sibling groups, and disabilities. Each state sets its criteria within broad federal guidelines. Researchers have exploited such variation in adoption subsidies within states over time to evaluate their impact on adoption outcomes. Using data from the 1996-2003 Adoption and Foster Care Analysis and Reporting System (AFCARS), Hansen (2007) shows that federal and state policies promoting the adoption of children from foster care significantly increase the demand for adoption. She finds that an increase of $100 in average adoption assistance payments would reduce the number of children waiting in foster care by almost 4,200 per year.

State child welfare agencies also offer financial compensation to foster caregiver families or institutions. These financial incentives take the form of basic monthly foster care payments. The average figures indicate that these subsidies compensate foster caregivers about 56%-77% of their monthly expenditures for a 9- to 11-year-old child, depending on the level of family income (Doyle and Peters 2007). Hence, the subsidies constitute a significant portion of the monetary costs associated with childrearing and seem to improve the placement of foster children (Doyle and Peters 2007; Duncan and Argys 2007). Not only is adoption cheaper than foster care (in terms of monthly subsidies), but it might also yield better outcomes than foster care placement. Barth et al. (2006) perform a careful comparison of long-term foster care with adoption using a propensity score matching method. They find that the cost of foster care to the government is on average twice as much as the cost of adoption.

Blome (1997) finds that children who age out of foster care are significantly more likely to have discipline problems in school and drop out of high school than a matched group of nonfoster care children. Doyle (2007) also reports significant adverse effects of foster care on child outcomes for school-age children and youth. Using an instrumental variable (IV) approach to identify causal effects of foster care on longterm outcomes, he finds that children in foster care have higher delinquency rates, higher teen birth rates, and lower earnings compared to those who remain at home. Another cost-benefit analysis carried out by Hansen (2008b) indicates that adoptions provide far greater longterm benefits. Her analysis takes into account a wide range of outcomes including the child's health, behavior, education, criminal activity, and employment. The estimated total private benefits range from $88,000 to $150,000, while the estimated total public benefits are even higher at about $190,000 to $235,000 in 2000 dollars. Her findings indicate that, from the public policy point of view, $1 spent on adoption from foster care yields about $3 in benefits.

B. Literature on ART Utilization

The second strand of literature related to our article is on ART use, and evaluates how the increased availability of treatments influences ART utilization, fertility rates, and other outcomes. Most of these studies exploit the variation in state-level infertility insurance mandates that mostly took place in the late 1980s and early 1990s, well before the period of our analysis. While Bitler and Schmidt (2006) do not find any evidence that the mandates increased ART utilization, several others claim otherwise (Bundorf et al. 2007; Henne and Bundorf 2008). Some others found that the mandates affected the utilization of treatments and birth rates for older or more educated women but not necessarily for the entire female population of childbearing age (Bitler and Schmidt 2009; Schmidt 2007). Buckles (2007) and Machado and Sanzde-Galdeano (2009) confirm that the infertility insurance mandates increased birth rates among older women (ages 35+) but decreased those for younger women.

Finally, Bitler (2008) analyzes not only fertility but also infant health outcomes. She shows that the insurance mandates led to significant increases in the number of twin deliveries. As mentioned previously, multiple gestations may lead to complications for both the babies and the mothers. She finds that the mandates increased the use of infertility treatment by older women, which in turn led to small, albeit statistically significant, negative impacts on health outcomes for twins as well as singletons. None of the studies mentioned above focus on how adoptions and infertility treatment utilization might be related. To our knowledge, our study is the first to provide empirical evidence on the links between ART use and child adoption in the context of public policy.

III. DATA

Our empirical analysis uses state-specific longitudinal data from 1999 to 2006. The main variables of interest are the measures of ART utilization and child adoption. The former come from the national clinic registry data and were collected as a joint effort by the CDC, the American Society for Reproductive Medicine, and the Society for ART (SART). The foster care adoption figures are obtained from the AFCARS. The analysis also includes several annual, state-specific characteristics that were collected from various sources. The full list of sources and summary statistics is depicted in Table 1.

A. ART Utilization

We obtained ART utilization figures from the CDC's annual ART Reports. (6) These data refer specifically to ART procedures in which a physician surgically removes a woman's eggs in order to combine them with sperm and then returns them to her uterus. Among various ART techniques, IVF is the most common form. According to the 2006 report, IVF was the method of choice in 99.7% of all types of ART procedures using fresh non-donor eggs or embryos. Since 1995, each ART clinic that is a member of SART has been required to report certain indicators regarding its operations and success rates to the CDC as mandated by the Fertility Clinic Success Rate and Certification Act of 1992. Each year, the CDC publishes clinic-level data with a 3-year lag to track the outcome of the cycles performed.

The ART cycles performed correspond to the year in which they started. The data include information on the number of cycles performed as well as pregnancy and birth rates for women in various age groups. Our key outcome measure is the number of ART cycles performed using fresh nondonor egg transfers. (7) As shown in Table 1, this measure displays great variation across states and years. We normalize this measure (as well as the adoption figures) by the number of women within the age range 25-44 in each state by year. This is the population relevant for our purposes as it corresponds to the reproductive-age women who are most likely to use ART services (Henne and Bundorf 2008). (8) Specifically, the main ART utilization measure in this article is the number of ART cycles performed per 1,000 women, which steadily increased from 1.42 in 1999 to 1.92 in 2006 in our sample.

As part of our empirical analysis, we restrict our attention to women who are ages 35 or older (35+), given that impaired fecundity becomes more pronounced well before the onset of menopause. Moreover, ART is less effective among older women: on average, it takes fewer than three ART cycles per one live birth when all women are considered, while about five ART cycles are performed per one live birth among older women (35+). We construct an additional outcome measure for the number of ART cycles performed for women who are aged 35 or older and adjust this measure by the number of women in the age group 35-44. This outcome measure rose somewhat faster, increasing from 1.40 in 1999 to 1.96 in 2006, with an overall average of 1.72. Finally, we proxy the availability of ART procedures using the number of ART clinics for a given state and year (per 1,000 women). (9) The unadjusted annual average number of clinics, across all states, is about 7.5, and there is a great deal of variation across states and years. Adjusting this average for the number of childbearing age women in each state by year yields about 8.8 clinics for each one million women aged 25-44.

There are a few data limitations that need to be mentioned. First, the CDC ART reports lack information on individual ART users. As a result, an increase in the number of cycles performed could be due to an increase in the number of users and/or a higher number of trials per user. However, focusing on the number of ART cycles rather than the number of ART users is more policy relevant as the cost of utilization depends on the former. Another shortcoming of the CDC ART reports is the lack of figures on the utilization of infertility treatments other than ART, such as ovulation stimulating drugs and the surgical repair of reproductive organs.

Schmidt (2007) reports that most instances of impaired fecundity are treated by these less invasive therapies with costs ranging from $200 to $3,000 per treatment. Given the considerably lower cost of such alternative treatments, individuals with impaired fecundity would presumably use these kinds of infertility service before considering adoption or ART. Hence, we do not expect to see any significant relationship between child adoption and the utilization of infertility treatments other than ART.

B. Adoption and Foster Care

Data on foster care adoption measures come from the AFCARS, a federal source of child-specific information on all children covered by the protections of Title IV-B and Title IV-E of the SSA (NDACAN 2007). (10) Since 1995, states have been submitting data concerning each child in foster care and each child who has been adopted under the authority of the state's child welfare agency to the U.S. Children's Bureau on a semi-annual basis. (11) The AFCARS database has been designed to address policy development and program management issues at both the state and federal levels. It contains one file with the adoption data and another with the foster care data for each year. The data These adoptions do not involve a state agency and are not included in the publicly available version of the data." correspond to the reporting period of federal fiscal years, starting on October 1st and ending on September 30th of the following year. (12)

The AFCARS has some limitations. First, in the years prior to FY1998, some states did not report any adoption or foster care data and other states have incomplete data. For this reason, our analysis covers the period from 1999 to 2006, and the panel includes 378 state-year observations. (13) Second, the adoption figures reflect only the children adopted under the authority of the state child welfare agency, which accounts for about a third of all child adoptions in the United States (US DHHS 2004b). The process of child adoption takes a variety of forms, as children can be adopted through private or public agencies and from domestic or international sources. The AFCARS does not contain figures on international adoptions or domestic private agency adoptions. Our calculations using the 2007-2008 National Survey of Adoptive Parents (NSAP) reveal that the share of foster care adoptions is about 37%, while private agency adoptions constitute 38% of all adoptions, and the remaining 25% are international adoptions. (14) In what follows, we provide descriptive statistics based on the NSAP and conjecture that our estimates using the AFCARS constitute a lower bound of the true effect of adoption rates on ART use. In addition, we provide evidence to support this argument using data on international adoptions from the Department of State.

Despite its shortcomings, the AFCARS is still the only data source available on adoption figures along with adoption subsidies that are consistent across states and over time. The adoption file contains information on adopted children, including their gender, race, birth date, ethnicity, and prior relationship with the adoptive parents, as well as limited information on the characteristics of the birth and adoptive parents. In this article, we consider the number of child adoptions by unrelated individuals for the main analysis and later provide a robustness exercise using the number of adoptions by relatives. Thus, in what follows, adoptions refer to child adoptions from foster care by unrelated individuals, unless stated otherwise. In addition to considering all unrelated adoptions, we decompose this number by the age of the adoptive mother and the age of the adopted child. The motivation for the subsample of older women (35+) is explained above. We also study the adopted children who are above age 5 (5+) as an additional subsample. Individuals may pursue ART as a means to have children both because they might have a stronger preference for biologically linked children and because they might be particularly interested in raising an infant. If so, adoptions of younger children would be considered a more appropriate substitute for ART. We normalize each of these adoption figures by the number of women (and the number of women 35+), and these child adoption rates are the primary explanatory variables in our empirical analysis.

In our sample, the average number of unrelated children adopted from foster care is 711. The average number of children adopted by older adoptive mothers (35+) is 562, constituting almost 80% of all adoptions. The normalized figures indicate that, on average, there is 1.0 child adopted per 1,000 women (aged 25-44). When we consider adoptions by older women only, on average, there are 1.5 child adoptions per 1,000 women (aged 35-44). These figures suggest that adoptions are more popular among older individuals who are more likely to suffer from impaired fecundity and for whom ART is a relatively more costly and less successful alternative. During the period we consider, children adopted under the age of 5 constitute 43% of all adoptions.

Importantly, the AFCARS contains the adoption subsidies provided to the adoptive parents under adoption assistance agreements. Under such agreements, adoptive parents may receive monthly payments, medical coverage, and other services depending on the state and county policies if they adopt a child who meets certain eligibility requirements. According to the U.S. Department of Health and Human Services, 89% of all children adopted during FY2006 received adoption subsidies (US DHHS 2008). Most of these subsidies are federally funded under Title IV-E of the SSA. In addition, state and/or county funded subsidies might be available even if an adoption does not qualify for the Title IV-E subsidies. Using the AFCARS figures, which reflect the subsidies effective at the time the adoption is finalized, we construct average monthly adoption subsidies (in real 2000 US$100s). (15) The average monthly adoption subsidy in our sample is about $470. Note that the adoptive parents continue to receive these subsidies typically until the child reaches age 18. Thus, they could be sensitive to even small variations in adoption subsidies.

The AFCARS foster care file contains information on children in foster care including their gender, birth date, race, ethnicity, and availability for adoption. As Table 1 indicates, only about 6% of children in foster care end up being adopted. This is partly due to the fact that it is not easy to match children in foster care with adoptive families. More importantly, not all children in foster care are available for adoption as many are being kept to be reunited with their biological parents. We construct a measure of the "supply of adoptable children" using the number of children in foster care waiting to be adopted (about one-fifth of all children in foster care). This measure is also adjusted by the number of women in a given state and year. We use the gender and age composition of these children as additional identifying variables. As Hansen (2007) points out, the matching of adoptable children and adoptive parents depends on "attributes of both the adult population and the population of waiting children." Therefore, the composition of adoptable children may he an important factor if adoptive parents seek specific child attributes.

C. State-Level Characteristics

In all of our analyses, we include a number of state-level characteristics to account for the demographic and economic conditions that could be related to both child adoptions and to ART utilization. The 2000 Census figures indicate that adopted children are more likely than biological children to live with white, married couples with higher educational attainment, home ownership rates, and household income (Kreider 2003). Similarly, women who utilize infertility treatment have, on average, higher income and educational attainment, and are more likely to be white and to have health insurance coverage compared to other infertile women (Bider and Schmidt 2006; Hammoud et al. 2009; Staniec and Webb 2007; Stephen and Chandra 2000). Given that the state FE remove any unobservable time-invariant state-specific factors, the effects of all these state-specific characteristics are solely identified via within-state variation over time.

The state-level variables in our analysis include the number of ART clinics per 1,000 women, the share of the total resident population covered by private health insurance, the real income per capita, the percentage share of the total resident population with at least a college degree, the teen birth rate, the race ratio of the resident population (defined as blacks over whites), and the ratio of the female resident population between the ages of 35 and 44 to that between the ages of 25 and 44. This last variable indicates the portion of females with deteriorated fecundity among women of childbearing age. Female labor force participation rates play a crucial role in determining the demand for children and thus are included as an additional explanatory variable. We also include a dummy variable that equals one if the state has adopted health insurance mandates for infertility treatments. Such mandates provide an exogenous source of variation in the out-of-pocket cost of ART procedures, but most of the states passed the mandates in earlier years. During the period of our study, only Louisiana and New Jersey passed mandates, both in 2001. For this reason, it is not possible to carry out an alternative analysis of how infertility insurance mandates influence adoption rates.

IV. EMPIRICAL FRAMEWORK AND RESULTS

A. Empirical Framework

To conceptualize our estimation equations, we first present a simple framework. Suppose that women of childbearing age in a given state and year are distributed according to a joint distribution of four variables, [U.sub.0], [U.sub.b], [U.sub.t], and [U.sub.a]. Let [U.sub.0] denote the indirect utility of a woman when she remains without a(n additional) child, and let [U.sub.b], [U.sub.t], and [U.sub.a] denote the expected utility when she attempts to have an additional child through natural birth, ART, and adoption, respectively. (16) Next, we define the following probabilities:

[[pi].sub.0] = Pr([U.sub.0]> [U.sub.b], [U.sub.0] > [U.sub.t], [U.sub.0] > [U.sub.a])

[[pi].sub.b] = Pr([U.sub.b] > [U.sub.0], [U.sub.b] > [U.sub.t], [U.sub.b] > [U.sub.a])

[[pi].sub.t] Pr([U.sub.t] > [U.sub.0], [U.sub.t] > [U.sub.b], [U.sub.t] > [V.sub.a])

[[pi].sub.a] = Pr([U.sub.a] > [U.sub.0], [U.sub.a ]> [U.sub.b], [U.sub.a] > [U.sub.t]).

Each probability, [[pi].sub.j], represents the proportion of women who choose option j. (17) Consequently, [[pi].sub.t]N is the number of women who choose to use ART to have a child, where N is the total number of childbearing age women in a given state and year. One empirical implication of this set-up is that a simple cross-sectional relationship between ART utilization ([[pi].sub.t]) and adoption ([[pi].sub.a]) might be spurious if the other two probabilities, [[pi].sub.0] and [[pi].sub.b], are omitted from the analysis. For example, if there is an increasing trend of childless marriages, this alone will lead to a positive correlation between [[pi].sub.t] and [[pi].sub.a].

Now suppose that there is an increase in the adoption subsidy (or any exogenous shock) that makes adoption more attractive than other options, holding everything else constant. Let s > 0 denote the corresponding increase in [U.sub.a]. It is difficult to think of any plausible scenario in which such a change would have an impact on ART utilization rates except through their effect on child adoptions. Therefore, there is no reason for [U.sub.j], j [not equal to] a, to change. The probabilities after the new adoption subsidy can be expressed as follows:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

There are some women who start to seek adoption, (([[pi].sub.0] - [[pi].sub.'0]) + ([[pi].sub.b] - [[pi].sub.'b]) + ([[pi].sub.t] - [[pi].sub.'t])) N, where ([[pi].sub.'t]- [[pi].sub.t])N represents those who switch from ART to adoption. The following term allows us to gauge the degree of substitutability between ART use and adoption:

(1) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

The numerator above indicates the percentage change in the number of women who use ART, while the denominator represents the percentage change in the number of women who decide to adopt as a result of the policy change. (18) This term, which can be interpreted as an elasticity, is what we aim to identify.

The existence of a measurable mass of ([[pi].sub.t]-[[pi].sub.'t])N requires that there are at least some women who do not have lexicographic preferences, and therefore, may be willing to switch between ART and adoption. Our calculations using the 2002 NSFG suggest that there might be significant overlaps in the characteristics of women who seek medical assistance with getting pregnant and those who adopt an unrelated child. The former are more likely to be infertile (29% vs 23%), white (83% vs 78%), and college educated (79% vs 66%). However, there are no statistically significant differences in terms of age (36% vs 38%), labor force participation (76% vs 74%), or full-time work status (47% for both).

Whether there are some women who switch from ART to adoption is an empirical question. To test this, we specify the following equation determining ART utilization:

(2) [ART.sub.s,t] = [beta][Adopt.sub.s,t] + [X.sub.s,t[gamma]] + [[micro].sub.1,t] + [v.sub.1,s] + [v.sub.s,t],

where the dependent variable is the number of ART cycles undertaken in state s and year t, normalized by 1,000 women aged 25-44. [Adopt.sub.s,t] is the number of unrelated foster care adoptions in state s in year t, which is also normalized. Our main interest is the estimate of [beta], which quantifies the average elasticity of ART cycles performed with respect to adoptions. The vector X includes various state characteristics as listed above that may impact ART use. Equation (3) also includes year- ([[micro].sub.1,t]) and state-specific ([[v.sub.,1,s]) FE. The former capture any nationwide trend in infertility treatment utilization, while the latter pick up any unobserved time-invariant state heterogeneity. Lastly, [v.sub.s,t] is a standard error term.

An econometric problem here, as our conceptual framework indicated, is the potential endogeneity of Adopt because child adoption and ART utilization are related choices. This problem may be resolved to a degree with the inclusion of state and year FE (Doyle and Peters 2007; Duncan and Argys 2007). The FE jointly provide unambiguous controls for the state- and year-specific unobserved characteristics and attitudes influencing both adoptions and ART use. We expect that the inclusion of FE will alleviate any potential endogeneity bias to the extent that the bias comes from unobservable, state-specific time-invariant characteristics and year-specific effects. To better address the endogeneity problem, we also use a FE two-stage least squares method (FE-2SLS). Specifically, we estimate the following first-stage equation:

(3) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where the adoption rate is formally modeled as a function of adoption policy and foster care variables, state FE ([v.sub.2,s]), year FE ([[micro].sub.2,t]), and state characteristics that are included in the second-stage equation.

The IVs are selected based on previous findings. The literature on child adoption reviewed above has documented that the adoption rate significantly depends on the number of adoptable children (Bernal et al. 2008; Bitler and Zavodny 2002; Gennetian 1999) and the policy variables that influence the cost of adoption, including the adoption subsidies (Doyle and Peters 2007; Duncan and Argys 2007; Hansen 2007). We include four IVs: Subsidy, KidsWait, PercGirls, and PercInfants. Average monthly adoption subsidies within states vary substantially over time as states can define their criteria for special needs. Furthermore, Hansen and Pollack (2005) show that the subsidy payments typically result from a bargaining process between the child welfare authority and the adoptive family. Thus, subsidies vary due to variation in the characteristics of children waiting to be adopted as well. For robustness, we also use an alternative measure: the basic adoption assistance rates for 2-year-olds as previously utilized by Hansen (2007). These basic rates serve as benchmark amounts in the bargaining processes, and the data are collected by the North American Council on Adoptable Children (NACAC). The AFCARS, on the other hand, contains the actual rates effective at the time of adoption finalization and thus displays greater variation. (19)

While KidsWait indicates the number of children in foster care waiting to be adopted, PercGirls and PercInfants reflect the demographic composition of these children, as measured at the end of the previous fiscal year. The last two variables are defined as the percentage of girls and the percentage of children under age 5 who are in foster care waiting to be adopted, respectively. The number of children waiting to be adopted (KidsWait) represents the supply. Given that there is an excess supply of children waiting for adoption, one may expect that an increase in KidsWait would not necessarily affect the number of children finally adopted. However, a higher number of adoptable children improves the matching quality between children and adoptive parents, thus increasing the number of adoptions. Moreover, the number of children waiting to be adopted might directly or indirectly affect the monetary or nonmonetary costs of adoption, as well as the decision processes regarding case goal determination for the foster care children. In addition to the definition of special needs and the determination of monthly subsidy rates, Bower and Laws (2002) identify a dozen other policy tools used by the states to support adoptive parents. These include, but are not limited to, nonrecurring adoption expense reimbursements, subsidized guardianships, subsidies for children over 18 years old, and eligibility for Medicaid. We omit such variables from our analysis due to the absence of state-specific data, but they are partially and indirectly captured by KidsWait.

The inclusion of the two demographic composition measures, PercGirls and PercInfants, is also motivated by the literature on "revealed preferences" for adopted children. First, it has been documented that girls are more likely to be adopted than boys (Kreider 2003) both because adoptive mothers prefer girls over boys and girls seem less risky to adopt (Baccara et al. 2009; Chandra et al. 1999; Gravois 2004; Groze 1991). Accordingly, we would expect PercGirls to increase the adoption rate. Second, Chandra et al. (1999) report that the majority of women (58%) prefer to adopt an infant under the age of 2, while 28% expressed a preference for an adopted child aged 2-5, and only 6.8% for a child aged 6-12. Therefore, to the extent that adoptive parents prefer infants over older children, PercInfants is expected to be positively related to adoption rates. (20)

The assumption is that our instruments do not directly affect ART use after we control for various state characteristics as well as for year and state FE. It might be plausible that states determine monetary incentives, such as adoption subsidies, according to the number of children waiting in foster care or the number of children adopted. However, many researchers have defended the assumption of exogeneity, especially in the presence of state and year FE (Doyle and Peters 2007; Hansen 2007; Hansen and Hansen 2006). Doyle and Peters (2007) argue that the subsidy rates are not set to clear the foster care market; rather they may be "set to partially reimburse foster parents for child expenditures and are often tied to rates set in other welfare programs." The level of adoption subsidies is decided through a process of bargaining between the adoptive family and the child welfare authority, with the aim of determining the appropriate level of financial assistance that would cover the costs of raising the adopted child. In fact, the adopted child's and adoptive parent's characteristics can explain most of the variation in subsidies (Avery and Mont 1992; Dalberth et al. 2005). We discuss the validity of our instruments, together with other issues regarding potential biases, in the next subsection.

All equations are estimated using the total number of women as weights to reflect underlying microdata. In addition, standard errors are clustered by state in order to account for possible serial correlation in the error terms. We interpret our FE-2SLS estimates in the spirit of the local average treatment effect. The FE-2SLS estimate identifies the average response of individuals whose decision is influenced by the specific IVs used in the analysis (Angrist and Imbens 1995). In this case, we can interpret the FE-2SLS estimate as the effect of one additional adopted child on the ART utilization when the decision to have children is influenced by the excluded IVs we utilize. These variables include public policy measures such as adoption subsidies. Therefore, even if these individuals might not be representative of the entire population, they represent the group whose infertility treatment utilization responds to changes in adoption markets.

B. State-Specific Time Trends, Validity of Instruments, and Potential Bias

It is possible that both FE and FE-2SLS estimates are biased. In this subsection, we argue that the potential bias is likely to yield estimates that are upwardly biased toward zero. First, our FE estimates may be biased as we cannot control for state-specific time trends. For example, suppose that there is a state-specific unobservable trend toward smaller families. The trend would be negatively correlated with both ART use and child adoption. Similarly, if there is a state-specific trend among women to postpone motherhood, it would be positively correlated with both the number of ART cycles and adoptions. Considering these and various other plausible scenarios, we conjecture that our estimated elasticity without controlling for state-specific trends is likely to be biased upwardly toward zero, thus underestimating the extent that individuals might substitute ART use for adoption.

The FE-2SLS estimates might also be biased if our instruments are not valid. In terms of the validity of our instrument Subsidy, one might think that the subsidies are endogenously determined conditional on the demand and supply of children in adoption markets. More specifically, it might be the case that some states increase the subsidies in an effort to promote adoption when faced with an increase in ART use. In this case, Subsidy would be correlated with the error term in the second-stage equation, yielding IV estimates that are biased. However, note that this scenario indicates a positive correlation, that is, a positive (or negative) shock in ART use leads to an increase (or decrease) in adoption subsidies. Therefore, like the FE estimates, our FE-2SLS estimates can be also interpreted as lower bounds of the true effect.

Regarding KidsWait, one might consider the possibility that higher ART utilization rates reduce adoption rates, thereby increasing the number of children available for adoption. However the suggested mechanism also indicates a positive correlation between the instrument and the second-stage error term. Again, our FE-2SLS estimate is a lower bound of the true substitution effect in absolute terms. Furthermore, KidsWait, PercGirls, and PercInfants are all based on the foster care files, which reflect the stock values at the end of each fiscal year (i.e., the number of children remaining in foster care as of September 30th), whereas adoption figures are flow variables, which reflect the number of children who were adopted throughout the entire fiscal year. Therefore, we merge adoption variables with foster care variables for the previous fiscal year. For example, we merge 2005 adoption variables (corresponding to the period of October 2004 to September 2005) with 2004 foster care variables (corresponding to the period of October 2003 to September 2004). The timing structure of these measures supports the validity of our IVs. Below, we present the results of formal diagnostic tests for the relevance and exogeneity of the excluded instruments.

C. Estimation Results

In Table 2, we present the ordinary least squares (OLS) and the FE estimation results corresponding to Equation (3). Panels A, B, and C display the results for all unrelated adoptions, unrelated adoptions by older women (ages 35+), and unrelated adoptions of older children (ages 5+), respectively. Each column corresponds to a different specification: in column 1, the estimation method is OLS with control variables and year FE, but no state FE. Column 2 features the same specification as column 1 except that it includes state FE. For brevity, we do not report the estimated coefficients on state and year FE, nor those on the control variables. (21) We focus on the estimated coefficient of our main interest, 13. We report the coefficient estimate for the adoption variable followed by its standard error in parentheses and the implied elasticity in italics.

First, we present the results for all unrelated adoptions and ART use in panel A of Table 2. The OLS model without state FE (column 1) yields a positive coefficient on child adoptions that is only marginally statistically significant (p < .10). Including state FE (column 2) yields a negative coefficient of about -0.22, which is highly significant (p < .01). The implied elasticity of ART use with respect to unrelated adoptions is about -0.13. A comparison between columns 1 and 2 reveals a pattern that is consistent with our conjecture that a cross-sectional correlation between adoption and ART utilization is likely to be spurious due to omitted variables related to [[pi].sub.0] or [[pi].sub.b]. The FE estimate implies a degree of substitutability between these two options for having children.

Panel B in Table 2 displays the results for unrelated adoptions by older adoptive mothers (ages 35+). The implied elasticity of ART use with respect to adoptions (-0.17) is somewhat larger in absolute terms compared to the results presented in panel A. A higher responsiveness among older women is expected because ART success rates are lower for them. As older women need more ART cycles for a successful delivery, an increase in adoptions will reduce ART use more compared to younger women. Alternatively, older women might have weaker preferences for biological children. The estimate in panel B is also statistically significant (p < .01). The final panel in Table 2 reports the results for the adoptions of older children (ages 5+). The FE estimate is about -0.30 (p < .01) and yields an implied elasticity of -0.10. This suggests a greater substitutability between ART use and infant adoption, likely reflecting strong preferences for infants among those who undertake ART.

Next, we present the findings from the FE-2SLS models. Table 3 displays the results for the first-stage equation (4). The odd-numbered columns refer to the results using the IV set I, which includes average monthly adoption subsidies from the AFCARS, while the even-numbered columns report the results using the IV set II, which includes basic adoption assistance rates for 2-year-olds from the NACAC. The first two columns present the results of our baseline specification for all unrelated child adoptions. The results are consistent with our expectations: increases in the average amount of subsidies and the number of waiting children both significantly increase adoption rates. Our point estimates suggest that a $100 increase in adoption subsidies would increase the unrelated adoption rate by 0.04-0.07, depending on the specification. The implied elasticity is about 0.14-0.29. These estimates are similar to those of Hansen (2007), who estimated an elasticity of 0.16. She found that a $100 increase in adoption subsidies would increase adoptions by about 80 children, in an average state and a given year. Hansen and Hansen (2006), on the other hand, found that a $100 increase in subsidies is associated with an increase of about 28 additional adoptions per year. Our estimates indicate about 40 additional adopted children, with 31 additional unrelated adoptions, in response to a $100 increase in subsidies.

We also find that the adoption rates rise as PercGirls increases, indicating a preference for girls over boys. Finally, the adoption rates among older women (35+) and of older children (5+) both increase as PercInfants decreases. (22) The finding that adoption rates among older women increase as PercInfants decreases might reflect the positive correlation between adoptive mothers' age with the age of the adopted children. While most states do not specify an age for adoptive parents, some states require a minimum age difference between the adoptive parent and the adopted child. The instruments are all statistically significant, and the F-statistic for the joint significance ranges between 17 and 21. Based on Hansen's J-statistics, we cannot reject the null hypothesis of over-identification, thus validating our instruments.

Columns 1 and 2 in Table 4 display the results for our second-stage regressions using the IV sets I and II, respectively. The two sets of IVs yield very similar results in all three panels. In panel A, we start with all unrelated adoptions. Recall that the FE estimate reported in Table 2 is about -0.22. The FE-2SLS estimates of [beta] in columns 1 and 2 of Table 4 indicate even larger effect sizes of about -0.26 and -0.25, respectively, and both are highly statistically significant (p < .01). This pattern implies that unobserved factors that are correlated with both ART and Adopt in the same direction, bias the FE estimate toward zero. Hence, this finding provides further confirmation that our FE estimates constitute lower bounds of the true effect of child adoption on ART use. The implied elasticity of ART utilization with respect to unrelated adoptions is -0.15, slightly larger in absolute terms. In panel B, we restrict our attention to older adoptive mothers (35+). The estimated coefficients are associated with somewhat larger elasticity figures in absolute terms: between -0.18 and -0.19. Finally, panel C displays the results for older adopted children (5+), with estimated elasticities of about -0.11. Thus, the adoption of older children seems to be a relatively weak substitute for ART use.

The findings regarding the control variables included in the analysis (not reported) are also worth noting. First, we find that one additional infertility specialty clinic per 1,000 women increases the number of cycles performed by about 66 per 1,000 women (p < .01). Second, infertility insurance mandates increased ART utilization rates, but the effect is statistically insignificant probably due to the limited within-state variation. Third, the ART use increases with increasing per capita income (p < .05), as the financial cost is a major obstacle to ART. Fourth, higher teen birth rates apparently reduce ART utilization (p < .01), which might be picking up the effect of private agency adoptions on ART use. Lastly, ART use is also positively associated with the ratio of the female resident population between the ages of 35 and 44 to that between the ages of 25 and 44 (p < .01). As explained above, this fraction represents the relative size of the female population experiencing relatively more severe fertility problems among all women of childbearing age.

D. Robustness Checks

Table 5 presents the results of several robustness checks using FE models. First, we run a regression, including both related and unrelated adoptions. Related adoptions are quite different in nature than unrelated adoptions as the former are probably undertaken primarily due to motives independent of fertility status and economic factors (e.g., people adopting young offspring of their siblings or spouses). This falsification exercise yields basically zero coefficient and elasticity estimates for related adoptions, while the coefficient on unrelated adoptions is virtually the same as reported above (panel A).

As discussed above, the AFCARS figures reflect foster care adoptions only rather than all child adoptions. Our calculations using the NSAP reveal that foster care adoptions make up little more than a third of all adoptions. Those who adopt from foster care might be different from those who pursue international or private agency adoptions. In fact, the summary statistics using the NSAP suggest that foster care adopters are less affluent than the rest, based on the level of their household income. Second, foster care adopters are less likely to suffer from infertility. About 38% of foster care adopters report infertility, compared to 52% and 72% for private agency and international adopters, respectively. Therefore, foster care adopters might be less likely to pursue ART, which further supports our conjecture that our estimates constitute lower bounds of the true effect of child adoption on ART use.

To test this conjecture, we obtained data on international adoptions from the Office of Children's Issues under the U.S. Department of State. These data include the number of all children who received immigrant visas issued during a given fiscal year for the purpose of adoption in the United States. On average, there were 0.5 international adoptions per 1,000 women during our period of analysis (Table 1). Panel B in Table 5 reports the estimation results when we include both international and domestic unrelated foster care adoptions as explanatory variables. The FE regressions of ART use yields a coefficient of -0.761 on international adoptions (p < .05). The coefficient on unrelated foster care adoptions remains the same as before (p < .01). The estimated elasticity on international adoptions is -0.22, indicating that changes in international adoptions have larger effects on ART use than domestic adoptions from foster care.

As another robustness exercise, we consider all ART cycles using nondonor and donor egg/embryo transfers, including both fresh and frozen types. Inclusion of frozen nondonor embryo transfers together with fresh nondonor ones does not affect the results (panel C). The coefficient estimate is still highly statistically significant with an implied elasticity of about -0.13. On the other hand, the regression using donor egg/embryo transfers (both fresh and frozen) reveals a negative but statistically insignificant coefficient estimate (panel D). We can interpret this finding as evidence for no substitutability between adoptions and ART cycles performed using donor eggs/embryos only, which might be first because such cycles are a small portion (12%) of all ART cycles performed. Moreover, they are associated with even higher price tags as well as additional difficulties of finding a donor.

We perform one final robustness exercise in an attempt to address any concerns due to potentially significant time delays in child adoptions from foster care. (23) Hansen (2006) finds that about half of all adoptions are completed within 1 year once the child is available for adoption, that is, the date of the TPR. She also reports that about 80% of adoptions are completed within 2 years of termination, while very few adoptions take place within 4 years of termination. Note that there may also be significant delays between the time a decision is made to pursue IVF and its actual implementation because the period during which drugs are used to stimulate egg production before the eggs are retrieved from the ovaries varies, depending on the circumstances (Staniec and Webb 2007). We re-estimated our main specifications (OLS, FE, and FE 2SLS) by using adoption figures that are further lagged 1 year. This exercise yields negative and statistically significant but smaller elasticity figures that support our main conclusions.

V. CONCLUSIONS

Our estimates indicate that changes in adoption markets significantly affect ART utilization. In our efforts to gauge the substitutability between ART use and child adoption, we also address the potential simultaneity bias and show that our empirical findings are likely to provide lower bounds of the true effect of adoptions on ART utilization. The empirical evidence we present suggests that promoting adoption encourages some individuals to adopt, rather than pursue ART as a means to have children. Recently, Congress passed the Fostering Connections to Success and Increasing Adoptions Act of 2008. This legislation, among other things, allows for revisions in the adoption incentive programs, including the Title IV-E adoption assistance. Our findings suggest that such changes in adoption and foster care policies may have significant long-term effects on both child adoption rates and ART utilization.

Since the early 1980s, treatment options for infertility have become more widely available due to rapid technological change and legislation in some states regarding infertility treatment insurance. The Family Building Act, introduced in the House of Representatives in 2009, proposes to expand the coverage for infertility treatment to those who work for firms that self-insure. In Europe, ART is being considered as a possible tool to increase fertility rates in an effort to mitigate the consequences of population aging (Grant et al. 2006). In the future, new technological advances and increased competition among infertility clinics are expected to continue reducing the ART cost and to accelerate its increasing use. The question as to whether these developments in ART markets might lead to unintended consequences in child adoptions is left for ongoing and future research.

While ART provides individuals with the possibility to have children who are biologically linked to themselves, concerns have been voiced about the demanding nature of these types of treatments, not only financially but also physically and psychologically, considering the low odds of success. At the same time, keeping children in foster care poses social costs in terms of social services and adverse long-term child outcomes. Therefore, it might be socially desirable to implement policies to encourage child adoption at the expense of ART. In light of these issues, the regulations on child adoption and ART utilization need to be reviewed to level the playing field as adoptions are subject to tougher restrictions. The issue of ART use lends itself to several controversial questions concerning who should have access to such technologies, who should pay for the treatments, and whether the providers should be regulated. These considerations need to be addressed given that ART not only impact women's birth and health outcomes but also potentially their marriage, education, and career decisions.

ABBREVIATIONS

AFCARS: Adoption and Foster Care Analysis and Reporting System

ART: Assisted Reproductive Technology

ASFA: Adoption and Safe Families Act

CDC: Centers for Disease Control and Prevention

FE: Fixed Effects

FE-2SLS: Fixed Effects Two-Stage Least Squares Method

IVs: Instrumental Variables

NACAC: North American Council on Adoptable Children

NSAP: National Survey of Adoptive Parents

NSFG: National Survey of Family Growth

OLS: Ordinary Least Squares

SART: Society for ART

SSA: Social Security Act

TPR: Termination of Parental Rights

REFERENCES

Allen, M., and M. Bissell. "Safety and Stability for Foster Children: The Policy Context." The Future of Children, 14, 2004, 49-73.

Angrist, J. D., and G. Imbens. "Two-Stage Least Squares Estimation of Average Causal Effects in Models with Variable Treatment Intensity." Journal of the American Statistical Association, 90(430), 1995, 431-42.

Angrist, J. D., and W. N. Evans. "Schooling and Labor Market Consequences of the 1970 State Abortion Reforms," in Research in Labor Economics, Vol. 18, edited by S. Polachek. Greenwich: JAI Press, 1999.

Archibold, R. C. "Octuplets, 6 Siblings, and Many Questions." New York Times, February 4, 2009.

Avery, R. J. "Adoption Assistance under P.L. 96-272: A Policy Analysis." Children and Youth Services Review, 20(1/2), 1998, 29-55.

Avery, R. J., and D. Mont. "Financial Support of Children Involved in Special Needs Adoption: A Policy Evaluation." Journal of Policy Analysis and Management, 11(3), 1992, 419-41.

Baccara, M., A. Collard-Wexler, L. Felli, and L. Yariv. "Gender and Racial Biases: Evidence from Child Adoption." Unpublished Working Paper, 2009.

Barth, R. P., C. K. Lee, J. Wildfire, and S. Guo. "A Comparison of the Governmental Costs of Long-Term Foster Care and Adoption." Social Service Review, 80(1), 2006, 127-58.

Bernal, R., L. Hu, C. Moriguchi, and E. Nagypal. "Child Adoption in the United States: Historical Trends and the Determinants of Adoption Demand and Supply, 1951-2002." Unpublished Working Paper, 2008.

Bitler, M. P. "Effects of Increased Access to Infertility Treatment on Infant and Child Health: Evidence from Health Insurance Mandates." Unpublished Working Paper, 2008.

Bitler, M. P., J. Gelbach, and H. Hoynes. "Welfare Reform and Children's Living Arrangements." Journal of Human Resources, 41(1), 2006, 1-27.

Bitler, M. P., and L. Schmidt. "Health Disparities and Infertility: Impacts of State-Level Insurance Mandates." Fertility and Sterility, 85(4), 2006, 858-65.

--. "Utilization of Infertility Treatments: The Effects of Insurance Mandates." Unpublished Working Paper, 2009.

Bitler, M. P., and M. Zavodny. "Did Abortion Legalization Reduce the Number of Unwanted Children? Evidence from Adoptions." Perspectives on Sexual and Reproductive Health, 34(1), 2002, 25-33.

Blome, W. W. "What Happens to Foster Kids: Educational Experiences of a Random Sample of Foster Care Youth and a Matched Group of Non-Foster Care Youth." Child and Adolescent Social Work Journal, 14(1), 1997, 41-53.

Bower, J. W., and R. Laws. Forever Families: Support for Families of Children with Special Needs--A Policy Analysis of Adoption Subsidy Programs in the United States. St. Paul, MN: North American Council on Adoptable Children, 2002.

Buckles, K. "Stopping the Biological Clock: Infertility Treatments and the Career-Family Tradeoff." Unpublished Working Paper, 2007.

--. "Understanding the Returns to Delayed Childbearing for Working Women." American Economic Review Papers and Proceedings, 98(2), 2008, 403-7.

Bundorf, M. K., M. Henne, and L. Baker. "Mandated Health Insurance Benefits and the Utilization and Outcomes of Infertility Treatments." NBER Working Paper No. 12820, 2007.

Caplan, A. "Ethics and Octuplets: Society is Responsible." Philadelphia Inquirer, February 6, 2009.

Caucutt, E. M., N. Guner, and J. Knowles. "Why Do Women Wait? Matching, Wage Inequality, and the Incentives for Fertility Delay." Review of Economic Dynamics, 5, 2002, 815-55.

CDC. Assisted Reproductive Technology Success Rates: National Summary and Fertility Clinic Reports. Atlanta, GA: CDC, various years.

Chandra, A., J. C. Abma, P. Maza, and C. A. Bachrach. "Adoption, Adoption Seeking, and Relinquishment for Adoption in the United States." Vital Health Statistics, No. 306, 1999.

Chandra, A., G. M. Martinez, W. D. Mosher, J. C. Abma, and J. Jones. "Fertility, Family Planning, and Reproductive Health of US Women: Data from the 2002 National Survey of Family Growth." Vital Health Statistics, 23(25), 2005, 1-160.

Cohen, M. "A Search for a Surrogate Leads to India." Wall Street Journal, October 2009, p. D1.

Collins, J. "Cost-Effectiveness of In Vitro Fertilization." Seminars in Reproductive Medicine, 19(3), 2001, 279-89.

Dalberth, B., D. Gibbs, and N. Berkman. Understanding Adoption Subsidies: An Analysis of AFCARS Data. Washington, DC: Office of the Assistant Secretary for Planning and Evaluation (ASPE), US Department of Health and Human Services, 2005.

Doyle, J. J. Jr. "Child Protection and Child Outcomes: Measuring the Effects of Foster Care." American Economic Review, 97(5), 2007, 1583-610.

Doyle, J. J. Jr., and H. E. Peters. "The Market for Foster Care: An Empirical Study of the Impact of Foster Care Subsidies." Review of Economics of the Household, 5(4), 2007, 329-51.

Duncan, B., and L. Argys. "Economic Incentives and Foster Care Placement." Southern Economic Journal, 74(1), 2007, 114-42.

Gennetian, L. A. "The Supply of Infants Relinquished for Adoption: Did Access to Abortion Make a Difference?" Economic Inquiry, 37(3), 1999, 412-31.

Gibbs, D. A., B. T. Dalberth, N. D. Berkman, and D. Weitzenkamp. "Determinants of Adoption Subsidies." Adoption Quarterly, 9(2/3), 2006, 63-80.

Goldin, C., and L. F. Katz. "The Power of the Pill: Contraceptives and Women's Career and Marriage Decisions." Journal of Political Economy, 110(4), 2002, 730-70.

Grant, J., S. Hoorens, F. Gallo, and J. Cave. Should ART Be Part of a Population Policy Mix? Cambridge, UK: RAND Documented Briefing DB-507-FER, 2006.

Gravois, J. "Bringing Up Babes. Why Do Adoptive Parents Prefer Girls?" Slate, January 2004.

Groze, V. "Adoption and Single Parents: A Review." Child Welfare, 70(3), 1991, 321-32.

Hamilton, B. H., and B. McManus. "The Effects of Insurance Mandates on Choices and Outcomes in Infertility Treatment Markets." Unpublished Working Paper, 2010.

Hammoud, A. O., M. Gibson, J. Stanford, G. White, D. T. Carrell, and M. Peterson. "In Vitro Fertilization Availability and Utilization in the United States: A Study of Demographic, Social, and Economic Factors." Fertility and Sterility, 91(5), 2009, 1630-35.

Hansen, M. E. Age of Children at Adoption and Time from Termination of Parental Rights to Adoption. AFCARS Adoption Data Research Brief, No. 2. St. Paul, MN: North American Council on Adoptable Children, 2006.

--. "Using Subsidies to Promote the Adoption of Children from Foster Care." Journal of Family and Economic Issues, 28(3), 2007, 377-93.

-- "The Distribution of a Federal Entitlement: The Case of Adoption Assistance." Journal of SocioEconomics, 37, 2008a, 2427-42.

--. "The Value of Adoption." Adoption Quarterly, 10(2), 2008b, 65-87.

Hansen, M. E., and B. A. Hansen. "The Economics of Adoption of Children from Foster Care." Child Welfare, 85(3), 2006, 559-83.

Hansen, M. E., and D. Pollack. "Unintended Consequences of Bargaining for Adoption Assistance Payments." Family Court Review, 43(3), 2005, 495-511.

Henne, M. B., and M. K. Bundorf. "Insurance Mandates and Trends in Infertility Treatments." Fertility and Sterility, 89(1), 2008, 66-73.

Kane, T. J., and D. Staiger. "Teen Motherhood and Abortion Access." Quarterly Journal of Economics, 111 (2), 1996, 467-506.

Kreider, R. M. Adopted Children and Stepchildren: 2000. Census Special Reports, CENSR-6RV. Washington, DC: US Census Bureau, 2003.

Machado, M. P., and A. Sanz-de-Galdeano. "Coverage of Infertility Treatment and Fertility Outcomes: Do Women Catch Up?" Unpublished Working Paper, 2009.

National Center for Health Statistics (NCHS). "Trends in Twin and Triplet Births: 1980-97." Vital Health Statistics, 47, 1999, 1099-120.

National Data Archive on Child Abuse and Neglect (NDACAN). AFCARS, 2000-Present, User's Guide and Codebook. Ithaca, NY: NDACAN, 2007.

Neumann, P. J. "Should Health Insurance Cover IVF? Issues and Options." Journal of Health Politics, Policy and Law, 22(5), 1997, 1215-39.

Neumann, P. J., S. D. Gharib, and M. C. Weinstein. "The Cost of a Successful Delivery with IVF." New England Journal of Medicine, 331(4), 1994, 239-43.

Schmidt, L. "Infertility Insurance Mandates and Fertility." American Economics Review Papers and Proceedings, 95(2), 2007, 204-8.

Staniec, J. F. O., and N. J. Webb. "Utilization of Infertility Services: How Much Does Money Matter?" Health Services Research, 42(3), 2007, 971-89.

Stephen, E. H., and A. Chandra. "Use of Infertility Services in the United States: 1995." Family Planning Perspectives, 32, 2000, 132-7.

Swann, C. A., and M. S. Sylvester. "The Foster Care Crisis: What Caused Caseloads to Grow?" Demography, 43(2), 2006, 309-35.

US DHHS. Cost of Adopting: Factsheet for Families. Washington, DC: Child Welfare Information Gateway, US DHHS, 2004a.

--. How Many Children were Adopted in 2000 and 2001? Washington, DC: Child Welfare Information Gateway, US DHHS, 2004b.

--. The AFCARS Report 14: Preliminary FY 2006 Estimates as of January 2008. Washington, DC: Child Welfare Information Gateway, US DHHS, 2008.

Van Den Akker, O. B. A. "Adoption in the Age of Reproductive Technology." Journal of Reproductive and Infant Psychology, 19(2), 2001, 147-59.

Van Voorhis, B. J. "In Vitro Fertilization." New England Journal of Medicine, 356(4), 2007, 379-86.

Waldfogel, J. "Welfare Reform and the Child Welfare System." Children and Youth Services Review, 26, 2004, 919-39.

(1.) Another option is surrogacy if there are no problems with male fertility. In surrogacy, a fertile surrogate mother is artificially inseminated with sperm and then carries the baby to term. In most states, there are no laws regarding surrogacy arrangements, and in others, surrogacy agreements are prohibited altogether. Moreover, this is an expensive method as the average estimated cost ranged between $70,000 and $130,000 in 2009 (Cohen 2009), which is considerably higher than the costs associated with the other two alternatives discussed here.

(3.) Alternatively, one can estimate the effect of ART use on child adoptions. One possibility would be to exploit the infertility insurance mandates that led to the exogenous variation in ART utilization. Unfortunately, this approach is not feasible due to data limitations; during the sample period, only two states, New Jersey and Louisiana, adopted such mandates. Thus, there is not enough variation to obtain reliable estimates.

(4.) The federal adoption tax credit, offered to families adopting a child with special needs from foster care, was set at $13,170 per child for the 2010 tax year. In addition, some employers provide adoption benefits to their employees, and several agencies make adoption loans and grants available to couples.

(5.) Bernal et al. (2008) provide an excellent review of child adoption in the context of institutional background and historical trends in the United States for the period of 1951-2002.

(6.) CDC ART reports are publicly available online at http://www.cdc.gov/ART.

(7.) In 2006, the most commonly performed ART cycles were transfers using fresh non-donor eggs/embryos (72%), followed by frozen non-donor eggs/embryos (16%), and fresh and frozen donor eggs/embryos (8% and 4%, respectively). Below, we provide robustness checks based on these alternative measures of ART cycles performed.

(8.) Normalizing variables by 1,000 births instead of 1,000 women leaves our results unchanged.

(9.) As it is not the primary focus of our analysis, the number of ART clinics is treated as exogenously given. Hamilton and McManus (2010) provide an analysis of the market structure for infertility treatment and examine the determinants of new ART clinic entry.

(10.) Information on how to obtain permission to access the AFCARS database can be found at http://www.ndacan.cornell.edu/NDACAN.

(11.) AFCARS data description states, "Other adoptions, such as those involving children who were not in the public foster care system and were placed for adoption by tribal or private agencies, are voluntarily reported to AFCARS.

(12.) For example, the 2005 data corresponds to October 2004 through September 2005. The 2005 AFCARS adoption file is then merged with the 2005 CDC ART report which corresponds to the 2005 calendar year. Note that this accords well, as we focus on the effect of adoption on ART utilization, as explained in further detail below.

(13.) Mainly due to missing information on adoption subsidies, we exclude Nevada and New York from our sample, together with 14 other state-year observations. The OLS and FE estimations can be carried out using data on 400 observations. However, the FE-2SLS estimations require information on the number and composition of the children in foster care waiting to be adopted. To allow for ease of comparison, we report the OLS and FE results for the sample with 378 observations. The results based on the full sample of 400 observations are virtually identical to those reported below.

(14.) Bernal et al. (2008) report that the share of international adoptions in total adoptions rose from 17.2% in 1996 to 21.7% in 2002. The share of public agency adoptions also increased from 37% in 1996 to 44.2% in 2002. Finally, domestic private adoptions decreased from 45.8% in 1996 to 34.1% in 2002.

(15.) Following Dalberth, Gibbs, and Berkman (2005), we set the monthly subsidy amount to missing in the few cases when they were reported as greater than $10,000.

(16.) In reality, some women may concurrently pursue ART and adoption. Our estimates for the effects of adoption on ART utilization correspond to those women who switch from ART exclusively to adoption or vice versa.

(17.) The probabilities are represented by integration over the distribution. For example, we have [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], where f denotes the joint distribution.

(18.) We assume that in both ART and adoption markets, the equilibrium quantity is determined by the demand.

(19.) In our sample, a substantial portion of the variation in adoption subsidies is due to within-state variations over time. The average monthly adoption subsidy is $473 with an overall standard deviation of $143. The between-states standard deviation is $127 and the within-states standard deviation is $66. The average basic adoption assistance rate (also measured in constant 2000 dollars) is $381 with an overall standard deviation of $119. The between-states standard deviation is $111, and the within-states standard deviation is $44.

(20.) We also tried to include a measure of the racial composition for the foster children waiting to be adopted as an additional IV, but this variable was never statistically significant.

(21.) The full set of results can be obtained from the authors upon request.

(22.) To interpret the reported elasticities on these two variables, one should consider a one-unit change which implies a doubling of the variable given that they are expressed as fractions rather than percentages.

(23.) We thank an anonymous referee for suggesting this idea.

GULCIN GUMUS and JUNGMIN LEE*

* Part of the data used here were made available by the National Data Archive on Child Abuse and Neglect (NDACAN), Cornell University, Ithaca, NY, and have been used with permission. Data from the Adoption and Foster Care Analysis and Reporting System (AFCARS) were originally collected by the Children's Bureau. Funding for the AFCARS project was provided by the Children's Bureau, Administration on Children, Youth and Families, Administration for Children and Families, U.S. Department of Health and Human Services. The collector of the original data, the funder, the Archive, Cornell University and their agents or employees bear no responsibility for the analyses or interpretations presented here. We gratefully acknowledge the staff at NDACAN, in particular Michael Dineen, for their assistance. We thank Mary E. Hansen for providing us with the data on adoption assistance rates as well as for her suggestions. We also thank the conference participants at the 2009 Latin American Meetings of the Econometric Society, the 2010 Annual Meeting of the American Economic Association, and the 2010 Biennial Conference of the American Society of Health Economists for helpful comments. We are grateful to Mabel Andalon, Marianne P. Bider, Luca Bossi, Kasey Buckles, Kate Bundorf, Carlos A. Flores, Lisa A. Gennetian, Chiaki Moriguchi, Elizabeth Peters, Tracy L. Regan, Anna Sanzde-Galdeano, Justin Wolfers, and two anonymous referees for their constructive feedback and to Jennifer R. Nimmo for research assistance.

Gumus: Assistant Professor, Department of Management Programs, College of Business, Florida Atlantic University, 777 Glades Road, Fleming Hall, Room 313, Boca Raton, FL 33431; IZA, Bonn, Germany. Phone 1-561297-2115, Fax 1-561-297-2675, E-mail ggumus@fau.edu

Lee: Associate Professor, School of Economics, Sogang University, 1 Sinsu-dong, Mapo-gu Seoul 121-742, South Korea; IZA, Bonn, Germany. Phone 82-2-7058504, Fax 82-2-705-8180, E-mail junglee@sogang.ac.k

doi: 10.1111/j.1465-7295.2011.00401.x

TABLE 1
Summary Statistics (N = 378 Unless Indicated Otherwise)

 Standard
 Mean Deviation
Child adoptions from foster care (a)
 Adoptions 711 885
 Adoptions per 1,000 women aged 25-44 0.994 0.524
 Adoptions by adoptive mothers 35+ 562 725
 Adoptions by adoptive mothers 35+ per 1,000 1.456 0.807
 women aged 35-44
 Adoptions of children 5+ 408 515
 Adoptions of children 5+ per 1,000 women 0.588 0.351
 aged 25-44
 Average monthly adoption subsidy ($100) (b) 4.728 1.428
 Average monthly basic adoption assistance 3.812 1.191
 rate for 2-year-olds ($100) (b,c)
 Children in foster care 12,374 19,328
 Children in foster care waiting to be 2,380 2,841
 adopted
 Children in foster care waiting to be 3.262 1.816
 adopted per 1,000 women aged 25-44
 Children in foster care waiting to be 6.106 3.647
 adopted per 1,000 women aged 35-44
 Percentage of girls in foster care waiting 0.466 0.048
 to be adopted
 Percentage of children 0-4 in foster care 0.291 0.073
 waiting to be adopted
 Related adoptions 214 532
 Related adoptions per 1,000 women aged 0.259 0.275
 25-44
Child adoptions from abroad (d)
 International adoptions 366 343
 International adoptions per 1,000 women 0.499 0.203
 aged 25-44
ART utilization (e)
 ART clinics 7.503 10.089
 ART clinics per 1,000 women aged 25-44 0.009 0.006
 ART clinics per 1,000 women aged 35-44 0.017 0.013
 ART cycles performed 1,477 2,083
 ART cycles performed per 1,000 women aged 1.743 1.778
 25-44
 ART cycles performed for women 35+ 789 1,252
 ART cycles performed for women 35+ per 1.723 2.247
 1,000 women aged 35-44
 ART cycles performed using nondonor eggs 1,747 2,447
 (fresh and frozen)
 ART cycles performed using nondonor eggs 2.045 1.997
 (fresh and frozen) per 1,000 women
 aged 25-44
 ART cycles performed using donor eggs 225 407
 (fresh and frozen)
 ART cycles performed using donor eggs 0.223 0.198
 (fresh and frozen) per 1,000 women aged
 25-44
State characteristics
 Total number of women aged 25-44 (f) 801,357 920,363
 Total number of women aged 35-44 (f) 426,134 480,674
 Health insurance mandate on infertility 0.280 0.450
 treatments
 Percentage of resident population under age 0.721 0.063
 65 with private health insurance
 coverage (h)
 Income per capita ($1,000) (b,h) 29.306 4.828
 Percentage of resident population (25+) 0.264 0.053
 with at least a college degree (i)
 Percentage black (j) 0.170 0.258
 Female labor force participation rate (h) 0.612 0.042
 Teen birth rate (births to teenage mothers 0.111 0.028
 under age 20 as a percent of total
 births) (k)
 Ratio of the female resident population 0.535 0.026
 aged 35-44 to that aged 25-44 (f)

(a) AFCARS Adoption and Foster Care Files. Figures reflect
unrelated child adoptions unless indicated otherwise.

(b) A11 figures are expressed in constant 2000 US$.

(c) North American Council on Adoptable Children, NACAC (N = 373).

(d) U.S. Department of State.

(e) CDC ART Reports. Figures reflect ART cycles performed using
fresh nondonor eggs only unless specified otherwise.

(f) U.S. Census Bureau.

(g) Bitler (2007) and Schmidt (2007).

(h) Current Population Surveys.

(i) U.S. Bureau of Economic Analysis.

(j) National Center for Health Statistics.

(k) National Vital Statistics Reports.

TABLE 2
OLS and FE Regression Results, 1999-2006

(A) Dependent variable: ART cycles performed
(per 1,000 women)

 Adoptions (per 1,000 women) 0.2058 * -0.2189 ***
 (0.1167) (0.0728)
 0.1174# -0.1249#
 R-squared 0.7129 0.5988

(B) Dependent variable: ART cycles performed
for women 35+ (per 1,000 women)

 Adoptions by adoptive mothers 35+ (per 0.1332 -0.1951 ***
 1,000 women) (0.1078) (0.0574)
 0.1126# -0.1648#
 R-squared 0.7508 0.6189

(C) Dependent variable: ART cycles performed
(per 1,000 women)

 Adoptions of children 5+ (per 1,000 women) 0.3578 * -0.3044 ***
 (0.1956) (0.0864)
 0.1207# -0.1027#
 R-squared 0.7152 0.5996

Estimation method OLS FE
Year FE Yes Yes
State FE No Yes
Controls Yes Yes
Observations 378 378
States 49 49

Notes: For each regression, we report the estimated coefficient
followed by estimated standard error in parentheses and implied
elasticity in italics. All specifications include state and year
FE as well as the following controls: number of ART clinics (per
1,000 women), indicator for health insurance mandate on
infertility treatment, percentage covered by private health
insurance, real income per capita, percentage with at least a
college degree, percentage black, teen birth rate. female labor
force participation rate, and ratio of the female resident
population between the ages of 35 and 44 to that between tile
ages of 25 and 44. Estimates are weighted using total number of
women and standard errors are clustered by state. R-squared
values in column (2) are calculated without taking the
contribution of the state FE into account.

* and *** indicate statistical significance at 10% and 1%,
respectively.

Note: For each regression, we report the estimated coefficient
followed by estimated standard error in parentheses and implied
elasticity indicated with #.

TABLE 3
First-Stage Regression Results, 1999-2006

 Dependent Variable:
 Adoptions

 IV Set I IV Set II
 (I) (2)

Average monthly adoption subsidy 0.0430 **
 ($100) (0.0209)
 0.2045#

Average monthly basic adoption 0.0353 *
 assistance rate for 2-year-olds ($100) (0.0203)
 0.1375#

Number of children in foster care 0.1929 *** 0.1939 ***
 waiting to be adopted (per 1,000 (0.0359) (0.0363)
 women) 0.6432# 0.6349#

% girls in foster care waiting to be 1.6628 *** 1.7318 ***
 adopted (0.2980) (0.3394)
 0.7921# 0.8225#

% children 0-4 in foster care waiting to -0.6429 -0.7865 *
 be adopted (0.3958) (0.4169)
 -0.1914# -0.2324#

Observations 378 373

States 49 49

R-squared 0.6373 0.6363

F-statistics for the excluded instruments 19.04 17.37

Overidentification test of all excluded 0.4045 0.2816
 instruments

(p value for the Hansen J-statistic)

 Dependent Variable:
 Adoptive Mothers 35+

 IV Set I IV Set II
 (3) (4)

Average monthly adoption subsidy 0.0675 **
 ($100) (0.0321)
 0.2192#

Average monthly basic adoption 0.0619 *
 assistance rate for 2-year-olds ($100) (0.0328)
 0.165#

Number of children in foster care 0.1644 *** 0.1648 ***
 waiting to be adopted (per 1,000 (0.0315) (0.0321)
 women) 0.7022# 0.6918#

% girls in foster care waiting to be 2.9271 *** 3.0680 ***
 adopted (0.4383) (0.5017)
 0.9542# 0.9972#

% children 0-4 in foster care waiting to -1.4089 ** -1.6628 **
 be adopted (0.6009) (0.6247)
 -0.2871# -0.3363#

Observations 378 373

States 49 49

R-squared 0.6507 0.6521

F-statistics for the excluded instruments 20.61 19.82

Overidentification test of all excluded 0.4417 0.2760
 instruments

(p value for the Hansen J-statistic)

 Dependent Variable:
 Adopted Children 5+

 IV Set I IV Set II
 (5) (6)

Average monthly adoption subsidy 0.0358 **
 ($100) (0.0138)
 0.2881#

Average monthly basic adoption 0.0281 **
 assistance rate for 2-year-olds ($100) (0.0129)
 0.1853#

Number of children in foster care 0.1503 *** 0.1520 ***
 waiting to be adopted (per 1,000 (0.0270) (0.0273)
 women) 0.8482# 0.8424#

% girls in foster care waiting to be 1.4305 *** 1.5065 ***
 adopted (0.2162) (0.2325)
 1.1534# 1.2111#

% children 0-4 in foster care waiting to -0.8698 *** -0.9980 ***
 be adopted (0.2655) (0.2704)
 -0.4384# -0.4992#

Observations 378 373

States 49 49

R-squared 0.6885 0.6923

F-statistics for the excluded instruments 17.80 19.46

Overidentification test of all excluded 0.4090 0.2847
 instruments

(p value for the Hansen J-statistic)

Notes: For each explanatory variable, we report the estimated
coefficient followed by estimated standard error in parentheses
and implied elasticities in italics. All specifications include
state and year FE as well as all the following controls: number
of ART clinics (per 1,000 women), indicator for health insurance
mandate on infertility treatment, percentage covered by private
health insurance, real income per capita, percentage with at
least a college degree, percentage black, teen birth rate,
female labor force participation rate, and ratio of the female
resident population between the ages of 35 and 44 to that
between the ages of 25 and 44. Estimates are weighted using total
number of women and standard errors are clustered by state.
R-squared values are calculated without taking the contribution
of the state FE into account.

*, **, *** indicate statistical significance at 10%, 5%, and 1%,
respectively.

Note: For each explanatory variable, we report the estimated
coefficient followed by estimated standard error in parentheses
and implied elasticities indicated with #.

TABLE 4
FE 2SLS Regression Results, 1999-2006

 (1) (2)

(A) Dependent variable:
ART cycles performed (per
1,000 women)

 Adoptions (per 1,000 -0.2602 *** -0.2452 ***
 women) (0.0704) (0.0776)
 -0.1484# -0.1376#
 R-squared 0.5980 0.6059

(B) Dependent variable:
ART cycles performed for
women 35+ (per 1,000 women)

 Adoptions by adoptive -0.2232 *** -0.2192 ***
 mothers 35+ (per 1,000 (0.0551) (0.0587)
 women) -0.1886# -0.1818#
 R-squared 0.6182 0.6226

(C) Dependent variable:
ART cycles performed (per
1,000 women)

 Adoptions of children 5+ -0.3337 *** -0.3135 ***
 (per 1,000 women) (0.0869) (0_0965)
 -0.1125# -0.1040#
 R-squared 0.5994 0.6069

 Estimation method FE 2SLS (IV set I) FE 2SLS (IV set II)

 Observations 378 373

 States 49 49

Notes: For each regression, we report the estimated coefficient
followed by estimated standard error in parentheses and implied
elasticity in italics. All specifications include state and year
FE as well as all the following controls: number of ART clinics
(per 1,000 women), indicator for health insurance mandate on
infertility treatment, percentage covered by private health
insurance, real income per capita, percentage with at least a
college degree, percentage black, teen birth rate, female labor
force participation rate, and ratio of the female resident
population between the ages of 35 and 44 to that between the
ages of 25 and 44. IV set I and II are presented in Table 3.
Estimates are weighted using total number of women and standard
errors are clustered by state. R-squared values are calculated
without taking the contribution of the state FE into account.

*** indicates statistical significance at 1%.

Note: For each regression, we report the estimated coefficient
followed by estimated standard error in parentheses and implied
elasticity indicated with #.

TABLE 5
Robustness Checks Using FE Models,
1999-2006

(A) Dependent variable: ART cycles performed
(per 1,000 women)

 Unrelated adoptions (per 1,000 women) -0.2188 ***
 (0.0750)
 -0.1248#

 Related adoptions (per 1,000 women) -0.0015
 (0.2502)
 -0.0002#

 R-squared 0.5986

 Observations 378

 States 49

(B) Dependent variable: ART cycles performed
(per 1,000 women)

 Unrelated foster care adoptions (per -0.2178 ***
 1,000 women)
 (0.0732)
 -0.1243#
 International adoptions (per 1,000 -0.7611 **
 women) (0.3707)
 -0.2180#

 R-squared 0.6074

 Observations 378

 States 49

(C) Dependent variable: ART cycles performed
using fresh and frozen non-donor eggs (per
1,000 women)

Adoptions (per 1,000 women) -0.2657 ***
 (0.0837)
 -0.1291#

 R-squared 0.6387

 Observations 378

 States 49

(D) Dependent variable: ART cycles performed
using fresh and frozen donor eggs (per 1,000
women)

 Adoptions (per 1,000 women) -0.0217
 (0.0276)
 -0.0971#

 R-squared 0.5413

 Observations 378

 States 49

Notes: For each regression, we report the estimated coefficient
followed by estimated standard error in parentheses
and implied elasticity in italics. All specifications include
state and year FE as well as the following controls: number
of ART clinics (per 1,000 women), indicator for health
insurance mandate on infertility treatment, percentage covered
by private health insurance, real income per capita,
percentage with at least a college degree, percentage black,
teen birth rate, female labor force participation rate, and
ratio of the female resident population between the ages of
35 and 44 to that between the ages of 25 and 44. Estimates
are weighted using total number of women and standard
errors are clustered by state. R-squared values are calculated
without taking the contribution of the state FE into
account.

** and *** indicate statistical significance at 5% and 1%,
respectively.

Note: For each regression, we report the estimated coefficient
followed by estimated standard error in parentheses and
implied elasticity indicated with #.