The Economics of Adoption: Project Summary

Adoption, as an alternative to child bearing, is a widely accepted means of forming a family in modern societies. Today, adopted children comprise roughly 2.5% of all children in the United States. Data indicate that Americans adopt more children, domestically and internationally, in total and per thousand births, than any other country in the world. Who adopts children and who places children for adoption? In spite of rapidly declining fertility rates worldwide, why do people adopt children who are not biologically related to them? Is the number of adoptions in the U.S. rising? Despite the quantitative importance of adoption and the welfare implications associated with it, unlike sociologists and demographers, economists have paid little attention to this subject.

Child adoption, however, is no less an economic issue than is child bearing in the classical Becker model. In fact, some of the peculiarities of adoption should be of particular interest to economists. For example, in an “adoption market,” prospective parents and adoptable children are matched by intermediaries using primarily non-price mechanisms; adoption consists of different types of adoption that are imperfect substitutes for one another (e.g. domestic vs. intercountry adoption); and, adoptive parents face costs and benefits that are specific to adoption (e.g., uncertainty over child characteristics, long waiting time). Even though there is an extensive economic literature on fertility decisions, to our knowledge, child adoption has never been modeled explicitly as an alternative to child bearing.

To fill this gap, we propose to undertake the first systematic economic analysis of child adoption. Our project has three components. First, we construct historical trends in adoption in the U.S. using aggregate-level data and correlate them with trends in key demographic variables to motivate research and policy questions. Second, we investigate the demand for and supply of adoption separately, using two individual-level data sets. Third, we develop and calibrate a formal model of adoption seeking and conduct counterfactual policy experiments. The intellectual merit of our project is twofold. Empirically, we provide new evidence on adoption demand and supply, using rigorous econometric analysis and new data sources, and advance our understanding of their underlying determinants. Theoretically, we build a dynamic model of adoption seeking, in which women can form a family by child bearing or adoption, and provide policy analysis and derive welfare implications using computational methods.

More specifically, we estimate individuals’ propensity to adopt and to relinquish children, using two complementary data sources: the National Survey of Family Growth (NSFG) Cycles 1-6, 1973-2002, and the Survey of Income and Program Participation (SIPP), 1984-2004. We improve on previous research by (i) forming and testing competing hypotheses concerning adoption demand and supply using logistic regressions and providing the first multivariate analysis of the determinants of adoption supply, (ii) introducing so far unexploited SIPP data to study adoption demand and pooling NSFG data across cycles to generate a larger sample, and (iii) using time-series variations as well as the variations across adoption types in the data to evaluate various hypotheses. Furthermore, we develop a dynamic model of adoption seeking in which heterogeneous women make labor supply, childbearing, and adoption decisions given their time and resource constraints. We calibrate the model to the relevant moments in the micro data, and use it to simulate the effects of changes in the supply of adoptable children, adoption law reforms, progress in reproductive technology, changes in labor market policies, and changes in marriage market dynamics on adoption market outcomes, such as the number and types of adopted children and the characteristics of adoptive parents. From these experiments, we expect to gain better understanding of how different policy interventions can influence adoption and fertility decisions of individuals.

We expect our project to have broader impacts through (i) publishing the findings from this project in appropriate outlets to further the scientific understanding of adoption, (ii) involving graduate students as research assistants and exposing them to empirical and theoretical methods proposed in this project, (iii) presenting the results in interdisciplinary conferences and promote collaborations among economists, demographers, and sociologists, (iv) building data sets, publicly available to researchers interested in adoption, that contain historical time-series and machine-readable micro data from the NSFG and SIPP, and (v) preparing policy briefings useful to policy-makers and accessible to broader audiences.

The Economics of Child Adoption

1 Motivation and Overview

Adoption, as an alternative to childbearing, is a widely accepted means of forming a family in many modernsocieties. Today, adopted children comprise roughly 2.5% of all children in the U.S. (U.S. Census Bureau(2000)). The number of foreign-born children adopted by U.S. citizens has more than tripled in the last�fteen years (U.S. Dept of Homeland Security (2005)), generating much press coverage. 1 Data indicate thatthe U.S. adopts more children, domestically and internationally, than any other country in the world bothin total and per 1,000 births (Selman (2002)). These facts pose important questions. Who adopts children,and who places children for adoption? In light of the declining demand for children in general in the U.S.and in other developed countries, why do people demand children who are not biologically related to them?Why has the number of foreign-born adoptions risen so much recently in the U.S.? Is it due to historicalor cultural reasons speci�c to the U.S., or can it be explained more generally by demographic and economiccharacteristics?

Despite the quantitative signi�cance and potentially important welfare implications of the answers tothese questions, economists have paid little attention to child adoption. In fact, adoption has been studiedalmost exclusively by sociologists, demographers, and psychologists (Fisher 2003), and there virtually is noeconomic literature on adoption.2 This is due partly to the scarcity of data, but primarily to the lack ofinterest on the part of economists. Child adoption, however, is no less an economic issue than is childbearingin the classical Becker (1981) model. Adoption is an outcome of the decision making process undertaken byindividuals who weigh its costs and bene�ts, and as such, it �ts squarely in the �eld of family economics.Moreover, adoption is a way to reallocate children between di�erent parents, and given the recent economicresearch suggesting that early childhood investments by parents are critical in the formation of human capital(see Cunha and Heckman (2004), Cunha et al. (2006), Bernal (2006), and Bernal and Keane (2006)), itcould have a large impact on the accumulation of human capital by the a�ected children. Even thoughthere is a sizeable economic literature on fertility decisions (e.g., Becker and Barro (1988), Eckstein and andWolpin (1989), or, more recently, Caucutt et al. (2002), and Greenwood et al. (2003)), to our knowledge,child adoption has never been modeled explicitly as an alternative to childbearing. The primary goal of thisproject is to undertake the �rst systematic economic analysis of child adoption.

Child adoption has a number of features distinct from childbearing that require careful investigation.Just to discuss a few, unlike childbearing, individuals on the demand side and the supply side of children donot coincide, giving rise to an �adoption market� in which the two sides are matched by intermediaries usingprimarily non-price mechanisms. As the adoption market deviates greatly from the standard competitivemarket, institutional factors (e.g., adoption agencies, laws, and government policies) play a major role indetermining its outcomes. In addition, as we discuss below, there are several types of adoption that areimperfect substitutes for one another in the adoption market. Most notably, the characteristics of adoptablechildren may di�er systematically across di�erent sources of supply (foreign-born versus domestic foster carechildren), which may result in excess demand in one category and excess supply in the other. Furthermore,adoptive parents face costs and bene�ts that are speci�c to adoption. For example, unlike childbearing,non-economic motives such as religion or altruism may be important in driving adoption demand. Adoptiveparents may incur monetary, time, and psychic costs that are speci�c to adoption (e.g., long expected waitingtime, uncertainty over quality of the match between prospective parent and a child, and the possible stigmaassociated with adoption (Fisher, 2003)).

Our proposed research project has three components. First, after providing a comprehensive review ofthe literature and available data sources, we document historical trends using macro data and correlate thetrends in adoptions with the trends in key demographic variables to motivate research questions. Second,

1For example, �Madonna Urges More People to Adopt from Africa,� January 12, 2007, Reuters News; �China TightensAdoption Rules for Foreigners,� December 20, 2006, New York Times; and �Rules Set to Change on Foreign Adoptions,�November 2, 2006, Wall Street Journal.

2Notable exceptions are Landes and Posner (1978), Medo� (1993), Bitler and Zavodny (2002), and Hansen and Hansen(2006). Case et al. (2000, 2001), Plug (2004), Plug et al. (2003, 2005), Bjorklund et al. (2006), Das and Sjogren (2002), andSacerdote (2000) provide child outcome studies utilizing data on adopted children, but their primary focus is not on adoptionper se.

1

we investigate the determinants of adoption demand and supply using two complementary micro data sets.Namely, we formulate alternative hypotheses regarding demand and supply and test these hypotheses inmultivariate analysis using micro panel data from the Survey of Income and Program Participation andpooled micro cross-sectional data from the National Survey of Family Growth. Third, we develop a dynamicmodel of adoption seeking in which women make decisions about labor supply and family formation throughchildbearing or adoption. After calibrating the model to the relevant moments in the micro data, weinvestigate whether it can explain the changes in the adoption market observed in the past twenty years inresponse to changes in the environment faced by prospective parents. We then use our model to investigatethe e�ect of several counterfactual policy experiments pertaining to adoption.

Our main contributions to the adoption literature are twofold. Empirically, we provide new evidence onadoption demand and supply, using rigorous econometric analysis and novel micro data. Theoretically, webuild a formal model of adoption seeking and conduct policy experiments using computational methods.

2 Historical Trends in Adoption in the U.S.

2.1 Background

For the purpose of our research, it is important to distinguish several types of adoption. First of all, wefocus on formal adoptions, i.e., legally approved adoptions, since no systematic data on informal adoptionsexist. Adoptions can be divided into unrelated and related adoptions where the latter is de�ned as adoptionsby relatives or stepparents. In our research project, we primarily focus on unrelated adoptions as a way ofbuilding a family. Limited data indicate that, in recent years, 80% of related adoptions in the U.S. havebeen stepparent adoptions (Flango and Flango (1995)). Since these adoptions are in�uenced heavily by thedecisions to divorce and remarry, they are quite distinct from unrelated adoptions.

Adoptions can be also divided into domestic and intercountry adoptions. The �rst permanent legislationfor intercountry adoptions in the U.S. was in 1963, establishing a special visa category for foreign childrenapproved for adoption (Weil (1984)). By de�nition, almost all intercountry adoptions are unrelated adop-tions. In 1996, intercountry adoption comprised 9.4% of total adoptions and 17% of unrelated adoptions inthe U.S. (Marshner (1999)).

Starting in 1980, the federal government has placed an emphasis on �nding adoptive homes for childrenin the foster care system. In particular, the government has encouraged the placement of children with�special needs� by providing subsidies. The de�nition of special needs children varies from state to state, butit typically refers to children who are above a certain age, have physical or mental disabilities, or are froman ethnic minority. In 1996, special needs children comprised 22% of total adoptions and 40% of unrelatedadoptions in the U.S. (Marshner (1999)).

In the U.S., adoptions can be arranged through public agencies, private agencies, or without involving anagency. The primary functions of adoption agencies are to evaluate prospective applicants, arrange suitableplacements, and process court applications. A birthmother may relinquish all rights to her child to anagency. All private agencies are licensed and subject to state regulations. Most private agencies are non-pro�t organizations, but some states permit pro-pro�t agencies (O'Halloran (2006)). The cost of adoptionvaries substantially by agency and by adoption type. Estimated total costs are up to $2,500 for adoptionof special needs children through public agencies, $4,000 to $30,000 for domestic adoption through privateagencies, and $7,000 to $25,000 for intercountry adoption through private agencies. Expected waiting timefor healthy infants is between two and four years.3

2.2 Trends in Adoption in the U.S., 1944-2001

To motivate the empirical analysis using micro data, we �rst summarize the historical trends in the U.S.based on aggregate-level data. Figure 1 presents adoption rates (de�ned as the number of adoptions per1,000 live births) in the U.S. from 1944 to 2006. Note that the time-series data are from four disparatesources and thus not homogeneous across all years. Note also that after 1975 the data become sporadic(in particular, no estimates after 2001), making it di�cult to establish national trends. These limitations

3Adoption.org website: http://www.adopting.org/adoptions/learn-about-adoption-costs-and-fees-2.html.

2

notwithstanding, the most striking fact is that the adoption rates in the U.S. increased sharply in the 1960'sand declined in subsequent decades. As a result, adoption rates in recent years (31.6 per 1,000 live births in2001) are substantially lower than the historical peak reached in 1971 (47.5 per 1,000 live births). 1

Figure 1. Adoptions per 1,000 Live Births in the U.S.

10

15

20

25

30

35

40

45

50

1944 1948 1952 1956 1960 1964 1968 1972 1976 1980 1984 1988 1992 1996 2000 2004

Maza (1984)Marshner (1999)Flango (1995)NAIC (2004)

In terms of composition, the percentage of unrelated adoptions in total adoptions was constant at around53% in 1955-70, then dropped sharply from 51% in 1970 to 37% in 1975, and increased from 38% in 1982to 55% in 1996. The percentage of intercountry adoptions in total adoptions increased steadily from 1.0%in 1965 to 4.4% in 1975, �uctuated between 4% and 9% between 1982 and 1992, and then rose sharplyfrom 5.1% in 1992 to 15% in 2001. It is worth noting that, while the number of intercountry adoptions hasincreased dramatically since the early 1990's, they still make up a relatively small share of total adoptionsin the U.S.

Why is the adoption rate in 2001, for both related and unrelated adoptions, still 30% below the 1971level? There are several possible explanations that we explore more formally in our analysis below. It canbe attributed to the decline in the supply of adoptable children due to the availability of oral contraceptives,abortion legalization (Bitler and Zabodny (2002)), or the increasing share of unmarried mothers who decidenot to relinquish their babies for adoption (Bachrach (1986)). Alternatively, it can be attributed to thedecline in the demand for adoption due to the advancement in infertility treatment or to the increasingpreference for biological children given smaller family sizes.

Upon closer inspection, however, there are several countervailing factors we must account for. On thesupply side, the number of children born out of wedlock has been rising sharply (Ventura and Bachrach(2001)), which may mitigate the fall in adoption supply due to falling relinquishment rates. In addition, thenumber of adoptable children in the foster care system has increased since 1980 due to the aforementionedpolicy change. The rise in divorce and remarriage rates since the 1970's also suggests an increasing supply ofchildren available for stepparent adoption. Finally, the fact that intercountry adoption became a signi�cantcomponent of total adoptions only in the last decade despite its availability since the 1960's poses a puzzle.

Similarly, a number of factors on the demand side suggest a potentially large increase in adoption demand.The dramatic rise in the educational attainment of women and their labor force participation rates sincethe 1970's implies increasing opportunity costs for childbearing due to work interruptions (Ellwood et al.(2004), Olivetti (2006)). Moreover, both the age at �rst marriage and the age at �rst childbirth have increasedsubstantially over the last thirty years. As delayed childbirth increases a woman's risk of facing fecundityimpairment before achieving her desired family size, these changes presumably increased the demand foradoption. The availability of in vitro fertilization and other assisted reproductive technology has certainlymitigated this e�ect, but note also that such technology became widely available in the U.S. only in themid 1980's. Lastly, in terms of transaction costs, we expect that the decline in search costs in the adoptionmarket due to recent advances in information technology would lead to higher adoption rates.

3

In short, in order to explain the historical patterns, we need a better understanding of adoption supplyand demand by di�erent adoption types. Therefore, we turn to micro data to explore the determinants ofthe demand and supply in the U.S. adoption market.

3 Analysis of Adoption Demand and Supply Using Micro Data

3.1 Literature Review

The empirical literature on adoption demand using micro (individual-level) data is relatively small andwritten almost entirely by sociologists. Most micro studies (e.g., Bonham (1977), Bachrach (1983), Bachrach(1986), Bachrach et al. (1991), Freundlich (1998), Chandra et al. (1999), Hollingsworth (2000)) use single-year cross-sectional data from the National Survey of Family Growth (NSFG), and some (e.g. Bachrach et al.(1990), Waddoups (1997)) use the data from the National Health Interview Survey. These papers typicallycompare the characteristics of women who have adopted (or have ever sought to adopt) a child with thosewho have never adopted (or have never sought to adopt). The common �ndings are that women who seekto adopt are on average more likely to be white, married, older, fecundity impaired, of higher socioeconomicstatus, and exhibit higher labor force participation. While most studies present only descriptive statistics,some employ multivariate analysis such as logistic regressions but do not formulate and test hypotheses (e.g.,Bachrach et al. (1991), Waddoups (1997), Hollingsworth (2000)).

The empirical literature on adoption supply using micro data is even scarcer. Two studies, Bachrach(1986) and Chandra et al. (1999), use cross-sectional data from the NSFG to �nd relinquishment ratessubstantially higher among unmarried white mothers than among unmarried black mothers and decliningtrends in relinquishment rates among white mothers. Their analyses, however, are purely descriptive, assmall sample sizes of the single-year data preclude the use of multivariate analysis. A few economists haveused aggregate-level data to estimate adoption supply. In particular, Bitler and Zavodny (2002) use state-level panel data from the National Center of Social Statistics, 1961-75, and careful econometric analysisto estimate the e�ect of abortion legalization on adoptions. They �nd that it could explain much of thedecline in adoption rates that took place in the early 1970's. Medo� (1993) uses state-level cross-sectionaldata from the 1982 National Committee for Adoption survey and shows that adoption rates, interpreted asa proxy for relinquishment rates, are negatively related to women's marital status, welfare payments, andthe unemployment rate, and positively related to education and religious a�liation.

3.2 Data

To summarize, preceding studies have mostly used descriptive statistics instead of multivariate analysis; havefocused on unrelated adoption without distinguishing di�erent types of adoption; have not fully exploitedvariation in the data across time by focusing on a single cross-section at a time; and have not explicitlytested speci�c hypotheses concerning the determinants of adoption demand and supply. To address theseshortcomings, this project not only takes full advantage of the NSFG data by pooling all six cycles of databetween 1973 and 2002, but also introduces micro panel data from the Survey of Income and ProgramParticipation (SIPP) from 1984 to 2004. We describe each data set in turn.

3.2.1 National Survey of Family Growth, Cycles 1-6 (1973-2002)

The National Survey of Family Growth, conducted by the National Center for Health Statistics, consists ofsix cycles of repeated cross-sectional data in 1973, 1976, 1982, 1988, 1995 and 2002. The sampling universehas expanded slightly over time: through personal interviews, the data were collected from a nationallyrepresentative sample of ever-married women and single mothers 15 to 44 years of age in Cycles 1-2; allwomen 15 to 44 years of age in Cycles 3-5; and all women and men 15 to 44 years of age in Cycle 6. Thesample sizes range from 7,600 to 11,000 females over the six cycles.

The primary strength of the NSFG is the availability of detailed information on women's marital history,fertility history, reproductive and sexual health, including use of contraceptive and infertility services, inaddition to women's socio-demographic characteristics such as age, race, religion, income, education history,and work history. In particular, in all cycles, women are speci�cally asked if they have ever adopted a child

4

and ever considered adopting a child. In addition, starting in Cycle 4, women are also asked if they haveever taken any steps toward adopting a child. NSFG data thus enable us to study not only ful�lled demandbut also potential demand for adoption. To the extent that ful�lled demand for unrelated adoption could beconstrained by the supply of adoptable children in a non-clearing market, the analysis of potential demandmay provide additional insights. Although stepparent adoption is explicitly excluded from the data in the�rst two cycles, in Cycles 3-6 we can identify unrelated adoptions, stepparent adoptions, and other relatedadoptions, as well as intercountry adoptions. Starting in Cycle 3, women are also asked if they have everrelinquished a child for adoption. To our knowledge, the NSFG is the only micro data source based on anational survey that provides information on child relinquishment.

3.2.2 Survey of Income and Program Participation, 1984-2004

The Survey of Income and Program Participation (SIPP), conducted by the U.S. Census Bureau, is a fairlycontinuous series of national panels from 1984 to 2004, with a national sample of 12,000 to 44,000 U.S.households in each panel. The duration of each panel is one to four years. The SIPP data provide richinformation on individual household members, including spouses, partners, and cohabitating children. Inaddition to demographic characteristics, the SIPP provides detailed information on individuals' labor marketstatus (e.g. hours worked, wages, work history, and occupation), sources of income and assets, as well as theirparticipation in various welfare programs. From the Household Relationship topical module, which elicitsrelationships between all household members, we can identify the relationship of each child (biological,step, or adopted) to their mother and their father. In addition, using the information on disability statusof children, we can identify adoption of disabled children. Unfortunately, the SIPP does not contain anyinformation on child relinquishment.

Although it has never been exploited, the SIPP has a number of advantages for studying adoptiondemand. First, its large sample sizes and frequent panels (13 panels between 1984 and 2004) provide us witha su�cient number of observations of adoptive parents to conduct multivariate analysis. Second, we candistinguish di�erent types of adoption, including unrelated adoption, stepparent adoption, and adoptions ofdisabled children. Third, in contrast to the NSFG whose focus is women of childbearing age, the SIPP ismore representative of the population, including single men as well as women aged 45 years or older whoare part of adoptive families. Fourth, while the NSFG has better data on women's reproductive healthand fertility history, the SIPP contains better information about labor market status and other economiccharacteristics.

In short, the two data sets, NSFG and SIPP, complement each other in many important aspects. Byusing both sources of data, we can address di�erent issues that require di�erent types of information. And,to the extent that the two data sets overlap, we are able to cross-examine the reliability of the di�erent datasources and check the robustness of our empirical �ndings.

3.3 Empirical Analysis of the Demand for Adoption

We examine the demand for unrelated adoption, using both the NSFG and SIPP data sets. Our primary goalis to formulate various hypotheses concerning adoption demand and test these hypotheses using multivariateanalysis. Partly to showcase the socioeconomic and demographic variables available in the respective datasets, we present descriptive statistics for selected years (see Table 1 for the NSFG, where, due to spacelimitations, we include only the results from the two most recent cycles, and Table 2 for the SIPP). In eachtable, the distribution of a set of characteristics for adoptive mothers and biological mothers is reported inseparate columns. Note that due to low adoption rates in general, in both data sets, the sample size ofadoptive mothers in any single-year sample is small. When we pool data across all available cycles or waves,however, the sample sizes increase to 719 and 1,565 in the NSFG and SIPP data sets, respectively.

According to Table 1, in the NSFG sample, compared to biological mothers, adoptive mothers on averagehave higher education, higher household income, and fewer children than biological mothers, and are morelikely to be white, fecundity impaired, married, and religious; they are also more likely to work full time, lesslikely to participate in welfare programs, and more likely to be a foster parent. These results are consistentwith the �ndings of preceding studies using earlier cycles of NSFG data. According to Table 2, our SIPPsample exhibits similar patterns except for labor force status, in which the two groups show little di�erence.

5

Table 1Biological, Adoptive, and Relinquishing Mothers by Observed Characteristics, NSFG

Characteristics All biological Adoptive Relinquishing All biological Adoptive RelinquishingTotal Number 6,911 91 40 4,413 58 33

Education at InterviewNo High School diploma or GED 17.4 5.5 25.0 18.6 7.0 33.3High School diploma or GED 82.6 94.5 75.0 81.4 93.0 66.7

Race and Hispanic OriginHispanic 16.1 8.8 17.5 24.5 8.6 15.2Non-Hispanic white 55.8 70.3 75.0 48.7 60.3 66.7Non-Hispanic black 25.2 18.7 7.5 22.5 24.1 9.1Other 3.0 2.2 - 4.3 6.9 9.1

Total household income Under $10,000 12.4 1.1 22.5 11.9 6.9 12.110000-$19,999 16.2 6.6 15.0 21.0 5.2 24.2$20,000-$29,900 15.1 7.7 17.5 16.2 12.1 15.2$30,000-$39,999 13.9 9.9 17.5 13.1 5.2 18.2$40,000-$49,999 12.1 23.1 12.5 8.2 13.8 3.0$50,000-$59,999 9.2 12.1 2.5 7.9 6.9 9.1$60,000 or more 21.1 39.6 12.5 21.7 50.0 18.2

Number of children in householdNo Children 10.0 15.4 25.0 10.6 12.1 33.31 Child 34.8 41.8 40.0 35.0 43.1 39.42 Children 33.7 24.2 17.5 33.3 17.2 15.23 Children 15.4 12.1 10.0 14.9 15.5 12.14 Children 4.6 5.5 5.0 4.4 8.6 -5 Children or more 1.4 1.1 2.5 1.8 3.5 -

Fertility Status1

Not Sterile 47.4 25.6 34.8 69.6 62.1 54.6Surgically Sterile 47.6 48.7 65.2 28.3 27.6 45.5Nonsurgically Sterile 5.0 25.6 - 2.1 10.3 -

Formal Marital StatusMarried 63.9 76.9 47.5 56.2 67.2 42.4Widowed 1.2 3.3 - 1.0 - 6.1Divorced 11.9 17.6 12.5 12.0 6.9 15.2Separated 5.9 - 20.0 6.1 3.5 21.2Never Married 17.1 2.2 20.0 24.6 22.4 15.2

Labor Force StatusWorking full time 42.3 49.5 30.0 43.9 55.2 42.4Working part time 15.5 15.4 5.0 14.8 17.2 12.1Working but temp. illness, vacation 2.2 1.1 5.0 3.9 3.5 9.1Working but on family leave 0.6 - - 1.3 - -Unemployed (searching for job) 1.4 - 7.5 4.6 3.5 3.0In school 2.3 2.2 - 0.5 - -Keeping house 28.4 17.6 35.0 27.8 17.2 30.3Other 7.2 14.3 17.5 3.2 3.5 3.0

ReligionNo religion 9.2 7.7 5.0 12.5 8.6 24.2Catholic 29.1 31.9 35.0 31.0 24.1 27.3Protestant 57.1 53.9 55.0 51.9 62.1 39.4Other 4.7 6.6 5.0 4.7 5.2 9.1

Welfare ParticipationYes 14.6 3.3 22.5 37.1 17.2 27.3No 85.2 96.7 75.0 63.0 82.8 72.7

Pregnancy Loss HistorySpontaneous Pregnancy Loss 26.7 51.7 32.5 28.7 48.6 36.4No Spontaneous Pregnancy Loss 73.3 48.3 67.5 71.3 51.4 63.6

Abortion HistoryInduced Abortion 20.2 15.0 25.0 20.5 21.6 27.3No Induced Abortion 79.8 85.0 75.0 79.5 78.4 72.7

Importance of Religion in Daily LifeVery Important 57.0 67.0 57.5 63.6 66.7 64.0Somewhat Important 35.7 27.5 37.5 32.1 27.8 28.0Not Important 7.2 5.5 5.0 4.0 1.9 8.0N/A 0.1 - - 0.2 3.7 -

Being a Foster ParentYes 1.3 15.4 - 0.7 27.6 -No 98.7 84.6 100.0 99.3 72.4 100.0Soure: NSFG (Cycles V and VI)Biological mothers: All women who report giving birth to at least one child.Adoptive mothers: All women who reported to have ever adopted a non-biological child.Relinquishing mothers: All women who report to have ever placed a child for adoption.1 In 1995, this question was not asked to the subsample of unmarried women (or not co-habitating with a male partner) and womenthat reported not having had intercourse for the first time.

1995 2002

% of mother type (biological, adoptive or relinquishing)

Table 2Biological and Adoptive Mothers by Observed Characteristics, SIPP

Characteristics Biological Adoptive Biological Adoptive Biological AdoptiveTotal Number 4,882 97 8,500 159 12,067 225

Education at InterviewNo High School diploma or GED 25.0 18.5 20.3 16.4 16.6 8.5High School diploma or GED 39.8 37.2 36.0 30.9 31.2 27.2Some college no bachelor's degree 20.6 26.9 26.4 25.7 22.1 18.8Bachelor degree or higher 14.6 17.4 17.2 27.1 30.1 45.6

Race and Hispanic OriginHispanic 7.7 3.0 12.1 4.4 16.2 4.7Non-Hispanic white 78.5 89.7 74.3 85.8 69.0 79.5Non-Hispanic non-white 13.8 7.3 13.7 9.8 14.7 15.8

Total household income Under $10,000 14.0 7.5 10.8 6.1 6.8 7.310000-$19,999 20.1 17.1 14.3 6.1 9.8 9.0$20,000-$29,900 21.5 18.0 16.2 15.8 11.1 5.7$30,000-$39,999 16.6 20.9 14.4 14.8 12.2 9.0$40,000-$49,999 10.9 15.7 12.3 12.5 11.2 12.3$50,000-$59,999 6.5 9.3 9.1 11.9 9.6 8.7$60,000 or more 10.3 11.7 22.9 32.8 39.3 48.1

Number of children in householdNo Children 20.5 23.8 18.8 23.9 18.7 18.61 Child 31.2 49.7 30.5 34.5 32.2 36.92 Children 30.5 11.9 30.5 20.6 30.6 19.33 Children 12.1 9.0 13.5 9.3 12.8 15.74 Children 3.8 2.0 4.2 7.8 3.9 3.95 Children or more 1.9 3.6 2.5 4.0 1.9 5.6

Formal Marital StatusMarried 73.6 85.6 68.3 80.9 67.5 77.2Widowed 7.1 6.3 6.6 7.5 6.4 7.1Divorced 9.4 6.8 12.0 6.6 11.4 6.0Separated 3.9 0.0 4.3 1.9 4.0 2.2Never Married 6.1 1.2 8.8 3.1 10.8 7.5

Labor Force StatusWorking full time 33.5 30.3 40.4 39.2 41.0 31.8Working part time 18.1 16.2 16.5 17.5 22.6 28.6Working, missed one or more weeks 4.7 7.1 2.3 2.9 3.1 3.6Unemployed 4.1 1.3 4.1 3.8 2.2 1.1In school 1.2 1.2 - - - -Other out of the labor force 38.5 43.9 36.8 36.8 31.1 35.0

Welfare ParticipationNot applicable 93.9 96.7 91.2 97.8 97.0 97.0Yes (recipiency this month) 5.9 3.3 8.4 1.2 2.6 3.0No (recipiency this month) 0.2 - 0.4 1.0 0.4 -

Being a Foster ParentYes 0.3 2.0 0.2 3.5 0.1 2.9No 99.7 98.0 99.8 96.5 99.9 96.9Source: SIPP (1984 panel waves 6 and 8, 1993 panel wave 2, and 2001 panel wave 2)Biological mothers: All women who report giving birth to at least one child.Adoptive mothers: All women who report to have ever legally adopted a non-biological child.

% of mother type (biological or adoptive)

20011984 1993

Although these descriptive statistics are informative about how each characteristic is correlated with thelikelihood of adopting, they cannot tell us what its partial e�ect on adoption is (i.e., holding other charac-teristics constant), nor can they reveal the relative importance of the di�erent characteristics in determiningadoption demand. For example, the observed big di�erence in welfare participation rates between adoptiveand biological mothers in Table 1 above could just re�ect the di�erences in their education and income. Infact, in our preliminary logit regression analysis (results not reported here), the coe�cient on the indicatorvariable for receiving public assistance is not statistically signi�cantly di�erent from zero once women's ed-ucation and household income are controlled for. Descriptive statistics also do not provide a convenient wayto examine interactive e�ects of di�erent characteristics. To address these issues, we proceed to multivariateanalysis and implement a logistic regression of adoption demand on a set of explanatory variables. We useboth ful�lled demand and potential demand as the dependent variable, and compare the results of these twospeci�cations. To guide the choice of explanatory variables, we consider the following competing hypothesesthat can explain why people adopt, rather than bear, children.

1. Opportunity cost hypothesis: individuals adopt children because it is costly to interrupt a job spell forchildbearing in terms of labor market outcomes (i.e., lower accumulation of work experience or human capitalresulting in lower wage or less favorable occupation). Proxy variables for the opportunity cost, denoted asX1, include educational attainment, employment status, work experience, wages, and occupation, all priorto adoption.

2. Infertility hypothesis: individuals adopt children because they are not able to produce biological chil-dren but would like to derive utility from an o�spring. Proxy variables for infertility, denoted as X2, includefecundity impairment, reproductive health, pregnancy loss history, use of infertility services, (available onlyin the NSFG), age at adoption, the absence of biological children (available also in the SIPP).

3. Humanitarian hypothesis: individuals adopt children, especially those who are disadvantaged, due tohumanitarian or altruistic motives. Unlike the above two hypotheses, under this hypothesis adoptive childrenare not substitutes for biological children. Proxy variables for such motives, denoted as X3, include religiousa�liation and importance of religion in daily life (in the NSFG) and charity donation (in the SIPP).

4. Resource constraint hypothesis: individuals adopt children, incurring substantial monetary cost (e.g.adoption fees) and psychic costs (e.g. anxiety from long waiting, uncertainty over the quality of matchbetween a prospective parent and a child), if they have better �nancial resources and psychological support.Proxy variables for such resources, denoted as X4, are marital status, family income, and assets.

The �rst three sets of variables (X1, X2, and X3) capture di�erent motivations for adoptions, while thefourth (X4) can be thought of as measuring resources that can facilitate adoption. Because these hypothesesare not necessarily mutually exclusive and can jointly impact individuals' decision to adopt, all three factorscould have an e�ect in a regression of adoption demand. The advantage of the multivariate analysis that wepropose is that we can obtain estimates of the marginal e�ects of each factor (i.e., holding other characteristicsconstant), and uncover the relative importance of the di�erent characteristics in determining adoption. Inaddition, we will examine interactive e�ects of di�erent characteristics that would allow us to (i) test theimportance of the various hypotheses and (ii) explore how di�erent motivations have changed over time. Todo this, we exploit variations in the data across adoption types and across time.

First, because di�erent motivations likely manifest themselves in di�erent types of adoption (e.g., relatedvs. unrelated adoption, domestic vs. intercountry adoption, adoption of special needs children vs. healthyinfants), by distinguishing adoption types, we can better disentangle the competing hypotheses. Speci�cally,in our logistic regression, in addition to the basic variables (X1, X2, X3, and X4), we include dummies fordi�erent adoption types and their interactions with certain regressor(s). Then we can examine variationsin the relative importance of the chosen regressor(s) across di�erent types of adoption. Example 1: underthe humanitarian hypothesis, the proxy variable X3 would be more important relative to other factors foradoption of disabled children (identi�ed in the SIPP) than for other types of adoption. Example 2: under theresource constraint hypothesis, intercountry adoption would be preferred to domestic adoption if it entailslower monetary costs and more e�cient adoption process (e.g., adoption from China). In this case, theresource proxies X4 would be less important relative to other variables for intercountry adoption (identi�edin the NSFG Cycles 3-6) than for domestic adoption. Furthermore, because resource constraints wouldbe less binding in potential demand than in actual demand, the proxy X4 would be less important in theregression using the former dependent variable than the latter.

The second strategy relies on comparisons over time. In our regression, in addition to the basic variables

8

(X1, X2, X3, and X4), we include time dummies and their interactions with certain regressor(s). Thenwe can examine di�erential time changes in the coe�cients associated with those regressor(s). Example 1:under the infertility hypothesis, the availability of in-vitro fertilization (IVF) treatment after 1985 wouldmake the importance of fertility variable X2 relative to other socioeconomic factors become less importantafter 1985 than before (we compare NSFG Cycles 1-3 sample and Cycles 4-6 sample). Given the high costsof IVF, this e�ect should be stronger among higher income group. Example 2: under the opportunity costhypothesis, increasing returns to education and work experience for women since the 1970's would increasethe opportunity cost of childbearing, thus the importance of labor market attachment variables X1 relativeto other socio-demographic characteristics would become more important over time (we use both NSFGdata, 1973-2002, and SIPP data, 1984-2001).

3.4 Empirical Analysis of the Supply of Adoption

Employing a similar methodology, we also provide an empirical analysis of adoption supply using the datafrom the NSFG Cycles 3-6 (1982, 1988, 1995, and 2002). Table 1 presents descriptive statistics comparingrelinquishing mothers with all biological mothers. According to Table 1, white mothers were much morelikely to place children for adoption than black and Hispanic mothers. In addition, compared to all biologicalmothers, relinquishing mothers tend to be less educated, more concentrated in the lower income brackets,somewhat less likely to participate in the labor market, and less likely to be married at the time of theinterview.

In order to study the propensity to give up a child for adoption, we again propose to use a multivari-ate analysis using micro data. To our knowledge, no such analysis exists for adoption supply, thus, thiscomponent constitutes an important contribution of this research proposal.

Given that the sample size of relinquishing mothers in each cycle is very small (40 in 1995 and 33 in2005), we pool the data across four cycles to increase the sample size and gain statistical power. Theresulting data set includes 286 relinquishing mothers, enabling us to conduct the �rst multivariate analysisof child relinquishment at the micro level. Namely, we run a logistical regression of relinquishment on acomprehensive set of explanatory variables. To guide the choice of explanatory variables, we consider thefollowing competing hypotheses that can explain why women give up biological children for adoption ratherthan keeping and raising them by themselves:

1. Opportunity cost hypothesis: individuals give up biological children for adoption because it is costly tointerrupt their human capital accumulation process (particularly, in terms of education and thus, subsequentlabor income). Proxy variables for the opportunity cost, denoted as Y1, include educational attainment, workstatus, work experience, wages, and occupation, all measured at the time of relinquishment.

2. Resource constraint hypothesis: women give up children for adoption simply because they do not havethe required �nancial resources to raise the child. Proxy variables, denoted as Y2, include wage, householdincome and living arrangements at time of relinquishment. Unfortunately, it is not possible to construct mostof these variables at the time of relinquishment. However, we could potentially use age of the mother at thetime of the birth of the relinquished child, living arrangements at age 14, and the education of the woman'sfather as proxies for the woman's earnings capacity and resource constraint at the time of relinquishment.

3. Unwanted child hypothesis: women relinquish children for adoption because they have very lowpreference for motherhood, but fertility is only partially controlled by the woman's choices. Proxy variablesfor the �unwantedness,� denoted Y3, could include desired number of children, history of induced abortions,and number of children at the time of the interview (post-relinquishment date).

4. Stigma hypothesis: women relinquish children because there is a signi�cant stigma associated withbeing a single mother. In other words, even if the woman has the resources to raise the child, has partiallyor fully completed her human capital accumulation process, and derives utility from o�spring, she mightstill want to give up a biological child for adoption because society might discriminate her for having a childout-of-wedlock. Proxy variables, denoted Y4, could include marital status at the time of relinquishment.

Note that these hypotheses are not necessarily mutually exclusive and can jointly impact individuals'decisions to give up children for adoption. Thus, all factors included in Y1, Y2, Y3 and Y4 can have ana�ect on the propensity to relinquish. The advantage of the multivariate analysis that we propose is thatwe can obtain estimates of the marginal e�ects of each factor (i.e., holding other characteristics constant),and uncover the relative importance of the di�erent characteristics in determining relinquishment.

9

However, either because we have small sample sizes or due to the speci�c time period that the NSFGdata cover, it is more di�cult to examine interactive e�ects of di�erent characteristics that would allow usto (i) test the importance of the various hypotheses and (ii) explore how di�erent motivations have changedover time. For example, under the unwanted child hypothesis, the availability of more e�ective contraceptionmethods and abortion that enabled women to selectively prevent or terminate unintended pregnancies wouldmake the importance of variable Y3 relative to other socioeconomic factors become less important. However,because both oral contraceptives and abortions had become widely available by the early 1970's, we cannotuse time variation in NSFG to test this hypothesis (we refer to Bitler and Zavdony 2002, for evidencesupporting this hypothesis).

Nevertheless, there are still a number of things we can do. First, under the opportunity cost hypothesis,increasing returns to education and work experience for women since the 1970's would increase the oppor-tunity cost of childrearing, thus the importance of labor market attachment variables Y1 relative to othersocio-demographic characteristics would become more important over time. Second, one could argue thatdue to changes in the nature of marriage markets and social norms over the past decade (e.g., increasingshare of children born out of wedlock, increasing number of marriages after parenthood, etc.), the relativeimportance of the stigma hypothesis must have declined over time. In this case, one would observe thatthe importance of Y4 relative to the other set of explanatory variables would decrease over NSFG cycles.Third, Fang and Keane (2004) report that mean income of single mothers from public assistance and foodstamps dropped substantially from $2,892 in 1980 to $1837 in 1996 just prior to the Welfare Reform (asgrant levels declined in real terms over time in almost every State). This trend in the welfare system wouldhave implied, under the resource constraint hypothesis, that variables Y2 became relatively more importantover time, especially for unmarried women (who are usually more likely to be welfare recipients). We couldtest these by including time dummies and their interactions with the explanatory variables.

Although the time comparison will not help us with the unwanted child hypothesis, we might be able touse a group-comparison. For example, religion itself should not explain the propensity to relinquish a child,unless women of a particular religious a�liation are more likely to be characterized be one of the hypothesesmentioned above (as measured by variables Y1, Y2, Y3 and Y4). So conditional on these variables, underthe unwanted child hypothesis, we would expect the proxy variable Y3 to be less important relative to otherfactors for child relinquishment in the case of religious women since they would be more likely to morallyoppose not wanting a child than less religious women.

There is evidence that changes in the marriage market have been stronger among whites than blacks. Forexample, Ventura and Bachrach (2000) report that while birth rates for unmarried black women remainedrelatively stable at approximately 80 per 1,000 births between 1970 and 1998, birth rates for unmarriedwhite women more than doubled from 18 per 1,000 births in 1970 to 40 in 1998. Note that race itself shouldnot signi�cantly explain the propensity to give up a child for adoption, unless women of a particular raceare more likely to be characterized be one of the hypotheses mentioned above (as measured by variables Y1,Y2, Y3 and Y4). Thus, under the stigma hypothesis, the relative importance of variable Y4 should be lowerfor blacks than for whites and moreover, should have remained constant over time for black women whileit should have decreased relative to other characteristics over time in the case of white women. Note thatmarital status at the time of relinquishment could a�ect both, the resource constraint hypothesis and thestigma hypothesis. If the test just described shows what we expect, we would be more con�dent that maritalstatus is capturing stigma and that Y2 satisfactorily captures the resource constraint.

4 Model of Adoption Seeking

In addition to carrying out the regression analysis described in Section 3, we propose to further the under-standing of the adoption market by developing a novel theoretical model of adoption seeking. To capturethe fact that childbearing and adoption are two alternative ways to become a parent, we include in ourmodel not only the decision to adopt children, but also the decision to bear children of one's own. Ourmodel shares some features with other dynamic models of fertility choice that describe the trade-o� betweenlabor-market advancement and childbearing,4 but adds the novel dimension of adoption to the picture. The

4See Eckstein and Wolpin (1989), Caucutt et al. (2002), Mullin and Wang (2002), Erosa et al. (2002), and Greenwood etal. (2003).

10

model is explicitly designed to capture the hypotheses regarding adoption demand described in Section 3and, unlike the regression analysis, can be used to perform counterfactual policy analysis. Similarly to ouranalysis in Section 3, we abstract from related adoptions. To simplify our analysis, we also abstract frombargaining between prospective parents and assume that all childbearing and adoption decisions are madeby prospective mothers.5

4.1 Environment

Technology Consider the following dynamic optimization problem faced by a woman entering adulthoodat adult age t = 0. For the �rst T b periods of her adult life, every period she makes childbearing/adoptiondecisions, and resource and time allocation decisions. After the �rst T b periods, she only makes resourceand time allocation decisions until the end of her planning horizon after T r > T b periods.

A woman's decision to have children is in�uenced by the prospects she faces in the labor and marriagemarkets. In particular, each woman has a �xed level of education, e, where e ∈ {e1, .., eK}. Each woman hasa �xed productivity p which comes from the set Pi =

{pi1, .., p

iNi

}if her education level is ei. A woman's

productivity together with her endogenous labor-market experience, which starts out at x0 = 0, determineher wages. To keep the model simple, we do not explicitly model the marriage market, rather we assume anexogenous process by which women become married. In particular, we assume that all women start out theiradult life single (m0 = 0), and single women become married (mt = 1) at age t with probability πm (t, e).

When a woman is married, her husband's income becomes part of the household income and he becomesan adult sharing in household consumption. The husband's income is determined by his productivity and hislabor-market experience, which corresponds one-to-one to his adult age, since men in the model are assumedto work n hours every period of their adult life. The age of the husband is given by th = t + ∆, where∆ is the age di�erence between spouses. The productivity of the husband

(ph

)is distributed according

to the distribution function H(ph | p

), where ph is stochastically increasing in p, capturing the positively

assortative nature of marriage markets. A husband's education level (eh) is a deterministic function of hisproductivity. Marriage is assumed to be an absorbing state and it is assumed that the marital status of awoman remains the same after period T b.

Since infertility appears to be a major reason for adopting a child as suggested by our empirical analysisand by analysis of self-reported reason for adoption (Berry et al. (1996)), we model the process by whicha woman's fecundity evolves with age. In particular, we assume that all women start out their adult life ina high fecundity state

(ω0 = ωh

)and a woman's fecundity becomes low

(ωt = ωl

)with probability πω (t),

where low fecundity is an absorbing state.To keep the size of the state space more manageable, we do not keep track of the age of each child a

woman has, rather we assume that the state of her children can be summarized by the number of biologicaland adopted children she has (kb

t and kat , respectively) and their average ages (yb

t and yat , respectively). Thus

the total number of children a woman has is kt = kbt + ka

t and their average age is yt = ybt kb

t+yat ka

t

kbt+ka

t.

The timing of a woman's decisions at adult age t with kt children of average age yt is as follows. First, sheneeds to decide whether she wants to have an additional biological child (in which case wb

t = 1, else wbt = 0).

Then she learns whether she will have a biological child at the end of the period (in which case bt = 1, elsebt = 0), the probability of which is given by wb

tπb (t, ω), so that it depends on her desire to have a child, her

fecundity status, and her age. It is assumed that pregnancy and caring for a newborn requires τ b amountof time during the period. After learning the outcome of bt, a woman needs to decide whether she wants toadopt a domestic child (in which case wd

t = 1, else wdt = 0) and whether she wants to adopt a foreign-born

child (in which case wft = 1, else wf

t = 0). The di�erence between adopting domestic and foreign-bornchildren is in the time and monetary cost of adoption, denoted by τd, rd and τf , rf , respectively, and in theage distribution of children available for adoption, denoted by F d and F f , respectively. If a woman decidesshe wants to adopt a child, she will be able to do so at the end of the period for sure, and the age of theadopted child (yd and/or yf ) is drawn from the appropriate distribution at the time of adoption.

After making childbearing and adoption decisions and knowing the outcome of bt, a woman needs tomake time-allocation decisions for period t. It is assumed that caring for a woman's existing children requiresτ c (kt, yt) amount of time, thus a woman's available time in period t is 1− btτ

b −wdt τd −wf

t τf − τ c (kt, yt),

5Our model is a unitary preference model similar in this regard, for example, to Regalia and Rios-Rull (2001).

11

which she needs to allocate between work (nt) and leisure (lt). Finally, consumption per household memberin period t is determined by

ct =(

11 + mt + qkt

)ν [it − wd

t rd − wft rf

], (1)

where q ≤ 1 is the adult-equivalent of a child and ν ≤ 1 measures the returns to scale in household size.Household income, it, is given by

it = phrh(eh, t + ∆

)nmt + prw (e, xt) nt, (2)

where rh (·, ·) and rw (·, ·) are functions determining the returns to experience by education for husbandsand women, respectively. Finally, a woman's labor-market experience evolves according to

xt+1 = xt + I(nt ≥ n),

where n represents the minimum amount of work required for a woman to enhance her work experience.Notice that our model does not allow for a savings decision. This is, of course, an extreme assumption(applied �rst by Eckstein and Wolpin (1989) in a model of endogenous fertility), but one that is often madein this literature to abstract from the complex issue of asset accumulation.

Preferences Since humanitarian reasons seem to be an important motivating factor in adoption for asizeable minority of adoptive parents (Berry et al. (1996)) and there is substantial variation in the numberof children women choose to have, we allow for heterogeneity in women's preferences over adoption andchildren. In particular, we allow women to di�er by their type, j = 1, ..,M , where a woman of type j derivesutility from consumption, leisure, and `parental satisfaction' in each period and maximizes

E0

T∑t=0

βtu(ct, lt, P

jt

),

where ct is determined by Equations (1) and (2),

lt = 1− btτb − wd

t τd − wft τf − τ c (kt, yt)− nt, (3)

and parental satisfaction of the woman is given by

P jt = ξjρ1−mt

(ηjf

(kb

t , ybt

)σ−1σ +

(1− ηj

)f (ka

t , yat )

σ−1σ

) σσ−1

, (4)

where σ is the elasticity of substitution between biological and adopted children, ηj is the weight type jwomen put on biological children, ξj is the weight type j women put on parental satisfaction, ρ ≤ 1 capturesthe decrease in parental satisfaction a woman has from having children out of wedlock (which could be aresult of stigma associated with single motherhood), and f is a parental satisfaction input function, whichincreases in its �rst argument.

4.2 Characterization of Optimal Decisions

Consider a type j woman. At adult age t, her state variable is st =[p, xt, ωt,mt, p

ht , kb

t , ybt , k

at , ya

t

]. Given

the �nite horizon of a woman's decision problem, we can solve it recursively backwards by starting in thelast period from V j

T r+1 = 0. Since women cannot bear children after t = T b, the decision problem of a typej woman at age t ∈

{T b + 1, ..., T r

}can be characterized by:

V jt (st) = max

nt≥0

[u

(ct, lt, P

jt

)+ βV j

t+1 (st+1)],

where st+1 =[p, xt + I(nt ≥ n), ωt,mt, p

ht , kb

t , ybt + 1, ka

t , yat + 1

]and the maximization is subject to Equa-

tions (1), (2), (3), and (4), evaluated at bt = wdt = wf

t = 0.

12

During her child-bearing periods t ∈{0, ..., T b

}, a woman's decision problem can be characterized by

V jt (st) = max

wbt∈{0,1}

[wb

tπb (t, ωt) V j1

t (st) +(1− wb

tπb (t, ωt)

)V j0

t (st)],

where V jbt (st) is the value function of a type j woman who will have b additional biological children at the

end of the period. This value function, in turn, can be characterized by

V jbt (st) = max

wdt∈{0,1},wf

t ∈{0,1},nt≥0

[u

(ct, lt, P

jt

)+ βEt

{V j

t+1 (st+1)}]

,

where

st+1 =[p, xt + I(nt ≥ n), ωt+1,mt+1, p

ht+1, k

bt + b, yb

t+1, kat + wd

t + wft , ya

t+1

],

ybt+1 =

kbt

(yb

t + 1)

+ b

kbt + b

, and yat+1 =

kat (ya

t + 1) + ydwdt + yfwf

t

kat + wd

t + wft

,

with ωt+1, mt+1, pht+1, yd, and yf determined stochastically as described above and with the maximization

subject to Equations (1), (2), (3), and (4).

4.3 Baseline Model Performance

Due to the complexity of the above dynamic model, it is not possible to obtain analytical solutions describinga woman's decision. Therefore we use numerical simulations to solve the model and to show that themechanisms embedded in the model allow us to match the stylized facts on adoption demand reportedearlier. For example, we expect women with higher productivity to delay forming a family due to human-capital accumulation considerations, and thereby be more likely to be faced with low fecundity at the timethey decide to have children. This e�ect and the higher opportunity cost of childbearing for high productivitywomen both induce these women to adopt children at higher rates, as observed in the data.

Functional Form Assumptions We assume that the period utility function takes the form

u(ct, lt, P

jt

)= ln ct + Blγt + P j

t .

Furthermore, we adopt a piecewise-linear speci�cation in age for the functions τc and f . In particular,assume that there is a grid on ages [y0, y1, ..., yG] , where y0 = 1 and yG corresponds to the age after whichall children in the household become adults, and assume that

τc (k, y) ={

0 if y ≥ yG

αick

β if y ∈ [yi, yi+1) for i = 0, ..., G− 1

and

f (k, y) ={

αGf kθ if y ≥ yG

αifkθ if y ∈ [yi, yi+1) for i = 0, ..., G− 1 .

Baseline Calibration We calibrate the above model to match several observations from the early 1980's,using the 1982 cycle of the NSFG and the 1984 panel of the SIPP where possible. We consider two educationgroups: those with a college degree or more education and those with some college or less education. Wechoose πω (t) to match average fecundity statistics by age and πb (t, ω) to match average time until child-bearing by fecundity status and age in the NSFG. We set τ b to equal half a year. τd, τf , rd, and rf arecalibrated to re�ect the average waiting time for a domestic and international adoption multiplied by thefraction of time parents spend on the adoption process during waiting and the average cost of domesticand international adoption based on Marshner (1999). Fd and Ff are calibrated to the age distribution ofchildren adopted domestically and internationally as reported by the Adoption and Foster Care Analysisand Reporting System (U.S. Children's Bureau, AFCRS Reports, various years) and the Immigration and

13

Naturalization Service (U.S. Department of Homeland Security (2004)). ρ is chosen to match the incidenceof child-bearing out of wedlock in the NSFG, while πm (t, e) and ∆ are chosen to match the distribution ofage at �rst marriage for women by education group and the average age di�erence between husbands andwives in the SIPP Marital History Supplement.

The variables in�uencing labor-market outcomes are calibrated as follows. rh and rw are derived byseparately estimating the model-implied log-wage equation for gender g and education group e

lnwit = ln rg

(e, xi

t

)+ εi

t i ∈ (e, g) ,

non-parametrically (that is, allowing for a distinct value of rg for each possible value of education andexperience) using data from the SIPP Employment History Supplement. The distribution of productivityfor gender g and education group e is then calibrated to approximate the distribution of εi

t by a �nitedistribution. Given the support of ph, H (ph | p) is calibrated to match the distribution of productivity ofthe husband conditional on the productivity of the wife in the SIPP. Finally, γ is calibrated to match theaverage labor-supply elasticity of women as reported by Blundell and McCurdy (1999), while B is calibratedto match average hours worked by women in the SIPP.

As in Caucutt, Guner, Knowles (2002), we set q = 0.5 and ν = 0.5 as the midpoints of the intervalsprovided by Cutler and Katz (1992), who report ranges of estimates for these parameters, based on theiranalysis of the U.S. poverty line and other available estimates. As for the preference and time cost parameters,ξj are chosen to match the average number of children and their distribution in the SIPP; ηj and σ are chosento match average adoption rates and the incidence of adoption-seeking by fecundity status in the NSFG;the age grid and αi

c are calibrated to time use data provided by Robinson and Godbey (1999); and αif is

calibrated to replicate the di�erence in the adoption of domestic versus foreign-born children by choosingthe payo� from very young children appropriately.

4.4 Changes in Adoption Demand Since the 1980's

After carefully calibrating our model to data from the early 1980's, we examine whether our model is capableof capturing the observed changes in the adoption market in response to changes in the environment facedby women. In particular, we allow for the following changes in the model

1. Change in the distribution of educational attainment among women.

2. Increase in the the age at �rst marriage captured by a change in the function πm (t, e).

3. Improvement in female labor-market opportunities captured by a change in the distribution of p andan increase in the returns to experience rw.

4. Changes in the availability of in-vitro fertilization treatments (IVF) captured by a change in πb(ωl, t

).

5. Changes in the time and monetary costs of adoption (due to subsidies and pro-adoption policies)captured by changes in τd, rd and τf , rf .

6. Changes in the age distribution of children available for domestic adoption due to changes in relin-quishment patterns captured by a change in the distribution F d.

7. Changes in the stigma associated with single motherhood captured by a change in ρ.

We then compare outcomes predicted by our model to those observed in the data. In terms of outcomes, wefocus on the total number of adoptions, the incidence of adoption by the mother's demographic and fertilitycharacteristics, and the mix of adoptions between domestic and foreign adoptions. We then examine theimportance of each of the above described changes separately in explaining adoption market outcomes.

4.5 Policy Experiments

In addition to capturing the changes in adoption market outcomes due to observed changes in the environmentin which women make childbearing and adoption decisions, our model provides a useful tool for analyzingcounterfactual policy changes aimed at in�uencing adoption market outcomes. We intend to experiment

14

with di�erent policy alternatives such as an increased subsidy for all adoptions, an increased subsidy fordomestic adoptions only, or for domestic adoptions involving children of a certain age, or an improvementin the legal process of domestic adoption leading to a decrease in the waiting time for a domestic adoption.

5 Broader Impacts of the Proposed Project and Future Directions

This study is an attempt to �ll an important gap in the literature on fertility decisions by explicitly modelingadoption as an alternative to childbearing. We expect our proposed research to make several importantcontributions both to scholarly research and to policy making by providing new evidence and valuableinsights. First, we contribute to the literature on family economics by formally investigating adoption asa means of family formation. Second, through the rigorous econometric analysis of the determinants ofadoption demand and supply using micro data sources, we provide empirical results that are useful forresearchers in economics and other disciplines, as well as practitioners. Third, by developing and calibratinga formal model of adoption seeking, we provide a vehicle for analyzing the e�ect of possible policy changeson adoption outcomes.

In this proposal, we develop and study a model of adoption seeking that we expect to be able to matchsalient features of observed fertility and adoption choices of women in the data. Our model has severalsimplifying assumptions, however, and relaxing some of these may provide further insights for better policydesigns. In future work, we consider the following extensions. First, we plan to add additional dimensionsin which adoptable children could di�er, most importantly, in health status and race. As data indicate thatchildren with disabilities or from ethnic minorities tend to have a lower chance of �nding adoptive parentsat a given age, incorporating these variables would allow us to examine a broader set of policy issues. Thesecond extension departs from the assumption that all children require a �xed amount of child care timeand allows for �exibility in parents' investment in their children, both in amount and in form (time ormonetary). By incorporating this extension, we can examine the impact of di�erent policies on the humancapital accumulation of adoptive children. Third, we plan to endogenize marriage choice and allow for thepossibilities of divorce and remarriage. With this more realistic characterization of the marriage market, wecan analyze not only unrelated but also related (stepparent) adoptions in our model. Last, but not least, weplan to extend the model by adding the possibility of having an unintended birth and then explore women'sdecision to relinquish their children for adoption. After incorporating both the demand- and supply-side ofthe adoption market, we can then study the general equilibrium e�ects of possible policy changes.

6 Results from Prior NSF Support

Éva Nagypál has been granted NSF support in the amount of $182,853 (under award number SES-0551352)for the period March 1, 2006-Feb 28, 2009 for her project �A Quantitative Study of the Extent, E�ciency,and Cyclical Behavior of Job-to-Job Transitions�. To this date, Nagypál has completed a manuscript �Labor-Market Fluctuations, On-the-Job Search, and the Acceptance Curse�, which is currently under review, andprepared a second manuscript �On the Extent of Job-to-Job Transitions�, which has been presented at NBERSummer Institute and at various invited seminars.

15

REFERENCES CITED

• Bachrach, C. (1983): “Adoption as a Means of Family Formation: Data from the National Survey of Family Growth,” Journal of Marriage and the Family, 45(4), 859-865.

• Bachrach, C. (1986): “Adoption Plans, Adopted Children, and Adoptive Mothers,” Journal of Marriage and the Family, 48(2), 243-253.

• Bachrach, C., P. Adams, S. Sambrano and K. London (1990): “Adoption in the 1980’s,” Advance Data from Vital and Health Statistics, No. 181. Hyattsville, MD: National Center for Health Statistics.

• Bachrach, C., K. London and P. Maza (1991): “On the Path to Adoption: Adoption Seeking in the United States, 1988,” Journal of Marriage and the Family, 53 (3), 705-718.

• Becker, G. (1981): A Treatise on the Family. Cambridge, MA: Harvard University Press. • Becker, G. and R. Barro. (1988): “A Reformulation of the Economic Theory of Fertility,”

Quarterly Journal of Economics, 103: 1-25. • Bernal, R. (2006): “The Effect of Maternal Employment and Child Care on Children’s

Cognitive Development,” Northwestern University, manuscript. • Bernal, R. and M. Keane (2005): “Quasi-structural Estimation of a Model of Child Care

Choices and Child Cognitive Ability Production,” Northwestern University, manuscript. • Bernal, R. and M. Keane (2006): “Child Care Choices and Children’s Cognitive

Ahievement: The Case of Single Mothers” Northwestern University, manuscript. • Berry M., R. Barth, B. Needell. (1996): “Preparation, Support, and Satisfaction of

Adoptive Families in Agency and Independent Adoptions,” Child and Adolescent Social Work Journal, 13:157–183.

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

• Bjorklund, A., M. Lindahl and E. Plug (2006): “ Intergenerational Effects in Sweden: What Can We Learn from Adoption Data?” Quarterly Journal of Economics, forthcoming.

• Blundell, R. and T. McCurdy. (1999): “Labor Supply: A Review of Alternative Approaches,” in O. Ashenfelter and D. Card (eds.), Handbook of Labor Economics, volume 3: 1559-1695. Amsterdam: North-Holland.

• Bonham, G. (1977): “Who Adopts: The Relationship of Adoption and Socio-Demographic Characteristics of Women,” Journal of Marriage and the Family, 39(2), 295-306.

• Case, A., I. Lin and S. McLanahan (2000): “How Hungry Is the Selfish Gene?” Economic Journal, 110(466), 781-804.

• Case, A., I. Lin and S. McLanahan (2001) "Educational Attainment in Blended Families," Evolution and Human Behavior, 22 (4), 269-289.

• Chandra, A., J. Abma, P. Maza and C. Bachrach (1999): “Adoption, Adoption Seeking, and Relinquishment for Adoption in the United States,” Advance Data from Vital and Health Statistics, No. 306. Hyattsville, MD: National Center for Health Statistics.

• Caucutt, E., N. Guner, and J. Knowles. (2002): “Why do Women Wait? Matching, Wage Inequality, and the Incentives for Fertility Delay,” Review of Economic Dynamics, 5(4): 815-855.

• Cunha, F. and J. Heckman. (2004): “The Technology of Skill Formation,” University of Chicago, manuscript.

• Cunha, F., J. Heckman, L. Lochner, and D. Masterov. (2006): “Interpreting the Evidence on Life Cycle Skill Formation,” in E. Hanushek and F. Welch (eds.), Handbook of the Economics of Education, 697-812. Amsterdam: North-Holland.

• Cutler, D., and L. Katz. (1992): “Rising Inequality? Changes in the Distribution of Income and Consumption in the 1980s,” American Economic Review, 82(2): 546-551.

• Das, M. and T. Sjogren (2002): “The Intergenerational Link in Income Mobility: Evidence from Adoptions,” Economics Letters, 75: 55-60.

• Eckstein, Z., and K. Wolpin. (1989). “Dynamic Labor Force Participation of Married Women and Endogenous Work Experience,” Review of Economic Studies, 56: 375–390.

• Ellwood, D., T. Wilde, and L. Batchelder. (2004): “The Mommy Track Divides: The Impact of Childbearing on Wages of Women of Differing Skill Levels,” Russell Sage Foundation Working Paper.

• Erosa, A., L. Fuster, and D. Restuccia. (2002). “Fertility Decisions and Gender Differences in Labor Turnover, Employment, and Wages,” Review of Economic Dynamics, 5: 856–891.

• Fang, H. and M. Keane. (2004): “Assessing the Impact of Welfare Reform on Single Mothers,” Brookings Papers on Economic Activity, volume 1: 1-116.

• Flango, V. and C. Flango (1995), “How Many Children Were Adopted in 1992?” Child Welfare 74: 1018-1032.

• Fisher, Allen (2003): “Still ‘Not Quite As Good As Having Your Own?’ Toward a Sociology of Adoption,” Annual Review of Sociology, 29, 335-361.

• Freundlich, M. (1998): “Supply and Demand: The Forces Shaping the Future of Infant Adoption,” Adoption Quarterly, 2, 13-46.

• Greenwood, J., N. Guner, and J. Knowles. (2003): “More on Marriage, Fertility, and the Distribution of Income,” International Economic Review, 44(3): 827-862.

• Hansen, M. and B. Hansen. (2006): “Economic Analysis of Children from Foster Care,” Child Welfare, 85(3): 559-583.

• Hollingsworth, L. (2000): “Who Seeks to Adopt a Child? Findings from the National Survey of Family Growth (1995),” Adoption Quarterly, 3, 1-23.

• Marshner, Connaught (1999). Adoption Factbook III. Washington, D.C.: National Council for Adoption.

• Maza, P. L. (1984): “Adoption Trends: 1944-1975,” Child Welfare Research Notes #9 (U.S. Children’s Bureau, August 1984), pp. 1-11. Child Welfare League of America Papers, Box 113, Folder: “Adoption-Research,” Social Welfare History Archives, University of Minnesota.

• Maza, P. L. (1999): “Recent Data on the Number of Adoptions of Foster Children,” Adoption Quarterly, 3, 71-81.

• Medoff, M. (1993): “An Empirical Analysis of Adoption,” Economic Inquiry, January 1993; 31(1): 59-70.

• Mullin, C. and P. Wang. (2002): “The Timing of Childbearing among Heterogeneous Women in Dynamic General Equilibrium,” NBER Working Paper, #9231.

• NAIC (2004): U.S. Department of Health and Human Services (2004). How Many Children Are Adopted in 2000 and 2001? Washington, D.C.: National Adoption Information Clearinghouse.

• National Center for Social Statistics, Adoptions, 1968-1975. Washington D.C.: National Center for Social Statistics; Child Welfare Statistics (later, U.S. Department of Health, Education, and Welfare).

• O’Halloran, K. (2006): The Politics of Adoption: International Perspectives on Law, Policy and Practice. Dordrecht: Springer.

• Olivetti, C. (2006): “Changes in Women's Aggregate Hours of Work: The Role of Returns to Experience,” Review of Economic Dynamics, October 2006; 9(4): 557-587.

• Plug, E. (2004): “Estimating the Effect of Mother's Schooling on Children's Schooling Using a Sample of Adoptees,” American Economic Review, March 2004; 94(1): 358-68.

• Plug, E. and W. Vijverberg (2003): “Schooling, Family Background, and Adoption: Is It Nature or Is It Nurture?” Journal of Political Economy, June 2003; 111(3): 611-41.

• Plug, E and W. Vijverberg (2005): “Does Family Income Matter for Schooling Outcomes? Using Adoptees as a Natural Experiment,” Economic Journal, October 2005; 115(506): 879-906.

• Regalia, F. and V. Rios-Rull. (2001): “What Accounts for the Increase in Single Households?” University of Pennsylvania, manuscript.

• Robinson, J. and G. Godbey. (1999): “Time for Life: The Surprising Ways Americans Use their Time,” University Park, Pennsylvania: The Pennsylvania State University Press, 2nd edition.

• Sacerdote, B. (2000): “The Nature and Nurture of Economic Outcomes,” NBER Working Papers #7949.

• Selman, Peter (2002). “International Adoption in the New Millennium; The ‘Quiet Migration’ Revisited,” Population Research and Policy Review, 21(2): 205-225.

• U.S. Census Bureau. (2000): Adopted Children and Stepchildren: 2000. www.census.gov/prod/2003pubs/censr-6.pdf

• U.S. Children's Bureau, Child Welfare Statistics – Adoptions, 1951, 1955, 1957-1967. Washington, D.C.: U.S. Children's Bureau Statistical Series.

• U.S. Children's Bureau. Adoption and Foster Care Statistics (AFCRS) Report, 1997-2005. http://www.acf.dhhs.gov/programs/cb/index.htm

• U.S. Department of Homeland Security. Yearbook of Immigration Statistics, 2002-2005. http://www.uscis.gov/graphics/shared/statistics/yearbook/

• U.S. Immigration and Naturalization Service. Statistical Yearbook of the Immigration and Naturalization Service, 1978-2001. Washington, D.C.: U.S. Immigration and Naturalization Service.

• Ventura S. and C. Bachrach. (2000): "Nonmarital childbearing in the United States, 1940-1999." National Vital Statistics Reports, 48(16).

• Weil, R. H. (1984): “International Adoptions: The Quiet Migration,” International Migration Review 18(2): 276-293.

• Waddoups, J. (1997): “Female Labor Supply: Adoption and the Labor Force Participation Decision,” American Journal of Economics and Sociology, April 1997; 56(2): 243-55.