The Determinants of Children's Attainments: A Review of Methods and Findings

by Robert Haveman, Barbara Wolfe
Citation
Title:
The Determinants of Children's Attainments: A Review of Methods and Findings
Author:
Robert Haveman, Barbara Wolfe
Year: 
1995
Publication: 
Journal of Economic Literature
Volume: 
33
Issue: 
4
Start Page: 
1829
End Page: 
1878
Publisher: 
Language: 
English
URL: 
Select license: 
Select License
DOI: 
PMID: 
ISSN: 
Abstract:

The Determinants of Children's Attainments: A Review of Methods and Findings

BY ROBERT HAVEMAN and

BARBARAWOLFE

University of Wisconsin-Madison

The authors gratefully acknowledge financial support froin the Russell Sage Foundation, from a grant by the Assistant Secretay of Planning and Evaluation in the U.S. Department of Health and Human Seroices to the Institute for Research on Pooerty, and from the Economics Program of the National Science Foundation. We are oey grateful to Kathryn Wilson for her assistance with this paper, especially her significant contributions to the tables; to Elizabeth Eoanson for her editorial work; and to Dawn Duren for her excellent typing. The paper has benefited from comments and criticism of four anonymous referees, John Antel, Jere Behrman, Andrea Beller, Mary Corcoran, Jonathan Crane, Sheldon Danziger, Greg J. Duncan, B. J. Kiker, Susan Mayer, Robert Plotnick, and David Ribar.

I. Introduction

HE LEVEL of the nation's investment in children is enormous. Government expenditures on elementary, secon- dary, and postsecondary schooling alone totaled about $372 billion in 1991, about 6.6 percent of GNP (U.S. Bureau of the Census 1992, Table 211). In addition, hundreds of billions of dollars are spent for housing, feeding, and clothing chil- dren, for transporting them, for provid- ing nonparental care services, and for as- suring provision of health care services. Perhaps the largest of all costs is the im- plicit value of the time that parents spend nurturing, monitoring, teaching, and caring for their children.

Table 1presents a rough-and lowerbound-estimate of the annual social in- vestment in the 66.5 million American children (ages 0-18), who in 1992 com- prised about 27 percent of the popula- tion. Our calculations distinguish the in- vestments made by the public sector from those made by sarents and other

i I

individuals. We estimate that annual ex- penditures on children total about $898 billion, nearly 15 percent of GDP. The largest component of the private cost of children is the direct exnenditure of nar-

L L

ents. Another important component of parental costs is the time spent by par- ents who forgo either work or leisure time in caring for children in the home.1

1We assume that the total time that a mother spends in either work or child care is 2080 hours per year. If she works full time, year round, child care time is taken to be zero. The child care time for a nonworking mother is taken to be the time worked by women with her educational level but without children. This procedure presumes that, be- cause of children, the mother sacrifices this coun- terfactual level of work time. The child care time for a mother who works part time, year round (taken to be 1040 hours) (note continued on p. 1831)

TABLE 1 EXPENDITURESON CHILDREN

1992 Dollars (in millions)

By Parents

Direct costs" ($7,579 per child)
Food 20.7%
Housing 14.4%
Transportation 22.6%
Other (recreation, health care, clothing, etc.) 42.3%

Indirect costs (opportunity cost of mother's child care
time; $1,693 per child)*

By Government

Elementary and secondary educationc
Federal 5.8%
State 45.3%
Local 40.5%
Private and others 8.4%

Social services program"
Early childhood developmente
Foster care
Delinquency, abuse, and violence serviced
Child welfare
Child support enforcement
Summer youth employmentc
Other

Legal system Crimes by childrenc Crimes against childrend (includes

juvenile justice and missing programs) Housing Lower-income housing assistance Low-rent public housing Federal food programs Food stampsd National school lunch programc School breakfast and milk Child and adult care Summer feeding of children Commodities donated to child nutrition Health carec Maternal, child health programs Medicaid Social security benefits to childrenc Aid to Families with Dependent Children (AFDC)c Earned Income Tax Credits

By NonproJ'zt Organizationsc

TOTAL^

Expenditures per child (in dollars)
Total expenditures as percentage of GDPi

a From Thomas Espenshade (1984), updated using 1992 CPS data to correct for changes in the labor force
participation patterns of women and the size of families.
b Calculated using 1992 CPS data for women's labor force participation (including full-time and part-time) by level
of education and presence of children by age, and the U.S. Bureau of the Census (1992) for average earnings by
level of education for full-time and part-time work.

TABLE 1(Cont.)

CU.S. Bureau of the Census (1992). Expenditures on higher education total $157 billion (federal, 12.5%; state,

28.9%; local, 2.5%; other, 56.1%). U.S.Office of Mangement and Budget (1993; for Fiscal Year 1994). Including services through Head Start, comprehensive child development centers, day care assistance, and depen-

dent care. fincluding services through runaway youth, child abuse, emergency protection, abandoned infants, dmg abuse protection, and family violence. gwendell Primus, Federal Expenditures on Children-FY 1994, communication from staff of the U.S. Department of Health and Human Services, Fall 1994. h~otalis net of AFDC, food stamps, housing, foster care, and Earned Income Tax Credit. 'U.S. Congress. Joint Economics Committee (1993) (for GDP).

Neglecting any costs of mother's child care time beyond 40 hours and all fa- ther's child care time, and assuming that child care time equals the difference be- tween actual work time and an estimate of the mother's work time if she did not have children, we arrive at a lower bound estimate of the opportunity costs of child care time of nearly $1,700 per child.2 The public share totals $333 bil-

is taken to be the time in excess of 1040 hours that women with her educational level but without children actually work. This difference, then, is as- sumed to be the reduction in work time due to the presence of children, and taken to be the opportu- nity cost in terms of hours of the children. For all women, opportunity cost child care hours are val- ued by a predicted wage rate based on women of the same educational level. The estimated value of child care time for each mother is divided by the number of children less than 18 years in her household. This calculation neglects the o nity cost of father's child care time and of Pportu

orgone leisure.

2If we had assumed that all deviations of mother's work time from full-time, year-round work are the opportunity cost of child care, our per child cost estimate more than doubles, to about $4,000 per year. With this assumption, the total parental cost of children increases from $.616 trillion to $.756 trillion, and the total cost of chil- dren from $.898 trillion to $1,038 trillion. This ex- cludes the opportunity cost of father's child care time and leisure beyond 40 hours per week. An upper-bound estimate could be obtained by as-

lion, including about $236 billion on education (excluding all expenditures on higher education) and $63 billion on transfer programs targeted on children in low-income families.3 Using our lower-bound estimate of opportunity cost, we estimate that average annual to- tal costs per child are approximately $13,500 in 1992, of which slightly more than one-third (35 percent) are public expenditures.

suming that one-half of Mayer's (1994) estimate of 76 hours of available parental child care time Tr week is actually used in child care. Valuing t is time by the average female wage rate yields a per child cost estimate of nearly $20,000 er year. This upper-bound estimate, which incrudes father's opportunity cost and forgone leisure, yields total parental costs of children of $1.835 trillion, as compared to our lower-bound estimate of $.616 trillion, and total costs of children of $2.117 tril- lion as com ared to our estimate of $.898 trillion. This upper-%ound estimate is about $32,000 per child er year, or 34 percent of GDP.

3TRe share of total expenditures devoted to children equals the proportion of recipients who are children in federal food programs, publicly provided health care, and AFDC. We do not in- clude the value of AFDC, Food Stamps, housing assistance, and foster care in the total in order to avoid double-countin . In our calculation of the public share of chil%ren2s expenditures, we include these as public expenditures and subtract them from parents' direct costs.

The amount of family resources allocated to children, the nature of these re- sources, and the timing of their distri- bution influence the attainments of children in the family. Children are also affected by choices made by parents re- garding such things as the number of their siblings, the type of neighborhood in which they grow up, and the number of location moves and family structure changes.

The most full-bodied statement of this model is in the work of Becker; in par- ticular Becker and Tomes (1986; see also

5 Gary Becker and Nigel Tomes (1986) refer- ence a number of early studies that attempt to in- tegrate famil behavior regarding fertility, marital patterns, andhuman capital investment into mod- els of income transmission and inequality. See also Martin Browning (1992).

Haveman & Wolfe: The Determinants of Children's Attainments 1833 Heredity

A

Parents' Abilities / Parents' Education I // Final Schooling. Level /+Income
/Fam~lv Income Post-school Investment I
Figure 1.Home Investments in Childrena  
aAdapted from Leibowitz (1974).  

Becker and Tomes 1979; and Becker 1967, 1981). In this framework, children begin life with a genetic endowment transmitted by their natural parents, apart from any decisions by parents to alter the endowment. The transmission of the endowment (and in some versions the augmented or eroded endowment) is described by a Markov process, in which the degree of "inheritability" is taken to be greater than zero, but less than one. On average, parents with levels of educa- tional attainment far above the mean will produce children who attain high levels of schooling, but not so high relative to the mean as those of the parents. By much the same process, children also in- herit cultural endowments-for example, a commitment to learning or musical skills. These "inheritances" translate into human capital, and into earnings when rented on the labor market.

Parents care about the economic capa- bilities and success of their children, and can influence their human capital and earnings by making "expenditures on their skills, health, learning, motivation, 'credentials,' and many other characteristics" (Becker and Tomes 1986, p.

S5).6 These expenditures depend on pa- rental preferences, income, and fertil- ity.7 Parents can also affect the economic

6The model has similar implications if it is as- sumed that parents value the utility of their chil- dren, rather than their human capital or attainments. See Becker and Tomes (1979).

7If nonhuman capital can be bought or sold in efficient markets, if the debts incurred by parents for investments in their children can be assigned to them when they become adults, and if the "en- dowed luck" of children is known by parents to their investment in them, parents can act::: the (private) optimal level of human capital and attainments of their children by borrowin the future earnings of their offspring In t%tg,"g:: no reduction in parental consumption is necessary to attain the optimal level of children's human capital (or, in some versions of the model, chil- dren's well-bein ), and the level of parental in- vestment in chilfren's human capital need not be directly related to parental income. If capital mar- kets are inefficient, however, arents can invest in the human capital of their chiydren only by reduc- ing their own consumption or the consumption of their children, selling assets, or raising their own work effort or that of their children. In this case,

arental income is a direct determinant of chil- %en's human capital, as the cost of resources allo- cated to children is no longer constant across fami- lies. These market constraints also imply that public expenditures on children will increase the total amount invested in them, and that there ex- ists a tradeoff between the number of children and the investment in each (Hanushek 1992).

This basic framework can be expanded by al-

10 Both the Becker-Tomes and Leibowitz mod- els assume that all household decision makers share a common utility function (or, that families have a dictator who makes choices based on his own preferences or those of another person). This assumption has been subject to criticism by those who view family decisions to be the outcome of bargaining within the household among individu- als with specific utilit functions. Marjorie McEl- roy (1990) has formuited Nash-Stackelberg type bargaining models of the family decision process; Duncan Thomas (1990) and Mary ean Horney and McElroy (1988) have attem ted to test the common preference vs indivijual preference models, without particularly convincing results (Paul Schultz 1990). More recent results, and analyses, suggest behavior more consistent with in- dividual preference models (Behrman, Pollak, and Taubman 1995; and Behrman, Rosenzweig, and Taubman 1994).

Other Perspectives. A variety of other hypotheses regarding the determinants of children's development and attainments can be found in the social science literature. One of the most prominent of these is the "working mother perspec- tive." Mother's absence from the home, it is postulated, may reduce the level of control, guidance, and monitoring given to the child. Conversely, mother's work may also be associated with increased pa- rental income, offsetting the reduction

in child care time.1" second of these

more specialized perspectives is known

as the "economic deprivation perspec

tive," which suggests that growing up in

poverty has adverse effects on children's

development, net of those factors that

are related to low family income. The

deprivation perspective is difficult to dis-

tinguish from the "welfare culture per-

spective," which ascribes harmful effects

on children's aspirations and attainments

from the dependence and stigma associ-

ated with the receipt of public assistance

(Jacqueline Macauley 1977).

These perspectives have much in com- mon with a nonlinear version of the eco- nomic framework with inefficient capital markets. In such a model, parental in- come-and factors related to income, such as family structure, asset holdings, and work effort-become a direct deter- minant of investments in children, as the resource cost of investments in children varies with family traits and choices (see footnote 8). In such a framework, it is difficult to distinguish variables reflect- ing environmental/cultural factors emphasized by sociologists and psycholo- gists from economic variables such as family income, living in a single-parent (earner) family, or the receipt and level of welfare benefits.

111. Toward a More Comprehensive Economic Perspective

Both economic and other social sci- ence perspectives on the determinants of children's attainments have emphasized the role of parental (or family) circum- stances and choices, often to the neglect of other im~ortant considerations. A

L

12 Mavis Hetherington, Kathleen Camara, and David Featherman (1983) discuss the effects of parental time and presence on children's develop- ment An economic interpretation of this perspec- tive would emphasize the tradeoff between mone- tary and time resources in the production of child quality.

more com~rehensive framework would

I

view the attainments of children as de- pendent on three primary factors-the choices made by the society (or govern- ment) that determine the opportunities available to both children and their Dar- ents (the "social investment in children"), the choices made by the parents regarding the quantity and quality of family resources devoted to children (the "parental investment in children"), and the choices that children make given the investments in and opportunities avail- able to them.

In such a framework, society (govern- ment) employs a wide variety of policy instruments-taxing, spending, and regu- latory policies, judicial pronouncements, moral suasion-in setting the basic envi- ronment within which families and chil- dren make their choices. All of these in- vestments entail both costs and benefits, and government can be viewed as choos- ing among these options in order to maximize its objective (e.g., the collec- tive well-being, somehow defined) sub- ject to both economic and political con- straints.13 With reference to Figure 1, government's investment in children can be viewed as indirectly affecting chil- dren's attainments through its effect on the level of home (or parental) invest-,

13Among the most important of government's choices that affect children's attainments are those involving resources devoted to schooling and to the quality of the neighborhoods in which children grow up (e.g., crime rates, lack of sanitation, park and communi facilities). Even more broadly, governmental 7ecisions set the social and cultural environment and make clear what are and are not society's standards and ex ectations for minimally acceptable behavior an$ erformance; current

ublic admonitions designel to reduce the preva- fence of nonmarital teen births are an exam~le. James Coleman (1988) em hasizes the poteGtial role of "social capital" in chijbren's attainments in his view, social (e.g., governmental) organizations can create a structure of su port, trust, expecta- tions, and nurture based on tie character of social relations-social capital-that contributes to chil- dren's attainments apart from more explicit and measurable inputs such as resources or personnel.

ment, and as directly affecting children's success. The choices made by parents also in- fluence how children develop, and the

A'

level of success thev achieve. Parents

J

have objectives, just as do governments. In making choices that reflect these ob- jectives-subject to their resources and other constraintsl4-families make deci- sions concerning household size and structure, consumption levels and saving, work and leisure, and the allocation of income and time. Even more basicallv.

2'

parents choose the sort of monitoring, disciplinary, nurturing, and expectational environment in which their children are raised. Taken together these choices de- termine the level of "parental investment in children."

The maximizing framework that underlies decisions made by society (gov- ernment) and parents that affect chil- dren's success applies as well to the decisions made by children themselves. In this view. chicdren are also decision makers seeking to make themselves as well off as possible.l5 It is presumed that they have weighed carefully the benefits and costs associated with the options available to them, and have made the choices which we observe, given their re- sources and the constraints that they face.16

14111 some cases, social (governmental) deci- sions may serve as constraints on parental choice. For example, some parents are constrained in what the earn or whether they work by the qual- ity of education made available to them by the public school system when they were growing up, or by the failure of public macroeconomic policies to ensure full employment.

15 Baerbel Inhelder and Jean Pia et (1958) sug- gest that children's ca acity for un dperstanding the relationship between pehavior and outcomes-the ability to reason in an "if/then" framework-is developed by the age of 13to 15.

16For example, a teenage girl observed to have given birth out of wedlock is inter reted as havin made a choice that leaves her at Last as well off given the constraints that she faces, as the alterna- tive choice of not iving birth, at least in terms of expected utility. Tfe decision to give birth out of

This more comprehensive economic framework reflects a choice-based view of the world-governments, parents, and children all have their own utility func- tions and resource constraints, and they make choices that best serve their inter- ests in light of these. The choices made by society (government) and parents re- late to investments made on behalf of children; only the choices made by chil- dren themselves reflect their own utility functions and constraints, and these choices determine the investments they make in themselves.

Our characterization of the process of children's attainment reflects a sequen- tial view of the world. Society (govern- ment) acts first, making some direct investments in children, but more impor- tantly setting the economic environment in which both parents and children oper- ate. Given this environment, parents choose how much to work and earn and how much time to spend with their chil- dren and then, given their income, they decide how much time and income to devote to their children. They also make decisions about family structure and lo- cation that serve their own interests, but which also affect their children. Finally, given their own talents, the resources that have been invested in them, and the incentives that they confront, children pake choices about their education,

wedlock carries with it gains in the form of ublic welfare, social service, health care, and jo%-spe- cific training and education benefits. This decision also offers independence from parental control that may be perceived as oppressive. The costs of the nonmarital birth choice include the sustenance cost of the child, child care costs (if market work or continued schooling is chosen), and the forgone earnings from employment opportunities or mar- riage attributable to the presence of the child. Other effects of the choices may include reduced pressure to attend traditional schools (with poten- tial discipline, failure, and boredom correlates) or to work in unpleasant low-skill jobs, increased feelings of worth and "being needed," and the ability to form a community with other young women in like circumstances.

their fertility and family structure, and their work effort.17 We observe the out- come of these choices-children's attainments.

IV. On Modeling the Determinants of Children's Attainments

As in other areas of empirical research, both the question posed and the data and modeling constraints influence the design of studies seeking to identify the determinants of children's attainments and to measure the relative im- portance of them.18 These considerations also determine the extent to which research is able to reveal the nature of the process by which children succeed or fail.

Consider, first, the question posed. At one extreme, researchers may have little interest in the process of attainment, and seek only a measure of particular gross relationships. The numerous "mobility studies" illustrate this approach (Sections V.A and V.B, below). Here, the question is, "To what extent are children's outcomes (e.g., earnings) related to those of their parents?" The higher the estimated correlation between parents' and children's earnings, the more structured is the social and economic system; a lower correlation implies a more economically mobile society. To estimate this simple relationship, re

17In fact, these decisions may be interdepen- dent, rather than sequential. For example, if par- ents are viewed as not providing certain services to children that society views as desirable (e.g., im- munizations), government may elect to provide these services directly, or to regulate parents to ensure that such services are provided.

18Here, children's "attainments" are taken to be those outcomes in young adulthood with impor- tant implications for ultimate economic success- human capital (education), work and earnings, and (for young women) teen nonmarital childbearing and welfare recipiency. The noneconomic litera- ture on children's attainments covers a variet of other children's outcomes, including school bebav- ior, delinquency, and emotional security.

searchers may use data on annual (or life-cycle adjusted, or multi-year) earn- ings for parents and their children, and nothing else. And the estimated relation- ship gives no insight into the process by which parents' success influences that of their children.

At the other extreme, researchers may seek to understand the nature of the at- tainment process. The research process in this case is more difficult, as illustrated by both the Leibowitz-style model of Figure 1 and the comprehensive eco- nomic perspective on children's attainment outlined in Section 111. The latter framework, for example, implies the need for a multi-stage structural model. In a first stage, the determinants of gov- ernment choices affecting the opportu- nity sets or constraints of parents and children would be modeled. A second stage would characterize parental choices that affect children's develop-ment-choices regarding family size and structure, income, and time allocation- as a response to opportunities and con- straints that have been affected by gov- ernmental choices. Finally, children's own choices would be modeled as reflecting their individual tastes, opportu- nities, and constraints, the last being af- fected by both social and parental choices. Such a multi-equation causal framework could specify multiple chil- dren's attainments-as well as parental and social choices-as independent, se- quentially related, or jointly determined. In turn, these relationships would imply a complex pattern of direct and indirect links, with diverse lag structures.

The lesson is clear: Depending on the question which is posed, both reduced- form and complex structural models are able to yield interesting and important findings regarding the determinants of children's attainments.

In addition to the question posed, modeling and data constraints also influ-

ence the design of studies of children's attainments, and the extent to which they reveal the underlying process. For example, besides the complex modeling problems which our comprehensive eco- nomic perspective poses, the data requirements for estimation of such a pro- cess-revealing model are daunting. First, to estimate the effect of social invest- ments (e.g., school, health care, neigh- borhood quality) on children's attainment, the level of each type of investment must be associated with each child at various ages during childhood and young adulthood. Because these in- vestments reflect social choices, the de- terminants of each form of decision must be modeled. Similar demands are imposed on the level and quality of data describing numerous parental characteristics and choices. In addition to in- formation on basic family characteristics (e.g., parental education and number of children), measures of numerous aspects of the home environment-such as fam- ily structure, parental interactions, atti- tudes, and expectations, and the level of intrahousehold stress-must be available. This information too must be asso- ciated with individual children at various times during their childhood. If parental decisions are modeled in a maximization framework, data on the opportunity set and constraints facing the families are also required. Finally, information suffi- cient to model children's own choices is necessary, including information on the benefits and costs to the adolescent of each of several available options.

Although the studies that compose the empirical literature on the determinants of children's attainment are designed to reveal some important aspect of the un- derlying attainment process, without ex- ception they are constrained by both data and modeling limitations. While some of the estimates do derive from causal models, the typical study explores the reduced-form relationship between a limited number of parental characteristics or choices (e.g., family structure, income, and welfare recipiency) and some aspect of children's attainment, controlling for as many other rele- vant factors (e.g., parental education or neighborhood characteristics) as the data permit.

Implicit in this approach is the assumption that the coefficient on (or simulated effect of) a particular variable reflects its total effect on children's at- tainment; that the variable is an exoge- nous determinant of attainment, and as such independent of other potential de- terminants. Because this assumption is often violated, but to widely varying de- grees, the estimated relationships revealed in the studies must be interpreted with caution. In this, as in most other areas of empirical economic research, what one learns about important relationships from reduced-form estimates is not devoid of meaning; however, attrib- uting causality to the estimates requires evaluation of the independence of spe- cific determinants from other variables, whether observed or not. As with the re- lated literatures on the effects of school inputs on student test scores (educational production function studies) and the human capital studies of the returns to education, unassailable estimates of causal relationships describing the underlying process are not yet attainable.19

V. Empirical Studies of the Determinants of Children's Success

Efforts to identify those factors which affect children's attainments span the so- cial sciences. While we focus on studies

19 See Hanushek (1986) for a discussion of the data and modeling limitations that constrain the reliability of estimates of educational production functions; on related grounds, Charles F. Manski (1993a) casts doubt on the reliability of prevailing estimates of the returns to schooling.

by economists, we also discuss studies from other disciplines which emphasize economic factors, beginning with the early studies of sociologists and demog- raphers. We begin our review of recent contributions by discussing research on the question of intergenerational income mobility-the degree to which income status is transmitted from one generation to the next. The simple bivariate relationship estimated in these studies sug- gests the need for further exploration of the role of a more comprehensive set of social and family characteristics in deter- mining children's success. We organize our review of the recent literature by presenting more comprehensive estimates of social and family effects by out- come, moving from educational attainment to nonmarital fertility to economic status measures reflecting labor market success (earnings) and welfare recipiency.

A. Family, Education, and Children's Socioeconomic Status: An Overview of Early Studies

Within the social sciences, empirical study of the determinants of economic success dates as far back as the 1920s. The issue in these early studies was the relationship of father's occupation to son's occupation, and cross-tabulations known as occupational mobility tables measured this link. Little theory guided these studies; the purpose was simply to measure the extent of social mobility, taken to be the strength of the relation- ship between father's occupation and son's occupation.

The first causal model of this process (Otis Dudley Duncan and Ralph Hodge 1963) envisioned a socioeconomic life cycle of three stages: family, schooling, and job. Success on the job was taken to be the socioeconomic status (SES) of a person's occupation, measured by the now-famous Duncan Index of occupational prestige (0. D. Duncan 1961). Family background and schooling were viewed as determinants of SES, but schooling was treated as an intervening variable, determined in part by family background but also making an indepen- dent contribution to occupational status.

This simple model served as the moti- vation for an important line of empirical research, first in quantitative sociology and then in economics. A prominent and ambitious study by Peter Blau and 0. D. Duncan (1967) still stands as a classic in this area. In it, a system of recursive re- gression equations were fit to data from a special supplement to the March 1962 Current Population Survey (CPS). The resulting estimates described the relationships among time ordered, life cycle family background characteristics and children's schooling attainments, and in turn the effect of both of these factors on SES.

Robert Hauser and Featherman (1977) and Featherman and Hauser (1978) supplemented the standard life cycle occupational attainment framework with a human capital model in which family background and schooling choices determine wage and earnings outcomes. Using data from a 1973 replication of the 1962 CPS survey, they estimated the changes in generational mobility pat- terns from the 1960s to the 1970s by comparing the coefficients on similarly specified models using the 1962 and 1973 data.

Two other studies during the 1970s heavily influenced subsequent work in both quantitative sociology and econom- ics: the Wisconsin Longitudinal Study and the study of inequality by Jencks and his coauthors (1972). The empirical base of the Wisconsin study was a uniquely rich body of longitudinal data on 9,000 Wisconsin youths who were high school seniors in 1957. Estimates of the deter- minants of attainment relied on the life- cycle framework. However, because of the substantially larger number of fam- ily, school, and aspiration variables in- cluded in the analyses, the principal researchers using these data20 acknowl- edged the complexity of the attainment process far more openly than had earlier studies and made fewer claims regarding the causality of the relationships reported.

Whereas these studies sought to understand the determinants of the level of individual attainment, the volume by Jencks and his colleagues (1972) focused on the role of family characteristics and education in explaining inequality in SES and income. Again the life-cycle model served as the j?rimary organizing framework. Rather than exploiting data from a single survey, Jencks et al. was a synthesis, using information and estimates from a variety of studies to reach conclusions on the relative roles of fam- ily characteristics, genetic inheritance, and schooling on success. A central (and highly controversial21) finding of this work is that "luck-interpreted as fac- tors other than family characteristics, schooling, and genetic inheritance-ex- plains more than 50 percent of the variation in SES and 75 percent of that in income. From this they conclude that, in America, schooling has not had a large effect in reducing economic inequality.

This line of research revealed a substantial effect of family background (e.g., parental occupation or education) on children's occupational status or income,

20 Among the most prominent of the more than 60 published papers and research mono raphs us- ing the Wisconsin data is the work of Wifliam Sew- ell and Hauser (1975).

21 Publication of encks et al. (1972) was some- thing of a cause cdkbre within the social science community. An entire issue of the Harvard Educa- tional Review (1973) was devoted to reviews and critiques from several perspectives; see especially the essays by Alice Rivlin, Lester Thurow, and Coleman.

with up to 30 percent of the variation in attainments explained by variation in background.22 Children's schooling also played an important intervening role, with as much as 30-40 ~ercent of the

I

variation in attainment attributed to their educational choices. The extent of the indirect effect of parental background on children's atdnment operat- ing through the children's educational choice variable was an important issue in this research. Although these early stud- ies suggested that up to one-third of the measured role of education on attainments reflects the influence of family background (leaving, say, a 20 percent net effect of education), critics argued that both the limited number of family background variables included in the studies-and the fact that they are often measured with error-causes even this to be an overestimate of the role of edu- cation and, hence, an understatement of the mobility-retarding effects of family backeround.23 At least one-half of the

"

variation in attainment was attributed

220ne of the more prominent estimates of in- ter enerational transmissions was that of Sewell an$: Hauser (1975), who reported only a 0.18 cor- relation between sons' earnings and parents' earn- ings. (See Section V.B below.)

23 Samuel Bowles (1972) demonstrated analyti- call that, if family characteristics are measured wit; error, the resulting estimates dl1 understate the role of family and overstate the effect of own schooling. Bowles also showed that if the degree of measurement error in parental variables (e. , father's schooling and occupation) exceeds that Er children's outcomes, the measured effect of family characteristics will again be understated and that of schooling overstated. Because parental charac- teristics are reported in interviews of children, there is a strong presumption that measurement error is greater for parental than for children's education and income. By means of an ad hoc se- ries of adjustments usin estimated relationships from other studies, Bowfes revised previous esti- mates to reflect the effects of different degrees of error in measuring family background. He concluded that family characteristics are more impor- tant determinants of economic success (SES or earnings) than earlier researchers had found, and (as a corollary) that own schooling choices play a less important role. See also Becker (1972).

to factors other than measured family background and own education"luck," in the terms of Jencks et al. (1972), plays an important role. This research also revealed that the roles of family background and education choices on children's attainments varied by ethnic background, and that the low economic status of black offspring could be explained only partly by low-status family background; a role was attributed to educational, occupational, and wage discrimination. Studies that compared patterns in these measured determinants over time discerned a shrinking effect of family background and an up- ward trend in the role of schooling, lead- ing to an optimistic conclusion that economic mobility was on the rise and that opportunity was becoming more equa1.24

This early literature stimulated four important lines of empirical economic research that have continued to the present. The first is a rash of studies de- signed to improve estimates of intergen- erational income correlations through improved measures of fathers' earnings and adjustments for life-cycle bias (see Section V.B). The second is the extensive research using data on siblings in random- and fixed-effects models to control more completely for common family in- fluences on children's attainment, and hence to measure more accurately the effect of schooling on attainment.25 A third line of research-seeking to improve measurement of the effects of schooling on attainment-attempts to address the measurement error problems raised by Bowles (1972) through estimat-

24See Glen Cain (1974), Duane Alwin and Ar- land Thornton (1984), and Robert Haveman (1987) for earlier reviews and critiques of this early status attainment and mobility research.

25Griliches (1979) reviews the early studies. See also Behrman et al. (1980), John Bound, Griliches, and Bronwyn Hall (1986), Hauser and Sewell (1986), and Gary Solon et al. (1991).

ing the reliability and validity of survey reports of family variables.26 A final body of research has built on Becker's model of "home investments" in children. This post-1980s empirical literature employs more complete descriptions of both the characteristics and choices of children's families when they were growing up and their own education, fertility, and labor market outcomes when they are young adults. It is also distinguished by exten- sive reliance on several detailed and long-duration panel micro-data sets de- veloped during the 1970s and 1980s. The following discussion emphasizes this last body of empirical research.

B. The Determinants of Children's Attainments: A Review of Recent Findings

Recent Studies of Intergenerational Mobility. A long research tradition in both sociology and economics has sought to measure the simple correlation between fathers' socioeconomic status-oc- cupational SES, earnings-and that of their sons (see Sections IV and V.A). Most of the early studies used either the income or occupational status of the fa- thers and sons and were inconsistent in the extent to which they controlled for life-cycle considerations. Cross-sectional data were used in which sons reported on parental economic status; hence eco- nomic status measures were "one-yearwindow snapshots." These proxy reports by adult children of current parental in- come (taken as an indicator of parental income when the child was growing up) or their recollections of parental income when they were growing up also contain serious measurement error, suggesting downward bias in the estimated relation- ship. Moreover, adjustments made to eliminate life-cycle bias resulting from

26Becker and Tomes (1986) review several of these studies.

TABLE 2a RECENTSTUDIES USING PANEL DATA:AND ESTIMATION

MOBILITY DATA

OF ECONOMIC

Study Data
  -
Behrman and PSID: Sons and
Taubman (1990) daughters 18-34 in
  1984 in own
  households
Solon (1992) 348 father-son pairs
  from random sample
  of PSID: sons have
  positive earnings in
  1984 and are 2533,
  fathers have positive
  earnings 1967-71
Zimmerman (1992) 876 father-son pairs,
  both work full-time,
  250 maximum pairs
  used in reported
  estimates NLS
Buron (1994) 253 father-son pairs
  from random sample
  of PSID, sons have +
  earnings 1984-88
  and are 2533 in 1984

Time Period

Outcome, 1984: determinants, 1975-84

Outcome, 1984: father's earnings, 1967-71

Outcome, average over 1965-1981 or 1981 only, some control variables, 1965; father's earnings over 4 years

Outcome: 1984-88; father's earnings, 1967-71

Definition of Outcome Variables -Annual In earnings Estimation Method OLS
Ln earnings, In wage, In family income OLS, IV
Ln earned income, In wage, SES index OLS, IV, and GMM
Average In earnings OLS

Note: Abbreviations and table notes appear at foot of Table 2b.

observing fathers and sons at different ages are both crude and inconsistent. The intergenerational earnings correla- tions were on the order of .16-.20, suggesting substantial mobility, although studies based on more permanent char- acteristics such as occupational status found somewhat higher correlations and lower mobility.27

Tables 2a and 2b summarize four re- cent studies that use longitudinal data and, hence, are able to employ longer- term (and consequently more permanent) measures of direct reports of in- come. These studies are also more consistent in adjusting for life-cycle dif-

"These early studies are reviewed in Becker and Tomes (1986), Becker (1988), Solon (1992), and Lawrence Buron (1994).

ferences in outcome measures than were the earlier studies.28All of these studies find correlations approximately twice as high as those of the earlier studies, in part as a result of the errors in variables and life-cycle problems affecting the ear- lier studies. Their findings call into ques- tion Becker's conclusion in 1988 that "low earnings as well as high earnings are not strongly transmitted from fathers to sons" (p. 10).

Social and Family Determinants of Children's Attainments. In Tables 3-6 we describe several recent studies of the social and parental determinants of a va- riety of attainments of adolescents or

28All four of these studies are based on small numbers of observations and are sensitive to the exclusion of zero earners.

TABLE 2b RECENTSTUDIESOF ECONOMICMOBILITYUSINGPANELDATA:RESEARCHRESULTS

Background Characteristics,

Social, Parental, and Own

Choices Controlled Study (Adjusted) for:

Behrman and Ageo Taubman (1990) GendeP

Race"
Solon (1992) Zimmerman (1992) Age Income adjusted for family size (needs) in alternative estimation Age (life-cycle) Experience

Buron (1994) Experience, race-ed-occ experience

Estimated Coefficient (t-statistic), Log Son's Earnings

on Log Father's Earnings

Father's recent year earnings: .27 (9.80) Average of father's earnings: .80 (24.40) Father's recent year earnings: .25 (3.38) Average of father's earnings: .41 (4.44) Father's recent year earnings: .36 (5.76) Average of father's earnings: .54 (6.90) Exp only, .37 (5.50); race-ed- occ " experience .46 (7.01)

Other Results

Parental income: stronger the longer the observed period of parental income

Parental income: stronger the longer the observed period of father's earnings

Father's earnings: stronger the longer the observed period of father's economic status

Correlation higher if people with disabilities excluded

Shows estimates sensitive to treatment of zero earners.

Abbreviations: PSID -Michigan Panel Study of Income Dynamics

NLS -National Longitudinal Survey

SES -Index of Socioeconomic Status

OLS -Ordinary Least Squares

IV -Instrumental Variables

GMM -Generalized Method of Moments

a In regressions with interaction terms only

youths-high school graduation (Tables 3a and 3b), years of schooling (Tables 4a and 4b), teen nonmarital fertility (Tables 5a and 5b), and labor market success (Tables 6a and 6b). All of these studies extend the simple intergenerational mo- bility estimates by estimating the rela- tionships between children's outcome variables and more or less extensive sets of social and family background variables.

Three criteria have guided our decisions regarding which of the many exist- ing studies to include in our discussion. The first criterion is the quality of the studies, based on our appraisal of the data and estimation methods used. Sec- ond, all of the included studies have some "economic" orientation; each emphasizes one or more social or parental choices or characteristics which reflect economic conditions (e.g., state welfare benefit levels, family income, or poverty status). Finally, we have emphasized those that rely on longitudinal (panel) micro-data.

Each of the columns of Tables 3-6 describes a particular characteristic of the studies. The "determinant" variables are roughly grouped into categories describing social (governmental), paren- tal, and ''own" choices, consistent with

TABLE 3a STUDIESOF THE FAMILYAND NEIGHBORHOODDETERMINANTS DATA

OF HIGH SCHOOL GRADUATION: AND ESTIMATION

Definition of Estimation
Study Data Time Period Outcome Variables Method
Mayer (1991) HSB: 26,000 10th graders Outcome: 1980-1982; Dropped out of OLS and logit
  in 1980, followed up in determinants: 1980 high school models
  1982   between 1980  
      and 1982 = 1  
Ribar (1991) NLSY: 4741 women aged Outcome: graduated Graduated high Bivariate probit
  14-21 in 1979, from high school by age school by age model
  1979 to 1985 20; determinants: 20 = 1  
    1979    

Astone and HSB: About 10,000 high Outcome: 1986: Hieh school -Probit models

McLanahan (1991) school sophomores in determinants: completion by
1980, interviewed over 1980-1982 1986 = 1
1980-1986

Haveman, Wolfe, PSID: 1258 children aged Outcome: 1987; Graduated high Probit model Spaulding (1991) 0-6 in 1968,19-23 in events and school by 1987 circumstances: 1987 = 1 1968-1983

probit, nonparametric models

Sandefur, McLanahan, NLSY: 5246 youths aged Outcome: 1985; High school, GED OLS and probit and Wojtkiewicz 14-17 in 1979 living determinants: 1979 graduation by models (1992) with parents 1985 = 1;

college attendance Brooks-Gunn et al. PSID: 1132 black and Outcome: age 20; Dropped out of Logistic

(1993) 1214 white women determinants; age high school by
between 14 and 19 in 14 age 20
1968-1985

Note: Abbreviations and table notes appear at foot of Table 3b.

our comprehensive economic frame-ated from high school), and continuous work.29 variables, indicating the extent of attain-

A variety of measures of the attain-ment (e.g., annual earnings). While some ments of children are used in the stud- of the studies use ordinary least squares ies, including both categorical dummy estimation methods, most employ maxi- variables (e.g., whether the child gradu- mum likelihood techniques (e.g., probit,

tobit); a few employ simultaneous esti- "In most cases, we interpret neighborhood

mation methods designed to characterize

variables as social investments. However, neigh- borhood variables also reflect parental choices, interrelated or joint outcomes (e.g., hav- and could be included in that category. ing a teen nonmarital birth and sub

TABLE 3b STUDIESOF THE FAMILY DETERMINANTS RESEARCHRESULTS

AND NEIGHBORHOOD OF HIGH SCHOOL GRADUATION:

Background
Study Characteristicsa Social Investment Choicesa
Mayer (1991) Black = 1: (-) 1% School mean SES:(-) 5%
  Hispanic = 1:(-) 1% School proportion black:
    (-) ns
    School proportion Hispanic:
    (+) 1%
    School mean math score:
    i-)1%~,
    School mean educational
    expectation: (+) 10%
    School mean parents female
    head: (+) 1%
Ribar (1991) Black = 1:(+) 10% State max. AFDC benefits
  Hispanic = 1:(-) ns (-) ns
    Max, food stamp benefits
    (-) ns
    Urban resident at age 14
    = 1: (-) 5%
    South resident at age 14
    = 1:(-) 5%
Astone and Gender NR Region NR
McLanahan Race NR  
(1991)    

Parental Investment
Choicesa

Parents SES: (-) 1% Female-headed family in loth grade = 1:(+) 1%

Mother only = 1:(-) 1% Mother and stepfather = 1(-) 1% Other family structure = 1:(-) 1% Number of siblings: (-) 5% Foreign language at home = 1:(-) ns Mother's education: (+) 1% Mother in labor force = 1:(-) ns Magazines at home = 1:(+) 1% Newspapers at home = 1:(+) 1% Family library card = 1:(+) 5% Single-parent family = 1:(-) 5% Stepparent family = 1:(-) 5% Other family = 1: (-) 5% Parent's aspirations: (+) 5% Mother monitors school progress = 1:(+) 5% Father monitors school progress = 1:(+) ns General parental supervision = 1:(+) 5% Parent talks with child = 1:(-) ns Parental SES (parental education, occupation, income, possessions) NR Number of siblings -NR

Own Choice Determinants"

Math score: (-) 1% Own education expectation in 10th grade: (-) 1%

Gave birth before age 20 = 1:(-) 1% PDV earnings no HS: (-) ns PDV earnings HS: (-) ns Attend religious services often = 1:(+) 1% Attend religious services infreq. = 1:(+) 1% Study Haveman, Wolfe, and Spaulding (1991) Manski et al. (1992) Haveman & Wolfe: The Determinants of children's Attainments 1847 TABLE 3b (Cont.) Background Parental Investment Characteristicsa Social Investment Choicesa Choicesa Own Choice Determinantsa Nonwhite = 1: (+) ns Years in SMSA: (-) 5% Parental time in Female = 1:(+) ns preschool years: Nonwhite (x)female: (+) ns (+) ns Father high school grad. Firstborn = 1: = 1: (+) 1% (+) ns Father some college Head foreign born = 1:(+) 1% = 1: (+) ns Father college grad. Grandparents = 1:(+) 1% poor = 1: (+) ns Mother high school grad. = 1:(+) 1% Mother some college = 1: (+) 1% Mother college grad. = 1: (+) ns Number of years in poverty: (-) ns Years in poverty x AFDC: (-) 10% Years mother worked: (+) 1% No. of location moves: (-) 1% No. of parental separations: (-) ns No. of parental remarriages: (+) ns No, of other changes in family: (-) ns No. of siblings: (-) ns Catholic = 1:(+) 5% Jewish = 1:(+) ns Protestant = 1:(+) ns Black = 1:(+) 1% North central resident Mother high school grad. Hispanic = 1:(-) 1% = 1: (+) nsb = 1: (+) 1% Female = 1:(+) 1% West resident = 1: (-) ns-other some college South resident = 1:(+) ns" 1:(+) 1% Southern born = 1:(-) nsb Mother college grad. = 1: (+) 1% Father high school grad. = 1: (+) 1% Father some college = 1: (+) 1% Father college grad. = 1:(+) 1% Nonintact family at age 14 = 1:(-) 5% Mother's education > father's = 1: (+) 1% TABLE 3b (Cont ) Background Study Characteristics" Sandefur, Race NR McLanahan, Gender NR and Woj tkiewicz (1992) Brooks-Gunn Calendar year at age et al. (1993) 14 (+) 1% Black (-) ns Social Investment Clloices" Otllers want child in college = 1:(+) nsc Others don't want cllild in college = 1:(-) 1% Attitudes of others missing = 1: (-1 1% % families with income < $10,000 (-) ns % families with income < $30,000 (-1) 1% % income < $10,000 x Black (+) ns % income < $30,000 x Black (+) 5% % income < $10,000 x Inc/needs < 1.5 (+) ns % income > $30,000 x Indneecls < 1.5(+) ns % black (+) 10%~' % families wit11 children and female headed (+) 5%1'( % with pblic assistance (+) nsil % males not in labor force (+) ns(1~ 40% + poor and < 10% families \vith income > $30,000 (-) nscl Parental Investment Choices" Stepparent at age 14 = 1:(-1 1% Single parent at 14 = 1:(-1 1% No parent at 14 = 1: (-) 1% Stepparent at ages 14-17 = 1: (-1 1% Single parent at ages 14-17 = 1: (-) 1% No parent at ages 14-17 = 1: (-) 1% Intact to step- or single parent at ages 14-17 = 1:(-) 1% Other family structure changes at ages 14-17 = I. (-) 1% Adjusted family income: (+) 1% Parent doesn't want child in college = 1: (-) 1% Parent education NR Number of siblings N R Income/needs (-) 1% Mother's education (-1 1% Female head (+) 5% Own Choice Determinants" Self-esteem: (+) 1% Study Background Characteristics" Social Investment Choicesa Brooks-Gunn et al. (1993) ManageriaUprofessional < 5% (+) 1% ManageriaUprofessional 5% -10% (+) ns Abbreviations: PSID Michigan Panel Study of Income Dynamics HSB High School and Beyond Survey PUMS Public Use Microdata Samples NLSY National Longitudinal Survey of Youth CPS Current Population Survey NLS National Longitudinal Survey Parental Investment Choicesa Own Choice Determinants" SES -Index of Socioeconomic Status OLS -Ordinary Least Squares AFDC -Aid to Families with Dependent Children ns -not significant NR -No results reported a(-), (+) = negative, positive coefficient; statistical significance level indicated. bBased only on probit model. refers to person child identified as the most influential person in their life. If parent, listed as parent's opinion. "his neighborhood variable was the only neighborhood variable included in the regression. The regression did control for the family-level variables. Results reported are for tract-level neighborhood variables. Zip-code level variables had different sign or significance level. sequent welfare recipiency). Several of the studies of the determinants of teen out-of-wedlock births view this outcome as an age-dependent probabilistic phe- nomenon, and employ hazard rate esti- mation methods.30 The extensiveness of variables describ- ing social and parental investments in children ranges widely across the stud- ies. Only a few use specific indicators of social (governmental) choices affect- ing children's attainments; the use of local indicators of abortion accessibility in studies of nonmarital childbearing (Shelly Lundberg and Plotnick 1990, 1995) is noteworthy. Most of the studies that attempt to measure the effects of social investments rely on variables that describe living environments (e.g., crime 30 Hazard function estimation methods have also been emplo ed when the dependent variable is right-censored The problem of truncated records due to the duration-constrained nature of most longitudinal data sets is regularly encountered when attainments during young adulthood (e.g., years of school completed) serve as the de- pendent variable. rates or poverty incidence) at the re- gional or neighborhood level to capture differences in living conditions that may be affected by policy measures. Informa- tion on parental investments in children range from very sparse family-based in- formation to extensive and detailed time- specific variables constructed from longi- tudinal data describing a wide range of parental decisions or family characteristics. Only a few of the studies attempt to estimate a structural model of the choices of the adolescent child or youth; the study by Greg Duncan and Saul Hoffman (1990) of the nonmarital birth decisions of African American girls and the study by Behrman, Rosenzweig, and Taubman (1994) of postsecondary educa- tion are the exceptions. However, a num- ber of the studies attempt to characterize (at least partly) the opportunity costs or the benefits associated with a particular choice. Few of the studies include prior choices made by the child. Those that do TABLE 4a DETERMINANTS AND ESTIMATION STUDIESOF THE FAMILYAND NEIGHBORHOOD OF YEARSOF SCHOOLING:DATA Definition of Outcome Study Data Time Period Variables Estimation Method Datcher (1982) PSID: 552 male heads age Outcome: 1978; Years of schooling OLS, race specific 23-32 in 1978, living determinants: 1968 with parents in SMSA in 1968 Hill and Duncan PSID: 854 youths living Outcomes: ages 27-29 Years of schooling OLS, gender specific ( 1987) with parents, age 14-16 (1983); determinants: ages in 196%72 and 27-29 in 1416 (196%72) 1983 Krein and Various NLS surveys: 2544 Outcome: 1980; event and Years of schooling OLS; race-gender Beller (1988) matched mother-son and circumstance completed at age 26 specific mother-daughter pairs determinants: ages 0-18 Case and Katz Poor Boston Outcome and determinants: Years of schooling OLS and probit (1991) neighborhoods in 1989: 1989 models 1200 youths aged 17-24 Crane (1991) 1970 Census PUMS: Outcome and determinants: Graduated or in high Piecewise linear 96,000 1619-year-olds 1970 school = 1 logit model living with parents Duncan (1994) PSID: 3,439 white and Outcome: 1991; Years of schooling OLS, race and sex black teens observed determinants: ages 1&16 completed specific between the ages of 16- 22 in the period 196% 1991. Only individuals in metropolitan areas of the U.S. Graham, Beller, March/April CPS match Outcome: 1988; Five education variables, OLS, probit and Hernandez file (including child determinants: 1987-88 including years of (1994) support data): 5038 schooling completed, children ages 16-20 high school dropout, living with mother in entered college 1988 Behrman et al. NLSH72, resurveyed in Outcome: 1986; Postsecondary attainment Multinomial logit (1994) PETS, 1973-86: 9110 determinants: 1972 (none, two years, four youths who were high years) school seniors in 1971-72 High school test scores Quality of post- secondary school Note: Abbreviations and table notes appear at foot of Table 4b. tend to include only religion (or religios- titative magnitude of the effects of the ity), or a measure of school performance. social, parental, and "own" choice vari- Only for the earnings outcome (Tables ables estimated in the studies. These 6a and 6b) have children's own choices measures would answer such questions (e.g., own schooling) been formally in- as "What is the effect of growing up in a corporated into the models. poor family on children's schooling at-We attempted to summarize the quan- tainment?" They would also allow one to Haveman & Wolfe: The Determinants of Children's Attainments 1851 TABLE 4b STUDIESOF THE FAMILYAND NEIGHBORHOOD OF SCHOOLINGRESEARCH OFYEARS RESULTS DETERMINANTS Background Own Choice Study Characteristicsa Social Investment Choicesa Parental Investment Choicesa Determinantsa Datcher (1982) Age: (+) l%Vercentage neighborhood white: Father's education: (+) 10%; (-) ns; (+) ns (black)" (+) 1% (black)" Average neighborhood income: Mother's education: (+) 10%; (+) 1%; (+) ns (black)" (-) ns (black)" Rural origin = 1:(-) ns (+) 1% Parent's family income: (black)" (-) ns" City origm = 1: (+) ns; (+) 1% No. siblings: (+) 1%; (black)" (-) ns (black)" Southern origin = 1: (-) nsL Parent's educational aspirations for the child: (+) 1%" Parent's own job decision important = 1:(-) ns; (+) 1% (black)" Hill and Duncan Black = 1: (+) South = 1: (-) nsC Father's education: (+) 1%~ Catholic (1987) 5%~ City size: (-) nsc Mother's education: (+) 1%; = 1: (+) nsc (+) ns (women). Father's SES: (-) ns; (-) 5% women)^ Head self-employed = 1: (+) ns; (+) 1% (women). Mother's work hours: (-) 5%; (-) ns women)^ Father present = 1: (+) nsc No. of siblings: (-) 1%~ Total family income: (+) 1%~ Father's labor income: (+) 1%~ Mother's labor income: (+) nsc Any father's labor income = 1: (-) ns; (-) 1% women)^ Any mother's labor income = 1: (-) ns; (-) 10% women)^ Any asset income = 1: (+) 1% Any welfare income = 1: (+) ns; (-) 1% women)^ Any other income = 1: (-) nsc Add'l dollars of father's labor income: (+) 1%~ Add'l dollars of mother's labor income: (+) ns; + 5% (womenp Add'l dollars of asset income: (-) ns; + 1% women)^ Add'l dollars of welfare income: (-) ns; (+) ns women)^ Add'l dollars of all other income: (+) ns; (-) ns women)^ 1852 Journal of Economic Literature, Vol. XXXIll (December1995) TABLE 4b (Cont.) Background Study Characteristicsa Social Investment Choicesg Krein and Beller South = 1: Mother's education: (+) 1%" (1988) (+) nsd Father's education: (+) 1%; (+) ns (black)" Know father's education = 1: (-) 1%; (-) ns (black men); (+) ns (black women)" Family income during high school: (+) 1%; (+) ns (women, black men)" Years single-parent family: (-) 5%; (-) 1% (black men); (-) ns (women)" Preschool years in single-parent: (-) 5%; (-) ns (black men, white women); (-) 10% (black Elementary years in single- parent: (-) ns; (+) ns (women)d High school years in single- parent: (+) ns; (-) ns (women, black men)" No. of siblings: (-) 1%; (-) 5% (black men); (-) ns (white women)" Mother ever worked = 1: (-) 1%; (+) ns (black men); (-) ns (women)" Reading materials in home = 1: (+) ns; (+) ns (black men); (+) 5% (~omen)~ Case and Katz Female = 1: (+) (1991) ns Black = 1:(+) 5% Crane (1991) Percentage high-status workers in neighborhood: (+) 5% Parental Investment Choicesa Own Choice Determinantsa Both parents present = 1: (+) 1% Mother less than 20 at birth = 1: (-) ns Parents not married = 1:(-) ns Parents' years of schooling: (+) 1% Family member in jail = 1: (-) 10% Family member with drng/alcohol problem = 1:(-) ns Adults in family attend church often = 1: (-) ns Background Study Characteristics0 Duncan (1994) Calendar year child turned 16 (-1 1%; (-) ns (black males, white females) Graham, Beller, Age: NR and Hernan- Male = 1: NR dez (1994) Black = 1: NR Hispanic = 1: NR TABLE 4b (Cont.) Social Investment Choicesa % families income < $10,000 (+) 1%; (+) ns (black males, white femalesy % families income > $30,000 (+) 1%; (-) ns (black males) % individuals black (-) ns; (-) 1% (black males); (+) ns (black females) % female-headed families with children (-) ns; (-) 1% (black females) % adult women working 26+ weeks (+) 5%; (+) ns (black males, white females); (-) 1% (black females) Northeast (+) ns; (+) 5% (white females); (+) 1% (black females) North central (+) ns; (+) 1% (blacks) West (+) ns; (+) 1% (blacks) Northeast = 1:NR North central = 1: NR South = 1:NR SMSA = 1:NR Central city = 1: NR Own Choice Parental Investment Choicesa Determinantsa Income /needs (+) 1% % time in mother-only family (-) ns; (+) ns (white females) (+) 1% (black females) Mother's schooling (+) 1% % income from welfare (-) ns; (-1 1% (blacks); (-) 5% (white females) % time mother worked (-) ns; (-) 1% (blacks) Mother's education: (+) 1% Own child = Age mother gave birth: (+) ns 1: NR Mother works = 1:(+) 1% Total family income: (+) 1% Child snpport eligible = 1: (-) 1% Child support eligible not receiv- ing child support = 1: (-) 1% Child snpport eligible receiving child support = 1: (-) ns Awarded child support = 1:(+) ns Due child support = 1: (+) ns Received child snpport = 1: (+) 1% Nonintact family = 1: (-) 1% Widowed-mother family = 1: (-) Child snpport income = 1:(+) 10% Father has visitation rights = 1: (+) nsf Father lives in same state = 1: (+) nsf No, of days of parental contact: (-) nsf Other non-child-support = 1: (-) 5% TABLE 4b (Cont.) Background Study Characteristics" Behrman et al. Female = 1: (-) 5%; (-) 10% (2 year). Birth order: (+) ns; (-) ns (white) Hispanic = 1: (-) 10%; (+I ns (2 year). Some Investment Choicesa ent rate age 18-21: (+) ns; -) ns (nonwhite, 4 unemplOr" year). Average state tuition, 4 year col- lege: (-) ns; (-) 5% (for 2 year); (+) ns (nonwhite 4 year); (+) 5% (nonwhite 2 year)e StateAocal higher education ex- penditures: (-) ns; (+) 5% (2 year);#(+) ns (nonwhite 4 year) Ave. family income in the high school (-) ns; (-) 10% (2 year); (+) 5% (nonwhite 4 year); (+) ns (nonwhite 2 year); > 50% high school teachers with grad degree = 1: (+) 5%; (-) 10% (2 year); (+) ns (non- white) Private or Catholic high school (+) 10%; (+) 5% (nonwhite 4 year); (+) ns (nonwhite 2 year) Abbreviations: PSID -Michigan Panel Study of Income Dynamics HSB -High School and Be ond Survey PUMS -Public Use Microdata Samples NLSY -National Longitudinal Survey of Youth CPS -Current Population Survey NLS -National Longitudinal Survey NR -No results reported Parental Investment Choices" Mother high school = 1:(+) 10%; (+) ns (nonwhite 4 year); (-) ns (white 2 year, nonwhite 2 year)^ Mother some college = 1: (+) 5%e:,,i+), 10% (nonwhite 4 year); (+) ns (nonwhite 2 year) Mother 4+ years college = 1: (+) 5%; (+) ns (nonwhite 4 year); (-) ns (nonwhite 2 year)e Dad high school = 1:(+) 5%; (+) ns inonwhitey Dad some college = 1:(+) 5%; (+) ns (nonwhite 2 year); (+) 10% (white 2 year, nonwhite 4 yeary Dad 4+ years college = 1:(+) 5%; (+) ns (white 2 year, non- white 4 year); (-) ns (non- white)~ Family income $6,00&8,999 = 1: (-) 5%;(+) ns (nonwhite, white 2 yearp Family income $9,000-11,999 = 1: (-) 5%; (+) ns (nonwhite, white 2 year)e Family income $12,000-14,999 = 1:(+) ns Familv income > = $15.000 = 1: . , (+) 5%;(+) ns (2 year non- white). No, of siblings: (-) 5%; (-) ns (2 year)# Father SES index: (+) ns; (-) ns (2 year). Miles to closest 2 year school: (+) 5%; (-) 5% (2 year); (+) ns (nonwhite 4 year); (-) 5% (2 year)^ Miles to closest 4 year school: (-) 5%; (+) ns (nonwhite) (+) 5% (white 4 year). Own Choice Determinantsa Achievement test score: (+) 5%% Opportunity wage: (-) 5%; (+) ns (2 year); (+Ins (non- white 4 year); (-) ns (nonwhite 2 year)' College//high school earn- ings differ- ential (+) ns; (+) 5% (2 year); i+)ns (nonwhite). SES -Index of Socioeconomic Status OLS -Ordinary Least Squares AFDC -Aid to Families with Dependent Children PETS -Post Secondary Education Transcript Study NLSH -National Longitudinal Study of the High School Class of 1972 a (-), (+) = negative, positive coefficient; statistical significance level indicated. First entry for whites; black shown only if sign or stat. sign, differs. First entry for men; second for daughters (women) shown only if sign or stat. sign. hffers. "irst entry is for white males; other race-gender groups shown only if sign or stat. sign, differs. First entry is for white, four year; other educationlrace groups shown only if sign or stat. sign. differs. f Sample limited to those eligible for child support. g When treated as endogenous, predicted scores are (+) 10% ((white); (-) ns (nonwhite). judge the robustness of the findings and to assess policy suggestions based on them. Our efforts to accomplish this-by inferring elasticities of response, or the simulated change in a dependent vari- able from a specified change in an explanatory variable-were largely unsuccessful. A primary obstacle to accom- plishing this is the substantial variation amonsz the studies in the extent to which U account is taken of the potential endogeneities among the determinant vari- ables included in the estimates.31 For ex- ample, if part of the impact on children's educational attainments of the level of parental education acts through its effect on family income, and if this relationship is not explicitly accounted for in the esti- mation, inferences regarding the effect of both parental education and family in- come on children's education derived 31Numerous other difficulties were also en countered in our attempt to make such estimates, including substantial variation in estimation meth- ods (e.g., ordinary least squares, tobit, and hazard models); the distressingly small overlap in the ex- lanatory variables included in the models; the Lrge variation in specification of the variables de- signed to indicate the same phenomenon (deter- minant) across the studies; differences among the studies in the a e of the child at which the paren- tal or neighborgood variables are measured; the aucity of studies that either report simulated ef- fects of quantitative magnitudes or provide suffi- cient sample statistics to allow calculation of the numerical magnitude of effects; the large variation in the number and the character of the covariates included in the estimation models; inconsistency among the studies in reporting effects for entire samples as op osed to specific subsamples of the po ulation (of!en ethnic and gender groups); and sutstantial variation across the studies in the specification of the outcome variable of interest (e.g., a dummy variable for high school gradu- ation, number of years of schooling completed or highest grade attained). A few studies have attempted a "meta-analysis" designed to estimate the quantitative magnitude of effects. While no technique for combining the findings from many diverse studies gains universal applause, we have found the few studies of this sort to be helpful in understanding which measured relationships have sufficient consistency in sign, significance, and magnitude to be labeled "robust," and to some insight into the quantitative magnitude oft: fects (Paul Amato and Bruce Keith 1991). from the respective estimated coefficients will be unreliable and quite differ- ent from inferences based on estimates in which this relationship is modeled. Because of these difficulties, our tables report what is comparable across the studies-primarily the signs of the esti- mated relationships between the variables of interest and the outcomes, and the extent to which the estimated rela- tionships are statistically significant. The Determinants of Children's Edu- cational Choices. Variables describing parental characteristics or choices are the most commonly used in studies of children's educational attainment. Among these, perhaps the most funda- mental economic factor is the human capital of parents, typically measured by the number of years of schooling at- tained. This variable, emphasized in the earlier studies of the intergenerational transmission of socioeconomic status, is included in virtually every study de- scribed in Tables 3a, 3b, 4a, and 4b; it is statistically significant and quantitatively important, no matter how it is defined. The human capital of the mother is usu- ally more closely related to the attain- ment of the child than is that of the fa- ther. Parental completion of high school and one or two years of postsecondary schooling are typically found to have a larger effect on children's schooling than years of parental schooling beyond that level. The income level of the family in which a child grows up is perhaps the best measure of the level of economic resources devoted to the child by the parents, and is often included in the studies of children's educational attain- ment.32 With but one exception (Linda 32It should be noted, however, that the family income variable may be a rather crude proxy of the economic resources available to a child. Often famil income is recorded only in a single and Kence measures permanent income witbeti: Datcher 1982),33 the family income vari- able is positively associated with the edu- cational attainment of the child, and the variable is statistically significant in more than half of all cases where a positive relationship is estimated. Simulated changes in family economic resources, however, are associated with relatively small changes in educational at- tainments.34 The range of elasticities is wide-about .02 to .2-for reasons we have discussed (see footnote 32). In one of the most careful explorations of the relationship between family income and children's education (Martha Hill and Greg Duncan 1987), a 10 percent increase in family income (controlling for a large number of other variables) was as- sociated with an increase in educational attainment of less than one percent.35 ror. Moreover, it may convey little about family allocation of income to children and fail to capture other economic resources devoted to the child (e.g., parental time allocation). The measurement of this variable varies widely across the studies. A few studies employ either a single year of family income or an average of income over a limited number of years; most employ the ratio of the in- come level of the family to the income needs of the family, reflecting its size and structure. Others measure the effect of having annual family income below the official family size specific U.S. line. A few of the studies use an indicator of SES which attempts to summarize the combined effect of a variety of economic resource factors. A single study (Mayer 1991) focuses on the differ- ences between income, expenditures, and the ma- terial well-being of children. Education itself can be interpreted as a measure of permanent income, a point that suggests that the full effect of income (education) would, to some extent, include the ef- fect of education (income). 33In Datcher (1982), the income level of the family's neighborhood is included in the analysis, in addition to the famil 's own income level. The otentially high (thoug{ unreported) correlation getween these two variables probably accounts for the insignificant effect of the family's own income. 34When income is measured over a long period of time or few other variables are taken into ac- count, the estimated impact of income or of pov- reater. See for example, Children's De- c:Sk% (1994). 35 See also Haveman, Wolfe, and James Spauld- ing (1991) and Gary Sandefur. Sara McLanahan, There is some evidence that the source of income matters; for example, while earned income has a positive effect on children's schooling, income from wel- fare programs tends to have a smal ler positive-or even a negative effect.36 The contribution to children's educa- tion by two other parental investment variables has been regularly stud- ied-family structure (e.g., living in a one-vs. two-parent family) and the extent of mother's work. In all of the studies that included information on family structure, growing up in a one-parent family (or experiencing divorce or marital separation) is negatively related to the level of schooling attained and in most cases is statistically significant. Es- timates of the magnitude of effect of this determinant range from modest to rather large. For example, Haveman, Wolfe, and Spaulding (1991) estimate that the probability of high school graduation of the mean child experiencing two paren- tal separations during ages 6-15 is about five percent lower than that of the child growing up in an intact family. Sande- fur, McLanahan, and Wojtkiewicz (1992) report that a prototypical child living in a one-parent family during ages 14-17 has a 16 percent smaller prob- ability of graduating from high school than a child living in an intact family during these years. The smaller simu- and Ro er Wojtkiewicz (1992). Becker and Tomes (1986) ind elasticities in the 01-02 range in their survey of earlier literature on this topic. Sheila Krein and Beller (1988) and John Graham, Beller, and Pedro Hernandez (1994) find elasticities of .O1 to .04. 3GHill and Greg Duncan (1987) find that the participation of the family in welfare pro rams has a negative and significant effect on the etucational attainment of daughters; Haveman, Wolfe, and Spaulding (1991) and Beller and Graham (1993) find a significant negative effect of welfare receipt combined with being poor, and Graham, Beller, and Hernandez (1994) find that child support has a larger impact than other sources of support. lated impact of family structure in the former study is perhaps explained by the substantially greater number of vari- ables controlling for other parental choices and circumstances (e.g., family income) which mav i be related to change in marital status.37 Evidence on the effects of mother's work on chil dren's educational choices is mixed, some studies finding a negative and sig- nificant effect on children's attainment, others finding either no significant effect or a positive impact.38 Several additional ~arental investment L factors have been found to have statisti- cally significant and quantitatively large effects on children's educational attain- ment, including the number of geographic moves during childhood,39 the number of siblings, religiousness, school- related parenting practices, and the pres- 37Amato and Keith (1991) report the results of a "meta-analysis" of the effects of living in a di- vorced or intact family during childhood on educa- tional attainment. Their analysis is based on 18 studies of this relationship, and reflects the inclu- sion of a variety of other family characteristics in the estimation. They conclude that, on average, having parents who are divorced reduces educa- tional attainment by nearly .2 standard deviations. (Using the mean and standard deviation of the sample of children included in Haveman, Wolfe, and Spaulding 1991, the effect of ex eriencing a divorce and living in a nonintact fami 9y translates into a reduction of about 10 ercent in the prob- ability of graduating from hi& school, and about one-third of a year of schooling attained.) Mc- Lanahan and Sandefur (1994) compare the pro- portion who graduate from high school having rown up in a one- versus two-parent family across Eve data sets: the differential ranged from 7 to 16 percentage points. 38Hill and Greg Duncan (1987) find that the level of mother's earnings (when the child was aged 14-16) had a less positive effect on children's educational attainment than income from other sources. See also Krein and Beller (1988) and Datcher (1982). Haveman, Wolfe, and Spaulding (1991) find a positive association between mothers working while the child is 12-15 and the child's educational attainment; Beller and Graham (1993) have similar findings for those aged 16-20. 39 Haveman, Wolfe, and S aulding (1991) and Haveman and Wolfe (1994) find a negative, sig- nificant, and large effect of this variable. ence of reading materials in the home. Most of the studies find that race is not associated significantly with educational attainment when family income and other background characteristics are in- cluded in the models; indeed, when these factors are controlled for, African American children, especially daughters, have more schooling.than do non-African Americans. The effect of the characteristics of children's neighborhood on their educa- tional attainment is the focus of several recent studies. After controlling for a wide variety of parental choice and back- ground characteristics, the income, occu- pational status, and other supportive characteristics of the neighborhood are positively associated with children's educational attainment, but often of marginally statistical significance? Estimates of neighborhood effects should be interpreted with caution; the zip code or census track data used rarely correspond to a "true" neighborhood. More important, parental choice of neighbor- hood is likely to be dependent on family economic resources; no studies have suc- cessfully modeled this causal relationship. Because the variation in family income within narrowly defined neigh- borhoods is typically small, it is difficult to separate the independent effects of family and neighborhood economic cir- cumstances on children's educational at- tainment (William Evans, Wallace Oates, and Robert Schwab 1992; and Manski 1993b). There is some evidence that the relationship between neighborhood char- acteristics and children's schooling is nonlinear; for example, children living in neighborhoods with adverse characteristics that are several standard devia- tions from the mean have significantly 40 Jencks and Mayer (1990) review the effects of nei hborhood characteristics on a wide variety of chifdren's outcomes. 1858 Journal of Economic Literature, Vol. XXXIII (December 1995) TABLE 5a OF ADOLESCENT OUTCOMES: STUDIESOF THE FAMILYAND NEIGHBORHOODDETERMINANTS OUT-OF-WEDLOCKFERTILITY DATA Study Data Hogan and Kitigawa Young Chicagoans (1985) Survey: 1078 black females aged 13-16 in 1979 Ante1 (1988) NLSY: 2302 females less than 19 years living at home in 1979 Bumpass and NSFG: 7969 females McLanahan (1989) ages 15-44 from 1982 Duncan and Hoffman PSID: 874 black (1990)f females aged 14 during 1968-1980 period Lundberg and NLSY: 1181 white Plotnick (1995)gs11 females aged 1416 in 1979 Plotnick (1992) NLSY: 1142 non- Hispanic white females aged 14-16 in 1979 who were never married Hayward, Grady, and NSFG Cycle 111: 1004 Billy (1992) females less than 20 years old An, Haveman, and PSID: 872 females Wolfe (1993) aged 0-6 in 1968 and older than 21 in 1988 Wu and Martinson NSFH: 4776 females (1993)k born after 1937 Brooks-Gunn et al. PSID: 1132 black and (1993) 1214 white women between 14 and 19 in 1968-1985 AND ESTIMATION Time Period Outcomes: 1979; determinants: 1979 Outcomes: 1981-85; determinants: 1977- 79 Outcomes: ages 1544; determinants: child's age 0-14 or age 14 Outcome: ages 14-19; determinants: age 14 Outcome: 1979-86; determinants: 1979, child's age 14, 1980-85 Outcome: 1979-84; determinants: 1979 Outcome: ages 14-19; determinants: age 14 Outcome: child's ages 13-18; determinants: child's ages 615 Outcome: child's age < 19 years, 1987- 88; determinants Outcome: age 20; determinants: age 14 Definition of Outcome Variables Estimation Methods Pregnancy = 1 Continuous time semi- Markov (hazard) model Birth by age 21 = 1 Two-stage probit Out-of wedlock birth model (mother's by age 21=1 welfare participation is first equation) Out-of-wedlock birth Proportional hazard = 1 model, race specific Out-of-wedlock birth Two-stage logit model with AFDC (in 2 years) = 1 Out-of-wedlock birth without AFDC = 1 Out-of-wedlock Nested logit model childbearing = 1; abortion = 1; pregnancy = 1 Premarital pregnancy Two-stage nested = 1;conditional on logit model pregnancy: abortion = 1; prebirth marriage = 1;premarital birth = 1 Pregnancy = 1 Continuous time semi- Markov (hazard) model Out-of-wedlock Bivariate probit model birth, ages 13-18 = 1 Out-of-wedlock birth Hazard regression < 19 years model Out-of-wedlock birth Logistic by age 20 Note: Abbreviations and table notes appear at foot of Table 5b. TABLE 5b STUDIESOF THE FAMILYAND NEIGHBORHOOD OUT-OF-WEDLOCKFERTILI~Y DETERMINANTS OUTCOMES: OF ADOLESCENT RESEARCH RESULTS Background Study Characteristicsa Hogan and Months since ex- Kitigawa act age = 11: (1985) (+) 1% Sister who is teenage mother = 1 (+) 5% Ante1 (1988) Black = 1: (+) 1% Hispanic = 1:(+) 10% Bumpass and McLanahan (1989) Duncan and Hoffman (199O)f Lundberg and Plotnick (1995)g.l~ Social Investment Choicesa West side = 1: (+) 10% Neighborhood quality = low: (+) ns Neighborhood quality = medium: (-) ns Welfare guarantee: (+) 1% State effective welfare tax rate: (-) ns Northeast, central city = 1:(+) ns; (-) ns (black). Northeast, not central city = 1:(-1 10%; (-) 1% (black). North central, not central city = 1:(+) ns; (-) ns (black). South, central city = 1: (+) nse South, not central city = 1:(-) nse West, not central city = 1: (+) ns; (-) ns (black). Northeast = 1:(+) 1% North central = 1:(+) ns West = 1:(+) ns Medium size city = 1: (+) ns Small city/~ural= 1: (+) ns Ln of ma. AFDC benefits for family of 2 in state: (+) ns Welfare guarantee: (+) 1% Restrictiveness of abortion funding: (+) 1% Abortion availability: (-) 1% Conservative welfare and abortion policies: (+) Liberal welfare and abortion policies: (-) Conservative abortion policy climate: (+) Restrictions on contraceptives: (+) 1% Own Choice Parental Investment Choicesa Determinantsa Social class = middle: (+) 10% Career aspirations = Social class = lower: (+) 5% low: (+) ns Parents not married = 1:(+) 5% No. of siblings > 4 = 1:(+) 1%" Parent control of dating = high: (+) 5%~ Parent control of dating = low: (+) l%c Mother's years of schooling: (-) 1% No. of children: (+) 5% Mother < age 20 at &st birth = 1:(+) 5% SMSA residence = 1:(-) ns Mother welfare recipient = 1: (+) l%d Both parents = 1: (-) l%@ Mother's education 12 + years = 1:(-) 1%. Father's education 12 + years = 1:(-) 1%;(-) 10% black)^ Parent income $10-20,000 = 1: Ln of earned family (-1 5% Parent income > $20,000 = 1: income at age 26: (-) 5% (-) 1% Parent income AFDC = 1:(+) 5% Mother only = 1:(+) 1% Baptist = 1:(+) Mother's education: (-) 1% Catholic = 1: (-) Mother worked = 1:(-) 1% No. of siblings (+) 5% TABLE 5b (Cont.) Study Hayward, Grady, and Billy (1992) Background Characteristicsa Age: (+) 1%;(+) 5% (blackp Age squared: (-) 1% (black) (not included in white regres- sion) Social Investment Choicesa West = 1:(-) nse South = 1:(-) nse Central city/urban = 1:(+) ns; (+) 1% (black). NoncentraVrural = 1:(+) ns; (+) 1% black)@ Noncentral/urban = 1:(+) ns; (+) 5% (black). Non-SMSNrural = 1:(+) nse Parental Investment Choicesa Mother's education: (-) 1% (pregenancy);(+) 5% (premarital birth) Mother worked = 1: (+) ns (pregnancy);(-) 1% (premarital birth) No. of siblings: (-) ns (pregnancy;(+) 1% (premarital birth) Mother only: (+) ns (pregnancy);(+) 1% (premarital birth) Mother/stepfather: (+) 5% (pregnancy);(-) 5% (premarital birth) Ln mother's education: (-) l%e Single-parent family at age 14 = 1:(+) 10%;(+) ns (black). Own Choice Determinantsa Self-esteem:(+) ns (prep);i-)1% (pre- birth) Locus of control: (-) 1% (preg); (+) 5% (prebirth) Family/gender role attitudes: (+) 1% (peg); i+)5% (prebirth) Attitude toward school: (-) 1% (prep);i-)ns ipre- birth) Education expecta- tions: (-) 1% (preg);(-1 1%(pre birth) Protestant: (-) 1% (peg);i+) ns (pre- birth) Catholic: (-) 1% (peg);i+)1% (pre- birth) Jewishlother:(-) 5% (peg);(+) 1% (pre- birth) Baptist: (-) ns (preg); (+) ns (prebirth) Time since first sexu- ally active: (-) 5%@ Prior sexual activity = 1:(+) nse Fundamentalist = 1: (-) ns; (+) ns (black)E,l Catholic = 1;(-) ns; (+) ns (black).,! No religion =1:(-) ns; (+) ns black).,^ Contraceptive use at first intercourse = 1:(-) 1%. Current contracep- tive method, pill = 1: (-) l%@ Current contracep- tive method, con- dom = 1(-) 5%~ Current contracep- tive method, other = 1: (-) nse Background Study Characteristicsa An, Haveman, Black = 1:(+) 5% and Wolfe Any religion = 1: (1993) (-) 5% Wu and Martinson (1993)k Brooks-Gunn Calendar year at et al. (1993) age 14 (+) 5%; black (+) 1% Social Investment Choicesa No. of years in SMSA, ages 6-15: (+) 10% Bad neighborhood in 1976 = 1:(-1 ns County average unemploy- ment rate, ages 6-15: (-) ns Average state welfare gener- osity, ages 6-15: (+) ns % families with income < $10,000(-) nsft % families with income > $30,000 (-) 1% % incomes < $10,000 X black (+) ns % income > $30,000 X black (+) 1% % income < $10,000 X Indneeds < 1.5(-) ns % income > $30,000 X Indneeds < 1.5(-) ns % black (-) nsl % families with children and female headed (+) nsl % with public assistance (+) nsl % males not in labor force (-) 10%' Parental Investment Choices" No. of siblings: (+) ns Mother's age at 1st birth: (-) ns Mother high school graduate = 1:(-1 1% Mother out-of-wedlock birth = 1:(+) ns No, of household location moves, ages 6-15: (+) ns No. of parental separations, ages 6-15: (+) 1% No. of parental remarriages, ages 6-15: (-) 5% Predicted average welfare ra- tio, ages 6-15: (-) nsi Parental welfare recipiency, ages 615 = 1:(+) ns Mother-only family at birth = 1:(+) nse Mother-only family at least 75% of ages 0-5 = 1:(+) nse Mother-only family at least 75% of childhood = 1:(+) nse Current family mother only = 1:(+) ns; (+) 5% (black). Current family stepfamily = 1: (+) 5%;(-) ns (black). Current family other = 1:(-) ns; (+) ns (black). No. of changes in family situ- ation: (+) 1%;(+) 5% (black). Own Choice Determinants" Background Study Characteristicsa TABLE 5b (Cont.) Own Choice Social Investment Choicesa Parental Investment Choicesa Determinantsa 40%+ poor and < 10% families with income >$30,000(+) nsl ManageriaVprofessional< 5% (+) 5% ManageriaVprofessional5% 10% (+) ns Abbreviations: NLSY = National Longitudinal Survey of Youth NSFG = National Survey of Family Growth PSID = Panel Study of Income Dynamics NSFH = National Survey of Families and Households AFDC = Aid to Families with Dependent Children SES = Index of Socioeconomic Status (-) (+) = negative, positive coefficient; statistical significance level indicated. b When the model controls for sister who is teenage mother, the estimate is insignificant. Omitted categoly is daughter does not date. d From a prior probit explaining mother's welfare participation. First entry is for whites; blacks shown only if sign or stat. sign differs. f Other variables included were whether age-26 subsample +, age-26 subsample X AFDC +, and age-26 subsample X earned income. All the coefficient estimates were insigiiificant. g Sample size and no reporting problem make nonwhite estimates unreliable. h Results from a nested logit model including possible final outcomes: pregnancy (vs, no pregnancy), abortion (vs. carry), marital birth, out-of-wedlock birth. Unconditional probability of a teen non-marital birth is product of probabilities of getting pregnant, carrying to term, and not marrying before the child is born, all of which are part of estimation, and hence carry explicit coefficients and standard errors. The sign of the effect of a variable on the probability of a teen out-of-wedlock birth is known from simulation analysis, but there is no explicit level of statistical significance level for many of the variables. Other religions is the omitted category. j Welfare ratio is defined as after tax income divided by appropriate poverty line. k Variables describing SES of father, parental education, own education, number of siblings, age, religion are also included, but results are not reported for these variables. 1 This neighborhood variable was the only neighborhood variable included in the regression. The regression did control for the family-level variables. lower educational attainments (Crane 1991).*1 The Determinants of Teenage Non-marital Childbearing. Tables 5a and 5b summarize the primary micro-data analy- ses of the determinants of teen nonmari- tal childbearing. While all of the studies have a rather rich characterization of family background and parental choice 4lOther studies, however, have failed to find evidence consistent with such an "epidemic" or "contagion" conjecture (Rebecca Clark 1992). determinants of the probability of a teen nonmarital birth, few explicitly model the girl's own choice as a response to the opportunities and constraints with which she is confronted. Greg Duncan and Hoffman (1990) contains the most explicit structural model, with variables reflecting economic opportunities (including the gen- erosity of welfare benefits) available to black women who do and do not experi- ence nonmarital birth as a teen. Both of these expected economic opportunity variables have the predicted sign; the statistical significance of the variable in- dexing economic opportunities without a birth suggests that poor employment opportunities may encourage teen non- marital childbearing. A 25 percent in- crease in the income-at-age-26-without- a-nonmarital-birth variable is simulated to reduce the probability of a nonmari- tal birth by two percentage points (from 25 to 23 percent, or by about 10 per- cent; see also Lundberg and Plotnick 1990). Few studies attempt to measure the effect of social decisions on children's attainments. However, Lundberg and Plotnick (1990, forthcoming) estimate the effect of public family planning decisions on the teen fertility outcome by constructing state-specific indicators of abortion accessibility/costs and contraceptive availability matched to individual records. These variables have expected signs and are quantitatively important determinants of the nonmarital birth outcome. For example, for white adolescents, moving from the average level of restrictions on the public funding of abortions to the ab- sence of public support of abortion costs increases the probability of an out-of-wedlock birth by about 15 per- cent. As with the studies of educational at- tainment, parental economic resources and schooling are included as determinants of the teen nonmarital birth deci- sion. The level of parental education (typically that of the mother) is negative and statistically significant in all of the studies in which it is included; parental income is negative and usually, but not always, significant. Again, there is some evidence that the source of family in- come matters; welfare receipt generally has a positive effect on the probability that teens will choose to give birth out of wedlock. The few reports of the quanti- tative effects of simulated changes in variables suggest that decreases in pa- rental income and schooling and increases in the likelihood of parental wel- fare recipiency or the generosity of available welfare benefits will lead to small increases in the probability that teen girls will experience a nonmarital birth. A number of other determinants of the nonmarital birth outcome are often statistically significant, and have relatively large estimated coefficients, including indicators of parental attitude, expectation and personality, parental monitoring and control of children, con- traceptive practice, family structure, and a variety of family stress factors (such as family disruptions and geographic moves during childhood). It is important to note that none of the studies of this nonmarital childbearing outcome includes information on the male partners of the women. Surely their backgrounds, experiences, and characteristics are relevant determi- nants of the childbearing outcome, as pregnancy, abortion, and marriage deci- sions are often jointly made. No avail- able longitudinal data contain informa- tion on mothers' male partners other than husbands-or on absent fathers-that is linked to the child whose attain- ments are being studied. This is an im- portant limitation of all of these studies, and conveys the impression that only the choices of the female teenager are rele- vant. The Determinants of Labor Market Outcomes. Tables 6a and 6b summarize the primary studies of the effects of so- cial, parental, and own choices on the wages and earnings (and, in a few cases, hours worked) of young adults. The few available longitudinal micro-data sets containing information on both the ef- fects of social and parental investments during childhood and work-related out- comes when children are in their mid- TABLE 6a OF EARNINGS AND ESTIMATION STUDIESOF THE FAMILYAND NEIGHBORHOODDETERMINANTS OF YOUNGADULTS:DATA Study Data Datcher (1982)b PSID: 552 male heads aged 23-32 in 1978, living with parents in SMSA in 1968 Kiker and PSID: 334 males 1932 Condon (1981) in 1974 not living with parents Hill and Duncan PSID: 645 persons 14 (1987) 16 and living at home in 1968 and who worked >500 hours for at least 1year, ages 25-27 Corcoran et al. PSID (smaller samples (1992) in some models); 841 male heads, ages 25- 32, 1983 Li and NSFH: Persons age 25 Wojtkiewicz 64 in 1987-88 (1992) Corcoran and PSID: 1347 males age Adams (1993)b 25-35 in 1988, and head of household at least 1year since age 24 Note: Abbreviations and table notes are at foot of Table 6b 20s accounts for the paucity of studies of these links.42 The most consistent finding of this body of research is the positive and sig- nificant effect of parental income while children are growing up on their later la- bor market performance. The elasticity estimates are in the range of .l to .3. 42At least 20 years of panel data are necessary to estimate the effects of social and parental char- acteristics and choices on the labor market attain- ments of children. Annual wage and earnings ob- servations are reliable estimates of permanent attainments only for individuals at or be ond their mid-20s. while the governmental and arental variables of relevance are those recordefduring childhood years. Time Period Definition of Outcome Variables Estimation Method Outcome: 1978; determinants: 1968 Ln 1977 annual earnings; In 1978 hourly earnings OLS Outcome: 1974; determinants: 1968 Outcome: ages 25-27 (1979- 83); determinants: ages 14-16 (1968-72) Ln 1974 annual earnings, and (as intervening variables) education, motivation and IQ score Ln wage in 1982$ OLS OLS Outcome: aged 25+ through 1983: determinants: family 1968-1982 or when son left home Outcome: 1987-88; determinants: retrospective Ln annual earnings; In wage; In hrs. Ln earnings, SES score Weighted least squares OLS Outcomes: age B+(19781988); determinants: ages Annual earnings age 25+, hourly wages, age 25+, annual hours, OLS 5-17 age 25+ Growing up in a poor family appears to have a particularly negative effect on later work and earnings, while parental schooling has no consistent effect on youth earnings. However, estimates of the effect of parental educational choices on children's labor market attainments are difficult to interpret because of in- consistency among the studies in model- ing the effects of parental choices on children's own education choices and, through this link, on their work and earnings. When the youth's own educa- tional decisions are included in models, parental education has an important im- pact on children's labor market attain- TABLE 6b OF EARNINGS RESEARCH STUDIESOF THE FAMILYAND NEIGHBORHOODDETERMINANTS OF YOUNGADULTS: RESULTS Background Own Choice Study Characteristics" Social Investment Choicea Parental Investment Choice" Determinants" Datcher Age: (+) 5%; (-) Average neighborhood in- Father's education: (-) ns; Years of education: (1982)b ns (black). come: (+) nsc (-) 1% (black). (+) l%c Parent neighborhood white: Mother's education: (+) ns; Union member (+) 1%; (+) 5% (black). (-) ns (black). = 1:(t)l%c Rural origin = 1:(+) nsc No. of siblings: (-) ns; (-) 5% Employed in durable City origin = 1:(+) 1%;(-) (black). goods industry ns (black) Parental family income: = 1: (+) l%c Southern origin = 1:(+) 5%; (-) ns; (+) 1% (black). Self-employed = 1: (+) 10% (black). Family receives transfer (-) ns; (+) ns income = 1: (-) 10%; (black). (+) ns (black). Ln nonlabor income: Parents usually carry out plans (-) ns; (-) 5% = 1:(+) ns; (+) 1% black)^ (black). Married = 1:(+) l%c No. of children: (+) ns; (-) ns (black). Kiker and Average (5-year) parental Years of schooling: Condon income: (+) 5%c (+) 1%;(-) ns (1981) Father's motivation score: (black). (-) nsc Motivation score: (+) Father's intelligence score: 5%~ (+) nsc Intelligence test Father's schooling: (-) nsc score: (+) 5%~ Father's SES: (+) nsc Years worked full- time: (+) Years of job tenure: (+) l%c Professional = 1:(+) 1%" Manager = 1:(+) 1O%c Clerical = 1:(+) 5%c Craft = 1: (+) 5%c Operator = 1:(+) 5%~ Hill and Black = 1:(-) South = 1:(-) 1%; (-) ns Father's education: (-) ns; Catholic = 1:(-) ns; Duncan 5%; (-) ns (w01nen)d (-) 1% (women)" (+) ns (women)" (1987) (wornen)d Citj size: (+) 1%" Father's SES: (-) ns, (+) Couutj unemployment rate: 1% (women)d (-) ns" Head self-employed = 1: (-) 1%; (-) 5% (women)" Mother's education: (+) l%d No. of siblings: (0)ns; (-) ns (women)d Father's labor income: (+) 1%; (+) 10% (women)" Other income: (+) ns" Father Dresent = 1: i+)5%: (+) ns (women) (controlling income)" TABLE 6b (Cont.) Background Study Cl~aracteristics" Social Investment Choice" Hill and Duncan (1987) Corcoran et Black = 1: (-) Percent on welfare in al. (1992) 1% neighborhood: (-) ns Percent on welfare in neighborhood x welfare income: (+) 5% Ln meman family income in neighborhood: (+) ns Male unemployment rate in neighborhood: (-) ns Percent female-headed families with children in neighborhood: (+) ns Parental Investment Choicea Father present: (+) ns (not cont1.011ing income)" Mother's labor income: (-) 5%;(-) ns (women)' Mother's work hours: (-) 5%; (-) ns (women)" Total family income: (+) 1%;(+) 5% (women)" Any father's labor income = 1:(-) ns; (+) ns (women)' Any mother's labor income = 1:(-) 10%; (-) ns (womeny Any asset income = 1:(-) ns; (+) ns (womenP1 Any welfare income = 1: (-) ns; (+) ns (women)" Any other income = 1: (+) ns; (-) ns (women)' Add'l dollars father's labor income: (+) 1%; (+) 10% (women)' Add'l dollars mother's labor income: (-) ns; (+) ns (wornen)" Add'l dollars asset income: (+) nsd Add'l dollars welfare income: (+) ns; (-) ns (women)" Add'l dollars other income: (+) ns; (+) 10% (women)" Ln family income: (+) 1% Ln family needs: (-) nsc Family incomelneeds: (+) 1%1 Father's earnings: (+) ns Motherlwife's earnings: (+) ns Motherlfemale head3 earnings: (-) ns Other family income: (+) nsi Nonwelfare income: (+, ), 10% Proportion of years in poverty: (-) 5% Families welfare income: (-) 1% Welfare incomelfamily income: (-) 5% Dummy for welfare income: (+) ns Father's hours of work: (+) ns Motherlwife's hours of work: (-) 10% Own Choice Determinants" Haveman & Wolfe: The Determinants of Children's Attainments 1867 TABLE 6b (Cont.) Background Study Characteristicsa Corcoran et al. (1992) Li and Black = 1:(+) nsa Wojtkiewicz Hispanics = 1: (1992) (-) ng Other = 1: (-) ns Female = 1: (-) 5%6 Age 2534 = 1: (-) 5% Age 35-44 = 1: (+) 5% Age 4554 = 1: (+) 5% Corcoran and Adams (1993)'~ Social Investment Choicesa Percent poor in neighbor- hood >30 = 1:(-) 5%; (+) ns (black). Percent of households in neighborhood with income >$30,000 = 1: (+) 10%;(+) ns (blackpa Percent of men in area unemployed when a child: (-) ns; (-) ns (black) Percent men in area unemployed when an adult: (-) ns; (-) 5% (black). Parental Investment Choicesa Mothedfemale head's hours of work: (-) ns Father's education: (-) ns Mother's education: (-) ns Mother high school = 1:(+) 5% Mother college = 1:(-) ns Parents receive public assistance = 1:(-) ns Mother-only family at age 15 = 1:(-) ns Mother/stepfather family at age 15 = 1: (+) 10% Other family type at age 15 = 1:(-) 5% Years in mother-only family: (+) ns Years in mother/stepfather family: (-) ns Years in other family: (-) 5% Change to mother-only family = 1:(+) ns Mother-only to mother/ stepfather family = 1:(-) ns Mother/stepfather to mother- only = 1: (-) ns Change during preschool years = 1:(+) ns Change during elementaly years = 1:(+) ns Change during high school years = 1:(+) ns Income/needs ratio = 0-1.25 = 1: (-) 5%. Income/needs ratio = 1.25 2.00 = 1:(-) nsc Income/needs ratio =3 or more = 1:(+) ns; (-) ns (black). Average welfare income = 1: (+) ns white)^ Welfare income = 1-5000 = 1:(-) ns (black)] Welfare income = 5000-7500 = 1:(-) ns black)^,^ Welfare income = 7500 or more = 1:(-) 5% (black)] Head's education: (+) nsc Head's average work hours: (-) ns; (+) ns (black). Own Choice Determinantsa Years of education (+) 5% SES of occupation: (+)5% Years of education: (+) 10%;(+) 5% (b1ack)c TABLE 6b (Cont.) Background Study Characteristics" Social Investment Choice" Corcoran and Adams (1993)" Abbreviations SES = Index of Socioeconomic Status PSID = Michigan Panel Study of Income Dynamics NSFH = National Survey of Families and Households AFDC = Aid to Families with Dependent Children OLS = Ordinary Least Squares Parental Investment Choice" Own Choice Determinants" Ever lived in female-headed family = 1:(-) 5%; (-) us black)^ Percentage years lived in female-headed family: (+) 5%;(+) ns black)^ "(-), (+) = negative, positive coefficient, statistical significance level indicated. Qeported results are for earnings study, which also does a similiar estimate using wage ratio as the dependent variable. cFirst entry is for whites; black shown only if sign or stats sign differs. "First entry is for males; females shown only if sign or stat. sign differs. "It is (-) 1% when the only other control variable is % on welfare in zipcode area. When other controls are added, it becomes insignificant. fIf proportion of years in poverty is included, then estimate is + but insignificant. g The four sets of fa~nily structure variables and the three dependent variables resulted in 72 interaction terms for the gender and black/white contrasts of family structure effects. In statistical tests not shown, only two of the gender interactions and four of the race interactions were significant. In general, the effects of family structure are the same for males and females and for blacks and whites. "The reported results are for earnings. Similar estimation done for hourly earnings. 'Data uses $15,000 for 1970 and $30,000 for 1980. iSpecification of welfare income variable(s) differ by race. ments through their own educational choices (see Section V.A). In a thorough exploration of the ef- fects of parental choices and neighbor- hood circumstances on children's earnings, Corcoran et al. (1992) include a very extensive set of parental and com- munity variables and find that the coeffi- cients on parental income and welfare receipt variables are significantly related to son's earnings and wages, both with and without inclusion of son's own edu- cation choice in the model. Their results are consistent with Bowles' (1972) dem- onstration that the coefficients on paren- tal choice variables (e.g., family economic resources) are reduced when son's own education choice is included in the model. However, they interpret the find- ing that such parental choices remain significant determinants of son's earnings even with this "own" choice variable included (a result which is at some vari- ance from the findings of earlier studies) as indicating that parental choices and background characteristics do not oper- ate primarily through children's own education choices (p. 72).43 The Determinants of Weljare Recipi- ency. Much of the research on the deter- minants of daughter's welfare recipiency derives from the hypothesis that welfare dependency is intergenerationally trans- 43 They note that their finding on this important issue is consistent with that of Sewell and Hauser (1975). mitted. Early studies generally found a small positive and significant link between the probability that a daughter will receive welfare and parental welfare receipt.44 Recent studies of the link between the mother's and daughter's welfare receipt have added three dimensions to the ear- lier research. The first concerns the po- tential bias in the earlier estimates due to the inability to observe parental choices or characteristics that may be re- lated to parental welfare receipt and causal to daughter's receipt. Using PSID data on a sample of sister pairs, Solon et al. (1991) estimate sibling models designed to control more c&npletely for unobserved common parental characteristics and find a high;r degree of wel- fare receipt resemblance among sisters than is estimated in the earlier studies. Second, the early studies are estimated using all daughters as sample observa- tions, whether or not their characteristics (e.g., marital status) would allow them to be benefit recipients. As Peter Gottschalk (1992a, 1992b) demonstrates, estimation of this link over a sample that includes daughters whose characteristics " exclude them from being categorically 44 See Martin Rein and Lee Rainwater (1978), Hill and Michael Ponza (1984), Greg Duncan, Hill, and Hoffman (1988), Rainwater (1987), Ante1 (1988), and McLanahan (1988). The more reliable studies of this set observe the recei t of welfare by mothers at the beginning of a pane ?'of longitudinal data and relate this outcome to the daughter's re- ceipt later in the panel. Because the primary data set used for these studies (PSID) does not readily reveal information on the sources of income of adult children who live with their parents, the re- sults may reflect relationships observed for dau h ters living independently, and hence may be %i: ased. While this problem for daughters living dependently can be reduced by careful case by case exploration of family composition for families receiving AFDC, errors may remain. This problem also afflicts some of the early studies of the deter- minants of wages and earnings described above, including those by Kiker and Carol Condon (1981) and Datcher (1982), which rely on observations of young males living apart from their parents. eligible for benefits may result in an esti- mate which is biased downward.45 Both of these recent advances suggest a larger degree of welfare dependence than was found in the earlier studies. In another more recent study (Gottschalk 1994) un- observed tendencies toward AFDC par- ticipation are modeled using AFDC par- ticipation after a daughter has left home. Finally, other recent contributions analyze the intergenerational welfare trans- mission process for a particular popula- tion-teen girls who give birth out of wedlock-and find the probability that an unmarried teen would receive welfare subsequent to an out-of-wedlock birth to be greater if the girl's mother is a recipi- ent (see Greg Duncan and Hoffman 1990 and Chong Bum An, Haveman, and Wolfe 1993). VI. What Have We Learned? A Heroic Summary Our review of recent research on the determinants of children's attainments has yielded numerous, though not always consistent, results. Moreover, because of both specification limitations and data constraints, they vary widely in their reli- ability. Here we summarize the most sa- lient and robust findings they have revealed in a series of brief propositions that may not always do justice to their subtleties. Our review of recent studies rests on a large earlier body of research on social mobility and status attainment, primarily by quantitative demographers, sociologists, and economists. These earlier stud- ies established the links between family 45Gottschalk's estimates of this link, using a sample of dau hters whose characteristics make them potentiah eligible for benefits given pro- gram rules, su gest substantial intergenerational transmission of welfare participation. However, because categorical eligibility is not a strictly ex- ogenous characteristic of these daughters, this re- sult too is open to challenge. background and children's later occupa- tional and labor market statuses as medi- ated by the child's characteristics (e.g., ability) and choices (e.g., education). In Section V.A we summarized the main findings of this research. Recent research on the extent of intergenerational earnings correlations are summarized in Table 2, above. These studies correct several data and statistical problems that affected earlier mobility research and find much less mobility across generations. They call into question the earlier conclusions that the nation is a highly mobile society, and leave far less room for "luck." The research on the determinants of children's attainments is described in Ta- bles 3-6. It has several characteristics that distinguish it from the earlier litera- ture. First, research by economists, often relying on Becker-type models of family behavior, has been far more prevalent. Second, additional measures of attainment-for example, dependence on pub- lic transfers and nonmarital childbearing-have been introduced, both as surrogates for ultimate "success" and as of interest in their own right. Third, a far more extensive list of variables describ- ing specific social and parental invest- ments in children (e.g., family structure, mother's work time, parental welfare recipiency, and neighborhood characteristics) has been studied as potential determinants of children's attainments. Finally, recent research is characterized by a heavy reliance on panel data, longer-term and more accurate measures of potential determinants of children's success, and more advanced statistical methods. The primary findings of these studies are these: 1. Children who grow up in a poor or low-income family tend to have lower educational and labor market attainments than children from more affluent families, suggesting that parental choices or attributes that result in reduced access by children to economic resources or opportunities increase the chances of low attainment. Being poor as a child also has an independent and negative effect on the probability of giving birth as a teen and of becom- ing a welfare recipient. Growing up in a family in which the mother chooses to work appears to have a modest adverse effect on educational attainment, suggesting a negative effect of the loss of child care time. However, mother's work choices do not appear to have an ef- fect on the probability that a girl will experience an out-of-wedlock birth in her teens, or be a welfare recipient, nor on educational attain- ment if the mother's work occurs during a child's teen years. In the last case, the role model or additional income effect appears to dominate. Growing up in a family that has re- ceived welfare increases the prob- ability that a girl, if she becomes a single mother, will choose welfare recipiency. The level of family in- come as well as its source appears to affect this indicator of children's success. Growing up in a welfare family does not appear to influence the probability that a teenaged girl will give birth out of wedlock. Economic incentives and opportuni- ties are often created by society through government decisions (e.g., available income support if not working, and contraception and abortion availability). These, together with market incentives and opportunities, appear to influence a variety of behaviors and attainments, including earnings, welfare recipiency, and the probability of a teen nonmarital birth. While vari- ables reflecting these incentives and opportunities typically have the expected sign and are statistically significant, the magnitude of their quantitative effect tends to be small relative to that of several forms of parental investment in children. Growing up in a single-parent or stepparent family (or experiencing a parental separation or divorce) has a negative effect on educational attainment, and larger effects are recorded for African Americans than for whites. Adverse effects of single-parent or stepparent living arrangements on the probability that a girl will experience a non-marital birth or a dissolved marriage are also recorded. There is some evidence that change in pa- rental living arrangements, rather than growing up in a single-parent family, plays a more significant role as a determinant of the probability of a teen nonmarital birth. Stressful events during childhood (e.g., changes in geographic loca- tion) appear to have large and inde- pendent negative effects on a vari- ety of indicators of children's attainments. Growing up in a neighborhood with "good" characteristics (e.g., residents with more education and in- come, and less unemployment and welfare recipiency) has a positive effect on a child's choices regarding schooling and earnings, while reducing the likelihood that a child will choose to have an out-of-wed- lock birth. There is some evidence of increasing negative marginal ef- fects of poor neighborhood quality. When family background and paren- tal choices are controlled for, being a racial minority does not appear to have a negative effect on schooling, but is positively related to welfare recipiency and the probability of a nonmarital birth. VII. A Critique of Research on the Determinants of Children's Attainments Research over the past quarter century, reflecting important advances in both data and methods, has substantially increased our understanding of the de- terminants of children's attainments. Cross-sectional survev data have in , eluded increasingly detailed information on parental choices and family circum- stances, and a growing number of continuing panel data sets with ever-longer observation periods have become avail- able.46 These rich and extended longitu- dinal data sets enable both early-in-life ex~eriences and circumstances. and out- cdmes during young adulthood'to be ob- served directly for the same individual, yielding more accurate information than the retrospective reports available in ear- lier cross-sectional surveys. However, the relatively short history of national longitudinal data collection efforts limits individual observations to a maximum of about 25 years. In addition to constrain- 46 For example, the Michigan Panel Study of In- come Dynamics (PSID) now contains detailed in- formation on a common set of about 5,000 families and the splits from these families over a 25-year period. These panel data on children's family envi- ronment and experiences early in life have permit- ted tests of a wide variety of hypotheses that researchers were unable to explore with cross- sectional survey data. Recently, data on neigh- borhood characteristics have been merged with the basic family data in the PSID, permit- ting researchers to examine the linkages of peer-group and neighborhood characteristics to children's attainments. Similarly, the National Longitudinal Surve of Youth (NLSY) now has 12 years of inLrmation on about 12,700 youths and their families; in many di- mensions this information is more extensive than that available on the PSID. However, the NLSY data has little information on youth's liv- ing arrangements and family characteristics prior to age 14. ing the accuracy of measures of both childhood experiences and environments and life-cycle accomplishments, this short time span restricts the ability to distinguish individual from cohort effects. Recent studies have also employed a variety of advanced econometric meth- ods designed to capture the effects of si- multaneous relationships among the social, parental, and own choice variables affecting children's attainment and to permit more confidence that the rela- tionships observed reflect true causal links rather than simple correlations. These methods have also enabled researchers to work reliably with longitudi- nal data afflicted with censoring prob- lems and missing information. Yet our review of recent research on the determinants of children's success reveals a number of shortcomings involv- ing research strategy, data, and methods. Our brief mention of them here can serve as a road map for future research. A. Modeling Issues and Hypothesis Testing Although the several studies summa- rized in our tables are nested in one or another social science "perspective," there is no common framework that has guided researchers regarding the choice of model specification and relevant vari- ables. While exogenous individual and family background effects appear on every researcher's list, choice among the large constellation of remaining variables is largely ad hoc. Some of the studies have been designed to test a specific hypothesis; for example, growing up in a single-parent household adversely affects children's at- tainments. In these studies, the sign and statistical significance of the coefficient on the variable of interest is estimated, controlling for a variety of other vari- ables. However, interpretation of results is difficult, as the definition of the vari- able and the selection of control vari- ables differ across the studies. As our brief summary of findings (Section VI) has indicated, there is empirical support for several of the hypotheses or conjec- tures that have motivated research in this area. Family structure (e.g., single- parent families), stress, role model, and welfare culture hypotheses can all claim some empirical support from the avail- able evidence. The economic deprivation perspective is supported in nearly all of the studies, though the marginal impact of income relative to needs does not ap- pear to be large. And the verdict is still out on the "working-mother" hypothesis. However, as we have emphasized above (Sections I1 and IV), the empirical impli- cations of the various social and psycho- logical perspectives overlap those of the economic models, making definitive tests of the various hypotheses impossible. Other studies use a wide variety of pa- rental and community variables available in a particular data set, and attempt to identify which among them appear to be significantly related to a particular attainment, holding constant the others. While researchers presume that the criti- cism of unobserved variables often levied at this body of research will be blunted by so restricting the domain of unob- served factors, these concerns remain. Even more important, there are but few studies that attempt to account sys- tematically for the interdependence among determinant variables so necessary for establishing true causal links. For example, variables measuring paren- tal education, employment, and marital status, number of siblings, and family re- ceipt of welfare benefits are often intro- duced into empirical models without rec- ognition that the parental choices they reflect may be jointly determined or causally related. Or again, both parental income and neighborhood quality indica- tors are often introduced into regression estimates without explicit recognition that the latter variables may represent choices determined by the former. Moreover, the studies we have reviewed are primarily reduced-form estimates, with little attempt to characterize the choices made as responses to economic incentives; there are few structural mod- els to be found. The wide variety of specifications and this neglect of poten- tial endogeneity problems makes any overall summary of findings less confi- dent and robust than desirable (see Sec- tion VI) and generalizations regarding the absolute and relative effects of po- tential determinants on attainment virtu- ally impossible (see Section V.B). Re- searchers in this field are not unaware of these problems, many of which are a di- rect result of serious constraints on the data and information necessary to make clean tests of alternative hypotheses and to reliably account for potential interde- pendencies and endogeneity. This critique notwithstanding, it must be recognized that difficult estimation problems arise as information describing social (governmental), parental, and "own" choices affecting children's attain- ments becomes increasingly extensive. While each of these factors may enter a correctly specified model, there is little theoretical guidance for model specifica- tion. Moreover, within any modeling framework, specifying the potential causal links among and between them and indicators of attainment increases exponentially as the domain of relevant determinant variables expands. While skilled draftsmen may be able to draw the spaghetti-like lines of exogenous ef- fects and causal and simultaneous inter- dependencies in such complex models, the constraints imposed by sample sizes, data reliability, correlation among the variables, and available econometric techniques for causal modeling make es- timation of the magnitude of these rela- tionships problematic. Given available sample sizes, statistical indicators of de- terminants often fail to have the required degree of orthogonality with respect to each other to enable reliable estimation of the independent effects. In the face of this problem, perhaps the case for reporting-and emphasizing-results that are statistically significant at lower than conventional levels (say, .25 rather than .O1 or .05) becomes more compelling. B. Data Constraints While the data available to researchers have constrained studies of the determi- nants of children's attainments, researchers have also failed either to ap- propriately use or to exploit available information. For example, the limited number and questionable reliability of variables measuring the effects of social (governmental) choices available in most micro-data sets has constrained the abil- ity of researchers to study the role of these choices in influencing parental choices and children's attainments. While a few researchers have creatively or effectively merged data reflecting these choices onto existing micro-data bases, such efforts are rare. In part because of the limited number of years of longitudinal information recorded for a specific child, few of the studies attempt to measure the differen- tial impact of social and parental deci- sions made at different times during childhood. For example, while family breakups may adversely affect a child's later attainments, there is little evidence indicating whether the effect is more or less negative if it occurs during adoles- cence or in early childhood. Because of the same constraint, a number of the studies have violated the requirement that variables describing potential deter- minants of children's attainments (e.g., parental income) reflect circumstances or parental choices that are prior to the child's choices. The lack of available data has also lim- ited the ability of researchers to sort out the life-cycle patterns of attainment. In- deed, individuals appear to follow quite different trajectories as they move toward their ultimate attainments in life. Reliance on outcomes measured during young adulthood (at best) may suggest misleading conclusions if, in fact, attain- ments by middle age or over the life course are ill-proxied by outcomes meas- ured during the mid-20s.47 While im~rovements in some of these J. dimensions seem feasible given existing data, major advances in our understanding of the determinants of children's attainments require improvements in data available to researchers. As we have indicated, longitudinal data, tracing large and national samples of children and their families over time, have become the cornerstone of research in this area. Yet few data sets contain in- formation covering a sufficiently long sweep of time to enable both parental and social (governmental) choices during childhood and an assessment of attain- ments and performance over the life course-or even to middle age-to be recorded. As a result, studies have often had to relv on observations of circum- , stances and events in late adolescence as proxies for the full range of childhood 47The sampling properties and interviewing protocols of some of the longitudinal studies also prejudice the reliability and generalizability of es- timation results. For example, in some of the stud- ies, detailed observations on attainment are avail- able only for young adults who are not living with their parents. If independent young adults differ systematically from those who have not left their parental home, basing conclusions on estimated relationships between the backgrounds and attain- ments of only the former group could be mislead- ing. This problem will become less severe over time as longitudinal data sets add waves of infor- mation, thus permittin observations of attainments of children beyon % young adulthood years. experience or outcomes during young adulthood as proxies for lifetime attain- ments (see Haveman et al. 1995). Continuation of existing panel data collection efforts seems a high priority. While longitudinal data on children and their families have improved in many dimensions over time, important problems remain. Some of these have already been mentioned-the limited period of observation on individuals, the interviewing protocols restricting obser- vations to independent youths, the lack of information on unmarried male part- ners and absent fathers. In addition, the information which is collected on indi- viduals could be enriched in several di- mensions. Most of the data sets are not explicitly intergenerational, focusing pri- marily on either parents or children; studies of the determinants of attainment could be improved with more ex- tensive and more balanced information on both parents and offspring. Such an inter-generational focus would also include in- formation on the interactions between parents and children-monitoring of ac- tivities, joint projects, or "nurturing time." Although substantial progress has been made in extending the richness of infor- mation collected on both parents and children there is still a serious problem of "variable scarcity." The following list of information on parents, neighborhoods, children, and the relationships among them, forms our assessment of the most pressing data needs in this area:48 4Qn implicit gain from the addition of individ- ual-specific information on these items is the im- proved potential for estimating statistical models that attempt to address the host of potential endo- geneity problems in this area. Such estimation re- quires identification, often in the form of constructed instrumental variables that reflect, say, parental choices (e.g., welfare participation) that are potentially endogenous to children's outcomes, which variables are unrelated to children's attain- ments. Data sets with extensive parent, child, and environment variables increase the possibility of constructing such instruments. Parental self-perceptions and self-esteem, parental expectations for and monitoring of children, parental involvement with children's school(s), and parental time spent with children for both mother and father. The health status of both parents and children. Behavioral and attitudinal attributes of parents, such as criminal activities (often resulting in incarceration), drug and alcohol abuse, and religious commitment. The interaction of children with both school authorities (e.g., truancy) and the criminal justice system (e.g., arrests, convictions, incarceration). The behavior and attainments of the siblings of children (e.g., their marital, crime, substance abuse, fertility, and labor market experiences). Contemporaneous information on the neighborhood and peer group characteristics of children as well as ties to and distance from other family members. The characteristics and qualities of the schools that children attend, and indicators of their school performance. Information enabling researchers low-cost investments designed to secure individual-specific information on a variety of these potentially important factors. ALWIN,DUANEF. AND THORNTON,ARLAND. "Family Origins and the Schooling Process: Early versus Late Influence of Parental Characteristics," Amer. Sociological Rev., Dec. 1984, 49(6), pp 784-802. AMATO,PAULR. AND KEITH,BRUCE."Parental Divorce and Adult Well-Being: A Meta-Analysis," J. Marriage and the Family, 1991, 53(1), pp. 43-58. AN, CHONG BUM; HAVEMAN,ROBERT AND WOLFE, BARBARA."Teen Out-of-Wedlock Births and Welfare Recei~t:The Role of Childhood Events and ~conGmicCircumstances," Rev. Econ. Statist., May 1993, 75(2), pp. 195 208. ANTEL, JOHN. "Mother's Welfare Dependency Effects on Daughter's Earl Fertility and Fertility Out-of-Wedlock." Unpu&ished paper, Department of Economics, U. of Houston, 1988. "The Intergenerational Transfer of Welfare Dependency: Some Statistical Evidence," Rev. Econ. Statist., Aug. 1992, 74(3), pp. 467 -" 1.3, ASTONE,NANMARIEAND MCLANAHAN, SARAS. "Family Structure, Parental Practices, and High School Completion," Amer. Sociological Rev., June 1991, 56(3),pp. 309-20. BECKER,GARYS. Human capital: A theoretical and empirical analysis with special reference to education. New York: Columbia U. Press, for the National Bureau of Economic Research, 1964. --. "Human Capital and the Personal Distribution of Income: An Analytical Approach." Woytinsky Lecture, no. 1, U. of Michi an, Institute of Public Administration, Ann Arfor, 1967. -. "Schooling and Inequality from Genera- accurately to characterize the op tion to Generation: Comment," J. Polit. Econ., portunities available (and implicit "prices" reflected) in organized labor markets, informal labor markets, marriage markets, and public program "markets" relevant to alternative opportunities facing children, youths, and young adults regarding schooling and other choices. MayIJune 1972, 80(3, Part 2), pp. S252-55. A treatise on the family. Cambridge, MA: Harvard U. Press, 1981. ---. "Family Economics and Macro Behavior," Amer. Econ. Rev., Mar. 1988, 78(1),pp. 1-13. "On the Economics of the Family: Reply to a Skeptic," Amer. Econ. Rev., June 1989, 79(3),pp 514-18. BECKER,GARYS. AND TOMES,NIGEL."An Equilibrium Theory of the Distribution of Income and ~nter~eneiational Mobility," J. Polit. Econ., Dec. 1979, 87(6),pp. 1153-89. -. "Human Capital and the Rise and Fall of Families," J. Lab. Econ., July 1986, 4(3, Part 2), Major gains in our understanding of the determinants of children's attain- DD. S1-39. BE~RMAN, JERER. ET AL. Socioeconomic success: ment could be obtained from relatively A study of the effects of genetic endowments, family environment, and schooling. Amsterdam and New York: North-Holland, 1980. BEHRMAN, JERE R. ET AL. "A Sequential Model of Educational Investment: How Family Back- round Affects High School Achievement, Col- kege Enrollments, and Choice of College Qual- ity." Department of Economics, Williams College, Williamstown, MA, 1994. BEHRMAN,JERE R.; POLLAK, ROBERTAND TAUBMAN, PAUL. From parent to child: Intrahouse- hold allocations and intergenerational relations in the United States. Chicago: U. of Chicago Press, 1995. BEHRMAN,JERE R.; ROSENZWEIG,MARKAND TAUBMAN, PAUL. "Endowments and the Alloca- tion of Schooling in the Family and in the Mar- riage Market," Polit. Econ., '1994, 102(6), pp. 1131-74. BEHRMAN,JERE R. AND TAUBMAN, PAUL. "The Intergenerational Correlation between Children's Adult Earnings and Their Parents' Income: Results from the Michigan Panel Study of Income Dynamics," Rev, Income Wealth, June 1990,36, pp. 115-27. BELLER, ANDREA H. AND GRAHAM,OHN W. Small change: The economics of chi1 d support. New Haven, CT: Yale U. Press, 1993. BEN-PORATH,YORAM."The F-Connection: Fami- lies, Friends, and Firms and the Organization of Exchange," Population Devel. Rev., Mar. 1980, 6(1), pp 1-30. BLAU, PETER MICHAEL AND DUNCAN, OTIS DUDLEY.The American occupational structure. New York: Wiley, 1967. BOUND, JOHN; GRILICHES, ZVI AND HALL, BRONWYN. "Wages, Schooling and IQ of Broth- ers and Sisters: Do the Family Factors Differ?" lot Econ Rev., Feb. 1986,27(1), p~. 77-105.. BOWLES, SAMUEL. "Schooling an Inequality from Generation to Generation," J. Polit. Econ., MayIJune 1972, 80(3, Part 2), pp. S219-51. BRONFENBRENNER,URIE. "Ecological Systems Theory," Annals of Child Development, 1989, 6, pp. 187-249. BROOKS-GUNN. ET AL. TEANNE "Do Neiehbor- hoods 1nfluen;e Child and Adolescent ~gelop- ment?" Amer. J. Sociology, Sept. 1993, 99(2), pp. 353-95. BROWNING, MARTIN. "Children and Household Economic Behavior," J. Econ. Lit., Sept. 1992, 30(3), pp 1434-75. BUMPASS,LARRY AND MCLANAHAN, SARA. "Un- married Motherhood: Recent Trends, Composi- tion, and Black-White Differences," Demography, May 1989, 26(2), pp. 279-86. BURON, LAWRENCE. "A Study of the Magnitude and Determinants of Intergenerational Earnings Mobility." Ph.D, dissertation, Department of Economics, U. of Wisconsin-Madison, 1994. CAIN, GLEN G. "Rev. Socioeconomic Background and Achievement, by Duncan, Featherman, and Duncan," Amer. 1.' Sociology, 1974, 79, pp. 1497-1509. CASE, ANNE AND KATZ, LAWRENCE. "The Com- pany You Keep: The Effects of Family and Neighborhood on Disadvantaged Youths." Na- tional Bureau of Economic Research, Working Paper 3705,1991. CHILDREN'SDEFENSE FUND. Wastin America's future: The childrenS defense fun% report on the costs of child poverty. Boston: Beacon Press, 1994. CLARK, REBECCA. "Nei hborhood Effects on Dropping Out of Schoo? amon Teenage Boys: No Evidence of an Epidemic Effect." Urban In- stitute Research Paper, Washington, DC, 1992. COLEMAN,JAMES S. "Social Capital in the Crea- tion of Human Capital," Amer. J. Sociology, Supplement 1988,94, pp. S95-120. CORCORAN, MARY AND ADAMS, TERRY. "Race, Poverty, Welfare and Neighborhood Influences on Men's Economic Outcomes." Institute for Social Research, U. of Michigan, Ann Arbor, 1993. CORCORAN,MARYET AL. "The Association Be- tween Men's Economic Status and Their Famil and Community Origins,'' J. Human Res., Fag 1992, 27(4), pp. 575-601. CRANE, JONATHAN. "The E idemic Theory of Ghettos and Neighborhoof Effects on Dropping Out and Teenage Childbearing," Amer. J. Sociology, Mar. 1991, 96(5),pp. 1226-59. DATCHER, LINDA. "Effects of Community and Family Background on Achievement," Rev. Econ. Statist., Feb. 1982, 64(1),fp. 32-41. DUNCAN, GREG J. "Families an Ne~ghbors as Sources of Disadvantage in the Schooling Deci- sions of White and Black Adolescents," Amer. J. Education, 1994,103(1), pp. 20-53. DUNCAN, GREG J.; HILL, MARTHA AND HOFFMAN, SAUL. "Welfare Dependence Within and Across Generations," Science, 1988, 239(4839), pp 467-71. DUNCAN,GREG J. AND HOFFMAN, SAUL. "Wel- fare Benefits, Economic Opportunities, and Out-of-Wedlock Births among Black Teenage Girls," Demography, Nov. 1990, 27(4), pp. 519-35. DUNCAN,OTIS DUDLEY. "A Socioeconomic Index for All Occupations," in Occupations and social status. Ed.: A. J. REISS ET AL. New York: Free Press, 1961, pp. 109-38. DUNCAN, OTIS DUDLEY AND HODGE, RALPH W. "Education and Occupational Mobility: A Re- gression Analysis," Amer. ]. Sociology, May 1963, 68(6), pp. 629-49. ELDER, GLEN H., JR. Children of the Great De- pression. Chicago: U. of Chicago Press, 1974. ESPENSHADE, THOMAS J. lnvestin in children: New estimates of parental erpen$itures Washington, DC: Urban Institute Press, 1984. EVANS, WILLIAM N.; OATES, WALLACE E. AND SCHWAB, ROBERT M. "Measuring Peer Group Effects: A Study of Teenage Behavior," J. Polit. Econ., Oct. 1992,100(5), pp. 966-91. FEATHERMAN, DAVID F. "Opportunities Are Ex- panding," Society, 1987, 16, pp. 4-11. FEATHERMAN,DAVID L. AND HAUSER,ROBERT M. Opportunity and change. New York: Aca- demic Press, 1978. FERGUSON,RONALD. "Paying for Public Educa- tion: New Evidence on How and Why Money Matters," Harvard Journal on Legislation, Summer 1991,28, pp. 465-98. GOLDBERGER,ARTHUR S. "Economic and Me- chanical Models of Intergenerational Transmis- sion," Amer. Econ. Rev., June 1989, 79(3), pp. 504-13. GOTTSCHALK,PETER. "The Intergenerational Transmission of Welfare Participation: Facts and Possible Causes," J. Policy Analysis and Management, 1992a, 11, p .254-72. "IS Intergeneration$ Correlation in Wel- fare Participation across Generations Spurious?" Paper presented at the annual meeting of the Association for Policy Analysis and Manage- ment, Oct. 1992b. --. "Is the Correlation in Welfare Participa- tion across Generations S urious?" Department K of Economics, Boston Co ege, 1994. GRAHAM,JOHN; BELLER, ANDREA H. AND HERNANDEZ, PEDRO. "The Effects of Child Sup- port on Educational Attainment," in Child sup- port and child well-being. Eds.: IRWIN GARFINKEL,SARA MCLANAHAN, AND PHILIP ROBINS. Washington, DC: Urban Institute Press, 1994, pp. 317-54. GRILICHES, ZVI. "Sibling Models and Data in Economics: Beginnings of a Survey," J. Polit. Econ., Oct. 1979, 87(5, Part 2), pp. S37-64. HANUSHEK,ERIC A. "The Economics of School- in Production and Efficiency in Public Scoos 1. Econ Lit., Sept. 1986, 24(3), pp. 1141-77. "The Trade-Off Between Child Quantity and Quality," J. Polit. Econ., Feb. 1992, 100(1), pp. 84-117. HANUSHEK, ERIC A. ET AL. Making schools work: Improving performance and controlling costs. Washington, DC: Brookings Institution, 1994. Harvard Educational Rev. "Perspectives on Inequality." Harvard Educational Review Reprint Series no. 8. Cambridge, MA, 1973. HAUSER,ROBERT M. AND FEATHERMAN,DAVID L. The process of stratqication: Trends and analyses. New York: Academic Press, 1977. HAUSER,ROBERT M. AND SEWELL, WILLIAM H. "Family Effects in Simple Models of Education, Occupational Status, and Earnings: Findings from the Wisconsin and Kalamazoo Studies,'y. Lab. Econ., July 1986, 4(3, Part 2), pp. S83- 115. HAVEMAN,ROBERT H. Poverty policy and pov- erty research. Madison: U, of Wisconsin Press, 1987. HAVEMAN,ROBERTAND WOLFE, BARBARA. Succeeding generations: On the effect of inuestments in children. New York: Russell Sage Foundation, 1994. HAVEMAN,ROBERTET AL. "The 'Window Prob- lem' in Studies of Children's Attainments: A Methodological Exploration." Mimeo. Institute for Research on Poverty, U, of Wisconsin- Madison, 1995. HAVEMAN, ROBERT; WOLFE, BARBARA AND SPAULDING,JAMES. "Childhood Events and Circumstances Influencing Hieh School Com- pletion," Demography, ~ib.k91, 28(1), pp. 133-57. HAYWARD,MARK D.; GRADY, WILLIAMR. AND BILLY, JOHN 0. G. "The Influence of Socioeco- nomic Status on Adolescent Pregnancy," Soc. Sci. Quart., Dec. 1992, 73(4), pp. 750-72. HETHERINGTON, KATHLEEN E. MAVIS; CAMARA, A. AND FEATHERMAN, DAVID L. "Achievement and Intellectual Functioning of Children in One-Parent Households," in Achievement and achievement motives: Psychological and socio- logical approaches. Ed.: JANET T. SPENCE. San Francisco: W. H. Freeman, 1983, pp. 208-84. HILL, MARTHA AND DUNCAN,GREG J. "Parental Family Income and the Socioeconomic Attain- ment of Children," Soc. Sci. Res., 1987, 16(1), pp. 39-73. HILL. MARTHA AND PONZA. MICHAEL. "Does ~ilfare Dependence ~e~et In- D~e~de~cy?" stitute for Social Research, U. o Michigan, Ann Arbor, 1984. HOGAN, DENNIS P. AND KITAGAWA,EVELYN M. "The Impact of Social Status, Family Structure, and Neighborhood on the Fertility of Black Adolescents," Amer. J. Sociology, Jan. 1985, 90(4), pp 825-55. HORNEY,MARY JEAN AND MCELROY,MARJORIE B. "The Household Allocation Problem: Em- pirical Results from a Bargaining Model," in Research in population economics. Ed.: T. PAUL SCHULTZ. Greenwich, CT: JAI Press, 1988, pp. 15-38. INHELDER, BAERBEL AND PIAGET, JEAN. The growth of logical thinkin from childhood to 5 adolescence. London: Rout edge & Kegan Paul, 1958. JENCKS,CHRISTOPHERET AL.Inequalit A reas- sessment of the effect of family and sc&oling in America. New York: Basic Books, 1972. JENCKS,CHRISTOPHERAND MAYER,SUSAN."The Social Conse uences of Growing up in a Poor ~ei~hborhoo!," in Inner-city poverty in the United States. Eds.: LAURENCE LYNN, JR., AND MICHAEL G. H. MCGEARY. Washington, DC: National Academy Press, 1990, pp. 111-86. KIKER, B. F. AND CONDON, CAROL M. "The In- fluence of Socioeconomic Background on the Earnings of Young Men," J. Human Res., Winter 1981, 16(1), pp. 94-105. KREIN, SHEILA F. AND BELLER, ANDREA H. "Educational Attainment of Children from Sin- gle-Parent Families: Differences by Exposure, Gender, and Race," Demography, May 1988, 25(2), pp 221-34. LEIBOWITZ,ARLEEN. "Home Investments in Chil- dren," J. Polit. Econ., Mar./Apr. 1974, 82(2,II), pp. Slll-31. LI, JIANG HONG AND WOJTKIEWICZ,ROGER A. "A New Look at the Effects of Family Structure on Status Attainment," Soc. Sci. Quart., Sept. 1992, 73(3), pp. 581-95. LUNDBERG,SHELLYAND PLOTNICK, ROBERT. "Effects of State Welfare, Abortion and Family Planning Policies on Premarital Childbearing among White Adolescents," Family Planning Perspectives, Nov./Dec. 1990, 22(6), pp. 246- 51, 275. "Adolescent Premarital Childbearing: Do Economic Incentives Matter?" J. Lab. Econ., forthcoming. MACAULEY, JACQUELINE. "Stereotyping Child Welfare," Society, 1977, 13, p 47-51. MANSKI, CHARLES F "Achescent Econometricians: How Do Youth Infer the Returns to Schoolin 2" in Studies of supply and demand in higher egcation Eds.: CHARLES T CLOTFELTER AND MICHAEL ROTHSCHILD. Chicago: U. of Chica o Press, 1993a, pp. 43-57. "Irkentification of Endogenous Social Ef- fects: The Reflection Problem," Rev. Econ. Stud., July 199313, 60(3), pp. 531-42. MANSKI, CHARLES F. ET AL. "Alternative Esti- mates of the Effect of Family Structure during Adolescence on High School Graduation," J. Amer. Statist. Assoc., Mar. 1992, 87(417), pp. 25-37. MARE, ROBERT D. "Social Background and School Continuation Decisions," J. Amer. Statist. As- soc., June 1980, 75(370), p 295-305. MAYER, SUSAN E "The Effe~ of Schools' Racial and Socioeconomic Mix on High School Stu- dents' Chances of Dropping Out." Harris Graduate School of Public Policy, U. of Chi- cago, 1991. "Measuring Income, Employment, and the Support of Children." Paper prepared for the Conference on "Indicators of Children's Well-being," Bethesda, MD, Nov. 17-18, 1994. MCCUBBIN,HAMILTONI. ET AL. "Family Stress, Coping, and Social Support: A Decade Review," J. Marriage and the Family, 1980, 42(4), pp. 855-71. MCELROY,MARJORIEB. "The Empirical Content of Nash-Bargained Household Behavior,"]. Human Res., Fall 1990,25(4), pp. 559-83. MCLANAHAN,SARA. "Family Structure and De- endency: Early Transitions to Female House- [old Headship,'' Demography, Feb. 1988, 25(1), pp 1-16. MCLANAHAN, SARA AND SANDEFUR, GARY. Growin up with a single parent: What hurts, what hjPs. Cambridge, MA: Harvard U Press, 1994. MINCER,JACOB.Schooling, experience, and earn- ings. New York: Columbia U. Press, 1974. MOORE, KRISTIN A. AND CALDWELL,STEVEN B. "The Effect of Government Policies on Out-of- Wedlock Sex and Pregnancy," Family Planning Perspectives, July/Aug. 1977, 9(4), pp. 164-69. PLOTNICK,ROBERT. "Welfare and Out-of-Wed- lock Childbearing: Evidence from the 1980s," J. Marriage and the Family, 1990, 53(3), pp. 735- 46. "The Effects of Attitudes on Teenage Premarital Pregnancy and Its Resolution," Amer. Sociological Rev., Dec. 1992, 57(6), pp. 800-11. POLLAK, ROBERT A. "A Transaction Cost Approach to Families and Households," J. Econ. Lit., June 1985,23(2), pp. 581-608. RAINWATER, LEE. "Class, Culture, Poverty and Welfare." Center for Human Resources, Heller Graduate School, Brandeis U., 1987. REIN, MARTIN AND RAINWATER, LEE. "Patterns of Welfare Use," Social Service Rev., Dec. 1978, 52(4), pp. 511-34. RIBAR, DAVID C. "A Multinomial Logit Analysis of Teenage Fertility and High School Completion," Economics of Education Rev., June 1993, 12(2), pp 153-64. SANDEFUR,GARY D.; MCLANAHAN, SARAAND WOJTKIEWICZ, ROGER A. "The Effects of Pa- rental Marital Status during Adolescence on High School Graduation," Social Forces, Sept. 1992, 71(1), pp. 103-21. SCHULTZ, T. PAUL. "Testing the Neoclassical Model of Family Labor Supply and Fertility," J. Human Res., Fall 1990, 25(4), pp. 599-634. SELTZER, JUDITH A. "Consequences of Marital Dissolution for Children," Annual Reu. Sociol- ogy, 1994,20, pp. 235-66. SEWELL, WILLIAM H. AND HAUSER,ROBERT M. Education, occu ation, and earnings: Achieue- ment in the ear$ career New York: Academic Press, 1975. SOLON, GARY. "Intergenerational Income Mobil- ity in the United States," Amer. Econ. Reu., June 1992, 82(3), pp. 393-408. SOLON, GARY ET AL. "A Longitudinal Analysis of Sibling Correlations in Economic Status," J. Human Res., Summer 1991, 26(3), pp. 509- 34. THOMAS, DUNCAN. "Intra-Household Resource Allocation: An Inferential Approach," J. Human Res., Fall 1990, 25(4), pp. 635-64. U.S. BUREAUOF THE CENSUS.Statistical abstract of the United States: 1992. 112th Edition. Washington, DC: U.S. GPO, 1992. U.S. CONGRESS. COMMITTEE. JOINT ECONOMICS Economic report of the President. Washington, DC: U.S. GPO, 1993. U.S. OFFICE OF MANAGEMENTAND BUDGET, EXECUTIVE OFFICE OF THE PRESIDENT. Budget of the United States Government: Fiscal year 1994. Washington, DC: U.S. GPO, 1993. WU, LAWRENCE L. AND MARTINSON, BRIAN C. "Family Structure and the Risk of a Premarital Birth," Amer. Sociological Rev., Apr. 1993, 58(2), pp. 210-32. ZIMMERMAN,DAVID. "Regression Toward Medi- ocrity in Economic Stature," Amer. Econ. Rev., June 1992, 82(3), pp. 409-29. ZIMMERMAN,DAVIDAND LEVINE, PHILLIP. "The Intergenerational Correlation in AFDC Partici- pation: Welfare Tra or Poverty Trap?" Wellesley College WorRing Paper ~3-07. Nov. 1993.

Manski et al. (1992) NLSY: 2800 males and Outcome: 1985; High school Probit, bivariate females aged 14-17 in determinants: 1979 completion by probit, 1979 or at age 14 age 20 = 1 trivariate

tions, and values (e.g., educational expectations) are taken to affect directly the cognitive and social-psychological development of children. While the channils of transmission in this frame- work are auite different from those em-

1

phasized by economists, the implications of this perspective are consistent with the economist's with respect to the po- tential effects of ~arental education. la- bor supply, and flrtility choices on chil- dren's attainments.11

The Life-Span Development Approach.

This perspective (also referred to as the "life course" or "ecological systems" per- spective) guides much of the research in the developmental psychology literature (see Urie Bronfenbrenner 1989). It em- phasizes that development occurs over an individual's life, and that events im- pinging on a person have different ef- fects depending on when they occur, the length of time since the occurrence of the-event, the experiences that occur subsequently, and the historical context in which it occurs. The development pro- cess is viewed as one of continuous ad- justment and adaptation to exogenous forces. where the nature of the adiust- ment deDends on the transactions anb in-

I

teractions in which a person engages. For example, the divorce of parents may affect children quite differently depend-

l1The socialization framework emphasizes the transmission from parent to child of a pattern of behavior. A variant of the socialization model em- phasizes the importance of having two parents present in a family in order to foster normal per- sonalit development. The presence of two par-

?'

ing on the age of the child (which may indicate the existence of peers for sup- port), the nature of the separation (ac- commodating or antagonistic), the sub- sequent remarriage of one or both of the parents, or the time spent with the ab- sent parent (Seltzer 1994). This attention to the timing of events has supported at- tempts by empirical analysts to identify the different effects of events on children's attainments depending on when during childhood they occur.

Stress Theory and Coping Strategies.

This psychological perspective suggests that stressful events during childhood- as opposed to the persistence of adverse circumstances-may dislodge an individ- ual from an equilibrium path of develop- ment (see Glen Elder 1974 and Seltzer 1994). Such events (e.g., the separation or divorce, incarceration, or unemploy- ment of parents) are viewed as creating emotional uncertainties that impede nor- mal development. In a related model, the psychological resources of parents- their abilities to cope-are presumed to positively influence the performance and attainments of children (see Hamilton McCubbin et al. 1980). These resources are reflected in such parental traits as ambition, trust, and motivation that may serve as standards internalized by chil- dren. The ability to cope is often viewed as offsetting the negative effects of stressful events, such as family breakup, on children's success.

ents a so strengthens parental control and moni- toring, and weakens the otential adverse influence of other role mode\ (Judith Seltzer 1994). Another variant-one emphasizing neigh- borhood or peer group characteristics-stresses the attributes and circumstances of, and aspects of behavior present in, the neighborhoods in which children grow up (Christopher Jencks and Mayer 1990).Interaction of a child with peers and neigh- bors may result in adoption of their attitudes, aspi- rations, and behavior as norms; neighbors and peers become role models.

In this paper, we review and critique the empirical research on the links be- tween investments in children and chil- dren's attainments. The studies that we include in our review emphasize the po- tential effects on children of family (pa- rental) choices and neighborhood char- acteristics, the latter taken to reflect social choices.4 While our focus is on the economics literature, we include relevant studies from other social sciences.

In Section I1 we summarize the pri- mary theoretical perspectives that have guided research on the determinants of children's attainments; a more general and comprehensive economic perspective on this issue is presented in Section

111. In Section IV we discuss a number of empirical issues that pervade research in this area. Section V is the heart of our review; the children's outcomes that we emphasize include educational attainment, fertility choices (especially non- marital births during teenage years), and work-related outcomes such as earnings and welfare recipiency. Finally, we sum- marize the principal findings of this research and offer a critique of it.

11. Perspectives on the Determinants of Children's Attainments

The past quarter century has seen a growing body of social science research on the processes that explain why some children achieve success in young adult- hood while others do not. In most of this literature, "success" is typically measured by schooling attainments, occupa- tion or earnings (income) levels, and the choice of certain behaviors or life situ- ations (e.g., teen nonmarital motherhood

4We do not include in our review analyses of the effects of ublic schooling and child care in- vestments on ckildren's attainments. Reviews and assessments of research on these links include Eric Hanushek (1986), Jacob Mincer (1974), Ronald Ferguson (1991), and Hanushek et al. (1994).

or welfare recipiency). Most of the earli- est explorations of this issue were empirical analyses by sociologists; the con- tributions of economists have come later. Relative to the earlier work, the economic studies are distinguished by atten- tion to more formal models of this attain- ment process.

A. Economic Perspectives

From the outset, economists have viewed the process of children's attain- ment to be an aspect of the theory of family behavior.5 The family is viewed as a production unit which employs real in- puts in order to generate utility for its members. Adults in the family (typically parents) make decisions regarding the generation of family economic resources (e.g., labor supply); they also determine the uses (e.g., consumption, asset accumulation, or investment in children) of these resources. Parents make a variety of other choices such as fertility, loca- tion, and family stability that both influ- ence the returns to productive efforts and directly affect the well-being of members of the family.

position of their children by transferring

gifts or bequests to them.8

While Becker and Tomes' framework provides testable hypotheses regarding the effects of a few of the family-based determinants of investments in children (e.g., parental income and family size), it yields little empirical guidance beyond this (Arthur Goldberger and Becker 1989). Arleen Leibowitz (1974), building on this general framework, presents an economic model of the process of chil- dren's attainments with additional impli- cations for empirical work (Figure 1).In this model, the genetic endowments of parents (e.g., their abilities in a number of dimensions) are to some extent passed along to children via heredity. The abili- ties of parents and their educational choices jointly determine the level of family income and the quantity and qual- ity of both time and goods inputs (or "home investments") that parents devote to their children. Children's ability and the levels of parental income and home investments in time and goods determine the schooling attained by children and, through schooling, the level of post-school investment (e.g., work experience). All of these, in turn, affect chil- dren's earnings and income.9 Given their

lowing individual members of the family to be self-interested and to engage in bargaining among themselves for resources, including the allocation of parental investments in children's human ca i tal (see Yoram Ben-Porath 1980 and Robert Polik 1985). In general, the results of these extensions are consistent with those of the more basic model

the relationship between parental in::%:%% their investments in children.

8Another view of the literature investigates whether arents com ensate for differences in ge- netic enlowments ofPtheir children or instead re- inforce these differences and whether they do so consistently across human and nonhuman capital investments. See for example, Zvi Griliches (1979) and Jere Behrman, Mark Rosenzweig, and Paul Taubman (1994).

9An empirical analogue to this framework is <he following three-equation, recursive system. Children's Ability = fi(Genetic Factors, Home Investment)

abilities, then, parents make a wide vari- ety of decisions-including parental schooling, work effort, consumption, time allocation. and beauests-that are ex~ected to be relate2 to children's

L

schooling and labor market attainments.10

B. Related Perspectives from Other Disciplines

Sociologists and developmental psycholo~ists have contributed to the litera-

"

ture on children's attainments in wavs

1

that complement the work of economists. Here, we briefly describe the most important conceptual perspectives that have guided thinking in these fields.

The Socialization/Role Model Perspec- tive. This explanation stresses the poten- tially important effect of role models and socialization (adults or peers to whom children or adolescents relate and who set norms of behavior and achievement to which they aspire) during childhood and adolescent years on achievements as young adults. The primary role models are parents and older siblings, and their behavior (e.g., work, fertility), aspira-

Home Investment, Family Income)

Children's Income = f3 (Home Investment,

Children's Schooling, Post-School Investment,

Children's Ability, Family Income)

Children's Schooling = f2 (Children's Ability,

Comments
  • Recommend Us