What are the contexts that families come in?

Jeanne Brooks-Gunn, Brett Brown, Greg J. Duncan, and Kristin Anderson Moore

Introduction

The last decade has witnessed a remarkable transformation in social science data and research on child and adolescent development. Coming from quite different starting points, child development researchers, sociologists, and economists have converged in their needs for rich, multilevel data based on large samples. Beginning from an interest in socioeconomic attainment, sociologists and economists have produced a burgeoning literature on the factors that foster and undermine attainment; they now find themselves needing to delve deeper into the processes labeled socioeconomic status to understand how individual characteristics and family processes interact with community influences to produce socioeconomic attainment. Developmentalists have a tradition of conducting rich and detailed studies using small samples to examine in depth the processes by which children's characteristics interact with parental socialization practices during childhood. This approach has produced a voluminous literature that now seeks to test its theories and findings with data based on larger, more representative samples. This intersection of interests from the fields of child development, sociology, and economics places great demands on existing data systems (Brooks-Gunn et al., 1991; Duncan, 1991; Cherlin, 1991).

Despite an almost exclusive concentration on problem behaviors and a paucity of theoretical models that evaluate the full set of factors that influence children's development (Bronfenbrenner, 1979), the evolving literature is demonstrating that individual, family, neighborhood, and school variables all contribute to children's development (Brooks-Gunn et al., 1993a; Rosenbaum and Popkin, 1991; Alexander et al., 1993; Furstinberg et al., 1987; Rutter, 1985; Maccoby and Martin, 1983; Eccles, 1983; Sigel, 1985; Werner and Smith, 1982; Moore et al., 1994; Duncan et al., 1994a). However, findings on the relative importance of these domains in determining varied outcomes await both better data and further research.

Common to most of this research is the predominant use of large national datasets compiled by federal agencies or by survey organizations funded by federal agencies. Some of these datasets, such as High School and Beyond and the National Educational Longitudinal Surveys, were designed explicitly by the National Center for Educational Statistics (NCES) for analyses of adolescent outcomes and transitions to adulthood. Others, such as the Child Supplements to the National Longitudinal Surveys of Youth, are based on question modules added to datasets conceived primarily for other (in this case, labor market) purposes. Researchers working with still other datasets, such as decennial census microdata files with matched family-and neighborhood-level data, have been able to conduct valuable research on adolescent behavior by exploiting existing information that has been organized into a more useful form.

Stimulated by the increasingly widespread and complex nature of social problems involving children and adolescents (National Commission on Children, 1991; Hernandez, 1993), federal efforts to initiate or supplement data collection activities appear to be increasing. This is reflected in plans in the Survey of Income and Program Participation for a new supplemental module on family processes and developmental outcomes, NCES's initiation of a large cohort study of 5-year-olds, and consideration by the National Center for Health Statistics of a 1996 Child and Family Health Survey as part of the National Health Interview Survey.

In this paper we suggest specific national data collection projects that could improve research on child and adolescent development.1 Our explicit aim is to encourage continued expansion of both the outcome domains covered and the explanatory variables measured, to enhance the richness and quality of the data obtained, and to improve the representativeness of the samples that are drawn. These improvements would serve both the policy and academic research communities in their efforts to specify and estimate causal models of child, adolescent, and young adult behavior.

To this end, we begin with developmental theory and summarize key elements of an emerging "resource" framework, which we believe provides an integrative framework for understanding how child and adolescent development is affected by the time, money, and emotional resources of parents; by the institutions and "social capital" present in communities and neighborhoods; and by government policies that shape the context within which parental choices are made.

Next we explain and provide empirical examples of key elements of datasets that have proved especially useful in testing and drawing policy conclusions from the theoretical framework we advocate. Some of the elements we list consist of the outcomes, resources, and family processes identified by theory as important. Others are important methods for implementing and estimating child development models. Many of the illustrations are based on results from smaller-scale studies; all have implications for the data collection improvements we outline in this paper.

As we detail in the sections that follow, when assessing outcomes, a number of features are critical:

• High-quality, longitudinal assessments of child outcomes, obtained from the child as well as the parent and, when necessary, by trained professionals who test the child directly, and assessing how the factors that affect children influence development over periods of a decade or more;

• Measurement of age-appropriate outcomes and transitions; and

• Outcomes measured across multiple domains of functioning.

Also, as we discuss, measurement of resources needs to attend to the following:

• High-quality, longitudinal measurement of family resources, that is. obtaining measures of a broad range of economic and social resources periodically over the years when a child or adolescent is growing up:

• Measures of time "inputs," including the amount of time, the activities engaged in, and the persons present and interacting with a child:

• Measurement of family-process mediators, such as communication patterns. disciplinary style, and teaching style:

• Multiple levels of measurement, including the child, the family, the school, the community, the neighborhood, and the state;

• Measurement of school conditions, such as school organization and the socioeconomic composition of the school;

• Exact measurement of intrafamily relationships; and

• Measurement of extended-family relationships, including relationships with grandparents, aunts, and uncles.

Methodological and sampling considerations are also important, including:

• ''Natural experiment'' methods of model testing;

• Leverage for policy analyses provided by state-to-state variation in program benefits and structure;

• Oversamples of minority groups;

• Multiple informants; and

• Procedures to minimize and adjust for attrition in longitudinal surveys.

We next review the content of 12 existing national data collections in light of our list of desirable design features of developmental modeling, including:

(1)

Consumer Expenditure Surveys;

(2)

Decennial census;

(3)

High School and Beyond;

(4)

National Crime Victimization Survey;

(5)

National Educational Longitudinal Survey of 1988;

(6)

National Health Interview Survey (NHIS)—Child Health Supplement 1988;

(7)

National Longitudinal Survey of Youth (1979 Cohort);

(8)

NLSY Child-Mother Data;

(9)

National Survey of Families and Households;

(10)

Panel Study of Income Dynamics;

(11)

National Survey of Children; and

(12)

Survey of Income and Program Participation.

Table 1 lists salient characteristics of each survey, including measures of family resources and processes, measures of the extrafamilial context (e.g., neighborhood, school, peer group, county, and state), special advantages and disadvantages, sample size, and periodicity. Table 2 summarizes available child outcome measures for each survey by age group.

TABLE 1

Review of Federal Survey Contents and Characteristics.

TABLE 2

Review of Child Outcomes Covered in Federal Surveys.

Our theoretical discussion and empirical illustrations lead us directly to a set of suggested improvements, involving both incremental and more substantial investments in federal datasets that would enhance their value for research on child and adolescent development. In some cases, the suggestions involve minor and quite inexpensive changes that would produce large analytic benefits. In others, more expensive changes could open up invaluable analytic opportunities. We conclude with ideas for an even more expensive undertaking, a new longitudinal survey of children, outlining key design elements of such a survey.

Family and Community Resources and Children's Development Over Time

Many different frameworks have been used to study how children develop and the factors that influence development during childhood, adolescence, and the early stages of adult life. Such frameworks include: family systems approaches, risk and resilience, family and extra-family ecology, the life course, and economic decision making. Almost all of them consider. at least in passing, the ecology in which development occurs (context), as well as the stage or phase of life in which an individual is placed (time). However, these frameworks differ markedly in their relative emphases on time and context. Moreover, they also differ in how they examine the individual moving through time and context. Increasing interest in interdisciplinary research has focused attention on the value of different frameworks as well as the importance of looking at multiple mechanisms underlying development in any one study (Brooks-Gunn, in press; Brooks-Gunn et al., 1991; Duncan, 1991; Cherlin, 1991).

A number of investigative teams now combine scholars of macro issues (economists, sociologists, and demographers) and scholars concerned with more micro-oriented issues (developmental and clinical psychologists, pediatricians). Examples of such endeavors include Cherlin et al. (1991), Duncan et al. (1994a), Baydar and Brooks-Gunn (1991), Baydar et al. (1993), and Desai et al. (1989). The National Institute of Child Health and Human Development (NICHD) has recently initiated a Family and Child Well-Being Research Network, comprised of seven researchers and their colleagues; the seven teams in the NICHD network represent all six disciplines mentioned above.

Discipline-focused perspectives can be integrated into a framework based on familial and extrafamilial resources. The model we employ borrows heavily from the work of Coleman on social capital theory (1988) as well as the recent work by Haveman and Wolfe on choice-investment theory (1994). However, it departs from these two efforts in making more explicit the links with disciplines that focus on familial and extrafamilial processes, e.g., systems theory, ecological theory, and psychological-resource or social support theory.

Resources

Like Haveman and Wolfe (1994), we view "resources" very broadly, defining them as consisting of the money, time, interpersonal connections, and institutions that parents and communities may use to promote the development of children. Resources actually spent on promoting child and adolescent development are considered "investments" since, independent of whether they add to a child's well-being immediately, time and money are expended with the intent of enhancing the future health, cognitive ability, and productive social behavior of children.

Resources with which investments in children are made take many forms and are derived from the various systems or, to use Bronfenbrenner's term, contexts in which the child develops. We classify these as family, kin, peer, school, neighborhood, community, and larger societal systems. Societal systems include government policies that provide (or deplete) resources for children in general or for particular subgroups of children.

Decisions about resource investments are made on an almost continual basis by parents and, in adolescence, by the children themselves, in the context of changing circumstances and opportunities. Communities and the national government make decisions about institutional investments less frequently, although investments are ongoing at these levels as well.

Constraints on Choices

Choices are always constrained. No parent can expend more than 24 hours in a day for work, parenting, sleep, and leisure-time activities. A poor mother faces a difficult choice if she wants to move to a better neighborhood or school district, in that her limited income must be spent for food, clothing, and other basic needs as well as housing. A mother who is physically ill or has a mental health problem may not have the energy to invest a great deal of time in providing her child with stimulating experiences or acting in a warm and responsive manner. Parents with two or more children must divide time, money, cognitive investments, and emotional resources among the children. Not only do parents with different levels of economic and psychological resources distribute them differently, but also individual parents with the same set of constraints may also make different choices (e.g., how much time to spend with a given child).

Analogous limitations constrain choices regarding the provision of extrafamilial resources. Voters and administrators make choices about how much to spend on schools, how much to invest in programs for disadvantaged schools, how to organize and staff schools, and how much to spend on extracurricular activities. All of these influence the institutional resources invested in a given child. These constraints may influence children directly (via a school's facilities and climate for learning) as well as indirectly (via the family's investment in the schooling process itself).

Types of Family Resources and Interactions among Resources

It is useful to concentrate on four general kinds of resources: financial, time, psychological, and human capital. Economic models have looked at the financial and time resources that are made available to the children in a family (Hill and Stafford, 1980; Lazear and Michael, 1987) and consider the human capital (e.g., schooling level) of the parent as an indicator of the likely ''quality'' of the parent-child interaction time. Much more is known about income than about time-use. Time-use diaries have been extremely useful in describing the activities on which parents spend time and how much time is spent in child-oriented activities (Timmer et al., 1985). Despite this work, very little is known about how income and time are distributed across children within individual families—that is, how much is allocated to various household members or the process by which trade-offs between income and time are made.

Human capital includes the parents' levels of formal schooling, together with special skills, training, and other characteristics that affect financial or "psychic" income. Psychological resources at the family-system level include characteristics of the parents as well as parenting behavior. Relevant characteristics include the mental and physical health of the parents, the quality of their relationship, the psychological importance to them of factors such as education and work, and beliefs about the parental role in childrearing. Parenting behavior includes a wide range of behaviors directed toward the child as well as interactions between parent and child. Some of the most important include provision of learning and stimulating experiences, communication and decision-making styles, warmth directed toward the child, disciplinary practices, monitoring and supervision, and engagement. All of these have been shown empirically to be associated with child well-being (Maccoby and Martin, 1983; Bornstein, 1995; Holmbeck et al., 1995).

Parents vary as to their ability to provide these resources (i.e., a mother with little education may not provide many learning experiences because she herself is unable to read, a parent with little money may not be able to purchase books). Parents also make decisions that make it more or less likely that such psychological resources will be available to the child. For example, the barely literate mother might be able to enter a literacy program or be part of a family resource program that could provide such services. The poor mother may not be able to buy books, but, if the neighborhood or school has a library outreach program, she would be able to bring books into the home. Such parental decisions, however, are constrained by factors such as time availability and the resources available in the neighborhood.

Parents face choices about allocating their limited resources. Residence in a single-parent household means less parental time is available to the child (Hill and Stafford, 1980; Nock and Kingston, 1988). Residence in a stepparent household also results in less time spent in parental interaction than would be expected given the presence of two parent figures (McLanahan et al., 1991; Thomson and McLanahan, et al., 1993: Hetherington, 1993). Other adults in the household may provide the child with more time with a parental figure. For example, in multigenerational households, the grandmother or grandmother figure often functions as a coparent with regard to responsibilities and time spent with the child. Coresidence in multigenerational households presumably would offer children more resources (see Furstenberg et al., 1987; Furstenberg, 1976; Kellam et al., 1982), except in cases in which the grandparents cannot help with child care or, if ill, might require care themselves (Chase-Lansdale et al., 1994).

If both parents work, or if a single parent works, time is severely constrained. However, one would expect the addition of extra income to compensate in part for the time constraint. With greater income, from example, parents are able to purchase better-quality child care services. For families at the low end of the income spectrum, however, the loss of time due to working is probably not offset by high-quality child care, in that child care choices are constrained by income if relatives are not available (Hofferth and Phillips, 1991).

Trade-offs also involve social capital and psychological resources. If mothers who work find juggling work and parenthood stressful or unsatisfying or too much of a time drain, they may put less effort into providing stimulating experiences for their children or may exhibit less warmth toward their children (Wilson et al., 1995; Weinraub and Wolf, 1983; Lerner and Galambos, in press; Zaslow et al., forthcoming). If a mother who is residing with her own mother is not working (or bringing money into the household), conflicts may arise over the roles of both generations in the care of the children, which may be translated into less warmth or less provision of learning experiences by either the mother or the grandmother (Brooks-Gunn and Chase-Landsdale, 1995; Chase-Lansdale et al., 1994).

Types of Extrafamilial Resources

Important time and money inputs also come from institutions (e.g., schools, youth centers) outside the family. In addition, social capital has recently been conceptualized as an important potential resource (Coleman, 1988). Extrafamilial social capital consists of the interpersonal connections that families establish with people outside their immediate families. Time (e.g., helping others, volunteer work) and money invested by the family in these connections build a stock of resources that the family can call on when necessary. A neighborhood rich in connectedness among families and individuals and with high expectations for its children has a level of trust and stability that could prove extremely beneficial to children.

A study by Garbarino (1991) is illustrative. He looked at various neighborhoods in terms of child abuse and neglect rates, fitting a regression line for the rates of abuse/neglect and neighborhood income. Of special interest were neighborhoods with negative residuals—that is, neighborhoods that would be expected to have higher child abuse rates than was actually the case. These neighborhoods were characterized by stability, supportiveness, and trust—in other words, they appeared to have high levels of social capital. Overall, however, we know relatively little about how connections to the community promote child well-being.

Development and Resources

Most resource models do not take into account the age of the child or adolescent. The volume and type of resources may be more highly associated with well-being at some ages than others. For example, the spacing of births is more linked to academic outcomes during the preschool years (i.e., school readiness) than during the adolescent years (Furstenberg et al., 1987). Having many preschool children is probably detrimental because young children benefit from high rates of parental interaction; time spent with parents decreases dramatically during late childhood and even more in adolescence (Feldman and Elliott, 1990). Consequently, having many children, all of whom are young, takes away parents' ability to spend time with each of them.

Moreover, resource models do not take into account individual differences in children. An illustration may be taken from the work on low-birthweight children. Very low-birthweight children (1,500 grams or less at birth) often have difficulties in regulating moods and states. Parents who have little psychological or human capital may not be as responsive to the special needs of their biologically vulnerable child. They may not avail themselves of community services to provide themselves with better parenting skills or the child with remedial intervention. Biologically vulnerable children may be more affected by familial or extrafamilial resources than children without biological problems. Similar arguments may be made with respect to children who have emotional vulnerabilities (i.e., the child who is temperamentally difficult, shy, or active) or cognitive vulnerabilities (i.e., the child with mild mental retardation or developmental difficulties).

Current resource theories also do not explicitly consider the intersection of familial and extrafamilial resources. That is, certain families may benefit more from resource-rich neighborhoods than others (Klebanov et al., 1994; Duncan et al., 1994a). Consequently, we wish to extend our resource-based framework to include age-sensitive measures, individual differences, and context. These additions provide links between resource frameworks and existing life-course and ecological (contextual) frameworks.

Crucial Design Features of Developmental Datasets

The breadth of the constructs included in the various theoretical perspectives that are incorporated into a resource framework requires a range of data to test these theories. In this section we identify key elements of datasets that have proved especially useful in testing and drawing policy conclusions from the theoretical framework we advocate. We begin with examples of measures of developmental outcomes, resources, and family processes identified by theory as important. We turn next to methodological issues, producing illustrations of survey design features that have proved invaluable in implementing and estimating the models. Many of the illustrations are based on results from smaller-scale studies; all have implications for the data collection improvements we outline in this paper.

High-Quality, Longitudinal Measurement of Child Outcomes

In measuring child outcomes, the goal is to identify objective indicators of competence or well-being. No matter how objective a parent is about the characteristics of his or her child, it is difficult for anyone to judge accurately the achievements of the child across a range of behaviors. Even in an instance in which parents are asked to describe simply whether a child has achieved a particular developmental milestone, variations in parental definitions can affect outcome rankings. For example, in responding to the Social and Motor Development Scale in the 1981 Child Health Supplement of the National Health Interview Survey, mothers with graduate-level educations were more critical of their young children. And teachers in a school attended by children of highly educated parents may be more critical of an average child than a teacher would be who saw the same child in the context of a community with a lower educational level. For these reasons, it is desirable to use standardized and nationally normed measures whenever possible.

Longitudinal assessment of outcomes is desirable since how changes in resources influence children's circumstances is a question for both theory and policy. Theoretically, issues center on how malleable development is, and what boundaries constrain development (Brim and Kagan, 1980; Hunt, 1961; Lerner, 1984). Of policy relevance is also whether changes in maternal circumstances, such as education, marital status, employment, and welfare receipt, have the potential for altering children's development (Wilson et al., 1995; Brooks-Gunn, in press). Programs targeting at-risk, often poor families are typically based on the assumption that altering family circumstances will benefit children as well as parents (Huston, 1992; Chase-Lansdale et al., 1994; Palmer et al., 1988). However, without outcome data on children prior to and following a change in family resources (such as maternal movement from welfare to work, maternal completion of schooling, family enrollment in a supplemental food program) or a change in community resources (such as entrance into a Head Start program or another early education program, use of a neighborhood family resource program), it is impossible to determine whether a child outcome is due to the resource change or to unmeasured differences between families who did and did not change their circumstances or receive a community resource.2 Thus, direct, longitudinal assessment of child outcomes, using nationally normed tests, represents an important component of any large-scale data collection effort.

Measurement of Age-Appropriate Outcomes

It is essential to assess child outcomes in ways that are optimal for the age and developmental stage of the child. Standardized cognitive and school achievement tests take into account the rapid changes that occur throughout the childhood and adolescent years. Even then, some tests are not appropriate over the entire first two decades of life. Infant intelligence tests are constructed quite differently from childhood tests, in part because of the limited language abilities of infants and toddlers (Brooks-Gunn and Weinraub, 1983). Intelligence tests like the Weschler series have different versions for preschool children and for older children and adolescents (WPPSI and WISC).

Instruments used in national surveys to assess cognitive ability (e.g., PPVT-R; Dunn et al., 1981) focus on one aspect of language competence—receptive language or vocabulary. Although appropriate for children from age 3 through adulthood, they may be of limited use at the younger ages, especially for children who have not had any experience with situations such as naming and pointing and for children who, because of adverse family circumstances or biological conditions, are not speaking at age 3. The Bracken Basic Concept Scale assesses school readiness and is also brief and easy to administer in a home setting, even for an interviewer untrained in psychometric testing.

School-related outcomes also are sensitive to developmental age or phase. Grade failure is not a particularly good question to ask about children in kindergarten and first grade to predict failure from earlier experiences, since school situations are often quite fluid in the early grades. Similarly, current grade repetition is not relevant for high school students, since so few are held back. Instead, dropping out of high school and school absence are better measures of academic problems during high school.

Age grading is also important for behavior and emotional problems. What is considered a problem at one age may not be at another. A good example is biting: many young children go through a phase of biting others. Consequently, behavior problem scales (parents or teachers are asked to rate the frequency or severity of a list of behaviors and feelings, typically on a 3-point scale) are usually standardized by age.

Measurement of Developmental Phases and Transitions

Ideally, data collection efforts assessing child outcomes and family resources should be designed to permit estimation of models during and through key developmental phases and transitions. Scholars of child and adolescent development differ as to how to identify a phase or a transition period, as well as on whether development is so continuous that it is meaningless to talk about phases or transitions. However, school transitions in our society mean at the very least a movement to a different organizational structure as well as a change in how family, peers, teachers, and neighborhoods relate to the child. These transitions include the transition from preschool to elementary school, the transition to middle (or junior high) school, the transition to high school, and the transition out of high school (into the work force, higher education, vocational training, or unemployment). We believe that much more work needs to focus on these transitional periods. It is clear that many children fare poorly at these transition points, which in turn undermines their life trajectories (Alexander et al., 1993; Natriello, 1987; Brooks-Gunn et al., 1993a; Simmons and Blyth, 1987). Other transitions are important to an understanding of children's development. The transition to parenthood is crucial because it sets the stage for subsequent parent-child relationships, family interchanges, and potential gender-role divisions in the care of offspring and residence with the offspring (Ruble et al., 1990; Deutsch et al., 1988; Belsky, 1984; Cowan and Cowan, 1990; Shereshefsky and Yarrow, 1973).

The experiences of the mother and her unborn child also influence the health and neurological competence of the newborn, with concerns about adequacy of prenatal care, prenatal drug use, and prenatal stress and support being paramount (Robins et al., 1993; Berendes et al., 1991; Institute of Medicine, 1985). The Child-Mother Supplement of the National Longitudinal Survey of Youth is one dataset that, by starting the study prior to childbearing of the original adolescent cohort, allows an opportunity to study this transition (Chase-Lansdale et al., 1991). Transitions that typically involve young adults (and in some cases adolescents) are not studied in great detail. We would include here the transition to a household separate from parents and parental figures and the transition into the work force (which does not always occur at the end of adolescence and high school).

Thus, in terms of national data collection efforts, questions should be tailored to at least five age groups—the four school groupings and the infancy period—and focus on transitions between these periods as well as transitions into adulthood and parenthood.

Multiple Domains of Outcomes

The important domains of child and adolescent outcomes include the cognitive, emotional, social, health, and school-related (Brooks-Gunn, 1990). Employment, earnings, and fertility also become important in the late teen and early adult years. Outcomes in these various domains are linked in different ways to family and community resources.

An example may be taken from the Baltimore Study of Teenage Motherhood, a 20-year longitudinal study of more than 300 young women who gave birth in Baltimore in the late 1960s, their mothers, and their children (Furstenberg et al., 1987; Brooks-Gunn et al., 1993b). Different kinds of family resources predicted school failure (grade failure by age 16 and high school dropout by age 19 to 20), early sexual experience (intercourse by age 16), and behavior problems (a scale score). Number of children in the family during the preschool years was associated with school failure but not early sexuality. And the presence of a father figure in the home during the early adolescent years was important in delaying the sexual debut but was not predictive of high school difficulties.

Another reason to include multiple domains is that, in many cases, children and youth who are not faring well in one domain are also having problems in another domain. Ongoing research is examining the ways in which adolescents' problems occur together and the sequencing of the emergence of problem behavior (Dryfoos, 1990; Jessor, 1992). Subgroups of adolescents exhibit clusters of risk-taking behaviors, and certain risk-taking behaviors show a somewhat orderly progression of acquisition (cigarette use in the elementary school being predictive of alcohol and marijuana use in the middle school being predictive of the use of other drugs in high school years; Yamaguchi and Kandel, 1984). Including a narrow range of out-comes in a study would not allow for an examination of the clustering and timing of problem behaviors. Thus surveys should consider measuring out-comes across different domains.

High-Quality, Longitudinal Measurement of Family Resources

Most theories of child and adolescent development view as important the overall level of family economic resources. The statistical relationship between available material resources and child well-being is well established in the literature (Kalmuss and Fenelly, 1990; U.S. Department of Health and Human Services, 1992; Duncan et al., 1994b; McLoyd, 1990; Shaw and Emery 1987; Dryfoos, 1990; Wilson, 1987; Hill and Duncan, 1987; Mare, 1980; Brooks-Gunn et al., 1993a).

Underappreciated in much of this literature, however, is that longitudinal studies find great temporal variability in family income, both within (Survey of Income and Program Participation) and across (Panel Study of Income Dynamics) years and both within and outside the United States (Duncan et al., 1993). One-quarter of U.S. families who are poor in one year are not poor the next (U.S. Bureau of the Census, 1989), and only about one-half of children who are poor in a given year are poor over longer periods (Duncan and Rodgers, 1991). What implications does this variability have for studies of child and adolescent development?

Duncan et al. (1994a) use data from the Infant Health and Development Study to relate patterns of short-and longer-term poverty over the first five years of life to IQ at age 5. After controlling for conventional demographic and socioeconomic measures (family structure, maternal education), they find powerful associations between IQs and family income levels. Of particular interest is the finding that the IQs of ever-poor children are affected by the persistence of their poverty. Net of numerous controls, children poor on all four occasions when family income was measured scored five IQ points lower than did children in families whose income histories showed only transitory poverty. This and other research suggests a crucial role for family economic resources in models of child development (see Duncan et al., 1994a).

It should be noted that detailed assessments of family income provide a great deal of additional information on the family's situation, including welfare receipt, labor market involvement, and income-generating assets (wealth).

Accordingly, we draw two conclusions. First, in addition to conventional measures of family socioeconomic status, high-quality measurement of family income is crucial for testing resource-based theories of child and adolescent development. And second, also important is longitudinal measurement of family income, enabling researchers to distinguish between temporary and persistently low levels of family economic resources.

Measurement of Time Inputs

Although given a prominent place in most developmental models, measures of the quantity and the quality of time spent by parents with their children are almost never available in the kinds of datasets we review. Maternal schooling level is typically used as a measure of the likely quality of the mother's time. Mother's supply of labor outside the home is usually measured and sometimes taken as a (problematic) indicator of time not available for parenting activities. Although several surveys ask questions regarding time spent in a variety of parent/child activities (e.g., television, homework, meals, special activities), there is virtually no recent, direct measurement of parental time inputs.

Careful measurement of such time inputs was achieved in a national time-use study conducted in the late 1970s and early 1980s (Juster and Stafford, 1985). Following methodological work suggesting that time diaries are the best method for obtaining unbiased measurement of time inputs, such diaries were obtained from a national sample of adults and children.

When coupled with teacher reports of academic achievement in elementary school, the coding scheme enabled Stafford (1987) to analyze the developmental consequences of parental time spent in direct learning activities, such as reading to children, as well as other parental activities at which children were present. Although it should be kept in mind that his results are based on small samples. Stafford found highly significant effects of direct activities on academic achievement.

Time diaries on children require roughly 15 minutes of interviewing time per recalled day and are thus a time-intensive method for collecting data on parental time inputs. Periodic comprehensive measurement of patterns of parental time-use are highly desirable. Some existing national surveys should consider adding shorter question sequences that would provide at least some data on parental time inputs.

Measurement of Family Process

Family processes or functioning are important to the health and development of children, both in their own right and as mediators of material resources and child outcomes. A recent review of the literature has identified a number of important categories of measures of family processes that affect child well-being, including communication (parent-child and parent-parent), parent-child time together and activities, degree of commitment to the family, degree of social connectedness, religious/spiritual orientation, capacity to adapt to new situations, and the existence of clear family roles (Krysan et al., 1990). In addition, family conflict and styles of parental discipline are often emphasized in this literature (Zill, 1983).

The statistical relationship between available material resources and child well-being is well established in the literature. Low income has been related to less adequate prenatal care (Kalmuss and Fenelly, 1990), low birthweight and higher infant mortality (U.S. Department of Health and Human Services, 1992), slower cognitive development (Duncan et al., 1994b; McLoyd, 1990; Shaw and Emery, 1987), higher rates of adolescent risk behaviors (Dryfoos, 1990), and lower levels of educational and socioeconomic attainment as adults (Wilson, 1987; Hill and Duncan, 1987; Mare, 1980).

Relatively little is known, however, about the processes internal to the family that can account for observed relationships between resources and outcomes.3 Our resource model indicates that family processes will to some extent mediate the relationship between available material resources and child outcomes (see, for example, Klebanov et al., 1994). However, it is often argued that observed relationships between material resources and child well-being are in fact the result of currently unmeasured or poorly measured family processes that correlate with material resources but are not determined by them (e.g., Murray, 1984). This is an important distinction, since the policy implications are vastly different depending on how one specifies these relationships. In either case, however, the proper measurement of family processes, and their inclusion in models relating material resources to child well-being are key to a deeper understanding of the links between material resources and child well-being.

Multiple Levels of Measurement

Although studies of community influences on human behavior have a long tradition (Park et al., 1967; Wirth, 1956; Shaw and McKay, 1942), only recently have researchers successfully combined individual, family, and neighborhood-level measures in national or multisite developmental datasets. Much of this work was inspired by William Wilson's (1987) theories of social isolation as inherent in neighborhoods with particularly high concentrations of poor people.

Among the notable empirical contributions to this recent literature are Crane's (1991) and Clark's (1993) respective analyses of specially linked family-tract cross-sectional files from the 1970 and 1980 census-based Public Use Micro-Sample (PUMS) files; Brooks-Gunn et al.'s (1993b) examination of neighborhood effects in the PSID and the Infant Health and Development project; and the Garner and Raudenbush (1991) study of data, analyzed with hierarchical linear models, on a sample of children in one education authority in Scotland in the mid-1980s coupled with areal data taken from the 1981 Census of Population. In the U.S. studies, neighborhood data are usually obtained from tract-level economic and demographic data from the decennial census and matched to families' addresses.

Most U.S.-based sources have found that, after controlling for family-level resources, the presence or absence of affluent neighbors is a more powerful predictor of adolescent well-being than is the presence of low-income neighbors—a finding that supports theoretical models of neighborhood effects based on beneficial neighborhood institutions (e.g., police protection, parks, schools) rather than the "epidemic" models based on peer-induced contagion effects.4 Given the early stage of this research, these findings should be viewed as tentative but intriguing, with much more still to be learned about the influences of neighborhood resources.

National household surveys should append neighborhood-based data to their data files. This task is often surprisingly easy, especially in the first wave of most national surveys. This is because the tract/Block Numbering Area (BNA) identifier of sampled addresses is routinely gathered as part of the sampling process.5 It is a simple and inexpensive matter to use these identifiers to merge neighborhood characteristics (e.g., poverty rate, extent of female headship, male joblessness, ethnic composition) from STF3 decennial census data files. The resulting merged family-neighborhood files offer a rich combination of interview-generated family data and census-generated neighborhood data that can be used as independent variables in various analyses of child development.

Given their geographic detail, such merged data raise confidentiality issues.6 In the case of census Bureau data collections, the use of these matched data could be limited to census analysts as well as outsiders serving as census fellows. As for data collections outside the federal government, the matched data could be released under special contractual conditions, such as those applicable to the PSID.

Whenever possible, appending information regarding other extrafamilial environments, most notably school characteristics, would also be valuable. NCES samples are generally drawn from schools, and school-level data are routinely made available in analysis files. Household-based surveys must incur more expense in gathering school-based data. When such information is drawn from a survey of the teachers of the children in these samples, the questionnaires (which can be set up to use less expensive self-enumeration) can be designed to provide information on both the achievements of the individual children and key aspects of the school environment.

There is also the possibility of asking families about key elements of their neighborhoods—danger, drugs, unemployment, etc. This would enable researchers to address the methodology question of the extent to which perceptions and tract-level variables match, as well as the more substantive issue of how perceptions and objective characteristics interact. For example, resilient families may live in bad neighborhoods and perceive them as bad, but they may then respond to these conditions with intensive efforts to monitor their children. Less resilient families may live in bad neighborhoods but not perceive them as such and therefore not engage in as much monitoring.

Measurement of School Contexts

Schools are an important extrafamilial environment for child and adolescent development. Information on school functioning can be collected from self-enumerated teacher questionnaires that are sent to the child's school after permission to contact the teacher (and the teacher's name and address) is obtained from the parent. Data that can be collected include: (1) characteristics of the child, such as school grades, school engagement, parental involvement, school absences, peer relationships/social competence, attention in the classroom, and classroom-related behavior problems and (2) school characteristics, such as social climate, ethnic mix of students (in a given classroom), and teacher experience/credentials.

In several national or multisite studies, response rates to such teacher surveys range between approximately 75 and 80 percent (e.g., National Survey of Children, Moore and Peterson, 1989; Infant Health and Development Program, Brooks-Gunn et al., 1993b; Study of Elementary School Outcomes of Low Birth Weight Children, McCormick et al., 1992). The data obtained from elementary school teachers and high school teachers are often different (e.g., high school teachers have much less contact with students) and teacher-based information on 5-to 7-year-olds is less useful than data on slightly older children, since many children in that age range have not experienced school-related problems or are experiencing transient problems that resolve themselves with emotional maturity or increased cognitive skills. By age 8 and third grade, however, school problems are predictive of subsequent school failure (Brooks-Gunn et al., 1993a; Baydar et al., 1993; Snow, 1983).

Measurement of Family Relationships

Increased attention should be placed on different household arrangements as well as changes in these arrangements. When the focus is on child well-being, national studies must document the relationship of various parents and parent figures to each child.

Most research has focused on marital or parental status, rather than on changes in households or on links between marital states and individual children (Brooks-Gunn, in press; Thomson and McLanahan, 1993). Distinctions of potential importance include: (1) stepparent families in which the father is the custodial parent; (2) stepfamilies in which the custodial mother has not been married prior to her marriage to the stepfather (often seen in the case of a never-married single mother, often a teenager, who marries later); (3) never-married single males who marry and become stepparents; (4) custodial parents who marry a third or a fourth time; (5) cohabiting adults with children; (6) biological parents who have separated or divorced and then reunited; and (7) cohabiting adults each with children from a previous marriage (Brooks-Gunn, in press).

It is most common to determine family relationships by selecting a head or reference person and then asking for the relationship between each family member and this reference person. This is not sufficient for many important purposes, since it often does not allow the researcher to distinguish important relationships among members of subfamilies. If, for example, there is a three-generation family with a grandmother, her two daughters, and the daughters' children, then the relationship ''granddaughter'' does not provide enough information to classify children as siblings or cousins. In fact, mothers and their children cannot always be connected. As explained in the next section, we view as crucial the collection of information about family relationships that identifies the natural and stepparents of all children living in the household.

Most work on family relationships has been cross-sectional in nature, not focusing on the effects of changes in these family arrangements (see, as exceptions using national longitudinal datasets in the United States and Great Britain, e.g., Baydar, 1988; Cherlin et al., 1991; Kiernan, 1992). Using a framework focusing on transitions dictates that data on family resources and child outcomes be collected over a long enough time period that sufficient numbers of family structure changes will occur (Hetherington and Arasteh, 1988; Hetherington and Clingempeel, 1992).7

It is important to note that transitions in family structure often cannot be inferred reliably in a longitudinal study from changes in reported marital states. A woman married in two consecutive waves may have divorced and remarried between the waves, and, even if an analyst checked for changes, measurement error in husband characteristics (e.g., age) may make it difficult to infer transitions without asking about it directly. Longitudinal studies should ask marital and fertility histories during each wave in a way that covers the entire time interval between waves.

Coupled with this work is a concern for the relationship of the child to the noncustodial parent as well as the resources provided to the child by the noncustodial parent (Maccoby and Mnookin, 1992; Garfinkel et al., 1995; Teachman, 1991). It would be valuable for at least some national surveys to collect information on noncustodial parents regarding economic, emotional, and time resources. Information on the level of child support payments awarded (if any), the level received, and the stability of receipt would be welcome additions to national data collection efforts as well (Garfinkel et al., 1995). Changes in compliance of noncustodial parents and award of payments as a result of the Family Support Act of 1988 allow for natural experiments, in that one can examine payment levels and contact with children prior to and following changes in the law, as well as state-to-state variations (see McLanahan et al., 1994, who estimated predicted child support for each state to model differences in child support enforcement policy, using data from the Current Population Survey Child Support Supplement).

Measurement of Extended-Family Relationships

There is great theoretical interest in the extent and effects of intergenerational relations, especially in the flows of time and money between children and their parents. Such flows can affect the resources available to children and the need for public resources.

We have already provided examples of instances in which young children and adolescents living with their parents receive time and some (usually undetermined) share of the family's total income. Once children enter the late teenage years, they face important decisions about postsecondary schooling, careers, and living arrangements. Transfers from parents, in the form of both time and money, are of great potential importance at this stage. When children continue to live with their parents, the bulk of transfers consist of in-kind services. Children's college attendance can be greatly facilitated if tuition payments are made by parents or grandparents. Choices about home ownership, jobs obtained through personal connections, and start-up capital for independent business endeavors can all be influenced by the financial and social capital of a child's extended family.

Although some of this information (e.g., whether any help was received with tuition payments) can perhaps be recalled reliably by the grown children, parental financial resources during the time in question cannot be addressed without reliable information on these resources gathered from the parents themselves. Thus it would be desirable for longitudinal studies of children making the transition to adulthood to continue to collect information from parents as well as children, regardless of whether the children are still living with their parents.

There is an added methodological benefit from such information, as illustrated by Gottschalk's (1992) analysis (based on the Panel Study of Income Dynamics) of the links between parental welfare receipt and welfare receipt by daughters as they enter their adult years. He attempts to purge the parental welfare receipt measure of its sources of noncausal correlations with daughter's receipt by using patterns of mother's welfare receipt after the daughter has left home to adjust for the effects of unobserved heterogeneity. His argument is that the future welfare use of the mothers cannot have caused prior decisions leading to possible welfare receipt on the part of daughters and therefore is a valid control variable for the heterogeneity. Only by following parents after children leave home is such an approach possible.

Natural Experiments

Alone among their disciplinary colleagues in their insistence on testing theories against real-world data, labor economists in the last decade have effected a remarkable transformation in their methodology. Structural models based on tenuous identifying restrictions and ever more sophisticated econometric modeling are being replaced by reliance on "natural experiments." We endorse this development and encourage other disciplines to consider (or, in some cases, continue) work along these lines.

Drawing on a long history of work from related disciplines, Currie and Thomas (forthcoming) use sibling data from the National Longitudinal Survey of Youth to compare the cognitive development of siblings, some of whom did and some of whom did not participate in the Head Start program. Because siblings share a similar family background (including material and social resources and aspects of genetic endowment ranging from cognitive ability to temperament and physical and mental health), a comparison of their cognitive differences constitutes a kind of natural experiment, since such differences will be largely free from the confounding effects of unmeasured family background. Geronimus and Korenman (1992) and Hoffman et al. (1993) follow a similar strategy with data from the National Longitudinal Survey of Youth, the original National Longitudinal Survey cohorts, and the Panel Study of Income Dynamics in estimating sisters' differences in adult economic and demographic status as a function of whether a given sister had borne a child during her teenage years.

More subtle natural experiments have been extracted from datasets that were never designed to be relevant for developmental research. Bronars and Grogger (1992) consider the birth of the second child in twin pairs as an exogenously induced increase in family size in estimating the effects of family size on a family's economic well-being. Only data released from the decennial census provide enough observations on twins; regrettably, Grogger and Bronars were hampered by the quality of the fertility information available in census files; still, the approach represents a creative use of secondary data to test theory.

Owing to the ingenuity of the researchers and the resulting idiosyncratic nature of the experiments, it is difficult to generalize about the literature that has made use of natural experiments. Many of these studies require very large samples, which suggests that child development studies should be added to the many important uses of data from the decennial census in discussions of future design changes.

Common to much of the developmental research based on natural experiments is the need for clear data on family relationships, especially in the identification of sibling pairs. As noted above, this task proves especially problematic in multigeneration households in which relationships are established between household members and the head or reference person in the household. As set out in more detail below, unambiguous data are needed showing the relationships among all possible pairs of individuals who reside in a given family. Also, Clean information on birth dates has proved valuable in much of this research.

Random-assignment experiments are even better than natural experiments in providing exogenous variation in right hand-side measures of interest. In 1988, Congress passed the Family Support Act, creating the JOBS program to assist families receiving Aid to Families with Dependent Children (AFDC) in obtaining basic education, job training, and job search services. Because many more AFDC recipients are eligible than are or can be currently served, mothers of preschoolers were assigned either to an experimental group mandated to participate in JOBS or to a control group not so mandated. The developmental effects of this program on the pre-school children of AFDC recipients are being evaluated by Child Trends, Inc., as a substudy of a much larger evaluation of the JOBS program being conducted by the Manpower Demonstration Research Corporation. It is essential that such government experiments with significant potential to affect children be evaluated in light of the possibility of providing experimental data on child development.

Leverage for Policy Analyses Provided by State-to-State Variation in Program Benefits and Structure

Another kind of natural experiment of special interest for policy research is rooted in the fact that states often differ from one another in the benefit levels and other key features of the social policy programs they administer. Payments to three-person families from the AFDC program ranged from $119 per month in Alabama to $688 in Alaska in fiscal 1991 (Committee on Ways and Means, 1993). Many studies have attempted to use this variation to infer what would happen to work effort or family structure if the overall level of benefits from programs such as AFDC were to be changed (Moffitt, 1992). For example, Moore et al. (1994) use state-level policy variables, such as the AFDC benefit level, state abortion policy, and the adequacy of each state's family planning coverage in an event-history analysis of the determinants of adolescents' age at first intercourse. Similarly, Garfinkel et al. (1994) examine the effect of state child support on the economic status of children with absent parents.

Common to all of these studies is the addition of a state-of-residence identifier to the household and individual-level data. Adding state identifiers as a routine matter to all national surveys would be invaluable. In cases in which sample sizes within smaller states are judged to be too small to preserve the confidentiality of responding families, we urge that the datasets identify as many states as possible.

Oversamples of Minority Groups

Knowledge of the development of minority children and youth in the United States is woefully inadequate (Spencer and Dornbusch, 1990; Earls, 1992; Spencer et al., 1985; McLoyd, 1990; Lamberty and Garcia Coll, 1994; Garcia Coll, 1990). Furthermore, studies of minority children and youth tend to focus on problem behavior, rather than on competence or normal development. This is true even though representative samples include minorities.

Generally, the national datasets reviewed in this paper include enough African-American children and youth for separate analyses. In part, this is due to the fact that African-Americans have been oversampled in many national studies (see Table 1). At the same time, current studies do not allow for a separation of black respondents who are not African Americans (such as Caribbean blacks). Most datasets do not include enough Hispanic-American children and youth, in part because oversampling has not always been done and, unless supplemented by a fresh sample, studies begun in the 1960s and 1970s are not representative of the Hispanic-American immigration patterns of the past 15 years. And only one of the surveys we review oversampled the Asian population (see Table 1).

Another issue complicates understanding of Hispanic-American children and families. Different groups of Hispanic-Americans are quite distinct. Cultural beliefs, place of origin, geographic location in the United States, and recency of immigration all differ for Cubans, mainland Puerto Ricans, Dominicans, and Mexicans, in addition to Hispanic-Americans from other Latin American countries. Groups also differ with respect to their status in the United States and in their country of origin. For example, Cuban-Americans are unable to travel back to Cuba, whereas mainland Puerto Ricans, as citizens, have no restrictions of movement. Other Hispanic-Americans have to contend with immigration restrictions. These differences alter social support systems in this country and the country of origin. In many research situations, it is unwise to treat Hispanic-Americans as a single group.

However, with the exception of the oversampling of Cuban-American and Puerto Ricans in the Hispanic sample of the Panel Study of Income Dynamics, none of the national datasets has enough Hispanic-Americans to separate out even the two largest groups: Puerto Ricans and Mexican-Americans. To be most useful, all new national datasets should oversample Hispanic-Americans as well as African-Americans. If resources permit, dataset designers should also consider oversampling Asian-Americans as well as Hispanic subgroups such as Puerto Ricans.

There are also arguments in favor of oversampling on risk factors such as low-income and high-poverty neighborhoods. One, based on experience with the Panel Study of Income Dynamics, which oversampled on the basis of low family income, is that such oversampling be done using relatively exogenous factors, such as neighborhood conditions and parental schooling levels, rather than endogenous conditions, such as family income. Whereas weighing adjustments handle the endogeneity problem in theory, in practice many analyses of income dynamics opt to avoid potential problems of endogeneity by excluding the low-income oversample.

Multiple Informants

There are several reasons to obtain the perspectives of more than one person about a child and his or her family. The primary reason is that many kinds of information are known better by some persons than others. For example, young adolescents are unlikely to be as well informed regarding their family's income and asset position as a parent. And the parent is unlikely to know mainly, if any, details about the child's use of alcohol or drugs or about his or her sexual behavior.

The perspective of a respondent may also be skewed simply because the individual lacks sufficient distance and objectivity to make a judgment. Assessments of children's behavior by teachers, for example, tend to be considerably less positive than assessments provided by parents (Moore, 1986). Using data from the 1976 wave of the National Survey of Children, Moore found that the evaluation of the classroom teacher seems to be the better predictor of later behavior. When multiple respondents address the same questions or issues, this can help researchers address and perhaps adjust for measurement errors of various sorts.

Although obtaining data from multiple informants appears to be an expensive addition to a study, it can often be done rather cheaply. For example, in face-to-face interviews, interviewers are already in the home and can provide assessments on a wide array of topics. The biggest incremental cost of obtaining additional information often consists of the careful training of the interviewers and the coding of the observational data. Similarly, data from adolescents can sometimes be obtained by giving the adolescent a self-administered questionnaire to complete while the parent is being interviewed, or to mail back. Following up on nonresponse can increase costs, as can possible additional incentive payments. However, the increment to data quality afforded by having a second respondent is so high in some instances (such as teenage substance use, delinquency, and sexual behavior) that even a substantial additional cost is warranted.

Minimizing Attrition in Longitudinal Surveys

The utility of data from all surveys is threatened by nonresponse in their initial (or, in the case of cross-sectional surveys, only) interviewing wave. Longitudinal surveys face the problem of additional nonresponse in subsequent waves. We view as crucial that surveys budget sufficient resources to minimize possible problems owing to attrition. Techniques for motivating interviewers and respondents have been developed to produce response rates in the initial wave in excess of 80 percent and to keep subsequent nonresponse to no more than a few percentage points. Better still are surveys that attempt reinterviews with nonrespondents from prior waves. Some 14 years after its initial interview, the National Longitudinal Survey of Youth routinely interviews more than 90 percent of its wave-1 respondents.

Some attrition is inevitable even in the best-run longitudinal surveys. An important but neglected challenge in using longitudinal data is to gauge the impact of attrition on estimates of developmental models. It has proved almost impossible to do this since the determinants of attrition overlap extensively with the righthand-side measures in the developmental models. Helpful here would be attrition-related experimental treatments on the part of the survey organizations. For example, one might offer a monetary payment to a random portion (say, one-fourth) of the sample that was well in excess of the payment offered to the rest of the sample. This information would be very helpful in discerning the impact of attrition on the behavior of interest.

Characteristics of Existing Surveys

We have chosen for review 12 major federally sponsored surveys containing data on children. The surveys have been grouped into three categories: long-term longitudinal, short-term longitudinal, and cross-sectional. In keeping with our emphasis on the importance of collecting data across time, all but two of the surveys reviewed are longitudinal to some degree. Specific suggestions for improving each survey are presented in the next section.

The salient characteristics of each survey are presented in Tables 1 and 2, reflecting our criteria for good survey data on children. Table 1 summarizes available data on family resources, family processes, and extrafamilial context; special advantages and disadvantages of each survey; and sample size and periodicity.8 Child characteristics are summarized in Table 2 by age group and, within age group, by substantive category. A complete list of possible characteristics for each column is presented at the end of each table, with brief explanations as needed.

Long-Term Longitudinal Surveys

Panel Study of Income Dynamics

The Panel Survey of Income Dynamics (PSID) has been conducted on an annual basis since 1968. In 1993, the survey involved some 7,900 households. It includes an oversample of black and Hispanic families, facilitating separate analyses of these groups. Adult family members (and their children) who leave to establish a new household are followed and included in subsequent waves. The survey collects very limited data on all children in the household. Children are interviewed only after leaving to establish their own households.

The PSID contains extensive and detailed information on family material resources; transfer income data are available on a monthly basis. It contains very limited information on family process; however, valuable information is available on family spending patterns. Census data relating to place of residence are available at the tract, zip code, county, and state levels.

Only very limited information is available concerning child outcomes before age 16, including birthweight and an occasional report of health status. Schooling, labor, and income data are recorded for all children ages 16 and older. Complete birth histories are also available for all female children age 12 and over. Children who go on to form their own households are given the full battery of survey questions.

National Longitudinal Survey of Youth

The National Longitudinal Survey of Youth (NLSY) is an annual survey begun in 1979. The original sample included over 12,000 young men and women ages 14–21. A special military subsample was included, as were oversamples for blacks, Latinos, and poor whites. Among respondents to the 1979 survey attrition has been very low, about 10 percent at the 1991 interview. As of 1994, respondents range in age from 29 to 36. Current plans are to continue the survey at least for the foreseeable future. Interviews will take place every other year after 1994.

Parents are interviewed about family income data only, and only when their children are under age 18. All other information on personal and family background is obtained in interviews with the youth. The survey is designed to look at labor force participation and the transition from school to work and so is especially useful for analyses relating to the transition to adulthood.

Information on material resources of the family of origin is extensive, but only for those respondents who were living with that family at the time of the first survey. A limited set of retrospective questions regarding the condition of the family at age 14 is available for all respondents. Process measures for family of origin are practically nonexistent, though some data might be gleaned from a special time-use survey filled out by the youth in 1981 for those youth still living at home. Available contextual data include population descriptors for county and state of residence, characteristics of the high school attended, and a few questions regarding peers. Data files with appended zip code level census data have been created and may be available on a restricted basis.

Outcome data are available in many substantive areas beginning with the first year of the survey, at which time respondents ranged in age from 14 to 21 . Many measures are available on a yearly basis into adulthood (ages 29–36 as of 1994). The dataset is particularly rich in schooling and labor force data. For example, labor force activity data are available on a weekly basis beginning with the calendar year previous to the first survey in 1979. Some retrospective data regarding fertility, drug use, and high school activities are available for all respondents.

National Longitudinal Survey of Youth: Child-Mother Data

Surveys of the children of female respondents of the NLSY were begun in 1986 and continue on a biennial basis. Following the NLSY sample frame, there is an oversampling of blacks and Hispanics. All children in each family are surveyed, allowing for analyses of sibling pairs.

Although the sample of women in the NLSY is nationally representative, the sample of their children is not representative at present, since they are disproportionately the children of early childbearers. As the survey continues and children of delayed childbearers become part of the sample, it will become increasingly representative.

All of the data collected in this biennial supplement are merged with the annual data collected in the NLSY main data file. Thus, there is a considerable amount of detailed income, education, and labor force information available on the parents of these children, particularly the mothers. Family-process data are limited but include a HOME environment scale and questions relating to marital conflict. Contextual data are limited to county and state characteristics.

Through a combination of mother interview and in-home child assessment, detailed and age-appropriate measures of each child's social, emotional, cognitive, and physiological development are taken. These measures include the Behavior Problems Index, the Peabody Picture Vocabulary Test-Revised, the McCarthy Scale of Children's Abilities, and the Peabody Individual Achievement Test. In 1992, questions were added concerning television-viewing habits and how older children spend their time after school. Pre-and postnatal information is also gathered for each child.

Although plans to interview teachers have been dropped, school data are being added to the Child-Mother data file. Starting in 1994, the children of the NLSY ages 15 and over will be given a full, personal interview of the kind given to their parents when the study first began. A new NLSY cohort study is to be fielded; it will probably include some of these children of NLSY mothers in the new sample.

National Educational Longitudinal Survey of 1988

The National Educational Longitudinal Survey (NELS:88) is a large, nationally representative longitudinal survey of eighth graders begun in 1988. designed to focus on school and school-to-work transitions. The survey has been taken every two years from 1988 to 1994. The next wave is planned for 1998. The initial sample size was 24,599 students from 1,057 public, private, and church-affiliated schools. The sample has been enlarged in 1990 and 1992 in order to make it representative of 10th and 12th graders for those respective years. Because of a two-stage cluster sample design that included approximately 24 children per sample school, hierarchical linear modeling (a multilevel analysis technique) can be applied to these data.

The survey includes responses to detailed questionnaires from students, their parents, several teachers, and the school administrator. Student achievement scores are also included. Parent questionnaires were left out of the 1990 wave, leaving an unfortunate gap in the data on family background.

Data on family material resources are sparse, limited to total family income and assets (parental report). Family-process measures include questions about parent-child activities and time together, rules, and patterns of parent-child communication. In 1992, questions were asked of parents regarding their intended financial contributions toward their child's postsecondary education. Contextual data are a great strength of this dataset. Detailed information regarding the characteristics of the school and the specific classes attended by the student is included. Some questions are asked about peer attitudes. Finally, census data for each school catchment area are available on a restricted basis to specially licensed data users.

Child outcomes concentrate on schooling and academic achievement and also include questions on work, prosocial and delinquent behaviors, and childbearing. Information on preschool attendance is the only retrospective child data gathered.

National Survey of Children

The National Survey of Children (NSC) is a longitudinal survey begun in 1976. with a sample of 2,301 children ages 7–11. Waves 2 and 3 (completed in 1981 and 1987) followed a large subset of these children through ages 18–22. In wave 2, the survey included all children of disrupted or high-conflict families and a subsample of the rest. The sample size at wave 3 was 1,147. No further waves are planned. Black households were oversampled. Two children were interviewed in a subsample of households (554 families in 1976), allowing for the possibility of sibling studies. Parents, teachers, and children were all interviewed. The initial wave included in-person child assessments. Similar questions are often asked of both parents and their children, allowing for interesting comparisons across respondents.

Data on family material resources are limited primarily to the year in which the interviews took place, although some retrospective information was obtained in wave 3 regarding welfare receipt and maternal employment patterns. Family income is recorded in categories. The family-process measures are the most detailed and numerous of any of the surveys reviewed here. They include measures of family activities, time together, parent-child and parent-parent communication and decision-making styles, styles of parenting, rules in the home, measures of parental conflict and violence, and detailed parental marital histories. Data on the quality of relations of both parent and child with the absent parent post-separation/divorce are also gathered. Contextual data include characteristics of the child's school, school classes, and neighborhood, as well as zip code and state-level characteristics, including state policy measures such as AFDC benefit level.

Child outcome measures include a broad array of age-appropriate measures of physical health, cognitive abilities and achievement, emotional well-being, and social development. The data have been used most frequently to examine the impact of family processes (particularly separation and divorce) on various measures of child development.

Short-Term Longitudinal Surveys

Survey of Income and Program Participation

The Survey of Income and Program Participation (SIPP) is a continuous survey in which panels are interviewed every four months for approximately two and one-half years. It is a representative survey of the U.S. civilian noninstitutionalized population. Panels have varied in size from 13,000 to 21,500 households. The 1993 panel has 20,000 households. Because of the overlapping design, cross-sectional analyses can be produced combining two panels, doubling the sample size in many years. Members age 15 and older (and their dependents) who leave the household during the period covered by the survey are followed.

The core questionnaire, repeated every four months, asks detailed questions concerning employment, income, and participation in federal social support programs. Much of the information is collected on a month-by-month basis. Questions are asked about all adults age 15 and over in the household. Special modules covering personal history and data on school enrollment and financing are administered once or twice to each panel.

In addition, there are a number of special topical modules. Some have been asked of every panel to date; others have been fielded only once or twice. Topics include child care arrangements, child support agreements, functional limitations and disability, utilization of health care services, support for nonhousehold members, and others.

Information on family processes is very limited. Marital histories are available for parents in the household. Detailed child care information is collected concerning each of the three youngest children in the household. In addition, the child support module (asked once of each panel) gathers detailed information regarding child support agreements, payments made, and time spent by children with the noncustodial parent.

Information in the core questionnaire is gathered for all children in the household who are age 15 or older. Information beyond basic demographic information (age, sex, etc.) is generally not available for children under age 15, unless collected in a topical module such as the module on child care.

Beginning in 1996, the Census Bureau plans to change the sample design and field nonoverlapping panels of 50,000 households, to be followed for a total of 52 months. In addition, a special SIPP panel, the Survey of Program Dynamics (SPD), is being designed to last for 10 years. The SPD will make special efforts to collect data on the children in SIPP households. It will be designed as an extension of the SIPP 1993 panel.

National Survey of Families and Households

The National Survey of Families and Households (NSFH) was designed to accommodate family-oriented research on a wide variety of topics and from many research perspectives. Households were originally surveyed in 1987–1988, and again five years later. There are no plans at this time to follow-up with a third wave, although respondents are being tracked, making a third wave possible. The original sample included over 13,000 households, with a total of 7,926 children represented in those households. Double samples were taken for black and Hispanic households, single-parent families, cohabiting and newly married couples, and households with stepchildren. The oversampling of less common family types makes this a uniquely valuable survey for family research.

In-person interviews were conducted with a randomly chosen adult over age 18. In addition, self-administered surveys were given to the respondent and to the spouse/partner. In the follow-up survey, brief interviews were completed with those who were focal children age 5 and over in the initial survey. Both focal children who had left the household and parents who had separated or divorced since the original survey were followed and interviewed. In addition, a parent of the main respondent was also interviewed in the follow-up survey.

A limited amount of sociodemographic and behavioral information is gathered on every child in the household. In addition, there is a much more detailed set of age-appropriate questions asked concerning a randomly selected focal child. These questions cover a broad range of outcomes in the areas of health, social development, behavior problems, and cognitive achievement. No direct assessments of the child were made in either wave of the survey, however.

Focal children were ages 0–18 at the time of the first survey. Thus, while this dataset allows one to look at a child only at two time points, five years apart, it does allow one to analyze in detail the transitions between all adjacent developmental phases, including the transition to early adulthood.

Information concerning a household's material resources is very detailed, although the data refer primarily to the status of the household and its members at the time of the survey only. Income is identified by source, and in many cases according to the person who generated it. Questions are asked concerning income flows to and from extended kin and between ex-spouses.

Family-process data are one of the great strengths of this survey owing to the breadth and the depth of information gathered. Exact relationships between the adult respondent, spouse/partner, and each of the children in the household are recorded, as are detailed marital and fertility histories of all adult respondents. Questions are asked concerning time spent and activities engaged in by the focal child and each parent (including noncustodial parents), rules, parent-child communication patterns and conflict. In addition, there are many measures of the quality of the spousal relationship, including interspousal conflict, conflict resolution styles, violence, global marriage quality, and sharing of household duties. The CES-D depression scale is also administered to all primary adult respondents.

Finally, there is a host of measures of social capital regarding the relationship between adult respondents and extended kin, friends and neighbors, and community organizations such as the church, the PTA, and recreational, civic, political, and professional groups. Some county-level sociodemographic and labor force characteristics are also included in the dataset.

High School and Beyond

High School and Beyond (HS&B) is a nationally representative longitudinal survey of high school sophomores and seniors begun in 1980, with biennial follow-ups through 1986. It has a two-stage sampling design. In the first stage, public and private high schools were chosen. The following school types were oversampled: alternative, high-performance private, Hispanic, non-Catholic private, and black Catholic private. A total of 36 sophomores and 36 seniors from each school were included in the sample design (allowing for hierarchical linear modeling). A total of 30,030 sophomores and 28,240 seniors were interviewed in the first wave. A subsample of approximately one-half of the students was included in the 1982 through 1986 follow-up surveys. A 1992 follow-up of 1,300 sophomores was also conducted. Special files are available that allow one to link twins, siblings, and friends within the file.

Questionnaires were filled out by students, school administrators, and up four of the students' teachers. Questionnaires were also filled out by parents of approximately 10 percent of the students. In addition, standardized test results are available for all students, as are complete high school transcripts for a large subsample of the sophomore cohort.

For the 90 percent of the sample whose parents did not fill out a questionnaire, data on family material resources are limited to a categorical income variable and a question on home ownership. The parent questionnaire, however, provides data on parental income, assets, and expenses, as well as parental education and employment.

Family-process measures in this survey include measures of parental aspirations and attitudes regarding secondary for their child and financial planning for college. Such information, however, is available only for the 10 percent of the sample whose parents filled out questionnaires. In addition, all youth respondents are asked questions concerning the degree of parental supervision of homework and whether parents usually know where the youth is at any given time.

Contextual data are limited primarily to the extensive information available on the high school. Data include enrollment, demographic breakdowns of both students and faculty, course offerings, participation in federal programs, funding sources, school discipline problems, and grading systems. Local labor market indicators for the county or metropolitan area of residence are also available. Finally, a ''friends'' file permits one to identify and link friends within the sample, which allows for the possibility of peer measures.

Child outcomes covered in this database include information on high school and postsecondary education measures, job training and employment history, fertility and marriage, and income. Job and employment data are arranged in an event-history format. Available cognitive test measures include vocabulary, reading, mathematics, science, writing, and civic education. Data on educational aspirations, personal attitudes, and beliefs are also included.

Consumer Expenditure Surveys

Consumer Expenditure Surveys (CEX) collect detailed data on consumer expenditure patterns. The data are used to determine the need to revise the Consumer Price Index (CPI). There are two surveys, each with a sample size of 5,000. Each is a nationally representative sample of the U.S. noninstitutionalized population. In the CEX Quarterly Interview Survey, respondents are interviewed quarterly for a total of five quarters; they are asked to report major expenditures and those that occurred on a regular basis over the preceding quarter. Examples include rent or mortgage payments, utility bills, and purchases of home appliances and furnishings. The survey is continuous, with one-fifth of the sample replaced each quarter. In the Diary Survey, respondents are asked to keep two one-week diaries in which they record smaller purchases (e.g., food, cleaning supplies, clothes).

Information on material resources is detailed, covering the amount and sources of income, assets, and tenure, as well as participation in federal social welfare programs. This is by far the most detailed survey of spending patterns collected on a representative national sample of families with children; however, family-process data are limited to spending patterns. There are no contextual measures.

Data on children under age 14 are limited to age, race, Hispanic origin, sex, and exact relationship to the respondent. Employment data are gathered for all persons age 14 and older.

National Crime Victimization Survey

The National Crime Victimization Survey (NCVS) is designed to collect data on the incidence of crime as reported by the victims of crime. The survey is very large, covering approximately 100,000 people in 49,000 households. All persons ages 12 and older in the household are interviewed. Household members are interviewed every six months for three years, for a total of six interviews. The survey is continuous, with one-sixth of the sample replaced every six months. Households that move during the period of the survey are not followed. The sample is nationally representative. In addition, representative state-level data are available for the 21 most populous states.

Data on family material resources are limited to a measure of total family income. In addition, information on Medicaid and private health insurance coverage is obtained for those who visit the hospital as a result of an injury received during a crime. Family-process data are limited to the reporting of family violence that a respondent considers to be a crime (e.g., assault, rape) and is willing to report. State and region of residence are not identified in publicly available versions of the data file.

Child outcome measures are limited to detailed reports of crimes that affect them (if they are 12 or older) or to their household (regardless of their age). This is nevertheless important and unique information on child well-being.

Cross-Sectional Surveys

Decennial Census, Public Use Micro-Sample

The Public Use Micro-Sample (PUMS) is a large sample of households drawn from the 17 percent of the population that fills out the decennial census long form. The 5 Percent Sample, as it is called, includes detailed population and housing characteristics for over 15 million persons. A 1 Percent Sample version is also available. Data are available for the household, for each family and subfamily within the household, and for each individual.

Data on income are fairly detailed, with income reported by source and for each member of the household age 15 and older. Home ownership is also reported; assets and health insurance are not covered. There are no family process data gathered in the census. State and county group (census-defined constellations of contiguous counties) are identified, allowing for the possibility that state and county-group characteristics can be appended to the data. In 1970, a special "neighborhood file" was produced that included sociodemographic characteristics of the neighborhood of residence. Under special arrangement with the Census Bureau, a private researcher, Rebecca Clark, created a similar file using data from the 1980 census. For reasons of confidentiality, the latter file is not publicly available.

For children under the age of 15, outcome data are limited to school enrollment. For those age 15 and older, measured outcomes include school enrollment and educational attainment, employment status and characteristics, income, welfare status, marital status, and whether they have a child of their own living with them. If there is a health condition that impedes employment, that is also recorded.

National Health Interview Survey—Child Health Supplement 1988

The National Health Interview Survey—Child Health Supplement 1988 was a special supplement to the National Health Interview Survey, an annual, cross-sectional, and nationally representative survey of the U.S. civilian, noninstitutionalized population. For this supplement, additional information was gathered for one child between the ages of 0 and 17 in every sample household containing children. A knowledgeable adult member of the household, most often the mother, served as proxy respondent for each selected child. The 1988 Child Health Supplement gathered data on 17,110 children. Black families are oversampled. A similar supplement was fielded in 1981.

Family income data are limited to total family income and welfare status. In addition, health insurance coverage is also ascertained. Family process measures include child care arrangements, frequency of contact with absent parents, maternal marital history, and habits regarding seat belt use and the regularity of children's bedtimes. Contextual data are limited to a history of residential moves.

There are extensive data related to child health, including information on the child's birth, recent accidents, injuries and poisonings, particular childhood conditions, chronic conditions, emotional and behavior problems (including a 28-item Behavior Problems Index), and use of health care services. There are, in addition, many useful outcome measures that are not directly related to health, including whether the child has repeated a grade, reasons for discontinuing school, and the presence of learning disabilities or developmental delays.

The National Health Interview Survey is currently being redesigned. Current plans call for regular occasional supplements on child health. This redesigned survey is scheduled to be fielded starting around 1996.

Improving Existing Surveys

We detail below changes in existing practices or new content in large national surveys that could significantly improve them. Before launching into survey-by-survey details, we begin with some generic procedural issues for surveys.

Research on the development of children using existing data is greatly facilitated if the survey data are organized on a child-by-child basis. In terms of questionnaire construction, this means that questions should be designed to obtain information on each individual child, using questions that are developmentally appropriate. As simple as this advice may seem, it is remarkable how rarely it is followed. For example, in the SIPP: (1) Federal Food Breakfast Program and Federal School Lunch Program participation in formation is not gathered on individual children; (2) child care information is asked only about the three youngest children; and (3) divorce/separation visitation data are gathered only on the oldest child if not all children have the same visitation schedule. This structure produces some bizarre data patterns on children when considered in a longitudinal context. For example, child care data would be collected on the oldest of three children, but not after the birth of a fourth child. The older of two children would have a longer time series that would cease if two younger siblings were born during the panel period.

Undoubtedly, decisions to limit the number of children about whom such questions are asked are based on interviewing time constraints. However, we believe that it is far better to have at least some information available on program participation, child care, visitation, and similar activities for all children (and included in those children's own data records) than to have details about only some of the children. If such data are collected on a grid, data collection time requirements are minimized. If time pressures preclude obtaining some data for all siblings, child age would represent a better decision rule than using the three youngest as the criterion for some questions and other factors as the criterion for other questions. Alternatively, a focal child or two focal children per family might be selected about whom detailed information will be consistently obtained.

One hopes that computer-assisted interviewing methods will reduce the perceived need for such unfortunate restrictions, but there is little reason for them, regardless of the mode of interview. Question sequences about children should be constructed so that the questions are asked about all children for whom they are appropriate.

Second, repeating a point made earlier, it is important to collect information on family relationships that allows researchers to classify families according to type (step or biological, for example) and to establish the relationships among all pairs of individuals in the household. Gathering information about each family member's relationship to a reference person does not provide the needed information. One option, chosen by SIPP in some of its panels, is to enumerate a relationship matrix. Family members are listed down the rows and across the columns of the matrix, and the interviewer then establishes the relationship (e.g., biological parent vs. stepparent; sibling, half-or stepsibling; cousin) between each pair of individuals. This option is somewhat cumbersome, especially in large families, particularly when computer-assisted personal interviewing (CAPI) is used. An alternative being developed by the Census for SIPP and the SPD, to be administered beginning in 1996, is to ask for relationship to the reference person as well as the identity (i.e., unique ID or listing line number) of each person's mother and father if that parent is also present in the household. Additional questions allow one to identify biological, step, and adoptive parent-child relationships and to identify virtually all remaining pair relationships within the household.

Third, child-centered analyses are facilitated by the release of data files in which separate records exist for each individual child or in which the data can easily be transformed into child-specific record form. In the case of a survey such as SIPP, this should be true regardless of whether the child was actually given his or her own interview. The medium increasingly preferred by analysts is CD-ROM. Retrieval software on the CDs is also desirable; the CD-ROMs prepared for the NLSY provide a model here.

Fourth, longitudinal studies based on families or households should develop rules for following individuals that properly track the experiences of all children. In both the PSID and SIPP, for example, the samples are defined to include individuals who were present in wave-1 households as well as births to individuals who were present in wave-1 households. Although minor children are properly part of the sample in both studies, following the rules in SIPP and, until recently, in the PSID did not provide for minor children to be followed if they moved to a household that did not contain at least one adult sample member. This procedure is clearly a mistake because it throws away the opportunity to track a relatively small but very important group of children who live in a variety of family structures during childhood. We know next to nothing about how frequent such occurrences are or about the developmental consequences of such arrangements.

We now turn to a survey-by-survey list.

Consumer Expenditure Surveys

The CEX provides a wealth of detail about household expenditures. With a few additions it could become an invaluable source of information for opening up the black box of intrafamily resources allocation.

We make a number of suggestions to improve the organization of existing information in the CEX:

1.

Collect better information on family relationships. It is crucial to know how children are related to all of the adults in the household, because analysts will want to be able to link the incomes of those persons with the expenditures for the children. The existing family relationship information should be supplemented by asking for the household-listing line number of the natural father or mother of each household member. It would also be desirable to ascertain whether children share the same biological absent parent (for a more detailed discussion, see the recommendations for the decennial census listed below).

2.

Identify state of residence for as many states as possible.

3.

Attach census tract data to each record and offer the data on a restricted basis. Purchases of neighborhood inputs through choice of residence are among the most important ways that parents spend money on their children.

4.

The Bureau of Labor Statistics should consider creating a special child subfile to facilitate research on children. Such a file would include expenditure data on children and basic family demographic and socioeconomic data; it would include identifiers allowing researchers to easily link child records with the more complete family records in the larger CEX data file.

5.

Make the data available on CD-ROM with good documentation and provide a program to facilitate data extraction such as that currently available for the NLSY.

Suggestions for low-marginal-cost additions to the data include:

1.

Collect greater detail on child care expenditures. The CEX currently collects great detail on many expenditure categories but has only a single question about child care expenditures. Some details on the type of child care services that were purchased would be desirable.

2.

Collect additional data on expenditures explicitly made for children. At present, expenditures on clothing are about the only such data in the survey. Additional measures might include expenditures on education (e.g., private tuition, books, classes in art, sports, and other special classes), camps, computers, and expenditures on outings, for example, to the zoo, museum, and concerts.

3.

For expenditures on clothing, establish the identity of the person for whom the clothing was purchased. Most expenditures families make benefit several family members at the same time. An exception is clothing, because most clothing is purchased for an individual member, even though some of it may find its way to the wardrobes of other family members. Thus, clothing expenditures constitute a promising indicator of a family's intrafamily resource allocations.

The Decennial Census

Although cross-sectional and limited to what can be collected on a self-enumerated form, data from the decennial census have been used for a surprising number of developmental studies. Interesting developmental measures include the school and fertility status of adolescents as well as the economic and demographic characteristics of the families in which children live. Most important, of course, is the fact that the PUMS file contains massive case counts. Researchers have exploited this feature by constructing samples of very rare populations, such as twins or immigrants from different countries or regions.

It would be presumptuous to believe that child development considerations should dictate substantial changes to the census form. However, we would be remiss if we failed to propose a useful change in the way that the census form collects its household relationship data:

1.

The information on household-composition collected in the census derives from the identification of the householder (the household member in whose name the dwelling is owned or rented), a listing of all other household members, and the following question for each household member: "How is this person related to PERSON 1 [i.e, the householder]?" Response categories include: (a) husband/wife; (b) natural-born or adopted son/daughter; (c) stepson/stepdaughter; (d) brother/sister; (e) father/mother; (f) grandchild; (g) other relative (specify); (h) roomer, boarder, or foster child; (i) housemate, roommate; (j) unmarried partner; (k) other nonrelative. This information could be supplemented on the long form with questions establishing whether the individual's natural father or mother live in the household and, if so, who that person is. This could be done by asking for the household-listing line number of the natural father or mother of each household member. It would also be desirable to ascertain whether children share the same biological absent parent.

2.

The Census Bureau should consider restoring the marital history question, which was eliminated from the 1990 census.

3.

Because the census asks only about completed schooling, it is not possible to identify children who are enrolled in but have not completed preschool. Rewording the question would allow for such identification.

4.

The Census Bureau should consider producing a matched file with family and neighborhood-level data from the 1990 census that is comparable to similar files produced for 1970 or 1980. A publicly available file such as that produced with the 1970 data would be clearly preferable, but it would be also considerably more expensive to produce because it involves generating new neighborhood-level data rather than using data from identifiable census tracts. A less expensive alternative would be for the Census Bureau to create a file that simply matches existing tract data to household records and make it available to researchers on a restricted basis.

5.

The Census Bureau should consider creatine family-level records in addition to household and person records for their PUMS microdata files, as is currently done for the Current Population Survey.

The National Crime Victimization Survey

The National Crime Victimization Survey is one of the largest ongoing surveys carried out by the federal government. Its information on crime victimization makes it a uniquely valuable resource for researching the well-being of adolescents and young adults (children below age 12 are not interviewed). There are a number of changes that would greatly enhance the value of this underutilized data set for scholarly research.

We make one suggestion to improve the organization of existing information for this survey:

1.

Continue to create longitudinal files containing data on all seven interviews over the three-year period during which households participate. At present there is only one such file available for surveys having taken place in the period 1987–1990.

Suggestions for low-marginal-cost additions to the data include:

1.

Expand the income and demographic data collected on all members of the household, which are currently minimal in this survey. Such improvements would include more detailed income and employment data, identification of the exact relationships among all respondents in the household, an expanded Hispanic-origin question to allow for the separate identification of major Hispanic subgroups, and welfare receipt. Data for children under the age of 12 (currently limited to age, sex, race, and national origin) should he expanded to include exact relationship to adults in the household and the grade in which they are currently enrolled.

2.

Append state identifiers and tract-level census data to each record and make the resulting file available on a restricted basis. State laws and neighborhood conditions are of particular interest to those who study crime. At present, neither state nor region of residence is identified on the public tile out of concern that victims may be identifiable because of the rarity of some crimes.

3.

Ask follow-up questions concerning the disposition of crimes reported in previous waves; at present they are not. This would enhance the utility of the crime data substantially.

Suggestions br higher-cost additions to the data include:

1.

Follow families who move out of the household. Currently, data for all seven waves of the survey are available only for those who do not change their place of residence. The Bureau of Justice Statistics has been considering doing this for some time and is monitoring the experience of the SIPP survey, which does follow families.

2.

Gather data on the criminal victimization of children under age 12. The physical and sexual abuse of young children is far too common and too important to be ignored in a major survey of crime victimization. We realize that there are many problems in gathering accurate data of this sort for children under age 12, particularly in the context of a telephone interview. Serious thought should be given to a special module on the victimization of children under age 12, to be administered in the home to a large subsample. Alternatively, it may be useful to ask retrospective questions of respondents ages 12–17 about incidences of assault or abuse at earlier ages.

National Educational Longitudinal Survey of 1988

The NELS:88 survey contains many of the characteristics that we have identified as desirable for a dataset to study child and adolescent development. Parents, teachers, administrators, and the students themselves are all interviewed. Detailed information on school and neighborhood characteristics us available. The respondents are interviewed regularly over the course of their transition from adolescence to early adulthood.

A number of improvements could have been made to this survey earlier on (for example, the fielding of a parent questionnaire for the 1990 and 1994 waves, more detailed family income and process measures). At present, the main opportunity for improvement would be to discard current plans to skip the 1996 wave, which will leave a four-year gap between the third and fourth waves of the survey. This four-year spacing between interviews during what is for most youth the key transition period to adulthood (ages 20–24) substantially weakens this dataset for studying the transition to adulthood in general and the school-to-work transition in particular. Reinstating the 1996 wave would be highly desirable. We have been informed by Department of Education personnel working on this survey that, at this late date, intervention by the secretary of education would be required to reinstate the 1996 wave. Recognizing that there are substantial financial constraints involved, we would therefore propose that a telephone survey be considered for a 1996 wave.

In addition, it would be desirable for the current survey to be continued in some form until at least the year 2002, when respondents will have reached approximately 28 years of age. Many adult attainment measures, particularly earnings and work patterns, do not stabilize until that age. It is important that researchers be able to relate the rich information available in this dataset to stable and long-term adult socioeconomic outcomes.

National Longitudinal Survey of Youth (1979 Cohort)

In 1994, respondents in this survey ranged in age from 29 to 36. It is important that this cohort continue to be followed, at least for the next 5 to 10 years in order to continue to track important developments in fertility, earnings, and other important adult outcomes. Given financial constraints and problems associated with the accuracy of retrospective data, continuation of the NLSY as an annual survey would have been preferable, switching to telephone administration to save costs. However, given that the decision to move to a biennial format has already been made, we agree with current plans to continue with in-person interviews.

National Longitudinal Survey of Youth—Child-Mother Data

The child mother data supplements to the NLSY have proved very useful to policy scholars in child development, economics, sociology and demography (Brooks-Gunn et al., 1991). They could be enhanced by a number of relatively low-cost additions:

1.

Augment information on parental relationships. It is difficult to determine the relationship between each child in a home and a residential father. Likewise, more information should be collected on noncustodial fathers about child support (financial, time, and psychological) as well as conflicts (also financial, time, and psychological) between the parents about the child.

2.

Add more detail on the adolescent years as children of the NLSY age, paralleling what is known about the mothers during their own adolescent years.

3.

Add more detailed information on mental and emotional health. collected from teachers, child, and mother (supplementing existing questions).

4.

When these children reach their early twenties, data could be collected by telephone to reduce costs, with only occasional in-depth personal interviews.

There were plans to interview the teachers of children of the NLSY. It is our understanding that the Bureau of Labor Statistics discarded these plans out of concern that teachers would be able to look up sensitive information on one of their students and their parents. A number of federally collected or sponsored datasets have managed to collect sensitive parent and child data and conduct teacher interviews without experiencing confidentiality problems, including the National Educational Longitudinal Survey of 1988 and the National Survey of Children. We urge the bureau to develop ways in which teacher data could become part of NLSY data collection.

National Longitudinal Survey of Youth—A New Cohort?

It appears that the Bureau of Labor Statistics has secured funding for a new set of NLSY cohorts. Since the 1979 NLSY cohorts have proved to be such a valuable source of data for developmental studies, we offer some suggestions about the design of the new panel.

Although data collected from the adolescents ages 14–21 who constituted the 1979 panel have also been used for many developmental studies of individuals in late adolescence and early adulthood, the 14–21 range is far from optimal for such studies. A key problem is the lack of arguably ''exogenous'' information for the period prior to the time when outcomes are assessed. Many of the processes that affect adolescent development begin in early adolescence or before. Owing to limits on the accuracy of retrospective recall, concurrent data collected from, say, a cohort of 19-year-olds will have little of the rich parental background information available from individuals in a younger age range initially observed while they were still living with their parents. Data on attitudes and expectations of the youth are also problematic, since, if gathered for the first time from 19-year-olds, such data reflect the influence of endogenous decisions already taken by the youth.

If, for example, age 16 is considered to be the point at which major decisions regarding schooling and fertility are first taken, then the 14–21 age span provides only two cohorts—age 14 and age 15—for causal analyses. (Indeed, a number of analyses of teenage fertility have limited their samples in precisely this way.) Thus, starting with a younger age range would increase the sample size for causal modeling of school, family, and work transitions in youth and early adulthood.

From the perspective of sibling-based analyses, an initial sample that includes ages older than about 18 introduces a problematic set of possible biases. Although some 18-year-olds live in their own households, most still live with their parent(s). Data on independent 18-year-olds living on their own will not be accompanied by observations on younger siblings in the chosen age range since those younger sibling will, by and large, still be living with parents. (It would be possible to find the parental families and interview such siblings, but that would be expensive.) Also, 18-year-olds regardless of living arrangements are unlikely to have older siblings still living with them. The younger the age range, the less severe are these possible problems.

We view sibling-based problems in the NLS samples as important, since sibling observations are a great strength of the data set, and we predict that growing numbers of analysts will rely on the natural experiments inherent in sibling comparisons to control for heterogeneous family effects. Coincidentally, sibling observations are also much cheaper than observations on children in separate families, since only one set of parental data needs to be gathered and it often takes only one personal visit from an interviewer to gather the necessary information from all siblings. Our proposal for a new sample maximizes the number of sibling observations.

These considerations argue for a younger age range in the new NLS, with an upper age limit of 17–18 and beginning at age 12 or younger. The Bureau of Labor Statistics and the National Institute for Child Health and Human Development should seriously consider combining the need for a younger NLS age cohort with the meed for a mew national survey of children (discussed below) by combining the samples into a single, dwelling-based sampling frame. Detailed longitudinal information should be gathered on two of the children, with some information gathered on all children in the household. This would result in a large, nationally representative sample of children ages 0 through 18. The combined sample would represent a very broad age range, be clustered within families to reduce interviewing costs, provide abundant observations on siblings, and provide data from a "whole child" perspective. Such a survey should be conducted on an annual basis; to reduce costs, a personal interview could be conducted every two years, with telephone interviews in between.

National Survey of Families and Households

The NSFH offers the most comprehensive picture of the American family available in ongoing federal surveys. Information on material resources, family processes, family relationships, and nonresident kin is particularly rich.

We make a number of suggestions to improve the organization of existing data:

1.

Release data from the second wave on CD-ROM as soon as the data are ready for public distribution. Current plans call for distributing the data on CD-ROM one year after the initial release of the data on 9-track tape. We believe that this unnecessarily restricts early access to this very rich data set. When errors have been reported and corrected over the course of the first year, a connected CD can be produced and distributed at little additional cost.

2.

Offer with the CD-ROM a customized data extraction program similar to that already available for the NLSY. Although there are fewer waves of data, the records for this data set are very complex, particularly for the second wave. This is a moderately costly undertaking, but it would enhance the accessibility of the data significantly.

Suggestions for low-marginal-cost additions to the data include:

1.

Append tract-level data for place of residence and offer the resulting file on a restricted basis. Although there are important confidentiality issues here, a similar arrangement currently exists for data from the PSID.

The high-quality of the family-level data in the NSFH makes tract-level data even more valuable.

2.

If a third wave of data is collected, make arrangements to append social security earnings data to the files. This was attempted in the second wave; subsequently, however, Social Security personnel declined to allow the records search. Since such arrangements have been worked out for the Health and Retirement Survey, it would be worthwhile to reexamine this issue for the third wave.

One suggestion for a higher-cost addition to the data is as follows:

1.

Collect a third wave of data in 1997–1998. This would offer three data points over a 10-year period covering a variety of developmental periods for children. The National Survey of Children, which had three waves also spaced five years apart, has been the basis for important research on child well-being from a developmental perspective. Every effort should be made to maintain the highest response rate possible for a third wave.

National Health Interview Survey—Child Health Supplement

The large sample size of the National Health Interview Survey makes it an important source of data about American children. However, the breadth of data available regarding the physical and mental health of children and their families, particularly regarding the factors that enhance or undermine the development of children, has been minimal. The augmented information about children provided by the 1981 and 1988 Child Health Supplements made the NHIS far more useful for understanding the occurrence of such conditions as behavior problems, chronic health conditions, and accidental injuries, as well as aspects of the child's environment related to health such as child care, school adjustment, and academic success. Data from the Child Health Supplements to the NHIS have been extensively used by researchers and provide the beginning points for time-series data analysis, if a new supplement is implemented in the next several years.

The National Health Interview Survey is currently undergoing a substantial redesign. Redesign plans call for an annual core survey to be supplemented by both periodic and occasional supplements on special topics. Current plans call for a special child supplement to be carried out in 1996 if additional funds can be secured, a matter that will be settled soon. If funding is not forthcoming at this time, it would be preferable to reschedule the child supplement rather than drop it.

For the design of the supplement, adding siblings to the sample would increase the sample size and provide for sibling analyses. In addition, data collection from teenagers themselves would enable information to be obtained on topics about which parents are poorly informed, such as the substance use and sexual activity of their adolescent children.

Information on family processes should also be obtained, particularly those that affect a range of health outcomes, such as the health-related behaviors of children, treatment obtained for health conditions, the preventive health care obtained for children, the risk-taking of children, and the differences in health status and behavior across children in a family. In addition, it is important to collect sociodemographic information specific to the children for whom health data are being collected, such as the exact biological relationship between the child and the parents in the home, contact with the absent parent, insurance coverage for the child including coverage provided by an absent parent, and any transfer income or child support payments made on behalf of that child.

The National Center for Health Statistics should adopt questions that will allow researchers to identify exact relationships between each child and their parent(s) within the household (i.e., natural, step, adoptive), both in the occasional child supplement and in the annual core survey. The National Center for Health Statistics has plans to test the relationship matrix used in SIPP, but it has tentatively decided to use a less rigorous set of relationship questions out of concern that the SIPP matrix is too cumbersome. The Panel Study of Income Dynamics has adopted an approach that identifies for each child the survey line number of each parent in the household and their exact relationship. From this information, exact relationships between each child and all other household members can be unambiguously constructed in most cases.

Finally, if funding were made available to conduct longitudinal data collection based on the child supplement, prospective analyses of the factors that lead to improvements or declines in health-related behavior, accidents, illness, and risk-taking could be carried out.

Panel Study of Income Dynamics

The PSID has gathered more than 25 years of data on the family and neighborhood environments of its representative sample of children and adolescents, most of whom have been observed since birth. No other data source comes close to assembling such a high-quality longitudinal time series on family and neighborhood resources. Lacking in the PSID, however, are outcome measures on children prior to age 16. Also lacking are family-process measures.

We make two suggestions to improve the organization of existing information in the PSID:

1.

Extend to the current interviewing wave the "relationship file" that the PSID has constructed for the years 1968–1985, showing year-by-year pairwise relationships and coresidential status of all pairs of individuals associated with the same original wave-1 family.

2.

Develop and make available at minimal cost retrieval software that enables analysts to assemble data subsets and codebooks easily. Although the PSID data are now routinely released on CD-ROM, such software does not exist.

Suggestions for low-marginal-cost additions to the data include:

1.

Addition a supplement to the regular telephone interview with parents, asking for outcome measures such as grade failure and behavior problems on the part of their children.

Our suggestion for higher-cost additions to the data is as follows:

1.

Incorporate an NLSY-type personal interview to be conducted with all PSID families with children, in which the cognitive abilities of children and mothers would be tested, behavior problem reports on the children would be gathered from the mothers, and time-use activities would be reported by the mothers (in the case of younger children) or by the children themselves (in the case of older children and adolescents).

Survey of Income and Program Participation

The unique strength of SIPP lies in its rich and detailed recording of data on income and program participation. For analyses of the development of children, however, SIPP currently provides very limited information. To conduct causal analyses of child well-being, data are needed on child outcomes and on the family processes that translate resources into child outcomes. In addition, a broader array of resources needs to be measured than the set of income and program participation measures currently available.

In terms of child outcomes, at present children's health and educational attainment are the only outcomes available for study. It is proposed that measures of children's cognitive development and behavior be obtained in the Survey of Program Dynamics (the extended panel planned to follow the 1993 SIPP sample through the year 2002). With the addition of a broader array of child outcomes and family resources, combined with a longitudinal design, multivariate causal analyses can be conducted to examine how varied types of investments translate into attainment. That is, researchers and policy makers can assess the independent and interdependent effects of income by source, program participation, and family influences on children's education, employment, income, and family formation through childhood and into the early to mid-twenties. Moreover, it will be possible not only to extend the time period during which child outcomes are assessed, but also to expand the range of the independent variables that are examined. For example, beyond point-in-time or short-term estimates of welfare receipt or poverty, the implications of spending many years on welfare or in poverty can be compared with the implications of a brief spell of welfare receipt or poverty.

Examining income and program participation longitudinally represents an important expansion in research potential. Beyond this, an expansion in the range of resources measured is also needed. The only resources that are currently measured in SIPP are financial resources and human capital. Time inputs, the parents' psychological resources, and social capital resources in the community are not currently measured. In the absence of measures of these constructs, the only factors that can be examined as determinants of children's development and well-being are measures of income, program participation, and the structural characteristics of the family, such as family size and the number of parents present. The absence of a full set of resource measures can lead to erroneous conclusions about the role of these factors. For example, knowing only about income without knowing about the time parents spend interacting with their children could lead to an inflated assessment of the importance of money relative to other family inputs.

However, even with rich data on resources and child outcomes, there remains a gap in SIPP with regard to information on those family and community processes through which resources affect children's development, for example, family processes such as monitoring, communication, discipline, teaching, modeling of work, substance use, and marital behaviors. Addition of measures of these constructs would add immeasurably to SIPP in that the linkages through which poverty and program services affect children could be explored.

The addition of contextual information about the census tract of residence would also add to the utility of this data base, as is true for most national surveys.

Despite its many strengths, one disadvantage of SIPP is the relatively small sample size. Efforts to minimize attrition and to follow any children who leave the household are essential. Also, expansion of the sample size to 50,000, as currently planned, will enhance the utility of SIPP for the study of children.

Given the rich longitudinal information already being collected with SIPP on income and program participation, the incremental cost involved in adding measures of child outcomes, family processes, time inputs, and social capital is modest relative to the base cost of fielding SIPP. The incremental information will be extremely valuable for developmental research.

A New Survey of Child and Adolescent Development?

Taken as a whole, our list of proposed additions to the many national data collection projects we review is long and expensive. Unfortunately, available funds are probably insufficient to support all of these augmentations and also fund a new national survey of children. Accordingly, it is also important to consider whether it would be better to allocate resources in a piecemeal fashion across existing surveys or to attempt to pool those resources and spend a substantial fraction of them on a new survey focused exclusively on child and adolescent development.

Our suggestions for no-cost and low-cost additions to existing surveys are clearly our top priority. However, considering the trade-offs between the high-cost additions to existing surveys that we consider and the fielding of a new survey of child and adolescent development, in our judgment a new survey is the best use of available resources. The primary reason for this conclusion is that, although many of the existing datasets are likely to provide very valuable information on child development, these existing studies were either designed for other purposes or are designed too narrowly to serve as a general resource for research on children.

Specifically, among the overtly developmental data sets: (1) none of the NCES datasets covers the important period between birth and the beginning of school; none provides comprehensive data on siblings; and all are focused on very narrow cohort ranges; (2) the NLSY-Child-Mother supplement sample never represents a well-defined population of children in any given year and will not represent certain subgroups (e.g., adolescents born to older mothers) for more than a decade; and the NLSY Hispanic subsample is no longer representative due to heavy in-migration since 1979; (3) SIPP's proposed supplements on child development are ambitious, but the total amount of interviewing time available in the SIPP survey, focused as it is on income and program participation, will never be enough to provide comprehensive measures of outcomes and family process. Also, Census Bureau restrictions on data release make it difficult to add and analyze important contextual information, and the agency's perspective regarding the gathering of sensitive but important outcome and process measures unduly restricts survey content.

Also, most databases funded by a given federal department necessarily (and understandably) tend to emphasize the topics and processes of concern to that department. Yet what is needed is a data collection effort focused on the "whole child," that is, a survey that obtains information on multiple domains of well-being and factors that contribute to development across these domains.

Our list of important design features of a new study of child and adolescent development includes the following:

1.

A new national survey of children should be longitudinal, with interviews at least every other year (and, if possible, every year), an initial personal interview, and regular subsequent personal interviews aimed at gathering high-quality developmental data.

2.

It should contain a national probability sample of at least 30,000 children under age 18 and expectant parents, with oversampling among blacks and Hispanics, probably drawn from a dwelling-based sampling frame.9 Asians should also be oversampled if resources permit, although their level of geographic dispersion makes oversampling relatively more expensive. A case count of 30,000 children amounts to roughly 1,500 cases per single year of age. The longitudinal nature of the file will produce a steady flow of new cases into a given age range each year; after five years, for example, roughly 7,500 children will have been observed making the transition into school.

Included in the sample should be institutionalized children who would otherwise reside in the sampled dwelling. All children should be followed regardless of subsequent institutional status or living arrangements. Proxy information should be gathered on all older siblings of sample children.

We have considered several options for drawing the sample for such a survey, each with its own advantages. One option for this survey is to draw a fresh sample of 30,000 children. A second option is to take all PSID children under age 18 and augment them to bring the total sample to 30,000. The PSID sample would supply approximately one-fourth of the total sample. The second option offers the advantage that longitudinal data from the PSID could be merged with the new survey data for the PSID portion of the sample, significantly enhancing its value for research. A third option, discussed earlier, is to expand the age range of a new NLS cohort to cover all children younger than (or including) age 18, with BLS funding data collection on adolescents and data collection among younger children funded by another agency or agencies.10

A fourth option is to build on SIPP's plan to add child development data to its 10-year Survey of Program Dynamics sample. Some 14,000 children could come from this source, with the SIPP core and supplemental questions providing many of the measures we advocate. Although the Census Bureau has shown an increasing willingness to consider questions on child development topics, its attitude regarding the gathering of the sorts of potentially sensitive information that must be part of any new survey of child and adolescent development prevents us from recommending SIPP as the preferred vehicle for such a survey at this time. This in no way reflects on the value of adding regular child modules to SIPP or the SPD, however. We heartily endorse the value of the SPD and of making child development an important focus of SIPP and the SPD.

3.

At least some developmental and family-process data should be collected on all children, including very young ones. For time-consuming outcome measures in households with three or more children, 2 siblings should be randomly selected, with oversampling of twins. Information should be gathered about all important domains, with a sufficient number of questions within each domain to ensure reliability of measurement, but no single domain consuming a disproportionate share of interviewing time.

4.

Parents should be asked for their permission to conduct periodic teacher interviews and to gather social security earnings histories.

5.

The design should include interviews from multiple informants, including parents, children, and teachers. Interviewer observation data should also be included.

6.

Neighborhood information from the most recent census should be matched to the family-level survey data.

7.

A new set of cohorts should be initiated at least every decade. At a minimum, the new cohorts should be chosen to represent children born since the initial wave of the survey. Sibling-based studies would profit from a somewhat more complicated design, in which newly born siblings of original sample children are added at birth, and the fresh cohort selection should be designed to overrepresent newborns without siblings in the original cohort range. Alternatively, one may wish to opt for a continuous survey design, with a new cohort of newborns added each year.

8.

Some consideration should be given to the possibility of adding on community-based intensive or observational studies of comparable samples of children. The national survey would provide the quantitative instruments and other procedures; local samples could supplement these data with richer process and contextual data.

9.

There should be a vigorous competition for design and field work among coalitions of researchers and survey organizations, both private and public. The process by which the Health and Retirement Survey was begun could serve as a model for this process.

Ideally, resources for studying children and families would be available to fully augment existing surveys and for fielding a new national survey of children. If sufficient resources are not available to meet both of these goals, we would give priority to the low-and moderate-cost augmentations we have outlined and dropping some or all of the high-cost augmentations in favor of a new national survey of children as described above. The actual costs associated with each of these proposals should be estimated to facilitate future decision making on these issues.

Acknowledgments

The authors shared equally in the preparation of this paper and are listed in alphabetical order. Work on the paper was supported by the workshop's sponsors and by the National Institute of Child Health and Human Development's Family and Child Well-Being Research Network, of which Brooks-Gunn, Duncan, and Moore are members. Two of the authors are affiliated with datasets reviewed in this paper: Moore with the National Survey of Children and Duncan with the Panel Study of Income Dynamics. In addition, Child Trends, Inc., and Brooks-Gunn are part of a consortium of organizations that have bid to design the Department of Education's Early Childhood Longitudinal Survey.

In writing this paper we have drawn freely upon a number of earlier efforts: Moore et al. (1994), Haveman and Wolfe (1994), Moore (1993), Watts and Hernandez (1982), Zill (1989), and Zill et al. (1984). We are indebted to Deborah Phillips for helping us conceptualize our task and to helpful comments from Don Hernandez, Gary Sandefur, Terry Adams, Dorothy Duncan, Valerie Lee, Susan Mayer, Robert Moffitt, Bruce Taylor, Diane Hansen, Dennis Carol, Susan Mayer, Jennifer Maddens, Michael Pergamit, Felicia LeClere, and the NICHD Research Network on Family and Child Well-Being.

References

• Alexander, K.L., D.R. Entwisle, and S.L. Dauber 1993. First-grade classroom behavior: Its short-and long-term consequences for school performance. Child Development 64(3):801–814. [PubMed: 8339696]

• Baydar, N. 1988. Effects of parental separation and re-entry into union on the emotional well-being of children. Journal of Marriage and the Family 50:967–981.

• Baydar. N., and J. Brooks-Gunn 1991. Effects of maternal employment and child-care arrangements in infancy on preschoolers' cognitive and behavioral outcomes: evidence from the children of the NLSY. Developmental Psychology 27(6):932–945.

• Baydar, N., J. Brooks-Gunn, and F.F. Furstenberg, Jr. 1993. Early warning signs of functional illiteracy: predictors in childhood and adolescence. Child Development 64(3):815, 829. [PubMed: 8339697]

• Belsky, J. 1984. The determinants of parenting: a process model. Child Development 55(1):83–96. [PubMed: 6705636]

• Berendes, H., editor; , S. Kessell, editor; , and S. Yaffe, editor. , eds. 1991. Advances in Low Birthweight: An Interactional Symposium . Washington. D.C.: National Center for Education in Maternal and Child Health.

• Bornstein, M., editor. , ed. 1995. Handbook of Parenting . Hillsdale, N.J.: Erlbaum.

• Brim, O.G., and J. Kagan 1980. Constancy and Change in Human Development . Cambridge, Mass.: Harvard University Press.

• Bronars, S., and J. Grogger 1992. The Economic Consequences of Teenage Childbearing: Results from a Natural Experiment. Paper presented at the National Institute for Child Health and Development Conference on outcomes of early childbearing, Bethesda, Md., May 1992.

• Bronfenbrenner, U. 1979. Contexts of child rearing: problems and prospects. American Psychologist 34:844–850.

• Brooks-Gunn, J. in press. Research on stepparenting families: integrating discipline approaches and informing policy. In A. Booth, editor; and J. Dunn, editor. , eds., StepParent Families with Children: Who Benefits and Who Does Not? Hillsdale, NJ: Erlbaum.

• 1990. Identifying the vulnerable young child. Pp. 104–124 in D.E. Rogers, editor; and E. Ginsberg, editor. , eds., Improving the Life Chances of Children at Risk. Boulder, Colo.: Westview Press.

• Brooks-Gunn, J., and P.L. Chase-Lansdale 1995. Adolescent parenthood. Pg. 10 in M. Bornstein, editor. , ed., Handbook of Parenting. Hillsdale, N.J.: Erlbaum.

• Brooks-Gunn, J., G.J. Duncan, P.K. Klebanov, and N. Sealand 1993. a. Do neighborhoods influence child and adolescent behavior? American Journal of Sociology 99(2):353–395.

• Brooks-Gunn, J., G. Guo, and F.F. Furstenberg, Jr. 1993. b. Who drops out of and who continues beyond high school?: a 20-year study of black youth. Journal of Research in Adolescence 3(37):271–294.

• Brooks-Gunn, J., P.K. Klebanov, F.R. Liaw, and D. Spiker 1993. c. Enhancing the development of low-birth-weight, premature infants: change in cognition and behavior over the first three years. Child Development 64(3): 736–753. [PubMed: 7687948]

• Brooks-Gunn, J., E. Phelps, and G.H. Elder 1991. Studying lives through time: secondary data analyses in developmental psychology. Developmental Psychology 27(6):899–910.

• Brooks-Gunn, J., and M. Weinraub 1983. Origins of infant intelligence testing. Pp. 25–66 in M. Lewis, editor. , ed., Origins of intelligence, 2nd Edition. New York: Plenum Press.

• Brown, B. 1990. The Effect of Neighborhood Characteristics on Teen Outcomes Related to Socioeconomic Attainment: In Search of the Underclass Neighborhood. Dissertation. University of Wisconsin, Madison.

• Chase-Lansdale, P.L., J. Brooks-Gunn, and E.S. Zamsky 1994. Young African-American multigenerational families in poverty: quality of mothering and grandmothering. Child Development 65(2):373–393. [PubMed: 8013228]

• Chase-Lansdale, P. L., F.L. Mott, J. Brooks-Gunn, and D.A. Phillips 1991. Children of the NLSY: a unique research opportunity. Developmental Psychology 27(6):918–931.

• Cherlin, A.J. 1991. On analyzing other people's data. Developmental Psychology 27(6):946–948.

• Cherlin, A.J., F.F. Furstenberg, Jr., P.L. Chase-Lansdale, K.E. Kiernan, P.K. Robins, and D.R. Morrison 1991. Longitudinal studies of effects of divorce on children in Great Britain and the United States. Science 252:1386–1389. [PubMed: 2047851]

• Child Trends 1993. Researching the Family: A Guide to Survey and Statistical Data on U.S. Families . Washington, D.C.: Child Trends, Inc.

• Clark, R. 1993. Neighborhood Effects on Dropping Out of School among Teenage Boys. Mimeo. Urban Institute, Washington, D.C.

• Coleman, J. 1988. Social capital in the creation of human capital. American Journal of Sociology 94:95–120.

• Committee on Ways and Means 1993. 1993 Green Book: Background Material and Data on Programs within the Jurisdiction of the Committee on Ways and Means . Washington, D.C.: U.S. Government Printing Office.

• Cowan, P.A., and C.P. Cowan 1990. Becoming a family: research and intervention. Pp. 1–51 in I. Sigel, editor; and A. Brody, editor. , eds., Family Research. Hillsdale, N.J.: Erlbaum.

• Crane, J. 1991. The epidemic theory of ghettos and neighborhood effects on dropping out and teenage childbearing. American Journal of Sociology 96(5):1126–1159.

• Currie, J., and D. Thomas forthcoming. Does Head Start make a difference? American Economic Review .

• Desai, S., P.L. Chase-Lansdale, and R.T. Michael 1989. Mother or market? Effects of maternal employment on cognitive development of 4-year-old children. Demography 26(4):545–561. [PubMed: 2583316]

• Deutsch, F.M., D.N. Ruble, A. Fleming, J. Brooks-Gunn, and C. Stangor 1988. Information-seeking and self-definition during the transition to motherhood. Journal of Personality and Social Psychology 55(3):420–431. [PubMed: 3171914]

• Dornbusch, S.M., L.P. Ritter, and L. Steinberg 1991. Community influences on the relation of family statuses to adolescent school performance: differences between African Americans and non-Hispanic whites. American Journal of Education 38(4):543–567.

• Dryfoos, J.G. 1990. Adolescents at Risk: Prevalence and Prevention . New York: Oxford University Press.

• Duncan, G.J. 1992. The economic environment of childhood. In Aletha Huston, editor. , ed., Children in Poverty. New York: Cambridge University Press. 1991. Made in heaven: secondary data analysis and interdisciplinary collaborators. Developmental Psychology 27(6):949–951.

• Duncan, G.J., J. Brooks-Gunn, and L. Aber 1994. a. Neighborhood Poverty: Context and Consequences for Development . New York: Russell Sage Foundation.

• Duncan, G.J., J. Brooks-Gunn, P.K. Klebanov, and P. Kato 1994. b. Economic deprivation and early-childhood development. Child Development 65(2):296–318. [PubMed: 7516849]

• Duncan, G.J., B. Gustafsson, R. Hauser, G. Schmauss, H. Messinger, R. Muffels, B. Nolan, and J. Ray 1993. Poverty dynamics in eight countries. Journal of Population Economics 6(3): 215–234.

• Duncan, G., and D. Hill 1989. Assessing the quality of household panel survey data: the case of the PSID. Journal of Business and Economic Statistics 7(4):441–451.

• Duncan, G.J., and W. Rodgers 1991. Has children's poverty become more persistent? American Sociological Review 56:538–550.

• Dunn, L.M., et al. 1981. Peabody Picture Vocabulary Test-Revised . Circle Pines, Minn.: American Guidance Services.

• Earls, F. 1992. Not fear, nor quarantine, but science: preparation for a decade of research to advance knowledge about the control of violence in youth. Pp. 104–121 in D.E. Rogers, editor; and E. Ginsberg, editor. , eds., Adolescents at Risk: Medical and Social Perspectives. Boulder, Colo.: Westview Press.

• Eccles, J.S. 1983. Expectancies, values and academic behaviors. In T.J. Spence, editor. , ed., Achievement and Achievement Motives: Psychological and Sociological Approaches . San Francisco: Freeman.

• Elliot, D.S., S. Huizing, and D.S. Agetin 1985. Explaining Delinquency and Drug Use . Beverly Hills, Calif.: Sage Publications.

• Feldman, S.S., editor; , and G. Elliott, editor. , eds. 1990. At the Threshold: The Developing Adolescent . Cambridge, Mass.: Harvard University Press.

• Furstenberg, F.F., Jr. 1976. Unplanned Parenthood: The Social Consequences of Teenage Childbearing . New York: The Free Press.

• Furstenberg, F.F., Jr., J. Brooks-Gunn, and S.P. Morgan 1987. Adolescent mothers and their children in later life. Family Planning Perspectives 19:142–151. [PubMed: 3678480]

• Garbarino, J. 1991. Not all bad developmental outcomes are the result of child abuse. Developmental Psychopathology 3:45–50.

• Garcia Coll, C.T. 1990. Developmental outcome of minority infants: a process-oriented look into our beginnings. Child Development 61(2):270–289. [PubMed: 2188805]

• Garfinkel, I., and S. McLanahan 1986. Single Mothers and their Children: A New American Dilemma . Washington, D.C.: Urban Institute Press.

• Garfinkel, I., editor; , S. McLanahan, editor; , and P. Robins, editor. , eds. 1994. Child Support Reform and Child Well-Being . Washington, D.C.: Urban Institute Press.

• Garner, C.L., and S.W. Raudenbush 1991. Neighborhood effects on educational attainment: a multilevel analysis. Sociology of Education 64:251–262.

• Geronimus, A.T., and S. Korenman 1992. The socioeconomic consequences of teen childbearing reconsidered. Quarterly Journal of Economics 107:1187–1214.

• Gottschalk, P. 1992. The intergenerational transmission of welfare participation, facts and possible causes. Journal of Policy Analysis and Management 11(2):254–272.

• Haveman, R., and B. Wolfe 1994. Succeeding Generations: On the Effects of Investments in Children . New York: Russell Sage Foundation.

• Hernandez, D.J. 1993. America's Children: Resources from Family, Government, and the Economy . New York: Russell Sage Foundation.

• Hetherington, E.M. 1993. An overview of the Virginia Longitudinal Study of Divorce and Remarriage: A focus on early adolescence. Journal of Family Psychology 7:39–56.

• Hetherington, E.M., editor; , and J.D. Arasteh, editor. , eds. 1988. Impact of Divorce, Single Parenting, and StepParenting on Children . Hillsdale, N.J.: Erlbaum.

• Hetherington, E.M., and W.G. Clingempeel 1992. Coping with marital transitions: a family systems perspective. Monographs of the Society for Research in Child Development 57(227):2–3.

• Hill, C.R. , and F.P. Stafford 1985. Parental care of children: time diary estimates of quantity, predictability and variety. Pp. 415–437 in F.T. Juster, editor; and F.P. Stafford, editor. , eds., Time, Goods, and Well-Being. Ann Arbor, Mich.: Institute for Social Research, Survey Research Center.

• 1980. Parental care of children: time diary estimates of quantity, predictability and variety. Journal of Human Resources 15:219–239.

• Hill, M. 1992. The Panel Study of Income Dynamics: A User's Guide . Beverly Hills, Calif.: Sage Publications.

• Hill, M., and G.J. Duncan 1987. Parental family income and the socioeconomic attainment of children. Social Science Research 16:39–73.

• Hofferth, S.L., and D.A. Phillips 1991. Child care policy research. Journal of Social Issues 47:1–13.

• Hoffman, S.D., E.M. Foster, and F.F. Furstenberg, Jr. 1993. Re-evaluating the cost of teenage childbearing. Demography 30:1–14. [PubMed: 8379973]

• Holmbeck. G.N., R.L. Paikoff, and J. Brooks-Gunn 1995. Parenting of adolescents. In M. Bornstein., editor. ed., Handbook of Parenting. Hillsdale, N.J.: Erlbaum.

• Hunt, J.M. 1961. Environment and Experience . New York: Roland Press.

• Huston, A., editor. , ed. 1992. Children in Poverty . New York: Cambridge University Press.

• Infant Health and Development Program 1990. Enhancing the outcomes of low-birth-weight, premature infants. Journal of American Medical Association 263:3035–3042. [PubMed: 2188023]

• Institute of Medicine 1985. Preventing Low Birthweight . Washington, D.C.: National Academy Press. [PubMed: 25009946]

• Jessor, R. 1992. Risk behavior in adolescents: a psychosocial framework for understanding and action. In D.E. Rogers, editor; , and E. Ginsberg, editor. , eds., Adolescents at Risk: Medical and Social Perspectives. Boulder, Colo.: Westview Press.

• Juster, F. Thomas, editor; , and F.P. Stafford, editor. , eds. 1985. Time, Goods and Well-being . Ann Arbor. Mich.: Institute for Social Research, Survey Research Center.

• Kalmuss, D., and K. Fennelly 1990. Barriers to prenatal care among low-income women in New York City. Family Planning Perspectives 22(5):215–218, 231. [PubMed: 2272380]

• Kellam, S.G., C.H. Brown, and M.E. Ensminger 1982. The long-term evolution of the family structure of teenage and older mothers. Journal of Marriage and the Family 44:539–554.

• Kiernan, K. 1992. The impact of family disruption in childhood on transitions made in young adult life. Population Studies 46:218–234.

• Klebanov, P.K., J. Brooks-Gunn, G.J. Duncan 1994. Does neighborhood and family affect mother's parenting, mental health, and social support. Journal of Marriage and the Family 56(2):441–455.

• Krysan, M., K.A. Moore, and N. Zill 1990. Research on Successful Families. A report on a conference sponsored by the Assistant Secretary for Planning and Evaluation, U.S. Department of Health and Human Services.

• Lamberty, G., editor; , and C. Garcia Coll, editor. , eds. 1994. The Reproductive Health of Puerto Rican Women Residing on the U.S. Mainland and the Growth Development of Their Children . New York: Plenum Press.

• Lazear, E.P., and R.T. Michael 1987. Allocation of income within the Household . Chicago: University of Chicago Press.

• Lee, V., J. Brooks-Gunn, E. Schnur, and T. Liaw 1990. Are Head Start effects sustained? A longitudinal comparison of disadvantaged children attending Head Start, no preschool, and other preschool programs. Child Development 61:495–507. [PubMed: 2344785]

• Lerner, J.V., editor; , and N.L. Galambos, editor. , eds. in press. The Employment of Mothers during the Childrearing Years . New York: Garland Press.

• Lerner, R.M., editor. , ed. 1984. On the Nature of Human Plasticity . New York: Cambridge University Press.

• Liaw, F., and Brooks-Gunn, J. 1994. Cumulative familial risks and low-birthweight children's cognitive and behavioral development. Journal of Clinical Child Psychology 23(4):360–372.

• Maccoby, E.E., and J.A. Martin 1983. Socialization in the context of the family: parent-child interaction. Pp. 1–102 in P.H. Mussen, editor; and E.M. Hetherington, editor. , eds., Handbook of Child Psychology: Socialization, Personality, and Social Development . New York: John Wiley and Sons.

• Maccoby, E.E., and R.H. Mnookin 1992. Dividing the Child: Social and Legal Dilemmas of Custody . Cambridge. Mass.: Harvard University Press.

• Mare, R.D. 1980. Social background and school continuation decisions. Journal of the American Statistical Association 75(370):295–305.

• McCormick, M.C., J. Brooks-Gunn, K. Workman-Daniels, J. Turner, J., and G. Peckham 1992. The health and development status of very low birth weight children at school age. Journal of American Medical Association 267:2204–2208. [PubMed: 1556798]

• McKey, R.H., L. Condelli, H. Granson, B. Barrett, C. McConkey, and M. Plantz 1985. The Impact of Head Start on Children, Families, and Communities . Final Report of Head Start Evaluation, Synthesis and Utilization.

• McLanahan, S., N.M. Astone, and N.F. Marks 1991. The role of mother-only families in reducing poverty. Pp. 51–78 in A.C. Huston, editor. , ed., Children in Poverty: Child Development and Public Policy. Cambridge, Mass.: Cambridge University Press.

• McLanahan, S., J.A. Seltzer, T.L. Hanson, and E. Thomson 1994. Child support enforcement and child well-being: greater security or greater conflict? In I. Garfinkel, editor; , S. McLanahan, editor; , and P. Robins, editor. , eds., Child Support Reform and Child Well-Being. Washington, D.C.: Urban Institute Press.

• McLoyd, V.C. 1990. The impact of economic hardship on black families and children: psychological distress, parenting, and socioemotional development. Child Development . 61:311–346. [PubMed: 2188806]

• Moffitt, R. 1992. Incentive effects of the U.S. welfare system, a review. Journal of Economic Literature 30(1):1–61.

• Moore, K.A. 1993. Children and Families: Data Needs in the Next Decade. Invited presentation given at the Interagency Family Data Workshop Group Meeting. Washington, D.C., May 25.

• 1986. Children of Teen Parents: Heterogeneity of Outcomes. Final Report to the Center for Population Research, National Institute of Child Health and Human Development, Department of Health and Human Services, grant # HD18427-02. May.

• Moore, K.A. and Peterson, J.L. 1989. Wave Three of the National Survey of Children: Description of Data. The Consequence of Teenage Pregnancy. Final report prepared under NICHD and ASPE/DHHS Grant #HD21537.

• Moore, K.A., and N. Snyder 1991. Cognitive attainment among firstborn children of adolescent mothers. American Sociological Review 56:612–624.

• Moore, K.A., D.R. Morrison, and D.A. Glei 1994. Welfare and adolescent sex: the effects of family history, benefit levels and community context. Journal of Family and Economic Issues .

• Murray, C. 1984. Losing Ground . New York: Basic Books.

• National Commission on Children 1991. Beyond Rhetoric: A New American Agenda for Children and Families . Washington, D.C.: National Commission on Children.

• Natriello, G., editor. , ed. 1987. School Dropouts: Patterns and Policies . New York: Teachers College Press.

• Nock, S.C., and A.W. Kingston 1988. Time with children: The impact of couples' work-time commitments. Social Forces 67:59–85.

• Palmer, J., editor; , T.M. Smeeding, editor; , and B.B. Torrey, editor. , eds. 1988. The Vulnerable . Washington, D.C.: Urban Institute Press.

• Park, R.E., E.W. Burgess, and R.D. McKenzie 1967. The City . Chicago: University of Chicago Press.

• Robins, L.N., J.L. Mills, J. Brooks-Gunn, and C. McCarthy 1993. Effects of in utero exposure to street drugs. American Journal of Public Health 83:1–32. [PMC free article: PMC1694390] [PubMed: 7504891]

• Rosenbaum, J.E., and S.J. Popkin 1991. Employment and earnings of low-income blacks who move to middle-class suburbs. Pp. 342–356 in C. Jencks, editor; and P.E. Peterson, editor. , eds., The Urban Underclass. Washington, D.C.: The Brookings Institution.

• Ruble, D.N., J. Brooks-Gunn, A.S. Fleming, G. Fitzmaurice, C. Stangor, and F. Deutsch 1990. Transition to motherhood and the self: measurement, stability, and change. Journal of Personality and Social Psychology 450–463. [PubMed: 2324937]

• Rush. D., Z. Stein, and M. Susser 1980. A randomized controlled trial of prenatal nutritional supplementation in New York City. Pediatrics 65:683–697. [PubMed: 6988785]

• Rutter, M. 1985. Family and school influences on cognitive development. Journal of Child Psychology and Psychiatry 26(5):683–704. [PubMed: 3900115]

• Shaw, C., and H. McKay 1942. Juvenile Delinquency and Urban Areas . Chicago: University of Chicago Press.

• Shaw, D.S., and R.E. Emery 1987. Parental conflict and other correlates of the adjustment of school-age children whose parents have separated. Journal of Abnormal Child Psychology 15(2):269–281. [PubMed: 3611524]

• Shereshefsky, P., and L. Yarrow 1973. Psychological Aspects of a First Pregnancy and Early Postnatal Adaptation . New York: Raven Press.

• Sigel, I.E., editor. , ed. 1985. Parental Belief Systems . Hillsdale, N.J.: Erlbaum.

• Simmons. R.G., and D.A. Blyth 1987. Moving into Adolescence: The Impact of Pubertal Change and School Context . New York: Aldine de Gruyter.

• Smeeding, T.M., and B.B. Torrey 1988. Poor children in rich countries. Science 242:873–877.

• Snow, C.E. 1983. Literacy and language: relationships during the preschool years. Harvard Educational Review 53(2):165–189.

• Spencer, M.B., G.K. Brookins, and W.R. Allen 1985. Beginnings: The Social and Affective Development of Black Children . Hillsdale, N.J.: Erlbaum.

• Spencer, M.B., and S. Dornbusch 1990. American minority adolescents. In S. Feldman, editor; and G. Elliot, editor. , eds., At the Threshold: The Developing Adolescent. Cambridge, Mass.: Harvard University Press.

• Stafford, F.P. 1987. Women's work, sibling competition, and children's school performance. American Economic Review 77:972–980.

• Stinnett, N., and Defrain 1985. Secrets of Straining Families . Boston: Little, Brown, and Co.

• Swithart, J. 1988. Characteristics of Strong Families. Unpublished paper, International Family Center, Logos Research Institute.

• Teachnsan, J.D. 1991. Who pays? receipt of child support in the United States. Journal of Marriage and the Family 53(3):759–772.

• Thomson, E., and S. McLanahan 1993. Family structure and child well-being: Economic resource versus parental behavior. Presented at the Annual meeting of the American Sociological Association. Washington, D.C., August.

• Timmer, S., J. Eccles, and K. O'Brien 1985. How children use time. In F. Thomas Juster, editor; , and F. Stafford, editor. , eds., Time, Goods and Well-being. Ann Arbor, Mich.: Institute for Social Research.

• U.S. Bureau of the Census 1989. Transitions in income and poverty status: 1984–1985. Current Population Reports, Series P-70. No. 15-RD-1. Washington, D.C.: U.S. Department of Commerce.

• U.S. Department of Health and Human Services 1992. Child Health USA '92 . DHHS Publication No. HRSA-MCH-92-6. Washington, D.C.: U.S. Government Printing Office.

• Watts, H.W., and D. Hernandez 1982. Child and Family Indicators: A Report with Recommendations . Report of the Advisory Group for Child and Family Indicators of the Advisory and Planning Committee on Social Indicators. Washington, D.C.: Social Science Research Council.

• Weinraub, M., and B.M. Wolf 1983. Effects of stress and social supports on mother-child interactions in single-and two-parent families. Child Development 54:1297–1311. [PubMed: 6354635]

• Werner, E.E., and R.S. Smith 1982. Vulnerable But Not Invincible: A Longitudinal Study of Resilient Children and Youth . New York: McGraw Hill.

• Wilson, J.B., D.T. Ellwood, and J. Brooks-Gunn 1995. Welfare to work through the eyes of children: the impact on parenting of movement from AFDC to employment. In P.L. Chase-Lansdale, editor; and J. Brooks-Gunn, editor. , eds., Escape from Poverty: What Makes a Difference for Children? New York: Cambridge University Press.

• Wilson, W.J. 1987. The Truly Disadvantaged: The Inner City, the Underclass, and Public Policy . Chicago: University of Chicago Press.

• Wirth, L. 1956. The Ghetto . Chicago: University of Chicago Press.

• Yamaguchi, K., and D.B. Kandel 1984. Patterns of drug use from adolescence to young adulthood: II. Sequences of progression. American Journal of Public Health 74(7):668–672. [PMC free article: PMC1651663] [PubMed: 6742252]

• Zaslow, M.J., K.A. Moore, N. Zill, and M.J. Coiro in press. Implications of the JOBS Program for Children . Washington, D.C.: Urban Institute Press.

• Zigler, E., and S. Muenchow 1992. Head Start: The Inside Story of America's Most Successful Educational Experiment . New York: Basic Books.

• Zill, N. 1983. Divorce, marital conflict, and children's mental health: research findings and policy recommendations. U.S. Senate, Committee on Labor and Human Resources. Broken Families: Hearings Before the Subcommittee on Family and Human Services, Washington, D.C.: U.S. Government Printing Office.

• Zill, N., editor. , ed. 1989. Research and Policy Uses of Federal Data on Children and Families. Recommendations from the Second Interagency Conference on Child and Family Statistics, Washington, D.C., March.

• Zill, N., J.L. Peterson, and K.A. Moore 1984. Improving national statistics on children, youth, and families. A report on recommendations made at the Interagency Conference on Child and Family Statistics, September.

1. There are several exciting new federal data collection efforts, currently in the planning stages, that are not reviewed in this paper. The Department of Education intends to fund an Early Childhood Longitudinal Educational Study. This is to be a large, school-based, nationally representative, longitudinal survey of kindergarten children. A second effort, recently funded by the National Institute of Child Health and Human Development, is a National Longitudinal Study of Adolescent Health. This survey has as its goal to provide a better understanding of the complex forces that promote good health and those that increase risk among the nation's adolescents.

2. Randomized trials allow for an estimation of the effects of a particular treatment, in this case a family or community resource. However, in many cases involving federal or state programs, it is impossible to conduct randomized trials (see, for example, the paucity of such research in the literature on Head Start and the Supplemental Food Program for Women, Infants, and Children: McKey, 1985; Zigler and Muenchow, 1992; Lee et al., 1990; Rush et al., 1980). In addition, when community-level resources are the target of intervention, randomized trials are often not appropriate (i.e., randomizing communities is difficult, since the sample size is based on communities, not individuals, and communities are usually not comparable on all the possible dimensions of interest).

3. In an analysis of the determinants of cognitive test scores of 3–7-year-olds, Moore and Snyder found that mother's education and poverty status were not significant predictors once controls for mother's score on the Armed Forces Qualification Test and measures of the home environment (as measured by the HOME scale), which were highly significant, were included.

4. Both Brooks-Gunn et al. (1993a) and Clark (1993) find that measures of affluent neighbors are more important than measures of low-income neighbors. Crane (1991) interprets his neighborhood measure (the percentage of workers in professional or managerial occupations) in terms of epidemic models, but it clearly measures the presence or absence of affluent neighbors. In one exception, Brown (1990), using the 1970 U.S. Census Neighborhood PUMS file, found evidence consistent with an ''underclass neighborhood'' or epidemic hypothesis for both white and black females when looking at teen nonmarital births and for white females when looking at high school completion. Evidence for white and black males regarding high school completion and idleness was not consistent with such hypotheses, however.

5. It is considerably more expensive to geocode addresses beyond wave 1 in longitudinal studies, since address matching would be involved and it is impossible to match addresses to census geocodes without at least some tedious map work. There would he substantial value in this additional geocoding. But the greatest value, especially given its low cost, would be in geocoding the wave 1 addresses and matching STF3 census data to these addresses.

6. One valuable piece of neighborhood information that could be distributed more widely is a scrambled version of the tract/BNA identifier. Clustered samples typically select several families per block (or adjacent blocks). It is analytically very useful to be able to sort children into groups—(1) same family (i.e., siblings), same neighborhood; (2) different family, same neighborhood; and (3) different family, different neighborhood—even if the actual characteristics of the neighborhoods are not known. These groups form the basis of an analysis-of-variance type of accounting of family and neighborhood effects. To perform this kind of analysis, one need not know any of the actual characteristics of the neighborhoods, but only whether survey families share the same neighborhood. Although this analysis-of-variance accounting capability would be quite useful analytically, it is less valuable than and no substitute for the actual decennial census measures matched to the family-and individual-level survey data.

7. An example of the importance of studying different family structures as well as the processes within families is the well-documented fact that the movement from a traditional family structure to a single-parent structure (divorce) results in less optimal functioning across domains. However, the effect is not just due to a single-parent household, in that remarriage, and the entrance of a stepparent, do not alter substantially the well-being of children (Hetherington, 1993; McLanahan et al., 1991; Garfinkel and McLanahan, 1986; Kiernan, 1992).

8. For more detailed descriptions of each survey, consult Child Trends (1993).

9. Possible national sampling frames include census-based dwellings and list-based schools. The former provides national samples of noninstitutionalized families and children; the latter national samples of school-age children. The former clusters children within neighborhoods and families; the latter within schools. As already noted, clustering provides analytic advantages in the form of sibling, neighbor, and classmate comparisons. Choice of sampling frame depends on the value of covering preschool children (which is part of dwelling but not school-based frames) as well as the comparative advantages of clustering by siblings and neighbors versus classmates. Although we see some arguments for school-based clustering, there is greater analytic value in clustering by family and neighbor and in coverage of preschool children.

10. A new cohort of the NLSY will be fielded in spring 1996. All adolescents ages 12–17 in the household will be interviewed and followed over time; however. there are no plans to interview children age 11 or younger.

Jeanne Brooks-Gunn is at the Center for the Study of Children and Families, Teachers College. Columbia University. Brett Brown and Kristin Anderson Moore are at Child Trends. Inc., Washington, D.C. Greg J. Duncan is at the Center for Urban Affairs and Policy Research, Northwestern University.

What is the context of family?

What is a concept of family?

What are the 3 main factors that have changed the form of families?

What are the factors of family?