Family Formation Processes: Assessing the Need for a New Nationally Representative Household Panel Survey in the United States

Abstract

The American family has undergone rapid transformation. Careful measurement attention to family formation is important because families are at the heart of numerous decisions, roles, and responsibilities with implications for understanding the well-being of families, adults and children. This paper considers whether there is a need for a new household panel study that addresses family formation. This paper consists of a review of the recent body of population-based, American surveys and finds a considerable gap in the ability to study the implications of families for the health and well-being of Americans. Earlier panel surveys used to assess family life anchored questions around marital events, but changes in family patterns require attention to a more diverse set of family forms. The paper concludes with recommendations for a multi-purpose panel study. The key challenge is to keep to pace with complexity and changes in American family life while at the same time maintaining a parsimonious set of survey questions.

The study of family formation processes (parenthood, marriage, cohabitation, re-partnering) in a new panel household survey presents several new opportunities as well as challenges. Earlier panels used to assess American family life anchored questions around marital events, but changes in family patterns require broadening our attention to diverse family forms. Surveys have broadened their focus to include a broader spectrum of family relationships as well as family ties that cross household boundaries. Careful measurement attention is important because families are at the heart of numerous decisions, roles, and responsibilities with implications for understanding of the well-being of families, adults and children. The key challenge is to keep to pace with complexity and changes in American family life while at the same time maintaining a parsimonious set of survey questions.

There are five key topics that are reviewed in this paper. First I present why the study of family formation processes is important including the key scientific and policy questions. Second, I speculate on whether a new panel household survey is the best venue to address these key questions. Third, I evaluate several standard approaches and assess what data should be collected. Fourth, I present several innovations on the horizon to cover these topics. Finally, I offer a rationalization for a new household panel survey. I have drawn heavily on findings of the Counting Couples, Counting Families final report [14] and associated research papers [69] along with the broader literature. Important reviews of family measurement include Hofferth and Casper [44] and the Federal Interagency Forum on Child and Family Statistics [28]. The National Center for Family and Marriage Research has several crosswalks that show how items are measured across surveys. As a reference, Appendix 1 lists the primary U.S. surveys with details about year of administration and sample selection. It is important to extend our review beyond the United States data infrastructure and reflect on the experiences of some of the most successful panel collections in the United Kingdom and Germany [16, 39].

Why is the study of family formation processes significant?

The measurement of family transitions, family instability, family structure, and family formation are necessary to assess child and adult well-being. The family is the primary source of care and socialization of children, the basic consuming unit, a safety net for young adults, and the preferred provider of care for older adults. To assess the implications of families for well-being an understanding of who is in the family, the relationships of family members to one another, and how they are spending time and sharing resources is required.

In addition, the accurate measurement of family life is necessary to assess policy interventions targeted at providing resources to support those in need. Family units are important in policy work because resources meant for children are provided to the parents. Access and the level of support provided to children and adults provided by the government in some cases is determined by the definition of family. Policies may also encourage family ties across households, such as child support and health insurance to adult children. Further, a few policies are aimed at changing family behavior (e.g., teen pregnancy or marriage) and as a result require accurate measurement of those family behaviors.

Traditional approaches to study American family life do not reflect the contemporary complexity of families faced by children and adults today. There is some debate about whether the American family has become more complex over the last twenty years or whether researchers have developed better measures to tap family complexity. Nonetheless, the plateau in family change but growth in family instability has been characterized as sustained diversity [34, 61]. A careful review of all the main family changes are beyond the scope of this paper, but key indicators of family complexity include the following: approaching half of children are born to unmarried parents, the age at marriage is at a historic high point, a stable and high divorce rate persists, an increase in cohabitation continues among young adults and older Americans along with growing numbers of cohabiting partners, a substantial share of parents are having children with multiple partners, a persistently high number of family transitions, and continual repartnering and remarriage [e.g., 22, 37,48, 52, 57, 100]. Research requires measurement of family life that captures these patterns and trends and a household panel provides opportunities to study the implications of the full range of family experiences on adult and child well-being.

There has been a growing gap in the family experiences of the most advantaged and least advantaged which has been labeled diverging destinies [65]. Family change and complexity is especially pronounced among the most disadvantaged groups and will probably persist as economic circumstances do not improve [59]. Thus, questions about the ways families respond to economic distress are important. Further, numerous questions about the implications of employment changes are key as families require two earners rather than one to make ends meet. The recent recession has had serious implications for Americans ability to form and maintain families and may further the social class divide in family life. Thus, shifts in family formation are not uniform across social class and race/ethnicity requiring surveys to include large numbers of respondents of varying social classes as well as race/ethnicity and nativity status. Further the family context can help or hinder intergenerational economic mobility in terms of social class. Recent evidence using the PSID shows the deleterious impact of divorce on mobility [25]. A new household panel will provide opportunities to track the implications of a growing divide in family formation patterns in the U.S. and trace the ability of Americans to move beyond their childhood social class.

The progression from adolescence into adulthood is not a linear track from dating, engagement to marriage [35, 78]. New research should capture the full range of relationships based on dating or romantic patterns, sexual intercourse, residence, and legal recognition of relationships. Increase in the age at marriage and the more variable order of life course events create for many an extended period of nonmarital romantic involvement in young adulthood [2, 83]. This is an important topic because these relationships set the groundwork for subsequent relationship trajectories and have implications for psychological well-being, physical health, and economic and educational attainment. In addition, nonmarital relationships are increasingly the setting for family formation so ignoring these relationships provides a narrow lens on family life. Further, dating and sexual relationships do not end in young adulthood and continue through middle and old age. Panel data that is not limited to one point in the life course allows research to address how these relationships are important contexts for support and well-being across the life course.

The broadening legal recognition of same-sex relationships across local areas and states along with federal court decisions means that social science surveys need to keep pace with changing definitions of same-sex partnerships [32]. To date about three-quarters of the U.S. population lives in a state that offers legal recognition of same sex couple marriages [3]. The number of states offering legal same-sex marriage has sky rocketed the Supreme Court will hear challenges to DOMA and Circuit Court rulings this year. The U.S. Census has proposed new measurement of same-sex marriage as current strategies are not adequately capturing same-sex marriages [99]. New surveys need to include the full range of sexual relationships of single and coupled gays, lesbians, transgender men and women, and bisexuals by broadening surveys to include indicators of sexual orientation and gender identity. Panel data that includes sufficient sample sizes of LBGT respondents will permit researchers to examine the pathways to family formation and the health of an understudied minority population. In addition, questions about the well-being of children raised in same-sex parent families are best answered with longitudinal data.

An important source of contributions to the health and well-being of children and adults are family members living outside of the household. Families often cross household boundaries, for example in 2009 40% of children were not living with both biological parents [52]. To assess child well-being it is important to understand the levels of social contact and financial support by nonresident parents, but surveys have varied in their success in identifying nonresident fathers measurement [30, 90]. In addition, support to young adults as well as aging parents is often provided by family members living outside the household. Surveys that are limited to family roles and responsibilities within the household are missing an important and critical source of support [43, 81].

Can a new panel household survey address these issues?

A new panel household survey is not necessary to address basic descriptive trends or a profile of American families. A new panel is necessary to best address the impact of contemporary family life on health and well-being as presented above. However, decisions about the best way to measure and assess family patterns depend to a large extent on the specific scientific goal of the study. Some studies are quite general, i.e., the National Longitudinal Study of Adolescent Health (Add Health), while others serve specific purposes, e.g., to study education and child development (Early Childhood Longitudinal Study, ECLS-K) or consumer expenditure patterns (CE). Scientific priorities need to be established to then determine the optimal means and methods of family measurement.

One challenge is the level of detail about family patterns. Certainly a large number of categories could be developed to measure family experience, but the categories need to be meaningful and large enough to remain substantively informative [18, 41]. A balance is required in the level of detail and parsimony in the measurement of family life. A panel study provides an opportunity to monitor social change so repetition of questions rather than changing measurement is important.

The sampling required to capture some family experiences that are meaningful, such as same-sex parent families, may present challenges. The survey needs to be sufficiently sized to measure some family forms that are relatively rare but of great public interest. Disadvantaged groups are harder to interview and research has shown that past household surveys frequently undercounted young, disadvantaged men [64]. Immigrant groups present special challenges as they may question providing details to authorities about their legal status. Further, the policy realm typically requires samples that permit local and state level analyses. Large samples or targeted samples are often required to capture the population of interest, typically low income respondents or those with social welfare program experiences.

There certainly appears to be a place for a new panel household survey. Prior panel studies are specific to a birth cohort (e.g., Add Health, HRS, NLSY79, NLSY97) so results cannot be generalized and reference a single age group. The SIPP follows respondents for a relatively short time window and does not provide a long time span. The PSID is unique by providing nearly 50 years of data and following family members across households, but is limited in terms of immigrant groups and sample sizes of subpopulations. Cross-sectional data collections provide an important perspective and reflect a broad understanding of family patterns at a single point in time. However, current cross-sectional data collections that focus on the family are limited to specific age groups (e.g., NSFG is limited to 15–44 year olds) or have not been recently collected (e.g., NSFH data were collected over 25 years ago).

What data should be collected to assess family formation?

Household and family membership

The traditional way household membership is measured is by determining how every household member is related to the ‘head,’ the household head approach. This method provides a limited lens on family relationships and prevents accurate measurement of subfamilies, stepparents, and half and stepsiblings. For example, analysis of the SIPP indicates that 11% of children are living with a parent who is not the household head [60]. Rosters work well to capture stable families where parents have not had children with other partners. It also relies on proxy measures of household members’ family experiences (e.g., nonresident children). The consensus is that the use of household roster data to measure stepfamilies is problematic [88]. Accordingly, household roster approaches have been characterized as “outmoded and should be replaced” (p.176) [9].

A better strategy is to follow the lead of Census Bureau surveys, SIPP and CPS, and include child pointers to identify the parents of children in the household. In addition, the SIPP and CPS incorporate relationship pointers to determine the marital and cohabitation relationships of household members to one another.

Several studies have directed questions about relationships of household members to a specific child (ECLS-B, Add Health, NSLY-97). Hofferth and colleagues [45] drew on the PSID Child Development Survey which randomly selected a focal child and established whether he/she lives with their father, mother, stepmother, or stepfather. This method is more cost-effective than selecting all children, but it is limited to one child’s perspective. It also does not ask the child about relationships to all household members. The new PSID Child Development Survey will interview all children allowing assessments of full family relationships.

A full portrayal of household membership can be gained by asking for household relationship matrices. In other nations, the matrices have been included in census data collections and have been recommended by the UN Economic Commission for Europe Statistical Division [98] especially if there is an interest in ‘reconstituted’ or stepfamilies. The matrices have been instituted in census data collected in several countries including Canada, Australia, Great Britain, Estonia, and New Zealand. Typically, a single “main” respondent is selected to provide proxy information on the relationship of all household members to one another. These are important because they identify not only relationships to the head but relationships to one another, including subfamilies and sibling relationships that would not be captured with traditional strategies. The SIPP is the only nationally representative survey to include full matrices and has been used to identify family complexity that cannot be identified with other strategies. For example, 11% of children lived with step or half siblings and 16% live in a stepfamily household [52]. The primary flaw with relationship matrix data and direct questions about household membership is that it represents just one person’s view of the relationships in the household. A further limitation is that this method does not recognize relationships that extend beyond traditional household boundaries, which have become more common and are increasingly important for our understandings of family life (e.g., children who reside part-time or for short periods of time).

It is often assumed that there is one family per households or that families do not live with extended kin. Household complexity (i.e., residence with extended family members) is growing, one-fifth of children live in extended households [52], but little work has assessed how these household living arrangements influence well-being. Some work has examined co-residence with grandparents, about 7% of children in 2010 live with a grandparent but typically with a parent and grandparent [103]. In these complex households it is important to establish who is the head of household. Also, further work is needed to distinguish between short-term spells of longer term living arrangements.

An important issue to consider is the identification of the primary respondent. Typically, one family member responds to all questions about family members. Yet, definitions of family membership depend on who is asked. For example, two children in a family may be living in two different family types as in the case of stepsiblings one child may be living in a two biological married parent family while the other is living in a stepfamily. Family living arrangements are based on the relationships of children to adults as well as children’s relationships to other children which can be used to establish complexity at both the parental (biological, step, adopted) and sibling (biological, half, step) level. Further, child and adult responses to family membership are not uniformly consistent [13]. Finally, not all members of the family are equally knowledgeable about the family histories and experiences of all household members (e.g., nonresident parenthood or dates of union formation and dissolution) [90].

Marriage and Cohabitation

Social science research has established ways to measure respondent’s marriage and cohabitation. Virtually all national surveys include questions used to identify current marital and cohabitation status. These are often established using household rosters with questions about the marital status (married, divorced, separated, widowed). These are legal statuses and seem relatively straightforward. Yet, evidence from ACS shows some reported divorces are not in fact legal and are separations [71]. Among women who have experienced a first divorce in the past year, half spent about 7 months separated prior to divorce; the average time between separation and divorce for women is 22 months [74]. Further, separation and divorce patterns differ according to socioeconomic status presenting challenges in the measurement of the end of relationships.

As a growing set of states legalize same-sex marriage there are opportunities to capture the full range of legal definitions of relationships (civil unions, registered partnerships, designated beneficiaries, municipalities recognition, and marriage). New plans by the census to add direct indicators of same-sex marriage will be implemented in the next few years [99]. Recent evidence indicates that the current approach (same-sex couples self identify as legally married) is resulting in false positives with many opposite-sex couples incorrectly filling out gender in the census forms [56]. Even though it may be challenging it is important to accurately measure legal status of relationships for all respondents in surveys as this legal status confers many important social and economic benefits.

Most surveys include indicators of current cohabitation status. The most common method is to rely on a household roster and the term “unmarried partner”, but provides limited utility [84]. Unmarried partner is typically at the end of the list of relationship options near roommates and unrelated others, not immediately following spouse. New plans by the Census have moved unmarried partner to follow the spouse question and add opposite-sex and same-sex to the items [99]. In one recent study, Gates [31] finds that 10% of individuals cohabiting with a same-sex partner recalled having described their relationship as roommates or other non-relatives on the Census 2010 form. Qualitative evidence suggested that unmarried partner was not a term that always resonates with cohabiting couples [62] and surveys include a range of definitions including boyfriend or girlfriend to bring a relational element into the question. The Add Health’s wave 4 relationship roster included “partner, boyfriend” and “partner, girlfriend.” Some surveys include direct questions about cohabitation referencing specifically different sex and not same-sex relationships. For example, “Not married but living together with a partner of opposite sex” (NSFG) or “opposite sex romantic partner” (PSID), while the NLYS97 has excluded the “opposite sex” portion of the question. Research should include questions about both different-sex and same-sex couples. As mentioned above, in the CPS there is a roster as well as a direct question, “Do you have a boyfriend, girlfriend, or partner in this household?”

Krieder [51] assessed discrepancies in the measurement of cohabitation drawing on direct questions about cohabitation and a household roster and found 20% difference in the share of cohabiting couples in the CPS. Similarly, Kennedy and Fitch [47] find that estimates of cohabitation based on rosters undercount cohabiting unions by 18% and miss 12% of children living in cohabiting parent families. A few surveys have used the term ‘marriage-like’ to describe cohabiting relationships (Add Health, ECLS-K, NLSY97) and the NLSY97 continues to rely on this definition. This is problematic because it references marriage and many cohabiting unions may not be on the pathway to marriage. Some surveys have included the number of nights in the home and “cohabiting all or most the time” or “cohabiting some of the time” (FFCWB). In fact, about 15% of couples have part-time cohabitation experiences that make it hard to determine whether it is, or is not, occurring at the time of interview [50]. Some surveys include a time reference to count a cohabiting couple ranging from one month to at least a year (NLSY79; PSID). New surveys are attempting to assess the meaning of cohabitation by examining enmeshment in terms of having other residences, pooling of income, having keys, or even where they are getting mail or keeping belongings [5, 49].

Family histories: Prior to interview and in the panel

In panel data, assessment of family structure at time of birth and each survey year along with high quality retrospective histories are necessary to capture full family transitions for adults and children. Family histories need to include the union status (cohabitation and marriage) along with biological relationship (step, biological, and adopted) and sex of parents. Retrospective data are necessary that reference the respondent as well as their spouse/partner. For example, couple based marital histories are necessary to establish whether a marriage is the first for both members of the couple. The fertility histories need to measure not only the dates of birth and number of children, but the relationships of respondents to the mother/father of the child. This level of refinement is necessary to establish multiple partner fertility (see below). The combination of marital and cohabitation histories along with fertility histories allow the assessment of family instability. Family instability includes types of family changes, number of changes, developmental stage of change, duration of family types [41]. This type of full history data is necessary to answer questions such as what share of children’s lives are spent in specific family types.

Typically marital histories are included in surveys and less often cohabitation histories, but our ability to analyze relationship instability across the population is limited [76]. The SIPP provides complete marriage histories, the NSFG provides an overview of men and women ages 15–44, and the ACS provides data on divorce over a 12 month period. The age cap of the NSFG prevents a comprehensive assessment of repartnering, with most recent data based on the ACS or SIPP. Once Americans divorce, they typically do not stay single [91] and about one in three (30%) recent marriages were remarriages [24]. Marital histories are included in specific birth cohort studies (Add Health, NLSY79, NSLY97, HRS) or have been added to panel data in supplements (PSID).

The divorce estimates across the SIPP and ACS are similar providing confidence about the quality of data across surveys [17, 76]. The ACS focuses on divorce and not separation which does not completely reflect marital dissolution. The length of time from separation to divorce is longer for more disadvantaged subgroups and our assessments of the level and timing of marital instability depend on whether date of separation or divorce is employed [8, 17]. It has been established in the marital literature that about one-third of those who separate reconcile [7] suggesting there may be some relationship churning and could have implications for well-being. Attention to the dates of separation and divorce are warranted.

Cohabitation histories are critical to assess children’s family histories as well as the relationship histories of adults. For example, excluding children’s experiences with cohabitation from instability measures undercounts instability by 30% among White children and 100% among Black children [77]. Further, cohabitation transitions are consequential in later life [12]. Cohabitation histories are available in cross-sectional surveys such as the NSFG and are available in cohort based studies (NLSY97, Add Health). Yet even new surveys in the field are not including cohabitation history data (PSID, SIPP). While panel data follow cohabitation during the course of the panel they lack data on date of current cohabitation (SIPP) as well as information on cohabitation prior to the interview (SIPP, PSID).

Definite beginning and end dates of cohabitation are difficult to identify in all major surveys. Couple level data in the NSFH show cohabiting couples offer different months as start dates for their relationships reflecting differences in the cohabitation process by each person in the relationship [62]. Evidence suggests that the quality of retrospective reports of cohabitation weaken with time since the event [42]. Teitler et al. [93] reported systematic bias in the retrospective reporting of unmarried parents’ cohabiting status based in part on relationship quality and relationship trajectory. Serial cohabitation is cohabitation with different partners. As the age at marriage increases and the age at cohabitation remains steady there are increasing opportunities for serial cohabitation [23, 53]. Technically these cohabitation histories could be applied to same-sex or different-sex relationships, except in the case of the NSFG which continues to ask about “opposite-sex” cohabiting unions. Further, new evidence speaks to ‘cyclical cohabitation’ or relationship churning getting back together with prior partners [40, 70, 101]. Across surveys it appears that at least one-fifth of cohabitors have experienced churning. Despite the potential complications, full cohabitation histories are an important indicator of family experiences for adults and children.

Sexual and romantic relationships: Non-coresidential

The share of Americans living alone has risen and has hit a 60-year high point of 14% [75]. Yet living alone does not mean that individuals are not engaging in sexual and dating relationships. These relationships have been largely ignored in prior work treating all ‘single’ women or men as equivalent. At times dating relationships in adulthood have been treated as equivalent to adolescent fleeting relationships without much thought to their implications for adult development and well-being. Indeed views of the end of dating have arisen in part due to attention to the ‘hookup’ culture characterized by a succession of sexual liaisons lacking intimacy and investment. Yet, evidence shows that in contrast to adolescence the early adulthood dating relationships are characterized by increases in feelings associated with romantic love and emotional rewards along with greater instrumental rewards and support [35]. In fact the levels of love and emotional rewards among dating and cohabiting couples appear similar [11, 35]. New relationship forms have emerged “living apart together” which are long-term committed couple relationships who do not live together because of geographic separation or preferences for independence. Estimates range from 7% to 12% of adults were involved in LAT relationships [89].

Certainly, decisions to cohabit and marry stem from some sort of relationship basis [62, 78]. In addition, a steady share of children (18%) are born to ‘single’ mothers who are not cohabiting or married [60]. Researchers have categorized those relationships as ‘visiting’ or ‘romantically involved and living apart’ using the FFCWB [20]. Some surveys have used the term ‘romantic relationship’ (Add Health, FFCWB) to describe dating or sexual relationships and the NSLY strategy has been to reference ‘dating’ relationships. The NLSY surveys include data to determine the progression of on-going dating relationships. Data collections should distinguish same- and different-sex couples as has been pursued in the NLSY97.

The focus has been on young adult dating and sexual relationships, but there is growing attention to these non-coresidential relationships among older Americans. The sustained high divorce rates along with widowhood indicate that middle age and older Americans have opportunities to experience dating and sexual relationships. In fact, one-third of baby-boomers are unmarried [54]. Research relying on the National Social Life, Health and Aging Project (NSHAP) have provided opportunities to assess relationships among older Americans with about 14% of single older Americans are in a dating relationship [15] and older Americans remaining sexual active [102]. Research needs to allow comparisons of early and later life dating and sexual relationships.

Measurement of sexual and dating relationships are challenging as researchers often try to put bounds around the relationships analogous to marriage or cohabitation. The fundamental challenge is that there are blurred lines surrounding the starts and endings of dating and sexual relationships. Halpern-Meekin et al. [40] reported that two-fifths of young adults who broke up reported having sex with an ex extending the relationship beyond the dating boundaries. Further, relationship churning (break up and reconciliation) was experienced by two-fifths of daters in young adulthood. A sizeable share of casual sex partners in early adulthood are with individuals they had formerly dated [35]. Flux in relationships can also be captured by asking about the number of nights couples spent together showing the range in couples time spent together [5]. There are concurrent relationships with issues surrounding sexual non-exclusivity [95] reflecting the reality that many relationships are overlapping and do not follow in a sequential manner. Given the flux in relationships a focus on current relationships would result in selection suggesting attention to current or most recent relationships. Panel data would avoid the problems associated with retrospective reports of these relationships. Recognition of the blurred lines is important and applying a simple marriage measurement model to dating and sexual relationships at any point of the life course do not seem appropriate. Relying simply on duration of a relationship to reflect the salience of it is inappropriate in the current context.

Relationship Quality

Relationship quality has implications for the stability of relationships as well as key predictors of child and adult well-being. From a policy perspective the emphasis has been on personal safety in terms of physical aggression and more recently the federal government has highlighted the importance of ‘healthy’ relationships with large investments in relationship education programs. Most large-scale data collections include some indicator of relationship quality. A snapshot of measurement of relationship quality for eight surveys is provided at http://www.bgsu.edu/ncfmr/resources/data/measures-snapshots/relationship-quality.html. Panel data is well suited to assess the implications of relationship quality and permits assessments of how relationships unfold over time. The tension in assessments of quality is often providing sufficient scope to capture positive and negative qualities with criticisms levied at the limited range and depth of indicators. Tinkew et al. [94] and Stanley [87] provide excellent reviews. Psychologists traditionally include multi-item indicators with attention to psychometrics and established scales resulting in a more in-depth measurement focus than other social scientists [87]. The established or “named” scales are at times copyrighted generating barriers to their wide use. Traditional domains include love, satisfaction, happiness, conflict, physical aggression, trust, future orientation, stress, and exclusivity. More refined assessments often include specific indicators of power and control, intimate disclosure, commitment, dedication, forgiveness, positive connections, and disillusionment [87]. There are calls in the field for new developments that capture the specific qualities and nature of nonmarital relationships as well as ensuring that measures are appropriate and reliable for a range of subgroups including racial minorities, sexual minorities, parents, and institutionalized populations.

Fertility and Childhood Experiences

Most surveys include fertility histories to establish the timing of entry into parenthood along with the age and number of children. The household rosters establish who is living in the household, but not all children born to parents are living with their parents (older children or nonresident children) so a different set of questions are asked about fertility behavior. Parallel information on the dates of marriage and cohabitation provide information on the childhood residence experiences as provided in data sets such as the NSFG. Not only are the dates of children’s births important, but coresidence with the biological father or mother is necessary to establish whether children were raised by biological parents. Data on adoption and fostering of children is also important to determine the length of time children have coresided with their parents. One critical issue is the quality of reporting from men and women on fertility histories. Analysis of the NSFG suggest that about 20% of all births, and particularly nonmarital births, are under-reported by males [46]. The quality of male fertility data depends to some extent on the questionnaire design as questions anchored around romantic partnerships resulted in higher quality of male fertility data [46, 55].

It is not only important to establish the timing or union context of births but whether a birth was intended. A key indicator of health and well-being for children, parents, and families is whether a birth was intended. Unintended fertility is recognized as a public health issue and has remain relatively unchanged since the 1980s with one in three births unintended [67]. Differentials in fertility according to education are due to unintended fertility [68]. Intentions are typically based on retrospective measures asked of mothers about each birth establishing whether she wanted to get pregnant at that particular time, wanted to get pregnant but not at that time (sometimes distinguishing a time period of two years), did not want to get pregnant, or did not care or know how she felt about getting pregnant. These items will help tap trends in unintended fertility and establish differentials in the patterning of unintended fertility. A panel design would avoid some of the retrospective issues associated with the measurement of fertility intentions [38].

Parents may have children with more than one partner which is referenced as multiple partner fertility. While this term is relatively new, it is not a new phenomenon in that parents have often gone on to have children with new partners once a relationship has ended. Multiple partner fertility is a relevant issue because it represents a form of family complexity with implications for investments in children and well-being (e.g., 19, 37). The assessments of multiple partner fertility have relied on relied on administrative data as well as reports from only men or both men and women [37]. At times MPF is based on partnership histories, which do not specify whether this was a partnership with a former biological mother/father. The bar is high to obtain data necessary to establish multiple partner fertility requiring information on children living within and outside the household. The NSFG provides estimates of men’s MPF but not women’s experiences [37]. Some cohort studies provide available data but are limited to relatively young parents who have not completed their childbearing years (FFCWB, Add Health, NLSY97). Panel data focusing on older women (NLSY79) have established MPF among women who have completed their childbearing [26]. New data collections are including direct questions asking parents whether their children have the same biological parents and to group their children by shared parents (e.g., SIPP). This strategy allows for more direct identification of multiple partner fertility.

Once the family context of the birth has been established the key question is to trace the children’s family experiences. The fundamental issue is in what family contexts are children being raised. This requires establishing the marital and cohabitation histories of parents and mapping those experiences on children’s life course. Indicators of family instability from the child’s vantage point requires detailed reports of parental coresidence as well as the coresidence with other adults (cohabiting or married stepparents, grandparents, adult siblings) as well as children (stepsiblings, half-siblings). Panel data are uniquely able to draw out these living arrangement changes to establish indicators of age at family experiences and share of childhood spent in specific family circumstances.

Nonresident Family Members

Surveys do a better job capturing household residents than nonresidents [43, 81], but these relationships can be quite consequential. Researchers have focused on children under age 18 with complications emerging based on the growth in joint custody as well as informal arrangements of children spending time across households [5, 80]. Questions asked about residence of children can be established in fertility histories and household rosters to determine whether respondents are nonresident parents. The quality of data on nonresident fatherhood depends on the household reporter or individual based sampling strategy as well as the order of questions [86, 90]. Custodial mothers more often report a child with a nonresident parent than men report being noncustodial parents [90]. Surveys relying on a household head to identify nonresident fathers report lower levels of nonresident fatherhood in part because about 60% of the time in SIPP or CPS the head is providing proxy reports of nonresident fatherhood. Surveys targeted at individuals (i.e., NSFG) produce considerably higher estimates of nonresident fatherhood. Certainly, the assessments of financial and social ties between nonresident children and their parents depend on our ability to measure nonresident parents.

Relationships of adults to one another across household boundaries are key sources of social and instrumental support [43, 81]. Growing attention to young adults moving in and out of their parental homes showcases the challenges of relying on cross-sectional rather than ‘ever’ reports of parent-adult child coresidence [29, 73]. Panel data that measure these short term living arrangements such as the NSLY97 provide estimates of boomeranging and launching of adult children [72]. A data set that is well suited to questions about links across households is the PSID as the PSID follows samples members who leave home offering opportunities to study the universe of potential nonhousehold family members with whom exchange can occur [43]. A new innovation in the PSID data is the family roster and transfer module available in 2013 for respondents with parents and children over age 18 living outside the allowing for the construction of complete networks of parents and adult children [6]. These relationships across households are available in some of the cohort studies (HRS) but do not provide a complete assessment.

Innovative approaches

Innovation Panel and Modules

A new U.S. household panel should follow the lead of the large European panel studies (i.e., German Socioeconomic Panel – SOEP and UK Household Longitudinal Study) by including an innovation panel. This provides an opportunity to test new material and conduct methodological evaluations and experiments. For instance, Manning and Balistreri [58] have demonstrated the value of collecting full household matrices from children and parents and this approach could be tested in a larger scale format in a panel. Carlson [18] has proposed including diagrams or “family maps’ to help identify family relationships. New work by Berger et al. [5] to address complexity could be evaluated in the innovative panel. In terms of support measurement, new items about how technological advances are used to enhance social contact across households or new measures of potential support could be tested.

Further, topical modules provide opportunities to include material that is not part of the regular panel, but provides data to trace change in relationship patterns, attitudes, and family indicators. The PSID has had success with diary modules and focus on specific age groups such as children or young adults as well as a module on family support across households. Topical modules could include relationship forms without much empirical attention such as living apart together, sibling relationships, or single gay and lesbian families.

Multiple Reporters

Most U.S. surveys interview one member of the household or family. The European studies such as the SOEP interviews every household member over age 16 and follows them in and out of the household [39] as well as the UK Household Longitudinal Study [16]. There are only a few U.S. national data sources that ask multiple reporters about household membership.

Limited research has considered child and adult reports of family circumstances. Brown and Manning [13] use the Add Health data to compare family structure reports of parents and adolescents. A limitation of prior work using the Add Health data are that the data only consist of one parent and child pairs and the items measuring family structure are not identical. Few other national data sets include family structure reports from both children and parents (exceptions are the National Education Longitudinal Study, the High School and Beyond Survey). Carlson [18] recognizes that adults and children may have different knowledge about their family situation. For example, the NLSY97 asks the adult respondent whether the youth knows that a father- or mother-figure is not their biological father or mother. The levels of discordance or concordance in reports of family relationships are important when assessing family processes and child well-being.

Often researchers want to capture couple-based measures and there have been couple components added in many surveys such as the FFCWB, NSFH or Add Health. There are sound theoretical and empirical reasons for asking both members of a couple questions about their relationship and family life. However, decisions about collecting couple measures should be carefully considered given the considerable cost. The primary empirical outcome is that there is concordance or discordance in reports of dimensions such as father involvement or relationship quality. There are massive selection issues involved in couple level data. Some strategies such as the Add Health included a subsample of couples so there are issues with who is initially part of the sub-sample and then the ability to collect data from the other member of the couple. Most surveys include sociodemographic proxy data about spouses and partners including age, race/ethnicity, education, and marital history. The NSFG includes separate male and female instruments but does not capture couples. A couple-based module might be an effective way to focus on the well-being of couples and subsequent relationship trajectories as well as examine work-family stress.

Another type of “couple” is composed of both biological parents. The efforts to collect data from both biological parents have lead to some bias towards those that are most stable and highly functioning. For example, in the FFCWB 59% of unmarried fathers are followed up by the time the child is age 5 and in the ECLS-B the response rates of fathers are low, around 50%. The PSID offers an opportunity to follow nonresident fathers as they leave the home and may offer certain efficiencies unavailable in traditional household surveys [43].

Attitudes and Expectations

Researchers often focus on behavior and not expectations or attitudes toward behavior. The GSS provides a comprehensive cross-sectional assessment of attitudes toward family life along with other cross-sectional surveys such as the NSFG. Attitudes mark being favorably or unfavorably disposed, and these along with social pressure or norms, shape expectations [1, 4]. Attitudes or expectations are early harbingers of social change and provide a signal of broader-based changes in social norms. This is so, in part, because barriers to a behavior (e.g., a poor economy) may prevent individuals from achieving behavioral goals but not necessarily the desire to realize that goal. For example, research on marriage shows that women expect to marry but do not always marry because of the high economic bar of marriage [27, 85]. Thus, assessments of the desirability of a behavior cannot be accurately assessed relying solely on behavioral measures. Another advantage of studying expectations is that they are proximate determinants of family behavior (e.g., 10, 36, 63, 66, 79]. The NSLY panel surveys include a number of questions on expectations to marry, have children, and cohabit as well as the timing of these behaviors. Innovative strategies to capture attitudes include Seltzer, Lau, and Bianchi’s [82] work on vignettes to investigate under what conditions certain family behaviors maybe a good or poor idea.

Population Subgroups

Sufficient sample size or targeting sampling is required to ensure that important subgroups are included in the panel [21]. Family patterns differ sharply according to race and ethnicity as well as nativity status. Measures that tap multiracial categorizations as well as various indicators of immigration processes will be important to accurately reflect contemporary family circumstances. Surveys traditionally under-represent disadvantaged populations resulting in serious implications for extending work to policy venues. The disadvantaged more often experience family flux and instability with some of the greatest share of family change occurring among the population without a college degree [22, 59]. Most surveys are by design targeting noninstitutionalized populations, not incarcerated or in the military, but these are important institutions in terms of family well-being and formation [33, 92, 96]. Family experiences into and out of institutions are important to tap in surveys. Sufficient sample sizes and measurement of sexual minorities have been at issue in much family research. New opportunities to measure same-sex relationships should be explored [97], but there is not much empirical evidence about the costs and benefits of different strategies.

Rational for new panel data collection

Overall, a new panel of data collection that includes attention to family circumstances will make important contributions to new scholarship. The primary advantage of a new panel data collection is to provide a contemporary examination of family life. The advantage of a panel rather than a series of cross-sectional designs is the ability to trace individuals over time, establish some temporal ordering of events, and analyze questions about they ways families influence health and well-being. Certainly a long-term panel offers analysis of intergenerational change. The benefits of a new data collection including a focus on family circumstances, relationships, formation, and stability are that it will provide an understanding of the implications of contemporary family life that is not available with current cross-sectional or birth cohort based longitudinal data collections. Survey data rather than administrative data are necessary because there no population registers (as in Europe) that track family change and no capacity to link vital statistics systems records on marriage and divorce to survey data. The potential shortcoming of a panel is that the cohort or sample does not reflect the contemporary climate or population. A serious consideration is the declining response rates and rising costs of survey research. Additional challenges include decisions on the measures that bear repeating and filling in data between interview waves versus inclusion in special topical modules.

The paper offers a synthesis of family formation measurement in surveys and provides a review of some key questions relating to American families. Final decisions about next steps depend heavily on scientific priorities. In terms of family measures a multi-purpose panel study should at a minimum include household rosters that include parental and partner pointers, marital and cohabitation status and histories for both same- and different-sex couples (with attention to marriage or cohabitation to biological parents), current/most recent relationship status and sexual orientation, relationship quality, fertility histories that distinguish multiple partnerships, and complete children’s living arrangements (current and histories). Several data collections are currently in the field including, new SIPP, and several are going in the field soon, such as surveys with continuous interviews the NSFG and Add Health. It will be important to distinguish the comparative advantage of a new household panel and to learn from their measurement strategies. A new household panel study has the potential to provide an important resource for the greater research community and provide an unparalleled ability to answer questions the implications of family change on the health and well-being of Americans.

Table 1.

U.S. Family Data Sources

			Information about Sample
Data Source	Acroynm	Years of Data Collection	Age Range	Generalizability
Longitudinal Panels
Early Childhood Longitudinal Study: Birth Cohort	ECLS-B	Wave 1 - 2001/02 Wave 2 - 2003/04 Wave 3 - 2005/06 Wave 4 - 2006/07	9 months old in 01/02 Kindergarten in 06/07	Nationally representative of children born in 2001.
Early Childhood Longitudinal Study: Kindergarten Cohort	ECLS-K	Wave 1 -1998/99 Wave 2 - 1999/00 Wave 3 - 2002 Wave 4 - 2004 Wave 5 - 2007	Kindergarten in 1998/99 8th grade in 2007	Nationally representative of children in kindergarten in 1998.
Fragile Families and Child Well-being Study	Fragile Families	Wave 1 -1998/2000 Wave 2-1999/2002 Wave 3 - 2001/2005 Wave 4 - 2003/2006 Wave 5 - 2007/2010	Born 1998/2000 9 years old 2007/2010	Representative of urban, nonmarital births between 1998 and 2000.
Health and Retirement Study	HRS	Wave 1--1992	51 + year-olds	Nationally representative of noninstutionalized adults over 50. Sample is replenished every six years to remain representative of non-instituionalized adults over 50.
		Reinterviewed every 2 years
National Longitudinal Study of Adolescent Health	Add Health	Wave 1 -1994/95 Wave 2 -1996 Wave 3 -2001/02 Wave 4 - 2007/08	7th – 12th grad in 94/95 24 – 32 years old in 2008	Nationally reperesentative of youths 7th–12th grade in 94/95
National Longitudinal Survey of Youth 1997	NLSY97	Annually from 1997–2011	12–16 year-olds in 1996	Nationally representative of individuals 12–16 in 1996.
Health and Retirement Study	HRS	Wave 1 -- 1992 Reinterviewed every 2 years	51 + year-olds	Nationally representative of noninstutionalized adults over 50. Sample is replenished every six years to remain representative of non-instituionalized adults over 50.
Midlife in the United States	MIDUS	Wave 1 -- 1995/96 Wave 2 -- 2002/06 Wave 3 -- 2011/16	25–74 in 1995/96	Nationally representative of individuals aged 25–74 in 1995/96.
National Survey of Families and Households	NSFH	Wave 1 - 1987/88 Wave 2 - 1992/94 Wave 3 - 2001/03	19+ year-olds in 1987/88	Nationally representative inidividuals and families in 1987/88.
National Social Life, Health, and Aging Project	NSHAP	Wave 1 -- 2005/06 Wave 2 -- 2010/11	Community dwelling 57–85 year-olds in 2005/06	Nationally representative of adults aged 57–85 in 2005/06.
Panel study of Income Dynamics	PSID	Annually from 1968–1997 Biennially from 1997-present	All household members.	Nationally representative of individuals, families, and households. The sample is replenished to remain nationally representative of more recent cohorts. The initial cohort was representative of individuals, families, and households in 1968.
Survey of Income and Program Participation	SIPP	1984 Panel (1984/85–1987) 1985 Panel (1985–1987) 1986 Panel (1986–1988) 1987 Panel (1987–1989) 1988 Panel (1988–89/90) 1989 Panel (1989–1990) 1990 Panel (1990–1992) 1991 Panel (1991–1993) 1992 Panel (1992–94/95) 1993 Panel (1993–95/96) 1996 Panel (1996–99/00) 2001 Panel (2001–03/04) 2004 Panel (2004–07/08) 2008 Panel (2008–2013)	All household members	Nationally representative of households, families, and individuals every month they are included in the panel.
Cross-sectional Data
Consumer Expenditure Survey	CE	1960–61 1972–73 1980–2013	All household members	Nationally represntative of households, families, and individuals as consumer units.
American Community Survey	ACS	2000–2013 yearly	All household members	Nationally representative of states, households, and individuals.
Current Population Survey	CPS	1968–2013 yearly	All household members	Nationally representative of households and individuals in households.
Nation Survey of Family Growth	NSFG	1973 1976 1982 1988 1990 1995 2002 2006–2010	15–44 year olds	Nationally representative of individuals 15–44 (men were included starting in 2002).
General Social Survey	GSS	Annually 1972–1994 (excluding 71, 81, and 92) Biennially from 1994-present	18 and over	Nationally representative of noninstutionalized individuals 18+.
Decenniel Census	Census	Every 10 years (i.e. 2000, 2010, 2020).	All indvidiuals	Entire US population