Abstract
Many studies purport to reveal the effects of family structure upon child outcomes. Important limitations in such research are discussed. First, many studies rely on current family structure, which overlooks the past environment to which a child was exposed. Therefore, little can be said about the “dose” of family structure(s) received by the child or how such exposures might have occurred at important developmental turning points in the child's life. Studies involving heterosexual and/or same-sex (LGBT) parent families often must deal with such limitations, and so are a good model for assessing how duration of exposure to different types of family structures and the child's developmental situation(s) during such exposures might affect outcomes. Literature examples are discussed, and one study is assessed in detail to demonstrate that in some cases, more information about how long a child spent in different family structures can be found through careful statistical detective work.
The role that family structure may play in influencing children's outcomes has been of interest to many social science scholars. Many family variables could be important, e.g., having two parents, a single parent, or step-parent, same-sex parents, or adoptive parents, siblings, or step-siblings, etc. In assessing the effect of family structure, correlating the current structure of a child's family and outcomes is not adequate. There are at least two problems with this approach. First, family structures change over time, so that a child often does not remain in the same structure from birth to reaching maturity. Furthermore, the duration of exposure to any particular family structure may differ from one family to another. Second, exposures may occur at different developmental stages or situations of the child (and their parents). There might be key developmental times where family structure has a greater or lesser effect on a child. Therefore, “snapshot” renditions of family structure are incomplete, if not misleading, in terms of the child's development.
Although not all studies with such limitations (e.g., Burston, Murray, Mooney-Somers, Stevens, & Golding, 2003) have been subjected to intense criticism (Redding, 2013), some have been (e.g., Amato, 2012; Barrett, 2012; Eggebeen, 2012; Gates, et al., 2012; Goldberg, Kashy, & Smith, 2012; Massey, 2012; Osborne, 2012; Sherkat, 2012; Siegel, Perrin, Dobbins, Lavin, Mattson, Pascoe, et al., 2012; Wright, 2012; Anderson, 2013; Becker & Todd, 2013; Moore & Stambolis-Ruhstorfer, 2013; Perrin, Cohen, & Caren, 2013; Reiss, 2014), although some have minimized the implications of such limitations (Destro, 2012; Johnson et al., 2012; Marks, 2012; Monte, 2013). 2 For example, one of the chief criticisms of Regnerus's (2012a, 2012b, 2012c) research was that those children who had been born into a mixed-orientation marriage (MOM) should not have been classified as children from gay or lesbian families even if the gay or lesbian parent had established a stable home thereafter.
A few illustrations of some of these problems, featuring several recent studies based on the National Longitudinal Study of Adolescent Health (known as the Add Health study), are reviewed below in chronological order of publication. The National Longitudinal Study of Adolescent Health began as a nationally representative sample of over 90,000 U.S. adolescents who were in Grades 7 to 12 during 1994–1995. There have been four longitudinal in-home interviews of the adolescents, for a sub-sample of approximately 15,000 adolescents, most recently in 2008. The Add Health survey has been used to investigate numerous aspects of well-being with ecological data on the social and community environment, as well as biological data at the individual level. Although not reviewed here, Tillman's (2007) study using ADD HEALTH data is one of the better recent such studies.
Studies of Heterosexual Parent Families
Add Health Studies
Cookston (1999) used Add Health data to compare child outcomes in terms of problem behaviors as a function of three family structures (single-mother, single-father, intact family), with parental supervision as a mediating variable. The adolescents in the study ranged in age from 11 to 19 yr. (M = 14.8). Although the study's in-home interviews involved a large sample of 15,000 adolescents, only 684 adolescents were studied because others (> 95%) were from families that could not be clearly coded into one of the three categories. Families were coded as single-parent only if the adolescent's biological parents had never married or had been married and later divorced. Of the 684 adolescents, 322 were from single-mother families, 106 from single-father families, and 256 from intact families. Cookston found that delinquency (d = 0.34 in single-mother families; d = 0.60 in single-father families), heavy drinking (d = 0.16 in single-mother families; d = 0.47 in single-father families), and illicit drug use (d = 0.17 in single-mother families; d = 0.49 in single-father families) were lower for adolescents from intact families or from families (more often intact ones) where parental supervision was high vs low (d = 0.25 for delinquency; d = 0.27 for heavy drinking; d = 0.22 for illicit drug use). While the study's conclusions might seem strong, the results were based on less than 5% of the Add Health sample and neither the duration of time having lived in any of the types of families nor the developmental stages of the adolescent were used in the analyses of child outcomes.
Davis and Friel (2001) also used Add Health data from 6,261 adolescent girls and 6,106 adolescent boys, between the ages of 11 and 18 yr. In terms of family structure, they did not report n's, but percentages, for intact two-parent family (65.4%, approximate n = 8,088), stepfamily (33.3%, approximate n = 4,188), single-parent family (1.3%, approximate n = 161), cohabiting parents (11%, approximate n = 1,360), and lesbian parents (2.1%, approximate n = 260). They did not find family structure to be strongly related to either of their two outcome variables (age at initiation of sexual activity and number of sex partners); the only variables consistently related to those outcomes for both girls and boys were adolescent religiosity (measured in five levels) and the adolescent's GPA (measured in four levels). For each unit increase in religiosity girls initiated sexuality 0.11 yr. later, boys 0.09 yr. later. Likewise, for each unit increase in religiosity, girls had 0.16 fewer sexual partners, boys 0.21 fewer. For each unit of higher GPA, girls initiated sexuality 0.17 yr. later, boys 0.22 yr. later. For each unit of higher GPA, girls had 0.36 fewer sexual partners, boys 0.29 fewer. However, their analyses did not take into account parental relationship stability or duration, or the adolescent's age at which changes in family structure may have occurred, relative to the adolescent's development.
Manning and Lamb (2003) also used Add Health data from 13,231 adolescents to assess child outcomes as a function of family structure in terms of single mothers (n = 3,593), cohabiting unmarried parents (n = 559), married stepparents (n = 1,352), and married two-biological parents (n = 7,727). They noted specifically, “… the Add Health data do not include details about family structure histories” (p. 881). They did control for the duration of the relationship for all the family types except single mothers as well as number of mother's marriages, parental supervision and closeness, family income, mother's age and education, child's age, importance of religion to child, number of children in the household, and other variables. In general, adolescents from two-parent married families fared better, although the importance of religion to the adolescent and closeness to mother were also significant predictors of the adolescent doing better. Compared to those with two biological married parents, adolescents from families with a cohabiting stepfather, married stepfather, or single mother reported a greater frequency of having been suspended or expelled from school (p < .001, all three comparisons), a higher rate of delinquency, a greater extent of school problems, and lower GPA. Scores on the Peabody Picture Vocabulary Test (PPVT) were only lower for adolescents from cohabiting stepfather (B = 2.36) families and single mother (B = 0.85) families compared to those from married biological parents. Differences on college expectations were not statistically significant for any of the comparisons. For example, in regression models with unstandardized coefficients, importance of religion was related to higher grade point average (B = 0.16), lower rates of delinquency (B = 0.72), fewer school problems (B = 0.25), and higher expectations of going to college (B = 0.11) while closeness to mother was related to higher grade point average (B = 0.22), lower rates of delinquency (B = 1.18), fewer school problems (B = 0.55), higher scores on the Peabody Picture Vocabulary Test (B = 0.96), and to higher expectations of going to college (B = 0.08). They acknowledged that family structure (e.g., whether the parents married) might have been influenced or caused by child problems or related to selection effects. Specifically, they said that “…. we may find that mothers with children who have greater behavior problems and poor school performance are more likely to cohabit than marry. Thus, there could be selection into family types based on the adolescent behaviors. In a similar vein, the causal nature of the covariates is not clearly specified in our models” (p. 891).
Demuth and Brown (2004) also used Add Health data to assess child outcomes as a function of single-parent or step-family family structure. In particular, they assessed outcomes for adolescents from “two-biological-parent, married-couple families (n = 9,505), single-mother families (n = 3,792), single-father families (n = 525), mother-stepfather families (n = 2,039), and father-stepmother families (n = 443)” (p. 66). Family structure was based on the adolescent's current situation at the time of their interview. Demuth and Brown also controlled for the presence of any other adult in the home, as well as parent involvement, closeness, household size, parent education, and child's age. They found that family structure, in general, no longer significantly and directly predicted child outcomes after controlling for the other variables, although the family process variables may have mediated the relations between family structure and child outcomes. While the Demuth and Brown (2004) study improved over the Cookston (1999) study by using more of the adolescent participants and by including stepfamilies, it still did not distinguish dose effects of family structure or child development effects.
Hawkins, Amato, and King (2006) used Add Health data to assess parental involvement as a function of family structure, including children with two married parents (n = 10,275), an unpartnered resident mother and nonresident father (n = 3,212), a repartnered (married or cohabiting) resident mother and nonresident father (n = 1,955), a nonresident mother and unpartnered resident father (n = 529), a nonresident mother and repartnered (married or cohabiting) resident father (n = 422), and two nonresident parents (n = 937). While they found greater parental involvement for mothers and resident parents, their data nonetheless did not include much in the way of longitudinal assessment of changes in family structure related to the developmental status of the adolescents.
Limitations of Heterosexual Family Studies
The Add Health dataset can be drawn from in various ways, depending on the definitions of family structure in each study. Definitions must vary substantially among different reports, as accounts of different types of families vary widely (Table 1). The criteria for such definitions tend not to be described in detail, possibly not enough to permit replication of or even a clear understanding of the results of such Add Health reports. For example, it is interesting to note that in most of the above studies diverse types of single-parent families would have met the inclusion criteria for the study. For example: (1) child lived with both biological parents until age 17 when the parents divorced and the child lived with her mother after the divorce, (2) the child's biological parents never married and the child is now 19 yr. old, (3) the child's parents divorced when the child was three yr. old and neither parent remarried, while the child is 11 yr. old now, (4) the child lived with both biological parents until they divorced when the child was 12 yr. old after which the father died when the child was 13 while the child continued to live with her mother for three yr. after the divorce. Lumping all of these situations together as if they would contribute to a child's development in the same way is clearly a serious limitation. The studies did not assess the number of transitions to which a child might have been exposed, the relationship stability of the child's parents, or the ages at which the child experienced various types of transitions. Some of the studies did not report effect sizes or enough information to calculate effect sizes.
Variations in Numbers of Different Types of Families in Different Reports Based on Add Health Data
Again, as noted, the sample sizes for each alleged type of family were very different in various studies (Table 1). Sample sizes for single-parent families, using the same national data set, ranged from 161 to 4,317. For stepfamilies, samples ranged from 1,352 to 4,188, and for two-parent, intact families the sample sizes ranged from 256 to 10,275. The sample size differences from the same data set indicate that very different definitions of family structures were used in these studies. Given such differences, valid and useful comparisons across the studies are virtually impossible.
Studies of Same-Sex Parent Families
Selected Studies on Same-Sex Families
In the Add Health data set, most studies have not assessed different structures among same-sex parent families, especially with respect to male same-sex parent families. Those that have categorized family structure indicate the same problem of criteria: Davis and Friel (2001) identified 260 lesbian mother families in the data set, but later researchers only identified between 18 (Patterson, 2009) and 44 (Wainright & Patterson, 2006) lesbian couple families. Thus, in all likelihood, the vague definitions of family structure and lack of family history have affected studies of same-sex parent families as they have heterosexual parented families.
In contrast to the literature on heterosexual parent families, many studies in the literature involving lesbian mothers or gay fathers have reported some heterosexual relationship history in the parents' past (Schumm, 2012). As early as 1991, Tasker and Golombok observed that “A further limitation to these studies is that most of the children who participated spent the early part of their lives in a heterosexual family” (p. 186). Recent studies that confirm that observation include Goldberg and Kuvalanka (2012), Goldberg and Allen (2013), Goldberg, et al. (2012), and Regnerus (2012a, 2012b, 2012c), reviewed below. A more detailed discussion of research on same-sex parent families by Golombok and her colleagues will follow (Golombok, Perry, Burston, Murray, Mooney-Somers, Stevens, et al., 2003; Stevens, Perry, Burston, Golombok, & Golding, 2003; Perry, Burston, Stevens, Golding, Golombok, & Steele, 2004).
Family Stability
Goldberg and Kuvalanka (2012) discussed how they obtained a sample of 49 children who had been born into 22 heterosexual relationships and 27 non-heterosexual relationships. Since the heterosexual relationships involved parents who came out later as gay or lesbian, it is probably safe to assume most of them divorced, although that percentage was not reported. But what of the 27 non-heterosexual relationships? Of the 27, 20 were born via donor insemination to lesbian couples, 2 were born to single lesbian mothers, one was adopted at birth by a lesbian couple, one was adopted as a toddler by a lesbian couple, while three more were born into or adopted by couples who were not romantically involved but were co-parents. Thus, there were 22 lesbian couples who gave birth to or adopted children (2 children had single mothers, 3 children had parents who were not romantically involved). Goldberg and Allen (2013) reported that of those 22, 16 broke up and usually re-partnered into new stepfamily configurations (73%, 15 born into lesbian couples via donor insemination and one adopted as a toddler by a lesbian couple); of the 20 lesbian couples who became parents by donor insemination, 15 (75%) broke up. Furthermore, of the 16 children from same-sex couples who broke up, the average age for that breakup was cited by Goldberg and Allen (2013, p. 533) as 6.34 yr., with a median of 5.50 yr. and a range of 1 to 13 yr. However, of the three children born to or adopted by non-romantically involved couples or to single lesbian mothers (n = 5), four (80%) had parents who later formed romantic partnerships that later dissolved (Fig. 1). Thus, of the 27 children born into or adopted by gay or lesbian parents (not originally born into a heterosexual relationship except for the one child adopted as a toddler by two lesbian mothers), 20 (74%) experienced a parental breakup. If all of the 22 children born into heterosexual relationships experienced a parental divorce or breakup, then as many as 42 (86%) of the 49 children overall experienced a parental breakup. This is far higher than would be expected in a population sample in the U.S. Amato (2000) indicated that about 40% of all children would “experience parental divorce before reaching adulthood” (p. 1269). Amato (2010) stated that data from the “National Survey of Family Growth indicate that 42% of non-Hispanic Whites and Hispanices divorced within the first 15 years of marriage” (p. 651), although that rate of divorce was lower for married couples with children. Lansford (2009) indicated that “between 43% and 50% of first marriages” (p. 140) would end in divorce, affecting 50% of American children.
Definitions of Family Structure
Regnerus (2012a, 2012b, 2012c) has been criticized extensively because his sample included adult children of same-sex parents where those children had spent part of their youth living with heterosexual parents before or after one or both of the parents came out as gay or lesbian. If similar criteria had been applied to previous studies of children from LGB families, many of the participating families would have had to been disqualified from participation. Even so, one of the limitations of the Regnerus (2012a, 2012b) studies was that respondents were classified as having been only from a LGB family even if they had never lived with the same-sex partner of their LGB parent. On the positive side, the NFSS data set does include data from birth to adulthood for each respondent on which caregivers the respondent had lived with on a year-by-year basis, providing the potential to overcome many of the limitations of previous heterosexual or same-sex parenting studies.
Family Structure “Dose”
While, as noted, many studies of lesbian, gay, bisexual, or transgender (LGBT) parenting have involved parents from previous mixed orientation marriages (MOMs), sometimes those studies have not explained the extent to which the samples involved MOMs nor the ages of the children when the parents re-partnered into LGBT families. Golombok, et al. (2003), Stevens, et al. (2003), and Perry, et al. (2004) published results from a data set that included 28 lesbian mother families in which the children had been born into a heterosexual relationship, as well as 10 (Stevens, et al., 2003, p. 347; Perry, et al., 2004, p. 470) or 11 (Golombok, et al., 2003, p. 22) other lesbian families where at least one child had been conceived by donor insemination. The researchers explained that the average age of the 28 children when their mother entered into a lesbian relationship was 4.1 yr. (approximately 49 mo.) with a range of 0 to 108 mo. (Golombok, et al., 2003, p. 22; Stevens, et al., 2003, p. 350; Perry, et al., 2004, p. 470). Furthermore, the average age of the child at the time of the study was 92.5 mo. (SD = 15.5) with a range of 62 to 116 mo. (Stevens, et al., 2003, p. 350). The goal of the study had to been study families that had a 7-yr.-old child, but for the lesbian families the age restriction was relaxed to include children as young as five and as old as nearly nine years of age in order to recruit more lesbian families as participants in the study.
In none of these studies did the researchers disclose how much time the children actually spent in a heterosexual family rather than in a lesbian family. Thus, we are left with a sense that a large percentage of the families had begun as MOMs (at least 72%, 28/39), but it is not clear how many of the children had spent more time in a heterosexual family environment than in an LGBT family environment. Other scholars have noted this issue. While discussing Golombok, et al.'s (2003) study, Goldberg, et al. (2012) reported that of the children of the 39 lesbian-mother families, 28 “had been born into a heterosexual-parent family, and were 4.1 yr. old, on average, when their mothers entered into a lesbian relationship. Thus, the children of lesbian mothers often spent time in other family structures during their early years” (p. 506). Schumm (2012) noted that, with respect to the Golombok, et al. (2003) research, “children who had spent most of their life in heterosexual families were, in fact, included within the operational definition of children from lesbian families” (p. 1360). In the Appendix, more detailed analyses are presented to reaffirm the observations of Goldberg, et al. (2012) and Schumm (2012) with respect to the Golombok, et al., studies, a situation partly necessitated because Golombok, et al. have not released detailed information on how long each of their participants spent in various family structures.
It is possible that the “dose” of LGBT or heterosexual parenting might have affected the child outcomes measured, possibly depending on the developmental status of the child when the changes in family structure had occurred. It is vital to establish the relevance or effect of “family structure dose” before grouping LGBT or heterosexual parent families, because sampling without regard to stability may lead to conclusions that can not be generalized to all families of either type. This principle has been recognized by others; e.g., Rosenfeld (2010, 2013) did not combine recently established and more stable LGBT families in his U.S. Census data analyses, implying that he did not think it wise to assume that all LGBT families were alike, regardless of how long they had been established or stable. Similarly, Allen (2013) also had to deal with similar issues in his analyses of Canadian census data.
Conclusion
Family structure does not appear to be static in either heterosexual or LGBT families. Recategorizing families because they have experienced changes over time would eliminate many from eligibility for research participation. Taking a “snapshot” approach also ignores the timing of changes which might tie in with the developmental status of the child and future child outcomes. Even so, it appears that researchers have often disagreed substantially on how to classify family structure, even when using the same data sets. If a child's entire family structure history is known, questions remain about how to classify that structural exposure overall.
Family structural histories can be complex and this is becoming more common as time goes on. Although it would be ideal to know year-by-year how a child's family structure changed over time, it might be helpful to know the type of family structure to which the child had the most exposure. If researchers do not release their data for independent analysis, careful statistical detective work can still indicate how many children in a data set spent more time in one family structure than in another (see Appendix). Such work is valuable in developing study designs to examine various difficult-to-study issues related to family structure, including, e.g., parenting styles, juvenile delinquency, drug use, sexual activity, academic and social success, etc. Longitudinal studies are so expensive and time-consuming that such statistical work on less-than-complete extant reports is well worth the effort. It is vital that peer-reviewed reports include far more information about samples and definitions of family structure.
Footnotes
2
At the same-sex marriage trial in Michigan (February 25, 2014), Dr. David M. Brodzinsky served as an expert witness for the plaintiffs. He testified (p. 65) that because the participants in the Regnerus study had been born into heterosexual families that were later disrupted, the study did not allow for any conclusions to be drawn about the effect of being raised by same-sex parents. However, many other studies with same-sex families have incorporated similar types of families and have not been challenged with respect to their legitimacy (Schumm, 2012; Redding, 2013).