Close To You? How Parent–Adult Child Contact Is Influenced by Family Patterns

Abstract

Objectives.

Intergenerational contacts occur in the context of other family relationships. We examine how in-person contacts among parents and all adult children affect each other, focusing on proximity and other predictors to assess whether and how visiting is correlated across adult children.

Methods.

We use a modeling approach derived from an adaptation of multilevel models to provide a convenient mechanism by which to write child-specific equations, each with its own set of predictors, and wherein one child’s attribute values can be attached to other children’s records.

Results.

We find that parent–adult child visiting is positively correlated across siblings, but the frequency of visiting within families is not directly reciprocated. Rather, visiting responds to common family factors. Visiting declines with distance, but there are strong discontinuities in the effect. Distance between parents and a focal child is positively associated with visiting with other children.

Discussion.

The empirical patterns we report can be framed within enhancement and compensation models. Positive correlations and cross-sibling interactions that juxtapose levels of visiting against not seeing a child in last 12 months are consistent with the enhancement model. The cross-sibling interaction for distance, whereby one child’s farther distance leads to more visits reported with others, provides evidence of a countervailing, though, weaker, pattern of compensation for proximity.

Contrary to earlier concerns about family decline, parent–adult child relations continue to be strong and influential for the well-being of both generations (Bengtson, Biblarz, & Roberts, 2002; Logan & Spitze, 1996; Silverstein & Bengtson, 1997; Umberson, 1992). Parents and adult children often live near each other, have positive feelings, and provide support when needed. Many phone and/or see each other regularly. For example, although patterns are difficult to compare over time or cross-nationally, it appears that most parents see a noncoresident child at least weekly, with little evidence of declining in contact in various Western countries (Kalmijn & DeVries, 2009; Treas & Gubernskaya, 2012).

Visiting, viewed as a central dimension of the intergenerational solidarity paradigm (Rossi & Rossi, 1990; Silverstein & Bengtson, 1997), has received considerable research attention in recent decades (e.g., Crimmins & Ingegneri, 1990; Lawton, Silverstein, & Bengtson, 1994; Sarkisian & Gerstel, 2008). It has been found to be related to characteristics of both parents and adult children, perhaps most importantly to their geographic proximity (Greenwell & Bengtson, 1997; Hank, 2007). It also relates to other dimensions of solidarity, such as helping (Rossi & Rossi, 1990) and affection (Lawton et al., 1994), and may set the stage for subsequent support to elderly parents. Proximity and contact can be argued to represent the basic opportunity structure for intergenerational interaction and support (Bengtson, 2001; Hank, 2007). Kalmijn (2006, p. 2) suggested that face-to-face contact is a “good indirect measure of intergenerational support.”

However, intergenerational contacts occur in the context of other family relationships, often in gatherings involving multiple adult children. We know much about the frequency of visits, but research has tended to focus on individual parent–child dyads (e.g., Logan & Spitze, 1996; Rossi & Rossi, 1990) or aggregated parent reports on relations with adult children (e.g., Crimmins & Ingegneri, 1990; Shapiro, 2003). We know little about how visiting between parents and particular adult children is influenced by contacts with other adult children or characteristics of those children. We also lack precise understanding of the implications of parent–child proximity for contact, and how location of one child may affect contacts with others.

As Matthews (2002, p. 7) notes, “families can be broken into pairs or dyads, but it is important to remember that such dyad members usually are part of a larger system in which the other members affect their interaction.” Recent work has begun examining these within-family effects to determine how adult children’s relations with parents influence those of other children in the family. However, most studies using data for multiple parent–child relations in a family have focused on help to older parents (e.g., Leopold & Raab, 2013; Leopold, Raab, & Engelhardt, 2014; Silverstein, Conroy, & Gaus, 2008; Tolkacheva, van Groenou, & van Tilburg, 2010; Wolf, Freedman, & Soldo, 1997), with less attention to within-family influences on parent–adult child visiting.

Using the National Survey of Families and Households (NSFH) and building upon recent approaches using multilevel modeling, we examine how in-person contacts among parents and all adult children are interrelated and how behavior and characteristics of one parent–child pair affect other pairs. We focus on in-person contact (visiting) and a key predictor, proximity, to assess (a) whether and how visiting is correlated across adult children; (b) how a child’s proximity affects visiting with parents; and (c) whether one child’s visiting and proximity to a parent affects visiting of other children. Later, we outline our conceptual models of within-family effects. We then explain our focus on contact and on our key predictor, proximity. Finally, we discuss research questions to be addressed using our new modeling approach, as well as describing and justifying other variables to be included in our empirical models.

Background

Toward Models of Within-Family Dynamics

As an early step toward analyzing family-level patterns of contact or help, some studies included measures of family structure (e.g., number of children) in models predicting relationship dimensions for individual pairs or summary measures of parents’ relationships with all children (e.g., Eggebeen, 1992; Hoyert, 1991; Rossi & Rossi, 1990). Logan and Spitze (1996) found that parents with more children had less contact with each individual adult child, although having more children increased overall contact. Ward, Spitze, and Deane (2009) found that having more adult children was associated with increased differences across children in both contact and reported quality of relations.

Going beyond empirical effects of family size, researchers also suggested models to explain relationship patterns. Uhlenberg and Cooney (1990) proposed three alternative processes through which family size might affect bonds in later years: (a) weaker bonds with less attention to each child; (b) stronger bonds reflecting a family orientation; and (c) higher quality but less contact with each child due to shared support among siblings. Their results supported the third hypothesis. Aldous and Klein (1991) suggested similar models in which a larger family might enact “familism” but might also experience size constraints. Logan and Spitze (1996) defined an additive model (more children yield more contact or help) and a substitution model (with limits on time, energy, or needed help, children may substitute for other siblings).

A more recent innovation is the use of multilevel models to analyze data for some or all adult children, with either an older parent or an adult child as respondent. Much of this research focuses on patterns of caregiving provided to frail older parents, whereas some has examined support from mothers to adult children or support in both directions. Researchers have examined how adult children’s characteristics relate to who becomes or would be chosen as a caregiver for their parent (e.g., Leopold et al., 2014; Pillemer & Suitor, 2006), as well as more generally who is identified as emotionally closest (Suitor & Pillemer, 2007). Increased hours of support by siblings has been found to reduce an adult child’s hours of helping (Wolf et al., 1997), but a child’s caregiving involvement is positively associated with siblings’ average caregiving (Tolkacheva et al., 2010). Using fixed-effects modeling, Silverstein and colleagues (2008) find within-family differences among adult siblings in support to mothers that appear to be at least partly based on expectations about ability and willingness of siblings to participate.

Diverging from the focus on caregiving to parents, others have focused on how adult children’s characteristics influence mothers’ allocation of support (Kalmijn, 2013; Suitor, Pillemer, & Sechrist, 2006) or patterns of support in both directions (Fingerman, Miller, Birditt, & Zarit, 2009). Leopold and Raab (2013) examine patterns of short and long-term reciprocity between parents’ financial support and adult children’s caregiving. Kalmijn (2006) used multilevel models and data for up to two adult children, parents, and/or siblings to examine how educational differences influence proximity and contact.

Our work builds on this body of research in two ways. Substantively, we diverge from the typical focus on assistance by studying in-person contact, a more broadly applicable dimension of parent adult–child relations, and by examining its relationship to proximity in more depth than has been done in the past. Analytically, we use a data file structure and methodology that treats the family as a system of K interrelated equations (where K is the count of adult noncoresident children). This approach allows us to examine both the influences on a parent’s relations with each child and of a child’s characteristics on the parent’s relations with other children without removing the sibling pair from its larger family context.

Our earlier work has provided a background to the theoretical models we use here. One study found that some relationship dimensions between a parent and adult child (visiting and feeling close) were higher the more that parent visited and felt close to that child’s siblings, while others (phoning and children’s help to parents) were lower the more the parent engaged in such exchange with the child’s siblings (Spitze, Logan, Deane, & Zerger, 1994). More recently, using the methods and data set described below for the current analysis, we examined how patterns of socioemotional support were correlated across adult children in families (Spitze, Ward, Deane, & Zhuo, 2012). Parents who report giving support to one child are more likely to give support to other children, whereas receiving support from one child is negatively related to receipt from others. These exchanges of socioemotional support between parents and a particular adult child appeared to be independent of the location (proximity) of other siblings, or of having a sister or a stepsibling.

The complexities of enhancement and compensation

Three general patterns can be derived from this previous work: positive, negative, or zero intrafamily correlations for visiting or other relationship dimensions; we have termed these enhancement, compensation, and independence models, respectively (Spitze et al., 2012). Each model may reflect various processes or emotional qualities. For example, enhancement could reflect a warm familistic culture in which contacts by siblings are shaped by shared family factors. Most pertinent here, sibling contacts may directly encourage each other’s involvement, including through joint visits and family get-togethers, so that “more promotes more,” or children may feel competitive and try to see parents as much as others. Of course, a positive intrafamily correlation for siblings’ contact is also consistent with “less leads to less,” reflecting generally negative relations across children that might better be labeled dysfunctional than enhancement. Similarly, compensation may reflect explicit coordination or attempts to make up for a lack from another source, patterns of favoritism, or simply situations where “more leads to less” elsewhere due to finite time, needs, or resources. A zero correlation may imply less coordination, perhaps a lack of shared family culture, or fewer constraints.

Our Focus on In-Person Contact

We focus on in-person visiting, “associational solidarity” in the intergenerational solidarity paradigm and a primary outcome addressed in previous research. Of course, parent–adult child contact may also occur through other means: telephone, letters, e-mail, or more recently, various social media. Previous research has sometimes included both visiting and phone/letter contacts (e.g., Greenwell & Bengtson, 1997; Logan & Spitze, 1996; Sarkisian & Gerstel, 2008). Although face-to-face and telephone contact may substitute (Kalmijn, 2006), their frequencies tend to be correlated (Rossi & Rossi, 1990) and they exhibit very similar patterns and correlates (Greenwell & Bengtson, 1997; Kalmijn, 2006; Lye, Klepinger, & Hyle, 1995; Sarkisian & Gerstel, 2008). Further, both are affected by proximity, although phoning less strongly (Logan & Spitze, 1996; Rossi & Rossi, 1990; Sarkisian & Gerstel, 2008).

A number of studies have reported frequency of parent–adult child contact in the United States and Europe (e.g., Crimmins & Ingegneri, 1990; Hank, 2007; Kalmijn, 2006; Sarkisian & Gerstel, 2008; Silverstein & Bengtson, 1997; Ward et al., 2009). As stated earlier, data from various Western countries suggest most parents see a noncoresident child at least weekly (Kalmijn & DeVries, 2009; Treas & Gubernskaya, 2012). This at least partly reflects proximity despite geographic mobility in modern societies: parents generally live near (5–10 miles) at least one noncoresident adult child (Lin & Rogerson, 1995; Logan & Spitze, 1996; Silverstein & Bengtson, 1997).

The Centrality of Proximity

Theorists have long recognized parent–child proximity as a central feature of family typologies, and an integral determinant of interaction (Greenwell & Bengtson, 1997; Litwak, 1960; Litwak & Kulis, 1987). Litwak noted that geographic mobility both reflects and affects intergenerational cohesion, though the “modified extended family” can overcome its greater dispersion to maintain solidarity. The intergenerational solidarity paradigm that has guided much research on intergenerational relations includes distance as a key dimension of structural solidarity that enables or constrains face-to-face contact and exchange of support (Bengtson, 2001; Lawton et al., 1994; Roberts et al., 1991; Silverstein & Bengtson, 1997). Proximity is perhaps the strongest predictor of contact (e.g., Hank, 2007; Greenwell & Bengtson, 1997), whether focusing on adult children (e.g., Kalmijn, 2006; Sarkisian & Gerstel, 2008; Waite & Harrison, 1992), parents and a particular child (e.g., Logan & Spitze, 1996; Rossi & Rossi, 1990), or parents and the most proximate child (Crimmins & Ingegneri, 1990; Hank, 2007). Proximity is also associated with higher quality of ties (Kaufman & Uhlenberg, 1998) and more assistance (Eggebeen, 2005; Hogan, Eggebeen, & Clogg, 1993; Hoyert, 1991; Joseph & Hallman, 1998).

Although the role of proximity in shaping intergenerational contact may seem self-evident, both conceptual and empirical questions remain. Studies on proximity have mostly been restricted to parent–child or sibling pairs, often measured as distance between a parent and a particular adult child, and sometimes as a family-level measure, such as nearest child. For example, Crimmins and Ingegneri (1990) found distance to closest child was the strongest predictor of aggregate parent–adult child contact. Silverstein (1995), however, recognized the need to consider other children, and how, for example, the closest child may influence family relations. Shapiro (2003) included measures of both nearest and farthest noncoresident child to reflect proximate and nonproximate support networks; his investigation of the influence of proximity on the impact of later-life divorce on parent–child contact is a rare example that has been (partially) responsive to Silverstein’s (1995) charge. Thus, the location of particular adult children may influence the behavior of others, and within-family effects of proximity likely are greater for visiting than other forms of contact. One of our goals is to investigate more directly how distance of particular children from parents may influence visiting patterns of others.

Further, we assess how varying degrees of proximity influence relationships. Since children often cluster in relatively close proximity, it is important to investigate potential complexities and nonlinearities to capture the effects of distance more precisely. Previous research has coded distance in various ways, including log of distance to nearest or farthest child (e.g., Crimmins & Ingegneri, 1990; Kaufman & Uhlenberg, 1998; Shapiro, 2003), dichotomies divided at 25 km, 25 miles, or 50 miles (e.g., Eggebeen, 1992; Greenwell & Bengtson, 1997; Hank, 2007), and hours of travel time (Lawton et al., 1994; Logan & Spitze, 1996). We explore detailed empirical patterns for distance, including assessment of potential break points in the relationship between proximity and visiting.

There are also issues of causal order and endogeneity in understanding the associations between proximity and other dimensions of solidarity (Greenwell & Bengtson, 1997; Silverstein et al., 2008; Wolf, 1994). Locational choices, for example, may be affected by parent needs for support or the importance given to family ties (Michielin & Mulder, 2007), and proximity and contact may be jointly determined by other factors (such as parent health or social class) which may inflate estimates of the effect of distance. This may be more evident for exchanges of support than for contact. Using longitudinal data, Greenwell and Bengtson (1997) found that distance had a weaker but still strong and negative relationship with face-to-face contact after adjusting for endogeneity. Similarly, Ward, Deane, and Spitze (2014) found that parent health was unrelated to change in contact, with proximity and locational change very strong predictors of contact change after controlling for other factors.

Research Questions

We investigate parent–adult child contact, using models that quantify within-family dependence and illustrate how one parent–adult child relation (or characteristic) can affect associations among other parent–child pairs. One focus is on the effects of proximity for contact with a particular child, and of that proximity on contact with other children. While contact may be negatively associated with distance between parent and child, distant siblings may encourage greater contact by others to compensate. The following questions are posed.

First, is visiting correlated across adult children, after controlling for parent, family, and child attributes? If so, does one child’s visiting increase (enhancement) or decrease (compensation) that of others? Do some visiting patterns of one adult child have a particular effect? For example, if one child has had no contact in the past year, how does this affect visiting with other children? Second, net of controls, how does an adult child’s proximity to parents affect their visiting? Are effects linear, or linear in some readily interpretable transformation of distance, or do particularly close distances have a stronger impact? Third, net of controls, does one child’s proximity to parent affect visiting with other children? If a child lives near parents, do other children visit more (enhancement) or less (compensation)?

Direct reciprocity between an adult child’s frequency of visiting with parents and the frequency of visiting by his/her siblings need not be the sole determinant of family contact dynamics. An alternative explanation for correlation of visiting within families is that adult children are responding to shared underlying family factors, such as family climate or norms. This also suggests that intrafamily responses to siblings’ contact may differ from responses to siblings’ distance. For example, contact may be positively associated with other siblings’ contact due to the net effect of family climate/norms (enhancement), whereas a specific aspect of common family factors such as distant siblings may encourage greater contact by others (which is consistent with compensation). We do not have direct measures of shared normative factors, but we use a standard set of control variables that have been found to be predictors of parent–adult child relations, including various dimensions of family solidarity or ambivalence (e.g., Leopold & Raab, 2013; Pillemer & Suitor, 2002; Silverstein & Bengtson, 1997; Silverstein et al., 2008; Spitze et al., 2012; Umberson, 1992; Ward, 2008; Ward et al., 2009). Thus, in addition to our focus on proximity, analyses include other relevant parent, child, and family characteristics.

Parent characteristics include gender, marital status, age, education, health, and race/ethnicity. Mothers tend to have more contact and more tight-knit relations with children (Rossi & Rossi, 1990; Sarkisian & Gerstel, 2008; Silverstein & Bengtson, 1997). Unmarried or remarried parents have less contact (Logan & Spitze, 1996; Lawton et al., 1994; Umberson, 1992), though divorce may decrease fathers’ contact but increase mothers’ (Shapiro, 2003). Better-educated parents may visit less with children (Crimmins & Ingegneri, 1990; Spitze et al., 1994), though Umberson (1992) did not find this. Older parents (Logan & Spitze, 1996; Umberson, 1992) and healthier parents (Kaufman & Uhlenberg, 1998) may see children more. Findings on whether Black and Hispanic families have more contact are mixed (Hogan et al., 1993; Silverstein & Waite, 1993; Umberson, 1992).

Child characteristics include gender, marital and parental status, biological/adopted versus stepchild status, and distance from parents. Some studies have found greater contact by parents with daughters (Silverstein & Bengtson, 1997; Umberson, 1992). Parents may also have less contact with married children (Musick & Bumpass, 2012; Sarkisian & Gerstel, 2008) and with children who have children (Merrill, 2011), reflecting other responsibilities for children, though the implications of these roles may differ for sons and daughters (Merrill, 2011) and some have found little effect (e.g., Logan & Spitze, 1996). Stepparents and adult stepchildren have weaker ties and less contact (Ganong & Coleman, 2006; Ward & Spitze, 2007; Ward et al., 2009).

Family characteristics include number of adult children and whether the parent coresides with any child. Larger families may reduce contact with a particular child, but increase total contacts with adult children (e.g., Logan & Spitze, 1996; Sarkisian & Gerstel, 2008; Shapiro, 2003; Treas & Gubernskaya, 2012). Number of children is associated with increased differentiation across children in both contact and reported quality of relations (Ward, 2008; Ward et al., 2009). Coresidence with minor children may reduce parents’ contact with noncoresident adult children (Umberson, 1992). However, there is no evidence that coresidence with an adult child affects contact with noncoresident children (Umberson, 1992) or quality of relations with those noncoresident children (Spitze et al., 2012).

Data and Methods

Sample

Data are from Wave 1 (1987–1988) of the NSFH, a representative national sample aged 19 and older that oversampled some demographic groups and household types (Sweet & Bumpass, 1996). Unweighted data are used here, accounting for NSFH’s complex design by including variables defining the sampling design (Johnson & Elliott, 1998; Korn & Graubard, 1991; Winship & Radbill, 1994). We selected parents who had at least one noncoresident adult child (aged 19+) who gave valid responses to questions about contact and proximity to children.

The data file was organized with adult children as the units. Due to the small number of very large families, children listed ninth and higher are excluded because their inclusion prohibits estimation convergence for an unstructured within-family error covariance matrix. This does not eliminate any families and loses only 20 cases (i.e., adult children). Families thus constructed have one to eight noncoresident adult (aged 19+) children. We retain families with only one noncoresident child because our statistical analyses are applications of multilevel/mixed models that estimate unique coefficients for each child while allowing those coefficients to be correlated across siblings. Following this approach, restricting the sample to families with two or more children is not necessary and would potentially bias estimates away from population parameters. We apply casewise deletion because missing values are minimal, yielding 9,474 observations on adult children (3,604 children assigned as first child within family, 2,550 children as second, 1,559 as third, 867 as fourth, 467 as fifth, 244 as sixth, 125 as seventh, and 58 as eighth child) of 3,604 parent respondents.

Assignment of adult children to a particular sibling position is based on the NSFH data file order of the child records. As a consequence, this assignment reflects a strong within-family rank order of children’s ages. We find no difference in visiting or proximity between children assigned to the first position and children assigned to higher-order positions and relatively few statistically significant differences among other child-based attributes. This is described in more detail below.

Dependent Variable

Parent respondents reported on relations with each child. We focus on frequency of visiting with each noncoresident adult child: “During the last 12 months, how often did you see (child)?” (1 = “not at all”, 2 = “about once a year”, 3 = “several times a year”, 4 = “one to three times a month”, 5 = “about once a week”, 6 = “several times a week”). With the exception of the large step between the third and fourth categories, assuming an interval scale to these six response categories approximately maps frequency onto a proportionate scale, wherein the frequency of visiting doubles with each unit increase. The NSFH does not identify the motivator or location of contact (e.g., whether parent or child initiated the contact or where); nor were parents asked about their frequency of contact with coresident children. Approximately one-fourth of parents lived in the same household with an adult child, for whom contact was coded “not applicable.”

The percentage distribution of response categories, based on adult children as units (N = 9,474), shows that just under 27% of noncoresident adult children saw the parent respondent “several times a week,” about 15% visited with the parent “about once a week,” 17% visited “one to three times a month,” almost 23% visited “several times a year,” just over 11% saw the parent “about once a year,” and almost 8% had no visit with parents during the last year. From the perspective of the nearly 3,600 families in our sample, over 14% have at least one noncoresident adult child with whom the parent respondent had not visited in 12 months prior to the survey.

Independent Variables

Proximity of adult children to the parent is our primary independent variable. Parents reported “how many miles away from here” each child lived (recoded to units of 100 miles in our regression analyses). As in prior research, there is substantial heaping of noncoresident children at short distances; for example, almost 12% are a mile or less from the parent, over 17% are 2 miles or less, almost 30% are within 5 miles, and just under 41% are within 10 miles. The median is only 21 miles between parent and child. The distribution skew is revealed by the extreme location of the average distance at just over 400 miles (Panel A in Table 1).

Table 1.

Descriptive Measures: Means, Standard Deviations, Min and Max Values

	Mean	SD	Min	Max	Diff(1 vs 2+)
Panel A. Full sample (9,474 adult child cases from 3,604 families)
Parental/family characteristics (fixed/constant within families)
Frequency of visiting during last 12 months	3.99	1.61	1	6
Parent’s age (in years)	60.16	11.53	30	95
Parent’s gender (1 = female)	0.65	0.48	0	1
Parent’s marital status (1 = married)	0.61	0.49	0	1
Parent’s self-reported health status (1–5)	3.76	0.95	1	5
Parent’s education (in years)	11.05	3.33	0	20
Black (1 = yes)	0.17	0.37	0	1
Hispanic (1 = yes)	0.06	0.23	0	1
One or more adult child coresides with parent (1 = yes)	0.29	0.45	0	1
Number of children	4.51	2.32	1	13
Characteristics of adult children
Child’s marital status (1 = married)	0.66	0.47	0	1
Child’s gender (1 = female)	0.50	0.50	0	1
Distance (in 100-mile units)	4.13	9.46	0.01	80
Adult child’s parental status (1 = has 1 or more child)	0.69	0.46	0	1
Biological or step child (1 = bio/adopted)	0.87	0.34	0	1
Panel B. Children Designated as Sibling 1 (3,604 adult child cases)
Parental/family characteristics (fixed/constant within families)
Frequency of visiting during last 12 months	4.01	1.60	1	6	0.03
Parent’s age (in years)	59.18	12.50	30	95	−1.59*
Parent’s gender (1 = female)	0.65	0.48	0	1	0.00
Parent’s marital status (1 = married)	0.59	0.49	0	1	−0.03*
Parent’s self-reported health status (1–5)	3.79	0.94	1	5	0.05*
Parent’s education (in years)	11.36	3.27	0	20	0.49*
Black (1 = yes)	0.16	0.37	0	1	−0.01
Hispanic (1 = yes)	0.05	0.23	0	1	0.00
One or more adult child coresides with parent (1 = yes)	0.32	0.47	0	1	0.06*
Number of children	3.54	2.03	1	13	−1.57*
Characteristics of adult children
Child’s marital status (1 = married)	0.66	0.47	0	1	0.01
Child’s gender (1 = female)	0.49	0.50	0	1	−0.02
Distance (in 100-mile units)	4.13	9.24	0.01	80	0.00
Adult child’s parental status (1 = has 1 or more child)	0.70	0.46	0	1	0.00
Biological or step child (1 = bio/adopted)	0.91	0.29	0	1	0.06*

Notes. Diff (1 vs 2+) shows difference in mean between child designated as sibling 1 versus siblings 2–8. * indicates that the difference is statistically significant using p < .05 for two-tailed test.

Control variables include parent, adult child, and family characteristics which in part reflect underlying family factors. Parent characteristics include: age, gender (1 = female), race (dummy variables identifying Black and Hispanic), marital status (1 = married), education (no. of years) as an indicator of social class, and health (self-reported 1 = very poor to 5 = excellent). Individual adult child characteristics include gender (1 = female), biological/adopted or stepchild (1 = biological/adopted), marital status (1 = married), and parental status (1 = has at least one child; i.e., parent’s grandchild). Family characteristics include (parent) respondent’s number of children and whether the respondent coresides with one or more adult children (1 = any coresident).

Panel A in Table 1 reports descriptive measures for our sample of 9,474 adult child cases. Mean age of their parents is 60.2 years, 65% of parent respondents are female, 61% are currently married, 29% coreside with an adult child, and the mean number of adult children per parent respondent is 3.7. One-half of their adult children are daughters, 17% are Black, 6% are Hispanic, 87% are biological children, 66% are married, and 69% have children.

Data Structure and Cross-Sibling Interactions

We adopt several analytic strategies to assess our research questions. Our primary approach fits within a multilevel modeling strategy derived for dyadic data and random errors models for multiple potential responses (e.g., Agresti & Liu, 2001; Kenny et al., 2002; Kurdek, 2003). Conventional multilevel model file structures have one row for each level-one case, a variate that identifies the level-two aggregate to which the case belongs, and a vector of shared attributes. This “long” structure typically results from transposition of an original “wide” file that has one row for each aggregate (e.g., family) and blocks of vectors of attributes for each subject (e.g., family member) within the aggregate. Our file structure is distinguished by being both long and wide. A wide file structure supports the estimation of multivariate analyses, that is, simultaneous regressions with multiple dependent variables and, potentially, different sets of predictors for each outcome. We use the MIXED procedure in the SAS system to estimate multivariate multilevel/mixed regression models that treat the dependence between sibling pairs as within-family random effects (SAS Institute, Inc., 2008).

We simultaneously estimate eight interrelated child-specific equations, each with an N determined only by the number of noncoresident adult children assigned to that sibling order, and each with its own set of predictors that may include attributes from other children (siblings). We refer to these terms as “cross-sibling interactions” because they allow us to directly examine likely sources of within-family dependencies. For example, we examine whether distance between parent and one child affects visiting with his/her siblings by including the selected child’s distance as a covariate in the equations of each of the other children.

We adopt this approach because it allows us to retain each family context from one to eight noncoresident adult children. Alternative methods for joint-system estimation either sample pairs of siblings from their larger family context or must eliminate families with fewer siblings than the chosen family size and select a subset of siblings from larger families in order to estimate within-family error covariances. For example, joint-system estimation for four siblings would eliminate families with fewer than four noncoresident adult children and impose some selection (preferably, random) rule to extract only four siblings from larger families.

Because we utilize selected characteristics of the child assigned to the first sibling position on behaviors of his/her siblings, we assessed whether there are systematic differences between them and their siblings (Panel B of Table 1). As expected, given our retention of families with one or more noncoresident adult children, first-order children have fewer siblings and younger, more highly educated parents, but other differences are nonsignificant or only modest; and, most importantly, there are no differences in frequency of visiting or distance. Thus, our modeled cross-sibling interactions are representative of the effects that another focal sibling would exert.

Assessing the Functional Form of Proximity

Despite extensive research on the importance of proximity in determining in-person contact, there has been surprisingly little effort to assess the functional form of proximity. We investigate more thoroughly how distance between children and parents influences visiting. Our method begins by fitting polynomial spline regressions for unknown number and locations of knots, and the functional form of the line segment between knots (Marsh & Cormier, 2001; Marsh, Maudgel, & Raman, 1990; Messner et al., 2005). This gives unique insight into the association between proximity and contact. If the relationship is linear, after adjusting for parent, child, and family covariates, this form is identified as the best fit. If the relationship is nonlinear, this method will evaluate forms ranging from a quadratic polynomial through piecewise models with all possible change points and segment shapes.

Results

Our first research question, derived from the compensation and enhancement models, asks whether and how visiting is correlated across adult children. A conditional means model (a conventional multilevel/mixed random effects model that controls for parent, family, and child characteristics) sets the stage for addressing these questions by allowing a variance decomposition of visiting (Singer & Willett, 2003, p. 96). The intraclass correlation coefficient, $\widehat{\rho },$ derived as the ratio of the between-sibling variance to the sum of the between- and within-sibling variances, is 0.30; about one-third of the outcome variation, net of parent, family, and child characteristics, is attributable to the association between contact with one child and contact with others. This evidence is consistent with the enhancement model.

This evidence of between-sibling dependencies, however, does not show how visiting is associated with parent, child, and family attributes. Table 2 reports results from four specifications that bring in information from these covariates. Model 1 is our multivariate baseline model in which visiting is regressed on parent, family, and child attributes, including the optimal form for the association between visiting and proximity. Model 2 adds the frequency of visiting of the child assigned to the first equation to the sibling equations (equations 2 through 8) to assess whether the within-family dependence in visiting evidences reciprocal causation (i.e., visiting begets visiting) or whether the positive within-family correlation is due to remaining (unmeasured) common family factors. Model 3 completes assessment of our initial research question via indicator variables (dummy variables) in other siblings’ equations of whether the child assigned to the first position of sibling order had visited infrequently (“about once a year” or “several times a year”), with moderate frequency (“one to three times a month” or “about once a week”), or frequently (“several times a week”), using “not visited” within the year as the reference category. Our second research question is addressed by examining results presented in Model 1 (see further discussion in section on distance below). Model 4 addresses our third research question by extending Model 3 with a cross-sibling interaction that adds distance between parent and the first-order child to the other children’s equations.

Table 2.

Simultaneous Equation Multilevel Regression Models of Visiting Between Parents and Children on Parent-, Family-, and (Adult) Child Attributes

	Model 1: baseline model with nonlinear (spline) distance		Model 2: Model 1 + cross-sibling interaction with sibling 1’s frequency of visiting		Model 3: Model 1 + cross-sibling interaction with selected visiting patterns by sibling 1		Model 4: Model 3 + cross-sibling interaction with sibling 1’s distance
	B (SE)	t-Ratio	B (SE)	t-Ratio	B (SE)	t-Ratio	B (SE)	t-Ratio
Parental/family characteristics
Parent age	0.0004 (0.001)	0.26	0.0004 (0.001)	0.33	0.0003 (0.001)	0.19	0.0003 (0.001)	0.21
Parent gender	0.134 (0.028)	4.73*	0.133 (0.028)	4.71*	0.131 (0.028)	4.67*	0.131 (0.028)	4.72*
Parent married	0.244 (0.029)	8.47*	0.243 (0.029)	8.46*	0.241 (0.029)	8.44*	0.241 (0.028)	8.54*
Parent health	0.038 (0.015)	2.63*	0.038 (0.015)	2.64*	0.038 (0.014)	2.60*	0.036 (0.014)	2.51*
Parent education	0.014 (0.005)	3.17*	0.015 (0.005)	3.29*	0.015 (0.005)	3.20*	0.014 (0.005)	3.12*
Number of children	−0.049 (0.007)	−6.54*	−0.048 (0.007)	−6.43*	−0.048 (0.007)	−6.47*	−0.048 (0.007)	−6.53*
Coresident child	0.095 (0.034)	2.79*	0.094 (0.034)	2.76*	0.094 (0.034)	2.77*	0.093 (0.033)	2.78*
Black	−0.066 (0.038)	−1.73	−0.065 (0.038)	−1.71	−0.065 (0.038)	−1.74	−0.063 (0.038)	−1.69
Hispanic	0.041 (0.060)	0.69	0.041 (0.060)	0.69	0.044 (0.060)	0.74	0.042 (0.059)	0.72
Characteristics of adult children
Child gender	0.129 (0.019)	6.87*	0.129 (0.019)	6.87*	0.128 (0.019)	6.80*	0.128 (0.019)	6.79*
Child married	0.022 (0.022)	0.98	0.022 (0.023)	0.96	0.019 (0.023)	0.84	0.019 (0.023)	0.84
Child has kids	0.029 (0.024)	1.19	0.029 (0.024)	1.20	0.031 (0.024)	1.29	0.032 (0.024)	1.32
Bio/adopted child	0.645 (0.036)	18.02*	0.644 (0.036)	17.98*	0.641 (0.036)	17.95*	0.641 (0.036)	17.98*
Focal covariates
Spline terms for proximity effect: distance (main term)	−31.550 (3.788)	−8.35*	−31.580 (3.789)	−8.34*	−31.725 (3.790)	−8.37*	−31.938 (3.796)	−8.41*
effect change at 2 miles	24.959 (3.989)	6.26*	24.987 (3.998)	6.25*	25.161 (3.999)	6.29*	25.398 (4.006)	6.34*
effect change at 13 miles	4.883 (0.412)	11.85*	4.884 (0.413)	11.83*	4.854 (0.413)	11.75*	4.829 (0.414)	11.67*
effect change at 83 miles	1.431 (0.089)	16.04*	1.431 (0.089)	16.00*	1.433 (0.090)	16.00*	1.433 (0.090)	15.97*
effect change at 300 miles	0.208 (0.035)	5.90*	0.208 (0.035)	5.89*	0.207 (0.035)	5.84*	0.208 (0.035)	5.88*
effect change at 820 miles (linear term)	0.044 (0.014)	3.13*	0.043 (0.014)	3.10*	0.044 (0.014)	3.12*	0.043 (0.014)	3.05*
effect change at 820 miles (squared term)	0.0003 (0.00007)	3.76*	0.0003 (0.00007)	3.71*	0.0003 (0.00007)	3.37*	0.0003 (0.00008)	3.50*
Sibling 1’s frequency of visiting			0.008 (0.010)	0.80
Sibling 1 infrequent visiting					0.148 (0.062)	2.40*	0.207 (0.063)	3.28*
Sibling 1 moderate frequency of visiting					0.107 (0.063)	1.70	0.209 (0.069)	3.02*
Sibling 1 frequent visiting					0.135 (0.065)	2.06*	0.256 (0.071)	3.59*
Sibling 1’s distance							0.006 (0.002)	2.55*
Intercept	4.855 (0.144)	33.59*	4.824 (0.146)	33.01*	4.794 (0.146)	32.74*	4.742 (0.146)	32.52*
	−2LL = 25140.3		−2LL = 25133.5		−2LL = 25106.7		−2LL = 25091.4
			∆−2LL = 6.8, ∆df = 7		∆−2LL = 33.6*, ∆df = 21		∆−2LL = 15.3*, ∆df = 7

Notes. Weighted average of estimated regression parameters and standard errors (shown in parentheses) are reported. Analytic sample is N = 9474 adult non-coresident children. Asterisks indicate test statistics, t-ratios (two-tailed test) and ∆−2LL, are statistically significant at p < .05; −2LL is based on maximum likelihood estimation (MLE) and model differences relative to additional parameters (Model 2 and Model 3 vs Model 1; Model 4 vs Model 3) indicate improvement in fit; parameter estimates shown in table are based on restricted MLE (REML).

Before further evaluating our specific research questions, we can make the following statements about controls regardless of the model: more frequent visiting is reported by mothers than fathers and with daughters compared with sons; and by parents who are married, have fewer children, are in better health, and have more education. Parents also report more visiting with biological children than stepchildren and if there is (one or more) coresidential adult child. Parental age, adult child’s marital and parental status, and race/ethnicity are not related to parent reports of visiting.

Frequency of Visiting

Model 2 does not support reciprocal causation in visiting among siblings. Net of parent-, family-, and child-covariates, frequency of visiting by one child does not directly affect visiting by his/her siblings. Model 2 suggests that enhancement results from shared factors predicting frequency of visiting rather than visiting per se.

This conclusion is also supported in a supplemental analysis reported in the Supplementary Table 1. We reexamined this research question using instrumental-variables regression via two-stage least squares (2SLS) because a regression in which the frequency of visiting by one child is a predictor of frequency of visiting by a sibling likely violates the OLS assumption that the regressors must be exogenous (or predetermined; We thank a reviewer for expressing this concern and suggesting alternative analytic sample structures and appropriate methods for endogenous regressors.). Accordingly, we randomly selected two siblings from each family with two or more noncoresident adult children (N _p = 2,550 sibling pairs), with random assignment as first or second child in the sibling pair. The resultant 2SLS (second-stage) estimates are reported in the Supplementary Table 1. While diagnostics reported at the bottom of the Supplementary Table 1 provide evidence in support of treating sibling 1’s frequency of visiting as an endogenous regressor (as indicated by DWH tests), estimated slopes and their standard errors, with the single exception of the term for coresident child(ren), closely resemble those reported in Model 2 in Table 2. Notably, sibling 1’s frequency of visiting, treated as an endogenous regressor, is nonsigificant.

Our first research question, whether visiting with one child increases or decreases visiting with other children, also asked whether some visiting patterns have a particular effect. Auxiliary analyses (not reported but available on request) treating visiting as a cross-sibling interaction revealed that the “not at all” category exerted a particularly strong response. Model 3 capitalizes on this insight by adding a set of dummy variables to Model 1 to reflect categorical levels of visiting (infrequent, moderate frequency, frequent visiting, using “not visited within the last year” as the reference category) of the child in the first sibling position to the higher order siblings’ equations. Inclusion of these cross-sibling interactions improves the overall model fit (∆−2LL relative to ∆df at bottom of Model 3) and shows that more frequent visiting, when compared with the “not visit” response, increases siblings’ visiting, net of controls. This pattern, like the positive intrafamily correlation we reported, is consistent with an enhancement model for contact.

Distance

Our second research question addresses whether and how distance is related to visiting with adult children. Our preliminary spline regression identified a high degree of discontinuity in the association between distance and visiting, with significant breaks at 2, 13, 83, 300, and 820 miles. The segments for less than 2, 2–13, 13–83, 83–300, and 300–820 miles were found to be linear, while the final segment (820 miles and beyond) was best fit as a quadratic. The negative association between distance and visiting weakens following each breakpoint, with the final segment essentially flat. To illustrate, the coefficients in Model 1 yield the following solutions to segment-specific derivatives (slopes): in the initial segment, visiting decreases by 0.314 per mile; between 2 and 13 miles, visiting decreases by about seven-tenths of a point per 10 mile; between 13 and 83 miles, the decline in visiting further slows to a little less than two-tenths of a point per 10 mile change; the decline in visiting is about three-tenths of a point per 100 mile unit change between 83 and 300 miles; and between 300 and 820 miles, visiting decreases approximately 0.07 per 100 mile. Beyond 820 miles the negative slope continues to flatten, reaching its root at approximately 950 miles and then becoming positive (We also examined the fits of linear and (natural) log forms of distance. As shown at the bottom of Model 1 in Table 2, the spline specification yielded a −2LL = 25140.3 (AIC = 25548.3). Linear distance gave −2LL = 31608.2 (AIC = 31920.2) and log distance fit was −2LL = 25339.4 (AIC = 25651.4). Smaller fit statistics indicate better fit.). Solutions for the effect of distance in Models 2, 3, and 4 are nearly identical.

Cross-sibling Interaction: Distance

Our third research question asks whether one child’s proximity affects visiting reported with other children. Model 4 in Table 2 extends Model 3 by adding (linear) distance between parent and the child in the first sibling position to the higher order siblings’ equations (Linear distance in the cross-sibling interaction was preferred over alternative forms. Model fit of linear distance was −2LL = 25091.4 (AIC = 25555.4) compared with −2LL = 25103.7 (AIC = 25567.7) for log distance and −2LL = 25031.2 (AIC = 25579.2) for spline form of distance. Given the additional loss of 42 parameters to fit the spline form of distance over linear or log distance, smaller AIC recommends linear distance.). Although the magnitude is small, the cross-sibling interaction between a focal child’s distance and other children’s visiting is positive and statistically significant, even after adjusting for parent, child, and family attributes and accounting directly for cross-sibling association in visiting. Thus, while distance between parents and a child reduces visiting with that child (in a highly nonlinear fashion), distance from one child increases visiting with that child’s siblings; and conversely, closer proximity between parent and one child reduces visiting with other children. This suggests a countervailing pattern of compensation for proximity against the enhancement observed for visiting, as we suggested in our presentation of research questions.

Discussion

Parents and adult children interact within the context of other family relationships, but much work on these relations has focused on individual parent–child dyads. Recent studies have shown significant within-family differentiation in these relations, and further, that relations between parents and one or some children may affect relations with others (e.g., Matthews, 2002; Silverstein et al., 2008; Spitze et al., 2012; Suitor & Pillemer, 2007; Tolkacheva et al., 2010). Much of this research has examined patterns of caregiving and assistance to frail older parents. We take a broader approach by focusing on visiting, a central aspect of parent–adult child relations that is related to other dimensions of solidarity (e.g., Sarkisian & Gerstel, 2008; Silverstein & Bengtson, 1997), and parents who are mostly healthy and whose ages range widely. Our focal predictor is geographic proximity, which is strongly related to multiple dimensions of parent-adult child relations (e.g., Bengtson, 2001; Eggebeen, 2005; Hank, 2007). We use a multi-equation multilevel modeling approach, extending previous analyses of family structural effects to more directly observe how relations of one parent–adult child pair may affect other pairs.

We ask several questions. First, is visiting correlated across adult children and if so, does visiting with one child have a particular effect on visiting with siblings? We find that parent reports of visiting with adult children are positively correlated across siblings, even after accounting for underlying parent, child, and family factors. In particular, we also find direct evidence to support an enhancement model: higher frequency of visiting (relative to “not visiting”) with one child is positively related to visiting with other children.

Second, how does an adult child’s proximity affect visiting with that child? As in past research (e.g., Crimmins & Ingegneri, 1990; Hank, 2007; Logan & Spitze, 1996), we find that frequency of visiting reported by parents declines with distance, but we are able to show more precisely the functional form of that association. The effects of distance are most pronounced at closer proximity, with substantial differences in visiting associated with only a few miles distance. Further, there are strong nonlinear discontinuities in distance effects, as the negative association with visiting weakens at greater distances.

Third, does one child’s proximity to parents affect visiting with other children? If a child lives near parents, are visits with other children more or less frequent? We show that distance between parents and a focal child is positively associated with parent reports of visiting with other children. Thus, although children’s visits are positively correlated, and distance reduces visits, if a child lives closer to a parent reported visits with other children are less frequent.

This effect of proximity notwithstanding, we also find a positive effect of coresidence with adult children that was not anticipated from our review of previous research. The NSFH lacks information on contact frequency with coresident children, but sharing a household likely yields frequent contact with parents. Such contact, or perhaps simply the presence of a sibling, appears to be a draw to the parental household for other noncoresident children. This offers further evidence for an enhancement model.

We also report associations of other parent, child, and family factors with parent–adult child visiting. These mostly conform to expectations based on literature reviewed earlier: parent reports of visiting are positively related to parent education and are higher for mothers and with daughters, for parents who are married and in better health, and with biological children; visits are less frequent with a particular child in larger families; race/ethnicity, and parent age are not related to visiting. We find no association between visiting and child marital or parental status, though we find elsewhere that children becoming married or a parent leads to reduced contact with parents, at least in the short term (Ward et al., 2014).

As we have suggested, the empirical patterns linked to our research questions can be framed within the enhancement and compensation models discussed earlier. The substantial within-family homogeneity indicated by the intraclass correlations, as well as the direct association shown in the dummy variables for more frequent visits relative to not visiting with one child and visits with other children, indicate that overall family dynamics are consistent with an enhancement model: visiting with one child appears to encourage contact with others. This is evident with controls for pertinent shared family factors, consistent with a view that contact with adult children directly increases contact with others; this likely at least partly reflects joint visiting and family get-togethers. Similarly, our previous work has found that socioemotional support from parents to one adult child appears to encourage (enhance) support to other children (Spitze et al., 2012). It should be emphasized, however, that these enhancement associations do not necessarily imply cordial family relationships. As we noted in the introductory discussion, a positive correlation for visiting is also consistent with a dysfunctional family dynamic; and we found that having a sibling with no visits appears to discourage visits with others. Indeed, our reference to the solidarity paradigm is not meant to ignore the existence of conflict and ambivalence in parent–adult child relations.

The cross-sibling interaction for distance, whereby one child’s farther distance leads to more visits reported with others, reminds us that even within an overall climate of enhancement, countervailing (though weaker) compensatory family dynamics can operate. This parallels another of our previous findings, that receiving socioemotional support from one child was negatively related to support from other children. These patterns suggest cooperative (or competitive?) dynamics among children, greater closeness felt between parents and some siblings than with others, or less felt obligation when another sibling lives closer (Spitze et al., 2012). This effect of distance would be consistent with the kind of compensatory negotiation that may occur among children in allocating assistance and caregiving (Silverstein et al., 2008).

Many of these questions revolve around issues of proximity between parents and all of their adult children, as well as the location of adult children relative to each other. We devoted substantial thought to the coding of proximity, as have others reviewed earlier. Konrad, Künemund, Lommerud, & Robledo (2002) discuss the “geography of families” more broadly, theorizing about how multiple children’s locations might respond cooperatively to parents’ needs. This problem quickly becomes complex; their theorizing and data focus on families with a maximum of two children. And we may not be able to assume that children’s relative locations imply cooperation. We have also noted endogeneity issues regarding joint determination of proximity and contact (or other relationship dimensions). Thus, many questions are yet to be addressed about how extended families make choices about residential location and moves, and how the presence of other family members may influence these decisions.

The NSFH survey analyzed here has unusually detailed reports of parent relations with each adult child, making it especially pertinent for our research questions, but it is situated in time and place. The average NSFH parent respondent (born about 1930) had children during the 1946–1964 baby boom, with larger families (mean = 3.54 adult children) that reflect their higher fertility (Fingerman, Pillemer, Silverstein, & Suitor, 2012), as well as blended families from remarriage. However, the median number of adult children is 3, and there is a range of parents born across five decades that includes cohorts before and after the baby boom with lower fertility. But patterns may differ some for recent cohorts not represented here; for example, lower fertility may yield greater contact with older parents (Treas & Gubernskaya, 2012). We did find greater parent–adult child contact in smaller families; and sibling contact may have greater influence in smaller families. In terms of family stage, the average NSFH family had middle-aged parents (mean age of 59) with relatively young adult children. Although we did not find that parent age was related to contact, patterns of contact and the role of predictors may change as the situations and needs of parents and adult children change with age.

Patterns in the NSFH data are also based in the U.S. context. Patterns may be different in other societal contexts with different family norms or patterns of mobility. There may be more frequent contact in Europe (Hank, 2007; Kalmijn, 2006), for example, that may partly reflect greater geographic dispersion in U.S. families.

There are other limitations to what we have been able to do here. “Visit” refers to parent reports of getting together adult children, but we do not know the location or how the visit was initiated. We also lack information about feelings and motivations that may account for visiting behaviors, a caveat that has also been noted for understanding patterns of parent-child support (Silverstein et al., 2008; Tolkacheva et al., 2010). Third, parents may give more positive reports of their relations with children (including somewhat inflated estimates of visiting) than would their children (Aquilino, 1999; Logan & Spitze, 1996).

Finally, our approach involves some advantages but also some limitations. Although causal inference can be strengthened by using (within-family) fixed effects models, we use random effects estimation. This allows explanation of between-family variation as well as within-family variation, but at the cost of susceptibility to omitted variable bias. We also treat distance and cross-sibling interactions as exogenous to contact which could introduce bias under reciprocal causation. Our model diagnostics, however, show almost no difference between estimates from our approach and those from methods for system estimation, while our sibling relations are not restricted by the selection of a single family size or structure that is required by joint-system estimators.

These results suggest additional directions for research. First, it would be useful to learn more about the process through which intergenerational contacts occur. This may require more detailed questions asked of one or both generations, or it may require other approaches, such as Matthews’ qualitative interviews (2002). Second, we focus here on in-person contact, but forms of parent-adult child contact are evolving with new technologies such as cell phones, texting, and social media. The measurement and meaning of these new forms of contact is likely to be a continuing subject of investigation (Treas & Gubernskaya, 2012). Other relationship dimensions can also be analyzed using this approach, including emotional closeness and support (in each direction), such as personal care, household assistance, or financial assistance. These dimensions may be analyzed to examine whether intergenerational relations operate independently across adult children, or have enhancement or compensation effects; as noted previously, visiting and socioemotional support exhibit both similar and different cross-sibling patterns. The role of proximity likely also varies for different dimensions; for example, we have found no cross-sibling effects of proximity for exchanges of socioemotional support (Spitze et al., 2012). Distance may also matter less for financial assistance and some forms of contact, but is likely to be influential for types of instrumental assistance that require proximity or frequent travel.

The principal conclusion here is that factors associated with one adult child affect relations between parents and other adult children. It is apparent that family structure and the network of family relationships need to be considered and accounted for in theory and research on family intergenerational relations.

Supplementary Material

Supplementary material can be found at: http://psychsocgerontology.oxfordjournals.org/

Funding

Data in this article are from the National Survey of Families and Households, funded by grants from the Center for Population Research of the National Institute of Child Health and Human Development (HD21009) and the National Institute on Aging (AD10266). The survey was designed and carried out by James Sweet and Larry Bumpass, Center for Demography and Ecology, University of Wisconsin-Madison. Analyses in this article were supported in part by a grant (1 RO3 HD048451-01-A2) from the National Institute of Child Health and Human Development to the University at Albany, Glenna Spitze, Principal Investigator, and Glenn Deane and Russell Ward, Co-Investigators.