Applying Within-Family Differences Approaches to Enhance Understanding of the Complexity of Intergenerational Relations

Abstract

Objectives

The role of family relationships in the lives of older adults has received substantial attention in recent decades. Scholars have increasingly looked beyond simple models of family relations to approaches that recognize the complex and sometimes contradictory nature of these ties. One of the most exciting conceptual and methodological developments is the application of within-family differences approaches. In this paper, we focus on the ways in which such within-family approaches can extend the understanding of patterns and consequences of intergenerational ties in adulthood.

Method

Following a review of the conceptual underpinnings of within-family differences approaches, we provide empirical illustrations of these approaches from three projects conducted in the United States: the Family Exchanges Study (FES), the Longitudinal Study of Generations (LSOG), and the Within-Family Differences Study (WFDS).

Results

Analyses from the FES, LSOG, and WFDS reveal differences in the consequences of patterns of intergenerational relations found when using within-family compared to between-family approaches. In particular, these analyses demonstrate considerable variation within families that shapes patterns and consequences of parent-adult child ties that is masked when such variations are not taken into account.

Discussion

Within-family differences approaches have been shown to shed new light on intergenerational relations. Despite the value of within-family designs, their use may be limited by the higher investment of finances and time required to implement such studies.

Relations between the generations have been a core component of gerontology since this field of scholarship arose. The 1970s and 1980s saw the development of theoretical perspectives that have guided the study of intergenerational relations for the past half-century. Most prominently, these include the family solidarity model (Bengtson, Olander, & Haddad, 1976) and life course perspectives on the family (Conger & Elder, 1994; Elder, 1985, 1994). A consistent theme in research guided by these theoretical perspectives has been a movement from simpler models of family relationships to orientations that take complexity into account (Suitor, Gilligan, & Pillemer, 2015), including an emphasis on dissensus as well as consensus, and on the ways in which family members’ lives are inextricably linked over the life course (Moen & Hernandez, 2009).

One of the most innovative conceptual and methodological developments that has emerged from the increasing emphasis on the complexity of adult families is the study of variations in intergenerational relations within families. Until a decade ago, most studies asked parents about their relationships with their adult children in the aggregate or asked parents to report on their relationship with only one target child (Suitor et al., 2015). Scholarship from developmental psychology, however, has long suggested that such an approach masks variations in parent–child relationships within the family, thus providing an incomplete and potentially inaccurate picture of relationships in intergenerational families (cf., Suitor, Sechrist, Plikuhn, Pardo, & Pillemer, 2008). In particular, beginning in the 1980s and 1990s, studies of younger families showed that parents of young and adolescent children differentiated between their children in terms of both affection and disapproval (Suitor et al., 2008). A small number of studies conducted in the same period indicated that such variations in parent–child relations within the family continued into adulthood (cf., Aldous, Klaus, & Klein, 1985). However, it was not until the early 2000s that attention was devoted to understanding the prevalence, predictors, and consequences of within-family differences in the middle and later years.

The study of within-family differences in adulthood is grounded in classic theories of social interaction in both sociology (Simmel, 1964) and psychology (Heider, 1958), which can be used to argue that the relationship between a parent and any one of his or her adult children is likely to be affected by the parent’s relationships with other adult children in the family. The within-family approach also draws from two other closely-related theoretical perspectives—family systems theories (Bowen, 1978; Cox & Paley, 1997), which focus on the interconnectedness of family ties and life course theories (Conger & Elder, 1994; Elder, 1985, 1994) which emphasize linked lives between family members both within and across generations. In bringing together principles of these classic theoretical perspectives, within-family approaches provide opportunities for studying intergenerational family ties in unique ways that shed new light on the complexity of kin ties. We use the term “approach” throughout the paper to encompass both the conceptual and methodological dimensions of studying within-family variations.

Scholars can apply within-family approaches to studying intergenerational relations using a variety of designs. For example, within-family data can be collected from a parent about her or his relationship with each child in the family (cf., Fingerman, Miller, Birditt, & Zarit, 2009; Suitor, Gilligan, & Pillemer, 2013a; Suitor et al., 2016; Ward, Spitze, & Deane, 2009), from multiple adult children in a family regarding their parent(s) (cf., Gilligan, Suitor, Kim, & Pillemer, 2013; Silverstein, Conroy, Wang, Giarrusso, & Bengtson, 2002; Suitor, Gilligan, Peng, Jung, & Pillemer, in press), from parents and multiple adult children in the same families regarding one another (Birditt, Hartnett, Zarit, Fingerman, & Antonucci, 2015; Fingerman et al., 2011), or from one or more members of multiple generations looking up and/or down the generations (Kim, Eggebeen, Zarit, Birditt, & Fingerman, 2013; Lendon, Silverstein, & Giarrusso, 2014; Suitor et al., 2013a; Suitor et al., 2016). Thus far, this line of research has shown that there is substantial variation in parents’ relationships with adult children in the same family across a wide range of dimensions, including emotional closeness (Birditt, Tighe, Fingerman, & Zarit, 2012; Suitor et al., 2013a; Suitor et al., 2016; Ward et al., 2009), ambivalence, tension, and disappointment (Birditt, Fingerman, & Zarit, 2010; Fingerman, Pitzer, Lefkowitz, Birditt, & Mroczek, 2008; Pillemer, Munsch, Fuller-Rowell, Riffin, & Suitor, 2012; Suitor et al., 2016), contact (Fingerman, Huo, Kim, & Birditt, in press; Ward, Deane, & Spitze, 2014), exchange of support (Cong & Silverstein, 2011; Fingerman et al., 2011; Spitze, Ward, Deane, & Zhuo, 2012; Suitor, Pillemer, & Sechrist, 2006), and preferences for support (Cong & Silverstein, 2014; Suitor & Pillemer, 2013; Suitor, Sechrist, & Pillemer, 2007; Suitor et al., 2013a, 2013b).

Given the strong conceptual underpinnings of the within-family approach, it is not surprising that studies that have employed it have shed new light on intergenerational relations. However, considering the substantial costs in terms of both time and money required to conduct such studies, it is important to consider whether, in fact, they extend our understanding of family processes and outcomes beyond what can be learned using standard between-family approaches.

Our goal in this paper is to present three well-known studies of intergenerational relations to demonstrate how within-family approaches make unique and important contributions to the study of middle and later-life families. To this end, we used data from the Family Exchanges Study (FES), the Longitudinal Study of Generations (LSOG), and the Within-Family Differences Study (WFDS) to address the same broad research question: How do within-family differences in ties between the generations affect well-being?

First, we present daily diary data from the FES to examine the implications of daily encounters with one or more grown children suffering problems on parents’ psychological well-being. Second, we use data from the LSOG to examine generational change and stability in religiosity over three decades, and their association with relational well-being between grandchildren and their grandmothers. Third, we use data from the WFDS to illustrate the effects of relationship quality with mothers on adult children’s psychological well-being using single-dyad standard measures of closeness and conflict, compared to within-family approaches that capture children’s perceptions of maternal favoritism and disfavoritism.

Taken together, these three studies provide strong evidence that within-family approaches provide greater insight regarding intergenerational processes and psychological and relational well-being than can be revealed using standard between-family designs that focus on a single family member or a single parent-child dyad. Further, the diversity of theoretically grounded research questions and applications of within-family designs across these three studies demonstrate that this method can be used fruitfully to study a single point in time or to study changes and continuities across generations and decades.

Family Exchanges Study

The FES included 633 three-generation families interviewed in 2008 and 2013. Anchor participants were midlife adults (aged 40–60) with a living parent and at least one grown child. Heavy recruitment in high density minority and lower income neighborhoods generated a sample that was over one-third African American and represented the full range of socioeconomic backgrounds. The anchor participant’s grown children, parents, and romantic partners also were invited to participate. (For a full description of FES procedures, see http://sites.utexas.edu/adultfamilyproject).

In both waves, FES participants completed a global survey (by phone or internet) assessing relationships with each aging parent and grown children. In wave 2, a random subset of participants (230 offspring, 247 midlife adults, and 207 aging parents) completed a 7-day diary study by phone, reporting on daily experiences with family members. This study examines the midlife adults who completed the diary study (Mean age = 56.42 years, SD = 4.96; See left column of Table 1 for sample description).

Table 1.

Demographic Information for FES, LSOG, and WFDS^a

	FES	LSOG	WFDS
Parents	(N = 247)	(N = 191 Grandmothers @T1)	(N = 95 Mothers in full families)
Age in years, M (SD)	56.42 (4.96)	42.84 (4.43)	77.73 (3.02)
Mothers, %	56	100	100
Race, %
Black	24	1	6
White	71	65	94
Other	5	28	—
Married, %	72	93	48
Years of education, M (SD)	14.51 (2.03)	—	—
Education, %
Less than high school	2	13	8
High school graduate	24	38	37
Post-high-school vocational	—	—	6
At least some college	28	36	13
College graduate	23	3	19
Some graduate school	22	10	17
Employed, %	61	49	—
Number of children, M (SD)	2.77 (1.47)	3.67 (1.61)	3.18 (1.35)
Adult children	(N = 627)	(N = 311 Parents @T3)	(N = 253 within 95 full families)
Age in years, M (SD)	28.72 (6.81)	36.30 (2.88)	49.27 (5.43)
Daughters, %	52	64	52.9
Married, %	29	81	76
Years of education, M (SD)	14.60 (1.92)	—	—
Education, %
Less than high school	4	3	4
High school graduate	18	24	12
Post-high-school vocational	—	4	4
At least some college	32	28	12
College graduate	31	24	38
Some graduate school	16	17	30
Employed, %	60	82	86
Parents	37	100	75
Grandchildren		(N = 502 @T8)
Age in years, M (SD)	—	28.20 (5.11)	—
Daughters, %	—	57	—
Married, %	—	46	—
Education, %
Less than high school	—	4	—
High school graduate	—	10	—
Post-high-school vocational	—	7	—
At least some college	—	38	—
College graduate	—	26	—
Some graduate School	—	14	—
Employed, %	—	81	—
Parents	—	41	—

Notes: FES = Family Exchanges Study; LSOG = Longitudinal Study of Generations; WFDS = Within-Family Differences Study.

^aData were not weighted in any of the analyses included in the paper.

The FES was designed to examine individuals’ perceptions of multiple family relationships in different generations, discrepancies in partners’ perceptions of relationships, and the overall family milieu (Fingerman et al., 2009; Fingerman et al., 2011). FES has a psychological focus, with an emphasis on family member’s subjective feelings about exchanging support and emotional qualities of relationships. In particular, FES has focused on grown children’s problems.

Children’s Problems and Daily Well-Being

Research has established that when a grown child suffers a major life problem (e.g., divorce, major health problem, addiction, job loss), parents experience diminished well-being (Fingerman, Cheng, Birditt, & Zarit, 2012; Umberson, Pudrovska, & Reczek, 2010). The FES asked how parents’ relationships with children who suffered problems differed from their relationships with children who did not. For example, Fingerman and colleagues (2009) found parents provide grown children suffering problems with more frequent practical, emotional, and financial support than their children not suffering problems. Similarly, parents feel as much affection for their grown children who suffer problems as they do for children not suffering problems, but also experience greater conflict with problem children (Birditt et al., 2010).

More important, in prior studies, the FES allowed researchers to test a threshold model (i.e., simply having one child with a problem pushes a parent over a threshold where well-being suffers) as well as a cumulative model (i.e., whether having more children with more problems is associated with even poorer well-being). Finally, researchers considered a mitigating model in which a successful child mitigates effects of a grown child suffering problems; the successful child might support the parent, help their sibling suffering problems, and offer “parental bragging rights.” For general well-being, the threshold model mattered; having one grown child suffering a life crisis was associated with poorer parental well-being. The cumulative model also garnered support—having additional children suffering such problems was associated with even poorer well-being (Fingerman et al., 2012).

In the present study, we asked whether grown children’s problems affect parents’ daily well-being. Parents and grown children typically talk or get together several times a week (Fingerman et al., in press; Fingerman, Kim, Birditt, & Zarit, 2016). On any given day, we expected parents to be more likely to experience stressful interactions or stressful thoughts with grown children suffering problems than with grown children not suffering such problems. This may be the case because parents are more likely to be involved helping children suffering problems and because such children are more likely to serve as a source of tensions (Birditt et al., 2010; Fingerman et al., 2009). We expected children who do not suffer problems to be involved in enjoyable encounters with parents.

Further, whereas a parent may shrug off a brief annoyance with a grown child who is doing well, encounters with problem children are likely to evoke negative mood due to heightened emotional concerns regarding those offspring. Thus, midlife parents may report poorer mood on days when they experience stressful interactions with even one child who suffers problems (threshold model) and more so, when they have stressful interactions with many grown children suffering problems (cumulative model). Parents may show an attenuated response to stressful experiences with problem-ridden children on days when they have enjoyable encounters with other children not suffering problems (mitigating model).

Measures

In the global survey, midlife parents indicated whether each grown child had incurred 8 life problems in the past 2 years (e.g., divorce, health problem, and addiction). In the diary study, each day, parents responded to questions regarding encounters with each grown child that day (offspring N = 627; person-day N = 1,631), including positive encounters (laugh, enjoyable visit), stressful encounters (irritation, got on their nerves), and stressful thoughts (worry, thinking about a problem in the relationship) with each grown child, as well as negative daily mood (Birditt et al., 2010; Fingerman et al., 2009).

Results

A majority of parents had at least one child suffering problems (63%). Most of those parents (90%) had contact with a child suffering problems during the study week.

We examined the likelihood of each type of daily experience with each child, coding child’s problem (1 = has a problem, 0 = no problems). We used logistic multilevel models with three levels: (a) children nested within days and (b) days nested within parents. As expected, parents were more likely to report stressful encounters and stressful thoughts regarding children suffering problems (Supplementary Table 1). However, a child’s problems were not associated with having a positive encounter with that child.

Next, we examined daily negative mood as a function of positive and stressful experiences (e.g., pleasant encounters, stressful encounters, and stressful thoughts) with grown children. We used a 2-level multilevel model for daily mood (i.e., mood is a continuous variable), with days nested within parents. The model included a categorical variable for each type of experience (e.g., positive encounter, stressful thoughts) with children: (a) no such experience with grown child suffering a problem, (b) experience with one grown child suffering a problem (threshold model; reference group), or (c) experience with more than one grown child suffering a problem (cumulative model). We also included the same type of categorical variables (i.e., no child, one child, and more than one child) for each daily experience with children not suffering problems to the model.

As can be seen in Table 2, encounters with children suffering problems were associated with daily negative mood, controlling for prior day’s mood as follows: (a) a stressful encounter with one problem child was worse for negative mood than no stressful encounter (threshold), (b) stressful encounters with more than one problem child were worse than stressful encounters with one child (cumulative), (c) stressful thoughts about one problem child were worse than no stressful thoughts (threshold), and (d) positive encounters with multiple children suffering problems alleviated negative mood. Encounters with children not suffering problems were not significantly associated with negative mood. Because the model involved a categorical variable coding two or more children in a single category, we also tested the cumulative effects with number of problem children (and number of nonproblem children) treated as continuous variables for each type of experience (e.g., the number of problem children about whom parents had stressful thoughts). Findings revealed that negative mood was increased for each additional child, consistent with the model presented in Table 2.

Table 2.

Multilevel Models Predicting Daily Negative Mood From Daily Experiences With Adult Children

Predictors	B	SE
Fixed effects
Intercept	1.31***	0.19
Children suffering problems
Positive encountersa
No child	0.03	0.02
More than one child	−0.08*	0.04
Stressful encountersa
No child	−0.14***	0.04
More than one child	0.34***	0.09
Stressful thoughtsa
No child	−0.05*	0.02
More than one child	0.07	0.04
Children not suffering problems
Positive encountersa
No child	−0.00	0.02
More than one child	−0.03	0.03
Stressful encountersa
No child	−0.04	0.04
More than one child	0.11	0.16
Stressful thoughtsa
No child	−0.05	0.03
More than one child	−0.03	0.05
Controls
Prior day negative mood	0.40***	0.02
Parent: Age	−0.00	0.00
Parent: Maleb	−0.04	0.03
Parent: Years of education	0.00	0.01
Parent: Self-rated healthc	−0.03*	0.01
Parent: Minorityd	−0.00	0.03
Parent: Number of children	−0.01	0.01
Random effects
Intercept variance	0.02**	0.01
Residual variance	0.06***	0.00
−2 log-likelihood	487.9

Notes: Parent N = 247; Person-day N = 1,631 (maximum number of days per participant = 7).

^aReference category = one child.

^b0 = female, 1 = male.

^c1 = poor to 5 = excellent.

^d0 = non-Hispanic white, 1 = racial minority.

*p < .05; **p < .01; ***p < .001.

Further, we examined whether positive encounters with a nonproblem child alleviated effects of experiences with a problem child on negative mood (mitigating model). We reran the models with interaction terms for positive encounter × stressful encounter and positive encounter × stressful thought for problem and nonproblem children. Interaction terms were not significant (not shown here).

Finally, we asked whether any type of contact with a grown child suffering problems generated negative mood (regardless of whether that encounter was stressful). We used a categorical variable for having any type of contact with: (a) zero problem children, (b) one child suffering a problem, or (c) more than one child suffering a problem. Contact with a grown child or children suffering problems was not associated with parents’ daily mood (not shown here).

Discussion

Grown children who suffer problems affect parental long-term well-being. Here, stressful encounters and even thinking about children suffering problems increased parental negative daily mood. Indeed, long-term implications of grown children’s problems for parental well-being may stem in part from these daily transactions that involve demands and irritations.

From a within-family perspective, findings suggest that stressful daily experiences with even one such child (passing a threshold) are associated with poorer daily mood. Moreover, cumulative effects across the family were evident; stressful encounters with more problem children put a parent at greater risk. Fortunately, positive encounters with more children who suffer problems also benefited parental well-being. Unfortunately, these positive encounters did not undo the deleterious effects of stressful experiences on daily negative mood, as both types of experiences with problem children were associated with mood at the end of the day. Moreover, we had expected that positive encounters would occur with children who do not suffer problems, and it would be these children who mitigated the effects of their problematic siblings on the parents’ daily well-being. Surprisingly, parents’ positive encounters did not differ with regard to their children’s problems. Further, experiences with their children who do not suffer problems were not associated with the parents’ daily mood. Consistent with prior studies, the “least happy” children shape parents’ daily well-being (Fingerman et al., 2012).

The Longitudinal Study of Generations

The LSOG began in 1971 with the purpose of examining solidarity in intergenerational relationships. The original sample of three-generation families was identified by randomly selecting grandfathers from the membership of a large health maintenance organization in Southern California. Seven subsequent surveys were fielded up to 2005. Beginning in 1991, a fourth generation entered the study as they reached age 16. In total, the LSOG consists of 3,681 respondents from 418 three- and four-generation families. (For a full description of the derivation of the LSOG, see Bengtson & Roberts, 1991).

The multi-generation/multi-panel design of the LSOG provides analytic leverage for demonstrating how social change is manifest within families and identifying diverse patterns of continuity and change across generations (Min, Silverstein, & Gruenwald, 2012). Such a design has the added advantage of providing first-person assessments of opinions and beliefs, about which proxy reports would be considered unreliable.

Generations and Religious Orientations

The United States is typically considered a strongly religious nation. However, religiosity considerably weakened during the late 20th century (Hout & Fischer, 2002), a secularizing trend that culminated in the dramatic decline of religious identification among young adults in the 21st century (Chaves, 2011). In 2014, one-third of millennials had no religious affiliation, more than double the proportion of Baby-boomers (Pew Forum, 2015). On the other hand, evidence points to religious continuity between generations, with family lineages serving as conduits through which religious traditions are passed from parents to children, and grandparents to grandchildren (Copen & Silverstein, 2007; King & Elder, 1999; Min, Silverstein, & Lendon, 2012). Thus, there may be segments of the population with different intergenerational religious profiles—one characterized by decline and another by stability.

Although there is evidence that religious discordance between parents and children is related to poorer quality relations (Stokes & Regnerus, 2009), we know less about how patterns of religious change across multiple generations influence the relational well-being of grandparent-grandchild relationships. To address this question, we used a WFD approach to examine whether emotional closeness between grandchildren and grandmothers was weakened by religious decline across three generations.

Sample Design

We used a lagged within-family design—in which measures were taken from different family members at different periods of time—to examine change in religiosity within 502 grandmother-parent–grandchild triads. (For the sake of brevity we do not consider grandfathers in the current analysis.) The sample consisted of 191 G2 grandmothers in 1971, 262 G3 parents in 1988, and 502 G4 grandchildren in 2005 (see middle column of Table 1 for sample description). Advantages of this lagged design included achieving greater consistency in age across generations, avoiding the exclusion of grandmothers who died over the course of the study, and incorporating more than three decades of historical change into the model.

Measures

Religiosity was assessed in four domains: (a) Literalist beliefs measured by strength of agreement with the following statements: God exists in the form as described in the Bible; All people today are descendants of Adam and Eve. (b) Civic value of religion measured by strength of agreement with the following statements: All children should receive religious training; Religion should play an important role in daily life. (c) Religious participation measured by the frequency of attendance at religious services: How often do you attend religious services these days? (4) Religious intensity measured by the question: How religious are you? All items measured on an agree-to-disagree scale were dichotomized into agree versus disagree. Religious intensity was dichotomized into very/pretty religious versus somewhat/not religious. Religious attendance was divided into three categories: weekly or more, monthly or several times per year, and never.

Analytic Approach

To assess multigenerational patterns of religiosity we first employed Latent Class Analysis (LCA) to develop a multidimensional classification scheme for religiosity and then applied Latent Markov Modeling (LMM) to examine change in religious categories across generations (Latent GOLD 5.0; Vermunt & Magidson, 2013). The multilevel option was used to account for family clustering and full information maximum likelihood estimation was used to adjust for missing data. LMM is typically used to describe within-individual change, but in this application, it is applied to the sequencing of different individuals linked within the same families.

Results

In assessing the LCA model, we first determined the optimal number of latent classes of religiosity by estimating progressively more complex models until the BIC statistic failed to diminish, the log-likelihood fit was close to non-significance, and an interpretable set of classes was obtained. Based on these criteria, a four-class model was selected as optimal.

Conditional probabilities that defined the four religious classes are shown in Table 3 (the measurement model was invariant across the three generations). These classes were labelled: strongly religious (strong on all measures), weakly religious (weak on all measures), privately religious (strong on literalist beliefs and civic importance but weak on attendance), and liberally religious (strong on all measures except weak on literalist beliefs and moderate on attendance).

Table 3.

Latent Class Measurement Structure of Four Classes of Religiosity in Three Generations

	Strongly religious	Weakly religious	Privately religious	Liberally religious
Indicator	Conditional probabilities
Better off if religion played important role
Disagree	0.002	0.914	0.192	0.121
Agree	0.998	0.086	0.808	0.879
Every child should get religious instruction
Disagree	0.007	0.732	0.148	0.068
Agree	0.993	0.268	0.852	0.932
God exists as described in Bible
Disagree	0.013	0.888	0.066	0.634
Agree	0.987	0.112	0.934	0.366
All people are descendants of Adam & Eve
Disagree	0.040	0.928	0.132	0.936
Agree	0.960	0.072	0.868	0.064
How religious are you?
Somewhat or less	0.010	0.882	0.468	0.275
Pretty or very	0.990	0.118	0.532	0.725
Attend religious services
Never	0.011	0.577	0.425	0.136
Monthly or less	0.255	0.391	0.503	0.573
Weekly +	0.735	0.032	0.072	0.291
Generation	Latent class probabilities
Full sample	0.481	0.257	0.124	0.137
Grandmothers	0.629	0.094	0.042	0.235
Parents	0.482	0.245	0.124	0.149
Grandchildren	0.333	0.432	0.207	0.028

Respondents were classified into distinct classes based on their highest probability of class membership. The distribution of classes (see latent class probabilities in Table 3) reveals that religious class membership followed a predictable generational pattern with the oldest generation expressing the strongest religiosity, the youngest generation expressing the weakest religiosity, and the middle generation somewhere in between.

Turning to the analysis of multigenerational triads, the most common religious pattern identified by LMM consisted of triads in which all generations were strongly religious (23%) (see Supplementary Table 2). Triads in which at least one younger generation was less religious than a strongly religious grandmother accounted for 40% of all triads.

We next examined the relationship between patterns of intergenerational religious change and emotional closeness between grandchildren and their grandmothers. To assess emotional closeness, we used three assessments from the grandchild’s perspective (how close do you feel; how well do you get along; and how good is communication), each rated on a 6-point scale and then summed (alpha = 0.85).

The main predictors were groupings of commonly observed patterns of religious change across generations, collapsing categories as necessary to avoid sparseness. This effort resulted in 300 triads representing three religious patterns (corresponding to G2-G3-G4 sequences): strong-strong-strong (38.3%), strong-strong-weaker (30.7%), and strong-weaker-weaker (31.0%)—where strong refers to strongly religious and weaker refers to all religious types other than strongly religious. This specification allowed us to compare the two most common patterns of religious “movers” (unilinear weakening) to the most common pattern of religious “stayers” (consistently strong).

Multilevel modeling was used to predict emotional closeness between grandchildren and grandmothers from the religious patterns noted above (grandchildren were nested within 117 grandmothers). Control variables included age, education, and gender of grandchildren, gender of parents, and religious denomination of grandmothers (see Bengtson, 2013 for denomination definitions). The results (see Table 4) revealed that grandchildren in strong-weaker-weaker triads were significantly less close to their grandmothers than grandchildren in the strong-strong-strong reference group (b = −1.18, p < .05); however, grandchildren in strong-strong-weaker triads were no different than the reference group.

Table 4.

Multilevel Model Predicting Grandchild’s Emotional Closeness to Grandmother From Triadic Patterns of Religiosity (N = 316 Grandchildren Nested Within 117 Grandmothers)

Predictors	B	SE
Fixed effects
Intercept	10.91***	1.64
Triadic religious pattern
Strong-Strong-Weakera	−0.07	0.54
Strong-Weaker-Weakera	−1.18*	0.60
Controls
Grandchild: Age	−0.04	0.05
Grandchild: Education	0.23	0.18
Grandchild: Femaleb	−0.10	0.44
Parent: Femaleb	1.20**	0.51
Grandmother: Evangelical Protestantc	−0.22	0.83
Grandmother: Catholicc	−0.50	0.76
Grandmother: Mormonc	0.75	1.17
Random effects
Intercept variance	4.84**	1.57
Residual variance	8.88***	1.10
-2 log-likelihood	−700.39

^aReference group = Strong-Strong-Strong.

^bReference group = Male.

^cReference group = Mainline protestant and other.

*p < .05; **p < .01; ***p < .001.

Discussion

In this section, we highlighted an intra-familial design where data are linked between family members in different generations. Our analysis revealed that most families became more secular through the generations, though many retained their strong religiosity. Taken together, the results demonstrated that families exhibited diverse patterns of religious change through the generations, belying a simple narrative of universal religious decline. Our results suggest that religious change across generations that tends toward secularization is associated with weaker grandmother–grandchild relationships, but only when both the parent and grandchild have moved in that direction. This type of cumulative change could not have been detected without data collected from multiple family members. Although our analysis cannot definitively establish causal direction, it nevertheless stands as proof-of-concept of the utility of WFD designs for understanding the consequences of intergenerational discontinuity for the well-being of intergenerational family relationships.

The Within-Family Differences Study

The WFDS began in 2001with the goal of investigating the prevalence, predictors, and consequences of within-family differences in parent–adult child relationships in later-life families. In 2001–2003, the WFDS collected data from 566 mothers ages 65–75 and 774 of their adult children. Between 2008 and 2011, a second wave of data was collected from 420 of the original mothers and 835 of their offspring. (For more details regarding sampling and procedures, see http://web.ics.purdue.edu/~jsuitor/within-family-differences-study). Mothers were asked about their relationships with each of their offspring and asked to differentiate among their children on a wide range of relationship dimensions; children were asked about their relationship quality with their mothers and their perceptions of their mothers’ differentiation.

The WFDS revealed that ties between mothers and each of their adult children were shaped by their relationships with their other offspring, and documented consequences of such differentiation on both mothers’ and adult children’s well-being (Peng, Suitor, & Gilligan, in press; Pillemer, Suitor, Pardo, & Henderson, 2010; Suitor et al., 2013a, 2013b, 2015, in press). However, does studying these complex patterns of differential parent–child relationships in the family increase understanding beyond what can be learned using individual-dyad measures of mother-adult child relationship quality? The analyses we present demonstrate that, in fact, within-family designs can do just that.

Comparative Parent–Child Relationship Quality and Psychological Well-Being

Based on theories of social comparison (Festinger, 1954; Suls & Wheeler, 2000), we proposed that the association between mother–child relationship quality and adult children’s well-being would be stronger when taking into account children’s perceptions of their relationship quality with their mothers relative to the mother–child relationship quality of other siblings in the family. Specifically, we suggested that the effects of relationship quality on adult children’s depressive symptoms would vary depending upon where the child felt that she or he fit in the family “pecking order” (Conley, 2009). Thus, perceiving oneself as disfavored by one’s mother, relative to one’s siblings, would translate into higher depressive symptoms than would a child’s perception of the quality of his or her relationship with his or her mother without such comparisons. Recent evidence has shown that perceiving oneself as favored—specifically perceiving oneself as the most emotionally close to one’s mother—also predicts depressive symptoms (Suitor et al., in press). Thus, we hypothesized that perceiving oneself as either favored or disfavored, relative to one’s siblings, would be a stronger predictor of depressive symptoms than would be “individual dyadic” reports of mother–child relationship quality.

Sample Design

To address this questions, we used data on the 296 adult children nested within the 95 full families in the sample—specifically, those families in which all members of the sibship were interviewed at T2. We have chosen to use the subsample of 95 families in which all siblings participated rather than the full sample of 360 families which also includes families in which some but not all siblings were interviewed, to allow us to compare reports from all siblings in the family. See the far right column in Table 1 for the demographic characteristics of this subsample.

Measures

To measure depressive symptoms, we employed the 7-item version of the Center for Epidemiological Studies Depression (CES-D) Scale (Ross & Mirowsky, 1988). In this sample, the scale ranged from 7–28, with a mean of 11.6 (SD = 4.6) and an Alpha coefficient of 0.84.

To create the perceived parental favoritism and disfavoritism measures, each offspring was asked to select: (a) To which child in your family is your mother the most emotionally close? and (b) With which child in the family does your mother have the most disagreements or arguments? Children’s responses to each question were coded: 0 = child does not perceive mother as favoring/disfavoring any particular offspring; 1 = child perceives that mother favors/disfavors him or herself; or 2 = child perceives that mother favors/disfavors another child in the family. Because being perceived as the child who is favored or disfavored (as opposed to perceiving mothers’ favoritism or disfavoritism toward siblings) predicts depressive symptoms (Suitor et al., in press) whereas other perceptions do not, we created dummy variables for each dimension. “Chose self” was coded 1; “perceived mother did not differentiate” and “chose sibling” were combined and coded 0.

To measure individual-dyad closeness, each child was asked: “Use any number from 1 to 7, where 1 is very distant and 7 is very close. What number would you use to describe the relationship between you and (your child/your mother) nowadays?” To measure individual-dyad conflict, each child was asked: “Sometimes no matter how close we may be to someone, the relationship can also at times be tense and strained. Use any number from 1 to 7, where 1 is not at all tense and strained and 7 is very tense and strained. What number would you use to describe how tense and strained the relationship between you and (your child/your mother) is nowadays?” Consistent with the way these items are generally used (Gilligan, Suitor, Feld, & Pillemer, 2015), we combined the lowest categories of each item, so that the scores ranged from 1 to 4. To allow for direct comparisons between the within-family and individual-dyad measures, we further transformed both closeness and conflict into dichotomies in which 1 = highest and 0 = less than highest.

Analytic Approach

Because the 296 adult children were nested within 95 families, we used multilevel modeling. Listwise deletion was used to handle missing data on the independent variables because there were fewer than 2% missing on any variable in the analysis (cf., Allison, 2010).

Results

In Table 5 we present the findings for the within-family comparison versus the individual-dyad measures of mother–child relationship quality. We begin, in Model 1, by including only the individual-dyad measures of intergenerational closeness and conflict. In this model, neither closeness nor conflict predicted depressive symptoms. Next, in Model 2, we present the findings including only measures of children’s perceptions of being favored or disfavored by their mothers regarding emotional closeness and conflict. This analysis revealed that both perceived favoritism and perceived disfavoritism predicted depressive symptoms.

Table 5.

Multilevel Model Results Predicting Adult Children’s Depressive Symptoms (N = 296 Nested Within 95 Full Families)

	Model 1		Model 2		Model 3
Predictors	B	SE	B	SE	B	SE
Fixed effects
Intercept	21.85***	2.51	20.69***	2.42	20.37***	2.47
Family level characteristics
Race (1 = nonwhite)	−1.52	1.12	−1.49	1.08	−1.62	1.09
Family size	−0.24	0.15	−0.11	0.15	−0.11	0.15
Child level characteristics
Daughter	−0.83	0.44	−0.96*	0.42	−1.03*	0.43
Age	−0.06	0.04	−0.06	0.04	−0.06	0.04
Education	−0.05	0.17	0.05	0.16	0.07	0.16
Married	−1.21*	0.51	−1.23*	0.50	−1.13*	0.50
Employed	−2.00**	0.63	−1.78**	0.61	−1.74**	0.61
Health	−1.17***	0.21	−1.23***	0.21	−1.22***	0.21
Individual dyad relationship quality
Closeness to mother	0.41	0.47	—	—	0.21	0.46
Conflict with mother	1.17	0.65	—	—	0.83	0.66
Perceived maternal favoritism/Disfavoritism (Chose self = 1)
Most emotionally close to mother	—	—	1.62***	0.44	1.62***	0.44
Has greatest conflict with mother	—	—	1.51**	0.55	1.31*	0.58
Random effects
Intercept variance	1.17	0.83	0.94	0.76	1.05	0.78
Residual variance	11.96***	1.19	11.45***	1.14	11.37***	1.14
−2 log-likelihood	1,578.416		1,562.879		1,560.020

*p < .05; **p < .01; ***p < .001.

In Model 3, both the individual-dyad and within-family comparison measures of closeness and conflict were included in the analysis; the within-family measures of favoritism and disfavoritism continued to predict depressive symptoms, but neither of the individual-dyad measures of closeness or conflict did so. To ensure that these findings could not be accounted for by shared variance, we conduct collinearity diagnostics; none of the VIFs were greater than 1.4. Further, to be sure that the effects of individual-level closeness and conflict were not attenuated by the transformation into dichotomies, we conducted separate analyses that revealed the same patterns when using the 1–4 scale generally used for these measures (tables not shown).

It might seem surprising that perceiving oneself as most emotionally close to one’s mother would be associated with higher depressive symptoms. However, as we have discussed elsewhere, adult children who perceive themselves as most emotionally close to their mothers experience greater tension and less closeness with their siblings (Gilligan et al., 2013; Suitor et al., 2009), and may have a heightened sense of responsibility, either for the “emotional care” or the actual future instrumental care of their mothers, all of which are associated with lower psychological well-being (Suitor et al., in press).

Discussion

These findings reveal that within-family approaches uncover patterns that cannot be seen using single-dyad approaches. In particular, asking about children’s relationship quality with their mothers relative to that of their siblings revealed a stronger association with depressive symptoms than was found using questions in which offspring reported on their relationships without comparisons to their siblings. Thus, it appears that within-family measures in which respondents report on their perceptions of relationship quality relative to that of their siblings are more powerful predictors of psychological well-being than are reports of relationship quality that do not take such comparisons into consideration.

Discussion and Conclusions

Over the past five decades, research on intergenerational relations has become one of the most vibrant areas of gerontological research. Coupled with this scientific interest has been increasing awareness of demographic shifts that have changed the dynamics of intergenerational relationships in later life (Suitor et al., 2015). In particular, the remarkable increase in shared lifetime has created relationships between parents and adult offspring that endure three decades or more (Suitor et al., 2015). Both the scientific promise of research in this area, as well as societal concern regarding the importance of reliable exchanges of support between parents and their adult children, make the need for further advances in this area compelling. In this paper, we have provided evidence that taking a within-family approach is critical to understanding the patterns and consequences of intergenerational ties.

First, the FES showed that midlife parents have different experiences with grown children experiencing major life problems than with other grown children. Prior research has shown that in circumstances when grown children incur life problems, parents report diminished well-being (Fingerman et al., 2012) and attempt to mitigate the child’s problems via provision of support (Fingerman et al., 2009). The present analyses showed that children suffering problems may undermine midlife parents’ daily well-being by generating stressful encounters and negative mood. Moreover, whereas parents may have pleasant encounters with other offspring, these pleasant encounters do not mitigate effects of the problem child on daily mood. Thus, a within-family approach answers questions about which children have the greatest effects on midlife parents on a daily basis.

Next, in the case of the LSOG, taking measurements from multiple generations allows us to examine heterogeneity in cross-generational religious change and investigate how patterns of religious change have an impact on the well-being of intergenerational relationships. These questions cannot be adequately addressed without reports from all three generations.

Finally, the analyses we present using the WFDS demonstrate the greater light we can shed on the role of parent–child relations in adult children’s psychological well-being. Specifically, we show that by using measures of perceptions of mothers’ differentiation, as opposed to traditional single-dyad measures of closeness and conflict that do not take into consideration the relative strength of these relational dimensions within the family, we can better demonstrate the consequences of intergenerational relations on adult children’s psychological well-being.

Taken together, the findings presented here add to a growing body of evidence that profound within-family differences in intergenerational relations exist in the second half of life, and that these differences have real consequences for individuals and families. Nevertheless, the designs of most studies of intergenerational relations in later life do not permit an examination of within-family differentiation. In fact, almost all studies continue to ask parents about their adult children in the aggregate rather than about each child separately, or focus on only one parent-child dyad (Suitor et al., 2015).

This raises the question of why within-family approaches have not become more commonly applied. The most compelling explanation for the fact that relatively few studies have yet taken this approach is that conducting within-family studies is costly in terms of both money and time. Although it might appear initially that these costs emanate from collecting data on simply a larger number of respondents, the roots of the additional costs are more complex. For example, in each of the studies we have used to illustrate this approach, recruitment required that the respondent who was the “entry point” or “conduit” to the family agree to provide contact information for other members of the family. Such an approach means that in some cases, efforts to recruit additional family members fail, and data from that “entry individual” cannot help to address the central research questions of the project, and a replacement “entry point” respondent must be recruited. Further, there might be selection in favor of “entry individuals” with more generally harmonious family relations being more likely to provide contact information, as well as for individuals to provide contact information only for the specific family members with they have more harmonious relationships. Another issue is that the recruitment of multiple family members increases the length of the data collection period, such that data analysis and manuscript preparation are lengthier than that of single-respondent studies. Finally, analysis of within-family data requires that the investigators employ analytic strategies that may be unfamiliar, necessitating the time and often funding to learn these skills. Thus, these approaches are not without a variety of costs to the investigators.

One question that is raised by the costs we have just outlined is what scholars might do, short of implementing designs in which data are collected from all members of one or multiple generations. First, much can be learned from collecting information from one member about all of the family members of his or her generation or the adjacent generations. For example, some studies in which a parent was asked about each of his or her adult children have shed new light on both affective and instrumental exchanges between the generations that could not have been gleaned using reports on single dyads (Deane, Spitze, Ward, & Zhou, 2016; Fingerman et al., 2009, 2011; Suitor et al., 2006; Ward et al., 2009, 2014). Collecting such data might seem very time-consuming; however, once reporting on all offspring is established as a pattern in the interview, the incremental cost in time of reporting on each additional child is quite small. Second, although as we have demonstrated, there are considerable advantages to collecting data from members of multiple generations, we propose that there is also great benefit in collecting reports from members of even two generations, as opposed to a single generation. The strength of comparing reports from multiple generations has been evident since Bengtson and Kuypers’ (1971) classic work on the “generational stake,” and recent research continues to demonstrate the strong benefits of collecting data from even one member each in adjacent and/or non-adjacent generations (Bengtson & Silverstein, in press; Lin, 2008; Mandemakers & Dykstra, 2008).

In sum, within-family approaches enabled by multi-actor data provide the opportunity to greatly broaden the picture scholars can provide of family processes and the consequences of those processes for a variety of important outcomes, such as physical and psychological well-being, caregiving, and intergenerational cohesion. In this paper, we have highlighted the effects of within-family complexity using three separate studies addressing diverse research questions; these three sets of analyses offer support for the assertion that research must take into account the fact that intergenerational dyads within a single family constitute separate micro-environments that differ among themselves. Such research can take into account family intergenerational complexity and offer significant knowledge gains necessary to understand rapidly changing patterns of intergenerational relationships. However, it is important to take into consideration that these impressive gains must be weighed against the measurable costs of such innovations in the study of families in the middle and later years of the life course.

Supplementary Material

Supplementary data is available at The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences online.

Funding

J. J. Suitor and K. Pillemer acknowledge funding from the National Institute on Aging (RO1 AG18869-01; 2RO1 AG18869-04). J. J. Suitor also acknowledges support from the center on Aging and the Life Course at Purdue University. K. Fingerman and K. Kim acknowledge grants from the National Institute on Aging (R01AG027769), the MacArthur Network on an Aging Society (John W. Rowe, Network director), and the National Institute of Child Health and Human Development (R24 HD042849) awarded to the Population Research Center at U. of Texas-Austin. M. Silverstein and V. Bengtson acknowledge support from the National Institute on Aging (R01-AG07977 and R56-AG007977-26) and the John Templeton Foundation.