Transitions in Relationships With Older Parents: From Middle to Later Years

Abstract

Objective.

Although intergenerational relationships have been extensively examined, studies applying dynamic multidimensional treatments are rare. Employing the life course framework and the intergenerational solidarity and ambivalence paradigms, a typology of intergenerational relationships was derived and propositions about dynamics of intergenerational relationships were tested.

Method.

Using latent transition analysis, we modeled 4 waves of panel data spanning 18 years from the Longitudinal Study of Generations to examine how older parent–child relationships (N = 938) transitioned in and out of complex relational configurations.

Results.

We derived 5 relationship types roughly corresponding to those found in earlier research. Transitions in relationship type occurred mostly when both generations were relatively young, and along the lines of what attachment, ambivalence, and latent kinship theories would predict. When change did occur, it was primarily structured by factors affecting the availability of adult children, as well as circumstances that elevated the dependency of older parents and promoted both positive and negative reactivity in their adult children.

Discussion.

This study has demonstrated how typological analysis captures both the complexities and dynamics of intergenerational relationships in mature families. By including behavioral, emotional, and normative aspects of later life intergenerational relationships, we told a story that was more about continuity than change.

Due to the demographic development of increased life expectancy, the number of years children share with their parents has substantially increased (Bengtson, 2001). In the United States and other Western nations, this increase in longevity is paralleled by a retrenchment of family support policies due to declining governmental resources, placing more responsibilities on families (Putney & Bengtson, 2003). These trends have increased the importance of intergenerational relations in contemporary Western societies, warranting a scientific interest in the factors that predispose adult children to provide their parents assistance across the later years of their lives.

Despite an abundance of research on intergenerational relations in the past decades (for an overview, see the decade reviews by Allen, Blieszner, & Roberto, 2000; Mancini & Blieszner, 1989; Silverstein & Giarrusso, 2010; Streib & Beck, 1980), surprisingly little is known how these relations change, or evolve, as the generations age. It can be expected, however, that intergenerational relations are, in part, shaped by or dependent on their histories (e.g., Belsky, Jaffee, Hsieh, & Silva, 2001; Parrot & Bengtson, 1999; Whitbeck, Hoyt, & Huck, 1994; Whitbeck, Simons, & Conger, 1991), such as previous exchanges of support and life events typically associated with aging, such as widowhood or a decline in physical health. Applying a longitudinal, dynamic approach can improve our understanding of the evolving nature of adult intergenerational relations in ways that cross-sectional studies cannot.

In addition, we propose that, in order to capture the complexity of intergenerational relations, researchers should simultaneously consider multiple characteristics of family relationships, rather than individually focus on relational characteristics, such as contact, help exchange, or affection. Examining the most prevalent combinations of characteristics more faithfully captures the complexity and variation in the bundle of attributes that make up family relationships. In recent years, a number of studies have taken such a multidimensional approach (e.g., Ferring, Michels, Boll, & Filipp, 2009; Silverstein, Bengtson, & Lawton, 1997; Silverstein, Gans, Lowenstein, Giarrusso, & Bengtson, 2010; Van Gaalen & Dykstra, 2006), often relying on dimensions differentiated by the intergenerational solidarity paradigm (Bengtson & Roberts, 1991; Roberts, Richards, & Bengtson, 1991). Although families often show strong correlational stability in single relational attributes over time, the degree to which they exhibit configural stability across multiple attributes is little known.

In this investigation, we continue the line of research on intergenerational typologies by employing longitudinal panel data that span a period of 18 years to describe patterns of continuity and change in complex relational configurations and examine the correlates of the most commonly occurring transitions. To allow a comparison of our results with previous studies, our typology of intergenerational family relations is informed by the interdependence of dimensions from the intergenerational solidarity and conflict paradigm (Bengtson, Giarrusso, Mabry, & Silverstein, 2002). We assess the (construct) validity of our typology by examining the associations of relationship types (at the first moment of observation) with theoretically relevant covariates. For the longitudinal part of our investigation, we draw on the life course framework (Elder, 1985, 1994) as an orientating theoretical perspective under which more specific explanatory theories and mechanisms on stability and change can be applied. Whereas our main focus is on the impact of changes and events associated with the aging of the older generation, potential influences from events in the lives of offspring are considered as well.

Characterizing Intergenerational Relations

The development of typologies has been a strategy to capture the complexity of family relationships based on their behavioral and cognitive-emotional attributes (see Ferring et al., 2009; Giarrusso, Silverstein, Gans, & Bengtson, 2005; Hogan, Eggebeen, & Clogg, 1993; Silverstein et al., 1997, 2010; Van Gaalen & Dykstra, 2006; Van Gaalen, Dykstra, & Komter, 2010). Although a priori decisions about which relational dimensions to consider have a bearing on the nature of the descriptive typology, the most inclusive studies (i.e., Silverstein et al., 1997; Van Gaalen & Dykstra, 2006) found almost identical typologies, distinguishing four similar relationship types (harmonious or tight-knit; affective or intimate-but-distant; discordant or detached; and obligatory). An additional fifth relationship type differed between the studies (i.e., sociable vs ambivalent) and can be attributed to the inclusion of conflict in the analysis of the more recent study by Van Gaalen and Dykstra.

Although discovering similar relationship types in different data sets points to the validity of these typologies, their associations with covariates on factors that structure family life provides additional evidence of their veracity. Because the focus of the present study is on continuity and change in intergenerational relationships and not on factors influencing intergenerational relationships at one point in time, as well as for the sake of brevity, we omit a thorough discussion on sociodemographic predictors. Nonetheless, we note that research on family relations has found that demographic characteristics of both generations and structural properties of the relationship (e.g., geographical proximity) affected relationship types. In order to establish construct validation of our typology, we will examine the associations between the derived types and theoretically relevant covariates, including—among others—the age, gender, marital and employment status of both generations, family size, and geographic proximity.

A Story of Continuity?

Although previous results point the way to a general model of intergenerational relations, many questions are left unanswered in these cross-sectional studies, most notably whether the derived types are trait-like (stable properties of relationships), state-like (adaptive to evolving family or personal circumstances), or a combination of both. Applying the life course perspective (e.g., Elder, 1985, 1994) to intergenerational relations, two principles are identified to be particularly relevant to this issue and used as guiding tools in our theoretical framework: the principle of life-span development and the principle of linked lives (Bengtson, Elder, & Putney, 2005; Elder, Johnson, & Crosnoe, 2003; Putney & Bengtson, 2005).

Central to the principle of life-span development is the notion that human development and aging are lifelong processes: Events and experiences that occurred early in life may have long-term consequences on the choices and behavior in later life, a phenomenon also known as path dependency. In line with this principle, attachment theory (Bowlby, 1969) postulates that, due to the internal working models formed during childhood, relationships are relatively stable “from the cradle to the grave” (Bowlby, 1977). Research has shown attachment styles to be characterized by a fair degree of stability, from infancy to adolescence (Hamilton, 2000) and young adulthood (Lewis, Feiring, & Rosenthal, 2000; Waters, Merrick, Treboux, Crowell, & Albersheim, 2000), as well as from early to middle adulthood (Antonucci, Akiyama, & Takahashi, 2004; Scharfe & Bartholomew, 1994; Zhang & Labouvie-Vief, 2004). Additionally, the continuity theory of normal aging (Atchley, 1989) posits that middle-aged and older adults attempt to preserve and maintain existing social behavior, habits, and lifestyles as much as possible as they grow older. Nonetheless, instability in attachment styles has also been found (Weinfeld, Sroufe, & Egeland, 2000; Zhang & Labouvie-Vief, 2004), where it may be noted that the lack of stability was mainly found for respondents with insecure attachment styles and unstable family environments with emotionally distant relationships (Bowlby, 1980; Vaughn, Egeland, Sroufe, & Waters, 1979). In line with attachment theory and continuity theory, we hypothesize that configurations of intergenerational relations between adult children and parents will show more stability and continuity over time than they will show change (H1). Yet, among the variety of intergenerational relationships, those that lack emotional closeness are expected to be less stable over time than those that are emotionally close (H2).

Or a Story of Change?

In contrast to the previous arguments advocating continuity, the principle of linked lives draws our attention to the influence of changes or events in the lives of individual family members on both the lives of, and relationships with, other family members. Indeed, there is a fair amount of evidence showing that intergenerational relationships change with the aging of the older generation, insofar as children are reactive to the age-related deficits of their parents (e.g., Field & Minkler, 1988; Guiax, Van Tilburg, & Broese van Groenou, 2007; Ha, Carr, Utz, & Nesse, 2006). Indeed, adult children often form a safety net for their parents and emerge as caregivers when the need arises (Riley & Riley Jr., 1993; Silverstein, Gans, & Yang, 2006). With older age comes a loss of social roles (e.g., retirement, widowhood) and deterioration in health. Such changes increase the practical and emotional salience of parent–child relationships, as older adults tend to increasingly focus on relationships that provide the most emotional satisfaction (Carstensen, 1992).

In addition, events in the lives of the adult children, such as marriage, divorce, or the birth of a (grand)child, have been found to induce change in intergenerational relationships as well (Bucx, Van Wel, Knijn, & Hagendoorn, 2008; Kaptijn, Thomese, Van Tilburg, & Liefbroer, 2010; Kaufman & Uhlenberg, 1998; Spitze, Logan, Deane, & Zerger, 1994; Suitor, Pillemer, Keeton, & Robison, 1995), although children’s life course characteristics have been found to explain less variance in instrumental support exchanges than parent’s life course characteristics (Klein Ikkink, Van Tilburg, & Knipscheer, 1999). Nonetheless, the impact of events in the lives of both generations will be considered in the current study as well.

On the basis of these considerations and findings, we hypothesize that, as the older generation ages, the proportion of intergenerational relations characterized by high degrees of involvement will increase, whereas the proportion of intergenerational relations characterized by low involvement will decrease (H3). Moreover, the occurrence of events in the lives of either generation is expected to advance transitions between relationship types, over and beyond the general trend toward more involved relationship types (H4). For the latter hypothesis, we will examine family events (i.e., transitions in marital and parental statuses), labor force participation (i.e., transitions into and out of employment, including retirement), and changes in opportunity structures that predispose relational engagement (i.e., changes in geographical proximity and physical health). The only events we omit (for an overview of life events previously studied, see Wrzus, Hänel, Wagner, & Neyer, 2013) are the ones associated with the educational career, as these careers have typically ended by the time we start observing both generations in our data.

Change and Intergenerational Ambivalence

We note, however, that negative life events, such as divorce, widowhood, illness, and the need for care, may disrupt intergenerational relations even when (or because) they bring generations physically and behaviorally closer together (Aquilino, 1994; Greenfield & Marks, 2006). The growing dependency of parents on adult children may breach a previously existing balance of exchange between generations and induce tension and conflict in the relationship leading to mixed feelings (Ha & Ingersoll-Dayton, 2008; Silverstein et al., 2010), a characteristic also known as ambivalence (Luescher & Pillemer, 1998; Pillemer & Suitor, 2002; Willson, Shuey, Elder, & Wickrama, 2006). Although, there is some variation in how ambivalence is conceived and measured (Lendon, Silverstein, & Giarrusso, 2014; Suitor, Gilligan, & Pillemer, 2011), the indirect approach by which ambivalence is defined as the simultaneous presence of mixed feelings or contradictory behaviors seems a popular strategy that lends itself very well to typological analysis.

Although typologies represent an efficient way to capture antagonistic aspects of intergenerational relationships, their exploratory nature leaves open whether some form of psychological ambivalence (i.e., the presence of contradictory feelings toward the same individual) or behavioral ambivalence (i.e., the presence of contradictory attributes or behaviors, see also Van Gaalen & Dykstra, 2006) will be identified in a typology of intergenerational relations. In either case, we hypothesize intergenerational relations to transition more often into an ambivalent type when a life event raises the need for help in either generation (H5).

Based on the assumption that the experience of ambivalence induces stress, reduces well-being, and, as a set of contradictory elements, is inherently unsustainable in the long term (Connidis & McMullin, 2002), we expect family members to attempt to negotiate, manage, and diminish the ambivalence they experience in relationships. As previous research does not provide us with any clues about the nature of the attempts of family members to resolve ambivalence, we formulate an undifferentiated hypothesis that the ambivalent relationship type is the most likely to transition into another relationship type (H6).

Method

Sample

The present study used data from the Longitudinal Study of Generations (LSOG), a study of more than 300 four-generation families who initially lived in southern California. The LSOG began in 1971 as a cross-sectional study of 2,044 individuals representing 349 separate three-generation families. Grandfathers were selected via a multistage stratified random sampling procedure from a population of 840,000 individuals enrolled in southern California’s first large health maintenance organization. Spouses, adult children, and adult grandchildren of these grandfathers were also invited to participate. Wave 1 surveys were sent to 3,160 respondents who were eligible for inclusion with a response rate of 65%. The LSOG sample is similar to the U.S. population in terms of age, marital status, and social class distributions, but men, ethnic minorities, and lower educated are underrepresented. The study became a longitudinal investigation in 1985 with follow-ups every 3 years up to 2000, followed by a 6-year follow-up in 2006. The LSOG has had high longitudinal participation rates since 1971, considering the age of the sample, the duration of the study, and the use of self-administrated mail surveys. Retention rates have averaged 74% (mortality adjusted). Using various diagnostic techniques, such as comparing the scores of our final (longitudinal) sample with the scores of attrited (baseline) respondents on key demographic variables, as well as the use of Heckman modeling to correct for potential attrition bias (Miller & Wright, 1995), little evidence was found for serious selection bias as a result of systematic sample attrition (Mabry, Putney, & Wang, 2002).

For this study, we focus on reports by G3 children, who averaged 36 years of age at the first time point in this study, about their relationship with one biological G2 parent. Measurements are taken at 6-year intervals in 1988, 1994, 2000, and 2006, hereafter referred to as T1 through T4. Six-year intervals were chosen to increase the opportunity for changes to occur in relationship characteristics. To be eligible for inclusion, G3 respondents had to have at least one living parent at T1 (N = 1,010). If both parents were living, one parent was randomly selected. If this randomly selected parent died during the study, the other parent was chosen instead if he or she remained alive until T4. Also, if the G3 respondent indicated that the selected parent was not his or her biological parent, then the biological parent was chosen instead. These criteria excluded those with no biological parent alive (n = 15), and those with no wave of complete data (n = 61), resulting in a final sample of 938 intergenerational relations, nested within 295 families. The analytical sample consists primarily of higher educated (43.2% college graduate), non-Hispanic white adult children (93.1% Caucasian) with higher than average household income levels (average $49,513 at T1, $98,486 at T4). Thus, like the complete LSOG sample, our analytical sample underrepresents ethnic minorities, the lower educated, and those with lower incomes. Moreover, compared with the U.S. population, our sample overrepresents the nonreligious (37.4%), and the Jewish population (8.1%), whereas both Protestants and Catholics are underrepresented.

Measures

Intergenerational relationship characteristics.

Key measures for constructing the typology of intergenerational relations were based on the intergenerational solidarity and conflict paradigm that includes emotional-cognitive (i.e., affection, conflict) and behavioral (i.e., association and support exchange) aspects of intergenerational relationships. All measures were constructed using information provided by adult children who were asked to answer questions about their relationship with their parents.

Associational solidarity was measured using the following question: “How often during the past year were you in contact with your parent?” In the questionnaire, a distinction was made between face-to-face contact, contact by phone, and contact by mail. Face-to-face contact was transformed to a range from 0 to 365 days per year. Contact by phone and contact by mail were similarly transformed and, assuming they do not occur simultaneously, combined to form a single variable measuring all contact that was not face-to-face, resulting in a range from 0 to 730 days.

Three items were used to measure affectual solidarity: “How close do you feel is the relationship between you and your parent at this point in your life?” “Overall, how well do you and your parent get along together at this point in your life?” and “How is communication between you and your parent?” Each question was answered on a 6-point Likert-type scale, with answers ranging from not at all (close/well/good) to extremely (close/well/good). Cronbach’s alpha for the three items at the four time points ranged between .88 and .90. The three items were combined into a single score by taking their average value.

Conflict was assessed using responses to the questions: “How much conflict or tension do you feel there is between you and your parent?” “How much do you feel this parent is critical of you?” and “How much does your parent argue with you?” Again, each question was answered using a 6-point Likert scale, with answers ranging from none at all to a great deal. Cronbach’s alpha for the three items at the four time points ranged between .73 and .78. The three items were combined by taking their average value.

For exchanges of support, we distinguished between instrumental, financial, and emotional support, as well as between giving versus receiving support. Instrumental support was measured using the following questions: “[Did your parent turn to you/Did you turn to your parent] for (a) help with household chores; (b) transportation or shopping; (c) help in case of illness?” Financial support was measured using the questions: “[Did your parent turn to you/Did you turn to your parent] for financial assistance?” Emotional support was measured using the following questions: “[Did your parent turn to you/Did you turn to your parent for] (a) information and advice; (b) emotional support?” For the majority of the questions, the answer categories were coded as 1 = yes or 0 = no. At the two last time points, respondents were asked how often they provided each particular kind of help in the past year, ranging from 0 = not at all to 7 = every day. In order to make the measures comparable over time, we dichotomized the continuous response option, setting the cutoff point at once a month or more. Diagnostics and robustness checks revealed this to be an adequate cutoff based on the observed trend of changes in the distribution over time. For each of the four waves of measurement, additive scales were computed when multiple items were available, which were providing instrumental support, receiving instrumental support, providing emotional support, and receiving emotional support. Descriptive information on these scales, the items measuring financial support, as well as the items measuring affection, association, and conflict can be found in Table 1 (in the columns labeled “Original”).

Table 1.

Descriptive Statistics for Items on Relationship Characteristics Used in Typology

	Time 1 (n = 697)			Time 2 (n = 633)			Time 3 (n = 605)			Time 4 (n = 539)
	Original		Altered	Original		Altered	Original		Altered	Original		Altered
	Mean (SD)	Rangea	M a	Mean (SD)	Rangea	M a	Mean (SD)	Rangea	M a	Mean (SD)	Rangea	M a
Face-to-face contact	37.892 (77.058)	0–365	0.550	31.985 (62.524)	0–365	0.443	35.334 (80.874)	0–365	0.450	32.679 (77.031)	0–365	0.424
Other forms of contact	72.886 (103.832)	0–377	0.876	66.702 (81.450)	0–730	0.894	74.975 (105.117)	0–417	0.879	72.566 (105.842)	0–730	0.864
Affectional solidarity	3.222 (1.192)	0–5	0.349	3.207 (1.180)	0–5	0.377	3.221 (1.205)	0–5	0.374	3.202 (1.196)	0–5	0.347
Conflict	1.110 (0.953)	0–5	0.192	0.877 (0.865)	0–4.667	0.127	0.870 (0.942)	0–5	0.140	0.847 (0.897)	0–5	0.132
Instrumental support received	0.327 (0.579)	0–3	0.274	0.330 (0.591)	0–3	0.272	0.206 (0.526)	0–3	0.158	0.163 (0.502)	0–3	0.119
Instrumental support provided	0.341 (0.644)	0–3	0.263	0.458 (0.759)	0–3	0.329	0.614 (0.982)	0–3	0.342	0.816 (1.102)	0–3	0.417
Emotional support received	1.455 (1.271)	0–3	0.725	1.495 (1.183)	0–3	0.794	1.193 (1.296)	0–3	0.839	1.024 (1.260)	0–3	0.856
Emotional support provided	1.059 (1.201)	0–3		1.472 (1.210)	0–3		1.882 (1.210)	0–3		2.042 (1.177)	0–3
Financial support received	0.227	0–1	0.227	0.200	0–1	0.200	0.165	0–1	0.165	0.130	0–1	0.130
Financial support provided	0.033	0–1		0.061	0–1		0.075	0–1		0.096	0–1

Notes. SD = standard deviation.

^aFor original variables, the observed range is reported; the theoretical and observed range of all altered variables is 0–1.

We note that geographic distance from children and norms of filial obligation were omitted from the typology but they will be included as predictors of relationship type. As theorized in the model of intergenerational solidarity (Bengtson & Roberts, 1991; Hogerbrugge & Komter, 2012), both normative and structural solidarity (including geographical distance) predispose relational engagement but are considered exogenous to the relationships themselves.

Predictor variables.

For each time point, the following predictor variables were constructed. Gender of the parent and child were coded as 1 = female and 0 = male. As the age of the parents and children are highly correlated, only the age of children, measured in years, was included in the analyses. Number of siblings of adult children is measured in absolute numbers. A dummy variable indicated whether adult children had any children of their own (i.e., grandchildren of the parent) who were living at home. Marital status of adult children was measured with four dummy variables indicating whether they were cohabiting, unmarried, widowed, or separated/divorced (married = reference). Parent’s marital status was measured with two dummy variables, indicating whether they were divorced or widowed (married = reference). The employment status of both parents and children was measured using four dummy variables indicating whether they were employed, involuntarily unemployed, retired, or a housekeeper (employed = reference). Information on the employment status of the parent was retrieved from the questionnaire completed by the parent. Parent’s health status was assessed with a 12-item checklist completed by adult children, indicating whether their parents had any chronic diseases or impairments. The checklist included items such as heart problems, arthritis or rheumatism, visual impairments, hearing impairment, and diabetes. The items were reverse coded and summed to form a composite scale measuring health with higher scores indicating better health. Geographic distance between the adult children and their parents was measured using a scale with six categories ranging between 0 = less than five miles and 5 = more than 500 miles. Finally, the degree to which adult children endorsed norms of family obligation was assessed by six items. Examples of items are “How much responsibility have adult children to help with household chores and repairs or to provide transportation for the elderly parents?” and “How much responsibility should adult children have to provide for the personal and health care needs of elderly parents?” (0 = none, 4 = major). These items were asked only at the second and fifth waves of the LSOG study. Composite scores were computed by taking the average of the items at each wave (α = .86 and .89 for the second and fifth waves, respectively).

To examine the effect of life events and changing circumstances, the following dichotomous measures were constructed using information from two adjacent time points; whether any (grand)children were born; whether adult children married, started cohabiting, divorced, or became widowed (reference = no change); whether parents divorced or became widowed (reference = no change); whether the parent or child transitioned into or out of paid labor, or whether the parent retired (reference = no change). Improvement and decline in parent’s health, and increases and decreases in geographic distance from parents were measured by taking the absolute differences between two adjacent time points. If there was an increase or decrease, the variable measuring the opposite direction was coded 0. Descriptive information about the covariates can be found in Table 2.

Table 2.

Descriptive Statistics for Covariates Used for Predicting Relationship Type (Supplementary Appendix A) and Transitions Between Relationship Types (Table 5)

	Time 1 (n = 697)		Time 1–2 (n = 522)		Time 2 (n = 633)		Time 2–3 (n = 468)		Time 3 (n = 605)		Time 3–4 (n = 462)
	Mean (SD)	Missing (%)	Mean (SD)	Missing (%)	Mean (SD)	Missing (%)	Mean (SD)	Missing (%)	Mean (SD)	Missing (%)	Mean (SD)	Missing (%)
Demographic characteristics
Parent’s gender (1 = female)	0.630	0.0			0.664	0.0			0.638	0.0
Child’s gender (1 = female)	0.575	0.0			0.586	0.0			0.579	0.0
Child’s age	36.102 (3.654)	0.29			42.009 (3.723)	0.0			47.731 (3.671)	0.0
Parental status
Child’s # of siblings	2.698 (1.741)	0.14			2.793 (1.840)	0.0			2.743 (1.741)	0.17
Child has (grand)child (1 = yes)	0.715	0.43	0.132	0.0	0.694	0.47	0.043	0.0	0.592	0.0	0.013	0.0
Marital status
Child married (1 = yes)	0.786	0.0	0.071	0.0	0.787	0.0	0.045	0.0	0.749	0.0	0.037	0.0
Child unmarried cohabiting (1 = yes)	0.052	0.0	0.025	0.0	0.051	0.0	0.017	0.0	0.058	0.0	0.013	0.0
Child single (1 = yes)	0.065	0.0			0.055	0.0			0.058	0.0
Child divorced (1 = yes)	0.092	0.0	0.046	0.0	0.093	0.0	0.060	0.0	0.122	0.0	0.015	0.0
Child widowed (1 = yes)	0.006	0.0	0.010	0.0	0.014	0.0	0.006	0.0	0.013	0.0	0.004	0.0
Parents married (1 = yes)	0.634	0.43			0.554	3.00			0.463	7.11
Parents divorced (1 = yes)	0.225	0.43	0.017	0.0	0.225	3.00	0.013	1.92	0.238	7.11	N/A	N/A
Parent widowed (1 = yes)	0.141	0.43	0.100	0.0	0.221	3.00	0.092	1.92	0.299	7.11	0.112	5.19
Employment status
Child employed (1 = yes)	0.848	1.15	0.082	1.92	0.872	1.58	0.075	2.99	0.863	1.16	0.048	13.85
Child unemployed (1 = yes)	0.029	1.15	0.025	1.92	0.029	1.58	0.024	2.99	0.043	1.16	0.023	13.85
Child retired (1 = yes)	0.003	1.15	0.004	1.92	0.006	1.58	0.079	2.99	0.093	1.16	0.023	13.85
Child housekeeper (1 = yes)	0.120	1.15	0.039	1.92	0.093	1.58	0.000	2.99	0.000	1.16	0.065	13.85
Parent employed (1 = yes)	0.513	46.63	0.059	47.89	0.298	45.34	0.055	49.79	0.229	50.25	0.035	51.08
Parent unemployed (1 = yes)	0.035	46.63	0.026	47.89	0.038	45.34	0.013	49.79	0.033	50.25	0.000	51.08
Parent retired (1 = yes)	0.129	46.63	0.221	47.89	0.532	45.34	0.119	49.79	0.608	50.25	0.071	51.08
Parent housekeeper (1 = yes)	0.129	46.63	0.063	47.89	0.133	45.34	0.064	49.79	0.130	50.25	0.084	51.08
Structural and normative factors
Parent’s health	10.749 (1.321)	0.0			10.597 (1.316)	1.11			10.210 (1.565)	4.96
Decline			0.514 (0.853)	1.15			0.714 (0.993)	5.13			0.752 (1.075)	4.11
Improvement			0.337 (0.731)	1.15			0.266 (0.591)	5.13			0.293 (0.672)	4.11
Geographic distance	3.009 (1.478)	5.74			3.228 (1.490)	1.58			3.277 (1.480)	2.15
Decrease			0.275 (0.781)	7.47			0.167 (0.625)	4.27			0.135 (0.502)	3.46
Increase			0.391 (0.910)	7.47			0.246 (0.731)	4.27			0.126 (0.555)	3.46
Child’s endorsement of family normsa	2.619 (0.562)	39.46			2.582 (0.621)	4.11			N/A	N/A

Notes. SD = standard deviation.

^aFor Time 1 (= Wave 3), information from Wave 2 is used, as these items were not included in the questionnaire of Wave 3.

Statistical Procedure

To construct our typology, we applied latent transition analysis (LTA), a dynamic extension of latent class analysis (LCA). Rooted in latent class theory (Goodman, 1974; Lazarsfeld & Henry, 1968), LCA is a person-centered approach that groups units of analysis—intergenerational relationships in the present research—based on similarities in the patterned interdependencies of the characteristics that describe them (Bergman & Magnusson, 1997). LCA posits that an underlying (latent) grouping of respondents can be inferred from a set of manifest indicators. LTA extends LCA by adding a dynamic element that assesses the probabilities that latent class memberships (or, more appropriately “latent statuses”) transition over time (Velicer, Martin, & Collins, 1996). This technique makes the Markovian assumption that the probability of being in a particular class depends only on membership in the immediately previous class and not on any classes prior to it (Bye & Schechter, 1986; Graham, Collins, Wugalter, Chung, & Hansen, 1991).

We used LTA to identify the most common patterns (or combinations) of relationship characteristics across the time periods that produce the fewest and most homogenous groupings. Assignment of respondents to a particular latent class is typically made based on the class of which each respondent has the highest probability of being a member. Once a satisfactory typology was identified, a descriptive cross-classification table described how membership in the types changed or remained stable over time.

A primary concern in LCA and LTA is how many classes the investigator should specify. We used relative fit statistics to assess the adequacy of competing models, as well as principles of parsimony and class interpretability. Absolute fit statistics are often less useful for models involving large degrees of freedom because the reference distribution for the likelihood ratio statistic is not known (Collins, Fidler, Wugalter, & Long, 1993). We relied on two relative fit statistics: the Bayesian information criterion (BIC), which captures overall model fit and includes a penalty for model complexity; and a measure of entropy that assesses the quality or precision of the assignment of individuals to latent classes.

Model estimation was carried out using Mplus 5.21 (L. K. Muthén & Muthén, 2008) employing full information maximum likelihood estimation. This procedure allows for missing data under the usual missing at random assumption. Thus, all respondents were included, even if they participated at only one time point. Although the data for respondents with fewer than four observations did not contribute to the estimation of all parameters, they contributed to the model wherever they had coverage, thereby improving overall estimates (B. O. Muthén, 2007). Standard errors were adjusted using a sandwich estimator to account for the clustering of respondents within families.

Results

To ease the substantial computational burden of estimating latent transition models, and to allow a comparison of our results with previous developed typologies, we followed the lead of Van Gaalen and Dykstra (2006) and combined measures of providing and receiving emotional support, whereas providing financial support to the parent was dropped because of its low prevalence. Moreover, all variables not directly measured as dichotomies were dichotomized based on theoretical and empirical criteria, which included the comparability of our measures with previous studies, and the meaningfulness of the resulting distributions. For both face-to-face and other forms of contact, the cutoff point was set at “at least monthly.” For the composite item measuring affectual solidarity, average scores indicating extremely or very close designated greater affection (vs pretty, somewhat, not too, or not at all). For the composite item measuring conflict, average scores corresponding to a great deal, quite a bit, pretty much, and some designated greater conflict (vs a little or none at all). For the additive support measures, scores greater than 1 were set to 1. Proportions of the dichotomized variables are reported for each time point in Table 1 (in the columns labeled “Altered”).

Models with different numbers of latent classes were compared to identify the number of classes that provided the optimal balance of fit and parsimony. Measurement invariance across the four time points was assumed in order to achieve a more stable solution and maximize interpretability. As the entropy measure did not vary greatly between models, we primarily relied on the BIC statistic when deciding the number of latent classes to accept. Based on BIC values (available upon request), a five-class model provided the best fit as well as the most conceptually appealing solution.

Three sets of parameters were estimated in the LTA: conditional item response probabilities, latent status membership probabilities, and transition probabilities. The item response probabilities are analogous to factor loadings and represent the measurement part of the model. They are informative about measurement precision and are used to interpret the derived classes. Latent class membership and transition probabilities represent the structural part of the model and provide, respectively, the estimated proportion of participants in each latent class at the first occasion of measurement, and the probability of moving to a subsequent latent class conditional on latent class membership at the previous time point.

Table 3 shows the result of the five-class model. To facilitate interpretation, item response probabilities for each class, we compared them with the average probabilities for the total sample (reported in Table 1). Item response probabilities that were above the average at all time points are reported in bold. Overall, our five-class typology fits with much of what the literature has shown when categorizing intergenerational relationships within a multidimensional framework (Silverstein et al., 1997; Van Gaalen & Dykstra, 2006). That is, our typology includes relationship types that can be labeled harmonious, affective, obligatory, ambivalent, and discordant. Note that, although we included measures for both affection and conflict, a psychologically ambivalent relationship type in which these relationship traits coexist was not identified in our data. Instead, a type with high probabilities for all support and contact items, and relatively high probability scores for conflict was identified, resembling the behaviorally ambivalent relationship type found by Van Gaalen and Dykstra (2006).

Table 3.

Latent Transition Analysis of Parent-Adult Child Relationships (Probabilities) (N = 938 Nested in 295 Families)

	Latent statusa
	Harmonious	Affective	Obligatory	Behaviorally ambivalent	Discordant
At least monthly face to face contact	.843	.016	.950	.895	.018
At least monthly other forms of contact	1.000	.981	.923	.961	.433
Feeling close to parent	.852	.439	.220	.128	.013
Having a conflict with parent	.000	.063	.168	.321	.307
Instrumental support received	.535	.079	.183	.328	.017
Instrumental support provided	.640	.156	.094	.888	.042
Emotional support exchanged	.989	.917	.629	.971	.358
Financial support received	.314	.157	.185	.212	.060
Distribution latent statuses Time 1	20.4%	23.5%	26.7%	11.4%	18.0%
Distribution latent statuses Time 2	21.7%	31.6%	14.4%	14.5%	17.8%
Distribution latent statuses Time 3	19.2%	34.4%	13.6%	17.5%	15.3%
Distribution latent statuses Time 4	17.0%	32.2%	12.2%	21.2%	17.5%

Note. ^aItem response probabilities above sample average of all time points are reported in bold.

Additional evidence for the validity of our typology was derived from a multinomial logistic regression analysis in which the personal and relational correlates of class membership at the initial period of measurement was examined. In general, the type of relationship that adult children have with their parents is dependent upon geographic distance between the generations, health status of parents, gender and marital status of both parents and adult children, number of siblings, and norms of filial responsibility endorsed by adult children (see Supplementary Appendix A for details; to ease interpretation, estimated marginal effects are presented). Moreover, these results are in line with previous findings (e.g., Bengtson, 2001; Bengtson, Biblarz, & Roberts, 2002; Rossi & Rossi, 1990; Silverstein et al., 1997; Silverstein, Parrot, & Bengtson, 1995; Van Gaalen & Dykstra, 2006) and provide further evidence that the typology we arrived at is reasonably valid.

Probability distributions of the latent classes are presented for each time point in the lower part of Table 3. At T1, 20% of the intergenerational relationships were harmonious, 24% were affective, 27% were obligatory, 11% were behaviorally ambivalent, and 18% were discordant. Examining aggregate change in the latent status probability distributions, we see a decrease in the obligatory relationship type, declining from 27% at T1 to 12% at T4, a corresponding increase in the prevalence of the affective type from 24% to 32%, and a near doubling of the ambivalent type from 11% to 21%. When classifying the harmonious and behaviorally ambivalent relationship types as highly involved (i.e., they both show high levels of contact and support exchange), we find partial support for our third hypothesis, which stated that the relative number of relationships characterized by high involvement would increase over time. When the percentages of the harmonious and behaviorally ambivalent relationships are combined, the proportion of highly involved relationships increases from 31.7% at T1 to 38.1% at T4. It should be noted, however, that this increase is mainly driven by the increase in behaviorally ambivalent types, as the proportion of harmonious relationships slightly decreases with 3.3 percentage point from T1 to T4.

Transitions in Relationship Types Over Time

In this section, we examine transitions among relationship types across the four waves of measurement. Table 4 presents transition probabilities for the three intervals considered. We observe that most transitions occurred between T1 and T2, with relationship type stabilizing from T2 onward. Overall, 33% of relationships transitioned into another type between T1 and T2, where only 17% and 13% transitioned between T2–T3 and T3–T4, respectively. Stated differently, more than 80% of the relationships remained stable after the second time point, thus confirming our first hypothesis that intergenerational relations would show more stability and continuity than they would show change.

Table 4.

Transition Probabilities for Parent–Adult Child Relationships (N = 938 Nested in 295 Families)

	Transition probabilities to latent status at time t + 1
	Harmonious	Affective	Obligatory	Behaviorally ambivalent	Discordant
Latent status at Time 1
Harmonious	.720	.113	.030	.109	.028
Affective	.035	.887	.000	.058	.021
Obligatory	.115	.115	.477	.192	.102
Ambivalent	.231	.163	.065	.488	.053
Discordant	.016	.199	.020	.018	.747
Latent status at Time 2
Harmonious	.823	.071	.054	.053	.000
Affective	.029	.888	.012	.000	.071
Obligatory	.000	.167	.770	.027	.037
Ambivalent	.028	.000	.000	.949	.023
Discordant	.000	.127	.062	.131	.680
Latent status at Time 3
Harmonious	.778	.023	.000	.198	.000
Affective	.030	.874	.000	.061	.036
Obligatory	.030	.000	.883	.000	.087
Ambivalent	.024	.066	.000	.876	.034
Discordant	.000	.027	.021	.000	.952

Note. Values in bold are proportions of relationships that do not transition into a different type between time points.

Some relational types were more apt to change than others, however. Whereas more than 50% of relationships typified as obligatory or behaviorally ambivalent at T1 transitioned into a different type by T2, only 11% of those typified as affective transitioned into another type, with 25% of those in discordant and 28% in harmonious types also transitioning. We thus find partial support for our second hypothesis that relationship types characterized by warmth and affection (in our typology, the harmonious and affective types can be characterized as such) are more stable than those lacking such warmth and affection. Our sixth hypothesis, which stated that of all relationship types ambivalent relationships transition most often into other relationship types, is only supported for the first period of observation. Surprisingly, at later time periods, ambivalent relationship types were found to be the most stable (T2–T3) or third most stable (T3–T4).

In terms of specific transitions between T1 and T2, we note relatively high transition rates from behaviorally ambivalent to harmonious (23%), discordant to affective (20%), and obligatory to behaviorally ambivalent (19%), all designating an increase in some element of solidarity or support. After T2, only two conditional transitions are above 15% (obligatory to affective between T2 and T3 and harmonious to behaviorally ambivalent between T3 and T4).

We now turn to an examination of heterogeneity in transitions by inspecting their correlates based on personal and relational characteristics. Because most transitions occur between T1 and T2, we restrict our analyses to this time span, to those respondents who participated at both time points (n = 522), and to transitions that are most prevalent. Table 5 presents a series of logistic regressions corresponding to three outcomes: whether any transition occurred (instability), whether a transition occurred into the behaviorally ambivalent type (one of the most common destinations), and whether a transition occurred out of a type that is characterized by low involvement (i.e., affective, obligatory, or discordant).

Table 5.

Structural Predictors for Transitions From Time 1 to Time 2, Controlled for Clustering at Family Level

Predictors	Transition occurring (n = 522)			Into behaviorally ambivalent type (n = 461)			Out of a low involvement type (n = 361)
	B	SE B	OR	B	SE B	OR	B	SE B	OR
Events occurring between T1 and T2
Grandchild born (1 = yes)	−0.249	0.317	0.780	−0.760	0.778	0.468	−1.024	0.889	0.359
Child married (1 = yes)	−0.716	0.574	0.489	−1.815†	1.054	0.163	0.112	1.094	1.118
Child divorced (1 = yes)	−0.001	0.565	0.999	−1.663†	0.854	0.190	−1.469	1.059	0.230
Parent widowed (1 = yes)	0.764*	0.378	2.147	0.910	0.605	2.484	1.239†	0.676	3.451
Child into paid employment (1 = yes)	−0.241	0.607	0.786	−2.534*	1.124	0.079	−1.579	1.156	0.206
Child out of paid employment (1 = yes)	0.912*	0.432	2.489	1.832**	0.609	6.243	1.682*	0.801	5.377
Parent into paid employment (1 = yes)	0.295	0.544	1.344	0.679	1.140	1.971	1.185	1.072	3.271
Parent out of paid employment (1 = yes)	0.222	0.430	1.249	0.017	0.815	1.017	0.042	0.903	1.042
Parent retired (1 = yes)	−0.129	0.369	0.879	−0.092	0.740	0.912	0.114	0.762	1.121
Decline of parent’s health	−0.204	0.135	0.815	0.427*	0.204	1.532	0.650*	0.288	1.916
Improvement of parent’s health	0.038	0.221	1.039	−0.708	0.538	0.493	−0.918	0.580	0.400
Decrease in geographic distance	0.507**	0.169	1.660	1.087**	0.240	2.966	2.029*	0.300	7.607
Increase in geographic distance	0.431**	0.122	1.539	−0.318	0.274	0.728	−0.494†	0.298	0.610
Characteristics at T1
Parent’s gender (1 = female)	0.139	0.271	1.149	0.470	0.501	1.600	1.129†	0.622	3.091
Child’s gender (1 = female)	−0.450†	0.254	0.638	−0.994†	0.522	0.370	0.550	0.527	1.733
Child’s age (1 = yes)	−0.066†	0.039	0.936	−0.020	0.058	0.980	0.045	0.071	1.046
Child’s number of siblings	0.058	0.064	1.060	−0.101	0.095	0.904	−0.164	0.108	0.849
Child has (grand) children (1 = yes)	−0.598*	0.282	0.550	−0.576	0.493	0.562	−0.256	0.579	0.774
Child married (reference category)
Child unmarried cohabiting (1 = yes)	−0.567	0.717	0.567	0.540	1.182	1.716	−0.213	1.363	0.808
Child single or widowed (1 = yes)	−0.637	0.576	0.529	−1.441	1.441	0.237	−1.856	1.680	0.156
Child divorced (1 = yes)	0.520	0.423	1.683	0.208	0.645	1.231	−0.008	0.628	0.992
Parents married (reference category)
Parents divorced (1 = yes)	0.990**	0.313	2.691	0.865	0.595	2.376	1.400*	0.583	4.057
Parent widowed (1 = yes)	0.628†	0.367	1.874	1.778**	0.612	5.919	2.102**	0.680	8.187
Child employed (reference category)
Child nonemployed (1 = yes)	0.510	0.509	1.665	1.781*	0.750	5.935	1.107	0.718	3.024
Parent employed (reference category)
Parent nonemployed (1 = yes)	−0.115	0.446	0.891	−0.938	1.108	0.391	0.105	0.993	1.110
Parent retired (1 = yes)	−0.150	0.356	0.861	−0.407	0.699	0.666	−0.553	0.773	0.575
Parent’s health	0.030	0.107	1.031	−0.454**	0.167	0.635	−0.474*	0.234	0.623
Geographical distance	0.054	0.121	1.056	−0.685**	0.256	0.504	−1.027**	0.315	0.358
Child’s endorsement of norms	0.173	0.213	1.189	0.587	0.438	1.799	0.940*	0.457	2.560
Harmonious relationship type (1 = yes)
Affective relationship type (1 = yes)	−1.089*	0.454	0.336	1.495†	0.876	4.460
Obligatory relationship type (1 = yes)	1.197**	0.358	3.311	1.895*	0.756	6.653	1.083	0.906	2.953
Ambivalent relationship type (1 = yes)	1.403**	0.420	4.066
Discordant relationship type (1 = yes)	0.016	0.445	1.016	−1.377	1.066	0.252	−1.352	0.994	0.259
Constant	0.113	2.117		2.590	3.193		−0.152	3.779
χ2 (df)	150.97 (33)			178.37 (32)			187.20 (31)
% transitioned	31.42			9.54			15.79

Notes. OR = odds ratio; SE = standard error.

^† p < .10. *p < .05. **p < .01 (two-sided test).

The effects of both T1 and T1–T2 dynamic variables were estimated. T1 variables were the same as those considered earlier when assessing the external validity of our typology (Supplementary Appendix A). Because there were so few widowed children, we added them to the category of single children (omitting them altogether from the analysis did not affect the results substantially). Also, as relatively few parents and children were unemployed, we added them to the category of housekeepers, with the new dummy thus representing the nonemployed. For parents, the category of retired was kept separate from the nonemployed category. Dynamic variables described the most frequent changes that occurred during the interval, including the birth of a (grand)child, changes in children’s and parent’s marital and employment statuses, improvements and declines in parent’s health, and increases and decreases in geographic distance between children and parents. The reference category for each change variable was that no change occurred in the relevant characteristic. To account for family clustering of the data, we adjusted the standard errors using the cluster function in Stata (StataCorp, 2011). Missing data for the covariates were handled by multiple imputation using imputation by chained equations (ICE; Royston, 2007).

Turning to the first equation in Table 5, we found that the probability that a relational transition occurred increased when parents became widowed, when the geographic distance between the generations either increased or decreased, and when children moved out of paid employment. None of the other events affected the probability that a relational transition occurred. We thus find only partial support for our fourth hypothesis that life events trigger intergenerational relational change. This is only true for changes in the health and marital status of the older generation and changes in the employment status of the younger generation. Turning to the coefficients that measured the characteristics of both generations at T1, we observe that adult children who had young children in their homes at T1 were less likely than others to experience a transition in their relationship with parents. Parents who were already divorced at T1 are likewise more prone to experience a transition in their relationship with their adult child. The effects of the dummy variables representing relationship type at T1 were in line with findings in Table 4; harmonious and affective relationships were more stable and obligatory and behaviorally ambivalent relationships were more apt to change.

The second equation shows predictors of transitioning into a behaviorally ambivalent type. Both T1 state of parental widowhood and declining health of parents increased the likelihood that parent–child relationships transitioned into a behaviorally ambivalent type, as did decreasing geographic distance from parents. We thus find support for our fifth hypothesis, which stated that events resulting in increased dependency of the older generation would increase the likelihood that an intergenerational relationship will transition into a behaviorally ambivalent type. However, we should note that even stronger effects are found for the transitions children make into and out of paid employment: When children moved out of paid employment, their relationship with their parent was more likely to transition into a behaviorally ambivalent type, whereas the reverse was found for entering the labor market.

The third equation shows that the transition out of relationship types characterized by low involvement is associated with a decline in parent’s health, as well as reductions in geographic distance from parents. As low involvement relationship types, in our definition, were characterized by the relative absence of support exchanges, it is not surprising that increased parental dependency and greater proximity would move children and parents out of this type of relationship. Again, we find that changes in the employment status of children are more important than changes in their marital status in predicting whether the relationship type will change (in this case out of a low involvement type).

Discussion

In this study, we extend previous typological studies of intergenerational relationships by examining transitions in intergenerational relationship types and the correlates of the most commonly occurring transitions. Our analyses using regional data revealed that a five-class typology of family relationships best fit the data and that these types closely correspond in meaning to those found in several other schemes developed with nationally representative data in the United States and Europe (e.g., Silverstein et al., 1997; Van Gaalen & Dykstra, 2006).

One aspect of our typology that differs from several similar derivations in the literature relates to the identification of an ambivalent type. Although the measures used in the development of our typology allowed for the identification of an emotionally ambivalent type, we found no such type in our best fitting model. Instead, a behaviorally ambivalent relationship type emerged. Although this does not mean that emotional ambivalence is absent in intergenerational relations, it does suggest that behaviorally ambivalence is more discernable and more common as a distinct family type (similar results were reported by Van Gaalen & Dykstra, 2006, in Dutch data). Behavioral ambivalent relationships are supportive but conflictual, highlighting possibly hidden vulnerabilities resulting from interpersonal strain in these otherwise highly engaged families. We believe that our understanding of intergenerational relationships can be improved if future research on intergenerational ambivalence would not only consider psychological, but behavioral ambivalence as well.

In line with the life course framework, and more specifically the principles of life-span development, linked lives and phenomenon of path dependency (e.g., Elder et al., 2003), transitions among the five derived types provided evidence for both change and continuity in intergenerational relationships. Evidence of change was mostly apparent over the first interval and many of these changes were related to deficits and shifts in the lives of the older generation. In accordance with the principle of linked lives, increased dependency of the older generation, as indicated by declining health and long-term widowhood, was associated with intergenerational relationships shifting toward types characterized by support exchanges. Further, our results are consonant with the latent kinship model of intergenerational family functioning (Riley, 1983; Riley & Riley Jr., 1993), which predicts that strong emotional connections trigger support when older parents experience an increased need for assistance. Among characteristics of the younger generation, employment transitions were most related to transitioning into an ambivalent type of relationships. Employment transitions that reflect job loss and job gain reflect status changes may affect the degree of economic dependence of adult children on their parents, a state known to induce relationship strain and ambivalence.

Intergenerational relationships exhibited greater continuity as time passed in our observational period, providing some evidence for the principle of life-span development and phenomenon of path dependency, which suggests that current configurations of intergeneration relations are dependent on the history of that relation. The crystallization of relationship styles with time is evidence of inertia that in line with the literature on attachment as an enduring family style (Bowlby, 1969, 1977). We find that intergenerational types characterized by emotional closeness—harmonious and affective—are the least apt to change. The inherent stability of strongly integrated relationships is well known earlier in family life, but our findings provide evidence supporting this contention within mature families. Conversely, more dissonant relationships, particularly ambivalent ones, are less stable, but only so in the first interval of observation. Once parents advance into an age that is more strongly associated with dependency, avoiding ambivalence can no longer be realized. The general increase in relational stability over time suggests a reification of relationship styles that may have to do with a desire for consistency in middle-aged children and diminution in the ability of very old adults to adapt to change.

We note several limitations of this research. Because the number of relationships in each type at each time point can be quite low, we were not able to predict particular types of transitions based on unique origins and destinations. Although Schenk and Dykstra (2012) used LTA on a larger sample of relationships, the short time span between two waves produced an equally small number of transitions. We think it is important that future research have samples of sufficient size to capture various types of transitions and be careful in the choice of time lags as families might react quickly to volatile events.

A second limitation of our study concerns the reliance on a single informant for relationship data. Indeed, a number of studies have shown that perceptions of relationships differ between generations with parents reporting stronger emotional connections than their adult children (e.g., Aquilino, 1999; Bengtson, Olander, & Haddad, 1976; Mandemakers & Dykstra, 2008). This could also be a reason why we did not find a psychologically ambivalent relationship type in our data. Our choice to use only the reports of adult children was mainly pragmatic; using reports of both generations would have inordinately reduced our sample. Future research should investigate what effect reporting discrepancies between parents and their children have on the measurement, distribution, and dynamics of intergenerational relationships.

Further, our covariates lacked information on the length of time a child or parent had been in a particular status, or how recently a specific life event was experienced. This limitation introduces a conservative bias, as more precise measures on the timing of events might have better identified hypothesized effects.

Finally, we urge care in generalizing our findings beyond our geographically specific sample in which men, ethnic minorities, and the lower educated were underrepresented. These biases may have produced distortions in the nature of the derived types, as well as their distributions and patterns of change, in ways that are difficult to predict. The decision to drop the indicator for financial support when constructing out typology may affected our results as well. Economic support is an important element of intergenerational relationships and may enable the purchase of instrumental services on the private market. Thus, we likely underestimate the prevalence of instrumental exchange between generations. The absence of gender differences gives us pause as the literature is replete with findings showing that daughters are more integrated than sons with their older parents; this may indicate that sons participating in the study are highly selective. Despite these aforementioned limitations, the consistencies we observed with typologies developed in other samples provide some confidence that our findings are robust to variation in sample composition and are reasonably valid.

In conclusion, this study has demonstrated how typological analysis captures both the complexities and dynamics of intergenerational relationships in mature families. Transitions in relationship type generally occurred along the lines of what the life course framework and subordinate theoretical perspectives on attachment, ambivalence, and latent kinship would predict. Our results tell a story that is more about continuity than change; however, change, when it did occur, was primarily structured by circumstances that elevated the social dependency of older parents and the economic dependency of adult children, triggering both positive and negative changes in their relationships.

Supplementary Material

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

Funding

This work was supported by the Netherlands Organization of Scientific Research (400-08-203) and based on data from the Longitudinal Study of Generations, which is supported by grant R01-AG07977 from the National Institute on Aging.