Older Parent – Child Relationships in Six Developed Nations: Comparisons at the Intersection of Affection and Conflict

Abstract

Intergenerational solidarity and ambivalence paradigms suggest that emotional relationships between generations consist of both positive and negative sentiments. We applied latent class analysis to measures of affection and conflict in 2,698 older parent – child relationships in 6 developed nations: England, Germany, Israel, Norway, Spain, and the United States (Southern California). The best fitting model consisted of 4 latent classes distributed differently across nations but with a cross-nationally invariant measurement structure. After controlling for demographics, health, coresidence, contact, and support, the following classes were overrepresented in corresponding nations: amicable (England), detached (Germany and Spain), disharmonious (United States), ambivalent (Israel). We discuss policy and cultural differences across societies that may explain why the prevalence of particular emotional types varied by nation.

Conceptual and methodological advances in the social scientific study of intergenerational family relations have moved scholars to reconsider traditional approaches to assessing the quality of relationships between older parents and their adult children. First, theoretical and empirical models increasingly incorporate the possibility that family members may simultaneously have both warm and antagonistic feelings toward one another—an emotional dissonance identified in the literature as ambivalence (Luescher & Pillemer, 1998). Second, recent studies have highlighted the utility of categorical measurement models that are able to identify forms of family relationships, one of which is characterized as ambivalent (e.g., Van Gaalen & Dykstra, 2006). Third, the availability of multinational data focusing on older adults and their kinship networks has allowed scholars to compare intergenerational family relationships with the same instrumentation across a variety of societal and cultural contexts (e.g., Katz, Lowenstein, Phillips, & Daatland, 2005; Lowenstein, Katz, & Daatland, 2005).

In this investigation, we bring together these three features of contemporary investigations to examine (a) whether measures of affection and conflict taken from the intergenerational solidarity – conflict paradigm identify meaningful and comparable types of emotional relationships between older parents and their adult children in six developed nations, (b) whether the distribution of relationship types formed by affection and conflict vary across nations, and (c) whether behavioral and structural aspects of parent – child relationships account for cross-national differences in the types of emotional relations maintained.

Intergenerational Solidarity – Conflict and Ambivalence

The intergenerational solidarity paradigm—a comprehensive scheme for describing sentiments, behaviors, attitudes, values, and structural arrangements in parent – child relationships— has guided much of the research on adult intergenerational relationships over the past four decades (e.g., Atkinson, Kivett, & Campbell, 1986; Lee, Netzer, & Coward, 1994; Markides & Krause, 1985; Rossi & Rossi, 1990; Starrels, Ingersoll-Dayton, Neal, & Yamada, 1995). Building on theoretical and empirical advances in the social psychology of small group cohesion (Heider, 1958; Homans, 1950), the initial model codified six building blocks of intergenerational solidarity: emotional closeness, social contact, geographic distance, supportive behaviors, filial obligations, and attitudinal agreement (Bengtson & Schrader, 1982). The solidarity paradigm remains the gold standard as a measurement model for assessing intergenerational relationships (Bengtson & Roberts, 1991; Roberts & Bengtson, 1990; Silverstein, Parrott, & Bengtson, 1995), particularly the aspect of the paradigm that captures emotional aspects of such relations (Roberts & Bengtson, 1996).

Continuing efforts to refine the model included the addition of conflict (or its absence) as a seventh principal dimension to what has come to be known as the solidarity – conflict paradigm (Clarke, Preston, Raksin, & Bengtson, 1999; Parrott & Bengtson, 1999) but led to questions about whether affection and conflict were the antithesis of each other or whether they could coexist in strongly bonded relationships (Bengtson, Giarrusso, Mabry, & Silverstein, 2002). Parallel conceptual and empirical developments brought about by the integration of conflict with the prevailing approach to intergenerational solidarity, a renewed interest in the concept of ambivalence, and the use of person-centered (rather than variable-centered) approaches to study intergenerational relationships aided the answering of these questions. We discuss the developments below.

Conflict has long been considered intrinsic to social relationships, but it became formally integrated into mainstream sociological thought only with Georg Simmel's (1918/1955) classic essay in which he reasoned that conflict might serve an integrative function in intimate social relationships by allowing parties to let off steam (for a reappraisal of Simmel, see Coser, 1956). As Simmel (1918/1955) noted, conflict may be the “only means for making life with actually unbearable people at least possible” (p. 19). As applied to later-life intergenerational relations, conflict between adult children and frail elderly parents may provide a face-saving way for parents to accept care from a child or open up lines of communication in an otherwise discomfiting situation (for a review of theories of family conflict, see Farrington & Chertok, 1993). From this perspective, apathy is more detrimental than conflict to the integrity of such close personal relationships.

The idea that forces of attraction and repulsion are simultaneously present in the closest personal relationships formed the basis for the psychoanalytic concept of ambivalence—mixed positive and negative emotions toward the same relational object—long considered a source of neurosis deriving from early childhood attachment problems (Freud, 1913). Sociologists moved the concept of ambivalence beyond its pathological implications by considering it an intrinsic property of human relationships structured by irreconcilable demands for opposing behaviors toward another (Merton & Barber, 1963). Coser (1956) elaborated that “converging and diverging motivations may be so comingled in the actual relationship that they can be separated only for classificatory or analytical purposes, while the relationship actually has a unitary character sui generis” (p. 64). Luescher and Pillemer (1998) extended the application of ambivalence to intergenerational relations, critiquing the either-or approach to affectionate and conflictual aspects of mature parent – child relationships. Given that family life has its basis in the tension between the desire for autonomy and the need for interdependence, it is not surprising that intergenerational relations—throughout the family life cycle—are among the most ambivalent of social relationships (Fingerman, Hay, & Birditt, 2004). Intergenerational ambivalence in later life is more common in relationships in which older parents are frail and in declining health, possibly because they may become dependent on the adult children to whom they were formerly providers (Fingerman, Chen, Hay, Cichy, & Lefkowitz, 2006; Willson, Shuey, & Elder, 2003; Willson, Shuey, Elder, & Wickrama, 2006).

Scholars have used several types of measurement strategies to identify intergenerational ambivalence: direct strategies that ask respondents to rate the degree to which they have mixed feelings toward a parent or child (Pillemer & Suitor, 2002) and indirect strategies that ask respondents to independently rate the degree of closeness and conflict with a parent or child, where the researcher locates ambivalence in the intersectional social space where both emotions are strong (Bengtson et al., 2002). Various techniques have been used to indirectly capture ambivalence, including additive scales of dissonant (positive and negative) measures that describe the intensity of opposing feelings (e.g., Willson et al., 2006) and classification procedures that use such measures to group relationships into ambivalent and nonambivalent types (Giarrusso, Silverstein, Gans, & Bengtson, 2005; Hogan, Eggebeen, & Clogg, 1993; Silverstein, Bengtson, & Lawton, 1997; Silverstein & Litwak, 1993; Steinbach, 2008; Van Gaalen, & Dykstra, 2006). In this investigation, we use latent class analysis, a modeling-based classification procedure, to identify ambivalent relationships because, unlike additive ambivalence scales, such an approach can differentiate several types of accordant relationships (i.e., those characterized by strong affection or low conflict and weak affection or high conflict) and discordant relationships (i.e., those characterized by strong affection or high conflict and weak affection or low conflict).

Multinational Applications of the Solidarity – Conflict Paradigm

The intergenerational solidarity paradigm, though developed and validated with data from the United States, has been used in many countries around the world—for example, in Japan (Koyano, 1996), Germany (Steinbach, 2008), the Netherlands (Van Gaalen & Dykstra, 2006), New Zealand (Hillcoat-Nalletamby, Dharmalingam, & Baxendine, 2006), and Canada (Rosenthal, 1987). Formal cross-national comparisons using the solidarity paradigm have been carried out between an urban center in the United States and Wales (Silverstein, Burholt, Wenger, & Bengtson, 1998); between the recently reunified East and West Germany (Szydlik, 1996); and among England, Germany, Israel, Norway, and Spain (Daatland & Herlofson, 2003; Lowenstein, 2007). Thus, the nomenclature of the solidarity model—particularly those models that describe and measure the elements of emotional succor and friction between generations—has been shown to be generalizable across societal contexts, with the potential to extend to complex emotional states such as ambivalence.

Considering the solidarity – conflict model— essentially a set of social psychological constructs—at the institutional level of analysis requires concepts that theoretically bridge micro- and macrorealms of family life. One such bridging concept is that of structural ambivalence. As developed by Connidis and McMullin (2002), structural ambivalence refers to relational ambivalence induced by institutional forces exerting competing claims on the resources of family members. Such forces may manifest as role conflict (e.g., between work and family roles) or at the macrolevel in terms of welfare production. The generosity or restrictedness of state governments variably releases or obligates filial duties among adult children, with likely consequences for the emotional tenor of their intergenerational relationships.

Attributing individual phenomena to particular national characteristics is a well-known challenge in cross-national research given the myriad ways that nations differ from one another. Of particular relevance is Schooler's (1996) well-reasoned treatise affirming that social structural and cultural factors are more likely to influence psychological processes (than the reverse) and that controlling for particular individual-level characteristics provides some leverage in narrowing potential theoretical explanations. Thus, in developing expectations for our research, we relied on knowledge of the political economies and family cultures of nation-states to speculate about the nature of cross-national variations in the emotional ties between older parents and their adult children.

At the macro-structural level, welfare state structures differ depending on the way in which welfare production is allocated among state, market, and family. Esping-Andersen's (1999) widely used classification model suggests a gradient ranging from social-democratic states, in which all citizens are incorporated under a single universal insurance system (e.g., Scandinavian countries), to residualist states, in which the state assists citizens only when they have exhausted their personal resources (e.g., Mediterranean countries), with the middle ground occupied by liberal-market states, where assistance is means tested and modest social insurance plans are found (e.g., middle Europe and the United States). Because of the focus on welfare production, most cross-national comparative research on intergenerational relations in later life has focused on helping and care rather than on emotional cohesion between generations (e.g., Brandt, Haberkern, & Szydlik, 2009; Broese van Groenou, Glaser, Tomassini, & Jacobs, 2006). Little research has examined how the quality of intergenerational relations varies cross-nationally, despite the calls of sociologists to study how larger social structures regulate emotional expression (Thoits, 1989).

The political economy gradient maps well with filial obligation, which tends to vary inversely with the degree of welfare development. Familistic values are generally stronger in residualist nations of southern Europe than in the social-democratic nations of northern Europe (Hollinger & Haller, 1990; Inglehart & Baker, 2000). With regard to intergenerational relations, older parents have greater interaction with, live closer to, and tend more to live with their adult children the more southern their location is on the European continent (Hank, 2007). Recent research suggests that the involvement of adult children with their older parents tends to be more volitional in the welfare states of northern Europe than in the residualist states of southern Europe, where parental involvement tends to be more compulsory (Brandt et al., 2009). For that reason, older parents in nations with more evolved welfare systems may also have less conflict with their children than do those in nations with more residualist policies. In contrast, older parents in nations with weaker social policies may exhibit both more affection and more conflict because of their greater involvement with and dependence on adult children in such nations.

Intimate relationships are also bound by particularistic cultural or personality styles that are more difficult to identify than political structures but are no less national in character (Turner, 1988). A social scientific investigation into national cultural profiles conducted by Peabody (1985) found tendencies among the British to “get along with others,” “inhibit hostility,” and exhibit “self-restraint … [and] a smiling affability” in their private relations (pp. 97 – 99). If taken at face value, this profile implies that older parents in England may be apt to emphasize cordiality and minimize conflict with their adult children. Peabody (1985) also found that Germans tended to value honesty and expressiveness in their interpersonal relationships, which leads us to expect greater conflict in the intergenerational ties of parents in our German sample. Knowledge of cohort and historical factors leads to the same conclusion. Here we refer to Szydlik's (1996) observation that the older generation in Germany, by virtue of their association with the National Socialist regime of the 1930s and 1940s, may have strained relations with their adult children who vehemently rejected the tenets of that past society and the destruction that it brought.

Further, we expect that affection and conflict in the parent – child axis of families will be more intense in nations that have strong familistic cultures. This gradient—ordered from more to less familistic nations—is bracketed by the extremes of Spain and Norway but is less clear among the middle countries because of the idiosyncratic ways that those nations differ. Although familistic nations also tend to have more restricted public service sectors, this relationship is far from deterministic. For instance, Israel embodies a set of paradoxical elements by having a strongly familistic culture, social-democratic policies and a developed service network, and a sociopolitical environment that legitimates a contentious cultural style of interpersonal relations. The United States is a more mobile society than four out of five of the other nations but has a less developed social welfare system that compels greater reliance on informal sources. Given the complexities in weighing the political, cultural, and historical differences across the six nations we study, our expectations regarding national differences in the emotional connections between older parents and their children are more than speculations but fall short of formal hypotheses.

To summarize the above discussion, adding conflict to the solidarity paradigm has allowed for the measurement and detection of ambivalence as an emergent property of intergenerational relationships and one that is undergoing increased scrutiny and conceptual refinement. Clustering approaches provide the means to represent various types of relationships formed at the intersection of affection and conflict. Incorporating comparable measures of affection and conflict into the protocols of multinational surveys has provided the opportunity to formally test how societal context—including structural and cultural conditions—shapes personal family relationships across diverse societies. In this research, we directed our attention to the emotional component of parent – child relationships, which has been found to be central to the quality of life of elders in various societies (e.g., Lowenstein, 2007). On the basis of the dimensions of affection and conflict, we developed a scheme for classifying emotional ties between older parents and their adult children, examining its structure and distribution in samples from six developed nations that vary in their social policies, cultural milieus, and social histories.

Although our focus is on national differences in the quality of emotional relationships, it is important to consider individual and family factors found to influence how emotional bonds are maintained across generations (see Szydlik, 2008). These factors include social characteristics (gender, age), resources (education), need (physical impairment), family structure that signals the availability of alternative relationships (number of siblings, presence of a spouse) and interactions or transactions with children (face-to-face contact, coresidence, support received).

Method

Samples

Data for cross-national comparisons came from two sources: the five-nation study known as Old Age and Autonomy: The Role of Service Systems and Intergenerational Family Solidarity (OASIS) and the Longitudinal Study of Generations (LSOG), concentrated in Southern California. Measures deriving from the solidarity – conflict model, developed in the LSOG, have been included in the OASIS surveys (Lowenstein, Katz, Mehlhausen-Hassoen, & Prilutzky, 2001), which provides the opportunity for direct comparisons between nations.

Funded by the European Commission, OASIS is a multinational study that investigated intergenerational relationships across five nations—England, Germany, Israel, Norway, and Spain—with different welfare regimes and various family cultures. The overall goal of the project was to better understand how families and formal systems interact to support autonomy and quality of life in old age (Katz et al., 2005). Data for the OASIS study were collected in 2000 – 2001 through face-to-face surveys of representative samples drawn from the urban populations of those age 65 and older, yielding more than 400 respondents in each of the five national samples (see Table 1). Professional survey research organizations in each nation conducted sample selection, recruitment, and interviews under the supervision of OASIS project staff. Sampling was done in each country using a stratified multistage cluster method that targeted urban areas only (cities with populations of at least 100,000) and included an oversample of the inhabitants aged 75 years and older. Overall response rates varied between 70% and 76% among the five countries (Lowenstein, 2007).

Table 1. Characteristics of Older Parents and Adult Children in Six Nations (N = 2, 698).

	Englanda	Germanya	Israela	Norwaya	Spaina	LSOGb	p valuec
Parent characteristics
Mothers (%)	66.1	63.5	50.7	54.2	61.9	61.3	.000
Married (%)	49.3	48.9	43.6	44.4	51.3	70.0	.000
Age in years (mean)	77.2	78.0	77.3	78.5	77.0	73.4	.000
College education (%)	8.5	11.9	36.4	39.0	4.3	34.9	.000
Problem climbing stairs (%)	47.7	42.7	39.2	23.5	50.7	31.4	.000
Child characteristics
Daughters (%)	47.9	51.2	53.2	51.5	55.2	57.3	.082
Married (%)	68.8	78.3	87.3	61.7	78.0	71.6	.000
Age in years (mean)	46.9	49.4	46.5	47.3	46.1	49.4	.000
Number of siblings (mean)	1.3	.9	1.6	1.3	1.7	2.3	.000
Relationship characteristics In-person contact/coresidence							.000
Live together (%)	8.9	7.9	5.3	4.7	22.5	3.3	—
Daily contact (%)	9.0	9.0	6.7	8.6	17.2	7.6	—
Weekly contact (%)	47.0	37.3	58.7	42.8	44.7	29.2	—
Less than weekly contact (%)	35.1	45.8	29.4	44.0	15.7	60.0	—
Parent receives help with chores (%)	17.6	17.2	10.2	9.5	14.7	8.7	.000
Sample size	492	472	442	410	417	465

^aOASIS.

^bLSOG.

^cChi-square test performed for nominal variables and ANOVA for continuous variables.

The LSOG began in 1971 with 2,044 respondents who were members of three-generation families within which the grandparent generation was living in Southern California. Grandparents (G1) 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 (HMO) (for further details, see Bengtson & Schrader, 1982). Adult children (G2) and grandchildren (G3) of the G1 grandparents were also invited to participate in the survey. Follow-up surveys were administered to original respondents in 1985, 1988, 1991, 1994, 1997, and 2000. All data were collected by mail-back surveys. In this analysis, we focus on 465 respondents, almost all from the G2 generation, who were at least 65 years of age in 2000. Respondent retention rates (mortality adjusted) in the older generations averaged more than 75% over the course of the study (Gans & Silverstein, 2006).

Although the LSOG is a regional sample with a large majority of respondents living in the five-county Southern California region, several aspects of this study provide it with advantages as a comparative data set to the OASIS samples. First, much like OASIS, the LSOG sample is primarily urban in nature. Second, the most recently available data from the LSOG were collected during the same time interval as the OASIS project. Third, under a collaborative agreement between the OASIS and LSOG scholars, the protocols used to measure intergenerational solidarity and conflict were identical across the samples.

The final pooled sample (OASIS nations and LSOG) was 2,698. Given that missing values were relatively rare in the data (maximum of 5.8% among independent variables), we used mean substitution as an imputation strategy. Only 2.7% of respondents had missing values on items used to construct the dependent variable and were dropped from the analysis (N = 74). Table 1 shows descriptive characteristics and sizes of the six samples. Cross-national differences were found in the distribution of all variables except for gender of child.

Measures

In each of the six samples, parents were asked to answer questions about their relationship with a single randomly chosen child. The key measures in this analysis capture affectual and conflictual dimensions of the solidarity – conflict model. Questions were originally developed in English and translated for use in non-English-speaking countries. In non-English-speaking nations, the accuracy of translations was ensured through back-translation methods and extensive pretesting of the instrument.

Affectual solidarity was measured using the following questions: How close do you feel to this child? How well do you and this child get along together? and How is communication between yourself and this child? Conflict was measured using the following questions: How much conflict, tension, or disagreement do you feel there is between you and this child? How much do you feel this child is critical of you or what, you do? and How much does this child argue with you?

Each of the affection and conflict questions was answered on a 6-point Likert-type scale. Affect responses range from not at all to extremely, and conflict responses range from none at all to a great deal. In each country, distributions of the six items departed substantially from normality, with a strong positive skew for affection items and a strong negative skew for conflict items. Thus, we created dichotomous indicators from the raw variables on the basis of their distributional properties. Affection items were dichotomized with stronger attachment indicated by responses of extremely or very close/well/good (vs. pretty, somewhat, not too, or not at all). Conflict items were dichotomized with more intense conflict indicated by having a great deal, quite a bit, pretty much, or some conflict/arguing/criticalness (vs. a little or none at all).

Personal characteristics of parents included in our model were gender (1 = mother; 0 = father), age (in years), marital status (1 = married; 0 = unmarried), education (1 = at least some college; 0 = less than college), and ability to climb stairs (1 = has difficulty; 0 = no difficulty). For children, we included the following variables: gender (1 = daughter; 0 = son), marital status (1 = married; 0 = unmarried), age (in years), and number of siblings.

Statistical Procedure

In developing a typology of the emotional structure underlying intergenerational relationships, we used latent class analysis (LCA) to group relationships into ideal types based on the dichotomous indicators of affection and conflict. Latent class analysis is a statistical tool that allows researchers to posit a set of unobserved latent classes that accounts for the associations among observed variables—a condition known as local independence (Clogg & Goodman, 1984; Lazarsfeld & Henry, 1968). This property underlies a likelihood ratio chi-square goodness-of-fit statistic (L²) testing the discrepancy between a particular theoretical model and the observed data. An L² that is not statistically significant provides the basis for assuming the adequacy of a given specification. However, because many models may adequately fit the data, preference is given to well-fitting models that have fewer latent classes or estimate fewer parameters. We also relied on the Bayesian Information Criterion (BIC) statistic (Raftery, 1986) when selecting among competing models because BIC advantages more parsimonious models and disadvantages model complexity in assessing goodness of fit (Clogg, 1995).

In our application, we used the software Latent Gold, V.2.0 (Vermunt & Magidson, 2002) to analyze the 64 response patterns formed by the cross-classification of 6 dichotomous indicators of affection and conflict within each of the 6 samples. For a given latent class structure, two types of statistics were generated: conditional latent class probabilities (CLCP) and latent class probabilities (LCP). The CLCPs described characteristics of identified latent classes based on the distributions of manifest items among individuals assigned to each class. We used those probabilities to profile and then define the latent classes. The LCPs described the distribution of latent classes across a population. Both sets of statistics were tested for their invariance across the subpopulations being studied.

After an optimal solution was identified, we examined the number and measurement structure of the latent classes and the probability distribution of class representation for each nation, and we then used multinomial regression to test for national differences in the odds of class membership. Our multivariate models were estimated in two stages. We first examined national differences with control variables included for substantive reasons and to ensure that national effects were independent of compositional differences across the samples. For instance, LSOG parents were younger and more likely to be married than OASIS parents because of oversampling of those 75 and older in the latter study. Next, social relationship factors were added to the equation to determine whether national differences in the emotional content of intergenerational relationships were explained by variation in social dimensions of those relationships—specifically coresidence and the amount of interaction with the focal child, and whether support was received from that child—with the demographic composition of the samples held constant.

Results

Our first goal was to develop a well-defined, descriptively meaningful, and generalizable typology of intergenerational emotional ties. To do this, we first determined whether the best fitting model in each national sample had the same number of latent classes and, if so, whether the classes had similar profiles and were similarly distributed across the samples. We tested a series of latent class structures in each nation separately, successively adding classes and observing the change in goodness-of-fit statistics for each successive model. In all nations, the L² statistic was statistically significant for the two-class and three-class models but not significant for the four-class model, which signified that four latent classes provided a good fit to the data of each sample (for fit statistics of the four-class model, see Table 2). Further, the BIC statistic dropped precipitously with each additional latent class (not shown)—indicating relative improvements in fit—up to four classes, after which it reached an asymptote and then increased, thereby confirming the superiority of the four-class model in each nation. Still, there was some variation in the goodness-of-fit of this model, with England and Germany attaining the lowest BICs (indicating better fit) and the LSOG attaining the highest BIC (indicating worse fit).

Table 2. Goodness of Fit for Four-Class Model in Six Nations (N = 2,698).

	L2	df	p value	BIC
Samples
Englanda	29.19	36	.78	– 193.96
Germanya	28.90	36	.79	– 192.76
Israela	31.65	36	.68	– 187.64
Norwaya	32.88	36	.62	– 183.70
Spaina	29.39	36	.78	– 187.80
LSOGb	42.83	36	.20	– 178.29
Models with samples aggregatedc
Model 1: No cross-national equality constraints	194.84	216	.85	– 1,511.62
Model 2: Conditional probabilities for item responses equal across nations	306.22	311	.56	– 2,150.76
Model 3: Latent class and conditional probabilities for item responses equal across nations	464.20	326	<.001	– 2,111.29
Model 2 vs. Model 1	L′-111.38	L′-95	.12	—
Model 3 vs. Model 2	L′-157.98	L′-15	<.001	—

^aOASIS.

^bLSOG.

^cAll item-by-item residuals and nation-by-conflict residuals are fixed at 0. Nation-by-affection residuals are free to vary to obtain best model fit.

The next series of models aggregated the six national samples with the goal of testing for cross-national invariance in latent class parameters. The best fitting and most parsimonious of the aggregated four-class models (both unrestricted and restricted) was achieved by first fixing all item-by-item residuals and nation-by-conflict residuals to 0 and allowing nation-by-affection residuals to vary. That nation-by-affection residuals were best left free suggested differential item response to affection items in the preferred four-class structure. Although this aspect of the model did not alter the basic organization and interpretation of the classes, it revealed some cross-national variation in the consistency with which affection items behaved in representing the four latent classes.

Restricted models were formed by imposing a series of equality constraints on the four-class model to test for cross-national invariance in the conditional probabilities (CLCP), constituting the measurement part of model, and the latent class probability distributions (LCP), constituting the prevalence part of the model. To do this, we compared the chi-square statistic of a particular equality constrained model against the chi-square statistic of a less constrained model using a difference test. The fit of the unconstrained aggregated model (calculated as the sum of the fit statistics of the nation-specific models) is shown as Model 1 in Table 2 and represents the baseline against which we compared the first equality-constrained model. Imposing cross-national equality constraints on CLCPs in Model 2 resulted in a reduction of model fit that was not statistically significant (ΔL² 111.38, Δ95df, p = .12), which suggests that the measurement parameters were invariant with respect to national context. (It should be noted that our expectation of a generalized structural model of intergenerational relationships was confirmed via a global test, and there still exists the possibility that specific countries were different from one another.)

When we imposed cross-national equality constraints on LCPs in Model 3, the resulting decrement in fit compared to Model 2 was statistically significant (ΔL² 157.98, Δ15df, p < .001), which indicates that nations differed in their latent class distributions. Comparing BIC statistics for the three models revealed that the unconstrained model (full independence across nations) had the highest value and the CLCP-constrained model had the lowest value. Therefore, we accepted a partially constrained four-class model—one in which latent classes have the same measurement profile but vary in their prevalence across nations—as the preferred model.

We used the CLCPs for the accepted model to profile each latent class. Following McCutcheon (2002), we interpreted the probabilities in both absolute and relative terms when defining the classes. As Table 3 shows, the first class had high probabilities on affection items and low probabilities on conflict items, thus suggesting an amicable type of relationship. The second class had low probabilities on both affection and conflict items, implying an emotionally detached type of relationship. The third class was characterized by low affection and high conflict probabilities, a type we referred to as disharmonious. Finally, the fourth class exhibited high probabilities on affection and conflict items, thus suggesting opposing feelings or an ambivalent type of relationship. We note that the classes identified correspond to relationship types found in our earlier research using only the LSOG sample (Giarrusso et al., 2005).

Table 3. Latent Class Probabilities and Average Latent Class Distribution for Partially Constrained Four-Class Model (N = 2,698).

	Conditional Latent Class Probabilities
Measures	Amicable	Detached	Disharmonious	Ambivalent
Closeness	.93	.25	.14	.92
Getting along	.99	.27	.01	.93
Communication	.91	.12	.03	.62
Conflict	.01	.05	.76	.40
Criticalness	.03	.06	.75	.61
Arguing	.01	.03	.69	.43
Nations	Latent Class Probability Distributions
Englanda	.75	.15	.03	.07
Germanya	.49	.43	.07	.01
Israela	.62	.15	.08	.14
Norwaya	.64	.19	.08	.09
Spaina	.63	.25	.08	.05
LSOGb	.51	.20	.20	.09
Total pooled	.61	.23	.09	.08

Note: Latent class probabilities greater than .4 are considered relatively high and are shown in bold.

^aOASIS.

^bLSOG.

The second panel of Table 3 shows nation-specific distributions of the four latent classes. We discuss the most divergent national results, all of which were significantly different from the total pooled distribution, also shown in Table 3. Three quarters of parents (75%) in England had amicable relationships with their children, compared to only about half in the LSOG (51%) and Germany (49%). Detached relationships were most common in Germany (43%) and most rare in Israel (15%) and England (15%). Disharmonious relationships were unusually common in the LSOG (20%) and relatively rare in England (3%). Ambivalent relationships were most prevalent in Israel (14%) but highly unusual in Germany (1%).

We next examined the predictors of latent class membership using multinomial logistic regression with the amicable type as the reference group. In these equations, we controlled for characteristics of parents and children and behavioral aspects of their relationships that may explain cross-national differences in relational quality (and that adjust for unplanned compositional differences between the samples). We operationalized national context as a set of effect coded variables, which means that national effects should be considered contrasts with the unweighted national average (the LSOG sample was the omitted group). Effect coding was preferred over the more common dummy coding approach because comparing nations to the average was considered a more generalized and informative contrast than comparing them to any one specific country. For illustrative purposes, we also show coefficients for the effect-coded LSOG sample but without showing tests of statistical significance. Estimated logit coefficients were transformed into odds ratios (ORs) for ease of interpretation. We estimated two sets of equations. In the first set, we included variables for national context and individual-level characteristics of parents and their referent children. In the second set of equations, we added relational characteristics: frequency of contact and coresidence with children, as well as receipt of help from children with household chores. These variables controlled for exposure to and reliance on children—family features that tend to be more common in traditional societies (Hank, 2007).

The first equations in Table 4 revealed national differences similar to those found in the bivariate analysis, which suggests that national patterns were robust to sampling variability and compositional differences across national contexts. Intergenerational relations in England were almost half as likely (OR = .54) to be detached and almost three quarters less likely (OR = .27) than average to be disharmonious than amicable. In Germany, relations were almost three times (OR = 2.8) more likely than average to be detached and 80% less likely (OR = .18) to be ambivalent than amicable. Relations in Israel were over 3 times more likely (OR = 3.6) than average to be ambivalent than amicable. In Norway, relationships were 24% less likely (OR = .76) to be detached than amicable. No differences were found for relationships in Spain.

Table 4. Multinomial Odds Ratios Predicting Class Membership as Derived From Partially Constrained Four-Class Model (N = 2,698).

	Odds Ratios of Latent Class Membership vs. Amicable Class
	Model 1			Model 2
Latent class predictors	Detached OR (CI)	Disharmonious OR (CI)	Ambivalent OR (CI)	Detached OR (CI)	Disharmonious OR (CI)	Ambivalent OR (CI)
Nation-sample (vs. average)
Englanda	.54*** (.43 – .69)	.27*** (.17 – .43)	.76 (.49 – 1.16)	.53*** (.42 – .68)	.27*** (.17 – .44)	.75 (.49 – 1.15)
Germanya	2.81*** (2.31 – 3.41)	1.05 (.73 – 1.52)	.18*** (.07 – .47)	2.66*** (2.17 – 3.25)	1.05 (.72 – 1.53)	.17*** (.06 – .44)
Israela	.78 (.61 – 1.00)	1.20 (.85 – 1.68)	3.59*** (2.57 – 5.02)	.83 (.64 – 1.06)	1.30 (.92 – 1.84)	3.60*** (2.56 – 5.06)
Norwaya	.76* (.60 – .97)	.96 (.68 – 1.36)	1.08 (.71 – 1.67)	.71** (.55 – .91)	.94 (.66 – 1.34)	1.14 (.74 – 1.75)
Spaina	1.14 (.91 – 1.42)	.93 (.65 – 1.33)	.82 (.51 – 1.32)	1.43** (1.13 – 1.81)	1.00 (.68 – 1.47)	.81 (.49 – 1.33)
[LSOGb]	[.96]	[3.29]	[2.33]	[.84]	[2.85]	[2.44]
Parent characteristics
Mother (vs. father)	.76* (.61 – .94)	.71* (.52 – .98)	.92 (.65 – 1.31)	.75* (.60 – .94)	.72* (.53 – .99)	.92 (.65 – 1.32)
Married (vs. unmarried)	.85 (.68 – 1.06)	.67* (.48 – .92)	.78 (.54 – 1.11)	.85 (.68 – 1.07)	.69* (.49 – .95)	.80 (.56 – 1.15)
Age in years	.98 (.96 – 1.00)	.97 (.94 – 1.00)	.98 (.94 – 1.01)	.98 (.96 – 1.01)	.97 (.94 – 1.01)	.98 (.94 – 1.01)
College (vs. no college)	1.03 (.79 – 1.33)	.93 (.66 – 1.31)	1.2 (.85 – 1.73)	1.01 (.78 – 1.31)	.94 (.66 – 1.32)	1.22 (.85 – 1.74)
Problem climbing stairs (vs. no problem)	1.21 (.98 – 1.50)	1.42* (1.04 – 1.92)	1.15 (.81 – 1.62)	1.25* (1.01 – 1.55)	1.47* (1.08 – 2.00)	1.09 (.77 – 1.55)
Child characteristics
Daughter (vs. son)	.76** (.62 – .92)	.76 (.57 – 1.01)	1.22 (.88 – 1.68)	.80* (.65 – .98)	.82 (.62 – 1.10)	1.22 (.88 – 1.69)
Married (vs. unmarried)	.79* (.63 – 1.00)	.50*** (.36 – .68)	.70 (.48 – 1.02)	.72** (.56 – .92)	.51*** (.37 – .72)	.76 (.52 – 1.13)
Age in years	1.01 (.99 – 1.02)	1.00 (.97 – 1.02)	.98 (.95 – 1.00)	1.00 (.99 – 1.02)	1.00 (.97 – 1.02)	.98 (.95 – 1.01)
Number of siblings	1.04 (.96 – 1.13)	1.01 (.90 – 1.14)	.97 (.85 – 1.11)	1.02 (.93 – 1.11)	1.00 (.88 – 1.13)	.98 (.85 – 1.12)
Relationship characteristics
Live together (vs. monthly contact or less)	—	—	—	.40*** (.26 – .61)	.86 (.50 – 1.48)	1.36 (.70 – 2.66)
Daily contact (vs. monthly contact or less)	—	—	—	.28*** (.18 – .42)	.32*** (.17 – .60)	.81 (.42 – 1.53)
Weekly contact (vs. monthly contact or less)	—	—	—	.49*** (.39 – .61)	.40*** (.28 – .56)	1.18 (.81 – 1.72)
Child helps with chores (vs. child does not help)	—	—	—	1.15 (.86 – 1.55)	.70 (.42 – 1.17)	1.67* (1.06 – 2.63)
Pseudo R2		.164			.202

^aOASIS.

^bLSOG.

^*p < .05.

^**p < .01.

^***p < .001.

The effects of parents' characteristics revealed that relative to having an amicable relationship, fathers were more likely than mothers to have detached and disharmonious relationships, unmarried parents were more likely than married parents to have disharmonious relationships, and parents with difficulty climbing stairs were more likely than parents without such difficulty to have disharmonious relationships with children. Turning to characteristics of children, parents were more likely to have detached relations with sons than with daughters and more likely to have detached and disharmonious relations with unmarried children than with married children.

In the second equation of Table 4, we added variables signifying social involvement with children (frequency of in-person contact, coresidence, and receipt of help with household chores). Not surprisingly, increasing intergenerational contact reduced the likelihood of having detached and disharmonious types of relations, and coresidence reduced the risk of detached relations. Parents who received help with household chores from a child were two thirds more likely (OR = 1.67) to have ambivalent relationships with that child. With the addition of contact, coresidence, and support receipt, the effects of national context changed little with one exception. Parents in Spain were more than 40% more likely than average (OR = 1.43) to have detached than amicable relationships with their children after taking into account their relatively frequent rates of intergenerational contact, coresidence, and support. That is, emotional detachment in Spanish families emerged as more common only when nations were equated on social involvement with their children.

Discussion

In this investigation, we examined the latent class structure of two core dimensions of family relations—emanating from but not restricted to the solidarity – conflict model—to describe and predict the quality of emotional bonds between older parents and their adult children in six nations. Using multinational data of older parent – child relationships, we have demonstrated that a four-class model best described the configurations formed by the dimensions of affection and conflict. Our results showed that the profiles of the four derived types—amicable, detached, disharmonious, and ambivalent— were similar across the six countries studied, which suggests that the underlying structure of emotional ties between generations in the family is robust across diverse developed nations. However, the relative prevalence of types varied across nations, with amicable relations more common in England, detached relations more common in Germany, disharmonious relations more common in the LSOG (United States), and ambivalent relations more common in Israel.

We found that once we considered social involvement with children, cross-national differences in the prevalence of each type roughly followed expected patterns, particularly at the extremes of family culture and state functions. In Spain, the high amount of intergenerational contact and high likelihood of coresidence in that nation initially suppressed the predominance of a detached style. When level of exposure to children was held constant, other social forces came to the fore that may explain why intergenerational relations were less emotionally connected in Spain than in the other countries. These forces include a more coercive family culture reinforced by a residualist public sector, which together may produce obligatory affiliations between older parents and their adult children. Alternatively, this may be due to a generation gap caused by rapid modernization of a traditional society that disproportionately affects the young (e.g., severely low fertility rates). We believe that it is instructive to frame this finding as a counterfactual of what intergenerational relationships in Spain would resemble if proximity and contact were in line with those of other countries in the study. That is, without their greater proximity and contact, Spanish elderly would be more likely than elderly parents in other countries to be detached from their children. This suggests a steep distance – contact gradient in relationship quality in Spain, where having distant and infrequently seen children is more deviant and has greater meaning than in other countries. Spanish elderly who have such children would be least likely to maintain intimacy at a distance with them.

Our finding with respect to Norway—that relations were more likely to be amicable than detached—is consistent with the idea that intergenerational relations are more voluntarily engaged when children are freed from the demands of onerous caregiving duties. These interpretations for Spain and Norway rest on the broad assumption that national political economies influence microlevel emotional linkages between generations, though they are but-tressed by European research showing that adult children are more likely to provide essential and intensive assistance to their older parents in nations with restrictive welfare provisions but more likely to provide discretionary and intermittent assistance in nations with generous welfare provisions (Brandt et al., 2009).

Intergenerational relationships in Israel were more likely than average to be ambivalent, which reflects a culture that emphasizes interpersonal engagement and legitimates forms of conflict as a social adhesive in intimate relationships. In this context, ambivalence describes a state more of emotional intensity than of emotional uncertainty (as connoted in most direct definitions of ambivalence). What this tells us is that an understanding of national and cultural context is important when attributing the meaning of relational qualities, such as ambivalence, to family actors in a given population.

Our findings were also anticipated with respect to the LSOG, a sample based in Southern California serving as the U.S. counterpart to OASIS. Compared to its national counterparts, the United States embraces a stronger individualistic ideology with respect to kinship ties (Hollinger & Haller, 1990) and has a weaker public service sector than other nations in our study, with the possible exception of Spain. Thus, it is not surprising that intergenerational relations in the LSOG were more than twice as likely as those in the other national samples to be characterized as disharmonious and detached—the two relational styles with higher levels of conflict.

Findings with respect to Germany and England are somewhat more difficult to explain, but other research provides clues. The greater prevalence of emotional detachment among parent – child relationships in Germany possibly reflects an intergenerational schism rooted in the association of older parents with the National Socialism regime, a history fully repudiated by their now middle-aged children (Szydlik, 1996). The finding that intergenerational relationships in England are more likely than those in other nations to be emotionally close and free of conflict could result from a cultural tendency to inhibit the expression of strong negative emotion (Peabody, 1985).

An increasing number of studies are taking advantage of recently available multinational data to study kinship patterns. Much of this research suggests that both socioeconomic development and sociocultural factors are responsible for variation in intergenerational support and contact across nations (Glaser, Tomassini, & Grundy, 2004; Hollinger & Haller, 1990). Although our investigation focused on the emotional nature of intergenerational bonds rather than on their functions, our results have direct relevance to support provision, as parents in poorer functional health tended more to have detached and disharmonious relationships with their children, and those who received help from children tended more to have ambivalent relationships with them. Together, the findings suggest that frailty and dependence on children introduce elements of friction and strain into intergenerational relationships. Whether these qualities are ultimately divisive and undermine the support potential of adult children will await further research on this topic.

We found evidence that marital status plays a role in shaping emotional ties between older parents and their adult children. Marriage, both of parents and children, increased the likelihood that relationships were tight knit, which is consistent with the notion of marriage as a greedy institution that crowds out other social relationships (Coser & Coser, 1974). Unfortunately, there was no information from the parent's perspective about the employment status and parental status of their children, which may serve as additional sources of role competition.

We note several limitations of this research. First, there are methodological differences in the collection of data in OASIS and the LSOG, with the former collected through in-person surveys and the latter collected through mail-back surveys. Although this inconsistency may compromise comparability, it is unclear whether and how it systematically influences responses across the two studies.

Second, we intentionally limited our typology to affection and conflict to map the emotional terrain of intergenerational relationships. Although we used several other dimensions of the solidarity – conflict paradigm (associational, functional, structural solidarities) as predictors in our model, these additional dimensions would have likely enriched the typology itself, and this remains an important topic for future study.

Third, only one child in each family was chosen to be the object of parental reports. Although it is important to reiterate that the children were randomly selected from their families, this approach does not allow us to identify intrafamilial consistency and variation in relationship styles. Future research using reports about multiple children will allow us to identify on what basis parents develop preferences for their children.

Fourth, our goal of establishing a generalizable model was limited by the relatively narrow range of developed nations and the strictly urban samples considered in our analyses. Much would be gained from testing our four-class solution in less developed nations of the world as well as in rural regions of the developed world. The findings that reflect national differences provided some evidence that macro-contextual factors are important considerations in how relationships are managed at the microlevel; however, without a larger sample of nations, it is not possible to identify the mix of economic, government, and cultural forces at work.

Finally, our investigation considered the perspective of the older generation but not that of the younger generation in these paired relationships. We suspect that older parents would be more likely to report affection and less likely to report conflict than their adult children. Thus, older parents may perceive their intergenerational relations as more amicable and less disharmonious than their offspring do, a perceptual difference predicted by the generational-stake hypothesis (Giarrusso, Stallings, & Bengtson, 1995). Addressing these questions will clarify the subjective and intersubjective nature of positive and negative sentiments in dyadic intergenerational relations, where several perspectives are considered. Further, intergenerational relationships are highly dynamic at later stages of life, when health and social changes in the older generation may be sudden. Consequently, our categorization may reflect transitory relational states in families. Longitudinal models will be needed to establish whether emotional ties between generations are stable or change in response to altering conditions.

The simultaneous presence of affection and conflict in intergenerational relationships reflects emotional complexities that are intuitively obvious to anyone who is part of a family. However, conflict, as a late addition to the initial solidarity paradigm, has remained somewhat conceptually apart from the original dimensions of the model. We suggest that advances in family science—particularly recent developments in ambivalence theory as applied to intergenerational relations and classical sociological perspectives on conflict as a socially integrative force—open up strategies to theoretically incorporate conflict into the multidimensional solidarity scheme. Our empirical approach differentiated several nonambivalent types as well, providing a more comprehensive map of the emotional organization of intergenerational relationships than an approach that relies solely on ambivalence as a metric.

The evidence in this investigation points to the importance of national context in structuring emotional ties between older parents and their adult children. Identifying the ecological conditions responsible—welfare state structure, economic development, and/or cultural values—will require a larger sample of nations on which to map these multiple pathways. Such future research will be best situated to trace specific linkages between the public sphere and the private realm of intergenerational family relations.