Social Participation across Mid- and Later-Life: Evidence from a Longitudinal Cohort Study

Abstract

Despite the benefits of social participation for individuals and communities, little is known about how social participation varies over the life course. Drawing upon data collected between 1957 and 2011 by the Wisconsin Longitudinal Study (22,023 observations from a cohort of 6,627), this study provides four valuable results. One, I find evidence of five distinct social participation trajectories; the majority of which demonstrate social disengagement as individuals age. Two, these decreases were primarily attributable to declines in meeting friends and group exercise. Three, the activities most likely to predict being a part of more-desirable trajectories were cultural event attendance, voluntary group membership, and joining charity groups. Lastly, I find that seven different types of high school activities were each associated with greater social activity counts, decades later. In total, these results highlight systematic differences in social participation trajectories; and suggest that age-graded participation changes are highly dependent on the underlying social activities.

INTRODUCTION

Social participation has been linked to favorable health outcomes across cultures and communities (Sirven and Debrand 2008); as well as increased happiness (Leung et al. 2013), a sense of belonging (Hyyppä and Mäki 2003) and life satisfaction (Elgar et al. 2011). In part, this is because specific activities that make up social participation (e.g., group sports, meeting friends) provide a number of direct and indirect benefits for both physiological and psychological wellbeing (Bath and Deeg 2005). Unfortunately, some scholars suggest that social participation has been declining for decades (Putnam 2000), resulting in increasing rates of social isolation (McPherson et al. 2006)—both of which may be more pronounced and problematic as individuals age (Cornwell et al. 2008).

The few studies that have examined social participation changes generally share three limitations; which this manuscript attempts to address. One, prior research has primarily focused on changes across decades or generations (Clark 2015; Rotolo and Wilson 2004), giving much less attention to social participation fluctuations across the life course (i.e., “age effects”) (Schwadel and Stout 2012). Two, the majority these studies utilized cross-sectional data or restricted scope longitudinal studies (e.g., fewer than 10 years). As such, they are limited in their ability to infer about individual-level changes across multiple decades. Three, these studies frequently use rudimentary social participation measures, such as focusing on only one or two particular activities. Because of this, it is difficult to draw conclusions about aggregate social participation changes; or determine how particular social activities contribute to any lifetime fluctuations.

This article studies social participation changes across mid- and later-life (between the ages of 35 and 71). In doing so, it attempts to answer three research questions: (RQ1) Is there evidence of discrete age-graded social participation trajectories?; (RQ2) How do particular activities explain any social participation patterns (as identified in in RQ1)?; and (RQ3) Which individual-level attributes are associated with social participation over multiple decades? By employing a life course approach and studying numerous social activities, these analyses also allow a testing of prominent psychosocial aging theories (e.g., disengagement, activity, and continuity), as it pertains to social participation.

THEORETICAL FRAMEWORK

Social participation—a form of structural social capital—measures what people do and how they are connected (Harpham 2008); and is often operationalized as “involvement with activities that provide an interaction with others in the community” (Levasseur et al. 2010). Prior investigations into whether social participation has declined across generations or decades largely ignores how social participation varies within cohorts as individuals age. This is important, since having a socially engaged populace over the life course provides numerous benefits to both society and individuals, including improved health outcomes, a source of support for others, and opportunities to exhibit shared values across social groups (Waite 2018). In addition, some of these benefits may be critical in mid- and later-life, when social networks potentially shrink. Identifying social participation patterns across the life course can provide evidence for and against the three prominent theories of aging and psychosocial change: disengagement theory, activity theory, and continuity theory. In addition (as detailed in the following paragraphs), broader life course theory (Elder et al. 2003) suggests that prior social participation opportunities and choices influence current and future social participation profiles.

Disengagement theory (Cumming and Henry 1961) suggests that social participation generally declines over the life course—starting in middle age and accelerating throughout older ages. This decline is stimulated by individuals naturally withdrawing from society; partly as a result of increased morbidity, as well as handing down opportunities to future generations (Cornwell and Waite 2009). In addition, family and friends may begin to die or move away; reducing social networks and activity choices. Life course theory informs disengagement theory’s impact on social participation; since activity choices made in younger ages may alter the risk and severity of later-life declines.

Critics of disengagement theory contend that plenty of individuals remain socially engaged throughout older ages and that these adults tend to have greater life satisfaction (Achenbaum and Bengtson 1994). Therefore, there is no reason to expect it is “natural” to withdraw from society during a life stage in which social isolation may be especially deleterious (Mabry and Bengtson 2005). Since disengagement theory focuses on “exiting roles” (e.g., worker, parents), it does not consider how many social activities (e.g. religious attendance) may operate independently of these roles. Further, it does not consider the disparity of physical and psychological requirements needed to take part in various social activities (e.g., intense group sports versus meeting a friend for lunch). Perhaps due to Robert Putnam’s (2000) cynical view of social capital trends, studies examining social participation changes over time have predominantly tested disengagement theory; with mixed results (Clark 2015).

As a counter to disengagement theory, activity theory (Havighurst and Albrecht 1953) proposes that social participation increases as individuals age. Although older adults may face structural challenges, they also have fewer responsibilities; resulting in role substitution and greater free time. For example, childrearing duties tend to decrease during the latter half of middle age; and most individuals leave the workforce by age 65 (Rosnick 2014). Life course theory also interacts with activity theory, since the repression of social activities in mid-life could amplify the desire for greater social activity in older ages. For example, individuals may finally join that hobby group they long desired to, but were previously occupied with work.

One limitation with prior activity theory research is that these studies tend to use cross-sectional data and focus on the positive relationships between social participation and life satisfaction; emphasizing older adult social activity as “ideal” (DeLiema and Bengtson 2017; Lemon et al. 1972). This approach gives scant consideration to how social participation varies over the life course, including whether it meaningfully “increases” in older age, even among active seniors. In addition, since activity theory tends to emphasize replacement social roles (e.g., time spent as a grandparent instead of time spent working), it largely ignores differences between activity typologies (e.g., formal, informal, family) or how particular social activities may substitute for each other as individuals age. Further, activity theory research has not, largely, incorporated cultural habitus as an important influence on social participation. That is, social activity in older ages—including the addition of “new” activities—is likely shaped by life chances and life choices from earlier ages; including adolescence and childhood (Nash 1990). Surprisingly, activity theory, particularly as it pertains to social participation, has motivated little recent empirical research, even though it is frequently mentioned as an integral component of “successful aging” (DeLiema and Bengtson 2017).

The continuity theory of aging (Atchley 1989), most closely attached to broader life course theory, asserts that social participation is consistent across time and principally influenced by earlier life stages. We may expect, for instance, that older adults who participate in group sports are likely to be the same individuals who were active in group sports during younger ages. In one example, a Swedish study found that older-adult hobbies were, generally, preceded by similar hobbies during mid-life (Agahi et al. 2006). This theory also suggests that even major life course disruptions and transitions—including those considered influential with respect to disengagement and activity theories—may have little effect on long-term social participation patterns; due to consistent predisposition and motivation at the individual level (Atchley 1989).

Unfortunately, as disengagement theory suggests, the ability to take part in many types of social activities will likely decline at some point; due to physical declines or practical constraints. In addition, individuals who generally demonstrate social continuity throughout the life course will still go through considerable adaption and goal revision in order to do so (Atchley 1999; Menne et al. 2002). For example, religious middle-aged individuals who attend a few services a month may, in older ages, be motivated to go more regularly and teach youth classes (Johnston 2013). Unfortunately, many studies of social participation across the life course have not considered how different social activities may be more or less agreeable to consistency across the life course. For example, staying “connected” to friends necessitates only a phone, while other social activities (e.g., travel with friends or taking part in certain sports) may require both good health and discretionary income.

SOCIAL PARTICIPATION TRAJECTORIES

Identifying comprehensive social participation trajectories as people age is one valuable way to test these prominent life course theories (George 2009). In other words, systematic age-graded trajectory patterns could provide evidence of (a) social withdrawal, (b) marked increases in social activity; (c) consistent social participation; or (d) various combinations of these types. While some studies have examined how involvement in particular social activities vary across the life course, very few have estimated trajectories that include numerous activities; and this is important for three reasons. One, prior work suggests that the benefits of social participation may be strongest among those who take part in numerous different activities (Morrow-Howell et al. 2014). Two, membership in some social activities may be low (e.g., fraternal organizations). If so, studies that employ these activities as proxies for social participation may not accurately capture population-level trends. Three, participation changes in one social activity over the life course may be offset by changes in other activities (Ang 2019).

Data limitations and methodological choices have occasionally led to less than ideal operationalizations of “social participation” in prior studies. For example, some researchers use binary indicators, coding social participation as a “yes” if survey respondents (a) are involved in any social group or social activity (e.g., Giordano and Lindstrom (2010)) or (b) have involvement with a particular social activity; such as meeting friends at least one a month (e.g., Verhaeghe and Tampubolon (2012)). A slightly more complex operationalization involves counting the number of activities people have involvement with; and then creating categorical variables related to this count (e.g., “High Participation=involvement with any two social activities” (Poortinga 2006)). This type of treatment, unfortunately, does not allow for an understanding as to how certain activities may propel people into various categories.

Underlying Activity Trajectories

Separate from broader social participation trajectories, there are three reasons to study the trajectories of underlying social activities that contribute to such measures. One, it can help determine whether social participation changes over the life course are primarily attributable to particular social activities. Two, it could provide evidence of activity substitution as individuals age. Three, it could help ascertain whether differences between trajectories are mostly attributable to a few select activities. Although more common than that of aggregate measures, studies which estimate age-graded trajectories of social activities are also limited; and have primarily focused on certain activities. Most often, this focus has been on “voluntary organization” membership (e.g., community organizations, service organizations). For example, Schwadel and Stout (2012) found that such membership tended to peak when individuals were in their 40s, before declining in later years. Greenfield and Moorman (2017) found similar results, with membership peaking around the mid-50s.

A few studies have also examined age-graded changes in meeting friends and religious participation. For the former, Cornwell (2008) found no evidence of significant fluctuations across mid- and later-life; while Schwadel and Stout (2012) found a moderate decline. Research on religious participation is also mixed: while one study found an increase over the life course (Cornwell et al. 2008), Johnston’s (2013) work suggests that any such increases are attributable to already-active members. Unfortunately, prior work on age-graded social activity changes has largely ignored other important forms of social participation, such as community-level events (e.g., festivals, cultural shows) and health-promoting exercise groups. By not examining multiple social activities concurrently, they also miss possible activity substitutions. For example, Ang (2019) found that meetings with friends generally decline over the life course; but may be offset by compensating activities.

Individual-Level Risk Factors

Although there has been little work identifying social participation trajectories across the life course, prior research provides some clues as to which individual-level characteristics may predict greater social participation. For instance, educational attainment is positively associated with most types of social connectedness (Helliwell and Putnam 2007); as well as more-active social profiles in later life (Morrow-Howell et al. 2014). While marriage may promote social participation through “shared social activities” (Kim and Waite 2014); death of a spouse may encourage widows to seek greater connectedness with others (Utz et al. 2002). Research also suggests that women are involved with a greater number of social activities; possibly due to richer social networks and greater internal motivation to be connected to others (Einolf 2011). Associations between having children and being socially connected, however, are less clear. On one hand, being a parent may limit the frequency and type of social activity, due to caregiving responsibilities and a focus on the child’s social participation. Conversely, becoming a parent may evoke greater social connectedness with others (Nomaguchi and Milkie 2003).

As informed by broader life course theory, childhood circumstances and activity exposure may shape a lifetime of social participation behavior. For instance, growing up in a privileged household could allow children the opportunity to take part in numerous activities; including those with a significant entry cost. Parents in these households may also encourage greater activity exposure throughout childhood, amplifying these resource advantages (Luthar et al. 2006). Outside of the family, adolescent extracurricular involvement may also predict a lifetime of social activity. In one of the only studies to look for such an association, Greenfield and Moorman (2017) found a positive relationship between high school-sponsored activities and social participation in later life. Unfortunately, that study did not place any emphasis on the type of school group, and was limited by its focus on one later-life activity (voluntary organizations).

DATA

Data are from the Wisconsin Longitudinal Study (WLS), a long-term survey of a random sample of 10,317 men and women who graduated from Wisconsin high schools in 1957; and were born around 1939 (Herd et al. 2014). After the original wave (W1) was collected a few months before respondents graduated high school, follow-up surveys were completed in 1975, 1992, 2003, 2011; when the modal age of respondents was 35 (W2), 53 (W3), 64 (W4), and 71 (W5), respectively. The WLS used in-person interviews for W1, mail surveys for W2, and a combination of in-person, mail, and phone surveys for W3-W5. The response rate (for either phone or mail surveys) during these four subsequent waves was 90%, 87%, 86% and 74%, respectively (Wisconsin Longitudinal Study (WLS) 1957–2019).

The WLS—unlike many similar surveys—includes detailed information on numerous social activities; and this study’s baseline is W2, when these questions were introduced. Although the WLS began asking some of these questions during W2, other questions were not implemented until W4. To account for this, the WLS, during W4, asked respondents to retrospectively report participation in these activities at W2 and W3. In addition, W2 respondents were omitted from the sample if their high school extracurricular activity information were unavailable (8.56% of surveys). Therefore, the baseline analytical sample are W2 respondents that had complete high school activity information and responded to the W4 survey (n=5,978). Similarly, the W3 sample (n=5,285) are those who completed W3 and W4 surveys; while the W4 and W5 samples (n=5,834 and 4,930; respectively) were those that completed both the mail and phone surveys during that year. In total, 6,627 individuals contributed 22,027 observations ($\stackrel{\_}{x}$=3.32 waves per person). Consequences of sample adjustments and survey attrition are considered in Sensitivity Analyses. Table 1 presents descriptive statistics at baseline.

Table 1.

Descriptive Statistics at Baseline, 1975 Wisconsin Longitudinal Survey (n=5,978)

	Mean (SD) / Percent	No Social Participation (%)	Low/Moderate Social Participation (%)	High/Very High Social Participation (%)
Age	35.46 (0.60)
Male	45.6%
Bachelor’s Degree	26.3%
Employed	71.3%
Income	$20,622 ($16,547)
Marital Status:
Married	89.2%
Divorced	0.8%
Widowed	4.0%
Never Married	6.0%
# of Children	2.42 (1.27)
Parental Education: ≥High School	37.5%
Social Participation, count (0–8)	4.11 (1.27)
Comprised of (range 0–1, each):
Meeting Friends	0.82 (0.31)	7.2	21.3	71.5
Religious Services	0.79 (0.34)	10.7	19.5	69.8
Talking on the Phone	0.71 (0.32)	7.8	43.1	49.1
Light Group Exercise	0.60 (0.39)	22.9	34.2	42.9
Attending Arts	0.44 (0.33)	28.7	54.8	16.5
Heavy Group Exercise	0.39 (0.40)	44.3	32.3	23.4
Voluntary Associations	0.28 (0.42)	67.1	10.8	22.1
Charity Groups	0.08 (0.25)	90.0	3.9	6.1
High School Activities, count (0–10)	2.71 (2.09)
Comprised of (binary: 0–1):
Sports	0.47 (0.50)
Music	0.39 (0.49)
Publications	0.28 (0.45)
Performance	0.27 (0.44)
Pep	0.25 (0.44)
Occupational	0.25 (0.43)
School Assistant	0.24 (0.42)
Student Government	0.19 (0.39)
Hobby Club	0.19 (0.39)
Service Club	0.18 (0.39)

MEASURES

This study employs eight social activities: (1) meeting friends, (2) talking on the phone with friends, (3) attending religious services, (4) joining any of four types of voluntary associations (hobby, community, neighborhood, civic), (5) light physical activity with others, (6) vigorous physical activity with others, (7) attending arts-related or cultural activities, and (8) joining charity/welfare organizations. These time-varying measures were chosen because they (a) represent all activities asked consistently across WLS’s social participation modules, (b) comprise all WLS questions about social interaction with (non-family) others in the community, and (c) include the four social activities (#1–4, above) most commonly employed in social participation studies.

Responses about involvement in these eight activities were coded into three ordinal categories, representing no participation (0), low/moderate participation (0.50), and high/very high participation (1.0) in that activity—similar to other social participation studies (Lam and Bolano 2018); including those using this dataset (Greenfield and Moorman 2017; Vogelsang 2016). For items which asked for quantitative responses (frequency), the activity was coded into categories as suggested by the WLS codebooks; and then adjusted to ensure consistency across waves. For example, the three categories for item 1 (meeting friends) represent individuals meeting friends “0”, “1–2”, and “3 or more” times during the past four weeks. For item 3 (religious attendance), these categories represent “no attendance”, “once a year to once a month”, and “more than once a month”. See Table 1 and Supplemental Table S1 for additional details on these categories. For the social participation questions, 88% of the WLS respondents had zero or one missing items and 8% left two or three of these items blank. All missing participation responses were coded as “0”; and this assumption is discussed in the Sensitivity Analyses.

For the LCA analysis (RQ1), age-graded social participation was measured by a count of all eight available activities (range: 0–8)¹, which had an approximately normal distribution. Respondents, on average, reported 4.11 activities at age 35, decreasing to 2.99 activities by age 71; with a sample mean of 3.57 activities. The RQ3 analyses include a number of time-varying demographic characteristics that are likely associated with social participation—educational attainment, work status, marital status, income, number of children, and age. Because the entire sample graduated from high school, educational attainment is dichotomized as to whether or not the respondent obtained a bachelor’s degree (26.3% at baseline). Work status is coded as “1” if the individual was “currently employed”; while marital status includes four categories: “currently married”, “divorced”, “widowed”, and “never married”. Income is recoded into quartiles; similar to other WLS studies (Greenfield and Moorman 2019); and survey wave is used as a proxy for modal respondent age during that wave. Race information is not included in the public dataset for privacy concerns (Herd et al. 2014); the implications of which are discussed in Limitations.

The RQ3 analyses also includes three other variables—gender, parental socioeconomic status (SES), and high school extracurricular activities. The majority of respondents’ parents did not graduate high school, so parental SES is coded as a “1” if the “head of household” did (37.5%). High school extracurricular activities were determined using yearbook data coded by WLS researchers. For these activities, more than one hundred activities were placed into ten primary categories; as suggested by WLS documentation: (1) sports (varsity, club or intramural); (2) music (e.g., band, chorus); (3) publications (e.g., yearbook); (4) performance (e.g., drama); (5) pep (e.g., cheerleading, pep club); (6) occupational (e.g., future farmers of America); (7) school assistant (e.g., monitors, committees); (8) student government; (9) hobby clubs; and (10) service clubs. Since the degree of involvement in these activities is not available, any participation is coded as a “1”. For baseline respondents, the mean number of high school activities was 2.71; with sports (47%) and music (39%) groups being the most common.

ANALYTIC STRATEGY

To investigate RQ1, RQ2, and RQ3, three separate analytic strategies are utilized. For RQ1 (social participation trajectories), latent class analysis (LCA) is employed; which attempts to identify a small number of unobservable and homogenous subgroups (latent classes) using observed variables (Bartholomew et al. 2008). For this study, the observed variables are the age-graded individual-level social participation counts. Unlike similar research using LCA to identify “activity profiles” (Lam and Bolano 2018), this approach allows modeling of intraindividual social participation change across time; while also accounting for between-person differences (George 2009). See Supplemental File Note S2 for a comparison to Lam and Balano’s approach.

To identify these latent trajectories, models that specify 1,2,3…k trajectory classes were estimated in an iterative fashion. As in virtually all studies employing LCA, the ultimate quantity of latent classes is determined using various formal and informal criteria—model fit statistics, theoretical concerns, class size (estimated population proportion), and class distinctiveness (e.g., Halpern-Manners et al. (2015), Lam and Bolano (2018)). After the “final” number of latent classes are determined, then (a) for each latent class, the marginal predicated social participation means were calculated at each wave; and (b) all observations were assigned to one of the latent classes, based upon marginal predicted posterior probabilities.

For RQ2 (trajectories of particular social activities), two analyses were performed. One, the sample’s mean participation value for each activity was plotted at each wave. This allows an evaluation as to whether each activity demonstrates evidence of patterns described in the “activity”, “disengagement”, or “continuity” theories. Two, if latent class patterns were identified in RQ1, they are evaluated for systematic activity differences. This could help ascertain, for instance, if the primary difference between “active” and “disengagement” trajectories is that individuals in the active classes are much more likely to participate in one particular activity.

To test RQ3 (correlates of social participation), a three-model sequence of multilevel ordinary least squares (OLS) regression models were created (using maximum likelihood estimation), with the social activity count as the dependent variable. These models—nesting observations (Level 1) within respondents (Level 2)—provide a number of key benefits, including (a) their ability to estimate subject-specific coefficients, after accounting for unobserved heterogeneity (the random effect); and (b) allowing all participants to contribute to the estimates, even in the case of missing observations (Rabe-Hesketh and Skrondal 2012). The random intercept is assumed to be independent across individuals, but constant among multiple observations of the same individual. The first model in the sequence (M1) is a baseline model, including only the variables for survey wave (age). The second model (M2) includes demographic covariates that are commonly posited to be associated with social participation—gender, educational attainment, work status, marital status, income, number of children, and survey wave (age). In model 3 (M3), high school extracurricular activities are introduced, which tests whether adolescent social activity can help predict social participation in mid- or later-life. M3 also includes parental SES, since high school involvement and prior activity exposure is likely shaped by a family’s relative social status. All statistical analyses are performed using Stata 15.1 (StataCorp. 2017).

RESULTS

For RQ1 (social participation trajectories), up to 9 classes (i.e., 1, 2, 3,…9) were created and evaluated before ultimately identifying 5 latent profiles. Figure 1 presents a graphical representation of these five trajectories, plotting the marginal social participation means for individuals in these classes, at ages 35, 53, 64, and 71. Three of these trajectories (Classes 4, 2, and 1) provide evidence of social participation disengagement (“High to Low”, “Moderate to Low”, and “Low to Very Low”, respectively); and were estimated to encompass 64.2% of the sample. Also notable, the estimated activity decline for these three trajectories were remarkably similar (49%, 36%, and 42%, respectively). The remaining two classes (Classes 3 and 5) provided evidence of social participation continuity (“Very High” and “High”). These two classes were estimated to represent about one-third of the sample (35.8%), and their predicted mean activity count remained above 4.0—even in older ages.

FIGURE 1.

Marginal Predicted Social Participation Activity Count Mean; by Latent Class and Age, Five Class Solution, Wisconsin Longitudinal Study (n=22,027)

For comparison, adding a sixth latent class did not introduce a theoretically or substantively distinct trajectory (i.e., creating a similar trajectory to “Disengagement: High to Low”, but with a slightly higher participation count in older ages). More importantly, it offered a negligible improvement to model fit statistics, and resulted in two of the six classes having a small population size (<8% each). The five-class solution, on the other hand, offered significantly better model fits than the four- and three-class solutions.

For RQ2 (social activity trajectories), the plotted means for each of the eight activities can be found in Figure 2. Similar to RQ1, these results generally supported disengagement theory; with the exception of charity groups (the least popular activity). That said, the severity of decline varied widely among activities. For three of the eight activities (heavy group exercise, light group exercise, and meeting friends) there was a 73%, 52%, and 22%, decrease, respectively; which accounted for the majority (68%) of all social participation decline. Conversely, declines in talking on the phone with friends, attending the arts, and joining voluntary associations were more modest (12%, 9%, and 5%, respectively). Religious attendance was an anomaly in that it was the only activity to increase between age 64 and 71. Yet, this later-life increase did not compensate for the sharp decline between ages 35 and 64.

FIGURE 2.

Mean Social Participation Scores for Eight Activities between ages 35 and 71, Wisconsin Longitudinal Survey (n=22,027)

As a part of RQ2, activity differences between the “continuity” and “disengagement” classes (as identified in RQ1) were examined. This analysis, in Table 2, reveals stark contrasts for three activities—attending arts, voluntary association membership and charity group membership. The mean activity differences for these three activities (0.30, 0.24, and 0.17, respectively) represented 42.5% of the overall variation (1.67 activities) between the continuity and disengagement groups. Further, the sum of the means for these three activities was 1.27 for the continuity classes, which was more than double that of the disengagement classes (0.56).

Table 2.

Social Activity Means by Predicted Latent Class Membership. “Continuity” vs “Disengagement”. Wisconsin Longitudinal Study. (N=22,027).

	“Continuity”		“Disengagement”
	(Classes 3 and 5)		(Classes 1,2,4)
Percentage of Sample	35.8%		64.2%
	Mean	(SD)	Mean	(SD)	Difference	%
Mean Social Participation, count (0–8)	4.64	(1.07)	2.97	(1.32)	1.67*	56.2%
Comprised of:
Meeting Friends	0.85	(0.27)	0.63	(0.39)	0.22*	34.9%
Religious Services	0.81	(0.34)	0.61	(0.43)	0.20*	32.8%
Talking on Phone	0.77	(0.29)	0.62	(0.43)	0.15*	24.2%
Light Group Exercise	0.59	(0.37)	0.36	(0.38)	0.23*	63.9%
Attending Arts	0.61	(0.34)	0.31	(0.34)	0.30*	96.8%
Heavy Group Exercise	0.35	(0.37)	0.19	(0.31)	0.16*	84.2%
Voluntary Associations	0.42	(0.44)	0.18	(0.35)	0.24*	133.33%
Charity Groups	0.24	(0.37)	0.07	(0.22)	0.17*	242.9%
Social Participation, age 35 (0–8)	4.77	(1.08)	3.76	(1.23)	1.01*	21.2%
Social Participation, age 71 (0–8)	4.35	(1.06)	2.22	(1.14)	2.13*	49.0%
Decrease, age 35 to 71, %	(8.80%)		(41.0%)
n	7,883		14,144

^*Statistically significant difference at ≤.05

For RQ3, Table 3 presents results from the regression models, conditional upon random effects. The baseline model (M1) provides further evidence of social disengagement, with those at age 64 and 71 estimated to be involved with 1.01 and 1.16 fewer activities, respectively, when compared to 35-year-olds. In M2, males were associated with 0.38 fewer activities, while having two or more children was associated with greater social participation counts (when compared to childless adults). Although the relationship between SES (both education and income) and social participation matched expectations; marital status had surprisingly little effect, with only widowhood associated with more participation. In M3, greater parental education was associated with greater social participation; similar to the time-varying SES measures in M2. With respect to the high school extracurricular activities introduced in M3, seven (of ten) activities were each associated with increased social participation in mid- and later-life (ß=0.06–0.15). Of these, the three activities with the largest coefficients were pep groups (0.15), service groups (0.14), and high school sports (0.10). Conversely, there were no associations between social participation and involvement in high school (a) music clubs, (b) hobby groups, or (c) occupational groups. There was strong evidence against multicollinearity concerns for these extracurricular activities, with post-estimation variance inflation scores ranging from 1.04 to 1.33.

Table 3.

Results from Hierarchical Linear Regressions, Estimating Social Participation, conditional upon random effects. Wisconsin Longitudinal Study. (N=22,027).

	M1		M2		M3
	ß	(SE)	ß	(SE)	ß	(SE)
Age: (Reference=35)
53	−0.15***	(0.02)	−0.15***	(0.02)	−0.15***	(0.02)
64	−1.01***	(0.02)	−1.06***	(0.02)	−1.06***	(0.02)
71	−1.16***	(0.02)	−1.24***	(0.02)	−1.24***	(0.02)
Education: ≥Bachelor’s Degree			0.56***	(0.03)	0.44***	(0.03)
Income: (Ref: Lowest Quartile)
Quartile 2			0.09***	(0.02)	0.08***	(0.02)
Quartile 3			0.14***	(0.02)	0.14***	(0.02)
Quartile 4 (Highest Quartile)			0.20***	(0.02)	0.18***	(0.02)
Male			−0.38***	(0.03)	−0.31***	(0.03)
Marital Status (Ref=Married)
Divorced			−0.00	(0.04)	−0.00	(0.04)
Widowed			0.12***	(0.03)	0.12***	(0.03)
Never Married			0.05	(0.07)	0.07	(0.07)
# of Children: (Ref.=0)
1			0.09	(0.06)	0.09	(0.06)
2			0.20***	(0.05)	0.19***	(0.05)
3 or more			0.27***	(0.05)	0.26***	(0.05)
Currently Employed			−0.10***	(0.02)	−0.10***	(0.02)
Parental Edu: ≥ High School Grad.					0.09**	(0.03)
High School Activity Involvement:
Sports					0.10***	(0.03)
Music					0.01	(0.03)
Performance					0.08**	(0.03)
Pep					0.15***	(0.03)
School Assistant					0.06*	(0.03)
Student Government					0.08*	(0.04)
Publications					0.09**	(0.03)
Hobby					0.02	(0.04)
Occupational					−0.05	(0.03)
Service					0.14***	(0.03)
Intercept	4.10	(0.02)	3.87	(0.05)	3.66	(0.06)
SD of Random Intercept	1.02	(0.01)	0.96	(0.01)	0.95	(0.01)
AIC	70,331		69,748		69,622
Log likelihood	−35,159		−34,856		−34,782

Note:

^*p≤.05;

^***p≤.01;

^***p≤.001; SD=Standard Deviation; SE=Standard Error, AIC=Akaike Information Criterion

SENSITIVITY ANALYSES

Numerous sensitivity analyses were conducted; the first group of which ((A)-(D), below) consider non-response, attrition, and other sample adjustments. Similar to a prior analysis of WLS response patterns (Hauser 2005), being male and not having a college education were found to be predicative of attrition (A). Since these individuals reported, on average, lower social participation counts, the “Disengagement: Low to Very Low” latent class would have likely captured a greater percentage of the sample had no one left the survey. In addition, two trajectories presented in Figure 2 (attending arts/cultural events, joining voluntary associations) would likely have had slightly lower mean values across all ages. Next, excluding (W2 and W3) respondents who did not respond to W4 did not appear to materially change results (B) with one exception: the relationship between social participation and having two or more children would likely have increased; since this association was primarily driven by W2 and W3 respondents.

Those who were dropped from the sample due to yearbook coding issues (C) were found not to be systematically different from those who remained in the sample, with respect to all individual-level characteristics. Also, had all social participation questions been answered by all respondents (D), then a few activities would likely have had marginally higher scores (approximately 0.02–0.04 points). Concerns over attrition and nonresponse were also mitigated by the following: (a) the WLS is recognized for its relatively high survey response rates (Herd et al. 2014), and (b) this group of high school graduates have been largely resilient; with only 15.6% of the W1 graduates dying by W5 (54 years later).

Four other sensitivity tests were conducted. One, the implications of using LCA for RQ1 was considered, since other methods (such as finite mixture models) have been shown to occasionally identify disparate trajectories and group memberships (Warren et al. 2015). To compare this paper’s results to another popular method, group based trajectory models (GBTMs) were used to estimate a set of latent classes, using the traj Stata plugin (Jones and Nagin 2013). Results from this analysis identified groups that essentially mirrored those presented in this manuscript; with one exception: the GBTM analysis preferred a six-class solution. Under this solution, the shape, trajectories, and estimated membership percentages for four of these classes (1, 3, 4, and 5) were virtually identical. The two remaining GBTM classes were, effectively, an equal split of (Class 2: Disengagement: Moderate to Low); with half showing a steep (approximately 50%) decline and half showing a more moderate (approximately 20%) decline.

Two, an estimated set of Poisson regression models (for RQ3) provided essentially the same results as those displayed in the manuscript. Three, there was a consideration of how social participation depth (versus breadth) may have influenced the results of RQ1 and RQ2. In other words, is social participation disengagement mostly the result of taking part in fewer activities; but becoming more engaged in the remaining activities? A subsequent analysis found that a decline in participation depth generally mirrors the disengagement trends found in RQ2. One exception to this was for attending arts/cultural activities; in which there was a slight increase, as individuals age, of those reporting high/very high involvement (16.5% at age 35; increasing to 22.8% at age 71).

Lastly, health status was considered as a possible confounder for (RQ3) results, since the ability to be socially active may be limited for individuals suffering from various health issues. Unfortunately, health questions were not asked during W2, and only limited health information was gathered during W3. That said, an additional set of regression models was estimated for W3-W5, which included three health measures: (1) self-rated health² (SRH), (2) a count of ever being diagnosed with (up to four) serious health conditions (cancer—other than skin cancer, diabetes, heart attack/disease, stroke) ($\stackrel{\_}{x}$=0.37), and (3) a count of (up to fifteen) chronic health conditions ($\stackrel{\_}{x}$=1.70), such as liver disease, osteoporosis, and severe back trouble. Results are displayed in Supplemental Table S3. Introducing health measures in this analysis (a) explained the additional social activity associated with being widowed, and (b) made the association between high school government involvement and social participation significant only at the (p<0.10) level. In addition, reporting poor/very poor SRH (2.1% of the sample) was found to be associated with 0.7 fewer social activities. A separate analysis (not presented) also found that, for W4-W5, each additional disability measure ($\stackrel{\_}{x}$=1.88) was associated with participating in −0.06 fewer social activities. There was no evidence that the exclusion of these health variables in the (RQ3) analysis had any other impact on the results or conclusions presented in this manuscript.

DISCUSSION

Most archetypes of “successful aging” include an active and engaged social life—whether continuing to participate in the same activities, or beginning new pursuits subsequent to retirement (DeLiema and Bengtson 2017). Using a rich dataset that follows individuals over multiple decades, this manuscript attempted to answer three questions (RQ1-RQ3) with respect to age-graded social participation. RQ1, which focused on aggregate trajectories, found that approximately two-thirds of the population experienced social disengagement between ages 35 and 71; with substantial activity declines of approximately 41% during this time. The other one-third of the sample, generally, demonstrated social participation continuity; with little or no decline across 35 years. Unfortunately, there was little evidence of significant social participation increases after age 50; which highlights some disadvantages of using the activity theory framework. That is, labeling older adults as “active” in prior studies may be more about between-person differences; and not a reflection of intra-individual changes over time.

Although some degree of social disengagement between age 35 and 71 may be expected, the steepest declines in this study, surprisingly, occurred prior to older ages—between 53 and 64. In other words, perhaps those labeled as “active” in older ages are primarily those who did not display disengagement prior to age 65. Future research would benefit by focusing on this decade, especially since there are multiple reasons to expect an increase in social activity precisely during this time (e.g., little caregiving responsibilities, possible decline in time spent at work, rapid health decline not yet expected). Although possible reasons explaining why individuals may be less socially active after age 50 (e.g., a greater focus on family ties) have been discussed elsewhere (Marcum 2013), there is a notable dearth of qualitative studies to help understand why some healthy adults choose to spend more time alone than they may need to.

By examining the underlying trajectories of particular social activities across the life course (RQ2), this paper makes numerous contributions. Perhaps most importantly, it highlights a limitation with using the (disengagement-activity-continuity) theoretical frameworks—fundamental differences among social activities may belie aggregate participation trends. For example, this paper’s results demonstrate that participation declines between ages 35 and 71 are primarily driven by group exercise and meeting friends. Identifying these “risk activities” provides some guidance for individuals, family and health care providers looking to prevent social disengagement. Exercise groups, when compared to other social activities, may be the most vital for wellbeing, since they include both a social component and an explicit health promotion component. Despite the well-known benefits of exercise for those in mid- and later-life, there remains obvious systematic and individual barriers. With respect to meeting friends, the declines found in this study were comparable to prior research (Ang 2019; Schwadel and Stout 2012); and future work should begin to test reasons for these changes—such as migration, morbidity or psychosocial explanations.

This study also highlights the intersection of social activity trajectories and aggregate social participation trajectories. In particular, it was some of the least common social activities (arts/cultural events, voluntary organizations and charitable organizations) that (a) demonstrated activity continuity across multiple decades; while also (b) providing the starkest differences between those in the continuity and disengagement trajectory classes. In addition, these three activities, unlike some others, explicitly link individuals to the (non-family and non-friend) broader community. This may be one reason why voluntary organization membership, despite being relatively unpopular, remains one of the most studied social activities (Greenfield and Moorman 2017; Putnam 2000). Since prior work has found that attending cultural events is associated with better health (Wilkinson et al. 2007), these findings also point to a possible long-term benefit for individuals becoming engaged with the arts early in life.

RQ3 tested how various individual-level characteristics were associated with social participation; and perhaps the most notable finding was that seven high school activities predicted increased social activity decades later. The magnitude of these relationships was not inconsequential, as the combined impact of being involved in three of these activities (sports, pep, and service groups) was estimated to be approximately equivalent to that of obtaining a bachelor’s degree. Although there is scant previous empirical work to compare this to, some of these associations are unsurprising. That is, adolescents who join sports teams or the drama club in high school (when compared to those who do not) would be more likely to join exercise groups and participate in cultural events, respectively, in mid- and later-life. In addition, three of these activities (i.e., school assistant, student government, service groups) centered volunteering or contributing to the “community”; which may cultivate a dedication to service across adulthood. Indeed, the increase in formalized service learning over the past decade has, in part, been motivated by a belief that such programs have a positive lifetime impact on social responsibility and leadership (Richards et al. 2013). Among activities that did not have relationships with social participation later in life (music groups, hobby groups, and occupational groups), the most surprising may be music groups, since joining a band or chorus would, ostensibly, encourage cultural event attendance across the life course. A subsequent analysis revealed that a positive relationship between joining music groups and later-life social participation was eliminated by the inclusion of gender in the models; with females twice as likely to have been involved with these activities (when compared to males).

Linkages between high school involvement and later-life social participation must be contextualized by acknowledging the evolving nature of extracurricular activities, including how family background shapes the ability and motivation to take part in these activities. For example, public education in the 1960s and 1970s, as part of its mission, was concerned with developing the “whole person”, including school activities that supplemented—but were separate from—the school’s curriculum (Holland and Andre 1987). Since then, schools have increasingly emphasized measurable activities that directly tie to the school’s academic outcomes or allow students to “enjoy life” (Kleiber and Powell 2005). This, along with other social changes, help explain why many popular activities among WLS respondents, such as “pep clubs” (separate from cheerleading), “service clubs” (including religious service clubs) and “occupational clubs” are less common in recent decades.

Student extracurricular involvement, while likely beneficial, is shaped by cultural factors and resource disparities that exist at both the school- and family-levels (Weininger et al. 2015). For example, strategic choices due to college application pressure may result in affluent youth taking part in more school activities (Luthar et al. 2006). Conversely, students from lower-income households may be more prone to work part-time jobs after school (Mortimer and Finch 1996). Although excluding parental SES from the regression models (result not presented) did not change any of the relationships between high school groups and later-life social participation, a subsequent analysis revealed that involvement in nine of these activities was positively associated with parental SES. In particular, parental education had the strongest relationships with being a member of performance groups, student government, hobby groups, and service groups. The only exception was occupational clubs, which were primarily focused on farming and homemaking, and were more popular among students with lower parental education. Similar, for WLS respondents, education was positively associated with six activities across mid- and later-life. Of these, being a college graduate was most predictive of participation in the arts, charity groups, and voluntary associations—the same three activities most likely to separate those in the “continuity” trajectories from those in the “disengagement” trajectories. Recent decades have seen the resource gap among schools widening (Baker 2018), along with increasing income inequality and shifting gender/educational norms (e.g., only 26% of WLS respondents obtained a bachelor’s degree and only 38% had a parent who graduated high school). Because of this, there are reasons to believe that associations between socioeconomic status and the nature/intensity of social participation across the life course will become even more pronounced in subsequent cohorts.

LIMITATIONS & CONCLUSION

Although the WLS has some of the richest data on social participation across the life course, its use comes with four important caveats. One, the WLS is only representative of white high-school graduates (Herd et al. 2014). As such, it is limited by its inability to (a) make claims about those with less than a high school education and (b) explore likely racial and ethnic differences in social participation (Wiertz 2016). That said, the relatively homogenous sample of the WLS can help temper the influence of unobserved variables associated with race and education that could otherwise bias the estimates (Herd 2010). In addition, despite increasing racial and ethnic diversity in the United States among more recent cohorts, the vast majority (approximately 80%) of older adults in the U.S. identified as non-Hispanic White in 2010; when this study’s respondents were 71.

Next, only four WLS surveys were administered between 1975 and 2011; averaging one every twelve years. Because of this, it is impossible to ascertain whether social participation fluctuations occurred between waves; and the number of trajectories identified in this study may have differed had there been more data points. Three, since some social participation data for W2 and W3 were retrospectively reported during W4, these results and trajectories may be susceptible to recall bias. Four, although the WLS asks detailed questions about numerous social activities, it is not exhaustive. The addition or substitution of other activities may alter the trajectories reported in this manuscript.

Since the average age at W4 was 64, this study (in a strict sense) only employed one wave of data (W5) that includes “older adults” (aged 65-plus). As such, it is unknown how the results presented in this manuscript may change as this cohort continues to age and confronts greater morbidity risks. In addition, certain challenges and time constraints may become more evident in older ages; such as raising grandchildren (Meyer 2014) and spousal caregiving (Aneshensel et al. 1995). While a supplemental analysis revealed that being a primary caretaker for grandchildren (14% of the WLS sample) had no association with social participation, understanding the possible impact of family commitments on non-family social participation remains an open issue. Importantly, the most recent WLS wave (2011) occurred prior to the exponential increase in older-adult smartphone adoption; and this change likely impacts older-adult social participation in unknown ways. Lastly, the inconsistent collection of health data across survey waves limited the ability to identify complex relationships between health status, social participation, and other individual-level characteristics (such as socioeconomic status and employment); although this concern was somewhat ameliorated by the sensitivity tests detailed earlier in this manuscript.

The present study provides some evidence that—at least for those born around the start of World War II—social participation generally declines as individuals age. As younger cohorts enter older ages, there is increased need to understand how the type of social activities has changed across generations (Rotolo and Wilson 2004). For example, while there has been a marked membership decline in some voluntary organizations (e.g., Lions, Kiwanis), there may be substitutes with family-focused activities (e.g., child sports leagues, group playdates). Given the nature of social participation, it is challenging to suggest ways in which policy makers and organizations can either promote it during midlife in order to thwart its decline in later life. One possibility is placing a renewed emphasis on social participation during adolescence and even younger ages. For example, youth sports participation among low-income families has declined over the past decade; while slightly increasing among high-income families (Project Play 2018). Understanding how social inequality shapes how individuals form their lifelong social habits and community bonds is a critical yet understudied component of cumulative (dis)advantage.