Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 May 28.
Published in final edited form as: Soc Psychiatry Psychiatr Epidemiol. 2026 May 26;61(8):1413–1423. doi: 10.1007/s00127-026-03114-3

Stability and Change in Social Network Size Over Four Decades: An Age-Period-Cohort Analysis

David W Sosnowski 1, Mahum Khizar 1, Yulan Ding 1, Paul T Costa Jr 2, Grace Tsai 3, Brion S Maher 1, Adam P Spira 1,4,5, William W Eaton 1
PMCID: PMC13215212  NIHMSID: NIHMS2175285  PMID: 42189185

Abstract

Purpose:

Interest in social networks has increased in recent decades amid concerns around social isolation and loneliness, despite increased access to social connections via electronics. Prior studies identified predictors and outcomes linked to network size, but few studies have examined age, period, and cohort effects simultaneously.

Methods:

Using four decades of data, we assessed age, period, and cohort effects, as well as sociodemographic and personality factors, on family and friend network size. In 1981, 3,481 participants (Mage = 47.8 years; 62% White; 38% female) were sampled from East Baltimore, with follow-up periods beginning in 1993, 2004, and 2021. Mixed effects models tested fixed effects of age, sociodemographic, and personality characteristics, along with random effects of period and cohort, on network size.

Results:

Results demonstrated that males, unmarried persons, and persons higher on neuroticism had smaller family networks; Non-White individuals and those higher on conscientiousness had smaller friend networks. Higher income, education, and extraversion predicted larger family and friend networks. No age or cohort effects were detected, though a period effect indicated larger friend networks during baseline data collection compared to other periods.

Conclusion:

Findings suggest family and friend networks are shaped by demographic and personality traits but remain largely stable across time and generation.

Keywords: Age-Period-Cohort Study, Social Networks, Aging, Personality

Introduction

Interest in changes in social networks and social support has increased over the past several decades due, at least in part, to concern that the quality of social life in the United States has declined [1,2]. This concern has prompted the creation of a Commission on Social Connection by the World Health Organization [3] and Minister of Loneliness by the British parliament [4]. The purported decline has occurred over the last half-century, and changes in the last decade may be associated with increased availability of social contact via electronic means that do not involve face-to-face or vocal interaction. Today, more than ever, social networks are expanding, and the support provided by these networks is becoming more varied and diverse.

Social networks can be defined as the social structures comprised of individuals and/or organizations and the ties between them. At the individual level, social networks provide social support, which refers to companionship and practical, informational, and esteem support. Georg Simmel authored early structural theories in sociology, emphasizing the dynamics of triads and the “web of group affiliations” [5]. Simmel’s theory proposes that the more opportunities an individual has to interact with others, the more social support will be available, which, in turn, benefits health, health-related behavior, and general well-being [6].

The structure (e.g., size, composition, frequency of contact) and function (e.g., support) of social networks have long been of interest to researchers due to their bidirectional relationship with psychological health [7]. Herein, we focus on social network size, in part because of the strong association between social isolation and mortality, which is comparable to that of traditional clinical risk factors (e.g., high blood pressure, obesity) [8]. Moreover, social network size has been linked to psychological health and psychiatric conditions among adults. For instance, prior research of individuals aged 60+ found that smaller friend network size was associated with greater depressive symptomatology [9]. Similarly, in a sample of Spanish adults aged 50+, smaller social networks were associated with increased probability of depression, and this association was larger among individuals who reported higher levels of loneliness [10]. A more recent review recent of research on social connectedness identified eight studies focused on network size, with the majority (n = 5) demonstrating an association between smaller social network size and increased probability of depressive symptoms or disorder at follow-up [11]. Research also supports associations between social network size and other psychiatric outcomes, including schizophrenia and anxiety symptoms [12,13], PTSD, and suicidal behaviors [14]. Altogether, extant evidence highlights the role social networks in mental health throughout the lifespan.

Considering the psychological health impacts of social networks, particularly as individuals age, two important questions arise: (1) What are the sources of stability and change in one’s social networks?; (2) How do differences in age, period (i.e., calendar date when an individual is asked about their current network size), and birth cohort influence stability and/or change in network size? Are the personal networks and supports we create and integrate influenced more by external measures such as socioeconomic status (SES) or internal factors, such as personality? And what happens to the structure of our social networks when agents of change, such as major life events (e.g., marriage, divorce, retirement), affect us, and do we correct for these disruptions?

An Age-Period-Cohort analysis method [21] can be leveraged to address these questions, as it helps disentangle the effects of individuals’ age, the time period in which the data were collected, and their birth cohort on social network size and stability. Age effects reflect changes that occur in social network size as individuals grow older, influenced by both self-perception and how others perceive them at different life stages. Changes in social networks occur throughout life, and the relative stability of networks has been attributed to “convoys of attachment” [15]. The convoy model assumes that there are basic norms of social relationships over time that help individuals maintain their well-being and cope with the stresses of life. These relationships vary in closeness, quality, function, and structure [16]. Carstensen’s socio-emotional selectivity theory argues that people tend to restrict their networks as they get older, feeling that with less time left to live, they should concentrate their affective energies on those most important to them [17]. Further, theories of the past suggest that people become more socially isolated as they age and thus have smaller networks than younger people [25]. However, Cornwell et al. argue that even though there are a lot of changes in the social networks of older individuals, the size of the network remains constant, i.e., older adults maintain homeostasis by forming new relationships [20]. Period effects capture the impact of historical or societal events (e.g., pandemics, economic recessions, wars) that affect all individuals regardless of age and birth cohort. For example, a potentially important source of change in social networks in recent years is the COVID-19 epidemic, which may have led to less social interaction. Cohort effects represent generational differences in social networks among people born in the same time period, shaped by shared experiences, cultural norms, and social contexts. There have been proposals that individuals born later than the baby-boom birth cohort have reduced the size of their networks, such as in the work by Robert Putnam in Bowling Alone: The Collapse and Revival of American Community [1].

A sociological perspective on network size emphasizes contextual elements. For instance, the sociological lens examines the structure and function of convoys, particularly as they link to various group characteristics such as gender, SES, and ethnicity/race [21–24]. Below, we describe individual and sociodemographic factors that may directly influence social networks over time.

Sex

Researchers have found that women report slightly higher levels of perceived support than men. Women were shown to have more multifaceted social networks than men and more friends, but about the same number of family members [25,26].

Race

The convoy model has also been applied to investigations of people of different ethnic and racial groups. For example, research suggests that African and Caribbean Americans tend to have smaller networks, more family-dominated convoys, and more frequent contact with their convoy members than White Americans [21,27]. Interestingly, these differences disappear with age; that is, older Black and White individuals report similar levels of contact frequency with their relatives [21].

Socioeconomic Status

Researchers in the United States and China have found that network size, degree of network participation [28,29], and levels of perceived support [30,31] increase with SES and employment [29,31,32]. Further, a cohort study in Amsterdam found that, compared to an earlier birth cohort (1908–1937), the more educated later birth cohort (1938–1947) had larger and more diverse networks at retirement, reflecting a cohort effect linked to expanded educational opportunities [26]. However, by age 71.5 years, the network sizes converged, indicating age effects – network size tended to shrink with age regardless of the birth cohort. The SES and employment-related differences can be attributed to increased social ties formed through school, college, and job, as a meta-analysis shows that network size grows significantly during adolescence, young adulthood, and while entering the workforce [33].

Personality

A key source of stability in networks and support is an individual’s personality structure. We hypothesize that disruptions in networks or supports are continuously repaired by the individual so that the network size, frequency of contact, and support quality meet a level of comfort associated with stable dispositions embodied in the personality structure. Under this hypothesis, personality traits should be strongly associated with the size, frequency, and supportive aspects of network contacts. There is some evidence suggesting that extraversion and agreeableness positively influence network size, with extraversion having a stronger effect on friend networks, particularly [34]. However, although these personality traits are linked to the initial size of one’s network, they do not appear to influence its expansion or contraction over time.

Life Events

An important source of change in social networks is the wide range of social disruptions that occur throughout life, including many so-called “life events,” such as marriage and divorce, the death of a relative or friend, changing residence, or any physical change or decline. Retirement is a major life event for older adults. As discussed earlier, older adults tend to experience a decline in their social networks over time. This has led to considerable interest in the influence of retirement on social network size. Evidence demonstrates that individuals lose network members after retirement, especially the more peripheral ones (Kauppi et al., 2021), but tend to strengthen relationships with family and close friends.

Other life events like marriage, parenthood, partner loss, divorce, and relocation contribute considerably to network changes. Married persons report higher levels of perceived support and extensive social ties compared to single persons [29,30,32,33,36]. Meanwhile, loss of a partner and divorce have been linked to a reduction in network size, with divorce affecting family networks the most [33].

The effects of the COVID-19 pandemic are a recent concern for public health researchers. This time-bound event introduced widespread disruptions in daily life, including social isolation, increased psychological stress, and shifts in socioeconomic conditions. Thus, it is essential to consider the potential period effects of COVID on social network size and composition. Observed differences during this period may reflect the broader societal impacts that could have influenced people’s social ties independent of their age or birth cohort. However, the extent of this influence may vary across cohorts and age groups due to differences in cultural norms, awareness, adaptability, and access to emotional support.

The present study investigated key sources of stability and changes in personal networks and social supports, using a 40-year longitudinal panel survey with measures at baseline, 12, 23, and 40 years later: the Baltimore Epidemiologic Catchment Area (ECA) Follow-up [37]. We conducted an Age-Period-Cohort analysis to investigate how age, historical period, and birth cohort shape social network size and structure, reflecting broader shifts in social demands, cultural norms, emotional needs, and life course experiences. We hypothesized that network size will vary across age groups and that both period and cohort effects will influence social network outcomes. Notably, data for wave 5 of the ECA study were collected during the COVID-19 pandemic, which may have impacted social network structure across different cohorts.

Method

Participants and Procedure

In 1981, 175,211 household residents of Eastern Baltimore were sampled probabilistically for participation in the Baltimore site of the Epidemiologic Catchment Area Program [38,39]. Persons over the age of 65 were purposively oversampled by designating all household members of that age group for interview, as well as whoever in the household was designated from a random selection. Four thousand two hundred thirty-eight persons were designated for the sample, and 3,481 completed the interview. One year later, 2,768 of these were interviewed, but since the time period is so short compared to the overall length of the follow-up, this analysis did not use data from Wave 2 in 1982. From 1993 through June of 1996, 1,920 of those interviewed in 1981 were interviewed again [40] (Since most of the interviews occurred in 1993, this year is used as shorthand to designate this wave of the cohort study). In 2004 and the first half of 2005, 1,071 of those interviewed in 1993 were interviewed again (likewise, the shorthand of “2004” is used for this wave). Most recently, in 2019 through 2021 (herein referred to as “2021” for this wave of the cohort study), data collection occurred among 512 of those interviewed in 2004. The current analysis draws from available data on participants from Waves 1 and 3–5. The study was approved by the Institutional Review Board of the Johns Hopkins Bloomberg School of Public Health and was conducted in adherence to standards in the Declaration of Helsinki.

Measures

Network Size

At each Wave, size of family networks and friend networks was measured independently by responses to two questions: “How many family members and relatives do you usually keep in touch with by telephone or by visiting?; and “How many friends, including neighbors, people you work with, and anyone else you consider a friend who do not live with you, do you usually keep in touch with by telephone or by visiting?” Response options were: 0 (none), 1 (2–3), 3 (4–5), 6 (6–10), 5 (11+), and 8 (don’t know). “Don’t know” responses were recoded as missing. These questions were asked identically at baseline and all follow-up interviews.

Sociodemographic and Personality Factors

Participants self-reported their age (in years), sex (male or female), race, marital and employment statuses, years of education, and household income during each wave of data collection. Based on response distributions, race, and marital categories were collapsed into two groups (i.e., Black versus non-Black; married versus not married). Employment status at each wave was categorized as 0 (unemployed), 1 (employed), or 2 (retired). The response values for income were large ordinal categories, which changed slightly from baseline to later waves as the population’s average income increased. The baseline values included ten $1000 increments from $1,000-$10,000 and then eight larger increments to a maximum of “over $50,000.” In 1993, 2004, and 2019, there were 26 categories, with the top category being “$200,000 and over” Last, dimensions of normal personality were assessed in 2004 with the NEO Personality Inventory, which assesses five major dimensions [41]. Scores for each dimension were standardized prior to analysis.

Analysis

All analyses were conducted using R version 4.4.0 [42]. First, we described the sample characteristics and calculated polychoric correlations between network size indicators at each wave to assess stability and change in friend and family network size over time. We then conducted an Age-Period-Cohort analysis to evaluate the association of sociodemographic characteristics with family and friend network size, including effects for age, period, and cohort. Specifically, we used a Hierarchical Age-Period-Cohort-Cross-Classified Random Effect Model (HAPC-CCREM) [43]. This mixed effects model treats age as a fixed effect and cohort and period as random effects. In this framework, the HAPC-CCREM is similar to a typical mixed effects model, in which random effects often represent contextual effects that influence the outcome. In this case, cohort represents the contextual effect of experiencing the same social and historical context at the same age within the same birth cohort, and period effect represents the contextual effect of experiencing the same social and historical context at a certain time period that affects all age groups. This model also allowed us to treat network size as a continuous variable since there were low counts for persons reporting “none” or “11” contacts in their network over time.

Three models were run for friend and family network size, respectively (i.e., 6 total models). Model A contained fixed effects for age and sociodemographic factors and random effects for period and cohort. Model B contained the same fixed effects, but only period as a random effect. Model C was similar to Model B but only modeled cohort as a random effect. Likelihood ratio tests were used to compare Model A to Models B and C (separately). A significant p-value (i.e., < 0.05) for the χ2 indicates that the model with random effects for cohort and period suggests that this model is a better fit to the data than the model with only cohort or period as a random effect. Once the appropriate model was selected, we interpreted those results and plotted the beta of the random effect(s) on the outcome, conditional on the fixed effects in the model. Significance was set at a threshold of p ≤ 0.05.

Results

Table 1 presents descriptive statistics for the ECA cohort across Waves 1 and 3–5 and Figure 1 shows reasons for attrition between Waves 1 and 5 of the ECA study. The sample was comprised of majority of White males, and approximately half of the sample reported being married across the follow-up periods. As expected, employment status changed over time, with the highest employment rates being reported during Waves 3 and 4, when participants were, on average, 55 and 58 years old, respectively. Retirement rates increased over time as well, with half of the available sample reporting being retired at Wave 5.

Table 1.

Distribution of Demographic and Network Variables in the Baltimore Epidemiologic Catchment Area Follow-up Cohort

1981 1993 2004 2021
n = 3,481 n = 1,920 n = 1,071 n = 512
Sex, n (% female) 1,322 (38%) 707 (37%) 397 (37%) 185 (36%)
Race, n (% White)* 2,193 (62%) 1,214 (63%) 662 (62%) 310 (61%)
Married, n (% yes)† 1,474 (42%) 865 (49%) 581 (56%) 259 (52%)
Employment Status†
 Unemployed, n (%) 1,940 (56%) 298 (34%) 288 (28%) 78 (16%)
 Employed, n (%) 1,463 (42%) 453 (51%) 604 (58%) 171 (34%)
 Retired, n (%) 78 (2%) 134 (15%) 146 (14%) 247 (50%)
Age, Mean (SD) 47.8 (19.9) 55.0 (16.9) 58.9 (12.9) 67.9 (8.1)
Years of Education, Median (IQR)† 11 (4) 14 (4) 14 (4) 15 (3)
Size of Relatives Network, n (%)†
 None 130 (4%) 48 (3%) 22 (2%) 6 (1%)
 One 242 (7%) 74 (4%) 43 (4%) 22 (4%)
 2–3 773 (23%) 337 (20%) 194 (18%) 90 (18%)
 4–5 669 (20%) 392 (23%) 206 (20%) 101 (20%)
 6–10 885 (27%) 523 (30%) 356 (34%) 177 (35%)
 11+ 623 (19%) 350 (20%) 231 (22%) 115 (22%)
Size of Friends Network
 None 253 (8%) 117 (7%) 79 (7%) 41 (8%)
 One 228 (7%) 149 (9%) 81 (8%) 40 (8%)
 2–3 738 (22%) 516 (30%) 278 (26%) 137 (27%)
 4–5 685 (21%) 389 (22%) 233 (22%) 120 (23%)
 6–10 769 (23%) 333 (19%) 238 (23%) 106 (21%)
 11+ 644 (19%) 220 (13%) 144 (14%) 67 (13%)
Open in a new tab

Note.

*

Race is based on self-reported data at Wave 1 to maximize available data at later follow-up waves.

†

Values and percentages reflect exclusion of missing data; complete data were available for age, sex, and race across all waves.

Figure 1.

Figure 1

Open in a new tab

Overview of Study Attrition from Wave 1 to Wave 5 of the Baltimore Epidemiologic Catchment Area Study

Regarding familial network size, the majority (96%) reported at least one relative in their network across all time points, and familial network sizes remained relatively stable over time. Similarly, at least 92% of participants reported at least one friend in their social network across all time points, and those with the largest friend network size at baseline (i.e., 19% with 11+ friends in 1981) appeared to demonstrate the largest decrease in friend network size at later ages (to 13% in 2021). Table 2 presents polychoric correlations for the network size variables over time, indicating moderate correlations across waves, suggesting stability in these network sizes over time.

Table 2.

Polychoric Correlations of Network Size Variable Across Waves of Data Collection

Relatives: Wave 1 Relatives: Wave 3 Relatives: Wave 4 Relatives: Wave 5 Friends: Wave 1 Friends: Wave 3 Friends: Wave 4
1. Relatives: Wave 1 -
2. Relatives: Wave 3 0.34 -
3. Relatives: Wave 4 0.28 0.49 -
4. Relatives: Wave 5 0.26 0.40 0.41 -
5. Friends: Wave 1 0.44 0.23 0.18 0.17 -
6. Friends: Wave 3 0.27 0.36 0.26 0.23 0.45 -
7. Friends: Wave 4 0.29 0.24 0.37 0.25 0.38 0.47 -
8. Friends: Wave 5 0.23 0.18 0.20 0.37 0.38 0.37 0.45
Open in a new tab

Note. Bolded values highlight correlations among the respective network size variable at each wave with the same variable at the subsequent wave of data collection.

APC Analyses

Familial Network Size

Next, we used the HAPC-CCRE model to examine age, period, cohort (and sociodemographic) effects on familial network size over time. Model A, which included random effects for both period and cohort, demonstrated a singular fit, which indicates a random effect is at or near zero (i.e., does not influence the outcome). This was the case for the random effect for period (i.e., wave of data collection). So, we retested this model, only including a random effect for cohort (i.e., Model C). Results for these analyses are presented in Table 3. Notably, there was no fixed effect of age (B = 0.00, p = 0.96), though we did detect that males (B = −0.15, p = 0.05) and persons who scored higher on neuroticism (B = −0.09, p = 0.04) reported smaller familial networks over time. Conversely, married individuals (B = 0.36, p < 0.001), persons with a higher annual income (B = 0.03, p = 0.002) and education (B = 0.04, p = 0.01), and persons higher on extraversion (B = 0.16, p < 0.001) and agreeableness (B = 0.07, p = 0.05) reported larger familial network sizes over time. As expected, results from the likelihood ratio test revealed no significant improvement in fit by including both cohort and period as random effects in the model (χ2(1) = 0.00, p = 1.00). Lastly, to estimate the effect of cohort, we plotted the conditional effect of cohort on familial network size. As shown in Figure 2, there is some variation in the effect by cohort, but the confidence intervals are overlapping, suggesting no significant effect. In summary, these results suggest that various sociodemographic factors influence familial network size over time, though there are no age, period, or cohort effects.

Table 3.

Results for Fixed Effects from Hierarchical APC-Cross-Classified Random Effects Models Testing Age, Period, and Cohort Effects on Familial and Friend Network Sizes Over Time

Family Network Size Friend Network Size
N = 1,258 N = 1,256
B (SE) p-value B (SE) p-value
Age 0.00 (0.00) 0.96 0.00 (0.00) 0.55
Sex (ref = female) −0.15 (0.08) 0.05 0.04 (0.08) 0.64
Race (ref = White) −0.03 (0.07) 0.64 −0.53 (0.08) < 0.001
Marital Status (ref = not married) 0.36 (0.08) < 0.001 −0.11 (0.08) 0.17
Employment Status (ref = unemployed) - - - -
 Employed −0.11 (0.09) 0.23 −0.05 (0.10) 0.64
 Retired −0.16 (0.17) 0.35 0.19 (0.19) 0.30
Income 0.03 (0.01) 0.002 0.03 (0.01) 0.001
Education 0.04 (0.02) 0.01 0.05 (0.02) 0.01
Personality Domain - - - -
 Openness −0.06 (0.04) 0.15 0.01 (0.05) 0.76
 Conscientiousness −0.06 (0.05) 0.23 −0.12 (0.05) 0.01
 Extraversion 0.16 (0.05) < 0.001 0.29 (0.05) < 0.001
 Agreeableness 0.07 (0.04) 0.05 0.04 (0.04) 0.33
 Neuroticism −0.09 (0.05) 0.04 −0.06 (0.05) 0.25
Open in a new tab

Note. Bolded values reflect significant associations at p ≤ 0.05. The best-fitting model for family network size only includes a random effect of cohort (i.e., Model C), whereas the best-fitting model for friend network size only includes a random effect of period (i.e., Model A). Sample sizes refer to persons with complete-case data across all waves of data collection.

Figure 2.

Figure 2

Open in a new tab

Non-Significant Effect of Cohort on Familial Network Size, Conditional on Fixed Effects in the Model

Friend Network Size

Next, we examined friend network size over time. Model A, which included random effects for both period and cohort, was estimated normally. There was no fixed effect of age (B = 0.00, p = 0.55), though we did detect that non-White participants (B = −0.53, p < 0.001) and participants who scored higher on conscientiousness (B = −0.12, p = 0.01) reported smaller friend networks over time. Conversely, persons with a higher annual income (B = 0.03, p = 0.001) and education (B = 0.05, p = 0.01) and persons higher on extraversion (B = 0.29, p < 0.001) reported larger friend network size over time. The likelihood ratio test comparing Model A to Model B revealed that including both period and cohort effects did not improve fit over a model that only included a random effect for period (χ2(1) = 0.02, p = 0.89); however, Model A was estimated to be a better fit to the data compared to Model C, which only included a random effect for cohort (χ2(1) = 9.20, p = 0.002). Thus, we selected Model B (i.e., random effect for period only) as our final model, the results of which are presented in Table 3. Lastly, to estimate the period effect in this model, we plotted the conditional effect of period on friend network size. As can be seen in Figure 3, there is a stronger effect during the baseline data collection period (1981) compared to other data collection periods. In summary, these results suggest that various sociodemographic factors influence friend network size over time, and there is a period effect.

Figure 3.

Figure 3

Open in a new tab

Significant Effect of Period on Friend Network Size, Conditional on Fixed Effects in the Model

Discussion

In the present study, we examined the sources of stability and change in social network size in the Baltimore ECA Follow-Up Study, and whether these vary by age, study period, or birth cohort. On the basis of previous theoretical and empirical work found in the social network and support literature (e.g., the Convoy Model), findings from the present study affirm that social networks are indeed influenced by sociodemographic (e.g., sex) factors, socioeconomic characteristics (e.g., income), and personality traits (e.g., extraversion). We also detected a period effect on friend network size, such that larger networks were reported in 1981 relative to other data collection timepoints. Contrary to our expectations, we did not observe either age or cohort effects on family or friend network size over time. Nonetheless, these results provide key insights into how network structures vary over time, and which aspects may be relevant to improving the quality of social life.[18][44]

We observe no significant associations between age and changes in social network size over the course of the Baltimore ECA study. In other words, both friend and family networks remained stable as individuals aged. This may be explained by the findings of Cornwell et al. (2021), which suggest that, although adults tend to lose some peripheral connections over time, they maintain social “homeostasis” by adapting and forming new relationships. As a result, the sizes of their core social networks remain stable. Similarly, we detected no effects of birth cohort on social network size over time. While Putnam (2000) suggested that social isolation increases among later cohorts, subsequent studies [26,45,46] have consistently reported stable network sizes and contact frequency across cohorts. Suanet et al. noted a brief increase in network size upon retirement in a 1938–1947 cohort, but this did not persist in that study, and overall, network sizes remained comparable to other cohorts. Similarly, Ang et al. and Liu et al. found no meaningful cohort differences in social connectedness. Our findings support this growing body of evidence, indicating that later birth cohorts (i.e., those born in the 1950s-1960s) may not be experiencing a decline in social network size. However, these findings pertain to network structure rather than relationship quality. Future research should examine whether perceived support and relationship quality vary across birth cohorts, particularly in the context of shifting communication norms and the influence of digital platforms.

Our study identified a small but statistically significant period effect on friend network size in 1981, with networks notably larger during that year compared to other periods examined. This finding can be contextualized within broader social trends documented by Putnam in Bowling Alone: The Collapse and Revival of American Community. Putnam observed that individuals reaching middle age around the 1980s showed decreasing interest in political and civic engagement. He highlighted the 1980s as a pivotal decade when American society began shifting towards a more individualistic lifestyle, accompanied by an accelerated decline in civic participation, political involvement, and informal socializing. These observations are further supported by Schwadel’s (2012) research, which demonstrates a continuing decline in neighborly contact in the years following the 1980s [47]. The period effect we observe may be reflective of social network structure prior to this decline, where, on average, social networks tended to be larger in older adults (i.e., ages 60+).

Regarding sociodemographic characteristics, we identified a declining trend in the number of relative-based ties for men over time (Table 3). Women, on average, maintained larger relative networks compared to men. Although women’s friend-based networks were smaller than men’s networks, this difference was not significant. These findings are consistent with earlier work by Antonucci et al (1987) and align with the theoretical frameworks offered by both the convoy model and DIRe (differential investment of resources) model [48]. According to the convoy model, gendered social roles lead to women maintaining larger kin networks due to their traditional caregiving responsibilities, which may also limit opportunities to expand friendship ties [19]. The DIRe model similarly posits that women are expected to invest more emotional energy and time into close relationships, particularly with family members [49]. Both models report that while younger women have larger and more diverse networks, these networks may shrink with age and increasingly become centered on close family ties, a pattern observed in our findings.

We also identified significant differences by race, such that individuals identifying as non-White reported smaller friend network sizes over time. These findings are consistent with those of Ajrouch et al. (2001), who found that African Americans had smaller social networks compared to Whites. However, in that study, the racial differences tended to diminish with age. Similarly, Liu et al. (2025) observed that although Black individuals initially had smaller social networks, these networks expanded during their 50s before declining again, ultimately reaching a comparable size to those of White individuals by around age 70. Lastly, we found that married individuals tended to report larger family network sizes. This is expected, as married individuals have additional family ties (i.e., in-laws) and are more likely to have children to expand their available network of familial contacts. Indeed, extant evidence [33] suggests that individuals living with a spouse tend to maintain larger and stronger close ties.

We assessed three factors to examine the association between socioeconomic status (SES) and network size: employment, education, and income. Consistent with prior research (Lei et al., 2015), both income and education showed a significant, positive relationship with the size of family and friend networks. Previous literature [22] indicates that men and women with at least a high school education had more diverse and larger non-kin networks compared to individuals who did not have a high school education. A study conducted in the United States found that individuals who attained four years of college education but came from low-SES backgrounds experienced the greatest gains in social networks, particularly in formal ties [50]. These findings suggest that, despite economic disparities, policies that improve access to education can meaningfully strengthen social connectedness and improve health outcomes among vulnerable populations.

Although sociodemographic factors influence the structure and function of social networks, intrinsic characteristics, such as personality traits, also play a significant role in individuals’ ability to form and maintain relationships. Interestingly, we found multiple aspects of personality that were associated with familial and friend network size. Specifically, individuals higher in extraversion reported larger familial and friend networks, whereas individual higher in agreeableness reported larger familial networks. These findings are not surprising, as extraverted individuals are expected to create more opportunities for contact both within and outside families, [34] and agreeableness can serve to maintain social ties, even during periods of stress or conflict [51,52]. In contrast, conscientiousness was associated with smaller friend networks, potentially reflecting a preference for fewer but more meaningful or structured relationships. Meanwhile, individuals higher in neuroticism showed smaller familial networks over time. This may be partially explained by Iveniuk’s (2018) findings, which showed that those higher in neuroticism often only reach out to friends more frequently to discuss health-related concerns [53]. Additional research by Litwin et al. (2022) reported that neurotic individuals tend to have less frequent contact with close social ties and report lower life satisfaction, which may help explain the observed reduction in familial connections but no change in friend network size in the current study [54]. Although most existing research does not identify a relationship between conscientiousness and network size, this trait has been linked to stronger social ties, higher-quality relationships, and greater life satisfaction [54]. Thus, while conscientious individuals may not maintain large networks, their social connections may be more stable, supportive, and fulfilling.

Limitations

Though this study captures data on social networks across multiple decades and birth cohorts, several limitations should be noted when interpreting our findings. First, although we were able to examine familial and friend network size over several decades, we did not examine the support (or lack thereof) obtained from these networks. Although beyond the scope of the present analysis, examination of how these social networks function (e.g., providing more emotional versus instrumental support) would improve our understanding of how these networks vary over time by age, period, and cohort. Second, there is potential for bias in the self-reports of social network size, effect size estimation due to missing data from sample attrition, and representativeness of the sample, which is limited to older adults (i.e., 50+) originally enrolled from an East Baltimore neighborhood in the State of Maryland. Lastly, there are many other life experiences (e.g., death of a loved one, personal health) that may influence social network size and engagement and were not assessed in the present study. Despite these limitations, this study provides key insights into the social networks of older adults in the United States.

Implications

The implications of our findings highlight researchers’ attempt to address pressing questions about social life in the United States. There is a growing concern about loneliness and social isolation despite our exponential increase in connectedness through the Internet and social media platforms. Results of this longitudinal study indicate that personality factors (e.g., extraversion, neuroticism) and indicators of SES (e.g., income, education) are key factors associated with social network size among older adults. Reasonable next steps would be to understand the mechanisms by which these factors may affect social network size over time. For example, whether persons high on neuroticism have smaller networks due to specific behaviors, and how higher SES may increase network size? Moreover, understanding the functions of familial and friend networks will clarify their role in aging, cognition, and other health-related outcomes.

Our finding that males and non-White individuals are at risk for smaller familial and friend network sizes, respectively, highlights the importance of outreach to ensure social networks are maintained as individuals age and navigate life transitions (e.g., retirement), which may look different for men and minority populations. For example, those who work with older adults (i.e., nurses, physicians, social workers) may inquire about a person’s social networks during annual health assessments. In addition, researchers should explore sociocultural factors that may improve the social relations among aging males and non-White adults in the United States. Lastly, it is notable that there neither age nor cohort predicted social network size. This suggests that the structure of social networks is similar across age and birth cohorts. In summary, this Age-Period-Cohort analysis identified robust stability of social network size over four decades in a population-based sample and identified several personal factors that influence this stability. Exploration of additional network aspects (e.g., frequency of contact, support) may further shed light on the role of social networks in the loneliness epidemic.

Acknowledgements:

This work was supported by National Institute on Aging Grants U01AG052445 and R01AG075996.

References

  • 1.Putnam RD. Bowling alone: The collapse and revival of American community. Simon & Schuster; 2000. [Google Scholar]
  • 2.Chapman A, Verdure AM, Moody J. Analytic Advances in Social Networks and Health in the Twenty First Century. J Health Soc Behav. 2022;63:191–209. 10.1177/00221465221086532 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.World Health Organizaiton (WHO). Social Isolation and Loneliness [Internet]. 2025. [cited 2025 Jul 20]. https://www.who.int/teams/social-determinants-of-health/demographic-change-and-healthy-ageing/social-isolation-and-loneliness. Accessed 20 Jul 2025 [Google Scholar]
  • 4.Pimlott N The ministry of loneliness. Canadian Family Physician. 2018;64:166–7. [PMC free article] [PubMed] [Google Scholar]
  • 5.Simmel G. Conflict and The Web of Group- Affiliations.. First Edition. Wolff Kurt H., Bendix R, editors. The Free Press; 1955. [Google Scholar]
  • 6.Cooper H, Arber S, Fee L, Ginn J. The influence of social support and social capital on health: A review and analysis of British data. London: Health Education Authority; 1999. [Google Scholar]
  • 7.Dunbar R. Structure and function in human and primate social networks: implications for diffusion, network stability and health. The Royal Society. 2020; [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Holtzman RE, Rebok GW, Saczynski JS, Kouzis AC, Doyle KW, Eaton WW. Social Network Characteristics and Cognition in Middle-Aged and Older Adults. The Journals of Gerentology. 2004;59:278–84. [DOI] [PubMed] [Google Scholar]
  • 9.Kuiper JS, Voshaar RO, Zuidema SU, Stolk RP, Zuidersma M, Smidt N. The relationship between social functioning and subjective memory complaints in older persons: a population-based longitudinal cohort study. International Journal of Geriatric Psychiatry. 2016;32:1059–71. [DOI] [PubMed] [Google Scholar]
  • 10.Sörman DE, Rönnlund M, Sundström A, Norberg M, Nilsson LG. Social Network Size and Cognitive Functioning in Middle-Aged Adults: Cross-Sectional and Longitudinal Associations. J Adult Dev. Springer New York LLC; 2017;24:77–88. 10.1007/s10804-016-9248-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Kelly ME, Duff H, Kelly S, McHugh Power JE, Brennan S, Lawlor BA, et al. The impact ofsocial activities, social networks, social support and social relationships on the cognitive functioning of healthy older adults: A systematic review. Syst Rev. BioMed Central Ltd; 2017;6. 10.1186/s13643-017-0632-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Casey ANS, Liu Z, Kochan NA, Sachdev PS, Brodaty H. Cross-Lagged Modeling of Cognition and Social Network Size in the Sydney Memory and Ageing Study. Journals of Gerontology - Series B Psychological Sciences and Social Sciences. Gerontological Society of America; 2021;76:1716–25. 10.1093/geronb/gbaa193 [DOI] [PubMed] [Google Scholar]
  • 13.Perry BL, McConnell WR, Peng S, Roth AR, Coleman M, Manchella M, et al. Social Networks and Cognitive Function: An Evaluation of Social Bridging and Bonding Mechanisms. Gerontologist. 2021;62:865–75. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Yoo I Social Networks as A Predictive Factor in Preserving Cognitive Functioning During Aging: A Systematic Review. Home Health Care Manag Pract. SAGE Publications Inc.; 2022. p. 72–80. 10.1177/10848223211030448 [DOI] [Google Scholar]
  • 15.Green AF, Rebok G, Lyketsos CG. Influence of social network characteristics on cognition and functional status with aging. International Journal of Geriatric Psychiatry. 2008;23:972–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.van Groenou MB, Hoogendijk EO, Tilburg TG. Continued and New Personal Relationships in Later Life: Differential Effects of Health. Journal of Aging and Health. 2012;25. [DOI] [PubMed] [Google Scholar]
  • 17.Kuiper JS, Zuidersma M, Oude Voshaar RC, Zuidema SU, ven den Heuvel ER, Stolk RP, et al. Social relationships and risk of dementia: A systematic review and meta-analysis of longitudinal cohort studies. Ageing research Reviews. 2015;22:39–57. [DOI] [PubMed] [Google Scholar]
  • 18.Röhr S, Löbner M, Gühne U, Heser K, Kleineidam L, Pentzek M, et al. Changes in Social Network Size Are Associated With Cognitive Changes in the Oldest-Old. Front Psychiatry. Frontiers Media S.A; 2020;11. 10.3389/fpsyt.2020.00330 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Stephen G, Muldoon Orla T, Bennett Kate M. Multiple group membership, social network size, allostatic load and well-being: A mediation analysis. J Psychosom Res [Internet]. Elsevier Inc.; 2021;151:110636. 10.1016/j.jpsychores.2021.110636 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Yang Y, Land KC. Age-period-cohort analysis of repeated cross-section surveys: Fixed or random effects? Sociol Methods Res. 2008;36:297–326. 10.1177/0049124106292360 [DOI] [Google Scholar]
  • 21.Antonucci TC, Akiyama H. Convoys of attachment and social relations in children, adolescents, and adults. In: Nestmann F, Hurrelmann K, editors. Social networks and social support in childhood and adolescence. De Gruyter, Walter; 1994. p. 37–52. [Google Scholar]
  • 22.Antonucci TC, Birditt KS, Ajrouch KJ. Social relationships and aging. In: Lerner RM, Easterbrooks MA, Mistry J, Weiner IB, editors. Handbook of psychology: Developmental psychology. 2nd Editio. John Wiley & Sons Inc.; 2013. p. 495–514. [Google Scholar]
  • 23.Carstensen LL, Isaacowitz DM, Charles ST. Taking time seriously: A theory of socioemotional selectivity. American Psychologist [Internet]. American Psychological Association (APA); 1999. [cited 2025 Aug 27];54:165–81. 10.1037//0003-066X.54.3.165, [DOI] [PubMed] [Google Scholar]
  • 24.English T, Carstensen LL. Selective narrowing of social networks across adulthood is associated with improved emotional experience in daily life. Int J Behav Dev. SAGE Publications Ltd; 2014;38:195–202. 10.1177/0165025413515404 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Antonucci TC, Akiyama H. Social networks in adult life and a preliminary examination of the convoy model. Journal of Geronotology. 1987;42:519–27. [DOI] [PubMed] [Google Scholar]
  • 26.Cornwell B, Goldman A, Laumann EO. Homeostasis Revisited: Patterns of Stability and Rebalancing in Older Adults’ Social Lives. Journals of Gerontology - Series B Psychological Sciences and Social Sciences. Gerontological Society of America; 2021;76:778–89. 10.1093/geronb/gbaa026 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Ajrouch KJ, Antonucci TC, Janevic MR. Social Networks Among Blacks and Whites: The Interaction Between Race and Age [Internet]. Journal of Gerontology: SOCIAL SCIENCES Copyright. 2001. https://academic.oup.com/psychsocgerontology/article/56/2/S112/555819 [DOI] [PubMed] [Google Scholar]
  • 28.Ajrouch KJ, Blandon AY, Antonucci TC. Social Networks Among Men and Women: The Effects of Age and Socioeconomic Status. The Journals of Gerontology: Series B [Internet]. 2005;60:S311–7. 10.1093/geronb/60.6.S311 [DOI] [PubMed] [Google Scholar]
  • 29.Stoller E, Pugliesi KL. Size and effectiveness of informal helping networks: a panel study of older people in the community. Journal of Health & Social Behavior. 1991;32:180–91. [PubMed] [Google Scholar]
  • 30.Peek MK, Lin N. Age differences in the effects of network composition on psychological distress. 10.1016/S0277-9536(99)00139-2 [DOI] [PubMed] [Google Scholar]
  • 31.Antonucci TC. Social Support: Theoretical Advances, Recent Findings and Pressing IssuesSocial Support: Theoretical Advances, Recent Findings and Pressing Issues. In: Sarason IG, Sarason BR, editors. Social Support: Theory, Research and Applications. Springer Dordrecht; 1985. p. 21–37. [Google Scholar]
  • 32.Suanet B, Huxhold O. Cohort Difference in Age-Related Trajectories in Network Size in Old Age: Are Networks Expanding? Journals of Gerontology - Series B Psychological Sciences and Social Sciences. Gerontological Society of America; 2020;75:137–47. 10.1093/geronb/gbx166 [DOI] [PubMed] [Google Scholar]
  • 33.Nguyen AW. Variations in Social Network Type Membership among Older African Americans, Caribbean Blacks, and Non-Hispanic Whites. Journals of Gerontology - Series B Psychological Sciences and Social Sciences. Gerontological Society of America; 2017;72:716–26. 10.1093/geronb/gbx016 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Thoits PA. Life stress, social support, and psychological vulnerability: Epidemiological considerations. J Community Psychol. 1982;10:341–62. 10.1002/1520-6629(198210)10:4<341::AID-JCOP2290100406>3.0.CO;2-J [DOI] [PubMed] [Google Scholar]
  • 35.Turner RJ, Marino F. Social support and social structure: a descriptive epidemiology. J Health Soc Behav. 1994;35:193–212. [PubMed] [Google Scholar]
  • 36.Ross CE, Mirowsky J. Explaining the social patterns of depression: control and problem solving--or support and talking? Journal of Health and Social Behavior. 1989;30:206–19. [PubMed] [Google Scholar]
  • 37.Thoits PA. Explaining Distributions of Psychological Vulnerability: Lack of Social Support in the Face of Life Stress. Social Forces. 1984;63:453–81. [Google Scholar]
  • 38.Lei X, Shen Y, Smith JP, Zhou G. Do social networks improve Chinese adults’ subjective well-being? J Econ Ageing. Elsevier; 2015;6:57–67. 10.1016/j.jeoa.2015.07.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Wrzus C, Hänel M, Wagner J, Neyer FJ. Social network changes and life events across the life span: A meta-analysis. Psychol Bull. American Psychological Association Inc; 2013;139:53–80. 10.1037/a0028601 [DOI] [PubMed] [Google Scholar]
  • 40.Rollings J, Micheletta J, Van Laar D, Waller BM. Personality Traits Predict Social Network Size in Older Adults. Pers Soc Psychol Bull. SAGE Publications Inc.; 2023;49:925–38. 10.1177/01461672221078664 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Kauppi M, Virtanen M, Pentti J, Aalto V, Kivimaki M, Vahtera J, et al. Social Network ties before and after retirement: a cohort study. Eur J Ageing. 2021;18:503–12. 10.1007/s10433-021-00604-y [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Maddock J, Gallo F, Wolters FJ, Stafford J, Marseglia A, Dekhtyar S, et al. Social Health and Change in Cognitive Capability among Older Adults: Findings from Four European Longitudinal Studies. Gerontology. S. Karger AG; 2023;69:1330–46. 10.1159/000531969 [DOI] [PubMed] [Google Scholar]
  • 43.Eaton WW, Kalaydjian A, Scharfstein DO, Mezuk B, Ding Y. Prevalence and incidence of depressive disorder: The Baltimore ECA follow-up, 1981–2004. Acta Psychiatr Scand. 2007;116:182–8. 10.1111/j.1600-0447.2007.01017.x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Eaton WW, Kessler LG. Epidemiologic Field Methods in Psychiatry: The NIMH Epidemiologic Catchment Area Program [Internet]. Eaton WW, Kessler LG, editors. Academic Press; 1985. 10.1016/C2009-0-02951-1 [DOI] [Google Scholar]
  • 45.Eaton WW, Regier DA, Locke BZ, Taube CA. The epidemiologic catchment area program of the national institute of mental health. Public Health Reports. 1981;96:319–25. [PMC free article] [PubMed] [Google Scholar]
  • 46.Eaton WW, Anthony JC, Gallo J, Tien A, Romanoski A, Lyketsos C, et al. Natural history of diagnostic interview schedule. Arch Gen Psychiatry. 1997;54:993–9. 10.1001/archpsyc.1997.01830230023003 [DOI] [PubMed] [Google Scholar]
  • 47.Costa PT, McCrae RR. Normal personality assessment in clinical practice: The NEO Personality Inventory. Psychol Assess. 1992;4:5–13. 10.1037/1040-3590.4.1.5 [DOI] [Google Scholar]
  • 48.R Core Team. R: A language and environment for statistical computing [Internet]. Vienna, Austria: R Foundation for Statistical Computing; 2024. https://www.r-project.org/ [Google Scholar]
  • 49.Yang Y, Land KC. A mixed models approach to the age-period-cohort analysis of repeated cross-section surveys, with an application to data on trends in verbal test scores. Sociol Methodol. 2006;36:75–97. 10.1111/j.1467-9531.2006.00175.x [DOI] [Google Scholar]
  • 50.Antonucci TC, Ajrouch KJ, Birditt KS. The convoy model: Explaining social relations from a multidisciplinary perspective. Gerontologist. 2014;54:82–92. 10.1093/geront/gnt118 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51.Ang S Life course social connectedness: Age-cohort trends in social participation. Adv Life Course Res. Elsevier Ltd; 2019;39:13–22. 10.1016/j.alcr.2019.02.002 [DOI] [Google Scholar]
  • 52.Liu J An epidemic of social isolation? Age and cohort trends of social connectedness among older adults, 2004–2018. 2025; 10.1016/j.ssresearch.2025.103212 [DOI] [Google Scholar]
  • 53.Schwadel P, Stout M. Age, period and cohort effects on social capital. Social Forces. 2012;91:233–52. 10.1093/sf/sos062 [DOI] [Google Scholar]
  • 54.Huxhold O, Fiori KL, Windsor T. Rethinking Social Relationships in Adulthood: The Differential Investment of Resources Model. Personality and Social Psychology Review. SAGE Publications Inc.; 2022;26:57–82. 10.1177/10888683211067035 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 55.Huxhold O, Fiori KL, Windsor T. Rethinking Social Relationships in Adulthood: The Differential Investment of Resources Model. Personality and Social Psychology Review. SAGE Publications Inc.; 2022;26:57–82. 10.1177/10888683211067035 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 56.Andersson MA. Higher Education, Bigger Networks? Differences by Family Socioeconomic Background and Network Measures. Socius. SAGE Publications; 2018;4. 10.1177/2378023118797217 [DOI] [Google Scholar]
  • 57.Jensen-Campbell LA, Graziano WG. Agreeableness as a moderator of interpersonal conflict. J Pers. Blackwell Publishing Ltd; 2001;69:323–62. 10.1111/1467-6494.00148 [DOI] [PubMed] [Google Scholar]
  • 58.Rapp C, Ingold K, Freitag M. Personalized networks? How the Big Five personality traits influence the structure of egocentric networks. Soc Sci Res. Academic Press Inc.; 2019;77:148–60. 10.1016/j.ssresearch.2018.09.001 [DOI] [PubMed] [Google Scholar]
  • 59.Iveniuk J Social networks, role-relationships, and personality in older adulthood. Journals of Gerontology - Series B Psychological Sciences and Social Sciences. Gerontological Society of America; 2019;74:815–26. 10.1093/geronb/gbx120 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 60.Litwin H, Levinsky M. Does personality shape the personal social networks of older Europeans? Eur J Ageing. Springer Science and Business Media B.V; 2022;19:717–27. 10.1007/s10433-021-00678-8 [DOI] [PMC free article] [PubMed] [Google Scholar]

RESOURCES