Development of social convoys: Trajectories of convoy structure and composition from childhood through adulthood

Abstract

Personal networks undergo changes in structure and composition throughout the lifespan, adapting to developmental transitions and changing circumstances in a dynamic way. This study examines stability and change in social convoys from childhood to adulthood and variation in trajectories of convoy characteristics by gender and race. Multilevel models for convoy structure and composition characteristics were estimated using three waves of longitudinal data spanning 23 years. The regionally representative sample included 193 children aged 8 to 12 in Wave 1 (1992) who were surveyed again in their 20s (Wave 2) and 30s (Wave 3). The Wave 1 sample was comprised of 52% girls, 32% Black children, and 59% White children with average maternal educational attainment of 13 years. Overall, changes in composition, proximity, and contact frequency were observed at each wave. Between Waves 1 and 2, the changes reflect age-normative trends toward network diversification typical of the transition to adulthood, whereas between Waves 2 and 3, the changes were consistent with those expected as young adults settle into stable roles. We also identified convoy characteristics that differed between men and women and between Black and White individuals, emphasizing the importance of considering personal characteristics to fully understand form and function of social relations. Social convoy trajectories early in the lifespan provide direction for more in-depth examinations of the implications of social ties during these critical life periods.

Changes in close, personal networks occur across the lifespan, especially during periods of transition. Early adulthood is a period of shifting social roles, obligations, and societal expectations, as well as changing interpersonal relationships (Arnett, 2000; Arnett et al., 2020). Theoretical life course perspectives, including the convoy model of social relations and cross-sectional empirical evidence (Antonucci et al., 2010; Kahn & Antonucci, 1980), indicate differences in personal network, or social convoy, characteristics at different points in the lifespan. Limited longitudinal research exists on changes in social convoys from childhood to adulthood or on the personal and situational characteristics that shape these trajectories. In this study, we capitalized on three waves of in-depth social relations data to investigate stability and change in social convoys from childhood to early adulthood by examining intraindividual changes in individual social convoy characteristics, as well as sociodemographic factors associated with these trajectories.

Convoy Model

The convoy model of social relations, which describes personal networks as multidimensional and dynamic, provides an overarching theoretical framework for the present study (Antonucci et al., 2010; Kahn & Antonucci, 1980). Dimensions of social relations include structure, function, and quality (Antonucci et al., 2010), each of which is linked to health and well-being across the lifespan. Convoy structure encompasses objective characteristics of social convoys, including size, composition (e.g., age, gender of, and relationship to network members), proximity to, and contact with network members. Structure provides the basis for function, which often refers to support features (e.g., types and amount of support exchanged) that have implications for well-being. Convoys are shaped by personal (e.g., gender, race) and situational characteristics (e.g., social roles, culture) that remain stable as well as those that change over time, making them dynamic (Antonucci et al., 2011; Antonucci & Akiyama, 1987). As children develop into adolescence and adulthood, changes in situational characteristics like social roles, societal expectations, and demands promote changes in social convoys (Marsden, 2018).

The convoy model has been used successfully with people of all ages (Antonucci, Ajrouch, & Webster, 2019; Manalel & Antonucci, 2020). Although descriptions of social relations at different points across the lifespan are informative, it is also important to acknowledge change over time and with age. Changes in social relations can refer to either the formation or loss of social ties, or to changes in the nature of existing social ties, such as geographic proximity or decreased contact (Feld et al., 2007). An optimally functioning social network changes in ways to meet the demands of developmental periods (Antonucci, Ajrouch, & Webster, 2019).

Early Adulthood Social Relations

Evidence points to changes in social network size, composition, proximity, and contact frequency that occur within the context of emerging and established adulthood. These changes are driven primarily by developmental tasks and transition markers of early adulthood, like obtaining higher education, advanced vocational/skilled training, romantic partnership, family formation, entry into the workforce, and living independently (Arnett, 2000; Mehta et al., 2020; Settersten, 2007). Socioemotional selectivity theory (SST) (Carstensen, 1992, 1995) describes how social motivations of developmental periods drive individuals’ intentions to form, maintain, or adapt different types of social ties. In early adulthood, when unlimited time horizons prioritize future-oriented information acquisition goals, individuals seek numerous social ties with diverse social partners (e.g., romantic partners, friends, new acquaintances) that allow for varied social experiences (Charles & Carstensen, 2010). Changes in convoy composition include the addition of romantic partners, as well as adult friends, co-workers, and children (Wrzus et al., 2013). Studies of social networks across the lifespan consistently report that global, personal, and friendship networks increase in size throughout adolescence and early adulthood, while family networks remain stable. Based on SST and prior empirical evidence, from childhood to early adulthood, we expect to observe increases in network size and greater diversification of network composition (e.g. more friends and other non-kin, less family, more same-aged peers). Across early adulthood, stabilization of network size is expected. Changes in network composition to reflect settling into adult roles, like family formation, might include a relative increase in immediate family and decrease in friends.

The convoy model suggests that the closest relationships usually remain stable across time, forming a core network, while ancillary or role-based ties are more susceptible to changes (Antonucci & Akiyama, 1987; Kahn & Antonucci, 1980). These role-based ties, however, are another indicator of network diversification predicted by SST. Thus, we expect to observe a decrease in the relative number of closest social ties during the transition to adulthood, followed by stabilization across early adulthood.

Studies focusing on social contact (e.g., frequency, mode) are less common than those on personal network size and composition. Consistent with the convoy model, the frequency of visits with family remains stable over time in a large sample of adults, while non-family visits decreased (Sander et al., 2017). In contrast, overall family contact has been found to decrease over time across emerging adulthood (Sneed et al., 2006). Although the presence of certain social ties remains stable, the nature of these ties may change. For example, marriage and parenthood often result in less frequent contact with friends and other non-kin (Kalmijn, 2012). Thus, we expect to observe decreases in overall contact frequency during the transition to and across early adulthood.

Longitudinal research focused on changes in the geographic proximity of network is rare, but cross-sectional studies can be informative. Children’s social networks are geographically proximate, with most network members living in the same household or within an hour’s drive (Manalel & Antonucci, 2020). Younger adults typically have more proximate networks than older adults (Ajrouch et al., 2001; Antonucci & Akiyama, 1987), but this somewhat counterintuitive finding may be due to loss-related changes of older adults’ social networks. Developmental tasks of emerging and established adulthood suggest that network proximity decreases during the transition to adulthood. For example, relocating for college or employment may result in less proximate networks, as the geographic distance from family of origin or childhood friends increases. Thus, we hypothesize that network proximity decreases from childhood across early adulthood.

Personal Characteristics

Variations in social relations across different personal characteristics can provide information about the distinct social resources, or lack thereof, among various groups or identities. Cross-sectional studies indicate gender and race differences in certain convoy characteristics across the lifespan. For example, girls tend to have larger networks than boys and this difference persists into adulthood with women’s networks being larger than men’s (Ajrouch et al., 2005; Antonucci et al., 2004). With regard to gender composition, women are more likely to be included in personal networks overall (Antonucci, Ajrouch, & Webster, 2019; Antonucci & Akiyama, 1987). However, there is a well-established bias toward gender homophily in personal networks, meaning women often include proportionately more women than do men (Dunbar & Spoors, 1995).

Black adults, compared to White, have smaller networks, networks composed of a greater proportion of kin, and more frequent contact with network members (Ajrouch et al., 2001). These differences were more pronounced among young adults than older. Greater kin composition and extended family involvement characterize persons of color, whereas the social networks of White individuals consist of more friends (Levitt et al., 1993; Taylor et al., 2013). This pattern has been found in children’s social networks as well (Manalel & Antonucci, 2020). Similarly, Black Americans report more very close social partners than White Americans across adulthood (Fung et al., 2001). Some studies suggest that Black individuals have more proximate networks that White (Cantor et al., 1995), but others have found no difference (Ajrouch et al., 2001).

Although group differences in social networks have been established across gender and race, it is unclear whether these factors shape trajectories of change in convoy characteristics. For example, greater network size and contact are reported among women across different life stages, suggesting only mean-level differences over time between men and women (Ajrouch et al., 2005; Sander et al., 2017). Similarly, differences in convoy composition (i.e., kinship, close others) by race appear to be consistent across the life course (Ajrouch et al., 2001; Levitt et al., 1993; Taylor et al., 2013). Thus, we do not hypothesize that changes in social convoy characteristics vary by gender or race, but rather that gender and race differences remain stable from childhood across early adulthood.

Present Study

This study aimed to describe changes in social convoys from childhood through early adulthood. Identifying the social resources that are available to individuals and how they change helps determine how they can be leveraged to cope with the stressors that often accompany major life transitions, such as the transition to adulthood. Meaningful descriptions of personal network changes are an essential base from which to understand their implications on convoy function, health, and well-being. Using longitudinal data spanning 23 years, we assess change in social convoy characteristics over time through two research questions:

1. Do characteristics of convoy structure and composition change from childhood to adulthood, specifically, between three time points: middle childhood (i.e., 8-12); emerging adulthood (i.e., 20s); and established adulthood (i.e., 30s)?

2. Do convoy characteristics vary over time by gender or race? Specifically, are there mean-level differences in convoy characteristics over time between men and women and Black and White individuals. Additionally, we examine changes in convoy characteristics by gender or race.

Method

Sample

The Social Relations Study is a three-wave, longitudinal study that began in 1992 (see Antonucci & Akiyama, 1994 for sampling details). Recruitment efforts for child respondents targeted mothers who were participating in the study. For W1, consent was first obtained from mothers to interview their child. Assent was then obtained from children. Later waves were conducted by telephone, so oral consent was obtained. Procedures were approved by the Institutional Review Board at the University of Michigan (HUM00074983 for the Social Relations, Aging and Health Study). The current analytic sample included 193 children aged 8-12 in Wave 1 (W1). The average age of participants at W1, Wave 2 (W2; 2005), and Wave 3 (W3; 2015) was 10, 23, and 33 years, respectively. Over half of the participants were girls/women and about one third identified as Black (Table 1).

Table 1.

Social Relations Study Child Sample, Descriptive Statistics

	Wave 1 (N = 193)	Wave 2 (N = 143)	Wave 3 (N = 109)
Gender	52% girls	57% women	56% women
Race	59% White	65% White	66% White
	32% Black	30% Black	26% Black
Marital status
Married		12%	42%
Living with partner		13%	17%
Divorced		0%	6%
Separated		1%	1%
Never married		74%	33%
Have children		32%	58%
Working full time		48%	73%
Student		23%	8%
Age (years)	10.08 (1.38) [8-12]	23.36 (1.47) [21-26]	33.37 (1.46) [31-36]
Education (years)	4.96 (1.42) [2-8]	13.54 (1.94) [9-17]	14.70 (1.92) [8-17]
Income level		5.41 (3.07) [1-13]	7.17 (4.72) [1-13]
Household size	4.49 (1.45) [2-9]	3.25 (1.53) [1-9]	3.26 (2.16) [1-9]
Maternal education (years)	12.95 (1.94) [7-17]
Mother marital status
Married	69%
Living with partner	3%
Widowed	3%
Divorced	15%
Separated	3%
Never married	8%

Note. Means displayed with standard deviation in parentheses and ranges in brackets. Household size in W1, maternal education, and mothers’ marital status are mother-reported. Income levels: 1 = Less than $5,000; 2 = $5,000-$9,999; 3 = $10,000-$14,999; 4 = $15,000-$19,999; 5 = $20,000-$24,999; 6 = $25,000-$29,999; 7 = $30,000-$39,999; 8 = $40,000-$59,999; 9 = $60,000-$79,999;10 = $80,000-$99,999; 11 = $100,000-$149999; 12 = $150,000-$199,999; 13 = $200,000+

Attrition.

In W2, 149 (73.2%) of the W1 child sample was re-interviewed. In W3, 114 (57.1%) of the W1 child sample were interviewed again. Of the child sub-sample, 93 respondents (46%) participated in all waves, 56 respondents (28%) participated only in W1 and W2, 21 respondents (10%) participated only in W1 and W3, and 32 respondents (16%) participated only in W1. An attrition analysis comparing participants who participated in W2 and W3 with those who did not showed that respondents who participated in W2 were more likely to be women, χ(1)=4.11, p<.05, and White, χ(1)=8.13, p<.01. Those who participated in W3 were more likely to be White, χ(1)=5.06, p<.05. They did not differ in terms of age, mother’s educational attainment, mother’s marital status, or W1 convoy characteristics.

Measures

In-person interviews were predominantly obtained at W1 and by telephone for W2 and W3. Participants completed a network assessment using the hierarchical mapping technique, which allowed participants to use subjective evaluations of interpersonal closeness, rather than making assumptions based on role relationships or other characteristics (Antonucci, 1986). Respondents enumerated individuals they considered close and important, placing them in one of three concentric circles representing varying levels of emotional closeness. The closest enumerated individuals (i.e., network members) were placed in the inner circle. Respondents provided detailed information about up to the first 10 network members, including gender, age, and role relationship (e.g., mother, friend, etc.). Table 2 provides descriptive statistics of convoy characteristics and correlations across waves.

Table 2.

Means, standard deviations, and correlations of social network characteristics in Waves 1-3

	M	SD	W1	W2	Observed Range
Network Size
W1	8.66	4.99			1 to 20
W2	9.76	4.51	.10		0 to 20
W3	9.95	4.95	.08	.34***	1 to 20
% Closest Others
W1	58.18	27.76			0 to 100
W2	48.43	22.89	.09		0 to 100
W3	54.04	21.54	−.04	.27*	14.3 to 100
Contact Frequency
W1	4.34	0.58			2.0 to 5
W2	4.28	0.44	.19*		2.8 to 5
W3	4.09	0.49	.10	.31**	2.8 to 5
% Proximate
W1	87.93	20.31			0 to 100
W2	81.85	27.11	.11		0 to 100
W3	73.49	31.20	.13	.53***	0 to 100
% Immediate Family
W1	47.65	26.64			0 to 100
W2	49.90	21.73	.28**		0 to 100
W3	58.47	24.43	.19*	.55***	0 to 100
% Extended Family
W1	36.13	28.59			0 to 100
W2	17.87	18.23	.22**		0 to 66.7
W3	12.51	16.59	.13	.31***	0 to 62.5
% Friends
W1	18.60	25.53			0 to 100
W2	22.37	22.07	.12		0 to 83.3
W3	22.31	21.57	.14	.49**	0 to 77.8
% Agemates
W1	21.68	24.40			0 to 100
W2	29.06	20.57	.15		0 to 85.7
W3	23.46	18.97	.15	.32***	0 to 100
% Girls/Women
W1	53.09	20.70			0 to 100
W2	49.62	18.05	.30***		0 to 100
W3	53.05	18.43	.25**	.48***	20 to 100

Notes. M and SD are used to represent mean and standard deviation, respectively.

^*p < .05.

^**p < .01.

^***p < .001. Less than 4% of sample included more than 20 network members, so total network size was capped at 20. Immediate family in W1 consisted of parents and siblings, including stepfamily. In W2 and W3, this categorization was expanded to include spouses/partners and children. Friends in W2 and W3 include romantic partners who were not indicated as spouse/partner by participants. Extended family in W1 included grandparents, great-grandparents, cousins, and aunts/uncles. In W2 and W3 this categorization was expanded to include in-law.

Convoy structure.

Network size represents participants’ total number of network members. Contact frequency measured how often participants were in touch with each network member, from daily (5) to irregularly (1), and was averaged across network members. The following measures reflect percentages of total network (up to 10) to account for differences in network size (Antonucci & Akiyama, 1987). Closest others represent the percentage of the network placed in the inner circle. Proximity indicated the percentage of network members living within an hour’s drive of the participant.

Convoy composition.

Role relationships were categorized into immediate family, extended family, and friends and measured as percentages of total network (Manalel & Antonucci, 2020). Age-mates included network members who were aged within one year (older or younger) of the respondent in childhood and within two years in adulthood. Gender composition was assessed by the percentage of network members who were girls/women.

Sociodemographics.

Information about participant gender (binary), race, age, mothers’ educational attainment, and household size was collected from participant self-reports at all waves and mother-reports in W1. In W2 and W3, participants reported on marital status, number of children, work status, income level, and household size.

Analysis Strategy

Multilevel models were estimated for each convoy characteristics using SAS PROC MIXED (SAS Institute Inc, 2015). All available data were analyzed through maximum likelihood estimation and by including a random intercept with an unstructured covariance matrix (Allison, 2012). These models consisted of two levels, including upper level respondent characteristics (e.g., respondent gender, race, maternal education) and lower level time-varying measures (e.g., wave, participant age, and education).

To estimate change across two time lags, wave was a categorical variable with W2 serving as the reference group. W1 indicated the difference between W2 and W1, and W3 indicated the difference between W2 and W3. This allowed us to estimate the timing of and account for nonlinear patterns in convoy characteristics. Main effects of gender and race were used to determine mean-level differences in convoy characteristics. Interactions with wave were entered to estimate whether changes in convoy characteristics varied by gender or race. Significant interactions were determined by a significant effect (α=.05) and improvement in model fit between the main effects and interaction models, as determined by the −2 log likelihood (Singer & Willett, 2003). Respondent age, education, and maternal education were entered as covariates in all models. Continuous covariates were centered on the sample mean, and categorical covariates were effect coded. The study analysis code is available on request from the corresponding author. This study was not preregistered.

Results

Trajectories of Change

Results for the main effects of wave, indicating change over time, are displayed in Table 3. Between childhood (W1) and early adulthood (W2), participants reported significant decreases in percentage of networks comprised of closest others, extended family, and proximity, and a significant increase in percentage age-mates. During this period (W1 to W2), network size, contact frequency, percentage immediate family, friends, and girls/women in the networks remained stable. Across adulthood (W2 to W3), participants reported a significant increase in immediate family, and significant decreases in contact frequency, proximity, and age-mates. During this period (W2 to W3), network size, percentages of closest others, extended family, friends, and girls/women remained stable. Across both time lags, observed changes in convoy characteristics were in the expected directions. Sensitivity analyses assessed the robustness of these findings. Results were unchanged when controlling for mothers’ marital status (married/partnered versus others). We also expanded age-mates to include network members +/− 3 years of participants in W2 and W3 and observed the same pattern of findings.

Table 3.

Results of Models Estimating Change in Convoy Characteristics from Childhood to Adulthood

	Network Size Coef. (SE)	Closest Others Coef. (SE)	Contact Frequency Coef. (SE)	Proximate Coef. (SE)	Immediate Family Coef. (SE)	Extended Family Coef. (SE)	Friends Coef. (SE)	Age-mates Coef. (SE)	Women Coef. (SE)
Intercept	9.48*** (0.40)	49.34*** (2.06)	4.29*** (0.04)	82.08*** (2.05)	50.72*** (2.00)	17.42*** (1.82)	21.47*** (1.88)	27.92*** (1.73)	49.97*** (1.51)
Change over time (ref = Wave 2)
Wave 1	−0.97+ (0.49)	9.34*** (2.64)	0.06 (0.05)	5.73* (2.45)	−2.59 (2.28)	14.53*** (2.15)	−3.52 (2.30)	−7.04*** (2.09)	3.36+ (1.78)
Wave 3	0.06 (0.57)	4.99 (3.05)	−0.17** (0.06)	−8.49** (2.85)	9.60*** (2.67)	−4.08 (2.51)	−0.91 (2.67)	−7.66** (2.44)	3.35 (2.08)
Women	0.45+ (0.24)	1.53 (1.19)	0.03 (0.03)	1.62 (1.32)	0.64 (1.34)	0.52 (1.18)	−2.95* (1.17)	−2.11+ (1.09)	6.43*** (0.97)
White	0.68** (0.25)	−3.17* (1.25)	−0.06* (0.03)	0.88 (1.37)	−2.28 (1.39)	−2.64* (1.23)	4.51*** (1.22)	4.59*** (1.13)	−3.18** (1.02)
Age	−0.04 (0.18)	0.45 (0.90)	−0.003 (0.02)	−0.15 (0.97)	3.43*** (0.72)	−1.20 (0.87)	−1.66+ (0.87)	−2.03* (0.81)	−0.33 (0.72)
Education	0.21 (0.15)	−1.34+ (0.76)	0.003 (0.02)	−2.11** (0.75)	−2.16** (0.72)	0.05 (0.67)	2.07** (0.69)	2.21*** (0.64)	0.15 (0.55)
Maternal education	0.22+ (0.13)	0.18 (0.66)	−0.02 (0.01)	−2.41*** (0.72)	−0.75 (0.72)	0.13 (0.64)	1.10+ (0.64)	0.65 (0.60)	0.58 (0.53)
Variance estimates
Intercept	1.98 (1.24)	19.60 (33.93)	0.04** (0.02)	111.54*** (34.56)	152.04*** (37.98)	97.48*** (31.11)	68.88** (29.51)	65.65** (26.85)	66.52*** (20.46)
Within residual	19.70*** (1.70)	566.75*** (49.81)	0.22*** (0.02)	477.42*** (40.67)	410.07*** (36.36)	366.50*** (32.90)	423.97*** (36.86)	352.20*** (31.44)	251.38*** (22.15)
−2LL	2629.1	4098.9	656.4	4089.7	4057.1	3981.0	4015.6	3940.5	3812.8

Notes. Coef = unstandardized coefficients. SE = standard error of the coefficient. −2LL = −2 log likelihood.

⁺p < .10.

^*p < .05.

^**p < .01.

^***p < .001. For Wave 1 change over time, positive coefficients indicate a decrease from W1 to W2 (i.e., W1 value is greater compared to W2) and negative coefficients indicate an increase from W1 to W2. For Wave 3 change over time, positive coefficients represent an increase from W2 to W3 (i.e., W3 value is greater compared to W2) and negative coefficients represent a decrease from W2 to W3.

Variations by Personal Characteristics

Gender.

On average, women included significantly lower proportions of friends and higher proportions of girls/women in their networks over time, compared to men (Table 3). Significant interactions between respondent gender and wave were observed for contact frequency and percentage of girls/women in the network (Table 4). From childhood to adulthood (W1 to W2), women included proportionally fewer girls/women (b=−7.04, p<.01) in their networks, whereas percentage of girls/women in men’s networks remained stable (b=1.16, p>.05). Across adulthood (W2 to W3), contact frequency decreased significantly for men (b=−0.34, p<.001) and remained stable for women (b=−0.08, p>.05). There were no significant interactions between wave and gender for other characteristics, indicating that findings were similar in men and women.

Table 4.

Results from Models Estimating the Effects of the Interactions between Wave and Personal Characteristics on Convoy Characteristics

	Network Size Coef. (SE)	Closest Others Coef. (SE)	Contact Frequency Coef. (SE)	Proximate Coef. (SE)	Immediate Family Coef. (SE)	Extended Family Coef. (SE)	Friends Coef. (SE)	Age-mates Coef. (SE)	Women Coef. (SE)
Intercept	9.49*** (0.41)	49.38*** (2.14)	4.30*** (0.04)	83.01*** (2.11)	49.90*** (2.05)	17.89*** (1.88)	21.84*** (1.95)	27.94*** (1.79)	50.34*** (1.55)
Change over time (ref = Wave 2)
Wave 1	−1.02* (0.51)	9.29*** (2.74)	0.05 (0.05)	4.86+ (2.52)	−1.34 (2.35)	14.11*** (2.22)	−4.40+ (2.37)	−7.37*** (2.16)	2.94 (1.81)
Wave 3	0.21 (0.73)	4.94 (3.23)	−0.21** (0.06)	−10.96*** (2.99)	9.65*** (2.81)	−5.21* (2.64)	−0.43 (2.81)	−6.72** (2.56)	2.95 (2.16)
Women	0.18 (0.39)	2.10 (2.04)	0.02 (0.04)	1.15 (2.02)	1.27 (1.96)	−0.63 (1.80)	−3.84* (1.86)	−0.09 (1.71)	4.97*** (1.48)
x Wave 1	0.56 (0.50)	−0.13 (2.65)	−0.05 (0.05)	−1.37 (2.43)	−2.34 (2.28)	2.88 (2.14)	1.05 (2.29)	−3.17 (2.09)	4.10* (1.75)
x Wave 3	0.11 (0.58)	−2.20 (3.08)	0.13* (0.06)	4.28 (2.85)	2.14 (2.67)	−0.94 (2.51)	1.91 (2.68)	−2.45 (2.44)	−1.94 (2.05)
White	0.76+ (0.51)	−3.55 (2.16)	−0.08+ (0.04)	−1.92 (2.13)	−0.05 (2.07)	−3.60* (1.90)	3.40+ (1.97)	3.65* (1.81)	−3.61* (1.56)
x Wave 1	0.08 (0.51)	0.36 (2.75)	0.008 (0.05)	2.95 (2.53)	−4.20+ (2.37)	0.28 (2.23)	3.31 (2.37)	2.96 (2.17)	0.04 (1.82)
x Wave 3	−0.50 (0.60)	1.10 (3.21)	0.08 (0.06)	5.88* (2.98)	−1.01 (2.80)	3.80 (2.63)	−2.29 (2.80)	−1.86 (2.55)	1.95 (2.15)
Age	0.03 (0.18)	0.51 (0.90)	−0.005 (0.02)	0.05 (0.97)	3.48*** (0.98)	−1.07 (0.88)	−1.86* (0.88)	−2.13** (0.81)	−0.20 (0.73)
Education	0.21 (0.15)	−1.37+ (0.77)	0.005 (0.02)	−1.95* (0.76)	−2.26** (0.73)	−0.12 (0.67)	2.29** (0.70)	2.44*** (0.64)	−0.05 (0.55)
Maternal education	0.22+ (0.13)	0.19 (0.66)	−0.02 (0.01)	−2.39*** (0.72)	−0.77 (0.72)	0.18 (0.65)	1.08+ (0.64)	0.61 (0.60)	0.64 (0.53)
Variance estimates
Intercept	2.04* (1.23)	20.63 (33.99)	0.04** (0.02)	114.54*** (34.19)	151.83*** (37.47)	101.00*** (31.15)	74.63** (29.75)	70.49** (26.96)	72.50*** (20.59)
Within residual	19.52*** (1.68)	564.69*** (49.69)	0.21*** (0.02)	464.02*** (39.56)	402.12*** (35.63)	358.80*** (32.28)	413.78*** (36.12)	342.35*** (30.67)	239.48*** (21.22)
−2LL	2626.4	4098.2	644.0	4080.7	4050.0	3975.6	4010.4	3933.7	3801.2
Change in −2LL	2.7	0.7	12.4*	9.0	7.1	5.4	5.2	6.8	11.6*

Note. Coef = unstandardized coefficients. SE = standard error of the coefficient. −2LL = −2 log likelihood. Change in −2LL reflects change from main effects models (Table 3).

⁺p < .10.

^*p < .05.

^**p < .01.

^***p < .001. For Wave 1 change over time, positive coefficients indicate a decrease from W1 to W2 (i.e., W1 value is greater compared to W2) and negative coefficients indicate an increase from W1 to W2. For Wave 3 change over time, positive coefficients represent an increase from W2 to W3 (i.e., W3 value is greater compared to W2) and negative coefficients represent a decrease from W2 to W3.

Race.

On average over time, White participants had significantly larger networks and larger proportions of friends and age-mates in their convoys compared to Black respondents (Table 3). White participants also included smaller proportions of closest others, extended family, and girls/women in their convoys. There was a significant interaction of race with changes in proximity in adulthood (W2 to W3), indicating that Black respondents experienced a significantly greater decrease in network proximity than White participants (Table 4). However, adding this interaction term did not significantly improve model fit so we interpret with caution.

Discussion

Change in the structure and composition of social convoys across the lifespan is central to many theories of social relations. The present study investigated trajectories of change in social convoys from childhood through early adulthood. We examined whether specific characteristics of convoy structure change over time, and whether these trajectories varied by respondent gender and race. These findings bolster those in the literature, provide empirical support for the convoy model of social relations and other life course perspectives, and pave the way for more in-depth investigations into social convoy changes early in the lifespan.

Trajectories of Change

Observed changes in composition (i.e., family composition, closest others, age-mates), contact, and proximity are in the expected directions based on developmental shifts in social roles (Arnett, 2000) and social motivations (Carstensen, 1992) that are characteristic of the transition to adulthood and established adulthood. No significant changes in the overall size and in the proportion of friends were observed, contrary to hypotheses. The stability of network size indicates a need to focus instead on composition.

In terms of family composition, the proportion of extended family significantly decreased during the transition to adulthood, then remained stable. In contrast, the proportion of immediate family was stable, then significantly increased. It is important to consider change in the developmental meaning of role relationships over time. For instance, in childhood, immediate family refers to parents and siblings, but in young adulthood, the definition expands to include spouse and children. This is an example of heterotypic continuity, defined by continuity of concepts (e.g., close family), but differences in manifestation across developmental periods (Chen et al., 2017). Thus, the change in immediate family composition most likely reflects family formation typical of established adulthood.

The significant decrease in proportion of closest others within convoys, in favor of more peripheral ties, and increase in proportion age-mates is particularly demonstrative of the network diversification typical during the transition to adulthood, as suggested by SST (Charles & Carstensen, 2010). Notably, proportion of same-aged peers shows an inverted U-shaped pattern, increasing then decreasing. These patterns reflect the developmental tasks of early adulthood, namely higher educational attainment or vocational training and entry into the workforce (Arnett, 2000; Settersten, 2007).

We observed a consistent decrease in network proximity over time, and a significant decrease in contact frequency across adulthood. The decrease in contact frequency during this period is consistent with existing literature (Sander et al., 2017; Sneed et al., 2006), although there is also substantial evidence that this is heavily dependent on relationship type (e.g., family vs. non-family contact) and contact mode (e.g., in-person or remote). The present study aggregates across network members, limiting our ability to specify change by relationship type or contact mode. The literature suggests that this decline is primarily driven by a decrease in contact frequency with nonfamily or otherwise peripheral ties. Even when there is continuity in the presence of certain social ties, as is expected based on the convoy model, the nature of those ties can change (Feld et al., 2007). For example, core network members, like family of origin or close friends, may be included in the convoy across long periods of time; but contact with, proximity to, and even emotional closeness could change as individuals progress through adulthood.

Overall, the trajectories of change reflect increased diversification of convoys from childhood to the transition to adulthood (e.g., fewer close others, more same-aged peers, less geographic proximity). According to SST, personal network diversification serves to meet information acquisition goals and the transition markers of emerging adulthood, which could and should be considered a healthy developmental indicator (Carstensen, 1993). Similarly, the trajectories of convoy characteristics across young adulthood are consistent with the developmental tasks of established adulthood, including family formation and related roles shifts. Social convoys are shaped by life course events, role shifts, and developmental transitions (Antonucci, Ajrouch, Webster, et al., 2019; Marsden, 2018). For example, studies document changes in the composition and support of adults’ social networks following childbirth (Bost et al., 2002; Cronenwett, 1985). Similarly, adulthood educational attainment has been shown to underlie many differences in social relations (Ajrouch et al., 2005). Although the present study does not explicitly link convoy characteristics to life course events, the sample demographics suggest this is the case. At W2, most participants were never married, had no children, and only about half reported working full time. In contrast, by W3, most participants were married or partnered, had children, and were working full time. Future work that localizes the timing of changes in social relations occur relative to life course events would capture the nuances of these trajectories over the life course.

Personal characteristics

Gender.

On average, women reported significantly lower proportions of friends compared to men. Gender differences in family composition have been fairly consistent, with men’s personal networks consisting of a greater proportion and variety of non-kin, (Marsden, 1987). Alternatively, women have been found to include more kin in their personal networks than men. This difference was not observed in the present study, possibly because we distinguished between immediate and extended family rather than combining the two. However, this finding is consistent with the “kin keeper” role, organizing family events and keeping in touch with family, that many women adopt and appear to be socialized into from a young age (Rosenthal, 1985).

Change in contact frequency and gender composition of convoys also differed between men and women. Contact frequency decreased only for men. Studies of contact frequency among older adults find that men report overall less frequent and declining contact over time than women (Phongsavan et al., 2013). Our findings indicate that these gendered patterns may begin early in adulthood. Among women, proportion of girls/women in their convoys decreased, whereas it remained stable for men. This could be explained by a tendency towards gender homophily, especially in childhood, and diversification of convoy composition in early adulthood.

Race.

Average differences between White and Black participants over time supported our hypotheses and were consistent with previous literature. In particular, differences in convoy size and composition (i.e., family, friend, age, and gender) were consistent with previously identified patterns suggesting Black individuals maintain relatively small, close-knit, family-centric networks (Ajrouch et al., 2001; Fung et al., 2001; Levitt et al., 1993; Taylor et al., 2013). The current study indicates that these differences are established in childhood and persist across early adulthood. One explanation for this is that Black individuals cultivate close, social ties as children and maintain these protective networks, rather than diversify with potentially threatening outside ties (Antonucci, Ajrouch, Webster, et al., 2019). Our finding that network proximity decreases more for Black individuals is in contrast to this idea, and may indicate important age or cohort differences. Findings on race differences in network proximity have been inconclusive (Ajrouch et al., 2001; Cantor et al., 1995) and require more systematic study. The racial context of close ties is an area in critical need of further study, given evidence that race-related stress and discrimination determine the extent to which social relations are protective (Williams et al., 2019).

Limitations and Future Directions

This longitudinal study provided a unique opportunity to capture transitions in social convoy characteristics across emerging and established adulthood. The current descriptive findings on structure and composition provide a basis from which to examine other dimensions of social convoys, like relationship quality and social support exchanges. Incorporating these measures with consideration for heterotypic continuity of convoy function would provide developmental context for any observed stability or change in relationship function. A natural extension of this work is to investigate the effect of changes in convoy structure on health and well-being outcomes to determine which types of transitions or changes are beneficial. Studies have identified distinct profiles and transition patterns of social support exchange across adolescence and early adulthood, which are associated with well-being (Ciarrochi et al., 2017; Fang et al., 2020). Importantly, although these data spanned 23 years, no information is available from adolescence, when we expect many important developmental changes. A larger sample with more frequent time points, especially spanning adolescence, would allow us to localize trajectories of change with greater precision.

We acknowledge limitations in the generalizability of our findings. The present sample included only individuals who identified as boys/men or girls/women and Black or White. Thus, our findings are limited in depicting developmental trajectories of individuals with non-binary gender or diverse racial identities. Finally, there are likely cohort-specific effects given generational differences in the influence of technology and social media on how we form and maintain social ties (Antonucci et al., 2017). Certain convoy characteristics, including contact frequency and geographic proximity, take on significantly different meanings today given the advent of new communication technologies.

Conclusions

The present study describes stability and change in social convoys from childhood to adulthood by leveraging the strengths of a unique longitudinal dataset and applying innovative methodological approaches. Findings delineated aspects of close, personal networks that exhibit change across important developmental periods. The long reach of personal characteristics is evident through gender and race differences, emphasizing the importance of acknowledging these nuances in understanding form and function of social relations. These convoy characteristics early in the lifespan lay the groundwork for life course events, transitions, and role shifts. Ultimately, such findings could result in diagnostic profiles that detect individuals at risk for or undergoing problematic developmental transitions and identify strategies to leverage social relations for prevention or protection early in the lifespan.