Abstract
Background
The National Social Life, Health, and Aging Project (NSHAP) has collected 3 rounds of data on older adults’ egocentric social networks. We describe the structure of network data collection for different components of the sample and the data that are available for those groups. We also describe survey techniques that were used to track specific personnel changes that occurred within respondents’ networks during the 10-year study period.
Method
Descriptive statistics are presented for measures of network size, composition, and internal structure at all 3 rounds, respondent-level summary measures of change in these characteristics between and across rounds, and measures of change associated with the loss and addition of network members across Rounds 1, 2, and 3. Procedures that were used to clean the network change data are also explained.
Results
The NSHAP network change module provides reliable information about specific changes that occurred within respondents’ confidant networks. For returning baseline respondents, there is considerable overlap with respect to which confidants are named in successive rosters, but the norm is for Round 3 networks to be composed primarily of new confidants.
Discussion
These data provide new insights into the dynamic nature of networks in later life. Data limitations, and directions for future research, are discussed.
Keywords: Aging, Life course, Network change, Social isolation, Social networks
Scholars are increasingly interested in older adults’ social networks and how they change through later-life transitions (e.g., Cornwell et al., 2014; Fischer & Beresford, 2015; Goldman & Cornwell, 2018; Iveniuk et al., 2020; Menkin et al., 2016; Schwartz & Litwin, 2018; Spillman et al., 2019; Wrzus et al., 2013). There is growing evidence that how and why older adults’ social networks change has important consequences, including the fact that network changes have health effects above and beyond baseline levels of social connectedness (Cornwell & Laumann, 2015; Litwin et al., 2020; Petersen et al., 2016; Schwartz & Litwin, 2019). The National Social Life, Health, and Aging Project (NSHAP) collected the first nationally representative data on changes in older adults’ social network (see Cornwell et al., 2014) across two rounds of data collection spanning 5 years (Rounds 1 and 2 [R1 and R2]). These data have been critical in providing information about patterns of change in later-life network connections, such as how much network member turnover older adults experience (e.g., see Cornwell et al., 2014); how network turnover and confidant deaths are related to factors like social disadvantage (e.g., Cornwell, 2015; Goldman & Cornwell, 2018), residential mobility (Badawy et al., 2019), caregiving (Roth, 2020), and the ability to drive (Schafer, 2018); and the potential implications of these changes for important outcomes like health (e.g., Cornwell & Laumann, 2015; Schafer & Koltai, 2015).
The NSHAP collected additional data on network change at R3 using a survey technique, which we describe here, that captured specific changes within respondents’ networks between R2 and R3, as well as information about how R3 networks linked back to baseline respondents’ networks as measured at R1. We also describe measures of several potentially consequential elements of network change and present some preliminary findings.
NSHAP data are publicly available through the National Archive of Computerized Data on Aging within the Inter-university Consortium for Political and Social Research (ICPSR) (Waite, Linda J., Edward O. Laumann, Wendy Levinson, Stacy Tessler Lindau, and Colm A. O’Muircheartaigh. NSHAP: Wave 1. ICPSR20541-v6. Ann Arbor, MI: ICPSR [distributor], April 30, 2014. doi:10.3886/ICPSR20541.v6; Waite, Linda J., Kathleen Cagney, Benjamin Cornwell, William Dale, Elbert Huang, Edward O. Laumann, Martha McClintock, Colm A. O’Muircheartaigh, and L. Phillip Schumm. NSHAP: Wave 2 and Partner Data Collection. ICPSR34921-v1. Ann Arbor, MI: ICPSR [distributor], April 29, 2014. doi:10.3886/ICPSR34921.v1; Waite, Linda J, Kathleen Cagney, William Dale, Louise Hawkley, Elbert Huang, Diane Lauderdale, Edward O. Laumann, Martha McClintock, Colm A. O’Muircheartaigh, and L. Phillip Schumm. NSHAP: Wave 3. ICPSR36873-v1. Ann Arbor, MI: ICPSR [distributor], October 25, 2017. doi:10.3886/ICPSR36873.v1).
Method
The NSHAP team is interested in measuring changes in respondents’ social networks throughout later life, including over both 5-year periods (a) between the first round (2005–2006) and the second round (2010–2011), (b) between the second round and the third round (2015–2016), as well as (c) over the 10-year period between the first and third rounds. There were 3,005 R1 NSHAP respondents and 3,377 R2 respondents, including 955 partners of R1 respondents and 161 people who were approached but did not participate at R1. There were 4,777 R3 NSHAP respondents, 2,409 “return” respondents from earlier rounds comprising the first cohort, and 2,368 “new cohort” respondents and coresiding partners comprising the second cohort. The sampling strategy and overall study design are described in greater detail elsewhere in the supplemental issue.
New cohort members do not yet have any social network change data. Respondents who were enrolled at R2 do have one 5-year period’s worth of data on social network change. We present those data briefly, but our primary focus will be on novel features of the multiround data that provide new insights into egocentric social network change over a 10-year period.
Social Network Data Collection
The NSHAP study is comprised of four basic elements of egocentric network data: (a) network rosters, or lists of network members; (b) network interpreters, which are questions that probe for more information about any listed network members, and/or ego’s relationship with those alters (e.g., their gender, their frequency of contact with ego); (c) information on changes in the network over time (including who was lost from or added to the network over time); and (d) network roster references, which identify whether particular social phenomena are linked to particular network members later in the NSHAP survey. The primary task is the enumeration of respondents’ network members using name generators. At each time point, every respondent (“ego”) identifies the relevant set of people (“alters”) to whom s/he is connected. The NSHAP adopts the philosophy that respondents themselves are the best sources of information about who the most relevant network members are (see Cornwell et al., 2009; Marsden, 2011). They then provide information about their network members and their relationships with them, as well as alters’ relationships with each other—thus yielding egocentric social network data.
Identifying network members (rosters)
The NSHAP collected egocentric social network data from all respondents at all rounds using computer-assisted personal interviewing. At each round, the interview began with a module that records who the alters are. This module was placed at the beginning of the interview to minimize interviewer effects and respondent fatigue, which can affect the number of network members named (Paik & Sanchagrin, 2013). Four “rosters,” or lists of alters, were collected: A, B, C, and D. Roster A contained respondents’ “confidants.” To elicit the names of confidants, interviewers asked the following name generator: “From time to time, most people discuss things that are important to them with others. For example, these may include good or bad things that happen to you, problems you are having, or important concerns you may have. Looking back over the last 12 months, who are the people with whom you most often discussed things that were important to you?” Respondents could name up to five confidants. This name generator intentionally prompts respondents to consider with whom they have interacted over the prior year, which tends to elicit names of strong, frequently accessed, long-term contacts—ties through which normative pressures and social influence are likely to operate (Marin, 2004; Straits, 2000; cf., Bearman & Parigi, 2004). The vast majority of confidants identified in NSHAP at R1 were relatively strong ties in terms of frequency of contact, emotional closeness, and respondents’ likelihood of discussing health matters with confidants (Cornwell et al., 2009). Analyses for this paper will focus mainly on the characteristics of alters included in Roster A.
Rosters B, C, and D capture other potentially important network members. When respondents who had a spouse or romantic partner did not include that person in Roster A, that individual was recorded in Roster B. Otherwise, Roster B is not used. Following this, respondents were asked: “(Besides the people you already listed), is there anyone (else) who is very important to you, perhaps someone with whom you feel especially close?” If such an individual was identified, s/he was recorded in Roster C. This item was added to ensure inclusion of important contacts who may not have been captured by the main name generator. Finally, household members not captured in Rosters A, B, or C were recorded in Roster D.
All told, the NSHAP social network data corpus includes data on 38,890 network members from original baseline respondents, their partners, and new cohort members and their partners. This includes data on 13,125 alters at R1, 15,923 alters at R2, and 21,232 alters at R3. Of these, 30,136 are alters in Roster A (i.e., confidants), 1,612 are in Roster B (any spouses/partners who were not listed in Roster A), 3,177 are in Roster C (other important contacts), and 3,965 are in Roster D (household members who do not appear in an earlier network roster).
Assessing network composition and structure (name interpreters)
The second element of the network data collection involves what are known as “network interpreters,” which are questions about the network members who were named in the above rosters. NSHAP has consistently collected information about each network member and the type of relationships they have with the respondent, including their name (initials, first name, or nickname only), gender, coresidence status, and age (if that person is a coresident), as well as respondents’ frequency of contact with each person named in Rosters A, B, and C, and those alters’ frequency of contact with the other alters named in these same rosters. Items assessing issues such as frequency of contact can be especially useful for getting a sense of tie strength, or the extent of potential influence a given alter has on a given respondent, which is associated with outcomes like health (see Lin et al., 1985; Terhell et al., 2007).
Assessing network change
The third element of NSHAP’s network data collection involves a module that attempts to track changes in the membership of respondents’ social networks. This module was introduced at R2 to assess change in respondents’ baseline networks. This module involves asking respondents who had participated in the study in a previous round to identify which of the alters they named in Rosters A–C in the current round (R2 and/or R3) match up to the alters they named in those same rosters in the preceding round(s). This procedure, as it was first executed at R2, is described in detail in Cornwell and colleagues (2014).
One potential challenge arises when confirming some of the reported matches. This stems from NSHAP’s efforts to protect the identities of network members who are not themselves respondents. Respondents were asked to report only the first name, nickname, or initials of each alter. This, combined with other information about those alters (e.g., gender, relationship), was usually enough for the respondent to confirm the match. Coders reviewed and confirmed all matches, and suggested corrections for cases in which matched alters differed on certain characteristics (e.g., name, gender, relationship, or implausible age difference between rounds) and/or where it seemed that the alters could have been entered on incorrect line numbers.
To confirm matches, the NSHAP team employed a multicoder strategy. First, two separate coders analyzed together the rosters of 211 respondents from R1 to R3 (which included 1,798 total roster lines) to ascertain agreement of alter dispositions across those respondents’ three rounds, or waves, of network members. The coders agreed on 1,784 alter cases (99.2% of the time), suggesting sufficient intercoder reliability. Next, the coders split the remainder of the cases to confirm matches across rosters. Of these remaining cases, we were able to confirm matches in the vast majority. Just over 3% of respondents who were interviewed at all three rounds had any type of discrepancy or inconsistency in any of their alter matches across the three rounds of data.
With the matched roster data, one can distinguish between alters who were named at a previous round but not at the current round (“lost” alters), those who were named for the first time at the current round (“new” alters), those who were named at both (or all) rounds in question, those who were added then lost over a three-round period, as well as alters who were lost and then added back again at a third round. Regarding lost alters, additional information has been collected. At R2 and R3, respondents indicated whether any lost alter was still alive and, at R2, they provided information about why that alter was lost. (For more detailed information about why alters are dropped from confidant networks between rounds, see Cornwell et al. [2014] and Fischer & Offer [2020].) At both R2 and R3, respondents also reported on the duration of their relationships with any new alters.
Network roster references
The final element of NSHAP’s network data concerns a set of questions that appear later in the in-home interview. These questions ask about whether certain people who (a) have mistreated them, (b) serve as their primary medical decision maker, or (c) serve as their primary caregiver are mentioned somewhere on the rosters collected above. Respondents are then asked to provide the roster line on which that individual appears.
Network Data Collection at Different Rounds
Collection of the egocentric network data described in the previous section has changed across R1, R2, and R3. This is due to a number of factors, including time/budgetary constraints, as well as shifts in empirical priorities (e.g., increasing interest in the health effects of network change). In addition, new types of respondents have been added to the sample at different points, creating a somewhat complex network data structure. However, the NSHAP team has been mindful to preserve the potential to both track various aspects of networks in a comparable manner across rounds, as well as to enable comparisons of network connectedness across cohorts.
For ease of reference, Table 1 lays out the structure of data collection for different elements of respondents’ egocentric social networks throughout the NSHAP study. Most of the network measures are available in all three rounds for all members of the first cohort (e.g., original baseline respondents sampled in R1), as well as for both R2 and R3 for partners of members of that cohort. There will also be considerable opportunity to compare network elements across cohorts (e.g., respondents sampled in R1 and respondents sampled in R3), including for partners. There are also plans to extend the analysis of network change to a fourth round, which will allow for additional analysis of network change trends for Cohort 1 and their partners, as well as how that relates to network change in Cohort 2.
Table 1.
The Elements of the NSHAP Egocentric Social Network Data Collection Across Rounds
| Round(s) at which data are available for: | |||
|---|---|---|---|
| Network data element | Cohort 1 | Cohort 1 partners | Cohort 2 and their partners |
| Network rosters | |||
| A (confidants) | 1, 2, 3 | 2, 3 | 3 |
| B (previously unnamed spouse/partner) | 1, 2, 3 | 2, 3 | 3 |
| C (one other important contact) | 1, 2 | 2 | — |
| D (any other coresidents) | 1, 2, 3 | 2, 3 | 3 |
| Network interpretersa | |||
| Type of relationship | 1, 2, 3 | 2, 3 | 3 |
| Alter gender | 1, 2, 3 | 2, 3 | 3 |
| Alter coresidence status | 1, 2, 3 | 2, 3 | 3 |
| Alter age (if coresident)b | 1, 2, 3 | 2, 3 | 3 |
| Emotional closeness to alter | 1, 2 | 2 | — |
| Geographic proximity to alter | 3 | 3 | 3 |
| Frequency of contact with alter | 1, 2, 3 | 2, 3 | 3 |
| Likelihood of discussing health with alter | 1, 2 | 2 | — |
| Alter frequency of contact with other alters | 1, 2, 3 | 2, 3 | 3 |
| Network changea | |||
| Roster matching to previous waves | 2, 3 | 3 | — |
| Lost alters’ survival status | 2, 3 | 3 | — |
| Reasons for alter loss | 2 | — | — |
| Duration of relationship with new alters | 2, 3 | 3 | — |
| Network roster referencesc | |||
| Roster ID(s) of primary mistreatersd | 1 | — | — |
| Roster ID of medical decision maker | 1 | — | — |
| Roster ID of primary caregiver | 2, 3 | 2,3 | 3 |
Notes: NSHAP = National Social Life, Health, and Aging Project.
aThese name interpreters are asked of any alter named in Rosters A, B, or C.
bAlter age is also asked of those who are named in Roster D.
cRespondents can choose anyone from Rosters A to D.
dFor each one of the following items, respondents were asked for the roster ID (if applicable) of the primary person who: (1) “Is too controlling,” (2) “Has insulted you,” (3) “Has taken money or belongings,” and (4) “Hit you.”
Trends in Social Network Change
Perhaps the most complex, and novel, element of the NSHAP egocentric network data collection is the analysis of over-time change in respondents’ social network rosters. An unprecedented three rounds of nationally representative data on older adults’ egocentric networks now exists. We spend the remainder of this paper describing some preliminary findings.
Aggregate Changes in Network Composition and Structure
Table 2 displays name interpreter information about the composition and structure of confidant (Roster A) network members from among those groups of respondents who have had an opportunity to report on multiple rounds of network data (original baseline respondents and their partners). Briefly, this table shows that most confidants are female (62.01% among baseline respondents at R1), kin (65.80%), not coresident (82.03%), in relatively frequent contact with respondents (46.43% once or several times a week), and in some contact (at least once a year) with most of the respondent’s other confidants (76.19%). These general features are evident for both baseline respondents and their partners. Furthermore, these features are relatively stable across rounds. Information regarding the number and types of ties that are reported by respondents may be useful to researchers who are interested in analyzing the nature of ego–alter dyadic relationships in later life.
Table 2.
Prevalence of Roster A Network Members (Confidants) Who Possess Certain Characteristics, by Study Round and Type of Baseline Respondents in the NSHAP Studya
| Alters of Cohort 1 baseline respondents at: | Alters of Cohort 1 partners at: | ||||
|---|---|---|---|---|---|
| Alter characteristic | Round 1 | Round 2 | Round 3 | Round 2 | Round 3 |
| Gender | |||||
| Female | 6,318 (62.01%) | 5,524 (61.06%) | 3,854 (60.01%) | 2,206 (60.42%) | 1,730 (61.02%) |
| Male | 3,871 (37.99%) | 3,523 (38.94%) | 2,568 (39.99%) | 1,445 (39.58%) | 1,105 (39.98%) |
| Type of relationship | |||||
| Kin | 6,704 (65.80%) | 5,815 (64.23%) | 3,985 (61.82%) | 2,465 (67.50%) | 1,825 (64.19%) |
| Nonkin | 3,485 (34.20%) | 3,238 (35.77%) | 2,461 (38.18%) | 1,187 (32.50%) | 1,018 (35.81%) |
| Coresident? | |||||
| Yes | 1,831 (17.97%) | 1,448 (16.01%) | 955 (14.88%) | 807 (22.10%) | 530 (18.69%) |
| No | 8,358 (82.03%) | 7,599 (83.99%) | 5,463 (85.12%) | 2,844 (77.90%) | 2,305 (81.31%) |
| Frequency of contact | |||||
| Less than weekly | 1,666 (16.36%) | 1,586 (17.54%) | 1,253 (19.53%) | 641 (17.57%) | 536 (18.91%) |
| Once or several times a week | 4,727 (46.43%) | 4,244 (46.95%) | 3,039 (47.37%) | 1,624 (44.52) | 1,310 (46.21%) |
| Every day | 3,789 (37.21%) | 3,210 (35.51%) | 2,124 (33.10%) | 1,383 (37.91%) | 989 (34.89%) |
| Percent of R’s other alters in contactb | 76.19% | 75.14% | 72.14% | 76.88% | 74.29% |
| Total number of alters | 10,189 | 9,053c | 6,446d | 3,652 | 2,843d |
Notes: NSHAP = National Social Life, Health, and Aging Project.
aEstimates are unweighted. Data are reported for all baseline respondents for whom data are available at a given round, and are not restricted to respondents who participated at multiple rounds.
bCalculated as the percent of respondent’s other confidants with whom the respondent reports the alter in question has been in contact at least once a year. There are more missing data here due to the fact that valid data exist only where respondents reported having at least two confidants.
cIncludes alters of respondents who were in the Rounds 1 sampling frame but who did not participate at Round 1.
dIncludes alters of respondents who did not participate at Round 1 and/or Round 2.
As documented elsewhere (Cornwell et al., 2014), there is some evidence of a steady but slight trend between R1 and R2 of confidant networks becoming slightly less kin-centric, more inclusive of people who live outside of the home, involving less frequent contact between alters and respondents, and less dense (i.e., lower percent of alters who are in contact with a respondent’s other confidants). These trends appear to continue into R3, and are evident for both baseline respondents and their partners.
Patterns of Membership Change Across Rounds
Changes in roster membership
The Venn diagram in Figure 1 depicts the extent to which networks overlap and evince turnover among successive rounds of the NSHAP study. As a demonstration, this analysis pertains only to those individuals who completed confidant network rosters at all three rounds, as well as both rounds of the network roster matching exercise in R2 and R3 (N = 1,553). This means that the analysis does not include partners or new cohort members.
Figure 1.
Proportional Venn diagram representing the amount of overlap among the three rounds of confidant network rosters that were collected from baseline NSHAP respondents over a 10-year period (2005/2006–2015/2016).
Notes: NSHAP = National Social Life, Health, and Aging Project. These levels of change reflect movement into and out of the confidant network roster (Roster A) in the NSHAP interview. These numbers are calculated only for baseline respondents who provided valid social network data at all three rounds (N = 1,552 respondents; N = 11,037 confidants), excluding respondents who were present at only one or two rounds of data collection. The proportional diagram was created using the “pvenn2” command in Stata 14.2.
Each of the three gray circles in Figure 1 represents the set of confidants (Roster A) these respondents named in their rosters during the study period (N = 11,037 unique confidants named, in total, across rounds). This diagram is useful for conveying some important information about network size and turnover patterns. For one, notice that the left-most circle, which contains confidants who were named at R1, is smaller than the circle on the bottom, which contains R3 confidants. This reflects the fact that, overall, networks grew progressively larger throughout the study period. Among this subset of respondents, the average network sizes at R1, R2, and R3 were 3.55, 3.86, and 3.84, respectively.
We are also interested in patterns of overall network change. The sizes of the sections of Figure 1 where circles overlap are proportional to the numbers of confidants who were named by these respondents in two or more rounds. For example, 1,034 confidants were named in both the R1 and R2 confidant rosters but not in the R3 roster.
Generally, these networks are characterized by change (as opposed to complete stability) during this 10-year period. Fully 59.0% of the confidants named across the study period were named in only one round, while an additional 23.7% were named in two out of three rounds. Notice that there is, however, a relatively small, steady core of 1,908 confidants (17.3%) who are present at all three rounds.
At the alter level, patterns of change between R2 and R3 are similar to those observed between R1 and R2—as evidenced by the relatively proportionate levels of overlap that exist between the R1 and R2 circles and between the R2 and R3 circles. With respect to confidant losses, 46.6% of R1 confidants were lost between R1 and R2, similar to the 50.3% observed between R2 and R3. With respect to confidant additions, 50.9% of confidants at R2 were new confidants (i.e., they had not been characterized as confidants at R1), compared to 50.1% at R3. However, 517 (17.3%) of these “new” R3 confidants had been named at R1 but not at R2. Thus, some of the losses that occurred between R2 and R3 were supplemented by “resurrected” confidants that presumably had been “lost” earlier in the study period. This phenomenon will be discussed in greater detail below.
Respondent-level patterns of network change
Many analysts will likely be interested in respondent-level patterns of network change. There are six possible types of change in respondents’ network rosters between a given pair of rounds: (a) R lost at least one confidant who had been named in the earlier-round roster but did not name any new ones in the later-round roster (“Losses, no additions”); (b) R lost at least one confidant from the earlier-round roster, and added some new ones in the later-round roster, but not enough to offset the losses from the earlier roster (“More losses than additions”); (c) R added at least one confidant to the later-round roster who did not appear in the earlier round, and did not drop any from that earlier roster (“Additions, no losses”); (d) R added at least one confidant to the later-round roster who did not appear in the earlier roster, and lost at least one from the earlier roster, but, overall, added more than lost (“More additions than losses”); (e) R added to the later-round roster the same exact number of confidants s/he lost from the earlier round (“Equal roster turnover”); or (f) R’s roster remained completely unchanged between the earlier and later rounds.
Table 3 details the extent to which NSHAP respondents experienced these types of change. Between R1 and R2, the modal experience (28.71%) is equal roster turnover, as many respondents reported that they lost the same number of alters that they added. More respondents reported (net) network additions (37.88%) than (net) losses (26.67%). Relatively few respondents reported the exact same Roster A network members at both R1 and R2 (6.73%). These patterns reemerge between R2 and R3. There is a slight increase in the prevalence of equal roster turnover (35.54%) during that interround period, and the share of people who experience (net) network losses versus those who experience (net) network additions virtually equalizes (30.2% vs 28.84%, respectively).
Table 3.
Patterns of Change in Roster A (Confidant) Network Membership Across Rounds 1, 2, and 3 of the NSHAP Studya
| Prevalence of pattern of change between: | |||
|---|---|---|---|
| Pattern of change in Roster A membership | Rounds 1 and 2 | Rounds 2 and 3b | Rounds 1 and 3c |
| (1) Losses, no additions | 255 (11.30%) | 187 (12.04%) | 152 (9.55%) |
| (2) More losses than additions | 347 (15.37%) | 282 (18.16%) | 276 (17.35%) |
| (3) Additions, no losses | 356 (15.77%) | 160 (10.30%) | 190 (11.94%) |
| (4) More additions than losses | 499 (22.11%) | 288 (18.54%) | 440 (27.66%) |
| (5) Equal roster turnover | 648 (28.71%) | 552 (35.54%) | 477 (29.98%) |
| (6) Roster at earlier round = roster later | 152 (6.73%) | 84 (5.41%) | 56 (3.52%) |
| Total number of respondents | 2,257 | 1,553 | 1,591 |
Notes: NSHAP = National Social Life, Health, and Aging Project.
aEstimates are unweighted. Analysis excludes respondents who did not provide valid network data. Estimates do not include cases where respondents named alters at R2 but not at either R1 or R3.
bAnalysis excludes respondents who did not participate at R1.
cEstimates may include cases where respondents named particular alters at R1 and R3 but not R2.
Network changes over the 10-year period
The third column shows trends over the full 10-year study period. Here, estimates of stability are much lower (3.52%), and equal roster turnover continues to be the modal experience (29.98%). Looking across this period, (net) network additions are more common (39.60%) than are (net) network losses (26.90%). Note that we do not count instances in which a respondent named a network member at R2 but not at either R1 or R3 as instances of either loss or addition. Note also that some of what initially appeared to be instances of (net) network loss between R1 and R2 are due to respondents not naming R1 confidants again at R2 but then naming them again at R3. These cases are not counted as either network losses or additions in the R1–R3 change estimates. This leads us to our last issue—the return of confidants who were “lost” between R1 and R2 but who show up again in the R3 roster—which we address in the next section.
The Disposition of “Lost” Confidants
Table 4 provides information about alters who were named as confidants at R1 but not named again as confidants in at least one of the subsequent two rounds. The largest of these groups include alters who were included in Rosters A at R1 only (see column 1). Some (16.9%) of these losses are due to the deaths of those confidants. A similar number (17.9%) are due to residential mobility. Most of such “losses” are due to other reasons that were not recorded at all three rounds, including any actual weakening of ties due to life-course events like retirement, the onset of health problems, as well as generic “we drifted apart” or “we’re still in touch” types of explanations. It is worth noting that a nonnegligible number (10.1%) of those alters were not lost at all, but rather merely “demoted” and named instead by respondents in Rosters B or C in subsequent rounds.
Table 4.
Over-Time Dispositions of Alters Who Were Named in Roster A (Confidants) in Round 1 but Then Subsequently Droppeda
| Alters who were named at: | |||
|---|---|---|---|
| Reason for loss of confidant | Round 1 only | Rounds 1 and 2 only | Rounds 1 and 3 onlyc |
| Alter died | 347 (16.90%) | 725 (70.12%) | 0 (0.0%) |
| (Temporary) Roster demotion in next round(s) | 208 (10.14%) | 122 (11.80%) | 152 (29.40%) |
| Alter and/or R moved away | 367 (17.88%) | 169 (16.34%)b | 54 (10.44%) |
| Some other reason | 1,049 (51.12%) | 292 (56.48%) | |
| Missing/don’t know/refused | 81 (3.95%) | 18 (1.74%) | 19 (3.68%) |
| Total number of alters | 2,052 | 1,034 | 517 |
Notes:
aEstimates are unweighted. Analysis excludes respondents who did not provide valid network data at all three rounds. Estimates include only alters who meet the same analytic criteria as those included in Figure 1.
bIncludes any explanation apart from “alter died,” roster demotion, and missing/don’t know/refused, and thus may include “alter and/or R moved away.”
cEstimates may include cases where respondents named particular alters at R1 and R3 but not R2.
Table 4 also reveals that alters who were named as confidants at both R1 and R2 but not at R3 (see column 2) were likely disproportionately strong ties. Respondents indicated that 70.1% of those confidants were not named again at R3 because they had died. A further 11.8% of them were named again at R3, but this time they were named in Roster B (“forgotten” spouse/partner) instead.
The Resurrection of “Lost” Confidants
Respondents listed a total of 5,511 confidants at R1. Of these, nearly half (2,569, or 46.62%) were not named again at R2. An implication of the R3 network roster matching exercise is that it presents a unique opportunity to examine to what extent those confidants who were dropped between R1 and R2 were picked up again between R2 and R3. The data show that this happened with 517 of the alters who were named by respondents at any round (see Figure 1 and Table 4). This is just 4.68% of the total confidant pool across rounds, but this group constitutes one-fifth (20.12%) of the confidants who were dropped between R1 and R2.
Potential explanations for this are discussed further below. In the meantime, additional information about these “resurrected” confidants is provided in the third column of Table 4. (Note first that none of these resurrected confidants had been reported dead at R2, which speaks to the reliability of the social network change data.) Nearly one-third (29.4%) of these alters had in fact been named again at some point at R2, but not in Roster A. In other words, at R2 they show up as a “forgotten” spouse/partner (Roster B) or as one’s other “close” contact (Roster B). Other, more substantive, reasons accounted for over half (56.5%) of these resurrected confidants not having been named in the interim at R2 (e.g., tie decay, “we’re still in touch,” or another generic explanations).
Discussion
NSHAP collected the first nationally representative data on changes in older adults’ social network across two rounds of data collection spanning 5 years between 2005/2006 and 2010/2011. Scholars have used these data to document patterns in later-life connections, as well as changes in them and their associations with important predictors and outcomes. Here, we have described the survey techniques that were used to add a third round of data on social networks to the data set. These new data offer researchers an unprecedented opportunity to examine the prevalence and trends in a number of network elements—including network size, alters’ characteristics and their links to other alters, network change, and network links to key health experiences (e.g., primary caregivers)—across two 5-year periods and the overall 10-year study period. We have presented descriptive statistics of these elements for different components of the sample: Cohort 1 respondents, their partners, and Cohort 2 respondents and the data available for these groups.
There is evidence in these data that the structure and characteristics of older adults’ close confidant networks tend to remain stable over time, with some indication that they grow slightly less kin-centric and less dense (partly due to widowhood, which results in the loss of the most central of one’s core confidants). At the same time, the data also show considerable change in roster membership across rounds. Approximately two-thirds of respondents experienced net losses or net gains in confidants, and this is true across each of two 5-year time periods, as well as across the 10-year study period (i.e., between R1 and R3), with only a small, steady core of confidants named across all three rounds. But the modal experience across rounds is equal roster turnover.
We also find evidence of returning (i.e., “resurrected”) confidants: Nearly one-fifth of confidants named in R1 but not at R2 were named again at R3. This provides additional evidence, as noted elsewhere in recent work using other data (Fischer & Offer, 2020), that some “losses” that are recorded in egocentric network surveys may be incidental, reflecting not rapidly changing social environments, but rather environments that contain multiple people who serve as confidants from time to time under different circumstances. In other words, many returning contacts actually may not have been lost in the first place—rather they represent more interchangeable contacts who are part of ego’s broader social environment. This interpretation is supported by the fact that this resurrection was more common among respondents who had larger networks. This is further bolstered by the fact that some confidants who were named at R1 are included as spouses/partners or other “close” ties at R2, then named again as confidants at R3. Another potential explanation for returning ties is that they represent relationships that indeed weakened considerably and became “dormant” (e.g., see Marin & Hampton, 2019), but not so much that one or both parties could not rekindle it, perhaps due to some loss or life change. More lasting social network change appears to be due to more permanent, jarring, or exogenous kinds of events such as residential mobility or confidant death.
Despite these contributing mechanisms, the overall story is one of general stability, or homeostasis, in social networks (Cornwell et al., 2020). Several mechanisms that we cannot measure directly likely contribute to maintaining homeostatic pressure, depending on baseline network characteristics. These include a combination of individuals’ agentic efforts to maintain social environments with which they are familiar and/or that are beneficial to them, similar efforts at adaptation or compensation following life changes (e.g., Atchley, 1989; Baltes & Baltes, 1990), as well as the opportunity structures (e.g., Fiori et al., 2020) in which they are embedded (which can make it difficult to drastically change one’s network), and other structural factors that shape social network formation and change.
A number of limitations to analyzing network elements across rounds exist due to budget constraints, shifting analytic priorities, and the introduction of new respondent types in different surveys; additional challenges exist in linking specific alters across rounds. However, the NSHAP team has been thoughtful to maintain key items to allow for comparability across rounds, and substantial coding effort evidences high matching capabilities of network alters. Finally, the team anticipates extending the survey techniques discussed here in a fourth round, which will allow for novel empirical examination of network structure, trends, and comparability across cohorts and over time.
Additionally, recent critiques raise questions about the egocentric network measure and conceptualization. Egocentric networks do not capture the full set of alters in an individual’s social network, which renders our analyses and their implications specific to the set of core confidants who are represented in this measure. It also is worth noting that the NSHAP intentionally administers the social network module at the beginning of the interview in order to avoid issues around respondent and interviewer fatigue that can be problematic for the validity of “important matters” name generators (e.g., Paik & Sanchagrin, 2013). Studies of egocentric network change also raise questions about whether such changes represent significant or more fleeting social dynamics. It is possible that social network changes (e.g., additions, losses) take place between the survey rounds and are not reflected in the changes that we document here. Nevertheless, we emphasize the significance of the changes as they are measured using the methods described given the growing literature documenting how these patterns are consequential for health and other individual outcomes (e.g., see Bookwala, 2016; Cornwell & Laumann, 2015). Network change should not be considered a reflection of unreliability, and vice versa; rather, some amount of change or instability should be considered an expected property of social networks (Morgan et al., 1997). Future research can use the NSHAP data to shed light on how other network characteristics (e.g., composition, density) change over time (e.g., Litwin & Levinsky, 2020), and how these changes are associated with well-being.
We have analyzed these patterns not only because so little is known about them but also because they likely provide insight into important outcomes in later life. It is possible that the general tendency toward homeostasis reflects the beneficial effects of overall stability in one’s social environment. Given recent evidence that socially disadvantaged older adults’ social environments are more precarious and thus less likely to engender homeostasis (see Cornwell, 2015; Goldman & Cornwell, 2018; Schafer & Vargas, 2016), this may help to explain disparities between social groups with respect to health and mortality in later life. Future work should therefore consider the potential individual-level consequences of social–environmental stability as well as individuals’ abilities to rebalance following such changes, and how this differs by social groups. These dynamics might hold clues about enduring social disparities in health in later life. Finally, future research using a fourth round of network data collection may compare trends in network change between Cohort 1 and Cohort 2 to speculate around how increasing use of online communication and more widespread social media use influences network change. As older adults consider digital media to be a significant context for support exchange (Quan-Haase et al., 2017), it is possible that networks undergo higher rates of change with a wide range of options for forming and sustaining ties through different online platforms.
Acknowledgments
We thank Kathleen Cagney, Louise Hawkley, Erin York Cornwell, and two anonymous reviewers for providing useful suggestions that improved this paper.
Funding
This paper was published as part of a supplement sponsored and funded by the National Institute on Aging and the National Institute on Aging and the National Institutes of Health as part of the National Social Life, Health, and Aging Project (R01AG021487, R01AG033903, R01AG043538, and R01AG048511). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health or NORC at the University of Chicago.
Conflict of Interest
None declared.
References
- Atchley, R. C. (1989). A continuity theory of normal aging. The Gerontologist, 29(2), 183–190. doi: 10.1093/geront/29.2.183 [DOI] [PubMed] [Google Scholar]
- Badawy, P. J., Schafer, M. H., & Sun, H. (2019). Relocation and network turnover in later life: How distance moved and functional health are linked to a changing social convoy. Research on Aging, 41(1), 54–84. doi: 10.1177/0164027518774805 [DOI] [PubMed] [Google Scholar]
- Baltes, P. B., & Baltes, M. M. (1990). Psychological perspectives on successful aging: The model of selective optimization with compensation. In Baltes P. B. & Baltes M. M. (Eds.), Successful aging: Perspectives from the behavioral sciences (pp. 1–34). Cambridge University Press. doi: 10.1017/cbo9780511665684.003 [DOI] [Google Scholar]
- Bearman, P., & Parigi, P. (2004). Cloning headless frogs and other important matters: Conversation topics and network structure. Social Forces, 83(2), 535–557. doi: 10.1353/sof.2005.0001 [DOI] [Google Scholar]
- Bookwala, J. (2016). Confidant availability (in) stability and emotional well-being in older men and women. The Gerontologist, 57, 1041–1050. doi: 10.1093/geronb/gbu078 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Cornwell, B. (2015). Social disadvantage and network turnover. The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences, 70(1), 132–142. doi: 10.1093/geronb/gbu078 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Cornwell, B., Goldman, A., & Laumann, E. O. (2020). Homeostasis revisited: Patterns of maintenance and restoration in the social connections of older adults. Journals of Gerontology, Series B: Psychological Sciences and Social Sciences, 76(4), 778–789. doi: 10.1093/geronb/gbaa026 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Cornwell, B., & Laumann, E. O. (2015). The health benefits of network growth: New evidence from a national survey of older adults. Social Science & Medicine (1982), 125, 94–106. doi: 10.1016/j.socscimed.2013.09.011 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Cornwell, B., Schumm, L. P., Laumann, E. O., & Graber, J. (2009). Social networks in the NSHAP Study: Rationale, measurement, and preliminary findings. Journals of Gerontology, Series B: Psychological Sciences and Social Sciences, 64B(suppl. 1), i47–i55. doi: 10.1093/geronb/gbp042 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Cornwell, B., Schumm, L. P., Laumann, E. O., Kim, J., & Kim, Y.-J. (2014). Assessment of social network change in a national longitudinal survey. Journals of Gerontology, Series B: Psychological Sciences and Social Sciences, 69(suppl. 2), S75–S82. doi: 10.1093/geronb/gbu037 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Fiori, K. L., Windsor, T. D., & Huxhold, O. (2020). The increasing importance of friendship in late life: Understanding the role of sociohistorical context in social development. Gerontology, 66(3), 286–294. doi: 10.1159/000505547 [DOI] [PubMed] [Google Scholar]
- Fischer, C. S., & Beresford, L. (2015). Changes in support networks in late middle age: The extension of gender and educational differences. The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences, 70(1), 123–131. doi: 10.1093/geronb/gbu057 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Fischer, C. S., & Offer, S. (2020). Who is dropped and why? Methodological and substantive accounts for network loss. Social Networks 61, 78–86. doi: 10.1016/j.socnet.2019.08.008 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Goldman, A. W., & Cornwell, B. (2018). Social disadvantage and instability in older adults’ ties to their adult children. Journal of Marriage and the Family, 80(5), 1314–1332. doi: 10.1111/jomf.12503 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Iveniuk, J., Donnelly, P., & Hawkley, L. (2020). The death of confidants and changes in older adults’ social lives. Research on Aging, 42(7–8), 236–246. doi: 10.1177/0164027520917061 [DOI] [PubMed] [Google Scholar]
- Lin, N., Woelfel, M. W., & Light, S. C. (1985). The buffering effect of social support subsequent to an important life event. Journal of Health and Social Behavior, 26(3), 247–263. doi: 10.2307/2136756 [DOI] [PubMed] [Google Scholar]
- Litwin, H., and Levinsky, M. (2020). Always alone? Network transitions among detached older Europeans and their effects. Ageing and Society, 1–15. doi: 10.1017/S0144686X20000240 [DOI] [Google Scholar]
- Litwin, H., Levinsky, M., & Schwartz, E. (2020). Network type, transition patterns and well-being among older Europeans. European Journal of Ageing, 17(2), 241–250. doi: 10.1007/s10433-019-00545-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Marin, A. (2004). Are respondents more likely to list alters with certain characteristics? Implications for name generator data. Social Networks 26(4), 289–307. doi: 10.1016/j.socnet.2004.06.001 [DOI] [Google Scholar]
- Marin, A., & Hampton, K. N. (2019). Network instability in times of stability. Sociological Forum, 34(2), 313–336. doi: 10.1111/socf.12499 [DOI] [Google Scholar]
- Marsden, P. V. (2011). Survey methods for network data. In Scott J. & Carrington P. J. (Eds.), The SAGE handbook of social network analysis (pp. 370–388). Sage. doi: 10.4135/9781446294413.n25 [DOI] [Google Scholar]
- Menkin, J. A., Robles, T. F., Gruenewald, T. L., Tanner, E. K., & Seeman, T. E. (2016). Positive expectations regarding aging linked to more new friends in later life. Journals of Gerontology, Series B: Psychological Sciences and Social Sciences, 72(5), 771–781. doi: 10.1093/geronb/gbv118 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Morgan, D. L., Neal, M. B., & Carder, P. (1997). The stability of core and peripheral networks over time. Social Networks, 19(1), 9–25. doi: 10.1016/S0378-8733(96)00288-2 [DOI] [Google Scholar]
- Paik, A., & Sanchagrin, K. (2013). Social isolation in America: An artifact. American Sociological Review, 78(3), 339–360. doi: 10.1177/0003122413482919 [DOI] [Google Scholar]
- Petersen, J., Kaye, J., Jacobs, P. G., Quinones, A., Dodge, H., Arnold, A., & Thielke, S. (2016). Longitudinal relationship between loneliness and social isolation in older adults: Results from the Cardiovascular Health Study. Journal of Aging and Health, 28(5), 775–795. doi: 10.1177/0898264315611664 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Quan-Haase, A., Mo, G. Y., & Wellman, B. (2017). Connected seniors: How older adults in East York exchange social support online and offline. Information, Communication & Society, 20(7), 967–983. doi: 10.1080/1369118X.2017.1305428 [DOI] [Google Scholar]
- Roth, A. R. (2020). Informal caregiving and network turnover among older adults. Journals of Gerontology, Series B: Psychological Sciences and Social Sciences, 75(7), 1538–1547. doi: 10.1093/geronb/gby139 [DOI] [PubMed] [Google Scholar]
- Schafer, M. H. (2018). Gendered life course transitions: The case of driving cessation and social networks. In Alwin D. F., Felmlee D. H., & Kreager D. A. (Eds.), Social networks and the life course (pp. 245–262). Springer Publishing. doi: 10.1007/978-3-319-71544-5_12 [DOI] [Google Scholar]
- Schafer, M. H., & Koltai, J. (2015). Cancer diagnosis and mental health among older white adults: Moderating role for social networks? Society and Mental Health, 5(3), 182–202. doi: 10.1177/2156869315577631 [DOI] [Google Scholar]
- Schafer, M. H., & Vargas, N. (2016). The dynamics of social support inequality: Maintenance gaps by socioeconomic status and race? Social Forces, 94(4), 1795–1822. doi: 10.1093/sf/sow024 [DOI] [Google Scholar]
- Schwartz, E., & Litwin, H. (2018). Social network changes among older Europeans: The role of gender. European Journal of Ageing, 15(4), 359–367. doi: 10.1007/s10433-017-0454-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- Schwartz, E., & Litwin, H. (2019). Changes in social networks and cognitive decline. In Börsch-Supan A., J. Bristle, K. Andersen-Ranberg, A. Brugiavini, F. Jusot, H. Litwin, & G. Weber (Eds.), Health and socio-economic status over the life course: First results from SHARE waves 6 and 7 (pp. 219–226). De Gruyter. doi: 10.1515/9783110617245-022 [DOI] [Google Scholar]
- Spillman, B. C., Freedman, V. A., Kasper, J. D., & Wolff, J. L. (2019). Change over time in caregiving networks for older adults with and without dementia. Journals of Gerontology, Series B: Psychological Sciences and Social Sciences, 75(7), 1563–1572. doi: 10.1093/geronb/gbz065 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Straits, B. C. (2000). Ego’s important discussants or significant people: An experiment in varying the wording of personal network name generators. Social Networks, 22(2), 123–140. doi: 10.1016/S0378-8733(00)00018-6 [DOI] [Google Scholar]
- Terhell, E. L., van Groenou, M. I. B., & van Tilburg, T. (2007). Network contact changes in early and later postseparation years. Social Networks, 29(1), 11–24. doi: 10.1016/j.socnet.2005.11.006 [DOI] [Google Scholar]
- Wrzus, C., Hänel, M., Wagner, J., & Neyer, F. J. (2013). Social network changes and life events across the life span: A meta-analysis. Psychological Bulletin, 139(1), 53–80. doi: 10.1037/a0028601 [DOI] [PubMed] [Google Scholar]