Exploring the long-term effects of employment on social networks

Work brings people together, thereby shaping their social networks. Building on this premise, Cohn-Schwartz and Naegele (2023) investigated how years of employment and number of jobs affect social network size, type, and emotional closeness by gender. Using data from the Survey of Health, Aging and Retirement in Europe (SHARE), they found that (1) Working for longer years and in more jobs was associated with larger social networks and a higher likelihood of including children and friends within the networks; (2) working for longer years and in more jobs was related to lower emotional closeness with the network, but working for more years among men was related to better emotional closeness with the network; (3) women had larger social networks and were more likely to include friends; and (4) extensive work histories were not related to worse family relationships. A unique contribution of their paper is testing if varying tenure in workplaces serves as a social convoy from which gendered networks are constituted (Wrzus et al., 2013). This is a welcome addition to previous studies that have tended to focus a priori on the social networks themselves or in relation to health outcomes (cf., Smith and Christakis, 2008). These findings have significance for the literature on productive aging, particularly the increasing phenomenon of working in later life, and provide several directions for future research.

While not focusing on employment in later life exclusively, Cohn–Schwartz & Naegele’s sample of individuals aged 50 and over yields insights into the social relationships of respondents who remain working in mid-life and later life. As demographers have observed, the dual forces of lower birth rates and increasing life expectancy have radically altered the age distributions of societies around the world. To address labor force shortages, some societies are promoting late retirement and concomitantly prolonging pension age. In response, scholars of productive aging argue that society must better support older adults who elect to make economic contributions in later life, such as via employment or volunteering (Gonzales et al., 2015). Lower levels of wealth have been associated with a longer need to remain in the labor force, as well as reduced autonomy for selecting the types of transitions to retirement. Conversely, more advantaged statuses have been correlated with diverse patterns of retirement, such as phased retirement (Carr et al., 2021). These observations are confirmed by Madero-Cabib et al. (2015), who also utilized the retrospective job panel in SHARE; they established that late retirement after pension age was associated with accumulated disadvantages across the life course.

In light of these challenges, a significant finding in Cohn–Schwartz & Naegel’s study is that, irrespective of financial ability and other controls, individuals with longer working lives accrued emotionally meaningful relationships. There was a positive association between years in employment and emotional closeness with the social network for men and a positive association between years in employment and having a child confidant for both men and women. Further strengthening these social relationships or targeting their upstream factors in the workplace could counteract the disadvantages that accompany prolonged involvement in the labor market. Granted, this line of thinking has several caveats: It cannot be determined from the study if individuals with more years of working worked mostly prior to or after pension age, and no information about the level of volition for working beyond traditional retirement age was available. A solution would be to stratify the sample by age cohorts and retirement timing; however, as will be discussed later, incorporating characteristics of the type of work (e.g., part-time vs. full-time) will also capture older workers more vulnerable than others. Employment trajectories are certainly not uniform; some are more precarious, unpredictable, and less conducive to forming relationships than others (Shepherd, 2024). Furthermore, the timing of key life events such as marriage and retirement have shown to be significant precursors for subsequent health outcomes (Eisenberg-Guyot et al., 2020).

The unexpected dynamic of spousal relationships vis-à-vis other confidants found in Cohn–Schwartz & Naegele’s study suggest the need for sociocentric network analysis. It was found, for instance, that adults who worked in more jobs were less likely to include their spouses in their social networks. Furthermore, working in more jobs potentially resulted in less emotional closeness with one’s spouse. The use of a name-generator approach may limit the ability to explain these results, because it is not immediately clear if intervening upon the respondent in such cases would positively impact the named confidants as well. While this egocentric approach is more straightforward methodologically and commonly found in the social network literature, it neglects key processes within the broader social network (Ayalon and Levkovich, 2019). By considering which additional alters to include (i.e., the boundary specification problem)—such as mapping a broader network structure through a multi-informant approach—the whole network under study could potentially serve as the target for intervention (Koehly et al., 2015; Laumann et al., 1989). Thus, the question as to why a respondent was less likely to include their spouse in their networks should be addressed through both the respondent’s perspective and the perspectives of his or her alters. Since the SHARE study interviews respondents and their spouse or partner, a spousal dyadic approach is feasible and could yield insights into how joint employment trajectories affect each other’s social network composition. One could assess, for example, the influence of joint work histories on the emotional reciprocity of spousal dyads. To measure other network properties, such as network density, bridging potential, and network homogeneity, the Multinational Time Use Study is an alternative dataset (Cornwell et al., 2015).

With respect to the literature on work histories, future research should seek to distinguish additional types of employment and retirement. Gendered trends between full-time work and part-time work may have a differential effect on social networks. Also, participants who are still engaged in paid work while receiving a pension should be separately classified, given the unique dynamics of partial retirement or bridge employment (Beehr and Bennett, 2015)—although this kind of post-retirement data is not currently available in SHARE. The Job Episodes Panel used in the study, however, does contain information on unemployment spells, which has been found to have gendered effects on network composition and network emotional support (Rözer et al., 2020). Absent this information, an individual with a stable trajectory of full-time work would be analytically identical to an individual with the same number of jobs and total years of work but multiple spells of unemployment between jobs. To minimize bias, researchers should consider following Wahrendorf (2015) in excluding early retirees or those who reported chronic illness during their working ages. Another fruitful avenue is to adopt optimal matching analysis or related sequence alignment techniques to account for the ordering of different work trajectories (Chan, 1995; Liao et al., 2022).

There are some limitations in Cohn–Schwartz & Naegele’s work that should be noted. First, while the study has been framed around post-retirement social networks, it is unclear if the sample included only retirees or both retirees and individuals still in the labor force. Given that the analytical sample consisted of adults aged 50 and over, it is likely that many individuals in the sample have not yet retired. To focus on retirees, one strategy is to limit the sample to respondents who receive a pension income (Dingemans and Möhring, 2019). Second, the authors note that women working until retirement were found to engage in more volunteer activities, but the participants’ volunteering histories were not controlled in the models. Volunteering is more likely to follow employment cessation according to the activity substitution hypothesis (e.g., Carr et al., 2023; Wilson and Musick, 1997), and a large body of work has linked volunteering with social networks (e.g., Putnam, 2000). If controlling for volunteering histories is not possible, it would be helpful to use metrics such as the E-value to measure the strength of an unmeasured confounder (VanderWeele and Ding, 2017). Third, the authors’ concern about reverse causation may very well be an intractable problem due to the lack of time-varying covariates across the employment life course. As a concession, the authors could test if mid-life employment histories (for which panel data are available, e.g., the aged 50–60 cohort) are related to social relationships in later life. This fully exploits the longitudinal nature of the SHARE dataset and permits a wider range of methods to attenuate the risk of reverse causation (e.g., Studer et al., 2018). Finally, to correct for the observed attrition bias in the sample, the individual sampling weights provided by SHARE should be used.

Understanding the life course of employment, from entry to retirement and beyond, is a promising starting point for devising means to strengthen social relationships in later life. An added benefit when studying whole networks as the unit of analysis is that not only would individual older adults benefit, but as supradyadic studies show, the effects may spread from person to person (e.g., Christakis and Fowler, 2007; Valente and Davis, 1999). Formulating subgroups by age cohorts, as argued above, will be essential for learning when along the life course interventions can be most effective. Looking ahead, artificial intelligence, with its rapid evolution across multiple sectors, may exacerbate generational differences in employment patterns, and its effects should be closely monitored. Thanks to Cohn–Schwartz & Naegele, we now know how to proceed.