Computers in Human Behavior 153 (2024) 108103 Available online 22 December 20230747-5632/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). Social media contact with family members and happiness in younger and older adults Kristiina Tammisalo a,*, Mirkka Danielsbacka a,b, Antti O. Tanskanen b,c, Bruno Arpino d a INVEST Research Flagship Centre, University of Turku, Turku, Finland b Population Research Institute, Vaestoliitto, Helsinki, Finland c Department of Social Research, University of Turku, Turku, Finland d Department of Statistical Science and Department of Philosophy, Sociology, Education and Applied Psychology, University of Padua, Padua, Italy A R T I C L E I N F O Handling Editor: Dr. Catalina L Toma Keywords: Digital divide Family Happiness Social media Socioemotional selectivity theory Strong ties A B S T R A C T Prior literature suggests that social media may increase older adults’ well-being because they use social media to communicate with close social ties, such as family members. Despite this potential being recognized and asso- ciations between social media and well-being being established among older adults, the role of social media contact (SMC), particularly with family members, as a source of well-being has been little explored. In this study, we tested whether SMC with family members increases happiness for older (68–74 year-old) and younger and middle-aged (19–56 year-old) adults. Using population-based data from Finland, we examined the extent to which self-rated happiness is dependent on the respondents’ sustenance of SMC with given family members. This analysis involved 2807 social media users. The family members examined were children and siblings for both generations; in addition, grandchildren for the older generation and parents for the younger generation. Pro- pensity score matching was used to improve the credibility of the estimated associations. Partially supporting the socioemotional selectivity theory (SST), the results show that the older generation has more SMC with family members than the younger generations does; however, SMC with family members did not increase happiness among older adults, as SST suggests. Conversely, for younger adults, SMC with some family members was associated with increased happiness, suggesting that social media may gratify some family-related needs more specific to middle-aged adults than to older adults. This study adds to the knowledge about life-stage specific factors that contribute to subjective well-being in the digital era. 1. Introduction Social media platforms such as Facebook and Instagram increase the overall potential for communication by overcoming time and location barriers. These platforms allow users to communicate more spontane- ously and to more than one recipient at a time which can complement other forms of communication (Danielsbacka, Tammisalo, & Tanskanen, 2023; Dienlin, Masur, & Trepte, 2017). Social media may also gratify a range of idiosyncratic needs in that it provides a wide range of communication affordances, compared to traditional methods of communication (Taipale, 2019). Since social media can complement communication in close relationships, some scholars have considered social media’s potential as an enhancer of individuals’ well-being (Chan, 2015, 2018; Dienlin et al., 2017; Gazit, Nisim, & Ayalon, 2022; Ishii, 2017; Newman, Stoner, & Spector, 2021; Rosenberg & Taipale, 2022; Vriens & van Ingen, 2018). Numerous studies have found social media to enhance well-being particularly among older adults (see review by Newman et al., 2021). For instance, social media is associated with the preservation of personal social capital in older age (Simons et al., 2021) as well as various measures of well-being, including life satisfaction (Kim & Shen, 2020; Rosenberg & Taipale, 2022) and lower levels of social isolation and loneliness (Aarts, 2018; Gazit et al., 2022). However, not all studies have found positive effects of social media (Newman et al., 2021), emphasizing the need for improved under- standing of the circumstances in which social media affects well-being and the mechanisms of the effect. One significant factor that de- termines how social media affects users’ well-being is the composition of the online network. For example, the proportion of strangers (Seabrook, Kern, & Rickard, 2016) and that of contacts perceived as actual friends * Corresponding author. E-mail address: kristiina.a.tammisalo@utu.fi (K. Tammisalo). Contents lists available at ScienceDirect Computers in Human Behavior journal homepage: www.elsevier.com/locate/comphumbeh https://doi.org/10.1016/j.chb.2023.108103 Received 9 May 2023; Received in revised form 8 December 2023; Accepted 14 December 2023 Computers in Human Behavior 153 (2024) 108103 2 (Chang, Choi, Bazarova, & Lockenhoff, 2015) in a user’s social media network has been found to moderate the effect of social media on well-being. In the case of older adults, some have considered that essential for well-being is the possibility to sustain online contact with family members, in particular (Arpino, Pasqualini, Bordone, & Sole-Auro, 2021; Gazit et al., 2022). However the effect of family communication via social media has been little explored using rigorous quantitative methods and representative data. In this study, we explored the extent to which older and younger adults use social media to interact with family members and whether social media contact (SMC) with family members increases happiness. 1.1. Age-related changes in social networks Older adults’ social needs typically differ from those of younger adults. For example, it is common for older adults to face loss of mobility or diminishing social networks due to retirement and/or bereavement. These occurrences affect older adults’ social needs and, consequently, their communication patterns and media use for communication (David-Barrett et al., 2016; Rosenberg & Taipale, 2022). The question arising is whether social media gratifies these needs over and above the previously available methods of communication, such as face-to-face meetings and phone calls; that is, whether older adults supplement their communication needs with social media. Notably, our focus is not on specific activities on social media plat- forms (e.g., posting content, reacting to content, direct messaging, etc.), even though various activities can have distinct effects on well-being (Kim & Shen, 2020; Seabrook et al., 2016; Verduyn, Ybarra, Resibois, Jonides, & Kross, 2017). Rather, our focus is on whether social media platforms in their versatility offer complementary benefits to users of different ages in the context of family communication that are not accrued through other communication methods. As individuals age, their relationships, particularly with family, gain relative importance (David-Barrett et al., 2016; Wrzus, Hanel, Wagner, & Neyer, 2013). Among typically non-resident family members, the relationships that are the closest and more enduring across life transi- tions than non-family relationships are those between siblings, parents and adult children, and grandparents and grandchildren (Buchanan & Rotkirch, 2021; Roberts & Dunbar, 2011; Tanskanen & Danielsbacka, 2019). This age-related “family shift” also applies to social media net- works; that is, older adults use social media to connect with family members more than younger adults (Hutto et al., 2015; Matassi, Bocz- kowski, & Mitchelstein, 2019; Tammisalo, Danielsbacka, Andersson, & Tanskanen, 2022). Moreover, abundant evidence indicates that for older adults, family members are the primary reason for using social media. Not only does the number of children predict social media adoption (Tammisalo et al., 2022), older adults’ most frequent self-appointed reason for adopting social media is to connect with their children and grandchildren (Aarts, 2018; Matassi et al., 2019; Neves, Franz, Judges, Beermann, & Baecker, 2017; Newman et al., 2021) and they perceive social media to facilitate family interactions (Taipale, 2019; Taipale & Farinosi, 2018. 1.2. Theoretical considerations Some studies (Chan, 2015, 2018; Chang et al., 2015) have drawn on the socioemotional selectivity theory (SST) to explain older adults’ tendency to communicate with a smaller network of family and close friends. According to SST, with advancing age comes a motivational shift, from future-oriented and informational goals to emotionally gratifying goals, prompted by the realization of finite time towards the end of life (Carstensen, Isaacowitz, & Charles, 1999; Lockenhoff & Carstensen, 2004). It is well established that older adults selectively communicate with fewer close ties (Wrzus et al., 2013) and do so even in the social media environment (Chang et al., 2015; Newman et al., 2021; Tammisalo et al., 2022). However, whether older adults derive happi- ness from SMC with their family members via social media, as can be expected based on SST, has yet to be established. Drawing on the SST – although not focusing specifically on family relationships – two studies from Hong Kong have considered the role of social network composition in the association between multimodal communication (including social media) and well-being (Chan, 2015, 2018). These studies suggest that social media’s benefits to older adults are attributed to age-related changes in the proportion of strong ties (including close friends and family members) in their networks. In their findings, older age groups experienced positive emotions from using technology to connect with their strong ties, whereas younger age groups did not. Conversely, younger adults experienced a decrease in well-being from using media to connect with a larger network of weak ties (i.e., non-intimate friends and acquaintances). These findings sug- gest that age-related changes in network composition are important in determining how social media use affects well-being. Another related study (Chang et al., 2015), which also draws on the SST, found that smaller overall networks and a higher proportion of actual friends in the online network were more typical for older adults, and these were associated with lower levels of social isolation and loneliness. Similar to previous investigations, this study also supports the SST, that is, that social media’s effect on well-being is age-related, as explained by age-related changes in network composition. A notable gap in the research is that family relationships have not been examined separately (Chan, 2015, 2018; Chang et al., 2015) nor have different relationship types in the family. A separate examination of family relationship types is necessary, considering that each type of family relationship has distinct dynamics and new communication technologies can affect each relationship type differently (Tammisalo & Rotkirch, 2022). In addition to treating each relationship type separately, we acknowledge that there may be differences between genders and life stages in terms of typical social needs and the extent to which social media can gratify those needs. Examining genders and ages separately is justified from the perspective of digital divides as the literature has established that social media predominantly serves the typical needs of women and younger individuals (Newman et al., 2021; Robinson et al., 2015; Yu, Ellison, McCammon, & Langa, 2016). Interestingly, the family context has hardly been addressed in the digital divide literature (Lythreatis, Singh, & El-Kassar, 2022; Scheerder, van Deursen, & van Dijk, 2017). The relevance of family relationships for digital divides, however, is evident: family resources may be differentially accessible depending on how well technology serves to gratify the specific and typical family communication needs for men versus women and for in- dividuals in different life stages. On the other hand, social media may also mitigate divides as it can potentially compensate for the scarcity of offline social contacts. For example, as social networks decline with age, older adults may benefit from the affordable and efficient access to social support provided by social media (Hutto et al., 2015; Nguyen, Hunsaker, & Hargittai, 2022; Simons et al., 2021). To detect age- and gender-based divides, we examined SMC in a range of family relationships separately for young, middle-aged, and older adults and by gender (e.g., mother-daughter, mother-son, father-daughter etc.). To the best of our knowledge, the present study is the first to consider age-variant associations between SMC with family members and happiness. While studies on social media and various well-being out- comes have been extensive, they have excluded older adults or studied different age groups in isolation. To address this limitation, we used a dataset comprising a large age range (an older generation of 68–74-year- olds and their 19–56-year-old adult children) to investigate whether SMC with family members is associated with happiness differentially depending on age and gender. In addition, the data allowed controlling for a wide range of confounders. Existing studies on social media and well-being have rarely controlled for social network characteristics K. Tammisalo et al. Computers in Human Behavior 153 (2024) 108103 3 (number of close friends and family) (Newman et al., 2021) even though they are associated with both social media adoption (Tammisalo et al., 2022) and various positive well-being outcomes (Chan, 2015, 2018; Chang et al., 2015). We also controlled for the sociodemographic characteristics that are commonly controlled for in related studies that have investigated the associations between social media use and well-being (Newman et al., 2021). The control variables were age, gender, education, income, marital status, and health status. 1.3. Research questions Based on theoretical considerations and evidence from earlier studies, we formulated the following research questions: (Q1) To what extent do older and younger adults use social media for family communication? (Q2) Is the association between SMC with family members and happiness different in the older versus younger generation? Based on SST we can predict in the case of Q1 that the older gener- ation has more SMC with family members than the younger generation and in the case of Q2 that SMC with family increases happiness more in the older than younger generation. 2. Data and methods 2.1. Data We used the third wave of the Generational Transmissions in Finland (Gentrans) survey conducted in 2018 and 2019. The Gentrans survey gathers data about the living conditions, social networks, and well-being of Finnish adults. The third wave of the survey includes information about communication with kin via social media, as well as a range of sociodemographic background information. The data comprise two family generations: the older generation represents Finnish baby boomers born in the period 1945–1950 (n ˆ 2,663, response rate 66.4 %) and the younger generation represents the baby boomers’ adult children born in the period 1962–1999 (n ˆ 1,945, response rate 55.6 %). The data comprise a nationally and geographi- cally representative sample from Finland (excluding Åland). During the data collection period, the older and younger generation respondents were aged 68–74 years and 19–56 years, respectively. As the age range among the younger generation’s respondents is wide, we further divided them into 19–39-year-olds (young adults) and 40–56-year-olds (middle- aged adults) for further analyses, in addition to treating them as one age group in the main analyses. To investigate the association between SMC with family members and happiness, we excluded from the main ana- lyses those who were not social media users; the final sample sizes were 1,265 in the older generation and 1,542 in the younger generation (475 younger adults and 1,045 middle-aged adults). The analyses of SMC with each family member type (daughter, son, mother, father, grand- child, sister, and brother) involved only respondents who had the family member in question (sample sizes for each analysis are provided in the online supplementary material). Ethical approval for the Gentrans survey was granted by the Ethical Board of Statistics of Finland in 2006. The respondents provided informed consent regarding the use of their data for research and the data were anonymized. All users of the data committed to following the Statistics Finland ethical rules by accepting The Pledge of Secrecy of Holder of Permission to Use Data. 2.2. Measures Our dependent variable is based on the question: “How happy are you currently?”, with response options ranging from 0 (“Very unhappy”) to 10 (“Very happy”). The independent variables are based on the item: “With whom are you in contact with on social media?”, followed by a list of kin, affinal kin, and non-kin from which the respondents indicated with whom they interacted via social media. Examples of social media platforms given in the questionnaire were: Facebook, Instagram, Twitter, and Snapchat. In the absence of a frequency measure for SMC, we used dichotomous variables; similar variables have been used in previous studies (e.g., Yu, Mccammon, Ellison, & Langa, 2016). Based on the responses, we used the following indicators of SMC with family: 1) SMC with a daughter (yes/no) 2) SMC with a son (yes/no) 3) SMC with mother (yes/no, only for younger generation) 4) SMC with father (yes/no, only for younger generation) 5) SMC with a grandchild (yes/no, only for older generation) 6) SMC with a sisters (yes/no) 7) SMC with a brothers (yes/no) To determine whether and how SMC was associated with happiness across the types of family members, we controlled for a range of socio- demographic variables. The following sociodemographic variables were used as controls in all of the analyses: Age was calculated according to the year of birth. Gender was collected as a binary variable (man/ woman). Education was collected as a nine-level ordinal variable ranging from incomplete elementary school to doctorate degree and subse- quently recoded into a four-level ordinal variable from 1 ˆ completed or incomplete elementary school to 4 ˆ higher education. Self-rated eco- nomic situation was measured on a four-level ordinal scale ranging from 1 ˆ high income to 4 ˆ low income and subsequently inverted for analysis so that 4 indicates the highest income level. Marital status was collected as a categorical variable with six response options and subse- quently recoded into the following categories: 1 ˆ unpartnered, 2 ˆ cohabiting/married, 3 ˆ divorced, and 4 ˆ widowed. Finally, to control for Health, we included a dichotomous variable with 1 ˆ absence of long-term medical issues and 0 ˆ presence thereof. In addition, the control variables included two variables that captured the characteris- tics of respondents’ social networks: Number of children and Number of close friends. These variables were included because having close friends and relatives may affect both the independent variable (SMC) and dependent variable (happiness). Outlier values of 21 and above were set to 20. We did not include as confounders whether the respondent lived with the family member in question because, based on the data, less than 1 percent of the respondents did. Similarly, we did not control for co- residence with a non-relative other than a spouse, as less than 1 percent indicated that they lived with a non-relative. In Finland, it is relatively rare for adults to live with other relatives besides their own underage children or other non-relatives besides their spouses (Hamalainen, Danielsbacka, Hagglund, Rotkirch, & Tanskanen, 2021). A sensitivity analysis showed that co-residence with a child was not a significant confounder and did not affect the results. Therefore, even for the younger generation, many of whom are still likely to live with their children, we did not add co-residence with family members as a control variable. Conversely, co-residence with spouses was 99 percent congruent with marital status (option married/cohabiting); therefore, co-residence with spouses was not controlled for separately. In addition, the ages of daughters and sons in the younger age group and those of grandchildren in the older age group were added as controls in the corresponding analyses (i.e., SMC with daughter/son/grandchild) because some of the respondents’ children and grandchildren were less likely to use social media because of their young age. Age was approx- imated using available information. For children, the mean age of the four eldest children was used, and for grandchildren, the mean age of the youngest among the four eldest children was used. K. Tammisalo et al. Computers in Human Behavior 153 (2024) 108103 4 2.3. Methods Three estimates of the associations between SMC with family mem- bers and happiness are presented for each family member type: linear regressions without controls, linear regressions with controls (presented in detail previously), and the propensity score matching (PSM) approach (excluding the analyses in which sample sizes did not allow for PSM). In the linear regressions with controls, the control variables were included without interactions or higher-order terms. The same control variables were used in the linear regressions and PSM approach. These three methods were used to understand whether any association exists be- tween SMC with a family member and happiness, and whether the as- sociation concerns selection (i.e., whether those who are happy because of factors accounted for by the control variables are also more likely to engage in SMC with family members). After the regressions, the PSM approach was used to improve the credibility of the causal inference. 2.3.1. Propensity score matching PSM is a quasi-experimental method which is based on retrospec- tively forming comparable groups (“treatment” and “control”) from existing data. The “treatment group” comprises those who have SMC with a given family member, and the “control group” comprises those who do not have SMC with that given family member. PSM is conducted in two steps: First, to form comparable groups, a propensity score is calculated for each individual using the control vari- ables. The score is an estimate of the probability of an individual having SMC with a given family member, given the observed control variables. As in most applications, we used logit models to estimate the propensity scores. Observed variables that influence the probability of adopting social media and happiness (i.e., possible confounders) are included as controls: therefore, they contribute to the propensity score. After esti- mating the propensity scores for each individual, individuals with equivalent or similar scores in the treatment (SMC with a given family member) and control groups (no SMC with a given family member) were matched. Unmatched individuals were excluded from the analysis. In the second PSM step, outcomes are compared between matched individuals from the treatment and control groups. This step is non- parametric because it is based on the differences between outcomes without imposing a parametric model linking the outcome and treat- ment (and covariates). In this study, PSM allows us to estimate the counterfactual happiness of the respondents had they belonged to the other group by matching respondents from the treatment and control groups with equivalent or similar propensity scores values. The PSM results are presented in terms of average treatment effects on the treated (ATT). For PSM, all age variables (excluding the ages of the older genera- tion’s respondents whose age range was relatively narrow) were recoded into 5-year-intervals to guarantee finer matching of the age distribution. 2.3.2. Advantages of propensity score matching In addition to using the PSM approach, we also performed linear regressions, as most readers are more familiar with this method. Regression models have also typically been used in previous studies (Chan, 2015, 2018; Yu, Mccammon, et al., 2016). Both PSM and linear regression are based on the fundamental assumption that, to identify the effect of interest, adjusting for observed confounders is sufficient. However, the adjustment is performed differently in the two methods, and PSM has several advantages. First, PSM allows the evaluation of the extent to which the adjustment has been “successful” by assessing the covariate balance. The most commonly used measure of covariate bal- ance is the Absolute Standardized Bias (ASB). The causal inference literature suggests that ASB values below 5 percent are sufficiently low and can estimate effects (e.g., Cannas & Arpino, 2019). Another advantage is that PSM is non-parametric in its second step when out- comes are compared across treatment groups, whereas regression models impose a parametric function linking the outcome to the treatment and other independent variables. The non-parametric nature of the second step is an advantage also because of the possibility that the effects are heterogeneous. Regression models inherently assume ho- mogeneity of the effects of independent variables; therefore, they do not consider heterogeneity unless interactions are included (whereby het- erogeneity is limited to those specific interactions). Conversely, in PSM, heterogeneity is not limited in the non-parametric outcome assessment. This further explains why regression and PSM can provide different results. Methodological studies have found that regression models and pro- pensity score methods often provide similar results when the initial imbalance of covariates across the groups is low (Drake, 1993). In this case, a possible violation of regression assumptions (e.g., the functional form imposed to link the outcome to the treatment and covariates) does not have serious consequences on the bias of the estimated effects. Instead, when the two groups differ significantly, regression and PSM are expected to yield different results. When PSM and regression give different results (and the balance attained by PSM is sufficient), then PSM results are more reliable because the differences in covariates across groups are adjusted better (Cannas & Arpino, 2019) and because the second step of PSM is non-parametric, thus not subject to potential violations of the regression assumptions. All control variables’ ASBs for each analysis are presented in the online supplementary material (Tables 5–15). In the notes for the tables that present the main results (Tables 3 and 4), we specify the few cases where the average value of ASB remains higher than 5 percent. 3. Results 3.1. Descriptive findings In the older and younger generations, 47.5 percent and 79.3 percent of the respondents, respectively, reported using social media. Tables 1 and 2 present the descriptive statistics of social media users in the older and younger generations, respectively. The prevalence of SMC with family members varied across genera- tions (Figs. 1 and 2). In the older generation, SMC with children was high: 87.1 percent of those with at least one daughter had SMC with a daughter, and 78.2 percent of those with at least one son had SMC with a son. The corresponding percentages among the younger generation were 32.9 percent (SMC with daughter) and 27.1 percent (SMC with son). For Table 1 Descriptive statistics. Social media users in the older generation (68–74 years) (n and %/mean). n %/mean SD Age (mean) 1265 70.4 1.67 Gender (%) Women 802 63.4 – Men 463 36.6 – Education (%) Elementary school 333 27.1 – Upper secondary/Vocational 604 49.1 – College/Lower university degree 115 9.3 – Higher academic degree 178 14.5 – Economic situation1(mean) 1265 1.81 0.80 Marital status (%) Married/Cohabiting 888 71.1 – Unmarried 68 5.4 – Divorced 160 12.8 – Widowed 133 10.6 – Health (long-term medical issue) (%) Yes 767 60.6 – No 470 37.2 – Number of children (mean) 1265 2.2 1.59 Number of grandchildren (mean) 1265 3.6 5.10 Number of friends (mean) 1265 5.5 3.95 1 1 ˆ “low income,” 4 ˆ “high income”. K. Tammisalo et al. Computers in Human Behavior 153 (2024) 108103 5 the younger generation, 35.9 and 20.5 percent had SMC with their mothers and fathers, respectively. In the younger generation, friends and coworkers were the most typical SMCs. Overall, older adults’ social media use could be described as family oriented. 3.2. Linear regressions and PSM approach Using linear regression and PSM, we tested the two generations separately and by gender to check whether SMC with family members was associated with happiness. We tested this separately for contact with each kin type: daughters, sons, sisters, and brothers for both gen- erations; grandchildren for the older generation; and mothers and fa- thers for the younger generation. We analyzed each family member type separately rather than integrating them into one model, because the samples differed and because of possible collinearity among several SMC variables. Respondents who reported using social media were analyzed, and each analysis was performed only on respondents with the family member in question. Rather than integrating age and gender in the same models, we analyzed them in separate models to obtain separate esti- mates for each. More specifically, we have 2  3  5 ˆ 30 and 2  3  6 ˆ 36 models for the older and younger generations, respectively: linear regression with and without controls (2), men and women separately and together (3), and five and six types of relatives with whom SMC is sustained for the older and younger generations, respectively. Given the large number of models, we only present the results of interest, which is the estimated coefficient (and corresponding 95% confidence interval) in Tables 3–6. We also show the R2 of each regression model (i.e., R2 corresponds to the model used to estimate a coefficient) in parentheses. Table 3 presents the results for the older generation, Table 4 those for the younger generation, and Tables 5 and 6 those for the younger gen- eration divided into younger and middle-aged adults, respectively. For PSM, the balances across the control variables for the matched and unmatched samples are provided in the online supplementary ma- terial along with the specific matching algorithm used in each analysis (Tables 5–15 in the online supplementary material). In addition, sample sizes for each of the analyses are also provided in the online supple- mentary material (Tables 1–4 in the online supplementary material). 3.3. Older generation SMC with family members did not contribute to happiness among older adults; therefore, the analysis results were not as expected (Table 3). The only exception is men’s SMC with their daughters; however, this association was detected only using the linear regression with controls and with statistical significance at the 10 percent level (⠈ 0.516, p ˆ 0.081). Furthermore, the PSM method did not confirm the statistical significance of this association. In addition, women’s SMC with a brother showed a negative association in the linear regressions with controls, although also statistically significant only at the 10 percent level (⠈ 0.282, p ˆ 0.06). However, the PSM method did not Table 2 Descriptive statistics. Social media users in the younger generation (19–56 years) (n and %/mean). n %/mean SD Age (mean) 1542 42.0 5.85 Gender (%) Women 1071 85.5 – Men 471 68.0 – Education (%) Elementary school 27 1.8 – Upper secondary/Vocational 541 35.3 – College/Lower university degree 449 29.3 – Higher academic degree 516 33.7 – Economic situation1(mean) 1542 2.00 0.75 Marital status (%) Married/Cohabiting 1168 75.7 – Unmarried 231 15.0 – Divorced 137 8.9 – Widowed 6 0.4 – Health (long-term medical issue) (%) Yes 609 39.8 – No 921 59.7 – Number of children (mean) 1542 1.8 1.49 Number of friends (mean) 1542 4.3 3.27 1 1 ˆ “low income,” 4 ˆ “high income”. Fig. 1. Percentage of social media users in the older generation (aged 68-74) who sustain SMC with kin and non-kin. In the case of relatives, the number of re- spondents with the given relative is in parentheses. K. Tammisalo et al. Computers in Human Behavior 153 (2024) 108103 6 confirm the statistical significance of this association either. 3.4. Younger generation Conversely, in the younger generation, there were several age- and gender-specific significant associations (p < 0.05) between SMC with family members and happiness. Table 4 shows the results for the entire younger generation, including the results from the linear regressions and PSM approaches. PSM detected that SMC with daughters (ATT ˆ 0.371, p ˆ 0.045) and mothers (ATT ˆ 0.174, p ˆ 0.045) increase happiness. In addition, when examining the results by gender, positive associations that were significant at higher levels (p ˆ 0.05–0.10) suggested that the aforementioned positive associations were driven by women’s SMC with their daughters (ATT ˆ 0.429, p ˆ 0.055) and men’s SMC with their mothers (ATT ˆ 0.289, p ˆ 0.087). Regressions with the controls showed similar results. As an additional and relatively consistent result among men across generations, SMC with many of the tested family members was posi- tively associated with happiness before controlling for covariates (Table 4). This indicates that happy men are more likely to have SMC with family members. Tables 5 and 6 show the results for the younger generation, divided into younger (19–39-year-olds) and middle-aged (40–56-year-olds) adults. These analyses were based only on linear regression models Fig. 2. Percentage of social media users in the younger generation (aged 19–56) who sustain SMC with kin and non-kin. In the case of relatives, the number of respondents with the given relative is in parentheses. Table 3 Estimated coefficients for SMC with family members and self-rated happiness (1 ˆ very unhappy, 10 ˆ very happy) in the older generation (68–74 years). Each coefficient represents a unique model. The controls referred to in the table but not shown are: age, education, income, marital status, health, number of children, and number of close friends. Where genders are analyzed together, gender is controlled for. The mean age of grandchildren is controlled for in the respective model. Regression without controls Regression with controls PSM1 SMC with/by Coef. (R2) 95% CI Coef. (R2) 95% CI Coef. 95% CI2 Daughter/All 0.343* (0.004) 0.009 0.695 0.133 (0.091) 0.235 0.500 0.062 0.348 0.472 /Women 0.151 (0.001) 0.321 0.622 0.002 (0.100) 0.473 0.476 0.1323 0.476 0.741 /Men 0.599** (0.015) 0.063 1.135 0.516* (0.171) 0.064 1.097 0.3113 0.660 1.281 Son/All 0.048 (0.000) 0.247 0.343 0.060 (0.088) 0.369 0.248 0.129 0.419 0.160 /Women 0.002 (0.000) 0.376 0.371 0.119 (0.078) 0.513 0.274 0.097 0.484 0.289 /Men 0.137 (0.001) 0.346 0.621 0.140 (0.142) 0.367 0.646 0.253 0.320 0.827 Grandchild/All 0.145 (0.002) 0.079 0.369 0.161 (0.070) 0.094 0.416 0.117 0.178 0.413 /Women 0.051 (0.000) 0.249 0.351 0.214 (0.085) 0.133 0.552 0.194 0.231 0.618 /Men 0.338* (0.011) 0.006 0.682 0.095 (0.121) 0.279 0.468 0.016 0.389 0.421 Sister/All 0.252** (0.005) 0.020 0.484 0.060 (0.092) 0.182 0.302 0.006 0.251 0.262 /Women 0.175 (0.002) 0.117 0.468 0.043 (0.081) 0.257 0.344 0.050 0.240 0.339 /Men 0.345* (0.009) 0.048 0.738 0.144 (0.170) 0.267 0.556 0.193 0.284 0.669 Brother/All 0.033 (0.000) 0.259 0.194 0.160 (0.092) 0.396 0.076 0.163 0.422 0.097 /Women 0.112 (0.001) 0.397 0.172 ¡0.282*(0.091) 0.577 0.012 0.276 0.595 0.043 /Men 0.110 (0.001) 0.267 0.487 0.019 (0.167) 0.381 0.419 0.056 0.389 0.500 Notes: 1Coefficients are the average treatment effects on the treated (ATT). 2Bootstrapped confidence intervals. 3PSM sample mean bias >5 % (detailed information on sample biases available in the online supplementary material). Significance levels: *p < 0.1, **p < 0.05, ***p < 0.01. K. Tammisalo et al. Computers in Human Behavior 153 (2024) 108103 7 because the sample sizes did not allow the implementation of the PSM method in these groups. Based on these additional analyses, in which the younger genera- tion’s sample was separated into younger and middle-aged adults, the results acquired previously using the full sample appeared to be mostly driven by the older half of the sample. However, the limited sample sizes, especially in some analyses of the younger half of the sample, did not allow for further interpretation (see the online supplementary ma- terial for the sample sizes in each analysis). 4. Discussion In this study, we first asked to what extent older and younger adults use social media for family communication. Second, we asked whether this contributes to happiness differentially depending on generation (younger generation aged 19–56; older generation aged 68–74). In line with previous knowledge (David-Barrett et al., 2016; Hutto et al., 2015; Matassi et al., 2019; Newman et al., 2021; Tammisalo et al., 2022; Wrzus et al., 2013; Zickuhr & Madden, 2012), we found that older adults’ social media use was more family oriented than that of younger adults. In addition, we found evidence that SMC with family members increased happiness in some family dyads, however, not in those of older adults. This was unexpected because numerous earlier studies of older adults’ technology use have found social media use to be associated with various positive outcomes for older adults (Chan, 2015, 2018; Chang et al., 2015; Gazit et al., 2022; Hutto et al., 2015; Newman et al., 2021; Yu, Mccammon, et al., 2016). 4.1. Theoretical discussion The socioemotional selectivity theory (SST) posits that older adults selectively communicate with a smaller number of contacts and that they derive positive emotions from communicating with close ties such as family members. We conclude that, in the context of social media, the socioemotional selectivity theory (SST) was supported only partly. In line with SST, we found that older adults have proportionally more family members in their social media contacts compared to younger adults. However, our study did not find significant associations between SMC with family members and happiness among older adults, as the theory would suggest. In contrast, associations between SMC with family members and happiness were detected in the younger generation for some of the tested family dyads, including middle-aged mothers and their children as well as middle-aged men and their mothers. Notably, earlier studies report varying ages at which the selective communication predicted by SST occurs. Some studies have reported that the so-called family shift and well-being benefits occur in middle adulthood as early as after 35 years (Chan, 2015, 2018; Wrzus et al., 2013). For example, middle-aged individuals, who commonly have both dependent children and aging parents, may benefit from social media because such platforms provide a low-effort means of staying updated on the affairs of multiple family members (Robinson et al., 2015). In line with these considerations, our results suggest that social media may best cater to midlife-specific needs. Regarding older adults, while communication with family members is important for them, social media may be a suboptimal medium for it, as it lacks many qualities of rich and intimate communication such as voice tones and synchronized effort (Daft & Lengel, 1986; Good- man-Deane, Mieczakowski, Johnson, Goldhaber, & Clarkson, 2016; Tammisalo & Rotkirch, 2022). On the other hand, social media often provide instant messaging features that are able to create a synchronous and personal sense of “connected presence” (Cui, 2016). The use of instant messaging has been found to support well-being among older adults (Rosenberg & Taipale, 2022) as has the use of group-chat features that have recently risen in popularity (Gazit et al., 2022; Taipale & Farinosi, 2018, pp. 532–546). Notably these findings both included data from Finnish older adults, which emphasizes the importance of differ- entiating between types of social media activities that are common among the studied population. Considering that our data did not high- light instant messaging or group chats, it is possible that the respondents did not perceive the question as representing these activities. Future research should use refined measures that capture the more specific ways in which older adults use social media. As all individuals included in our analyses were social media users and social media use is associated with other forms of contact with family such as face-to-face meetings and phone calls (Danielsbacka et al., 2023; Shen et al., 2017), older adults may benefit emotionally from family communication using other means and not gain additional benefits from SMC. In other words, it is possible that our sample rep- resents those with above-average levels of contact with their families. Furthermore, it is important to acknowledge that our sample may mask Table 4 Estimated coefficients for SMC with family members and self-rated happiness (1 ˆ very unhappy, 10 ˆ very happy) in the younger generation (19–56 years). Each coefficient represents a unique model. The controls referred to in the table but not shown are: age, education, income, marital status, health, number of children, and number of close friends. Where genders are analyzed together, gender is controlled for. The ages of daughters/sons are controlled for in their respective models. Regression without controls Regression with controls PSM1 SMC with/by Coef. (R2) 95% CI Coef. (R2) 95% CI Coef. 95% CI2 Daughter/All 0.020 (0.000) 0.202 0.242 0.278** (0.182) 0.020 0.535 0.371** 0.008 0.733 /Women 0.028 (0.000) 0.230 0.286 0.310** (0.187) 0.001 0.619 0.429* 0.010 0.867 /Men 0.005 (0.000) 0.442 0.432 0.098 (0.209) 0.388 0.583 0.174 0.602 0.950 Son/All 0.015 (0.000) 0.238 0.207 0.204 (0.184) 0.053 0.460 0.202 0.129 0.533 /Women 0.063 (0.000) 0.316 0.188 0.237 (0.165) 0.060 0.534 0.328 0.147 0.802 /Men 0.114 (0.001) 0.342 0.569 0.072 (0.270) 0.439 0.584 0.1663 0.812 0.480 Mother/All 0.270*** (0.006) 0.086 0.454 0.182** (0.196) 0.012 0.351 0.174** 0.004 0.345 /Women 0.150 (0.002) 0.058 0.358 0.134 (0.152) 0.063 0.332 0.110 0.090 0.310 /Men 0.580*** (0.022) 0.216 0.944 0.205 (0.272) 0.124 0.533 0.289* 0.042 0.621 Father/All 0.281** (0.004) 0.050 0.512 0.070 (0.189) 0.144 0.283 0.041 0.146 0.229 /Women 0.078 (0.000) 0.176 0.333 0.032 (0.112) 0.277 0.213 0.049 0.290 0.192 /Men 0.739*** (0.024) 0.276 1.203 0.129 (0.311) 0.281 0.539 0.155 0.200 0.511 Sister/All 0.358*** (0.007) 0.088 0.627 0.154 (0.207) 0.094 0.403 0.104 0.170 0.377 /Women 0.094 (0.001) 0.223 0.411 0.015 (0.146) 0.288 0.319 0.010 0.319 0.339 /Men 0.682*** (0.025) 0.175 1.189 0.340 (0.353) 0.094 0.775 0.398 0.136 0.931 Brother/All 0.107 (0.001) 0.128 0.343 0.077 (0.189) 0.140 0.295 0.050 0.212 0.312 /Women 0.007 (0.000) 0.254 0.268 0.019 (0.149) 0.232 0.269 0.046 0.233 0.326 /Men 0.435* (0.011) 0.063 0.933 0.130 (0.309) 0.306 0.567 0.082 0.372 0.536 Notes: 1Coefficients represent the average treatment effects on the treated (ATT). 2Bootstrapped confidence intervals. 3PSM sample mean bias >5 % (detailed in- formation on sample biases available in the online supplementary material). Significance levels: *p < 0.1, **p < 0.05, ***p < 0.01. K. Tammisalo et al. Computers in Human Behavior 153 (2024) 108103 8 trends that apply to subgroups in specific life circumstances; for example, the benefits of social media may be more salient when it compensates for the lack of other communication options (Arpino, Meli, Pasqualini, Tomassini, & Cisotto, 2022). In terms of digital divides, which posit that technology use and its benefits are unequally distributed across demographics (Robinson et al., 2015; Yu, Ellison, et al., 2016), our results reveal some marked gender differences. Despite what the digital divide literature has established in contexts other than family, we did not find that women are more likely to benefit from social media in terms of family communication (Rob- inson et al., 2015; Yu, Ellison, et al., 2016). While we did find that among middle-aged women there was a positive association between SMC with their children and happiness, we also found that SMC with mothers was associated with happiness among middle-aged men. Although the results do not show consistent gender differences, they are in line with the notion that the mother-child bond is key to well-being, and that even in the context of social media communication, mothers form nodal points in family communication (Bracke, Christiaens, & Wauterickx, 2008; Tanskanen & Danielsbacka, 2019). Regarding age differences, the results suggest that, in the context of family communication, social media may best cater to the needs of middle-aged individuals, that is, it does not clearly benefit young adults, as emphasized in the digital divide literature, nor older adults, as predicted by SST. 4.2. Strengths and limitations Among this study’s strengths, the data used were representative and extensive, allowing the examination of multiple subgroups. Moreover, when N is high, a violation of the regression assumption does not have serious consequences for the estimates (e.g., violations of the normality of errors). The extensive number of variables in the data allowed con- trolling for multiple confounders, of which the possibility to control for the characteristics of social networks is a particular strength. Finally, the PSM method is an added strength, as it improves the credibility of the estimated associations by focusing on comparisons between groups only for comparable individuals, while regression models can be biased due to extrapolations. Furthermore, similar results from the two methods can signal the robustness of the regression estimates. Conversely, when the two methods provide different results, the PSM estimates are more reliable. When interpreting these results, the following limitations should be noted. First, despite controlling for a range of possible covariates that have been previously identified as factors determining social media use that are likely to affect happiness (Tammisalo et al., 2022), the effect of unobserved confounders cannot be ruled out entirely when using Table 5 Estimated coefficients for SMC with family members and self-rated happiness (1 ˆ very unhappy, 10 ˆ very happy) in middle-aged adults (40–54 years). The controls referred to in the table but not shown are: age, education, income, marital status, health, number of children, and number of close friends. Where genders are analyzed together, gender is controlled for. The ages of daughters/ sons are controlled for in their respective models. Regression without controls Regression with controls SMC with/ by Coef. (R2) 95% CI Coef. (R2) 95% CI Daughter/ All 0.169 (0.003) 0.100 0.439 0.331** (0.195) 0.034 0.628 /Women 0.269* (0.006) 0.049 0.587 0.442** (0.196) 0.081 0.803 /Men 0.097 (0.001) 0.609 0.414 0.021 (0.268) 0.567 0.525 Son/All 0.129 (0.002) 0.136 0.395 0.202 (0.203) 0.085 0.490 /Women 0.115 (0.001) 0.188 0.419 0.289* (0.174) 0.048 0.626 /Men 0.162 (0.002) 0.368 0.692 0.037 (0.325) 0.593 0.518 Mother/All 0.280** (0.006) 0.054 0.506 0.209** (0.210) 0.002 0.416 /Women 0.138 (0.002) 0.122 0.397 0.116 (0.171) 0.127 0.359 /Men 0.657*** (0.028) 0.226 1.089 0.346* (0.304) 0.039 0.732 Father/All 0.268* (0.004) 0.024 0.561 0.050 (0.211) 0.217 0.317 /Women 0.063 (0.000) 0.263 0.389 0.063 (0.138) 0.376 0.250 /Men 0.711** (0.021) 0.139 1.282 0.155 (0.356) 0.329 0.640 Sister/All 0.357** (0.007) 0.012 0.701 0.175 (0.220) 0.143 0.494 /Women 0.298 (0.005) 0.113 0.710 0.183 (0.160) 0.217 0.583 /Men 0.328 (0.006) 0.303 0.959 0.331 (0.396) 0.185 0.847 Brother/All 0.168 (0.002) 0.131 0.467 0.116 (0.207) 0.059 0.391 /Women 0.200 (0.003) 0.136 0.535 0.223 (0.180) 0.095 0.541 /Men 0.190 (0.002) 0.440 0.820 0.006 (0.367) 0.540 0.528 Significance levels: *p < 0.1, **p < 0.05, ***p < 0.01. Table 6 Estimated coefficients for SMC with family members and self-rated happiness (1 ˆ very unhappy, 10 ˆ very happy) in young adults (19–39 years). The controls referred to in the table but not shown are: age, education, income, marital status, health, number of children, and number of close friends. Where genders are analyzed together, gender is controlled for. The ages of daughters/sons are controlled for in their respective models. Regression without controls Regression with controls SMC with/ by Coef. (R2) 95% CI Coef. (R2) 95% CI Daughter/ All 0.262 (0.006) 0.727 0.202 0.134 (0.184) 0.415 0.682 /Women ¡0.485* (0.021) 1.011 0.041 0.157 (0.229) 0.803 0.488 /Men 0.407 (0.014) 0.565 1.379 1.356** (0.296) 0.198 2.515 Son/All 0.164 (0.002) 0.672 0.345 0.322 (0.165) 0.318 0.962 /Women 0.379 (0.012) 0.931 0.173 0.063 (0.188) 0.646 0.772 /Men 0.444 (0.011) 0.723 1.612 1.296 (0.233) 0.318 2.910 Mother/All 0.247 (0.005) 0.074 0.567 0.117 (0.189) 0.180 0.415 /Women 0.168 (0.003) 0.187 0.523 0.205 (0.160) 0.135 0.545 /Men 0.375 (0.009) 0.317 1.068 0.109 (0.275) 0.771 0.552 Father/All 0.302 (0.006) 0.072 0.675 0.107 (0.173) 0.244 0.459 /Women 0.102 (0.001) 0.303 0.508 0.032 (0.120) 0.357 0.421 /Men 0.807** (0.032) 0.011 1.602 0.129 (0.272) 0.664 0.923 Sister/All 0.366* (0.009) 0.067 0.799 0.145 (0.195) 0.261 0.551 /Women 0.244 (0.004) 0.739 0.250 0.137 (0.160) 0.613 0.345 /Men 1.408*** (0.116) 0.558 2.258 0.674 (0.363) 0.222 1.570 Brother/All 0.007 (0.000) 0.386 0.373 0.105 (0.185) 0.468 0.259 /Women ¡0.380* (0.015) 0.788 0.028 ¡.373* (0.149) 0.780 0.034 /Men 0.911** (0.058) 0.086 1.736 0.543 (0.303) 0.329 1.415 Significance levels: *p < 0.1, **p < 0.05, ***p < 0.01. K. Tammisalo et al. Computers in Human Behavior 153 (2024) 108103 9 non-experimental data. In other words, those with SMC with family members may differ in some unobserved systematic manner from those who do not have SMC with family, and this differentiating factor may also be associated with happiness (e.g., having more frequent social contacts or an extraverted personality). The second limitation pertains to the definition of social media. Social media is a rapidly transforming technology that lacks a precise definition. Our measure relies on the respondents’ interpretation of the question and the brief description provided (Facebook, Instagram, Twitter, and Snapchat were the given examples), therefore, not capturing nuanced information about the re- spondents’ social media use. More nuanced information on the quality of communication mediated by social media and the frequency of social media use for family communication is still lacking. Finally, the chosen outcome measure of subjective happiness was based on a single question, therefore, capturing only one aspect of overall well-being. Well-being generally includes multiple dimensions besides pleasant emotions, such as life satisfaction and the scarcity of unpleasant emotions (Diener, Oishi, & Lucas, 2002). Numerous mea- sures of well-being may be relevant when studying the effect of social media. However, considering the chosen theoretical approach (i.e., socioemotional selectivity theory), happiness was the most appropriate. This is because the theory views emotional gratification as the key motivator behind older adults’ tendency to communicate with their close ties. In future studies, a more suitable method for testing the socioemotional selectivity theory would be an experience sampling approach to happiness to capture emotional states arising from SMC with family members over an extended period. 4.3. Concluding remarks Although challenges posed by the concurrent trends of digitalization and population aging have been identified globally, caution should be exercised when generalizing these findings to other cultural contexts. This study was conducted in Finland, a society with high ICT penetration but in which families play a relatively small role as a source of support (Danielsbacka et al., 2020). Although we did not find SMC with family members to increase older adults’ happiness, social media use may be beneficial in other ways. In this study, we did not consider, for example, that social media may facilitate access to information and participation in society which contribute to a broader concept of well-being. Furthermore, for older adults, both sustaining contact with family and diverse use of digital media, including social media, can be considered to contribute posi- tively to active aging. Statement of ethical approval Ethical approval for the Gentrans survey was granted by the Ethical Board of Statistics of Finland in 2006. CRediT authorship contribution statement Kristiina Tammisalo: Conceptualization, Formal analysis, Investi- gation, Methodology, Writing - original draft, Writing - review & edit- ing. Mirkka Danielsbacka: Conceptualization, Data curation, Funding acquisition, Investigation, Project administration, Resources, Supervi- sion, Writing - review & editing. Antti O. Tanskanen: Conceptualiza- tion, Data curation, Funding acquisition, Investigation, Project administration, Resources, Supervision, Writing - review & editing. Bruno Arpino: Conceptualization, Formal analysis, Methodology, Re- sources, Writing - review & editing. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Data availability The authors do not have permission to share data. Acknowledgements The study is part of the NetResilience consortium funded by the Strategic Research Council at the Academy of Finland (Grant number 345183) and the INVEST flagship funded by the Academy of Finland (Grant number 320162). Additional funding was received from Acad- emy of Finland (Grant numbers 338869 and 331400 and 325857 and 317808). B. Arpino acknowledges funding for the project "Socio-de- mographic determinants of Information and Communication Technol- ogies (ICT) use among older people in Italy (ICTAGE)", Italian Ministry of Research (PRIN 2022). The financial sponsors have not influenced the design, execution, analysis or interpretation of data, or writing of the study. Appendix A. Supplementary data Supplementary data to this article can be found online at https://doi. org/10.1016/j.chb.2023.108103. References Aarts, S. (2018). Social media and loneliness among community-dwelling older adults. International Journal of Geriatric Psychiatry, 33(3), 554–555. https://doi.org/ 10.1002/gps.4769 Arpino, B., Meli, E., Pasqualini, M., Tomassini, C., & Cisotto, E. (2022). Determinants of grandparent–grandchild digital contact in Italy. Genus, 78(1), 1–22. https://doi.org/ 10.1186/s41118-022-00167-5 Arpino, B., Pasqualini, M., Bordone, V., & Sole-Auro, A. (2021). Older people’s nonphysical contacts and depression during the COVID-19 lockdown. The Gerontologist, 61(2), 176–186. https://doi.org/10.1093/geront/gnaa144 Bracke, P., Christiaens, W., & Wauterickx, N. (2008). The pivotal role of women in informal care. Journal of Family Issues, 29(10), 1348–1378. https://doi.org/10.1177/ 0192513X08316115 Buchanan, A., & Rotkirch, A. (Eds.). (2021). Brothers and sisters: Sibling relationships across the life course. Cham: Palgrave Macmillan. https://doi.org/10.1007/978-3-030- 55985-4. Cannas, M., & Arpino, B. (2019). A comparison of machine learning algorithms and covariate balance measures for propensity score matching and weighting. Biometrical Journal. Biometrische Zeitschrift, 61(4), 1049–1072. https://doi.org/10.1002/ bimj.201800132 Carstensen, L. L., Isaacowitz, D. M., & Charles, S. T. (1999). Taking time seriously. A theory of socioemotional selectivity. American Psychologist, 54(3), 165–181. https:// doi.org/10.1037/0003-066x.54.3.165 Chan, M. (2015). Multimodal connectedness and quality of life: Examining the influences of technology adoption and interpersonal communication on well-being across the life span. Journal of Computer-Mediated Communication: JCMC, 20(1), 3–18. https:// doi.org/10.1111/jcc4.12089 Chan, M. (2018). Digital communications and psychological well-being across the life span: Examining the intervening roles of social capital and civic engagement. Telematics and Informatics, 35(6), 1744–1754. https://doi.org/10.1016/j. tele.2018.05.003 Chang, P. F., Choi, Y. H., Bazarova, N. N., & Lockenhoff, C. E. (2015). Age differences in online social networking: Extending socioemotional selectivity theory to social network sites. Journal of Broadcasting & Electronic Media, 59(2), 221–239. https:// doi.org/10.1080/08838151.2015.1029126 Cui, D. (2016). Beyond “connected presence”: Multimedia mobile instant messaging in close relationship management. Mobile Media & Communication, 4(1), 19–36. https://doi.org/10.1177/2050157915583925 Daft, R. L., & Lengel, R. H. (1986). Organizational information requirements, media richness and structural design. Management Science, 32(5), 554–571. https://doi.org/ 10.1287/mnsc.32.5.554 Danielsbacka, M., Hamalainen, H., & Tanskanen, A. O. (Eds.). (2020). Suomalainen auttaminen: Tukiverkostot suurten ikaluokkien ja heidan lastensa elamassa. Gaudeamus. Danielsbacka, M., Tammisalo, K., & Tanskanen, A. O. (2023). Digital and traditional communication with kin: Displacement or reinforcement? Journal of Family Studies. https://doi.org/10.1080/13229400.2022.2040575 David-Barrett, T., Kertesz, J., Rotkirch, A., Ghosh, A., Bhattacharya, K., Monsivais, D., et al. (2016). Communication with family and friends across the life course. PLoS One, 11(11), Article e0165687. https://doi.org/10.1371/journal.pone.0165687 Diener, E., Oishi, S., & Lucas, R. E. (2002). In S. J. Lopez, & C. R. Snyder (Eds.), The Oxford handbook of positive psychologySubjective well-being: The science of happiness and life satisfaction (pp. 187–194). Oxford University Press. K. Tammisalo et al. Computers in Human Behavior 153 (2024) 108103 10 Dienlin, T., Masur, P. K., & Trepte, S. (2017). Reinforcement or displacement? The reciprocity of FtF, IM, and SNS communication and their effects on loneliness and life satisfaction. Journal of Computer-Mediated Communication, 22(2), 71–87. https:// doi.org/10.1111/jcc4.12183 Drake, C. (1993). Effects of misspecification of the propensity score on estimators of treatment effect. Biometrics, 49(4), 1231–1236. https://doi.org/10.2307/2532266 Gazit, T., Nisim, S., & Ayalon, L. (2022). Intergenerational family online community and older adults’ overall well-being. Online Information Review, 47(2), 221–237. https:// doi.org/10.1108/OIR-06-2021-0332 Goodman-Deane, J., Mieczakowski, A., Johnson, D., Goldhaber, T., & Clarkson, P. J. (2016). The impact of communication technologies on life and relationship satisfaction. Computers in Human Behavior, 57, 219–229. https://doi.org/10.1016/j. chb.2015.11.053 Hamalainen, H., Danielsbacka, M., Hagglund, A. E., Rotkirch, A., & Tanskanen, A. O. (2021). Sukupolvien suhteet - Ikaantyminen ja vuorovaikutuksen muutos suurten ikaluokkien ja aikuisten lasten elamankulussa. Vaestoliitto. Hutto, C. J., Bell, C., Farmer, S., Fausset, C., Harley, L., Nguyen, J., et al. (2015). Social media gerontology: Understanding social media usage among older adults. Web Intelligence, 13(1), 69–87. https://doi.org/10.3233/web-150310 Ishii, K. (2017). Online communication with strong ties and subjective well-being in Japan. Computers in Human Behavior, 66, 129–137. https://doi.org/10.1016/j. chb.2016.09.033 Kim, C., & Shen, C. (2020). Connecting activities on social network sites and life satisfaction: A comparison of older and younger users. Computers in Human Behavior, 105, Article 106222. https://doi.org/10.1016/j.chb.2019.106222 Lockenhoff, C. E., & Carstensen, L. L. (2004). Socioemotional selectivity theory, aging, and health: The increasingly delicate balance between regulating emotions and making tough choices. Journal of Personality, 72(6), 1395–1424. https://doi.org/ 10.1111/j.1467-6494.2004.00301.x Lythreatis, S., Singh, S. K., & El-Kassar, A.-N. (2022). The digital divide: A review and future research agenda. Technological Forecasting and Social Change, 175, Article 121359. https://doi.org/10.1016/j.techfore.2021.121359 Matassi, M., Boczkowski, P. J., & Mitchelstein, E. (2019). Domesticating WhatsApp: Family, friends, work, and study in everyday communication. New Media & Society, 21(10), 2183–2200. https://doi.org/10.1177/1461444819841890 Neves, B. B., Franz, R., Judges, R., Beermann, C., & Baecker, R. (2017). Can digital technology enhance social connectedness among older adults? A feasibility study. Journal of Applied Gerontology: The Official Journal of the Southern Gerontological Society, 38(1), 49–72. https://doi.org/10.1177/0733464817741369 Newman, L., Stoner, C., & Spector, A. (2021). Social networking sites and the experience of older adult users: A systematic review. Ageing and Society, 41(2), 377–402. https://doi.org/10.1017/S0144686X19001144 Nguyen, M. H., Hunsaker, A., & Hargittai, E. (2022). Older adults’ online social engagement and social capital: The moderating role of internet skills. Information, Communication & Society, 25(7), 942–958. https://doi.org/10.1080/ 1369118X.2020.1804980 Roberts, S. G. B., & Dunbar, R. I. M. (2011). The costs of family and friends: An 18-month longitudinal study of relationship maintenance and decay. Evolution and Human Behavior, 32(3), 186–197. https://doi.org/10.1016/j.evolhumbehav.2010.08.005 Robinson, L., Cotten, S. R., Ono, H., Quan-Haase, A., Mesch, G., Chen, W., et al. (2015). Digital inequalities and why they matter. Information, Communication & Society, 18 (5), 569–582. https://doi.org/10.1080/1369118X.2015.1012532 Rosenberg, D., & Taipale, S. (2022). Social and satisfied? Social uses of mobile phone and subjective wellbeing in later life. Human Technology, 18(1), 45–65. https://doi.org/ 10.14254/1795-6889.2022.18-1.4 Scheerder, A., van Deursen, A., & van Dijk, J. (2017). Determinants of Internet skills, uses and outcomes. A systematic review of the second- and third-level digital divide. Telematics and Informatics, 34(8), 1607–1624. https://doi.org/10.1016/j. tele.2017.07.007 Seabrook, E. M., Kern, M. L., & Rickard, N. S. (2016). Social networking sites, depression, and anxiety: A systematic review. JMIR Mental Health, 3(4), e50. https://doi.org/ 10.2196/mental.5842 Shen, C., Wang, M. P., Chu, J. T., Wan, A., Viswanath, K., Chan, S. S. C., et al. (2017). Sharing family life information through video calls and other information and communication technologies and the association with family well-being: Population- based survey. JMIR Mental Health, 4(4), e57. https://doi.org/10.2196/mental.8139 Simons, M., Reijnders, J., Peeters, S., Janssens, M., Lataster, J., & Jacobs, N. (2021). Social network sites as a means to support personal social capital and well-being in older age: An association study. Computers in Human Behavior Reports, 3, Article 100067. https://doi.org/10.1016/j.chbr.2021.100067 Taipale, S. (2019). Intergenerational connections in digital families. Cham: Springer. https://doi.org/10.1007/978-3-030-11947-8 Taipale, S., & Farinosi, M. (2018). The big meaning of small messages: The use of WhatsApp in intergenerational family communication. In Human aspects of IT for the aged population. Acceptance, Communication and Participation. https://doi.org/ 10.1007/978-3-319-92034-4_40 Tammisalo, K., Danielsbacka, M., Andersson, E., & Tanskanen, A. O. (2022). Predictors of social media use in two family generations. Frontiers in Sociology, 6. https://doi.org/ 10.3389/fsoc.2021.813765 Tammisalo, K., & Rotkirch, A. (2022). Effects of information and communication technology on the quality of family relationships: A systematic review. Journal of Social and Personal Relationships, 39(9), 2724–2765. https://doi.org/10.1177/ 02654075221087942 Tanskanen, A. O., & Danielsbacka, M. (2019). Intergenerational family relations: An evolutionary social science approach. Routledge. Verduyn, P., Ybarra, O., Resibois, M., Jonides, J., & Kross, E. (2017). Do social network sites enhance or undermine subjective well-being? A critical review. Social Issues and Policy Review, 11(1), 274–302. https://doi.org/10.1111/sipr.12033 Vriens, E., & van Ingen, E. (2018). Does the rise of the internet bring erosion of strong ties? Analyses of social media use and changes in core discussion networks. New Media & Society, 20(7), 2432–2449. https://doi.org/10.1177/1461444817724169 Wrzus, C., Hanel, M., Wagner, J., & Neyer, F. J. (2013). Social network changes and life events across the life span: A meta-analysis. Psychological Bulletin, 139, 53–80. https://doi.org/10.1037/a0028601 Yu, R. P., Ellison, N. B., McCammon, R. J., & Langa, K. M. (2016). Mapping the two levels of digital divide: Internet access and social network site adoption among older adults in the USA. Information, Communication & Society, 19(10), 1445–1464. https://doi. org/10.1080/1369118X.2015.1109695 Yu, R. P., Mccammon, R. J., Ellison, N. B., & Langa, K. M. (2016). The relationships that matter: Social network site use and social wellbeing among older adults in the United States of America. Ageing and Society, 36(9), 1826–1852. https://doi.org/10.1017/ S0144686X15000677 Zickuhr, K., & Madden, M. (2012). Older adults and Internet use. Pew Research Center. http://www.pewresearch.org/internet/2012/06/06/main-report-15. K. Tammisalo et al.