TABLE 3.
Associations between social connectedness and subsequent mental health (Well-Being Survey 2018–2019 and health insurance claims data 2017–2019, United States) a .
| Model specification | Depression | Anxiety |
|---|---|---|
| RR 95% CI | RR 95% CI | |
| Model 1 b | 0.730*** (0.637, 0.837) | 0.824** (0.719, 0.943) |
| Model 2 c | 0.554*** (0.438, 0.700) | 0.691** (0.550, 0.868) |
| Model 3 d | 0.858*** (0.794, 0.928) | 0.894** (0.825, 0.969) |
| Model 4 e | 0.784*** (0.696, 0.882) | 0.835** (0.743, 0.940) |
| Model 5 f | 0.669** (0.505, 0.884) | 0.735* (0.576, 0.937) |
Each analysis was controlled for demographics [gender (ref. = female), age (ref. = below 30), race (ref. = White), education (ref. = high school), marital status (ref. = not married)), having children at home (ref. = no), taking care of an elderly (ref. = no)], wealth and income [home ownership (ref. = no) and salary], lifestyle [voting in the last elections (ref. = no/not registered voter) and spiritual practices (ref. = at least once/week)] and work characteristics (number of work hours (ref. = ≤8 h), supervisor support, job control, job demand and job meaning). These variables were controlled for in the baseline wave, T = 0 (in the same wave as the exposure), since only two waves of survey data were available. Additionally, in each regression an outcome prior to exposure (T = −1) as well as the number of diagnosed health conditions (ranging from 0 to 37 possible diagnosed health conditions) prior to exposure (T = −1) were used as controls.
Model 1 is run on the full imputed analytic sample.
Model 2 is run after excluding participants with baseline mental health condition (i.e., either depression or anxiety).
Model 3 is run under the complete case scenario.
Model 4 uses a limited set of baseline controls (T = 0) including only the outcome prior to exposure (T = −1) and pre-exposure number of health conditions (T = −1).
Model 5 is run as a cross-lagged panel model using the set of controls from the primary analysis; under complete case scenario as Stata gsem command does not support mi estimate command used to account for multiple imputations; uses a logistic regression specification as gsem does not support robust Poisson regression. Odds ratio for the path from social connectedness to mental health outcome is presented. Full set of results is presented in the Supplementary Table SA3).
***p < 0.001, **p < 0.01, *p < 0.05; CI is confidence interval.
CI, confidence interval; RR, risk ratio.