Dyadic analytic strategy

Data from couples present special analytic challenges because the lack of independence among partners (Kenny et al, 2006). To address this issue, we used IBM SPSS AMOS v.23 a structural equation modeling (SEM) software. The SEM enables to estimate path coefficients describing theoretically-based complex network of relationships while accounting for dyadic dependencies. Moreover, SEM measures of model fit indices help assess how well a given model accounts for the relationships among partners, as well as relationships among scores of each individual. Non-significant chi-square value, CFI and TLI greater than 0.95, and RMSEA lower than 0.06 index excellent model fit (Kline, 2005).

Prior to estimating the model, we tested whether heterosexual partners were distinguishable by gender using independent t-tests (Supplementary Table S1) and found no gender effects, and thus treated all couples interchangeably in the SEM (Assad et al, 2007; Olsen and Kenny, 2006). Treating members as ‘interchangeable’ implies that assignment of participants to category of partner A or partner B is arbitrary (Kenny et al, 2006). This assignment requires imposing constrains on model paths, means, intercepts, and variances across respondent and partner, and also requires corrections for model fit indices (Olsen and Kenny, 2006). As a result of these constrains, the model path coefficients provide estimates of the intraclass covariances, the appropriate measure of association for dyadic data (Griffin and Gonzalez, 1995; Woody and Sadler, 2005).

To examine indirect association (mediation effects) between study variables we used Hayes's (2013) guidelines. Accordingly, the significance of an indirect effect is estimated using the 95% confidence interval (CI) of the cross-product of the predictor-to-mediator path and the mediator-to-outcome path. As recommended by Hayes (2013), we used 5000 bootstrapped samples to estimate the bias-corrected and accelerated 95% CI of the indirect path. When zero is not included in the 95% CI, it implies a significant statistical effect at α<0.05(Preacher and Hayes, 2008).