2.5. Statistical Analysis

All dyads were included as part of a single analysis. Repeated measures ANOVA using three time points (T1, T2, and T3) and partner status (person with stroke vs. care partner) was used to examine effects on depressive symptoms over time and to provide preliminary effect sizes. Exploratory post hoc analyses were conducted to identify subgroups who are most likely to benefit from the intervention (younger vs. older couples coping with stroke (based on young stroke definition of 55 years old or younger [34]), men vs. women, and affected cerebral hemisphere). Additional sensitivity analyses for these effects were tested using multilevel modeling (MLM). MLM is a recommended method for conducting intent-to-treat analyses using all available data. The magnitude, direction, and significance level of the results of MLMs were highly similar to those produced by ANOVAs [35].

Finally, an Actor–Partner Interdependence Model (APIM) was used to estimate the association between dyad members’ scores at each time point (interdependence effects). The model examines direct and indirect effects for how each partner’s T1 PROMIS-D-SF score predicts their own (actor effects) as well as their partner’s (partner effects) T2 and T3 scores. These models were estimated using a Bayesian structural equation model; this produces stable and unbiased parameter estimates with sample sizes n = 30 or larger [36].