Statistical Analyses

Consistent with other research using ecological momentary assessments (Shiffman et al., 2008), participants sometimes missed a survey during the day. Thus, we had unequal numbers of surveys on different days and estimated multi-level models using SAS 9.4 (Statistical Analysis System). All models adjusted for participant age, gender, education, marital status, racial/ethnic group, overall pain, number of health conditions, and depressive symptoms. We centered all continuous covariates on the sample means.

We first examined how older adults’ pain during the day was associated with their sleep disturbances at night. We estimated two two-level linear models (SAS PROC MIXED) considering the day level (level 1) nested within the participant level (level 2) to examine this link in two directions. In the model with sleep disturbances as the outcome, the predictor was older adults’ pain averaged across 3-hour assessments within each day. In the model predicting next-day pain as the outcome, we treated sleep disturbances the prior night as the predictor.

We then tested whether older adults’ positive encounters attenuated the bidirectional links between older adults’ pain and sleep disturbances at night. We added interaction terms to the two-level linear models using SAS PROC MIXED, as described above. We examined the moderating role of (a) the proportion of positive encounters and (b) the average pleasantness of encounters. In all models, variables included in the interaction terms were mean centered. We conducted simple slope analysis for significant interactions.

We then reestimated the moderation tests above; but this time, we examined positive encounters with close partners and with nonclose partners separately. These two moderators were entered into the same model.