We used a multilevel statistical model to investigate these associations separately for positive and negative RQ. Multilevel modeling accounts for the bias in standard errors and statistical tests that result from nonindependent data and effectively handles “unbalanced” or missing data at the level of repeated observations by using all available data for participants. Inferences are valid assuming missing data are missing at random (see Schafer & Graham, 2002). The models had two levels: within-dyad (over time) and between-dyad. Using the dyadic longitudinal approach described by Bolger and Shrout (2007), we included wives’ and husbands’ RQ in a single multilevel analysis to account for the fact that wives’ and husbands’ data were clustered within dyad. All analyses were conducted using the MIXED procedure in SAS (Version 9.1.3, 1997).
The within-dyad level of the analysis allowed each dyad’s RQ to be modeled as a function of husbands’ and wives’ anxiety. We predicted a given day’s husbands’ and wives’ RQ for a particular dyad and adjusted for number of days in the study and weekend effects. Because husbands’ anxiety may be highly associated with husbands’ evaluations of RQ, husbands’ anxiety was included to adjust for this effect. The model specified was as follows:
where Yijk is the RQ for dyad i for person j (j = 1 is wife’s report; j = 2 is husband’s report) on day k. When the outcome is the wife’s report (Wifeijk = 1 and Husbijk = 0), the first part of the model is selected and all of the b coefficients have the subscript w. Similarly, when the outcome is the husband’s report, Wifeijk = 0, Husbijk = 1, and the second part of the model is selected. Daysik is the number of days in the study; Weekendik indicates whether it is a weekend day or not; WAnxik is the wife’s report of anxiety; HAnxik is the husband’s report of anxiety; the residual components are represented by eijk. All predictor variables were within-person centered (Raudenbush & Bryk, 2002). Finally, the approach discussed by Bolger and Shrout (2007) allowed us not only to account for dependency within individuals across time (i.e., autoregressive) but also to account for dependency within dyads (pp. 292–295).
The between-dyad level of this analysis modeled individual differences in the coefficients specified in Equation 1. We fit a model that considered intercepts for both wives’ and husbands’ reports of RQ to be random (i.e., varying across persons). In addition, slope of day on wives’ RQ (b1w), slope of wives’ anxiety on wives’ RQ (b3w), and slope of husbands’ anxiety on husbands’ RQ (b4h) were modeled to be random for the positive RQ analysis; slope of husbands’ anxiety on husbands’ RQ (64h) was modeled to be random for the negative RQ analysis.1 Random effects were tested using the nested comparison of likelihood ratio (Singer & Willett, 2003, p. 119).
Table 3 presents results for both wives’ and husbands’ reports of positive RQ.2 Only variables of interest are reported here. Wives’ anxiety was not associated with their own positive RQ, b3w = −0.03, t(32) = −0.89, ns, but was significantly associated with husbands’ positive RQ, b3h = −0.14, t(32) = −3.05, p < .01. Specifically, on days when wives experienced higher anxiety, husbands reported less positive relationship quality. There was no significant variation around these effects.
Persecution Live 