For descriptive purposes, we examined the mean scores of the study variables in the experimental conditions, as well as the correlations between all study variables. Next, hierarchical moderated regression analyses were conducted to test the hypotheses concerning the extent to which cognitive set predicted participant's levels of relationship satisfaction and commitment, the two-way interactions between cognitive set and discontent with the partner on the one hand, and cognitive set and social comparison orientation on the other hand, and the three-way interaction between cognitive set and the two moderator variables.

In Step 1, cognitive set (0 = no-comparison thought-generating task and 1 = downward comparison thought-generating task), discontent with the partner, and social comparison orientation were entered simultaneously. In Step 2, the three two way interaction terms (Cognitive Set X Discontent, Cognitive Set X Comparison Orientation, Discontent X Comparison Orientation) were added to the regression equation. Finally, in Step 3, the three way interaction term (Cognitive Set X Discontent X Comparison Orientation) was entered.

To test and interpret these interactions, procedures outlined by Aiken and West (1991) were followed to calculate regression slopes and plots. Following Aiken and West, before the interaction terms were calculated, the criterion variables and all continuous predictor variables (i.e., discontent and comparison orientation) were standardized to reduce possible multicollinearity among the independent and interaction terms as well as to facilitate interpretation of the interaction effects. Furthermore, for each continuous predictor, participants with scores one standard deviation above the mean were labelled high social comparison orientation or high discontent with the partner, and those with scores one standard deviation below the mean were labeled low social comparison orientation or low discontent (following Cohen & Cohen, 1983). Where appropriate, significance values concern one tailed tests; two-tailed tests are reported if the direction of the slopes were contrary to the predictions or if no specific predictions were made.

Table 1 presents, separately for participants in the no-comparison and the downward comparison conditions, the means and standard deviations for the variables used in Study 2. Preliminary one-way analyses of variance with cognitive set (no-comparison versus downward comparison) as independent variable indicated that neither the moderator nor the dependent variables were related to cognitive set, Fs < 1, ns. As can be seen from Table 1, the variables measuring relationship quality were significantly correlated. Not surprisingly, especially relationship satisfaction and the moderator variable relational discontent were highly correlated. Furthermore, comparison orientation showed small, but statistically significant relations with the relationship variables.

Relationship satisfaction. The main effects of cognitive set, discontent, and social comparison orientation accounted for a significant amount of variance in relationship satisfaction, R² = .52, F(3, 112) = 40.37, p < .0001. In contrast to our prediction, the dummy coded term representing cognitive set did not yield a significant effect (B = .03, t(112) = .21, ns). Social comparison orientation (B = .15, t(112) = 2.32, p < .03) and discontent with the partner (B = -.68, t(112) = -10.30, p < .0001) were significantly related to relationship satisfaction, with those high in social comparison orientation and high in discontent experiencing less satisfaction.

The two-way interactions entered at the second step yielded a significant incremental amount of explained variance, R² = .07, F_change(3, 109) = 6.44, p < .001. Neither the predicted Cognitive Set X Comparison Orientation interaction (B = .03, t(109) = .03, ns) nor the Discontent X Comparison Orientation interaction (B = -.04, t(109) = -.03, ns) reached statistical significance, whereas the predicted Cognitive Set X Discontent interaction emerged (B = .53, t(109) = 4.24, p < .0001). Inspection of this interaction revealed that the downward comparison thought-generating task moderated the impact of discontent with the partner than the task in which noncomparative thoughts about the partner were generated: Discontent was less strongly negatively related to relationship satisfaction in the downward comparison condition (B = .43, t(109) = 4.82, p < .001) than in the no-comparison condition (B = -.97, t(109) = 10.80, p < .0001).

However, this interaction was qualified by the expected three-way interaction that accounted for an additional amount of explained variance in relationship satisfaction (B = .35), R² = .02, F_change(1, 108) = 6.85, p < .02. To clarify the nature of this interaction, the direction of the Cognitive Set X Discontent interactions were calculated in additional regression analyses at low and high levels (1 SD above and below the mean) of social comparison orientation, after which the interaction was plotted following standard procedures (Aiken & West, 1991).

As shown in Figure 1, and in line with the predictions, at low levels of social comparison orientation the Cognitive Set X Discontent interaction was not significant (B = .09, t(109) = .44, ns), whereas a significant interaction term was found at high levels of social comparison orientation (B = .78, t(109) = 5.04, p < .0001). The slopes of the standardized form of the equation, and their significance levels were as follows: Among participants low in social comparison orientation, discontent was an equally important predictor of satisfaction in the downward comparison condition (B = -.77, t(108) = 4.53, p < .0001) as well as in the no-comparison condition (B = .86, t(108) = 7.16, p < .0001). In contrast, for those high in social comparison orientation, discontent was less strongly related to satisfaction in the downward comparison condition (B = .27, t(108) = 2.53, p < .02) than in the no-comparison condition (B = -1.05, t(108) = 9.31, p < .0001).

Thus, in line with our prediction, the Cognitive Set X Discontent interaction appeared to be stronger for participants high in social comparison orientation, indicating that cognitive downward comparison led to a relatively better relationship quality among individuals high in discontent only when they were high in social comparison orientation.

Commitment to the relationship. The main effects of cognitive set, discontent, and social comparison orientation accounted for a significant amount of variance in commitment, R² = .23, F(3, 112) = 10.99, p < .0001. The dummy coded term representing cognitive set did not yield a significant effect (B = .16, t(112) = .95, ns). Discontent with the partner was a strongly significant predictor of commitment to the relationship (B = -.45, t(112) = -5.35, p < .0001), indicating that those who were discontent with the relationship with their partner felt less strongly committed to their relationship. Social comparison orientation was not significantly related to commitment (B = -.10, t(112) = -1.21, ns).

The entry of the two-way interactions at the second step did not account for further incremental variance in commitment, R² = .04, F_change(3, 109) = 2.05, ns. The Discontent X Comparison Orientation interaction (B = -.02, t(109) = -.28, ns) and the predicted Cognitive Set X Comparison Orientation interaction did not approach statistical significance (B = .14, t(109) = .84, ns). However, in line with our prediction, a Cognitive Set X Discontent interaction (B = .37, t(109) = 2.20, p < .02) indicated that downward comparison thought-generating was a more effective moderator of the negative effect of discontent on commitment than the task in which noncomparative thoughts about the partner were generated. Inspection of the two-way interaction revealed that discontent was less strongly negatively related to commitment in the downward comparison condition (B = .27, t(109) = 2.23, p < .02) than in the no-comparison condition (B = -.64, t(109) = 5.34, p < .0001).

As expected, this two-way interaction was qualified by the three-way interaction that accounted for a just significant portion of explained variance in commitment (B = .35), R² = .03, F_change(1, 108) = 3.83, p =.05. Additional regression analyses were performed to clarify the nature of this interaction in which the direction of the Cognitive Set X Discontent interactions were calculated at low and high levels of social comparison orientation. At low levels of social comparison orientation, the Cognitive Set X Discontent interaction was not significant (B = .077, t(109) = -.27, ns), whereas a significant interaction term was found at high levels of social comparison orientation (B = .62, t(109) = 2.96, p < .005). The slopes of the standardized form of the equation, and their significance levels were calculated to clarify these two-way interactions.

As shown in Figure 2, for participants low in social comparison orientation, discontent significantly predicted commitment in both the downward comparison condition (B = -.56, t(108) = 2.43, p < .02), and within the no-comparison condition (B = -.48, t(108) = 2.96, p < .005). In contrast, for those high in social comparison orientation, discontent was not at all associated with commitment in the downward comparison condition (B = -.15, t(108) = 1.05, ns), whereas in the no-comparison condition there was a highly significant association between both variables (B = -.78, t(108) = 5.06, p < .0001). Thus, especially for participants who were strongly inclined to engage in social comparison, cognitive downward comparison contributed to a relatively higher commitment among those experiencing a high discontent with the relationship.

As in study 1, it was examined again if the number of the qualities generated in both conditions could account for the results. There was no effect of cognitive set, F(1, 114) = .01, ns, upon the number of qualities generated. To determine whether the number of qualities did account for the effects on satisfaction, commitment and perceived relationship quality, the hierarchical moderated regression analyses were repeated after adding the number of the qualities in the first step. In these analyses, all of the major findings were replicated.