Participant's average satisfaction and commitment ratings were subjected to a two-way analysis of variance (ANOVA) with target (self versus partner) and cognitive set (no-comparison versus downward comparison) as the independent variables². Our primary prediction on these measures was that participants would perceive higher relationship quality following the downward comparison thought-generating task rather than following the no-comparison thought-generating task. The ANOVA revealed indeed that participants in the downward comparison condition reported more satisfaction (M = 7.67) than participants in the no-comparison condition (M = 6.79), F(1, 90) = 8.95, p < .01. There was no difference between participants who generated their thoughts about the self and the partner. Cognitive set did not interact significantly with target, F < 1.0, ns.

The ANOVA on the commitment measure indicated that participants who were induced to engage in enhancing self- and partner-evaluations felt more strongly committed (M = 7.51) to their relationship than participants who were instructed to make noncomparative evaluations (M = 6.83), F(1, 90) = 5.92, p < .02. There was again no significant difference between evaluations of self and partner on commitment, F = 2.9, ns. The interaction between both independent variables was not significant, F < 1.0, ns.

Response Latencies.   As is typically the case with response times (cf. Fazio, 1990), the average response latencies were positively skewed. Therefore, as recommended by Fazio, the response times were subjected to reciprocal transformations. Also, to reduce the possible impact of outlier response latencies on the analyses, those over 3 standard deviations from the mean were excluded. We then computed for each participant the latency indexes by averaging the recorded response times for each question of the dependent variable in separate indexes. Likewise, for each participant, a baseline response time was computed averaging the response times from presenting the biographical questions to entering the responses in one index. All data analyses were performed on the transformed scores, but for the sake of clarity, we report the retransformed latencies.

Subsequently, the transformed response latencies to the questionnaires of satisfaction and commitment were submitted to two-way (Target X Cognitive Set) analyses of covariance (ANCOVAs), with the baseline index for response time as a covariate³. In line with the above reported effects, cognitive set had significant effects on the latency indexes of satisfaction, F(1, 89) = 4.96, p < .03, and commitment, F(1, 89) = 5.13, p < .03. That is, as predicted, the average response latencies on both indexes were lower in the comparison condition than in the no-comparison condition. The adjusted means, following a retransformation back to the original metric of seconds, were 9.63 and 11.02 on the satisfaction index, respectively, and 11.29 and 12.60 on the commitment index, respectively. Thus, cognitive downward comparison seems indeed to increases the accessibility of the attitude toward one's relationship.

A number of supplemental analyses were done to control for a number of alternative interpretations of the data. In the first place, it would seem possible that those in the comparison condition generated more qualities of the self and the partner than in the no-comparison condition. Thus, the effects of cognitive downward comparison may have been primarily due to the greater number of qualities generated in the social comparison condition relative to the no-comparison condition.

We had available the number of qualities generated in the various conditions before the five most important ones were typed into the computer. There was no effect of cognitive set, F(1, 90) = 2.53, ns, upon the number of reasons generated. To determine whether the number of qualities did account for the effects on satisfaction and commitment, two 2 (Target) by 2 (Cognitive Set) analysis of covariance (ANCOVA) were performed, with the number of qualities included as a covariate. Both analyses revealed that the main effects of cognitive set on satisfaction and commitment remained statistically significant F(1, 89) = 7.33, p < .01, and F(1, 89) = 4.51, p < .05, respectively.

A second reason that those in the comparison condition reported more satisfaction and stronger feelings of commitment may be that the responses of participants were affected by the positivity of the qualities they had generated. That is, participants in the comparison condition may have generated qualities of the self or the partner that were more positive than participants in the no-comparison condition. It might thus be possible that the beneficial effects of cognitive downward comparison on perceived relationship quality are not due to the fact that individuals actively compared their relationship with that of others, but rather to the likely greater positivity of the qualities generated in the social comparison condition relative to the no-comparison condition.

This issue was dealt with by replicating all of the major analyses by controlling for the positivity of the generated qualities. First, participants were asked to rate the valence of the self-generated qualities. At the end of the experimental procedure, each quality that had been entered into the computer by the respondent (i.e., the thought-listing task asked them to enter the five reasons which they considered as most important into the computer) was presented on the computer screen again.

Participants were asked to rate each quality on 4-point scales ranging from not positive at all to very positive. The positivity ratings were subjected to two-way analyses of variance (ANOVA) with target and cognitive set as the independent variables. The analyses yielded no effects whatsoever, Fs < 1.06, ns. Further, to determine whether participants' positivity ratings did account for the effects on satisfaction and commitment, two 2 (Target) by 2 (Cognitive Set) analysis of covariance (ANCOVA) were performed after adding positivity as a covariate⁴. Both analyses revealed that the main effects of cognitive set on satisfaction and commitment remained statistical reliable, F(1, 89) = 7.93, p < .01, and F(1, 89) = 5.34, p < .03, respectively.

In addition, we had two independent judges rate the qualities on 5-point scales ranging from very negative to very positive. The judges were blind to the hypothesis and experimental conditions, and the Cohen's Kappa for interrater agreement was .88. The analyses of variance on the positivity of the qualities was repeated, and again, no effects reached statistical significance, F < 1.40, ns. Thus, the two judges did not rate the qualities in the no-comparison condition differently from those generated in the comparison condition.

The foregoing analyses of covariance was repeated treating the positivity ratings of the judges as covariate. Again, the main effects of cognitive set on satisfaction and commitment remained, F(1, 89) = 7.93, p < .01, and F(1, 89) = 5.34, p < .03, respectively. In sum, the results of these analyses suggested that our findings were not artifacts of the positivity of the qualities but that they were due to the effects of cognitive downward comparison.