Running Head: MODERATION and MEDIATION REGRESSION

Moderation and Mediation 1

Running head: MODERATED AND MEDIATED REGRESSION

Using Moderated and Mediated Regression:

Predicting Relationship Self-Efficacy and Relationship Intentions

Michael J. Walk

University of Baltimore

Using Moderated and Mediated Regression:

Predicting Relationship Self-Efficacy and Relationship Intentions

Research Question 1

People in serious dating relationships experience varying degrees of relationship self-efficacy. There are probably several factors that influence relationship self-efficacy; however, this study will examine the connection between relationship self-efficacy and time together (i.e., relationship length). Members of longer lasting relationships will most likely feel higher degrees of relationship self-efficacy—the length of the relationship is itself evidence for the efficaciousness of the actions therein. However, this association is probably moderated by how a partner reacts to and reciprocates the actions of the other partner. Perceptions of partner reciprocity (or the lack thereof) can be expected to impact one’s relationship self-efficacy. I hypothesize that time together (time2) will be a significant predictor of relationship self-efficacy (RSe), and that this relationship will vary across levels of perceived reciprocity (recip).

In order to investigate this hypothesis, I will run a moderated multiple regression. Moderated multiple regression is the appropriate statistical procedure to use in this scenario, because moderated regression includes the interaction of predictors as a term in the regression equation in order to examine whether or not the interaction of the predictors accounts for incremental variance in the dependent variable beyond the variance accounted for by main effects. Before running the regression, predictor variables were centered and an interaction variable (time2*recip) was created.

Results and Discussion

Results of the moderated multiple regression of RSe on time2, recip, and time2*recip are presented in Table 1. The model including only main effects of time2 and recip was significant, F(2, 331) = 10.99, R2 = .06, p < .01. The unstandardized regression coefficient for time2 was not significant, B = .02, t = .33, p = .74; however, the coefficient for reciprocity was significantly different from zero, B = .14, t = 2.38, p < .05, indicating that, for those people with average-length relationships, a one unit change in reciprocity was associated with a .14 increase in relationship self-efficacy. When the interaction term was sequentially introduced into the regression, the model (including time2, recip, and their interaction as predictors) was significant, F(3, 330) = 11.67, R2 = .10, p < .01, and accounted for significant incremental variance beyond the main effects, Fchange(1, 330) = 12.29, ∆R2 = .03, p < .01. The interaction term’s regression coefficient was significantly different from zero, B = .11, t = 3.51, p < .01, indicating that a one unit increase in the combined effects of time together and reciprocity is associated with a .11 increase in relationship self-efficacy. To understand and interpret the interaction of time together and reciprocity on relationship self-efficacy, an interaction plot is presented as Figure 1.

The data suggest that time together does not alter the relationship self-efficacy of those participants who experience average levels of partner reciprocity. However, for participants who experience high levels of partner reciprocity (i.e., one standard deviation or more above the mean), longer relationships are associated with higher levels of relationship self-efficacy. For participants who experience low levels of partner reciprocity (i.e., one standard deviation or more below the mean), longer relationships are associated with lower levels of relationship self-efficacy. Therefore, relationship self-efficacy can be predicted from time together, but only when partner reciprocity is included as a moderating variable. Longer relationships are associated with decreases in relationship self-efficacy if partner reciprocity is low and with increases in relationship self-efficacy if partner reciprocity is high. Time together is not associated with any changes in relationship self-efficacy when partner reciprocity is average.
Table 1

Moderated Regression Analysis Results: Predicting Relationship Self-Efficacy from Time Together, Reciprocity, and Their Interaction

Note. N = 334.

* p < .05. **p < .01.

Figure 1. Moderation of partner reciprocity on the relationship between time together and relationship self-efficacy.

Research Question 2

This study was conducted to determine the relationship between relationship satisfaction at Time 1 (relsat1), partner attractiveness at Time 1 (attract1), and intent to remain in the relationship at Time 2 (intent2). It is my hypothesis that while current relationship intentions can be predicted from earlier relationship satisfaction, this relationship is mediated by partner attractiveness. A mediated regression analysis (Baron & Kenny, 1986) is appropriate for this research question, because this procedure determines the degree to which a single predictor is related to a criterion through another predictor variable.

Results and Discussion

Results of the mediated regression analysis are presented in Table 2. Results from Step 1 indicated that relsat1 was a significant predictor of the mediator (attract1), B = .39, t = 6.78, p < .01. That is, a one unit increase in relationship satisfaction was associated with a .39 increase in perceived partner attractiveness. Participants who were more satisfied with their relationships also found their partners more attractive. Step 2 of the analysis indicated that relsat1 was also a significant predictor of the dependent variable (intent2), B = .34, t = 4.29, p < .01. That is, a one unit increase in relationship satisfaction was associated with a .34 increase in relationship intent. Participants with higher degrees of relationship satisfaction reported stronger intentions to remain in the relationship. Step 3 of the analysis (including both relsat1 and attract1 as predictors of intent2) indicated that the regression coefficient of relsat1 was no longer significantly different from zero, B = .09, t = 1.16, p = .25, but the coefficient of the mediator (attract2) was significantly different from zero, B = .61, t = 9.74, p < .01. The regression coefficient of partner attractiveness suggests that a one-unit increase in partner attractiveness is associated with a .61 increase in relationship intent. Participants who found their partners to be more attractive also reported stronger intentions to remain in the relationship.

The independent variable (relsat1), when included as the only predictor, significantly predicted both the dependent variable and the mediator variable. However, relsat1 lost its significance when attract1 was included in the regression as a predictor of intent2. This pattern is evidence of the full mediation of attract1, supporting my hypothesis. That is, relationship satisfaction is related to relationship intentions; however, this relationship exists only because relationship satisfaction predicts partner attractiveness, which then predicts relationship intentions.

Table 2

Results of Mediated Regression Analysis: Predicting Relationship Intentions from the Mediation of Physical Attractiveness with Relationship Satisfaction

Note. N = 334.

**p < .01.