Prior Performance and Goal Progress as Moderators of the Relationship Between Self-Efficacy and Performance

Abstract

Over the last several years, the conventional view of self-efficacy as a positive influence on performance has been called into question. Researchers have identified a negative relationship between self-efficacy and performance when examined via within-person analyses, even in the presence of large positive between-person relationships. The current study proposes that the within-person relationship between self-efficacy and subsequent performance is moderated by one's degree of prior success or failure. Using a multitrial task, support was found for the proposed model. Following poor or substandard performances, self-efficacy was positively related to subsequent performance. However, following more successful prior performances, self-efficacy was negatively related to subsequent performance. Implications of these findings for theory and research on work motivation are discussed.

Notes

¹Task complexity was manipulated for a related research question by varying whether any color could be repeated in the code. Under the simple condition, no color could appear in the code more than once. Under the complex condition, any color could appear in the code any number of times (up to four). Beyond a significant main effect of complexity on the number of tries required to find the solution set, no direct or interactive effects of complexity were observed on any of the study variables. Therefore, for clarity of presentation, complexity is not discussed in subsequent presentation of the study and its results.

²Aside from easing interpretation, this recoding had no impact on the results.

³The software program “Hierarchical Linear Modeling” that is commonly used for implementing the HLM analysis technique assumes that the residuals are independent, which may not be a valid assumption given the nature of the data in this study. By using the SAS mixed procedure to conduct the HLM analyses, we were able to utilize an unstructured error covariance matrix in all analyses, which requires minimal assumptions concerning the data. Alternative covariance structures were examined as well, with no change in the conclusions resulting from these analyses.

*p < .05.

***p < .001.

*p < .05.

***p < .001.