On the positive and negative effects of self-efficacy on performance: Reward as a moderator

ABSTRACT

We address the controversy regarding the sign of the effect of self-efficacy on performance. We propose that when reward is high, self-efficacy has a positive effect on performance, whereas when reward is low, it has a negative effect. This hypothesis was tested in three experiments (with sample sizes of 76, 99, and 111) in which reward and self-efficacy were manipulated and performance was measured using different manipulations and tasks in both within-factor and between-factor designs. The results supported the hypothesis that reward moderates the effect of self-efficacy on performance. When reward was high, the effect of self-efficacy on performance was positive, whereas when reward was low, the effect of self-efficacy on performance was negative.

Notes

¹ We note that although for the purpose of the current analysis the concepts of expectancy and self-efficacy are similar, there are contexts in which they are clearly distinct. For example, whereas self-efficacy does not depend on the social environment, expectancy does (see Bandura, 1982).

² Note that the model underlying Equation 3 assumes segregation between gain and cost (Kahneman & Tversky, 1979), unlike normative models, which view expected value as a multiplicative function of integrated gains and costs (value) multiplied by probability.

³ We chose two sequential trials (second and third), because internally attributed repeated success is a major determinant of self-efficacy (Bandura, 1977). If success occurred only once, it may have been attributed to luck. We also boosted self-efficacy in the fifth trial, because the nonmanipulated performance feedback in the fourth trial was most likely to be lower than the manipulated (positive) feedback in Trials 2 and 3, thus decreasing self-efficacy. In choosing the third try in Trials 2, 3, and 5 (the manipulated trials) to inform participants that they found the solution, we followed Vancouver et al. (2002). They reasoned that it was unlikely that one could find a solution by or before this try, but to do so did not represent an impossible level of performance, either. Indeed, Vancouver et al. (2001) reported that, on average, participants took 6.4 tries to get the solution.

⁴ To make the experiment less cumbersome, we did not use the strength measure of self-efficacy. We also note that the strength measure often yields results that are similar (e.g., Vancouver et al., 2002; Vancouver et al., 2001) or inferior (Maurer & Andrews, 2000; Vancouver & Kendall, 2006, Footnote 6) to those of the magnitude measure.

⁵ Because our contrasts hypotheses are unidirectional, the reported levels of significance are one-tailed.