All three theories could not be correct. Aided by an insight by Howard Garland, Locke, Motowidlo, and Bobko (1986) resolved the puzzle. When goal level is held constant, that is, within any given goal group, the positive linear relationship asserted by expectancy theory is correct. Between groups, when goal level is varied, the relationship is negative. This does not contradict expectancy theory, because expectancy theory assumes that the referent is fixed. When Bandura’s self-efficacy measure is used (which averages a person’s confidence estimates across multiple performance outcome levels) both the within and between group associations are positive. The curvilinear relationship between expectancy, or goal difficulty, and performance as suggested by Atkinson replicates only when there are a substantial number of people in the hard goal condition who reject their goals (Erez and Zidon, 1984; Locke and Latham, 1990).