The puzzle of study time allocation for the most challenging items.

Learners often allocate more study time to challenging items than to easier ones. Nevertheless, both predicted and actual memory performance are typically worse for difficult than for easier items. The resulting inverse relations between people's predictions of their memory performance (judgments of learning; JOLs) and self-paced study time (ST) are often explained by bottom-up, data-driven ST allocation that is based on fluency. However, we demonstrate robust inverted U-shaped relations between JOLs and ST that cannot be explained by data-driven ST allocation alone. Consequently, we explored how two models of top-down, strategic ST allocation account for curvilinear JOL-ST relations. First, according to the Region of Proximal Learning model, people stop quickly on items for which they experience too little progress in learning. Second, according to the Diminishing Criterion Model, people set a time limit and stop studying when this time limit is reached. In three experiments, we manipulated motivation with different methods and examined which model best described JOL-ST relations. Consistent with the Diminishing Criterion Model but not with the Region of Proximal Learning model, results revealed that curvilinearity was due to people setting a time limit.