One-trial learning with control of item-difficulty

Ratings of difficulty were obtained for each of 2,500 letter-number pairs. The ratings were used to construct three pools of items: Homogeneous Easy, Homogeneous Difficult, and Heterogeneous. Rock's (1957) experiment was conducted thrice, once with each pool of rated items. Each replication used 30 subjects. The manipulation of rated item-difficulty permitted a direct attack on the widely discussed problem of item-selection in Rock's procedure. Notwithstanding the control of item-selection, the performance of subjects in the drop-out condition did not differ significantly from the performance of those who learned in the ordinary way, with repetition. The geometric distribution provided a good fit to the obtained distributions of waiting times to the first correct response to an item. This finding accords well with an all-or-none view of associative learning.