Using must-have and can-have features to improve conceptual learning

Concepts can be taught by presenting examples and nonexamples and giving the learner feedback on whether they accurately identify the examples, but it is not clear how to select examples and nonexamples. Specifically, the degree to which examples and nonexamples should differ is unknown. Six experiments were conducted to compare conceptual learning for four stimulus sets (three sets of arbitrary stimuli and one set of biological stimuli) across up to three practice conditions: (a) nonexamples that were relatively similar to the examples, (b) nonexamples that were relatively dissimilar to the examples, and (c) examples only. Conceptual learning was measured before and after practice using tests with examples and nonexamples that were not used during practice. Including nonexamples in practice increased the likelihood of conceptual learning relative to including only examples. Using nonexamples that were more similar to the examples resulted in the most robust conceptual learning. Adding new but conceptually irrelevant features to the testing stimuli disrupted conceptual learning but less so when the practice included nonexamples that were more similar to the examples. The efficacy and efficiency of instruction for conceptual learning were affected by features of the stimuli used to practice and test conceptual learning.