Hypothesis testing in the learning of positive and negative linear functions

Subjects in two experiments had to learn to predict a numerical criterion, Y, on the basis of a numerical cue, X. Experiment 1 varied the algebraic sign of r XY and cue-criterion compatibility (similarity) in an attempt to discover why subjects have difficulty learning to use cues that are negatively correlated with a criterion. Compatibility was found to be the major determinant of learning. Experiment 2 showed that learning of negative linear functions was hindered by the presence of intermediate values of X, which were associated with intermediate values of Y. Removal of these intermediate values made the negative linear function much easier to learn. The results of both experiments were interpreted in terms of the subjects' ability to test hypotheses about functional relationships.