Individual Differences in Attention During Category Learning Michael D. Lee UC Irvine Ruud Wetzels University of Amsterdam
Kruschke (1993) Condensation Experiment 8 stimuli varying in their box height and interior line position Divided into 2 categories, so that both dimensions are relevant 40 participants did 40 blocks of trials with corrective feedback
Generalized Context Model
Results of Standard GCM Analysis Marginal posterior over the attention parameter indicates both dimensions are important Familiar story, and a strong temptation to stop there …
Posterior Predictive “Violin plots” of posterior predictive for each stimuli, together with aggregated data (black line) and individual data (broken lines)
Types of Individual Differences
Allowing for Individual Differences Continuous individual differences are modeled by drawing subject parameters from an over-arching hierarchical distribution Discrete individual differences are modeled as a latent mixture, so different subjects can be drawn from different group distributions Let WinBUGS do the heavy lifting, check chains for convergence, etc, …
Results of Individual Differences Analysis Suggests there are two groups, with different attention
Bayes Factor Savage-Dickey method gives approximate Bayes Factor of 2.3 in favor there being two groups (rather than one) “artist’s impression”
Posterior Predictive Distribution Posterior predictive distributions of categorization behavior are qualitatively different tracks people’s behavior at both the sub-group and individual level
Interpretation of Groups The two groups are shown in the panels The bars show the number of “A” vs “B” category decisions made for each stimulus
Interpretation of Groups The group on the left pays attention to position, and so makes mistakes with stimuli 4 and 5
Interpretation of Groups The group on the left pays attention to position, and so makes mistakes with stimuli 4 and 5 The group on the right pays attention to height, and so makes mistakes with stimuli 2 and 7