Categorical judgments do not modify sensory representations in working memory
Fig 11
Performance of Models 2b (resampled) and 2c (reweighted).
(A) Normalized log-likelihood comparison for Experiment 1. All models perform well compared to Model 1a (prior only), confirming that subjects maintain and reuse information about stimulus uncertainty on a per trial basis. Models 2b and 2c, however, perform consistently better than Model 2a. (B) Both models better predict subjects’ mean estimates than Model 1b (MSE and correlation; see Fig 5C). Predictions of Model 2c show a similar deviation from the data as the predictions of Model 2a. (C) Normalized log-likelihood comparison for Experiment 2. Model 2b consistently, but not always significantly, best predicts the data except for P1 (non-naïve). (D) In contrast to Model 1b, Models 2b and 2c both correctly predict mean estimates higher on the cw than the ccw side (see Fig 8B). Correlation and MSE measures indicate little difference between Model 2b and 2c; their predictions are better than those of Model 1b and 2a (see Fig 8C). All model predictions are parameter free and based on the parameter values of the fit self-consistent observer model to the data in correct trials. Errorbars indicate 95% confidence intervals computed from 200 bootstrapped samples.