Figure 5.

Regression analyses for an alternative measure of bias. We quantified bias as the probability of giving a “Category 1” response rather than as the SDT criterion and again observed a similar relationship between intrinsic bias, P_neutral (resp = Cat1), and the midpoint of the predictive criteria, P_PredMid (resp = Cat1). As in the SDT analyses, the obtained β value of .54 was significantly different than both 0 and 1. Each dot represents one subject and colors indicate the experiment in which the subject participated. The thick black line indicates the line of best fit, whereas the thin gray lines indicate the predictions of the “eliminate” and “maintain” hypotheses.