Median threshold and slope estimates obtained from 2,000 simulations each. Lapse rate was free to vary in all cases but using different fitting strategies. Leftmost symbols in each of the 'quads' of symbols is an attempted replication of Wichmann & Hill (2001) closely following their reported procedure. Significant and systematic biases in threshold and slope estimates are obtained. Second from left shows results obtained using same procedure but with modified placement: highest stimulus intensity in placement was changed to asymptotic performance level (specifically, that intensity at which the generating F evaluates to 0.9999). Third from left shows results obtained with modified placement but modeling probability of correct response at the asymptotic stimulus intensity as 1 - lapse rate (i.e., assuming that error at this intensity can only be due to lapses). Rightmost symbols shows results obtained with modified placement also. Probability correct at asymptotic level is again modeled as 1 - lapse rate, but threshold and slope are then estimated by fitting observed responses at other intensities using obtained lapse rate estimate as fixed value. (nAPLE: 'non-Asymptotic Performance Lapse Estimation', jAPLE: 'joint APLE', iAPLE: 'isolated APLE'). Error bars are minimum (thick lines) and maximum (thin lines) 'Quarter 68% Confidence Intervals', which would correspond to roughly 0.25 SD of the parameter estimate distribution. Click on link below for more information.
Nick Prins
Psychology
Ole Miss
Nick Prins