Figure 2.

Serial dependence effects for different conditions, for the aggregate observer

(A) Average bias (reproduced minus physical angle) of the current trial for neutral stimuli preceded by neutral stimuli, plotted against the previous minus current physical orientation (by convention, counterclockwise [CCW] is positive). The gray curve is the best fitting ∂G fit (see STAR Methods), with the star marking its center. Vertical error bars indicate ±1 SEM of the aggregate observer data (about 250–300 points each) and horizontal error bar ±1 SEM of the center of curve fit obtained by bootstrapping.

(B) Results for neutral trials preceded by illusory trials, separately for −15° (CW, dark green) and +15° (CCW, light green) flankers (conventions as for A). The color-coded vertical dashed lines show the average bias needed to offset that induced by the flankers for that condition. These pass within 1 SEM of the centers of the ∂G fits. Dashed vertical lines indicate point of maximal serial effect if the effect arose after (thick lines) or before (thin line) the illusion had exerted its effect.

(C) Results for illusory trials (navy blue −15° flankers; blue +15°) preceded by neutral trials (conventions as for A). Note the difference in the ordinates, reflecting the effect of the flankers on the current trial. Besides this difference, both best-fitting ∂G curves pass near zero, within 1 SEM. Dashed vertical lines indicate the predicted maximal serial effect, if the priors act after (thick lines) or before (thin line) the illusion had exerted its effect.