The detection frequency denotes the fraction of all trials that are estimated as detections by the model (
Figure 11). Note that the assumption underlying the model is that the participant either detects the stimulus with a certain precision, reports an opposite direction, or guesses. Thus, performance can be improved by increasing the number of detections or by increasing the precision for the detections. The detection frequency is, therefore, independent of the precision of the detection distribution (reported separately elsewhere in this article). There was an increase in detections with coherence for both response methods. The number of detections plateaus at approximately 50% coherence for responses reported with the trackball using BM and TM, whereas it seemed to increase further for WM at 100% coherence (
Figure 11A). For responses recorded with the rotating bar, the number of detections plateau in the BM condition at 50%, while they increased for TM and WM at 100% coherence (
Figure 11B).
A mixed-effects repeated-measures ANOVA showed a main effect of coherence on detection frequency, F(2.99, 197.26) = 633.89, p < 0.001 (Greenhouse–Geisser correction for sphericity). A post hoc pairwise comparison revealed a significant increase in detection frequency for each coherence from 0% to 50% (pairwise comparison p < 0.001, Bonferroni corrected for multiple comparisons), but not between 50% and 100% (p = 0.074, Bonferroni corrected for multiple comparisons). The RDK type had a main effect on detection frequency, F(2, 66) = 14.90, p < 0.001, as well. A post hoc pairwise comparison between stimulus types revealed that TM elicits fewer detections than BM (p < 0.001, Bonferroni corrected for multiple comparisons) and WM (p = 0.002, Bonferroni corrected for multiple comparisons), although no difference between BM and WM was found (p = 0.235, Bonferroni corrected for multiple comparisons). There was no main effect of response methods on detection frequency, F(1, 66) = 0.15, p = 0.704.