Purchase this article with an account.
Carlos Cabrera, Zhong-Lin Lu, Barbara Dosher; Testing the Stationary Variability Assumption in Signal Detection Theory. Journal of Vision 2014;14(10):56. doi: https://doi.org/10.1167/14.10.56.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Signal Detection Theory (SDT; Green and Swets, 1966) is possibly the most successful theoretical framework in cognitive psychology and features prominently in a variety of other research, clinical, and applied settings. Investigators frequently invoke SDT to estimate sensitivity and response bias by inferring observers' internal representations of stimuli using the z-transformed receiver operating characteristic (zROC). The zROC analysis assumes stationary distributions of internal representations at different criteria along the decision axis. Here we develop a procedure to test this assumption with a multi-pass paradigm (Burgess & Colborne, 1988; Green, 1964) in which subjects respond to multiple presentations of identical stimuli in order to estimate total variance in noise and signal+noise trials across different bias manipulations. We deployed this procedure in a multi-pass Yes/No visual detection experiment. Subjects responded 'Yes' or 'No' to stimuli consisting of a Gabor temporally combined with external noise (signal present) or external noise alone (signal absent). We estimated the total internal noise at three different bias manipulations: P[signal] = 70%, P[signal] = 50%, and P[signal] = 30%. Bias manipulations did not significantly alter mean signal strength, but did lead to significant differences in criterion placement and total internal noise. Changes in internal noise at different bias levels suggest that decision noise contributes to response variability and that this noise component depends on criterion position. We propose utilizing this multi-pass procedure at only a single, unbiased (P[signal] = 50%) condition to avoid altering underlying distributions with bias manipulations to provide a robust estimate of the ratio of the variability of the internal representations in noise alone and signal+ noise conditions. This procedure also avoids more costly and time consuming bias procedures, and sidesteps the varying decision noise of confidence ratings (Mueller & Weidemann, 2008; Wickelgren, 1968) or other criterion-dependent features of the internal representations (Balakrishnan, 1999).
Meeting abstract presented at VSS 2014
This PDF is available to Subscribers Only