Probability of target recollection varies with target-lure relatedness under the dual process signal detection model

Abstract

The ROC function plots hit and false alarm rates against one another. For item recognition the z-transformed ROC (Z-ROC) function is typically linear with slope less than one. The Dual Process Signal Detection (DPSD) and Unequal Variance Signal Detection (UVSD) models see the slope as reflecting the ratio of the lure and target evidence standard deviations. The DPSD model in addition sees the slope as co-varying with estimates of target recollection probability (R). This follows because the slope decreases with increases in the target evidence standard deviation which in turn increases with increases in the numbers of high evidence recollection items. However our results suggest that the lure evidence standard deviation, and thus the Z-ROC slope, can vary with factors seemingly unrelated to target recollection, posing problems for DPSD estimates of R. In word recognition the standard deviation of lure confidence ratings and the slope of the z-ROC function were larger with semantically related than unrelated lures. When the data for related and unrelated lures were fit separately, the standard two-parameter DPSD model, implausibly, set R lower for related than unrelated lures. The UVSD model, more plausibly, set the lure evidence standard deviation larger for related than unrelated lures.