A signal detection model applied to the stimulus: Understanding covariances in face recognition experiments in the context of face sampling distributions

Abstract

We provide a description and interpretation of signal detection theory as applied to the analysis of an individual stimulus in a recognition experiment. Despite the common use of signal detection theory in this context, especially in the face recognition literature, the assumptions of the model have rarely been made explicit. In a series of simulations, we first varied the stability of d ' and C in face sampling distributions and report the pattern of correlations between the hit and false alarm rate components of the model across the simulated experiments. These kinds of correlation measures have been reported in recent face recognition papers and have been considered to be theoretically important. The simulation data we report revealed widely different correlation expectations as a function of the parameters of the face sampling distribution, making claims of theoretical importance for any particular correlation questionable. Next, we report simulations aimed at exploring the effects of face sampling distribution parameters on correlations between individual components of the signal detection model (i.e. hit and false alarm rates), and other facial measures such as typicality ratings. These data indicated that valid interpretations of such correlations need to make reference to the parameters of the relevant face sampling distribution.