A Bayesian Analysis of Bias in Single Item Recognition for Continuous Word frequency

Abstract

The relationship between word frequency (WF), measured on a continuous scale, and recognition memory was examined in a single item recognition task. The aim was to more clearly map the relationship between word frequency and memory performance. In marked contrastContrary to standard findings of a linear relationship when between WF and recognition,is treated as discrete, we observed a curvilinear pattern. Specifically, discriminability (d’) is higher at both the low and very high ends of the WF continuum. In addition, we observe shifts in bias (C) with a conservative bias for very high frequency (HF) words between WF and memory performance. Variations of a Bayesian signal detection model were then applied to the data in order to better understand the influences WF on measures of d’ and C. The models examined contrast the current explanations of the WF effect in recognition where C does not influence performance with a model where C is free vary as a function of WF.of a linear relationship between WF and discriminability (d’) in recognition memory, with the curvilinear pattern for both d’ and bias (C) with the curvilinear patterns observed in the current data set. Implications for models of recognition memory are discussed.


Back to Table of Contents