False recognition and word length: A reanalysis of Roediger, Watson, McDermott, and Gallo (2001) and some new data

Abstract

Roediger, Watson, McDermott, and Gallo (2001) reported a multiple regression analysis of the variables that predicted rates of false recall and recognition across lists in the Deese/Roediger-McDermott paradigm. They concluded that false recollection was predictable from the backward associative strength of critical words and list words and the veridical recall level of list words, with no independent contribution of frequency, concreteness, or length of critical words. A reanalysis of their data shows that critical word length does contribute to false recognition when it is measured relative to the length of other words in the list. Relative word length in the form of an unsigned z-score has a larger correlation with false recognition than any of the variables used by Roediger et al. and is also independently predictive of false recognition. This relationship was confirmed by the results of two recognition experiments in which false positives were significantly less frequent for words of lengths that never occurred in a studied list than for words of lengths that did occur in the study list.