We investigate the effects of word characteristics on episodic recognition memory using analyses that avoid “language-as-a-fixed-effect” fallacy. Our results demonstrate the importance of modeling word variability and show that episodic memory for words is strongly affected by item noise, as measured by the orthographic similarity between experimental items. We found that the word frequency effect was not related to the item noise effects, whereas the effect of neighborhood density, which measures the similarity of a word to all other words in the lexicon, was greatly attenuated when item noise was controlled. Our results are also consistent with a likelihood based recognition decision mechanism that produces a mirror effect by taking into account item and subject characteristics.
Journal of Memory and Language Vol. 62, Issue 1, p. 1-18