Abstract
In two studies, participants studied 30 lists of 50 words and were tested on 30 lists of 100 words. Item-level multiple regression analyses were conducted on hits, false alarms, hits minus false alarms, d', and C. The predictor variables were objective frequency, subjective frequency, imageability, orthographic similarity, phonological similarity, phonological-to-orthographic N (PON), age of acquisition (AoA), and word length. The regression equations accounted for 45.9% of the variance in hit rates, 14.9% of the variance in false alarm rates, and 29.2% of the variance in hits minus false alarms. Other noteworthy results were that: (a) hit rates positively correlated with false alarms, (b) objective frequency negatively correlated with both hit rates and false alarm rates, (c) AoA positively correlated with hit rates and negatively correlated with false alarm rates, (d) length negatively correlated with hit rates and positively correlated with false alarm rates, (e) orthographic uniqueness was positively correlated with hit rates and negatively correlated with false alarms, (f) PON positively correlated with false alarm rates, (g) imageability produced the typical mirror pattern, and (h) imageability and length were the strongest predictors of performance. Results were largely compatible with predictions made by single- and dual-process theories of recognition memory.