Retrieval constraints and the mirror effect.

Six experiments tested the hypothesis that the mirror effect in recognition memory reflects a deliberate, postretrieval assessment of the test item's memorability. Both word frequency and concreteness were varied, and constraints at retrieval were manipulated in 2 ways: Three experiments compared recognition tested either alone or while performing a secondary task, and 3 experiments used the response-signal method to control recognition processing time. Contrary to the hypothesis, the mirror effect was not eliminated or attenuated by either kind of retrieval constraint. Moreover, both retrieval manipulations induced mirror effects of their own. Current recognition-memory theories appear inadequate to explain the results. It is suggested that the mirror effect be related to the full range of patterns of hits and false alarms, and that these 2 measures routinely be supplemented by measures of discriminability and bias. Variables that influence recognition memory commonly have opposite effects on the proportions of hits and of false alarms. That is, if subjects are more likely to call a studied item old in Condition A than in Condition B, they tend to be less likely to call a distractor item old in Condition A than in Condition B. This interaction between condition and memory status has been called the "mirror effect " (see Glanzer & Adams, 1985, 1990). The best known examples of the mirror effect are those attributable to item variables, such as the natural-language frequency and concreteness of words: Lowfrequency (LF) words typically produce higher hit rates and lower false-alarm rates than do high-frequency (HF) words, and concrete words typically give higher hit rates and lower false-alarm rates than do abstract words (Glanzer & Adams, 1985, 1990). However, manipulations producing the mirror effect also include subject variables such as age (Backman, 1991) and controlled variables such as presentation rate (Ratcliff, Sheu, & Gronlund, 1992) and recency (Glanzer,