On the difference between strength-based and frequency-based mirror effects in recognition memory.

Abstract

A mirror effect can be produced by manipulating word class (e.g., high vs. low frequency) or by manipulating strength (e.g., short vs. long study time). The results of 5 experiments reported here suggest that a strength-based mirror effect is caused by a shift in the location of the decision criterion, whereas a frequency-based mirror effect occurs although the criterion remains fixed with respect to word frequency. Evidence supporting these claims is provided by a series of studies in which high frequency (HF) words were differentially strengthened (and sometimes differentially colored) during list presentation. That manipulation increased the HF hit rate above that for low frequency (LF) words without selectively decreasing the HF false alarm rate, just as a fixed-criterion account of the word-frequency mirror effect predicts.In recent years, a well-known empirical regularity known as the mirror effect has commanded a great deal of attention. The mirror effect refers to the relationship between hit and false alarm rates in two conditions associated with different levels of recognition accuracy. Specifically, a mirror effect is said to exist when the condition associated with more accurate recognition performance is characterized by both a higher hit rate and a lower false alarm rate than the less accurate condition. This effect is so reliably observed that Glanzer, Adams, Iverson, and Kim (1993) described it as a "regularity of recognition memory."The mirror effect can be produced by manipulating either the class or the strength of the items presented for study. The most common class manipulation is based on word frequency (high vs. low), whereas strength is usually manipulated by varying study time or number of item presentations. The consistency of the mirror effect across different methods of manipulating recognition accuracy suggests the influence of a single underlying mechanism. In terms of signal detection theory, that mechanism is often thought to be a shift in the criterion for deciding whether to respond "yes" or "no" to a test item. Figure 1 illustrates the criterion-shift argument for a strength manipulation using the standard assumptions of signal detection theory. This model assumes that the decision axis represents a strength-of-evidence variable, such as familiarity. According to this account, the familiarity values associated with the target items and lure items are both normally distributed, with the mean of the target distribution Strength-Based Mirror Effects

Source-constrained retrieval influences the encoding of new information

“…We used a paired-samples t test to compare the recognition scores (hits -false alarms) for words that had been studied either once or three times. In line with the extensive repetition literature (e.g., Bentin & Moscovitch, 1988;Murnane & Shiffrin, 1991;Ratcliff, Clark, & Shiffrin, 1990;Stretch & Wixted, 1998), participants recognized items that had been studied three times significantly better than items studied only once, t(23)=5.76, p<.001. Separate t tests demonstrated that there were (1) more hits, t(23)=5.59, p<.001, and (2) fewer false alarms, t(23)=2.19, p<.05, for words presented three times than for words presented once.…”

Word frequency and receiver operating characteristic curves in recognition memory: Evidence for a dual-process interpretation.

“…Such a pattern of results (i.e., that higher hit-rates are often accompanied by lower false-alarm rates) has been dubbed the mirror effect by Glanzer and colleagues (Glanzer & Adams, 1985;Glanzer, Adams, Iverson, & Kim, 1993). Although the mirror effect is more general than manipulations of word frequency (e.g., encoding manipulations such as increased study time also produce mirror effects; Hirshman, 1995;Ratcliff, Clark, & Shiffrin, 1990;Stretch & Wixted, 1998), accounting for the word-frequency mirror effect has proven especially difficult for many theories of recognition memory. In particular, the word-frequency mirror effect has been difficult for global matching models (Gillund & Shiffrin, 1984;Hintzman, 1988;Murdock, 1982) to explain.…”

Towards a Competitive Learning Model of Mirror Effects in Yes/No Recognition Memory Tests

“…The approach takes inspiration from Bayesian models of recognition memory, especially from McClelland and Chappell's subjective likelihood model. 16 One of the most significant differences between the approach presented here and existing Bayesian models 8,12,16,17 is that we obtain a mirror effect without an explicit likelihood ratio calculation. We have shown here that the two basic classes of mirror effects can be generated from a simple, competitive learning neural network in which the values for the familiarity axis are directly generated from the activation of the winning detectors.…”

“…This is based on the observation that participants do not appreciate that low frequency items are more memorable, but instead adopt a single response criterion/threshold, even if provided with explicit cues about the item type. 8 …”

“…In contrast, weak old items elicited a lower level of memory activation, so that a less stringent criterion/threshold is adopted. 8 In the Type II mirror effect simulation (stimulus class based), a single decision criterion/threshold is calculated, which collapses new and old items across low-and high frequency. This is based on the observation that participants do not appreciate that low frequency items are more memorable, but instead adopt a single response criterion/threshold, even if provided with explicit cues about the item type.…”