Memory & Cognition 1992, 20 (6), 601-611 A multinomial modeling analysis of the mnemonic benefits of bizarre imagery DAYID M. RIEFER California State University, San Bernardino, California and JEFFREY N. ROUDER University ofCalifornia, Irvine, California A series of experiments was conducted to explore the cognitive processes that mediate the bi­ zarreness effect, that is, the finding that bizarre or unusual imagery is recalled better than com­ mon imagery. In all experiments, subjects were presented with noun pairs that were embedded within bizarre or common sentences in a mixed-list design. None of the experiments produced a bizarreness effect for cued recall; however, for two ofthe experiments, the bizarre noun pairs were remembered significantly better than the common pairsfor free recall. To determine ifthese differences were due to the storage or retrieval of the items, a multinomial model for the analy­ sis of imagery mediation in paired-aseociate learning was developed and applied to the data from the experiments. The model revealed that bizarre sentences benefited the retrieval of the noun pairs but not their storage within memory. The empirical and modeling results are discussed relative to previous findings and theories on the bizarreness effect. Many professional mnemonists (e.g., Lorayne & Lucas, their attention to possible mechanisms or theories to ac­ 1974) advocate the use of unusual or bizarre imagery as count for this effect. Currently no single, unified theory a method for improving memory. The mnemonic benefit exists that accounts for why bizarre imagery improves of bizarre over common imagery has been referred to as memory, or why it works under certain circumstances and the bizarreness effect, and over the past two decades a not in others. Instead, Einstein and McDaniel (1987) have number ofexperiments have been conducted to study it. described more than a half dozen different theories that However, the results of this research have been some­ can be used to account for the bizarreness effect. A care­ what mixed. Most early research indicated that bizarre ful analysis of these theories reveals that many focus on imagery is no more memorable than common imagery and storage versus retrieval explanations. For example, the that it may even have a negative effect (e.g., Bergfeld, attentional hypothesis (Merry, 1980; Wollen & Cox, Choate, & Kroll, 1982; Cox & Wollen, 1981; Wollen, 1981) states that bizarre images receive extra processing Weber, & Lowry, 1972). But more recent research (e.g., and are therefore stored better than common images. In Einstein, McDaniel, & Lackey, 1989; Hirshman, Whel­ contrast, the retrieval inhibition hypothesis (Hirshman ley, & Palij, 1989) has shown that, under certain circum­ et al., 1989; Einstein & McDaniel, 1987) postulates that stances, bizarre imagery can have a reliable and benefi­ bizarre images block the accessibility (or retrieval) of cial effect on recall. For example, McDaniel and Einstein common images. (1986) showed that the bizarreness effect occurs in mixed­ A key question, then, is whether the bizarreness effect list designs (with bizarre and common images both com­ is a storage or a retrieval phenomenon (or possibly is peting for recall) but not in between-list designs. Bizarre both). One approach to this issue is to conduct empirical images are also more memorable than common images experiments by using procedures designed to assess stor­ when they are tested for free recall (e.g., Pra Baldi, de age and retrieval separately. For example, a number of Beni, Comoldi, & Cavedon, 1985; Wollen & Cox, 1981) memory researchers have used the contrast between free or delayed recall (laccino, Dvorak, & Coler, 1989; recall, and either cued recall or recognition, to examine O'Brien & Wolford, 1982). storage and retrieval (e.g., Hanley & Morris, 1987; Ho­ Now that the bizarreness effect is a reasonably well­ gan & Kintsch, 1971; Schonfield & Robertson, 1966; established phenomenon, researchers have started to tum Tulving & Psotka, 1971). Specifically, ifa memory phe­ nomenon occurs for free recall but not for cued recall or recognition, this is often taken as evidence for a retrieval This research was supported by NSF Grant BNS-8910552to William explanation of the phenomenon because cued recall and Batchelderand D.M.R. We thank Williarn Batchelder, Gilles Einstein, and two anonymous reviewers for their helpfulconunents on this manu­ recognition presumably lessen or eliminate the importance script.Requests for reprints shouldbe sentto DavidM. Riefer, Department ofretrieval processes. This pattern of results does in fact of Psychology,California State University, SanBernardino, CA 92407. occur over many experiments on bizarre imagery. Ein- 601 Copyright 1992 Psychonomic Society, Inc. 602 RIEFER AND ROUDER stein and McDaniel (1987), in an excellent review, have as such provide a more theoretically motivated measure pointed out that, at least for immediate recall, the bizarre­ of cognitive processing than do ad hoc statistics. ness effect is strongest for free recall but is not obtained One example of this type of modeling, relevant to the for either cued recall (e.g., Andreoff & Yarmey, 1976; current issue, is a model developed by Batchelder and Webber & Marshall, 1978) or recognition (Emmerich & Riefer (1980, 1986) for measuring storage and retrieval. Ackerman, 1979; McDaniel & Einstein, 1986). Unfortunately, this model is inappropriate for studying In addition, McDaniel and Einstein (1991) and Einstein the bizarreness effect, because bizarre-imagery experi­ et al. (1989), in a series of experiments on sentence re­ ments typically employ a paired-associate paradigm in­ call, examined the effects ofbizarre imagery on two sep­ volving word pairs or sentences, whereas the storage­ arate measures: the number of sentences accessed during retrieval model analyzes category clustering data from a recall and the number of items recalled from accessed sen­ serial list-learning task. However, it should be possible tences. They showed that bizarre imagery improves the to develop a new multinomial model specifically for this first measure but not the second. These two empirical sta­ situation. The basic requirement for the development of tistics are analogous to measures developed by Tulving a multinomial model is an experimental paradigm that and Pearlstone (1966) to examine the availability (stor­ generates data in several discrete categories. Paired­ age) and accessibility (retrieval) of words that come from associate learning certainly provides such a paradigm, es­ taxonomic categories. Specifically, Tulving and Pearl­ pecially if the cued recall of paired associates can be sup­ stone used P( cat), the probability of recalling at least one plemented with some other measure, such as free recall. item from a category, as a measure of retrieval processes Thus, the purpose of this article is to present a new and [PC, the mean number ofitems recalled per category, multinomial model for the analysis ofimagery mediation as a measure of storage. Ifwe assume that a sentence and in paired-association learning and to apply the model to the elements within it comprise a "category," then num­ the issue of bizarre imagery. The model examines the free ber of sentences accessed corresponds to P(cat) and num­ and cued recall ofword pairs, and uses these data to mea­ ber of items recalled from accessed sentences corresponds sure separately the storage and retrieval processes in­ to [PC. Thus, by extending Tulving and Pearlstone's logic volved in their recall. The next section describes in de­ to this situation, one can conclude, as did Einstein et al. tail the experimental paradigm behind the model and (1989), that bizarre imagery improves the accessibility develops the model itself. Following this, the model is but not the availability of items. applied to a series ofexperiments on bizarre imagery, in Overall, the above pattern of results suggests that bi­ an attempt to test the validity of the model and explore zarre imagery aids the retrieval of items from memory the storage-retrieval basis of the bizarreness effect. but does not necessarily benefit their storage. However, a number of researchers have criticized the use of empir­ MULTINOMIAL MODEL ical statistics such as the ones above for separating stor­ age and retrieval factors. For example, some theorists Data Representation (Craik & McDowd, 1987; McNulty & Caird, 1966) have The experimental paradigm behind the model involves pointed out the methodological problems in comparing a paired-associate learning task, which traditionally is one recall and recognition performance. A detailed criticism of the methods used by researchers to study bizarre im­ has been given by Smith (1980), who argues that exam­ agery (e.g., Cornoldi, Cavedon, De Beni, & Pra Baldi, ining the differences in free recall versus cued recall or 1988; Iaccino & Sowa, 1989). In such a task, subjects recognition does not necessarily allow one to rule out en­ memorize word pairs consisting of a stimulus term (S) tirely encoding processes as a factor in recall. This is and response term (R). During test the subjects receive because successful retrieval inherently depends on a suffi­ the stimulus terms as cues to recall and are required to cient level of encoding, which might not always be equiva­ provide the response term for each stimulus. The trick lent across groups or experimental conditions. Another is to supplement these cued-recall data with a free-recall line ofcriticism