This article reports 4 experiments concerning the effect of repetition on rated truth (the illusorytruth effect). Statements were paired with differentially credible sources (true vs. false). Old trues would be rated true on 2 bases, source recollection and statement familiarity. Oldfalses, however, would be rated false if sources were recollected, leaving the unintentional influence of familiarity as their only basis for being rated true. Even so, falses were rated truer than new statements unless sources were especially memorable. Estimates showed the contributions of the 2 influences to be independent; the intentional influence of recollection was reduced if control was impaired, but the unintentional influence of familiarity remained constant.The truth of any proposition has nothing to do with its credibility and vice versa.-Parker's law of political statements (Bloch, 1979, p. 84) Our interest in this article is with the cognitive processes that influence ratings of probable truth. Ideally, a statement should not be accepted as true without factual evidence in support of its claims. However, people often rely on memories for that evidence. It is sensible to base truth ratings on whether expressed facts corroborate or contradict remembered facts. But memory is imperfect, and it is sensible to trust some remembered facts more than others. We propose that there are two independent bases on which remembered facts are given credence when people rate truth. One basis is recollection: A statement will be accepted as true if it corroborates remembered facts that are associated with a known, credible source, and it will be rejected as false if the facts are associated with a discredited source. The other basis is familiarity: A statement will seem true if it expresses facts that feel familiar. We propose, furthermore, that these two bases differ in the extent to which their influence is controlled rather than automatic. Recollection of source is a controlled use of memory, and its influence on rated truth is intentional. In contrast, increased familiarity is an automatic consequence of exposure, and its influence on rated truth is unintentional.This research was funded by Operating Grant OGP0008122 from the Natural Sciences and Engineering Research Council of Canada; a certificate of approval from the McMaster Ethics Committee is on file. Part of the research was conducted as an honors thesis by Suzanne Farinacci, who was supervised by Ian Maynard Begg.We are grateful for the advice and discussion from our McMaster colleagues, including Marcia Barnes, Lee Brooks, Vincenza Gruppuso, Larry Jacoby, Janine Jennings, Steve Lindsay, Douglas Needham, Rohan Robertson, James Taylor, Jeff Toth, and Andy Yonelinas. We also thank Colleen Kelley of Macalester College for her counsel. Special thanks are due to Jim Anas for being a male voice on the tapes.Correspondence concerning this article should be addressed to Ian Maynard Begg, Department of Psychology, McMaster University, Hamilton, Ontario, Canada L8S 4K1.Our thesis is ...

No abstract

(1995) claim to have shown that the independence assumption underlying the process-dissociation procedure (L. L. Jacoby, 1991) is not justified. They argued that correlations between processes at the level of items can result in an underestimation of automatic processes large enough to produce artifactual dissociations between process estimates. In contrast, the authors show that the effects of extremely high correlations at the level of items are likely to be trivial, and not differential across conditions. Curran and Hintzman's dissociations probably reflect violations of boundary conditions for use of the process-dissociation procedure, rather than violations of independence. It is important to distinguish between automatic and consciously controlled memory processes. For example, although amnesic patients are often unable to consciously remember previously presented words on direct memory tests, such as recall or recognition tests, they use the words on indirect memory tests, such as stem-or fragmentcompletion tests, more often than would be expected by chance (Moscovitch, Vriezen, & Gottstein, 1993). Similar dissociations are found in people with normally functioning memory (Roediger & McDermott, 1993). Comparing direct and indirect memory tests has significantly advanced our understanding of automatic and controlled processes. However, performance rarely reflects only one process acting in isolation; that is, controlled processes often influence performance on indirect memory tests (Holender, 1986; Toth, Reingold, & Jacoby, 1994), and automatic processes affect performance on direct memory tests (Jacoby, Toth, & Yonelinas, 1993). The process-dissociation procedure (Jacoby, 1991) was designed to separate automatic and controlled memory processes when both are affecting performance. As a brief introduction to the process-dissociation procedure, consider Experiment IB reported by Jacoby et al. (1993). Subjects studied words under full or divided attention and then were tested with word stems (e.g., mot for

Words that are generated as responses to incomplete stimuli are remembered better than complete words that are read. The present research shows that the generation effect occurs only if both reading and generating are done in the same list. Comparisons with unmixed reading controls reveal that the effect occurs because the read words are remembered poorly; the generated ones are no better than the controls. Therefore, the question for theories is not why generating helps memory, but why the demand to generate makes people lazy readers. Furthermore, cued recall of pairs of unrelated words is worse if they were generated than if they were read. We propose that generation is a demand to identify words as independent events. There is nothing special about generating, but the demand to generate is special; it hurts reading.

The generation effect occurs if people remember items they complete from fragments better than complete items they read. Four experiments investigate two questions. When does the effect occur, and why does it do so? Targets generated in related contexts are recognized better than read targets, and they are recalled better with the contexts as cues; the contexts are recognized equally well, and the relation between the context and target is not enhanced by generation. Furthermore, generated items exceed items read in pure lists even when the read ones from the mixed list are no worse than the controls. The generation effect is real; it is not an artifact. However, there is nothing special about generation. Generating is a type of encoding, and like any other type of encoding, its effects are maximal on tests that require subjects to do again whatever they did at study. Generating makes targets distinctive by contrasting them with other relatives of the context, and, as a result, the targets enjoy benefits in later discriminations within their family.Subjects need to do something to make ANIMAL-H_R_E meaningful that they need not do for ANIMAL-HORSE; horse can be read from HORSE, but it must be generated from H__R_E. The "generation effect" (Slamecka & Graf, 1978; cf. Jacoby, 1978) occurs if generated items are remembered better than items that are read. Many experimenters using many procedures have presented lists of complete items and fragments and have found that the items generated from fragments are remembered better than the read ones. The generation effect transcends even success or failure in the generation task; Kane and Anderson (1978) and Slamecka and Fevreiski (1983; see also McElroy & Slamecka, 1982) found no change in the size of the generation effect if subjects succeeded or failed in the attempt, provided that the complete target was given as feedback after the attempt.Slamecka and Katsaiti (1987) reviewed existing explanations of the generation effect and concluded that none were satisfactory. However, some recent accounts (Hirshman & Bjork, 1988;McDaniel, Waddill, & Einstein, 1988) seem promising. In agreement with these accounts, we will establish