Implications for Dual-Process Theories of Recognition Memory

Dual-process theory, which holds that recognition decisions can be based on recollection or familiarity, has long seemed incompatible with signal detection theory, which holds that recognition decisions are based on a singular, continuous memory-strength variable. Formal dual-process models typically regard familiarity as a continuous process (i.e., familiarity comes in degrees), but they construe recollection as a categorical process (i.e., recollection either occurs or does not occur). A continuous process is characterized by a graded relationship between confidence and accuracy, whereas a categorical process is characterized by a binary relationship such that high confidence is associated with high accuracy but all lower degrees of confidence are associated with chance accuracy. Using a source-memory procedure, we found that the relationship between confidence and source-recollection accuracywas graded. Because recollection, like familiarity, is a continuous process, dual-process theory is more compatible with signal detection theory than previously thought. A long-standing theory holds that recognition memory decisions are supported by two processes, namely, recollection and familiarity. The following anecdote, offered by Mandler (1980), describes a common experience that illustrates how these two processes sometimes unfold in real time: Consider seeing a man on a bus whom you are sure that you have seen before; you ‘‘know’’ him in that sense. Such a recognition is usually followed by a search process asking, in effect, Where could I know him from? Who is he? The search process generates likely contexts (Do I know him from work; is he a movie star, a TV commentator, the milkman?). Eventually the search may end with the insight, That’s the butcher from the supermarket! (pp. 252–253) The initial sense of familiarity refers to a memory signal pertaining to the item itself (based, perhaps, on its perceptual features), whereas the subsequent awareness of recollection refers to the retrieval of source information that is associated with that item. Familiarity is widely assumed to be a continuous process in the sense that it is experienced in degrees. Low degrees of familiarity are associated with low confidence and low accuracy, whereas high degrees of familiarity are associated with high confidence and high accuracy. By contrast, the recollection process is almost always thought to be categorical in that, theoretically, it either occurs (yielding high confidence and high accuracy) or does not occur. For continuous processes, the notion of a decision criterion almost inescapably comes into play. Thus, for example, on a typical old/new recognition memory test, the participant’s task is to distinguish between targets (e.g., words that were presented on a previous list) and lures (e.g., words that were not presented on a previous list). Although the targets are likely to be relatively familiar because of their recent appearance on a list, the lures are associated with some degree of familiarity as well. Thus, for a decision that is based on familiarity, a participant must decide how much familiarity is enough to decide that the item is old. In other words, the participant must set a criterion familiarity value. An early dual-process model proposed by Atkinson and his colleagues envisioned two criteria for familiarity-based decisions (Atkinson & Juola, 1973, 1974). According to this model, if the degree of familiarity associated with a test item was strong enough to fall above a high criterion or weak enough to fall below a low criterion, then a familiarity-based decision would be made (old or new, respectively). If the degree of familiarity instead fell between the two criteria (i.e., if familiarity was of intermediate Address correspondence to John T. Wixted, Department of Psychology, 0109, University of California, San Diego, La Jolla, CA 92093-0109, e-mail: [email protected]. PSYCHOLOGICAL SCIENCE Volume 20—Number 4 509 Copyright r 2009 Association for Psychological Science by Peter Wais on September 3, 2013 pss.sagepub.com Downloaded from strength), then a retrieval search would be initiated. That search was assumed to either succeed (in which case the item was declared to be old) or fail (in which case it was declared to be new). Thus, in this model, recollection was construed as a categorical process—one that does not involve a decision criterion. Mandler (1980) also pointed out that a decision criterion plays a role in familiarity-based decisions, but again treated recollection as a categorical process that either succeeds or does not succeed. The same approach to recollection and familiarity is taken in studies that use the process-dissociation procedure to obtain quantitative estimates of recollection and familiarity (Jacoby, 1991). In computing those estimates, recollection is again considered to be a categorical process, whereas familiarity is assumed to be a continuous process that involves a decision criterion (Jacoby, Toth, & Yonelinas, 1993). Finally, Yonelinas (1994) proposed a model in which recollection was assumed to be a categorical process that always yields high confidence and does not involve a decision criterion, whereas familiarity was regarded as a continuous signal detection process that does involve a decision criterion. A common feature of all of these dual-process models, in addition to the fact that they regard recollection as a categorical process, is that they assume that individual recognition decisions are based either on one process or on the other. That is, according to all of these models, old/new decisions about items that elicit recollection are based solely on recollection, whereas old/new decisions about items that do not elicit recollection are based solely on familiarity. This is a natural way to think if one begins with the assumption that recollection is categorical. That is, in the categorical view, the occurrence of recollection would yield high confidence that an item was previously encountered, thereby rendering unnecessary any consideration of familiarity. But when recollection fails completely, the only recourse would be to rely on familiarity. An alternative view is that recollection and familiarity are both continuous processes that are aggregated into a memorystrength signal (Wixted, 2007). According to this account, both processes play a role in an old/new decision about an individual test item. The core difference between this model and all earlier dual-process models is its assumption that recollection is a continuous process (i.e., it comes in degrees), not a categorical process. If recollection were a continuous process (e.g., Dodson, Holland, & Shimamura, 1998; Johnson, Hashtroudi, & Lindsay, 1993), then high degrees of recollection would result in high confidence and high accuracy, but low degrees of recollection would result in low confidence and low accuracy. In that respect, recollection would be like familiarity. In other respects, however, the two processes would remain distinct. That is, according to this view, familiarity is a fast process that involves the retrieval of information about the item per se, whereas recollection is a slower process that involves the retrieval of associated contextual information. But because recollection is assumed to occur in graded fashion, any degree of recollection that happens to occur would add to the extant familiarity-based memory signal instead of usurping it. The research we report in this article was concerned with differentiating between the categorical and continuous views of recollection. We investigated this issue by using a sourcememory procedure (Johnson et al., 1993), which is commonly used to study the recollection process. In this procedure, some items on a list are associated with one source attribute (e.g., the color red), and others are associated with a different source attribute (e.g., the color blue). On a later recognition test, participants are presented with test items in a source-neutral fashion (e.g., in black) and asked to recollect the original source attribute (i.e., a binary decision between Source A and Source B). In this experiment, however, participants were instead asked to rate their confidence in the item’s source using a 20-point scale, with 1 representing highest confidence in Source A (e.g., blue) and 20 representing highest confidence in Source B (e.g., red). In a test like this, the familiarity of the test item is not diagnostic of its source because the items from both sources recently appeared on the same study list. The categorical and continuous views of recollection make contrasting predictions about the relationship between the confidence in a source decision and the accuracy of that decision. The categorical view of recollection predicts that the relationship will be a step function. For example, an all-or-none version of the categorical model predicts that accuracy will be high for ratings made with the highest confidence (i.e., for ratings of 1 or 20) and will be no better than chance for all other ratings. By contrast, the continuous view of recollection predicts that the relationship will fall off in graded fashion (i.e., accuracy will be highest for ratings of 1 or 20, next highest for ratings of 2 and 19, and so on). In prior investigations into this issue, participants were first asked to make an old/new decision using a 6-point confidence scale and then asked for a source decision. Yonelinas (2001) reported that source accuracy was above chance only for old/ new decisions that were made with the highest level of confidence (a pattern consistent with the categorical view of recollection), but Wixted (2007) reviewed results from several other studies showing that source recollection was above chance even for old/new decisions that were made with low and medium levels of confidence (a pattern consistent with the continuous view of recollection). In the present experiment, we tested the relationship between confidence and accuracy for the source decision itself to directly test the categorical and continuous accounts, and we used a 20-point scale to examine the relationship over a wide range of confidence.