Participants in recognition memory studies are now often asked to partition recognized items into ones that are accompanied by some recollective experience (those they remember) and ones that are not so accompanied (but which they know were previously encountered), Rather than detecting separate memory systems, such attempts to distinguish between remembering and knowing are better understood as a division of positive recognition responses into those that lie above a second decision criterion (remember) and those that do not (know), As such, the amount of memory associated with knowing is strongly dependent on the placement of the decision criteria, A meta-analysis of published data and a simple experiment tested predictions from the decision process analysis of remember/know responses, The standard recognition memory task is one in which participants are presented with a list of items to be remembered and then are given a longer list containing the original items and a number ofdistractors. They are asked to say "yes" to those items they remember from the initiallist and "no" to those they do not remember seeing in the first list Participants are often uncertain about whether an item is an old one they saw before or not They can be induced to be more or less cautious in partitioning the items into old ones and new ones. They are more confident about responses to some items than others, and they can assign meaningful confidence ratings to their yes and no responses. Most researchers who use the recognition task now realize that the hit rate by itself, the proportion of times a participant says "yes" to old items, is not a good index of memory since it can be so easily influenced by the criterion the subject sets, One must combine the hit rate with the false-alarm rate, the proportion of times a participant says "yes" to new items, to calculate a measure of memory, such as d' or A', that is independent of a criterion or bias measure, such as Cor B D . Snodgrass and Corwin (1988) reviewed a variety of such measures, Recently, there has been an increased tendency to use the hit rate as ifit adequately measures memory. This has occurred when experimenters have asked participants to

Recent years have seen an expanded interest in recognition memory tasks. This resurgence of interest has also renewed concerns with measurement problems. Comparing 4 models of recognition memory, Snodgrass and Corwin (1988) found that measures of bias from the distribution-free (nonparametric) model were inadequate. However, their analysis was based on bias measures that can be shown a priori to be nonindependent of discrimination. This article traces the history of the nonparametric model and develops a better measure of bias. The consequence of developing this better measure is that the nonparametric model deserves serious consideration.

No abstract

Signal-detection analysis was applied to a continuous short-term recognition memory task for 3-digit numbers. The typical increasing false-positive rate accompanying progress through the task was indicated to be due solely to a shifting criterion and not to a buildup in proactive interference. This suggests that, in terms of memory capacity, a "steady state" has been obtained.Recent applications of Signal Detection Theory (TSD) (Green & Swets, 1966;Swets, Tanner, & Birdsall, 1961) to some of the phenomena of shortterm memory (STM) (Green & Moses, 1966;Murdock, 1965 Murdock, , 1966Parks, 1966), have yielded encouraging results. To date, however, the continuous recognition task introduced by Shepard and Teghtsoonian (1961) has not been analyzed within the context of the TSD model. Yet this procedure would appear to be a logical choice for two important reasons. First, of the many techniques available for the study of STM, the Shepard and Teghtsoonian task probably provides the closest methodological analogy to the psychophysical yes-no detection procedure de-

No abstract