The systematic discrepancy between A9 for overall recognition and remembering: A dual-process account

William James suggested as early as 1890 that, on the basis of subjective experience, at least two separable processes seemed to underlie the experience of recognition. An oft cited anecdote suggesting two recognition processes was later provided by Mandler (1980), who described the “butcher in the bus” phenomenon, where one steps onto a bus and immediately recognizes a particular face as having recently been seen but is, at least temporarily, completely unable to describe the contextual details that would situate the feeling of familiarity. Adopting a distinction later formalized by Endel Tulving in 1985, this might be characterized as knowing that the person has been previously seen, without remembering elements sufficient to specify the source or occasion. Tulving (1985) argued that remembering and knowing represented different forms of conscious awareness that reliably map onto separate underlying memory systems and, therefore, should be relatively easy to experimentally dissociate. Considerable research has supported this hypothesis, demonstrating dissociations of remembering and knowing with the use of encoding and retrieval manipulations (for a review, see Rajaram & Roediger, 1997), electrophysiological measurements (Duzel, Yonelinas, Mangun, Heinze, & Tulving 1997), and functional magnetic resonance imaging (Henson, Rugg, Shallice, Josephs, & Dolan, 1999). These dissociations have been taken by many as clear evidence for the presence of at least two separable mnemonic processes underlying explicit recognition. Despite the introspective appeal and experimental evidence supporting dual-process accounts of recognition, several signal detection theorists have argued that remember and know responses may be entirely accommodated within the equal variance signal detection model by simply assuming that subjects adopt a secondary, more stringent response criterion for providing remember reports (two-criterion signal detection model; e.g., Donaldson, 1996; Hirshman & Henzler, 1998; Hirshman & Master, 1997; Inoue & Bellezza, 1998). From this perspective, “remember responses represent nothing more than conservative yes responses” (italics added; Donaldson, 1996, p. 524). On the basis of this characterization Donaldson suggested that criterion-free estimates of accuracy (e.g., A9 or d9) should be equivalent for overall