Five experiments address the theoretical mechanisms governing false alarms in episodic recognition. Three major distinctions are drawn between extant theories of false recognition: those that address base-rate false alarms (background-familiarity and context-noise models); those that address similarity-based false alarms (lARs, recall-to-recognize, and category-based criterion shifts); and comprehensive theories (item-noise and spreading-activation models), which address both types of false recognition using the same set of theoretical assumptions. The Deese/Roediger-McDermott (DRM) paradigm (J. Deese, 1959; H. L. Roediger & K. B. McDermott, 1995) presents a challenge to theories of recognition memory in that it typically requires an 'explanation for both types of false recognition within the same experimental procedure: base-rate false alarms to unrelated distractors and similarity-based false alarms to related distractors.
Experiment 1 and 2 instantiated a variant of the DRM paradigm. In Experiment 1 (N = 203), a 2x2x2x2x2 mixed factorial design with one nested factor was employed; manipulating the type of conceptual category (taxonomic vs. associative), the category length (5-item vs. 1-item), the item tested within the category (first item in the 5-item categories and the only item in the litem categories vs.last item in the 5-item categories and the only item in the 1-item categories), the category separation type (blocked vs. distributed), and the type of recognition test (old-new vs. forced-choice). In both the old-new and two-alternative forced-choice procedures, the distractor items were either the most representative exemplar from the taxonomic categories, or the prototype from the associative categories. Additionally, participants in the forced-choice test received both within-category choices and between-category choices (nested factor) in which the two test alternatives were drawn from the same category or different categories, respectively. Experiment 2 (N = 168), employed distributed taxonomic categories exclusively, retained the item tested control, and introduced a test expectancy manipulation and forced-choice test conditions in which between-category choices were used exclusively.
The comprehensive theories (item-noise and spreading-activation models) predict a loss in discriminability as category length increases and a within-category choice advantage in forced-choice recognition. For the associative categories in Experiment 1, increases in category length produced both losses in discriminability and shifts to a more liberal decision bias. For the taxonomic categories in Experiment 1, the bias shifts were not accompanied by losses in discriminability. The Experiment 2 results were consistent with those of Experiment 1: increases in category length failed to produce losses in discriminability for the taxonomic categories, but again produced shifts to a more liberal decision bias. Furthermore, across both Experiment 1 and 2, there was no evidence for a within-category choice advantage in forced-choice recognition.
The control variables employed in Experiment 1 and 2 provided substantial support for the involvement of organizational processes in the DRM paradigm (e.g., categorization: encoding of same-category interitem associations and/or item-category associations). Experiment 3 and 4 were designed to pursue similarity-based theories of false recognition (recall-to-recognize; category-based criterion shifts) using simpler paradigms that afforded proper controls over the experimental variables of interest. Experiment 3 (N = 60) examined cued recall and cue recognition of related pairs. A high level of false recall and false recognition was observed for extralist cues related to both members of a study pair. The results support the involvement of cue-substitution driven recall-to-recognize in the DRM paradigm. Experiment 4 (N = 20) examined context-based categorization of unrelated items, demonstrating a context-like effect produced by item-by-item criterion shifts- evidence against static criterion accounts of within-list episodic recognition, and a foundation for a category-based criterion-shift account of false recognition in the DRM paradigm.
Experiment 5 (N = 60) examined false alarms to unrelated items under a source-monitoring lest manipulation. The materials and lists controlled for background familiarity (word frequency), context noise (context variability), and item noise (fixed set of old and new items). The observed materials-by-instructions interaction supports source-constrained retrieval and a context-noise model of recognition memory. These findings, in conjunction with failure to observe taxonomic category length effects and a within-category choice advantage in Experiment 1 and 2, provide convergent evidence against item-noise and spreading-activation accounts of false recognition. The utility of the DRM paradigm for investigating false recognition processes and the use of the term "false memory" to describe commission errors in list-learning paradigms are addressed in the General Discussion.