The invariance assumption in process-dissociation models: an evaluation across three domains.

The class of process-dissociation models, a subset of the class of multinomial processing-tree models, is one of the best understood classes of models used in experimental psychology. A number of prominent debates have addressed fundamental assumptions of process-dissociation models, leading, in many cases, to conceptual clarifications and extended models that address identified issues. One issue that has so far defied empirical clarification is how to evaluate the invariance assumption for the dominant process. Violations of the invariance assumption have, however, the potential to bias conventional process-dissociation analyses in different ways, and they can cause misleading theoretical interpretations and conclusions. Based on recent advances in multinomial modeling, we propose new approaches to examine the invariance assumption empirically and apply them in 6 studies to 3 prominent fields of application of process-dissociation models: to the Stroop task, to the interplay of recollection and habit in cued recall, and to the study of racial bias in the weapon task. In each of these content domains, the invariance assumption is found to be violated to a considerable extent.