Still no evidence for the encoding variability hypothesis: a reply to Jang, Mickes, and Wixted (2012) and Starns,...

Koen and Yonelinas (2010) contrasted the recollection and encoding variability accounts of the finding that old items are associated with more variable memory strength than new items. The study indicated that (a) increasing encoding variability did not lead to increased measures of old item variance, and (b) old item variance was directly related to the contribution of recollection. Jang, Mickes, and Wixted (2012) and Starns, Rotello, and Ratcliff (2012) wrote responses that, on the surface, appear to challenge those results. However, the issues raised about our first finding turn out to have no theoretical or empirical support. In addition, although Jang et al. replicated our second finding, they contested our conclusions on the basis of a perceived problem with the analyses (i.e., we used 5 rather than 2 points to calculate predicted zROC slopes). However, their concern was misplaced because the pattern of results is the same regardless of the number of points used in the analysis. They also conducted a simulation that, at first glance, appeared to suggest that our second finding was biased to favor the recollection account of the dual process model. We show that this conclusion arose because Jang et al. mistakenly built the contested correlation pattern into the simulated data. Overall, the 2 response articles serve to strengthen the main conclusions of our initial article by reiterating that there is no evidence in support of the encoding variability account, and the replication study by Jang et al. adds to the evidence favoring the recollection account.