Analysis of Reversal and Non-Reversal Training: Further Examination of the Retention Decrement Hypothesis

Abstract

Recently the retention-decrement hypothesis has been advanced to account for the reduction in errors over successive discrimination reversals (SDR). According to this hypothesis, error reduction results from increasing amounts of proactive interference that weakens the preference transferred from problem to problem. However, intra-problem transfer of a preference is also presumably weakened. Thus, in a 2-day design, errors on the first day of each reversal (R days) should show a gradual decline, but errors on the second (non-reversal) day should gradually increase across problems. To test the generality of the convergence of R and NR error functions, SDR data from 18 vertebrate species, obtained on a spatial and brightness task, were analyzed. In no instance was convergence of R and NR curves shown. Methodological conditions that might account for this discrepancy are discussed.