Higher probability of sampling hypothesis from the training domain influencing reversal and extradimensional shifts

Predictions from rccent transfer hypothesis (H) for reversal (R) and extradimensional (ED) shifts were tested directly using conjunctive concept identification problems. To assess a hierarchical probability of sampling of Hs within or across domains, four transfer conditions (R-1, R-2, ED-I, and ED-2) were arranged according to complete or half change in values or dimensions. The transfer performance was rank-ordered as follows: R-1≒R-2>ED-l >ED-2. These results were attributed to the fact that the subject first samples Hs within the previous correct domain, at the start of transfer, then tests some domain which differs in a single dimension from the training domain, and finally chooses Hs from other new domains with two changes in dimension.