Response Time Distributions in Memory Scanning

This article reports the results of a memory scanning experiment (S. Sternberg. 1966, Seicnrc. 153, 652-654) in which each of four subjects participated in about 1500 experimental trials per memory set sÍ7.c. These large samples made it possible to test a number of important non parametric (i.e., model-free) properties of the response time (RT) distributions. These properties place severe constraints on the various memory scanning models and they provide a deeper description of the data than summary statistics or goodncss-of-fit values. I7ivc conclu- sions stood out. First, increasing the size of the memory set induced the strongest possible form of stochastic dominance on both large! present and largcl absent trials. Second, the RT hazard functions were nonmonotonic, [hereby falsifying a large class of serial search models. Third, strong evidence was obtained against an exhaustive search. Fourth, some evidence was fun ml thai adding an item to the memory set inserts a stage with cxponenlially distributed duration into the processing chain, al liras! on lar-;el absenl liials. Fifth, (he data supported llu* hypothesis that thice oi llu: .snnjrcts stored the representations of the memory sei items ¡n a visual short-term memory system and the fourth subject used an acoustic short-term system. To our knowledge, the only extant model of memory scanning thai is consistent with all these results assumes thai search is parallel, self-terminating, and of very limited.