Stochastic Models of Visual Search: Queues, Pert Networks and Op Diagrams.

Abstract

For the most part current models of visual search assume that limits on capacity are not exceeded and that attention is not necessary in situations where the targets and distractors are consistently mapped to the same response. In this dissertation alternative models of search are developed. Limits are imposed on the number of comparison channels and constraints are placed on the process by which encoded stimuli arrive at the channels. The limited channel models are used to predict accuracy in three multiple frame experiments and to predict both accuracy and latency in three single frame experiments. Support for the models was indicated by the good fit of predictions to observations and by the reasonableness of parameter estimates. Additional support is provided by the rather surprising finding that the limited channel models which give the best fit require between 3 and 5 channels. This finding holds across different models, different tasks and different measures of performance. A second finding of some potential consequence emerged from Experiment 1. Subjects' performance was nearly equivalent in conditions which varied the stimulus exposure duration but kept constant the stimulus arrival rate. This finding is not easily predicted by the two alternative models given the serious consideration in this dissertation, but is easily predicted by a steady state limited channel model. The use of continuous time Markov processes to structure the limited channel models is discussed in some detail. It is suggested that continuous time Markov processes offer the investigator of complex cognitive tasks such as visual search is important. Special attention is given to the representation of visual search behaviors as stochastic PERT networks and by means of newly developed Order-of-Processing (OP) diagrams.