Comparing parallel and serial models: Theory and implementation

Some general distribution-free properties of parallel and serial models are reviewed, and predictions for mean reaction times (RTs) for parallel models for both unlimited and limited capacity conditions are derived. It is shown that when processing times differ across same and different comparisons and across spatial locations, serial and parallel models are identifiable at the level of mean RTs. Two experiments with 5 Ss included representative samples of such widely used tasks as memory and visual search and same-different comparisons. Results rule out exhaustive models in favor of self-terminating models. A self-terminating serial model fits mean RTs better than a fixed limited capacity parallel model across both experiments and across 2 levels of complexity of the 2 models. (37 ref)