Analysis of the Hockley and Murdock decision model

Abstract

Hockley and Murdock (1987) proposed a decision model to predict accuracy and response latency in recognition memory across a range of experimental paradigms. A decision is made when the evidence from a memory comparison process plus extraneous noise exceeds a lower or upper criterion. If neither criterion is exceeded, the distance between the criteria is reduced and a new sample of noise is added to the original memory comparison value. This is repeated until a criterion is exceeded. The model is shown to have two shortcomings: First, it produces a reaction time distribution that is multimodal; empirical distributions are generally unimodal. Second, application of the model to various speed-accuracy trade-off phenomena is found to be inadequate: Either the assumptions made to account for speed-accuracy data are post hoc or they are unable to mimic the data. An experiment that manipulates speed-accuracy tradeoffs demonstrates that the model cannot produce a trade-off of sufficient range. Alternative conceptions of the model (the addition of a guessing process and a zero-drift random walk) are unsuccessful. The diffusion model of Ratcliff ( Psychological Review, 85, 59–108 (1978); 88, 552–572 (1981)) provides an adequate account of these data and a more parsimonious account of other speed-accuracy phenomena.