Autocorrelation in category judgement

Jesteadt et al. discovered a remarkable pattern of autocorrelation in log estimates of loudness. Responses to repeated stimuli correlated to about +0.7, but that correlation was much reduced (0.1) following large differences between successive stimuli. The experiment reported here demonstrates the same pattern in absolute identification without feedback; if feedback is supplied, the pattern is much muted. A model is proposed for this pattern of autocorrelation, based on the premise: "There is no absolute judgment of sensory magnitudes; nor is there any absolute judgment of differences/ratios between sensory magnitudes." Each stimulus in an experiment is compared with its predecessor, greater, less than, or about the same. The variability of that comparison increases with the difference in magnitude between the stimuli, so the assessment of a stimulus far removed from its predecessor is very uncertain. The model provides explanations for the apparent normal variability of sensory stimuli, for the "bow" effect and for the widely reported pattern of sequential effects. It has applications to the effects of stimulus range, to the difficulty of identifying more than five stimuli on a single continuum without error, and to inspection tasks in general, notably medical screening and the marking of examination scripts.