A Probabilistic Palimpsest Model of Visual Short-term Memory
Figure 2
Distribution of errors in human subjects.
Top: Probability of errors between recalled and target colour (this particular experiment cued the location and required colour to be recalled), for 1, 2, 4 and 6 items (shown simultaneously). One can see that the tail of the distribution grows when an increasing number of items is stored. Bottom: Errors relative to non-target values presented in each array. Any bias towards 0 indicates misbinding. Error bars show one standard error of the mean, for 8 subjects. A resampling-based estimation of the probability of misbinding error was performed, and the p-value for a non-zero non-target response component is shown for N = 2, 4 and 6. Misbinding errors are significantly present in all conditions. See Methods for a description of the resampling analysis. The magenta lines (and outline showing standard error of the mean) show histograms obtained from randomly sampling from mixture models derived from the resampling analysis, removing inter-item correlations. Recalculated based on [7].