Control experiment, longer delays:
As mentioned in the main text, in order to validate that our results in Experiment1 are not related to temporal summation or disruption of iconic trace we repeated the experiment with longer time intervals between items: 600ms instead of 300ms. Participants were presented with a sequence of randomly oriented colored bars, and asked to reproduce from memory the orientation of one of these, the identity of which was specified by its color (Figure S1A). Angular errors for the different conditions and serial order of the target are presented in Fig S1B. Statistical analysis revealed qualitatively similar results to Experiment 1. Two-way repeated-measures ANOVA with condition (same/different location) by target's serial order (first to last) revealed a significant effect of condition (F(1,7) = 7.7, P = 0.027) , indicating larger errors in the same-location condition. The main effect of serial order was also significant (F(3,21) =16.2,P<0.001) and the interaction was not (F(3,21) <1.3, P > 0.3). Indicating similar effect of location for different serial position of the target.
Figure s1 | Memory for orientation with stimuli presented at same vs different locations
(A) The same paradigm as experiment 1 but with 600ms between items.(B) Error between target and response for the two conditions and four serial positions of the target in the sequence. (C)and(D) Frequency histograms of responses aligned with respect to orientation of the target and non-targets (respectively). Participants systematically reported non-target orientations significantly more often in the same location condition. Asterisks denote significance level of p<0.05 for two tailed t-test; error bars show SEM.
Next we analyzed the distribution of errors around the target orientation (the item that was probed; Figure S1C) and those around the non-targets (other items that were in the sequence but not probed; FigureS1D). The histograms in panels CD show the angular ‘distance’ from the target and non-target items in the sequence. A closer distance to the target means a response that was closer to the true orientation of the probed item. Responses plotted with respect to non-targets are shown such that each non-target orientation was aligned to zero degrees.
When objects were presented at the same location there were significantly fewer trials in which participants’ responses were less than 15° of the target orientation (only clock-wise errors reached significance in this version), compared to the different location condition (t(7)=2.7, p=0.03).
In addition, the histograms of responses centered on non-target items (Figure 1D) reveals that in the same location condition participants erroneously reported the orientation of non-target objects more frequently. When items were presented at the same location there was a significantly higher fraction of trials with <15° error clock-wise to non-target items, compared to the different location condition (t(7)=2, p<0.04). Thus the higher level of errors biased towards non-targets in the same location condition occurred with a corresponding decrease in reporting around the target item.
Figure s2 | Three sources of error and the effect of location, control experiment.
Subject's responses in the memory task with 600ms between consequent items. Responses were decomposed into three separate components; the different types of errors are illustrated by the colored regions in the illustrations at the bottom. Recall precision (right): Concentration parameter of the circular Gaussian (von Mises) distributions centered on the orientation values of the bars. Misreports (left): Circular Gaussian distribution centered on each non-target orientation value. Random guesses (middle): A uniform distribution, capturing random responses unrelated to any of the sample orientations. The model results were obtained separately for the same/different location conditions. Asterisk denotes a significant difference between conditions. Error bars denote SEM across participants.
Consistent with the histogram analysis, the mixture model analysis revealed that in same-location trials there was an increased probability to misreport the wrong item in memory (Figure s2left; t(7)=2.6, P = 0.03) but no significant change in responding with a random orientation (Figure s2 middle; t(7)=0.3 , p=0.77). The concentration parameter (κ), which represents the width of the underlying distributions, was also not statistically different between conditions (t(7)=0.3, p=0.8).
These findings replicate our findings from experiment 1 and demonstrate that our findings are robust also in longer delays which clearly hamper visual summation.
Mixture model analysis for experiments 2 and 3:
Figure s3 | Three sources of error in experiment 2 and 3.
Subjects' responses in experiment 2 & 3 (upper and lower subplots) were decomposed into three separate components; the different types of errors are illustrated by the colored regions in the illustrations at the bottom. Precision (right): Concentration parameter of the circular Gaussian (von Mises) distributions centered on the orientation values of the displayed items. Misreports (left): Circular Gaussian distribution centered on each non-target orientation value. Random guesses (middle): A uniform distribution, capturing random responses unrelated to any of the sample orientations. Error bars denote SEM across participants.