Corrigendum to “Clear evidence for item limits in visual working memory” [Cogn. Psychol. 97 (2017) 79–97]

In the online supplementary materials associated with this article, the “swap rate” was calculated incorrectly for Analysis S1 and Table S3. Correction of this error slightly changed the average swap rate for each response (corrected Table S3). However, the core conclusion of this analysis is the same; swaps did not increase monotonically as a function of response number (corrected Analysis S1). The corrected supplementary analyses are shown below and are available on the article’s Open Science Framework page (https://osf.io/kjpnk/). We would like to thank Sebastian Schneegans for alerting us to the error in our original analysis.

Analysis S1. Swap analysis for Experiment 1.

We considered what proportion of the uniform distributions observed for the later Set Size 6 responses could be explained by an increase in swap errors. First, we fit data from Experiment 1 using a Mixture Model with Swapping in Memtoolbox (Suchow, Brady, Fougnie, & Alvarez, 2013). Swapping represented a small proportion of Set Size 6 responses (Table S3). On average, swapping occurred for 10% of all Set Size 6 responses in Experiment 1a and 5% of all Set Size 6 responses in Experiment 1b.

Next, we ran a repeated-measures ANOVA on the Set Size 6 swap rates with Response Number as a within-subjects factor. In Experiment 1a, there was a main effect of response order on swap rate, F(5,105) = 6.40, p < .001. Critically, however, this significant effect was not due to a monotonic increase in swap rate as a function of response number. Instead, the effect was non-monotonic, peaking at the third response. In fact, the swap rate for responses 5 and 6 was no different than for response 1, p > .45. In Experiment 1b, there was no significant main effect of response order on swaps, F(5,95) = 1.50, p = .197. That is, swaps were no more likely to occur for the later, uniform responses than they were to occur for early responses. In sum, the uniformity observed for responses 5 and 6 cannot be explained by a sudden increase in swap errors.