Abstract
Original article: Pratte, M. S. (2018). Iconic memories die a sudden death. Psychological Science, 29, 877–887. doi:10.1177/0956797617747118
The original article stated that the rate of swap errors was relatively low. However, after the article was published, the data were shared with Paul Bays, and his analyses indicated that rates of swap errors were higher than reported. I am grateful that Dr. Bays took the time to share this result, as it led me to discover an error in my own swap-model analysis, which confirmed that the reported rates of swap errors were indeed inaccurate. Critically, although the prevalence of swapping is higher than was originally reported, the main result holds within the corrected swap models: The retention-interval manipulation had large effects on the proportion of in-memory responses but only small effects on memory precision.
The “Swap errors” section (p. 884) will be amended in two ways. First, it will be split into two paragraphs, with the first paragraph staying the same as in the original article and ending with “To assess nontarget responses in iconic memory, I fitted a version of the full mixture model that additionally included free swap-rate parameters for each retention interval (see the Supplemental Material for details).” The remainder of the section will be replaced with the following paragraph to reflect the correct swap-error results: The addition of swap errors to the full mixture model improved model fit in both Experiment 1a (29 of 55 participants; ΔAIC = 143) and Experiment 1b (8 of 8 participants; ΔAIC = 238), but the model without swap errors performed somewhat better in both Experiment 2a (35 of 42 participants; ΔAIC = 144) and Experiment 2b (7 of 8 participants; ΔAIC = 17). Critically, restricting precision in this swap model to be constant across retention intervals produced a better fit in Experiment 1a (44 of 55 participants; ΔAIC = 72), Experiment 2a (38 of 42 participants; ΔAIC = 79), and Experiment 2b (6 of 8 participants; ΔAIC = 21), with mixed results in Experiment 1b (4 of 8 participants; ΔAIC = 2.9 in favor of the unrestricted model). When the model was further restricted so the proportion of nontarget responses that were swap errors was fixed across retention intervals, it also outperformed the full swap model in Experiment 1a (50 of 55 participants; ΔAIC = 206), Experiment 1b (7 of 8 participants; ΔAIC = 54), Experiment 2a (40 of 42 participants; ΔAIC = 197), and Experiment 2b (7 of 8 participants; ΔAIC = 78) and often outperformed the fixed-precision models without swap errors (see Table S1 in the Supplemental Material). This model suggests that swap errors made up 35%, 48%, 15%, and 19% of nontarget responses in Experiments 1a, 1b, 2a, and 2b, respectively. Taken together, these results again suggest a fixed-precision model, but one in which a proportion of nontarget responses are centered around nontarget items. Further work is needed to understand the nature of this behavior, and examining how swap errors develop over the course of iconic memory decline may provide a new avenue for understanding their role in working memory tasks.
Table S1 in the Supplemental Material available online has also been amended to reflect the correct model-comparison statistics.