Visual working memory impairs visual detection: A function of working memory load or sensory load?

Abstract

Researchers have explored the influence of visual working memory (VWM) load on visual perception in the past decade. One of their key findings is that a high VWM load leads to reduced visual detection sensitivity to incoming visual stimuli. However, recent studies imply that persistent sensory processing continues after the memory array is offset. It is possible that the impaired visual detection is due to the sensory load of the residual sensory processing of the memory array (sensory load account) rather than the working memory load of the VWM task (VWM load account). We performed four experiments to examine the impacts on visual detection. Experiment 1 manipulated the retention time of the memory array while keeping the VWM load constant, revealing reduced visual detection along with retention time. Experiments 2 and 3 manipulated the VWM load while maintaining a constant sensory load, and visual detection was not affected. Experiment 4 affirmed that the findings in Experiments 1-3 were reliable. Together, the results of the current study suggest that the residual sensory load, rather than the VWM load of the VWM task, impairs visual detection. Public Significance StatementPrevious load theory studies claimed that the working memory load of the visual working memory (VWM) task impaired visual detection sensitivity to incoming visual stimuli. The current study demonstrates that the sensory load due to perceptual processing of the memory array is the real determining factor. This work provides new insights into the current debate on how VWM and perception interact and sheds critical light on load theory.

P-Values or Bayes Factors? Whichever, but not Both

“…Traditionally, statistical training for psychologists is based on the frequentist framework (see The "Orthodox" Frequentist Statistics section), with little to no focus on Bayesian approaches. Likely because of this lack of training in Bayesian statistics, some psychology researchers seem to still rely on frequentist approaches to test their hypotheses (for similar arguments see Dienes & Mclatchie, 2018;Scheel, 2022;Scheel et al, 2021), but concurrently report a BF along with traditional NHST and p-values (for recent examples see Chen et al, 2022;Guo et al, 2021). This practice is problematic.…”

“…Implementing multiple statistical tests enables researchers to identify and report any results that can be statistically supported (Friese & Frankenbach, 2019;Simmons et al, 2016;Wilson et al, 2022), a problem referred to as analytical flexibility (Dienes, 2023). Put simply, in studies using both approaches (Chen et al, 2022;Guo et al, 2021), the same statistical test is conducted twice, but with a different statistical approach, and provides an independent result for the frequentist test (i.e., p-value) and another independent result for the Bayesian test (i.e., BF). As such, the issue of analytical flexibility arises, which translates to increased researcher degrees of freedom, and consequently, bias.…”

“…As I discuss below, the answers to these questions do not lie in the superiority of one approach over the other, but instead, in the choice of appropriate scientific practices and, in general, the rigorousness of a given study. Still, part of the confusion could be attributed to the fact that, currently, BFs are wrongly used as a complementary statistic to the more common p-value, to quantify evidence (e.g., Chen et al, 2022;Guo et al, 2021). This misuse is also encouraged by attempts to compare the two approaches and investigate relationships between the two statistics (e.g., Huisman, 2022).…”

“…It should, however, be noted that both frequentist and Bayesian approaches for statistical inference should not be simultaneously used and presented because of the issues (amongst others) described above (see also Bayarri & Berger, 2004;Dienes, 2019;Dienes & Mclatchie, 2018;Wagenmakers et al, 2008). So far, the concurrent reporting of both a p-value and a BF was (wrongly) done so that further to NHST statistical significance (p < α), authors (e.g., Chen et al, 2022;Guo et al, 2021) could report on the strength of evidence based on the BF, which is information that the p-value cannot provide (Johansson, 2011). But, as explained above, the BF cannot complement the pvalue, simply because of the fundamental differences in the theories underlying each statistic.…”