Abstract
Measures of working memory capacity (K), derived from performance in change detection tasks (Cowan, 2001; Vogel et al., 2001), estimate that we can hold 3–4 items in visual working memory (VWM). Such measures assume that incorrect responses arise solely from a failure to store items, and do not take into account that comparison errors (CEs) may occur when internal VWM representations are compared to perceptual inputs during the probe stage of a VWM task. However, since these assumptions are likely incorrect (Awh et al., 2007; Zhang & Luck, 2008), K underestimates actual capacity according to the rule, Kmax = capacity * (1 - CE rate), where Kmax is the maximum estimate of K for any VWM set size (N). This equation also predicts that K will increase with increases in N until N corresponds to capacity. Thus, the value of N at which K peaks should be a better estimate of capacity than K. Here we experimentally demonstrate that this measure of capacity is independent of CEs by adding perceptual noise to the probe (85% of probe pixels assigned a random color) in order to increase comparison errors. In the task, participants had to memorize the color and location of 2–7 briefly presented squares. After a 2.4s retention interval, VWM was tested using a single-probe procedure in which participants indicated whether the probed color was shown in the correct location. Addition of perceptual noise reduced the K values measured at any setsize, demonstrating that K estimates are sensitive to CEs. However, K values peaked around N=4 regardless of the presence of perceptual noise, suggesting that this analysis provides a measure of WM capacity that is robust to CEs. We will discuss the application of this analysis for estimating the influence of stimulus complexity and sample-test similarity on VWM capacity.
This work was support by NIMH grant (R01-MH770776) to R.M.