The contribution of attentional lapses to individual differences in visual working memory capacity

Abstract

Attentional control and working memory capacity are important cognitive abilities that substantially vary between individuals. Although much is known about how attentional control and working memory capacity relate to each other and to constructs like fluid intelligence, little is known about how trial-by-trial fluctuations in attentional engagement impact trial-by-trial working memory performance. Here, we employ a novel whole-report memory task that allowed us to distinguish between varying levels of attentional engagement in humans performing a working memory task. By characterizing low-performance trials, we can distinguish between models in which working memory performance failures are caused by either (1) complete lapses of attention or (2) variations in attentional control. We found that performance failures increase with set-size and strongly predict working memory capacity. Performance variability was best modeled by an attentional control model of attention, not a lapse model. We examined neural signatures of performance failures by measuring EEG activity while participants performed the whole-report task. The number of items correctly recalled in the memory task was predicted by frontal theta power, with decreased frontal theta power associated with poor performance on the task. In addition, we found that poor performance was not explained by failures of sensory encoding; the P1/N1 response and ocular artifact rates were equivalent for high- and low-performance trials. In all, we propose that attentional lapses alone cannot explain individual differences in working memory performance. Instead, we find that graded fluctuations in attentional control better explain the trial-by-trial differences in working memory that we observe.

Figure 1

Results from Experiment 1a. (A) Illustration of the task design and stimuli in Experiment 1a. (B) Overall performance changes in a similar manner for change detection (blue) and whole report (red) across set-sizes. (C) The correlation between mean whole-report performance and change detection capacity at each set-size. The proportion of performance failures in whole-report (0 or 1 correct) increases across set-size (D) and explains more variance in capacity across set-size (E).

Figure 2

Whole-report performance distributions for Experiment 1b. (A) Correlation between mean whole-report performance and change detection capacity in Experiment 1b. (B) Performance distributions for participants split into extreme groups by their change detection score. (C) Performance distributions for all participants in Experiment 1b. Each column represents the performance outcome (number of items correct for a given trial), and each row represents a participant (sorted by capacity). Differences between participants are best characterized as a subtle upward or downward shift of the performance distribution (with a central tendency at 3 for most participants). (D) Performance outcomes correlate with change detection estimates of capacity for all levels of performance except for 3 correct.

Figure 3

Monte Carlo simulation of lapse performance and guessing inflation. (A) Results from a simulation of guessing without replacement from nine colors over six objects. (B) Results from a simulation of guessing inflation when participants get 3 items correct and guess without replacement from the remaining colors over 3 objects.

Figure 4

Monte Carlo simulation results for lapse and attentional control models of performance fluctuations. (A) The simulated mean number correct from the lapse model as a function of the actual mean number correct. (B) Data (gray bars) and lapse model fits (black lines) from the extreme groups split of participants. (C) The simulated mean number correct from the attentional control model as a function of the actual mean number correct. (D) Data (gray bars) and attentional control model fits (black lines) shown over the extreme groups split of participants.

Figure 5

Performance fluctuations over time in Experiment 1b. (A) Data from all participants and all trials is shown over time. Each subject is a row (sorted by whole-report performance), and the horizontal axis corresponds to trials over time. Tick mark color corresponds to the performance outcome; red lines delineate block breaks. (B) A summary of the prevalence of performance failures (0 or 1 items correct) over blocks. Gray bars show the mean performance level; red and blue lines illustrate a median split of subjects. (C) Reliability of mean number correct (left) and prevalence of performance failures (right) for even versus odd blocks.

Figure 6

Performance fluctuations do not relate to task compliance or sensory encoding. (A) The prevalence of poor trials (2 or less correct) before and after removing ocular artifacts. (B) The P1/N1 visual-evoked response as a function of performance outcome (good versus poor trials).

Figure 7

Spectrogram for good trials minus poor trials at all electrode sites measured. Each spectrogram represents spectral power at all frequencies from 4 to 30 Hz for each of the sites measured during Experiment 2.

Figure 8

Theta and alpha power as a function of whole-report performance. (A) Spectral power for good trials minus poor trials at frontal electrode sites. (B) Mean theta power (4–7 Hz) as a function of trial performance. (C) Spectral power for good trials minus poor trials at posterior electrode sites. (D) Mean alpha power (8–12 Hz) as a function of trial performance.