Alpha oscillations and stimulus-evoked activity dissociate metacognitive reports of attention, visibility, and confidence in a rapid visual detection task

Abstract

Variability in the detection and discrimination of weak visual stimuli has been linked to oscillatory neural activity. In particular, the amplitude of activity in the alpha-band (8-12 Hz) has been shown to impact the objective likelihood of stimulus detection, as well as measures of subjective visibility, attention, and decision confidence. Here we investigate how preparatory alpha in a cued pretarget interval influences performance and phenomenology, by recording simultaneous subjective measures of attention and confidence (experiment 1) or attention and visibility (experiment 2) on a trial-by-trial basis in a visual detection task. Across both experiments, alpha amplitude was negatively and linearly correlated with the intensity of subjective attention. In contrast with this linear relationship, we observed a quadratic relationship between the strength of alpha oscillations and subjective ratings of confidence and visibility. We find that this same quadratic relationship links alpha amplitude with the strength of stimulus-evoked responses. Visibility and confidence judgments also corresponded with the strength of evoked responses, but confidence, uniquely, incorporated information about attentional state. As such, our findings reveal distinct psychological and neural correlates of metacognitive judgments of attentional state, stimulus visibility, and decision confidence when these judgments are preceded by a cued target interval.

Figure 1.

Trial procedure and response options. (A) Each trial began with the words “Get Ready” presented on screen. After a fixed interval of 1 second, the RSVP sequence began, and a target image was presented once, on 50% of trials. Targets (shown outlined in blue) were presented in one of positions 3 to 8 in the RSVP stream. (B) After each trial, participants rated either their subjective confidence and attention (experiment 1), or (C) the perceived visibility of the target and their attention (experiment 2).

Figure 2.

Subjective responses to the same visual detection task. (A) In experiment 1, participants rated their decision confidence that a target was either absent or present, simultaneously with their subjective attention, with a single click in the response square. Orange dots indicate target present trials, purple dots represent target absent trials. (B) Increases in subjective confidence positively correlated with an increase in attention. Larger circle sizes correspond with higher click counts. (C) Average linear correlation coefficients were significantly positive for attention and perceived-presence (orange), as well as attention and perceived absence (purple). Error bars display 1 SD. (D) In experiment 2, participants rated the subjective visibility of targets on the x-axis. Color conventions are the same as in (A–C). (E, F) Subjective visibility did not positively correlate with attention ratings. Note: no correlation is calculated for “perceived absent” trials in experiment 2 because these trials were defined as having the same (zero) visibility rating on all trials.

Figure 3.

Objective and metacognitive accuracy in both experiments. (A) No significant difference was observed in objective accuracy across experiments. (B) Signal-detection theory measures of sensitivity (dʹ) were also similar across experiments. (C) Metacognitive sensitivity was greatest for confidence and visibility judgments and did not differ significantly between experiments. Metacognitive sensitivity based on attention was significantly stronger in experiment 1, although significantly weaker than metacognitive sensitivity based on confidence or visibility judgments in both experiments. (D, E) In both experiments, accuracy increased with the intensity of subjective attention. (F, G) In both experiments, metacognitive sensitivity also increased with subjective attention. In each box, the bottom, central, and top lines indicate the 25th, 50th, and 75th, percentiles respectively. Whiskers extend to the furthest data points. AUROC2, type 2 performance as the area under the receiver operating characteristic curve; ns, not significant.

Figure 4.

Preparatory alpha amplitude is negatively correlated with attention ratings. In experiment 1, strong preparatory alpha over occipital electrode sites correlated negatively and linearly with attention ratings on both target-present (B) and target-absent (C) trials, as well as in the pooled data (D). In experiment 2, a similar topography of alpha band activity (E) negatively correlated with attention ratings on target-present trials (F) and in a pooled analysis (H), but not reliably in target-absent trials (G). Error bars represent 1 SEM, corrected for within-participant comparisons (Cousineau, 2005). Black lines display linear lines of best fit. Asterisks denote significant linear effects. ***p < 0.001, *p < 0.05.

Figure 5.

Preparatory alpha amplitude is quadratically related to subjective visibility and confidence. (A–C) Decision confidence in the presence of a target is maximal at intermediate values of alpha amplitude on target-absent trials and when pooling all trial types. (D) Subjective target visibility is maximal at intermediate values of alpha amplitude on target-present trials. (E) No significant effect of preparatory alpha on visibility when targets are absent, or (F) when pooling across all target types. Error bars represent 1 SEM, corrected for within-participant comparisons (Cousineau, 2005). Quadratic lines of best fit are shown in black. Asterisks mark significant quadratic fits. *p < 0.05, **p < 0.01, ***p < 0.001

Figure 6.

Preparatory alpha amplitude and behavioral performance. Alpha amplitude quadratically modulates (a) accuracy, (b) hit rate, and (c) false alarm rate, but not (d) criterion, (e) dʹ, or (f) area under the curve (AUC) in combined experimental data (N = 21). Responses are normalized per subject, by dividing by the mean across alpha bins, and zero centered by subtracting by 1. *p < 0.05, **p < 0.01. For separate experiments, see Supplementary Figure S3.

Figure 7.

Preparatory alpha amplitude quadratically modulates event-related potentials. (A) Grand average whole-trial epochs for experiments 1 and 2. Gray-shaded regions note the time windows used to calculate the P1, and target-locked CPP (see Methods). (B) Grand average P1 from experiments 1 and 2. (C) Preparatory alpha quadratically modulates the amplitude of the early P1 component, evoked by the first image in our RSVP stream in experiment 1. (D) Grand average target locked CPP. Red shading indicates 250 to 550 ms relative to target onset. (E) Average CPP amplitude over a period 250 to 550 ms relative to target onset in experiment 1. In all plots error bars and shading indicate 1 SEM, corrected for within-participant comparisons (Cousineau, 2005). *p < 0.05, **p < 0.01.

Figure 8.

The subjective correlates of the CPP. CPP amplitude increases with reported confidence (A, B), and visibility (E, F), in experiments 1 and 2, respectively. CPP amplitude also varied as a function of subjectively rated attention in experiment 1 (C, D), but not in experiment 2 (G, H). Gray-shaded regions note 250 to 550 ms relative to target onset, used to calculate the CPP.