Figure 2.

Behavioral analysis. A, Proportion of errors committed in alert and drowsy periods across left and right stimuli. Multilevel modeling reveals that error rates depend on stimulus and state of the participant. Post hoc tests indicate that error proportion is reliably modulated by alertness but only for left stimuli. ***p < 0.001. ns, Not significant (or not reliable). Error bars indicate SEM. B, Mean reaction times for individual participants. Reaction times are variable and slower under drowsy conditions; further, a paired-samples t test indicates reliable difference (p < 0.001) in reaction times across alert and drowsy periods. C, Example fits of two different participants are shown. For each participant under each condition (alert, drowsy), the proportion of rightward responses was fitted to the stimulus angle varying from −60° to 60°. The size of the dots represents the normalized (per condition) number of trials under each angle. Here, we notice that the bias point (dotted line) shifts toward the left side as the participant becomes drowsy. In other words, as the participant becomes drowsy, they overestimate the right side of space. For another participant, we notice the opposite effect. D, Spatial bias. Slope parameter per participant (from psychometric fits). Bias level shifts toward the left side (indicating more left errors) for most participants. Negative bias values indicate shifts in the subjective midline toward the left, and positive values indicate shift toward the right. Further, a paired-sample t test indicates reliable difference (p < 0.01). Slope values of the psychometric fits also increased as participants became drowsy, indicating shallower slopes. Further, a paired-sample t test indicates reliable difference (p < 0.001). E, Group-level psychometric fits indicate the shift in subjective midline (dotted lines). Shaded regions represent CI bounds. Arrows indicate a gap in the asymptotes at −60° compared with 60° between alert and drowsy state, further evidence of inattention to the left side. F, Signal detection analysis shows that d′ (sensitivity) is strongly modulated by alertness compared with c (criterion) using Bayes factor.