Our ability to sustain attention for prolonged periods of time is limited. Studies on the relationship between lapses of attention and psychophysiological markers of attentional state, such as pupil diameter, have yielded contradicting results. Here, we investigated the relationship between tonic fluctuations in pupil diameter and performance on a demanding sustained attention task. We found robust linear relationships between baseline pupil diameter and several measures of task performance, suggesting that attentional lapses tended to occur when pupil diameter was small. However, these observations were primarily driven by the joint effects of time-on-task on baseline pupil diameter and task performance. The linear relationships disappeared when we statistically controlled for time-on-task effects and were replaced by consistent inverted U-shaped relationships between baseline pupil diameter and each of the task performance measures, such that most false alarms and the longest and most variable response times occurred when pupil diameter was both relatively small and large. Finally, we observed strong linear relationships between the temporal derivative of pupil diameter and task performance measures, which were largely independent of time-on-task. Our results help to reconcile contradicting findings in the literature on pupil-linked changes in attentional state, and are consistent with the adaptive gain theory of locus coeruleus-norepinephrine function. Moreover, they suggest that the derivative of baseline pupil diameter is a potentially useful psychophysiological marker that could be used in the on-line prediction and prevention of attentional lapses.
All data
Data for van den Brink, Murphy & Niewenhuis: Pupil diameter tracks lapses of attention. Three types of data are provided: 1) Raw data; 2) the processed data that were used to compute metrics for inferrential stistics; 3) and the metrics themselves. (1) Raw data are contained in the folder 'raw_data'. The folder 'pupil_data contains four sub folders: * edfs: Raw EDF files as produced by the EyeLink. * samples: ASCII file containing data points from the EDF files (so the pupil data). * events: ASCII file contaning event type and timing information. Type: 0 = Scrambled image; 1 = Mountain; 2 = City; 32 = Response (space bar press). * converted: MATLAB files containing the data imported into EEGLAB format. Each block is contained in a separate EEG entry within ALLEEG. The first channel is pupil diamter in pixels. The second and third channel are gaze x and gaze y respectively. Event type and timing are contained in EEG.event. The folder 'behavior' contains a MATLAB file per participant and block that contains the behavioral data. * The relevant matrix here is 'response', which is organized as trials (rows) by variables (columns). Relevant columns are: Column 1 contains trial types (0 = Scrambled image; 1 = Mountain; 2 = City), Column 2 contains key code (32 = space bar; 0 = no response), Column 5 contains RTs (RT = 0 if no response), Comlumn 7 contains response type (-1 = false alarm; 0 = miss; 1 = hit). (2) Processed data are contained in the folder 'processed_data.’ Within are text files that resulted from the sliding window analysis. In all files the first column is participant number, and the second column is block number. All following columns are data points (a value per window). These data served as regressors in all the major analyses. Folder and file names will tell you what's what. (3) Regression coefficients and slopes are contained in 'statistics'. All MATLAB files containing matrices on which the stats were run. * Slopes, indicative of linear changes over time, are contained in 'Slopes_xxx.mat;’ Size: participant (rows) by block (columns). * Linear regression coefficients are contained in 'Linear_betas_diameter/derivative.mat.’ * Quadratic regression coefficients are contained in 'Quadratic_betas_diameter.mat.’ * File suffix _noTOT indicates that these are regression coefficients after taking time on task into account. The matrices that contain regression coefficients are of size Participant by block by measure. Measure: 1 = False alarm; 2 = Slow quintile 3 = RT; 4 = RTCV.In all of the above, the participants are in the same order as in the text files in the folder 'processed_data'. Note that all statistics were run on the block-average of these matrices.
vandenBrinketal2016PONE.zip