Uncertainty Compensation in Human Attention: Evidence from Response Times and Fixation Durations

BACKGROUND Uncertainty and predictability have remained at the center of the study of human attention. Yet, studies have only examined whether response times (RT) or fixations were longer or shorter under levels of stimulus uncertainty. To date, no study has examined patterns of stimuli and responses through a unifying framework of uncertainty. METHODOLOGY/PRINCIPAL FINDINGS We asked 29 college students to generate repeated responses to a continuous series of visual stimuli presented on a computer monitor. Subjects produced these responses by pressing on a keypad as soon a target was detected (regardless of position) while the durations of their visual fixations were recorded. We manipulated the level of stimulus uncertainty in space and time by changing the number of potential stimulus locations and time intervals between stimulus presentations. To allow the analyses to be conducted using uncertainty as common description of stimulus and response we calculated the entropy of the RT and fixation durations. We tested the hypothesis of uncertainty compensation across space and time by fitting the RT and fixation duration entropy values to a quadratic surface. The quadratic surface accounted for 80% of the variance in the entropy values of both RT and fixation durations. RT entropy increased as a function of spatial and temporal uncertainty of the stimulus, alongside a symmetric, compensatory decrease in the entropy of fixation durations as the level of spatial and temporal uncertainty of the stimuli was increased. CONCLUSIONS/SIGNIFICANCE Our results demonstrate that greater uncertainty in the stimulus leads to greater uncertainty in the response, and that the effects of spatial and temporal uncertainties are compensatory. We also observed compensatory relationship across the entropies of fixation duration and RT, suggesting that a more predictable visual search strategy leads to more uncertain response patterns and vice versa.