Spatial probability AIDS visual stimulus discrimination

Abstract

We investigated whether the statistical predictability of a target's location would influence how quickly and accurately it was classified. Recent results have suggested that spatial probability can be a cue for the allocation of attention in visual search. One explanation for probability cuing is spatial repetition priming. In our two experiments we used probability distributions that were continuous across the display rather than relying on a few arbitrary screen locations. This produced fewer spatial repeats and allowed us to dissociate the effect of a high-probability location from that of short-term spatial repetition. The task required participants to quickly judge the color of a single dot presented on a computer screen. In Experiment 1, targets were more probable in an off-center hotspot of high-probability that gradually declined to a background rate. Targets garnered faster responses if they were near earlier target locations (priming) and if they were near the high-probability hotspot (probability cuing). In Experiment 2, target locations were chosen on three concentric circles around fixation. One circle contained 80% of targets. The value of this ring distribution is that it allowed for a spatially restricted high-probability zone in which sequentially repeated trials were not likely to be physically close. Participant performance was sensitive to the high-probability circle in addition to the expected effects of eccentricity and the distance to recent targets. These two experiments suggest that inhomogeneities in spatial probability can be learned and used by participants on-line and without prompting as an aid for visual stimulus discrimination and that spatial repetition priming is not a sufficient explanation for this effect. Future models of attention should consider explicitly incorporating the probabilities of targets locations and features.

Keywords: attention; perception; reaction time; vision.

Figure 1

(A) Left: The probability distribution for the left sided hotspot distribution is shown. The screen dimensions were 31° × 24° of visual angle and the center of the hotspot distribution was offset 5.6° to the left. The distribution had a linear decrease over a radius of 7.3° at which point it blended in to a background uniform probability. (B) Right: The actual experimental distribution collapsed across all 12 participants. Each point indicates the location of one trial for one participant. Note that for half of the participants the hotspot was actually on the right side, but we have reflected the data here to aid visualization.

Figure 2

Stimuli in Experiment 2 were located in concentric circles at 3°, 6°, and 9° of visual angle from fixation, with one circle containing 80% of stimuli and the other two containing 10% each. Plotted here is the full set of stimulus locations for one participant in the middle-heavy condition. Five centimeters near fixation subtended just over 4° of visual angle.

Figure 3

There is a benefit in RT and accuracy for individual participants when targets from the high-probability, hotspot, side are compared to targets from the low-probability side. Participants are ordered by the magnitude of the RT benefit. Trials with RT <150 ms or >1000 ms are excluded, and RT is computed for correct trials only.

Figure 4

Scatterplots of RT versus distance to high-probability hotspot for each participant in Experiment 1. The trend lines superimposed on each participant's data are for visualization purposes and are created from locally weighted scatterplot smoothing using the xylowess function in the R Statistical Language. Linear regression on the individual plots showed ten out of the 12 regression slopes were significantly positive, at p < 0.05.

Figure 5

Accuracy (above) and correct trial RT (below) in Experiment 1 as a function of whether the trial n back had the same stimulus color and thus required the same response. Mean values are seen at n = 0 and at the corresponding horizontal lines. A future stimulus cannot prime a current one, so the future trial contingencies serve as a visual reference for variability in the data.

Figure 6

Mean reaction time as a function of target location and probability; error bars are 95% confidence intervals for each location × probability condition. Targets were either on the inner, middle, or outer circle. For each circle, the right (red) point is for the trials where that circle was the high-probability location, and the left (black) point is for trials where the circle was a low-probability location. This graph is intended to visualize the magnitude and directions of the effects and thus excludes the data from two participants who had mean RTs two standard deviations away from the sample mean. Statistical analyses reported in the text include the data from all participants.