Feature-continuous motion judgements: Assessing different random dot motion displays
Riccardo Barbieri, Felix M. Töpfer, Joram Soch, Carsten Bogler, John–Dylan Haynes · Journal of Vision · 2018
In perceptual decision making experiments subjects are often asked to indicate a stimulus feature under different level of noise (Gold & Shadlen, 2007). In most cases the decisions require categorical judgements, for example binary choices between two different motion directions. Such categorical judgement tasks have important shortcomings: They do not reflect the inherently continuous nature of perception. And due to being binary they do not provide a graded trial-wise measure of precision. We developed a continuous version of the Random Dot Kinematogram (RDK – Newsome & Parè, 1988), in which the direction of motion varied from trial to trial, each time taken from the full range of directions. Each trial had a different level of motion coherence (Figure 1). We tested 12 subjects on three types of RDK (Transparent, Limited Lifetime White noise and Brownian) to observe which stimulus would lead to a better estimate of the motion direction (Pilly and Seitz, 2009). As expected, responses were less precise and slower with lower stimulus coherence (Figure 2). This is in line with predictions of a drift diffusion model extended to feature continuous tasks (Smith, 2016). Moreover we found that Brownian RDK is the most suited for the continuous report of motion direction. Both the Transparent and Limited Lifetime White-Noise stimuli induced a second component in the distribution of responses: the report of opposite direction (ROOD, Figure 3). Please note that in a standard RDK task with only two directions this would go unnoticed. The Brownian implementation yielded a single distribution of responses that spreads with decreasing coherence with no secondary peak. Therefore, Brownian RDK is most suited for studying continuous report of motion direction. We further conclude that feature-continuous decision making tasks capture perceptual performance, especially the accuracy, in more detail. Meeting abstract presented at VSS 2018