Abstract
Zhang, Daw & Maloney (2011, VSS) developed a 2-IFC task that allowed us to compare the motor error distribution implicit in a subject’s choices between targets differing in shape to his actual motor error distribution. Subjects’ implicit distributions typically differed from their actual (anisotropic Gaussian) error distributions. Here we use this task to estimate subjects’ implicit distributions with targets that were composed of multiple, disjoint regions. The weight that subjects gave to regions of the target near to and far from their aim point allowed us to estimate the tails and variances of their implicit distributions.
Methods: Targets were three equally-spaced vertical stripes (SSS) or a single vertical stripe (S). The widths of each stripe of the SSS and the distance between central and flanker stripes of the SSS were systematically varied across 12 experimental conditions (3 widths X 4 distances). Training. Subjects completed 300 trials of a speeded reaching task, in which they touched an SSS target on a touchscreen within 400 milliseconds. Task. On each trial, subjects judged which of two targets, an SSS or an S, was easier to hit. We used a staircase method to estimate the width of the S the subject judged to be as easy to hit as each of the 12 SSS targets. We tested three nested models: the true Gaussian, a Gaussian with a free variance, and a heavy-tailed t-distribution. Ten naïve subjects participated.
Results: (1) The true error distribution of all the ten subjects was Gaussian in the horizontal direction. (2) Most subjects overestimated the probability of hitting parts of the target distant from the aim point. (3) Among them, the implicit distributions of 3/10 subjects were Gaussian of a larger variance than their true distributions. (4) The implicit distributions of 4/10 subjects were heavier-tailed than Gaussian.
Meeting abstract presented at VSS 2012