Abstract
This project investigated how centroid estimates depend on the recent experience of the participant in the task. Stimuli were brief clouds comprising 18, barely visible dots, and on each trial the participant strove to mouseclick the centroid. A double- pass procedure was used: that is, each participant went through exactly the same set of stimulus sequences twice. The data are well-described by a model in which the response on a given trial is a weighted sum of the centroid of the current stimulus and the previous response location. Setting the mixture parameter to zero yields a nested model in which the current response is immune to influence from the previous response. A likelihood-ratio rejects this nested model supporting the claim that the current response is influenced significantly by the previous response. Since the double-pass procedure provides a model-free estimate of intrinsic process noise, it is a tool for model rejection using a least-squares criterion. Specifically, if the noise estimate provided by the double-pass procedure is significantly less than the noise estimate derived from the deviation of the model predictions from the data, then the model is rejected. In the current instance, the residual noise in the fit provided by the mixture model is very close to the model- free estimate of the response noise estimated from the double-pass procedure. We conclude that the mixture captures all of the structure in the data.
Meeting abstract presented at VSS 2018