Abstract
A key visual computation is figure-ground assignment. It is often assessed by participants’ reports regarding whether they perceive a figure on the left or right side of a bipartite display. We built a Drift Diffusion Model (DDM) of this task and assessed its ability to fit and extract meaningful parameters from human data. We assume that participants’ responses are determined by a decision variable representing the current state of evidence in favor of a figure on the left or right side. This variable varies over time according to a drift term towards the true figure side determined by figural priors and a diffusion term based on random neural fluctuations. The model “decides” when the decision variable crosses either a left or right threshold, thereby simulating both choice and response time (RT). By fitting the four free parameters of the model to participants' choice and RT, DDMs allow mechanistic insight into how experimental manipulations change the decision. The best practice is to conduct a “parameter recovery” analysis that fits the model to the simulated data, with known simulation parameters, to determine whether the fitting process can recover these simulation parameters. If fitted and simulated parameter values are correlated, then parameter recovery is successful and parameters fit to real data will be meaningful. We assessed parameter recovery in our model by simulating the behavior of 128 subjects performing 72 trials of a figure-ground task with the familiar configuration prior favoring one side. Each simulated subject had random simulation parameters in a range consistent with human data. The model was fit using a maximum likelihood procedure. We found strong correlations between simulated and fit parameters for all four free parameters (rs > 0.70) indicating good parameter recovery in the model and demonstrating that our model can be applied to different experimental conditions.