Abstract
Perceptual decision-making involves subprocesses, formalized using a Wiener Diffusion Model (WDM), including the rate of evidence accumulation (Drift Rate; DR) and the willingness to execute a decision with higher or lower amounts of evidence (Response Boundary; RB). WDM parameters have been used to successfully characterize links between behavioral and neural data, but typical model estimation methods rely on aggregation over many trials. Such aggregation limits the possible processes able to be investigated, and single-trial parameter estimates would allow for more precise links between behavioral and neuroimaging data. Here we present a simple-to-implement method for mitigating this limitation by estimating trial-level and participant-level WDM parameters using basis functions over the dimension of time (i.e., trial number). Basis functions can approximate arbitrarily complex nonlinear changes, and their implementation is straightforward as random effects within a conventional hierarchical Bayesian WDM. We analyzed data from a visual peripheral orientation discrimination task in both video game players and non-players, with manipulated presence of distractors and predictiveness of cues, and compared models with various combinations of basis function densities for both DR and RB. Model fixed effects indicated predicted reliable differences due to player status (in DR), cue validity (in both DR and RB), distractor presence (in both DR and RB), and post-error effects (in DR). Several fixed-effect interaction coefficients were also reliable. Recovery of raw data's accuracies and response times were not heavily influenced by basis function density (best model by-trial RT r = .67). Possible applications to BOLD data are considered in order to leverage the advanced flexibility of trial-basis-function WDM parameters for integrating conventional fixed-effect hypothesis testing with by-trial estimates.