Preview

PDF, English Download (7MB) | Terms of use

Abstract

The Diffusion Model (Ratcliff, 1978) is a mathematical model that allows the disentangling of several cognitive processes involved in binary decision tasks. Although the model has gained in popularity within the last decade, strikingly little is known about the prerequisites of diffusion modeling. In this thesis, based on both simulation studies and test-retest studies, the performance of different estimation procedures and numbers of trials was compared. The studies were based on fast-dm-30, the newest version of the program fast-dm, which has been extended to include three different optimization criteria (Kolmogorov-Smirnov, Maximum Likelihood, and Chi-Square). The three main guidelines that emerged from this research were: (1) Chi-Square performs worse than the other two criteria for small- to medium-sized trial numbers; (2) Increasing the number of trials beyond 500 is of limited advantage as the precision of parameter estimation improves only marginally; and (3) Fixing the intertrial variabilities of drift rate and starting point can bring about more reliable estimates of the four main diffusion model parameters (i.e., drift rate, threshold separation, starting point, and nondecision time).