ABSTRACT
Models of response time distributions are powerful models because they capture mean response times as well as variability and asymmetry. Some of these models are expressed with a closed-form equation (e.g. the Weibull distribution). However, a large number of others are not available in closed form, including convolutions. Fitting such distributions with the likelihood method is still possible with the help of numerical integration techniques. However, the time required to fit a single subject is a matter of days. We present an alternative to numerical integration based on rational approximations. This approach is 100 times faster so that fitting a single subject can be done in less than an hour. As shown with simulations, the approach can be fully automatized, and is both reliable and accurate.
