A Double Hazard Model for Mental Speed

This paper describes a model that combines these two kinds of data to estimate a subject latent variable usually called mental speed, but more appropriately called mental power. The model implies that the expected item score increases with invested time. Nevertheless, it allows for a decreasing expected item score with response time, which is sometimes found in experiments. This paradox is obtained by assuming that a
subject not only stops working on a problem because of time pressure, but also when he has solved the problem. The model builds on a familiar framework of IRT models. An MML estimation procedure is
developed, and model fit on the item level is evaluated using Lagrange multiplier tests.