14 - Constant Information Model on the Dichotomous Response Level

Abstract

For the purposes of measuring the abilities of individuals when they are assumed to vary over a wide range, it is important that all ability estimates have the same standard error. One way of achieving this is through adaptive testing with a termination rule based on a value of the standard error of measurement. As the standard error of measurement given a specified level of ability is inversely proportional to test information, this can be expressed as constant information over a defined range of θ. Another way to achieve constant information in the measurements is described by Samejima. This chapter describes Samejima's potentially useful constant information model and discusses some of its properties such as the simplified ability and error variance estimation procedures, which derive from the model. This model provides a constant item information for a finite interval of a latent trait. Latent trait theory enlarges its horizon if full use is made of information functions, enabling types of research to be conducted that could not otherwise be done.