Estimation of parameters in the generalized graded unfolding model using a genetic algorithm

Abstract

In the current study, a genetic algorithm was used in conjunction with the expectation-maximization algorithm to estimate parameters in a polytomous unfolding IRT model known as the generalized graded unfolding model (GGUM). One advantage of using a genetic algorithm for IRT parameter estimation is that this global optimization procedure is not easily affected by local maxima in the likelihood function – a condition that is often encountered in unfolding IRT models including the GGUM. Additionally, because genetic algorithms do not use derivatives to maximize the likelihood function, it is computationally simple and could be deployed efficiently with higher dimensional data. The focus of this study was to implement the genetic algorithm in the context of the GGUM, and then evaluate the speed and accuracy of the resulting parameter estimates Program development was done with the R computer language, and the efficacy of estimates was examined with simulation methods, which systematically vary sample size, test length and number of response categories. The resulting estimation strategy was also illustrated with real data from an abortion attitude questionnaire.