Study on the Initial Values of the Latent Ability Distribution When Estimating the IRM Parameters Using EM Algorithm
Yaxin Kong, Xichang Wang · 2019
Using the EM algorithm can estimate the item parameters and the latent ability distribution parameters in the item response models at the same time. However, the initial values for the latent ability distribution parameters have a noteworthy impact on the parameters estimation results. In this paper, a method of adjusting initial values for the latent ability distribution based on mean and variance is presented. It can reduce the influence of initial values on estimation results so as to make the estimation results within a reasonable range and promote the practical application of the estimation of latent ability distribution. The feasibility of this adjustment method is illustrated by comparing with the marginal maximum likelihood Bock-Aitken algorithm.