Paper Title
Improvements in Validity with New Bayes Estimation in the Item Response Theory Model
Abstract
In this study, we propose a simple Bayes-LINEX estimation of an item response model (IRT) under asymmetric
loss and improved validity of ability estimates in educational test data. More precisely, using the Markov Chain Monte Carlo
(MCMC) method, we propose a simple Bayesian estimator of the Rasch model under LINEX loss. [8] We call this estimation
a simple Bayes-LINEX estimation (BLE). The BLE was applied to a total dataset for adult literacy [1], and we found that
estimates using the ability parameter with the simple BLE showed better validity than the estimates using the conventional
Bayesian estimation. We conclude that our simple BLE is a suitable solution for the educational test data with respect to
validity.
Index Terms - Asymmetric loss, educational testing, item response theory, validity..