Modelling of response quality degeneration from lengthy survey using CRF

Kwonhyun Kim and Cheongtag Kim

Abstract

Kim et al.(2014) suggested three ways of quantifying the survey response quality, which are entropy rate, likelihood of IRT and likelihood of CRF. Entropy rate considers only the sequence of the response, whereas IRT considers the ability of the testee and the charateristics of the items. But in IRT, given the ability of the testee, the responses are independent each other, which means no cosideration in given to the sequence of the responses. The CRF model from Kim et al.(2014) considers the latent variables and the characteristics of the items but no consideration is given to the ability of the testee. So I suggest a model that considers the ability of the testee, characteristics of the items and the sequence of the responses simultaneously and show it outperforms any of the models suggested before.