Imputation for statistical inference with coarse data

Abstract

Coarse data is a general type of incomplete data that includes grouped data, censored data, and missing data. The likelihood-based estimation approach with coarse data is challenging because the likelihood function is in integral form. The Monte Carlo EM algorithm of Wei & Tanner [Wei & Tanner (1990). Journal of the American Statistical Association, 85, 699704] is adapted to compute the maximum likelihood estimator in the presence of coarse data. Stochastic coarse data is also covered and the computation can be implemented using the parametric fractional imputation method proposed by Kim [Kim (2011). Biometrika, 98, 119132]. Results from a limited simulation study are presented. The proposed method is also applied to the Korean Longitudinal Study of Aging (KLoSA). The Canadian Journal of Statistics 40: 604618; 2012 (c) 2012 Statistical Society of Canad