220
reads

Yen-Ling Kuo

2012-03-09
13:00:00 - 14:40:00

103 , Mathematics Research Center Building (ori. New Math. Bldg.)

The EM algorithm proposed in Dempster, Laird, and Rubin (JRSSB, 1977) is an efficient iterative procedure to compute the Maximum Likelihood (ML) estimate of parameters in statistical models, where the model depends on unobserved latent variables. Dempster et al. (1977) showed that the rate of convergence of the EM is governed by the fraction of missing information, and in the neighborhood of mle of unknown parameters. Then Meng and Rubin (JASA, 1991) proposed SEM algorithm to find the incomplete-data variance- covariance matrices for parameters utilizing this fact and complete- data variance-covariance matrices for mle of unknown parameters. In this talk, we present SEM algorithm and use random censoring data as an example to illustrate the idea and the proof argument used in Meng and Rubin (1991). When time allows, we will also illustrate how it can be used in the rater agreement problem.