ABSTRACT
Many statistical problems can be formulated as discrete missing data problems (MDPs), e.g., Dawid & Skene (1979) may be the first to apply the EM algorithm to find MLEs of observed error rates in the absence of a golden/well-accepted standard by treating the disease status as the missing data. Other examples include change-point problems (Carlin et al., 1992; Pievatolo & Rotondi, 2000), capture and recapture models (George & Robert, 1992; Robert & Casella, 1999, p.307), finite mixture models (McLachlan, 1997), normal mixture model with left censoring (Lyles et al., 2001), sample survey with nonresponse (Albert & Gupta, 1985; Chiu & Sedransk, 1986), crime survey (Kadane, 1985), misclassified multinomial models (Tian et al., 2003), inference from nonrandomly missing categorical data (Nordheim, 1984), zero-inflated Poisson models (Lambert, 1992; Rodrigues, 2003), medical screening/diagnostic tests (Johnson & Gastwirth, 1991; Joseph et al., 1995; Zhou et al., 2002) and bioassay (Qu et al., 1996).
