Pattern-Mixture Models

The high sensitivity of selection modeling results to the correct specification of the measurement model as well as the dropout model, about which little is often known, has been extensively documented. See also Sections 15.3, 15.4, 17.1, 17.2.2, and 17.5. This has lead to growing interest in pattern-mixture modeling, based on the factorization (15.2) (Little 1993, Glynn, Laird and Rubin 1986, Hogan and Laird 1997). After initial mention of pattern-mixture models (Glynn, Laird, and Rubin 1986, Little and Rubin 1987), they are receiving more attention lately (Little 1993, 1994a, 1995, Hogan and Laird 1997, Ekholm and Skinner 1998, Molenberghs, Michiels, Kenward, and Diggle 1998, Molenberghs, Michiels, and Kenward 1998).