Abstract:
This chapter provides some missing-data situations when, under missing-at-random assumptions, the observed/missing mechanism can be thought of as corresponding to an additional phase of sampling. The class includes some studies that are not normally thought of as multi-phase designs. Multi-phase sampling is the term used in the survey sampling literature where multi-stage sampling already has another well-established meaning. Recall the design variables are variables used in a sampling plan. The theory above assumes that the true selection probabilities depend only on design variables that are obtained at Phase 1 and are present in the regression model. The main design variable was an indicator of whether or not a child was a Sudden Infant Death Syndrome victim and it makes no sense to include this as a predictor of inoculation status. A particular feature of the Women's Health Initiative study, however, invites consideration of a more complex design.