ABSTRACT
Missing or incomplete data are a serious problem in many fields of research and is ubiquitous in medical research, randomized control trials. Missing data points is unavailable data, whether for the response or one or more covariate values or an entire questionnaire. The issue of missing data is in particular very frequent in longitudinal/repeated measures studies. Given that in many studies the problem of missing data is unavoidable, researchers are forced to accept a certain amount of missing data points. An important question is how much missing data is acceptable. Similar to the maximum likelihood method, except that multiple imputation generates actual raw data values suitable for filling in gaps in an existing database. The expectation maximization approach to missing data handling was documented extensively by R. J. A. Little and D. Rubin. The investigator can determine if the missing data pattern has a predictive power in the model, either by itself or in conjunction with another predictor.
