ABSTRACT
In longitudinal studies, observed data are often incomplete since it is difficult to observe all data for every variable at each measurement time, especially if the studies last a long time. Incomplete data are also common in other types of studies, including cross-sectional studies. In many studies, data may also be censored or mis-measured. Specifically, the following problems are common in practice: (i) missing data: the data are completely missing, (ii) censored data: the true data are not observed but they are known to be in certain ranges, (iii) mis-measured data: the true data are not observed but their mis-measured versions are observed, (iv) outliers: the observed data may or may not be the true values but they are inconsistent with majority of the observed data. For last three cases, although the true data may not be observed, partial information is available for the true values so they may also be called incomplete data. To avoid confusion, we use the term missing data to refer to data that are completely missing and use the term incomplete data to refer to the more general cases, i.e., one of the above four cases. In this chapter, we review some commonly used general methods for incomplete data problems.
