ABSTRACT
There are several methods that are often misused for the analysis of HRQoL studies when the data are not MCAR. The objective of this chapter is to convey the relationship between the assumptions of MCAR or MAR and commonly used analytic procedures. When missingness depends on observed data (Y obsi ) and the analytic procedure does not include all available data, the results are biased. If dropout is dependent on a covariate (Xi) and the covariate is not included in the analytic model, then the results are biased. Popular analytic methods such as repeated univariate analyses and MANOVA of complete cases are examples of procedures that exclude observed data; results from these analyses are unbiased only if the missing data are MCAR [105].
