Inserting missing values

Survey data almost always contains a considerable amount of missing values. In close-to-reality simulation studies, the variability due to missing data therefore needs to be considered. Three types of missing data mechanisms are commonly distinguished in the literature. For example, (Little and Rubin 2002): missing completely at random (MCAR), missing at random (MAR), and missing not at random (MNAR).

In the following example, missing values are inserted into the equalized household income of non-contaminated households with MCAR, that is, the households whose values are going to be set to NA are selected using simple random sampling. In order to compare the scenario without missing values of a scenario with missing values, the missing value rates 0 percent and 5 percent are used. The number of samples is reduced to 50 and only the contamination levels 0 percent, 0.5 percent, and 1 percent are investigated to keep the computation time of this motivational example...