Descriptive Statistics

The descriptive statistics table for tenure (Months with service) under the custom imputation model with constraints shows that the problem of negative imputed values for tenure has been solved.

Figure 2. Descriptive statistics for marital (Marital status)

The table for marital (Marital status) now has an imputation (3) whose distribution is more in line with the original data, but the majority are still showing a greater proportion of the cases estimated as being married than in the original data. This could be due to random variation, but might require further study of the data to determine whether these values are not missing at random (MAR). We will not pursue this further here.

Figure 3. Descriptive statistics for lninc (Log of income)

Like tenure, and all the other scale variables, lninc (Log of income) does not show negative imputed values. Moreover, the mean values for imputations are closer to the mean for the original data than in the automatic imputation run — in the income scale, the mean for the original data for lninc is approximately e^3.9291=50.86, while the typical mean value among the imputations is very roughly e^4.2=66.69. Additionally, the maximum values for each imputation are closer to the maximum value for the original data.