RDP 2003-09: Housing Leverage in Australia Appendix A: Income Imputation and Results

As discussed in Section 2.2, the nature of the missing data leaves us with the need to impute income for three separate types of missing cases. For Type I individuals we impute total gross financial year income. For Type II individuals we impute gross financial year wage and salary income and add this imputed income to their reported gross financial year non-wage and salary income. For Type III individuals we impute gross financial year non-wage and salary income and add this imputed income to their reported gross financial year wage and salary income. Table Al contains all the relevant results.

In all cases, missing values are imputed using the predictive mean matching (PMM) method outlined in Little (1988). In the first stage this involves estimating a regression on the variable to be imputed for individuals without missing values – in our case income. Next the model with the highest R² is used to predict the income of individuals with missing values. For every missing value we find the record with the nearest predicted value. The actual value of this ‘donor’ is then imputed for the missing value. The advantages of using the PMM method over other single imputation methods, such as simply imputing the conditional mean obtained from a regression, are that it ensures that only feasible values of the variable are imputed, and that a random error component is introduced so that imputed values have a similar variance to the reported values (ISER 2002).