Conferences

Comparing the income of tenants and home-owners is difficult since owners pay the service of their dwelling in a very different way than tenants. In order to allow a cross-European comparison of income, Eurostat suggests adding an imputed rent to the income of owners. Important indicators on poverty and social exclusion like the Laeken indicators are based on these adjusted incomes.

Besides owners there are tenants paying a rent which is below a market rent. For both groups rents have to be imputed. The preferred imputation method is based on a regression model for the rent which is developed for observed market rents. Rents for owners are then the fitted values based on this model. The model may have a selection bias since it is developed for tenants with market rent only. In principle, the Heckman correction allows an unbiased estimation. The Heckman method is a two-step procedure which involves a logistic regression to estimate the probability to rent at market price and a main model to estimate the amount of rent. The main model involves a correction term to account for the selection.
Application of the Heckman method to survey data needs adjustments: The models have to be estimated for the population and not only for the sample. In other words, the sampling weights should be taken into account. In addition missing values and outliers have to be taken into account.
The presentation discusses the application of the Heckman method to the data of the Swiss Statistics on Income and Living Conditions (CH-SILC). The modelling process including pitfalls will be shown.

The quality of the imputations depends strongly on the covariates at hand. Therefore the results for the imputation in SILC will be compared with the application of the method to other surveys.