Nonresponse Adjustment using Classification Trees

Nonresponse in household surveys can be a threat to the quality of statistics. Research shows that often the response to these surveys is selective with respect to demographic characteristics like age and household composition. For this reason estimators are usually adjusted to account for nonresponse.Nonresponse adjustment methods make use of covariates that are available for both respondents and non-respondents. A problem is the selection of covariates that relate both to the key survey questions and to the response behaviour. Therefore, often the process of selection is performed in two steps.We present a classification tree method that allows for the construction of weighting strata that simultaneously account for the relation between response behaviour, survey questions and covariates. We apply the classification trees to survey data of Statistics Netherlands.