|
Abstract:
|
Economic data are subject to many logical linear restrictions, such as for example the fact that most variables need to be non-negative. Standard imputation techniques do not take these restrictions into account and are therefore likely to provide imputations that do not satisfy all restrictions on the data. Additionally, it is difficult to find an appropriate multivariate model as economic survey data usually consist of a large number of variables, which may have several distributional forms and specific properties such as the linear bounds. In this paper the data are modeled through conditionally specified distributions. A sequential regression approach is used to univariately impute each variable, where the regression models depend on the type of variable that is imputed. The performance of this method is assessed by comparing it to other common imputation methods.
|
