Abstract:
|
The Economic Census collects information on the revenue obtained from products from all sampled units. The collection is quite challenging as establishments can report values from a long list of potential products in a given industry. Moreover, product descriptions are quite detailed, many products are mutually exclusive, and reported products are subjected to strict additivity constraints. Consequently, legitimate missing values occur frequently and nonresponse is quite high. Auxiliary data are not available, and other predictors such as total receipts are often weakly related. In the 2017 Economic Census, missing product data will be imputed using hot deck imputation, and variance estimates for product data will be published for the first time. The variance estimator must account for sampling variance, calibration weighting, and imputation variance. Thompson, Thompson,and Kurec (2016) present results of a simulation study that examines the first two factors. This focusses on the estimation of the imputation variance component. Using a simulation study, we compare the statistical properties of this component estimated both the model-based and a design-based (model-assisted) frameworks.
|