Publication at Faculty of Mathematics and Physics |
2012
Abstract
This study is aimed at variance computation techniques for estimates of population characteristics based on survey sampling and imputation. We use the superpopulation regression model, which means that the target variable values for each statistical unit are treated as random realizations of a linear regression model with weighted variance.
We focus on regression models with one auxiliary variable and no intercept, which have many applications and straightforward interpretation in business statistics. Furthermore, we deal with cases where the estimates are not independent and thus the covariance must be computed.
We also consider chained regression models with auxiliary variables as random variables instead of constants.