Imputation procedure and inference in presence of imputed data : application to industries

Nonresponse occurs in various types of surveys. Unfortunatly, missingness cannot be simply ignored. Imputation is often used to fulfil the missing values and it is well known that imputed values cannot be treated as observed ones. In the literature, several imputation methods are developed. They are created to be convenient in function of the type of missingness and the further analysis to realize. Complexe inference methods in presence of imputed data are also detailed in literature. A real challenge is to consider the range of methods when the survey data are delivered, while satisfying requirements of the study. Indeed, it may be laborious to unravel the complex theory and apply it correctly in respect to the data and the objectives of the survey. In this paper, the steps of a study with missing values are dissected and applied to two dataset on industries with missing values. The choice of imputation method is discussed and inference is realized with the filled dataset. This show how to use modern and advanced techniques to improve the accuracy of the estimates. The quality of the imputation method and the estimators of the population total of variables of interest are evaluated with the industry datasets.

, Fifth International Conference on Establishment Surveys, Geneva