Restrictive imputation of incomplete survey data

Missing data form an ubiquitous source of problems that most scientists or researchers cannot escape. For example, in survey applications, such as in social sciences or in official statistics, where vast amounts of data are collected, respondents often neglect to answer one or more items.

An often applied solution is imputation. With imputation, some estimation procedure is used to impute (fill in) each missing value, resulting in a completed dataset that can be analyzed as if the data were completely observed.

In this dissertation only plausible imputations—imputations that could be real values if they had been observed—are considered to be satisfactory. Plausibility should consider the imputed value and the relation that imputed value has to other (observed and imputed) values for the same entity in the data. For example, if variables sum up to a certain total, only those imputations are plausible that obey the structure of the sum.

This dissertation focuses on finding plausible imputations when there is some restriction posed on the imputation model. The restrictions, and resulting missing data problems are real-life situations frequently encountered across different domains of statistics, such as official statistics, social sciences, geology and medicinal sciences.

Vink G. (2015). Restrictive imputation of incomplete survey data. Dissertation, Utrecht University, handle:1874/308699.