Editing and estimation of measurement errors

Data that are collected for the production of official statistics or, more generally, for statistical analyses nearly always contain measurement errors. National statistical institutes, other statistical agencies and academic researchers have therefore developed methods to handle error-prone data. Two broad classes of approaches can be distinguished: editing methods that aim to identify and correct individual errors in the data and estimation methods that try to correct for measurement errors at the analysis stage, without adjusting the data themselves. The aim of this thesis is to contribute to the development of both approaches for dealing with measurement errors, with a particular focus on their extension and application to large data sets from administrative sources.

In particular, the following points are addressed in this thesis:

Firstly, current methods for automatic data editing—based on the seminal work of Fellegi and Holt (1976) —have limited practical applicability because they are based on rather restrictive assumptions. In this thesis, two new methods for automatic editing are developed that relax some of these assumptions.

Secondly, we discuss the estimation of measurement error models with latent variables in an official-statistics context. Here, often univariate descriptive statistics such as population totals and means are of interest. It is demonstrated how latent-variable models could be used to assess the suitability of new data sources for official statistics, to gain better insight into the accuracy of statistics and to improve the quality of statistical output.

Thirdly, an application at Statistics Netherlands is described in which a measurement error model is used to compare the quality of data before and after an automatic editing procedure.

Scholtus, S. (2018). Editing and estimation of measurement errors in administrative and survey data. Dissertation, Vrije Universiteit Amsterdam, handle:1871/55568.