Methods for combining administrative data to estimate population counts

Abstract

Governments require information about population counts and characteristics in order to make plans, develop policies and provide public services. The main source of this information is the traditional population censuses. However, they are costly, and the information collected by the decennial censuses goes out-of-date easily. For this reason, this thesis has two main aims: to develop methodologies to combine administrative data sources to estimate population counts in the absence of both a traditional census, and to produce uncertainty estimates for the estimated population counts.

Although, the methodologies are illustrated using administrative data sources from England and Wales, they can easily be applied to other countries' administrative data sources. The most comprehensive administrative sources in England and Wales are the NHS Patient Register and the Customer Information System. However, it is known that both of these sources exceed the census estimates. Therefore, it is crucial to use another source to adjust the bias to estimate population counts using these administrative sources.

Three substantial chapters assessing methodologies to combine administrative sources with the auxiliary information are presented. The first of these chapters presents a basis methodology, log-linear models with offsets, which is extended in the following chapters. The second chapter extends these models by using individually linked administrative sources. The third chapter improves on the basis models to produce measures of uncertainty.

This thesis evaluates different log-linear models in terms of their capacity for producing accurate population counts for age group, sex and local authority groups both within the classical and the Bayesian framework. On the other hand, it also presents a detailed perspective to understand which population groups tend to be missed by the administrative data in England and Wales, and how much they can be improved just by combining them with the specific association structures obtained from auxiliary data sources.

More information

Identifiers

ORCID for Peter Smith: orcid.org/0000-0003-4423-5410

ORCID for Peter Van Der Heijden: orcid.org/0000-0002-3345-096X

Catalogue record

Export record