A Census has to produce accurate and reliable estimates of the population, not just at the national level but also at lower geographical detail. However, it is also widely known that despite all the efforts of the census, there will be some people missed. Therefore, the final population estimates have be adjusted for this undercount.
In the UK Census at a national level, this is achieved through a combination of a post-enumeration survey, referred to as the Census Coverage Survey (CCS), and two statistical methodological approaches known as Dual System Estimation and Ratio Estimation. In addition, to obtain lower level estimates of the population, Small Area Estimation techniques are employed.
The use of model based small area estimation methods in the UK Census was introduced in 2001 within the One Number Census (ONC) methodology. The same approach will be adopted for the 2011 UK Census however new features were tested. The methodology aims at obtaining local level population estimates by age-sex groups adjusted for undercount.
The UK Census process includes: the Census itself; the Census Coverage Survey (CCS); the matching of Census and CCS; the use of Dual System and ratio estimation to estimate undercount; the process to obtain local area model based population estimates and the production of a census database with individual and household level records consistent with the estimates.
Small area estimation techniques based on regression models will be employed to produce local level estimates using data from the CCS and the Census. The underlying idea of small area estimation is to exploit similarities in order to borrow strength over areas. Regression models that relate the CCS and Census count are then used as a tool for describing the relationships so as to produce the local level model based estimates.
However, although more precise, the resulting model-based estimates are biased. In this study, various regression models were considered and the objective of the small area estimation procedure was to find the estimator that balances the trade-off between variance and bias, yielding estimates with good precision and as little bias as possible.
This paper reports the research undertaken for developing the small area methodology for the 2011 UK Census, the framework for evaluation and the results of the study.
Keywords: Census Methodology; Population Estimates; Small Area Estimation; Post Enumeration Survey
Biography: Dr Denise Silva is a Principal Methodologist at the Brazilian Institute of Geography and Statistics and a Senior Lecturer at the National School of Statistical Sciences in Rio de Janeiro, Brazil. She previously worked as a Lecturer at the Southampton Statistical Sciences Research Institute (S3RI) at the University of Southampton and as a Principal Methodologist at the UK Office for National Statistics (ONS). Her main research interests focus on small area estimation, time series analysis, survey sampling and statistical modelling in the social sciences. She completed her PhD in Statistics at the University of Southampton and has an MSc and a BSc in Statistics.