Authors:Kirsten Piller (Office for National Statistics, UK) (Speaker) Solange Correa-Onel (Office for National Statistics, UK)Alison Whitworth (Office for National Statistics)
The Office for National Statistics is looking at using more administrative and survey data to produce typical census outputs, in an Administrative Data Census. However, survey data often have very small or null sample sizes within areas, and administrative data may not cover the entire population. The census will continue to be a detailed source of information for some time after it is carried out, but for intercensal estimates it would also become outdated. This leads to the use of data sources in combination to produce reliable estimates for small areas and sub-groups of the population for which there are small or no samples. Small Area Estimation methods provide a framework for combining data sources in Official Statistics.
References:Luna Hernandez, A., Zhang, L., Whitworth, A. and Piller, K. (2015). Small Area Estimates of the Population Distribution by Ethnic Group in England: A Proposal Using Structure Preserving Estimators. Statistics in Transition New Series and Survey Methodology Joint Issue: Small Area Estimation 2014, 16(4), pp. 585-602.
Zhang, L. and Chambers, R. (2004). Small area estimates for cross-classifications. Journal of the Royal Statistical Society, B, 66(2), 479–496.
Generalised Structure Preserving Estimation (GSPREE) can combine the available data in a model to produce estimates for categorical population or household characteristics. The methods are flexible in that additional data sources can be introduced as available and existing sources excluded if they are no longer relevant.
GSPREE has been used to produce population estimates for ethnic groups by local authorities in England (Luna-Hernandez et al., 2015; Zhang and Chambers, 2004). The performance of the GSPREE estimator has now been assessed in a validation scenario where the population distribution is known (i.e. March 2011 Census), with a longer time period between 2001 Census auxiliary information and the reference date of the GSPREE estimates. A GSPREE model also considered the contribution of the different auxiliary information (2001 Census and 2011 English School Census) for various age groups. Uncertainty in the GSPREE estimates was estimated using a bootstrap.