How to deal with districts/cities that have zero sample points?
When integrating survey and population censuses, how do we deal with the difference in reference period?
What to do when geographical boundaries are not aligned across data sources?
Boundary systems in UK exist for varying purposes and very often do not co-exist well with each other. Moreover they often change over time. A particular problem for small area estimation is that the sample design is based on postcode sectors while the usual requirement for small area estimation is for administrative areas such as local authorities or smaller nesting units.
Within ONS, unit level census and sample records were available with an address georeference allowing placement in any of the available digitised boundary systems (this includes postal and administrative geographies). Administrative data were available at aggregated levels for administrative boundaries so the models were designed with a unit level response plus area level covariates and catered for the varying sampling and estimation geographies by separating the fixed effect and the random effect structures.