A Bayesian approach to deriving population estimates from multiple datasets (Joint Seminar with CPC) Seminar

Many countries are looking for new ways to exploit administrative data for population estimation. The talk will describe a Bayesian approach being developed at Statistics New Zealand. It will include an application to estimating population by age and sex in 66 regions of New Zealand. The approach has some nice features. There is no need to link datasets at the individual level. Random variation in demographic processes and measurement are dealt with naturally. Evaluations and weights for the datasets emerge automatically out of the estimation process. However, as the application demonstrates, population estimation is still difficult when the available datasets all have similar flaws.