Abstract

Population and migration statistics are some of Stats NZ's most widely used outputs, with high customer expectations of accuracy and timeliness. Current population estimates are based on a full field-enumeration census of the population and a post-enumeration survey used to measure and adjust for the coverage errors in the census. Between censuses, demographic accounting is used to adjust the population for aging, births, deaths, and internal and international migration. The flow components of the demographic accounting are currently largely derived from administrative data. Stats NZ has been researching alternative methods for producing the stock component (small domain population estimates) using primarily administrative data, employing methodology from Bayesian dual/multiple-system estimation and machine learning. In this talk, we present our research and development to date.