ABSTRACT
We apply Bayesian demographic accounting to the problem of estimating and forecasting population, births, and deaths in China. We recognize differences by age. We ignore international migration.
Our data sources are: (1) (unadjusted) Chinese census data for the years 1990, 2000, and 2010, disaggregated by 5-year age group; (2) United Nations estimates of births and deaths, by 5-year age group, for the periods 1991--1995, 1996–2000, …, 2011–2015.
Although each series looks plausible on its own, we uncover inconsistencies under an accounting framework. Many cohorts gain in size over at least one period. The steep drops in the size of the cohorts born in the 1990s and 2000s imply much higher death rates for children than are normally attributed to China over this period.
The posterior means for cohort sizes from our model respect the accounting constraints. No cohort increases in size over time. Furthermore, the sharp drops in cohort size immediately after birth have gone, implying much more plausible levels of childhood mortality.
We also obtain a rich set of outcomes: estimates and forecasts of population counts, birth counts, death counts, birth rates, death rates, and life expectancy, as well as coverage ratios for the data sources.
