中国1982年后人口普查和抽样调查中死亡漏报的估计——基于Bayesian分层回归模型  

Estimation of Death Underreporting in Population Censuses and Sample Surveys of China since 1982

作　　者：李成[1] 米红[1] Li Cheng;Mi Hong(School of Public Affairs,Zhejiang University)

机构地区：[1]浙江大学公共管理学院,浙江杭州310058

出　　处：《人口研究》2022年第1期19-36,共18页Population Research

摘　　要：利用Bayesian分层回归模型估计中国1982年后历次人口普查和抽样调查的死亡漏报率及其随时间的变化。结果表明:0岁死亡漏报率随时间的变化具有明显的阶段性,而1~4岁死亡漏报率随时间的变化趋势不明显;全年龄、5~14岁和15~59岁、60~89岁和90岁及以上死亡漏报率随时间的变化基本近似且同样具有阶段性。受漏报原因和不同时期社会背景的影响,女性1~4岁的死亡漏报率不一定低于0岁的死亡漏报率。人口普查成人阶段各年龄死亡漏报率差别较大,而人口抽样调查则较为一致。2000年人口普查男性和2010年全人口90岁及以上死亡率偏低是由比较严重的死亡漏报造成。死亡漏报对预期寿命计算误差的影响因年龄而异,二者关系在婴幼儿中较弱,其余年龄死亡漏报和预期寿命误差存在显著线性正相关。This paper estimates the death underreporting rates in population censuses and sample surveys and their time trend since 1982 through Bayesian hierarchical regression model. While the trend of death underreporting rates at age 0 has distinct phases, the trend of underreporting rates at age 1 to 4 is inconspicuous;distinct phases and similarity also exist in the changes of death underreporting rates over time at all ages, 5-14, 15-59, 60-89, 90 and over. Due to the changing causes of under-reporting and social environments at different periods, female death underreporting rates at age 1 to 4 are not necessarily lower than age 0. In censuses, death underreporting rates at different adult ages have large disparities, while those in sample surveys are relatively consistent. Lower mortality rates at age 90 and over in male of 2000 census and in both sexes of 2010 census are caused by serious underreporting. The effect of death underreporting on the calculation error of life expectancy varies with age, and the relationship between the two is weaker under age 5, while the relationships at other ages are significantly positively linear-correlated.

关 键 词：Bayesian模型 死亡漏报 人口普查和抽样调查 死亡率 数据质量 

分 类 号：C921[社会学—人口学]