中文摘要：

在近200年高龄人口死亡率研究中,学者们对高龄死亡率减速已达成共识,且认为人口异质性是导致减速现象的最常见解释。然而,近年来的研究则表明这种减速是源于夸大年龄、数据异质性、使用死亡概率指标而非危险率导致的估计下行偏差。为此,本文将死亡数据质量问题、数据同质性和差异性融入到5类经典Logistic死亡模型中,通过构建具有一致性的分层建模框架,在深度诠释我国高龄人口死亡率性别差异、区域差异及动态改善基础上,探讨我国高龄人口死亡率减速到底是估计偏差还是事实？研究表明,当数据质量良好时,描述死亡率减速的Logistic模型最优且对数据质量改变的敏感程度最低,即我国高龄人口死亡率减速不是估计下行偏差,而是事实。最后也对近年研究得出的下行偏差给出了解释。

英文摘要：

From the studies of mortality rates at old ages over the past 200 years, scholars have reached a consensus about mortality deceleration at extreme old ages or advanced ages, and population heterogeneity is now considered as the most common explanation of mortality deceleration phenomenon. However, in some recent studies it is believed that mortality deceleration at advanced ages is resulted from the downward biases in mortality estimates, such as the age exaggeration, the data heterogeneity, and the death probabilities instead of hazard rates. In the research, it incorporates mortality data quality issues and data homogeneity and heterogeneity into the five classical logistic-type models, and puts forward a consistent hierarchical modeling framework in order to explain and quantify the gender differences, regional differences, and dynamic improvement of morality at old ages, upon which investigations are based as to biases or facts in relation to deceleration in the age pattern of old-age mortality rates in China. The results show that if data have high quality, the logistic model describing mortality deceleration has the best goodness-of-fit, which has the lowest sensitivity for the changes of mortality data quality. Therefore, deceleration in the age pattern of old-age mortality rates in China is not a downward bias of mortality estimates, but a fact. Finally, we also provide some explanations for the downward biases in some recent studies.