Influencing factors of disability among the elderly in China, 2003-2016: application of Bayesian quantile regression

Background: More and more disabled elderly need long-term care as China becomes an aging society. In 2016, there were 220 million people over the age of 60, and nearly 10 million completely disabled elderly people who cannot complete Activities of Daily Living (ADLs). Therefore, the topic of influencing factors for disability among the elderly in China has attracted close attention from researchers, most of which use the traditional empirical methods, such as Ordinary Least Squares (OLS) and logistic. Objective: The purpose of this paper was to introduce the Bayesian Quantile Regression (BQR) method to the topic of the disabled elderly, which was achieved by using BQR to study the influencing factors of disability among the elderly in China during 2003-2016. Methods: This paper was the first attempt to use the BQR for the influencing factors of disability among the elderly in China. Furthermore, a comparison was made between the regression results of BQR, OLS, Quantile Regression (QR), and Bayesian Linear Regression (BLR). Results: It was found that there was a relatively stable relationship between chronic diseases and disability, although there was a little difference in different quantiles. In addition, the BQR can obtain results similar to the traditional method. For instance, the coefficient of chronic diseases (to total disability) obtained by OLS, QR, and BLR were basically consistent (around 0.778), which was similar to BQR. The BQR not only provided estimates for all the quantiles, but also provided upper and lower values of a certain confidence interval. Conclusions: By applying the BQR to the influencing factors of disability among the elderly in China, we reached the conclusion that BQR methods are adaptable for this research topic because of their characteristics and advantages over the traditional methods, such as less strict constraints, the estimates for all quantiles, and the combination of historical information with prior information. Moreover, the BQR method appropriately obtained the lower and upper values in a confidence interval, which can provide prediction space for the future.