基于logistic回归与决策树模型的社会资本对老年人自评健康的影响研究

The Impact of Social Capital on Elderly Self-rated Health based on Logistic Regression and Decision Tree Model

目的探讨社会资本对老年人自评健康的影响,为有针对性地开发利用社会资本提供依据。方法基于2018年中国老年健康影响因素跟踪调查(CLHLS)数据,以自评健康为因变量,以社会资本(人际信任、社会参与、社会支持)为自变量,并纳入个人人口学特征、社会经济特征、生活方式等作为协变量,分别建立二分类logistic回归模型和基于CHAID算法的分类决策树模型分析社会资本对老年人自评健康的影响,并分析比较两种模型的差异。结果logistic回归结果显示,人际信任水平、社会参与水平、社会支持水平高的老年人,其自评健康状况更好,OR值分别为1.141、1.918、1.377;分类决策树分析结果显示,在自评经济水平中等且患2种及以上慢病的老年人中,社会参与水平越高,其自评健康状况越好,而在自评经济水平中等且患3种及以上慢病的老年人中,人际信任越低,其自评健康状况越差;两种模型分析结果比较显示,二分类logistic回归模型的灵敏度为68.7%,特异度为56.9%;CHAID分类决策树模型的灵敏度为76.7%,特异度为45.1%。结论利用社会资本例如定期组织集体活动,加强社区文体娱乐设施建设,帮助家庭适应新时代的养老需求等对促进老年自评健康具有积极意义。两种模型均具有一定的分类预测价值,但各有优劣,logistic回归模型的特异度较高,分类决策树模型的灵敏度较高,二者可互补,从而使分析结果更具有实际意义。

Objective To explore the impact of social capital on the self-rated health of the elderly, in order to provide evidence-based decision-making basis for government to develop and utilize social capital in a targeted manner.Methods Based on the data from the Chinese Longitudinal Healthy Longevity Survey(CLHLS)in 2018,taking self-rated health as the dependent variable, social capital(interpersonal trust, social participation, social support)as the independent variable, and personal demographic characteristics, socioeconomic characteristics, lifestyle, etc.as covariates, a binary logistic regression model and a classification decision tree model based on CHAID algorithm were established, respectively.Furthermore, the differences between the two models were analyzed and compared.Results The logistic regression results showed that elderly people with high levels of interpersonal trust, social participation, and social support have better self-rated health.The OR values were 1.141,1.918,and 1.377 respectively.The results of classification decision tree based on the CHAID algorithm showed that among the elderly with ≤2 types of chronic diseases who have a moderate self-assessment economic level, the higher the level of social participation, the better the self-rated health status.While among the elderly with ≥3 types of chronic diseases in the moderate self-evaluated economic level, the lower the interpersonal trust, the worse their self-rated health status.From the comparison of the two models, it could be seen that for the binary logistic regression model, the sensitivity is 68.7%,and the specificity is 56.9% for the classification decision tree model, the sensitivity is 76.7%,and the specificity is 45.1%.Conclusion It is of positive significance in promoting self-rated health in the elderly to regularly organize collective activities, strengthen the construction of community cultural, sports and recreational facilities, and help families adapt to the needs of the elderly in the new era.Both models had certain classification prediction value, but each had its own advantages and disadvantages.The logistic regression model carried high specificity and the classification decision tree model had high sensitivity.Therefore, the two could complement each other, so that the analysis results would be more practical.