回归诊断在医学中的应用

Application of Regression Diagnostics in Medicine

本课题应用SPSS/PC^+TM对回归诊断在医学上的应用进行了研究。研究内容分为三个部分:残差分析、数据变换、影响分析。残差分析发现,原始数据在正态性、方差齐性和独立性上均偏离了回归假设。使用Box-Cox变换可使数据满足回归假设。影响分析发现,异常值和强影响点的存在对回归分析中的参数估计有重要影响。分析结果表明:当使数据满足回归假设后,回归分析的效果有了明显改善。

An application of regression diagnostics in medicine was studied utilizing SPSS/PC+TM. The project is diveded into 3 parts: residual analysis, data transformation and influence analysis.First, The departure of raw data from the hypothesises required in the regression analysis is found using residual analysis. Then we make the departure disappered throughout Box-Cox transformation and we have also found two outliers and influence cases based on influence analysis. Although the study includes 78 cases, the partial regression coefficients change large with deletion of one or two cases. Thus, even with resonably large data sets,single observation may have a great impact on the results Of an analysis. All procedures are illustrated on a study of primary glaucoma.