摘要
对条件ｌｏｇｉｓｔｉｃ回归模型配合适度的分析，除了总配合适度检验统计量外，还包括有对匹配组和匹配组内每一个体的残差分析和影响诊断。本文介绍了离差残差和Ｐｅａｒｓｏｎ残差的理论，采用一步近似法计算删除某匹配组或匹配组内某个体后的配合适度变化，包括参数估计值的改变量、参数可信区间的位移以及Ｐｅａｒｓｏｎ卡方改变量，用来识别影响大的匹配组和匹配个体。用１∶２匹配子宫内膜癌病例－对照研究资料为例，进行了残差和影响诊断的分析，并提出了对待强影响点的意见。
In addition to comprehensive statistics of goodness of fit,analysis of fitting conditional logistic regression models should includes residual analysis and influence diagnosis for each macthed set and every individuals within sets.The article illustrates the theories of deviance and Pearson residuals,introduces one step approximation to calculate the possible changes in goodness of fit after one set or one individual within the set been deleted from the model.With and without the set or the individual of the set in the model,the change amount of parameter estimates,the confidence region displacement of these estimates and the change amount of Pearson chi square are used to identify sets or individuals which give great influence on model fitting.An example of 1∶2 matched case control data is used to illustrate the procedure of residual analysis and influence diagnosis,Some points for handling strong influence points is suggested.
出处
《中国卫生统计》
CSCD
北大核心
1997年第1期13-16,共4页
Chinese Journal of Health Statistics
