Logistic Regression Multiple Linear Diagnosis And Improvement And Its Application In Medicine

This thesis studies the diagnosis and enhancing methods of the multi-collinearity of logistic regression model, to solve the problems caused by the multi-collinearity of variables in medical research, such as the instability of model coefficients and difficulty in conclusion explanations, and finally to enable medical researchers to establish correct and reasonable logistic regression model, deal with mixed factors, conduct disease predictions and identify and categorize them,In order to diagnose the multi-collinearity of logistic regression model, the author mainly adopts the following methods applicable in medical research:binary correlation coefficient r, variance inflation factor VIF, tolerance TOL, eigenvalue analysis. In connection with the enhancing methods, the author chiefly introduces principal component regression, namely, on condition that the loss of information is not significant and practical application is needed, to integrate the information of variables of originally higher interrelatedness into a certain number of main ingredients of lower interrelatedness, then substitute main ingredients for original variables and involve them in regression for the separation of interrelated variables. In addition, the author makes a brief introduction of the improvements conducted by partial least squares regression on logistic regression model.Through case analyses, Like multiple linear regression, logistic regression model is of same sensitivity to multi-collinearity. It can be concluded: in medical research, especially in the analyses of pathological factors of epidemics, the application of principal component regression and partial least squares regression to the improvements of multi-collinearity can reduce the multi-collinearity between matrices of variables, build up relatively ideal relation models and improve the reliability of results. SPSS 17.0 statistical software is employed in all calculations.