Çok değişkenli verilerde ayrımsama sorunu ve lojistik regresyon analizi

Abstract

Lastly it has been emphasized that logistic regression analy sis gives good results for data a e 1 3 which deoands on concrete measurements n hile it is insufficient for social applications.

IV SUMMARY In this study, the purpose is, in the case of survey sampling, to find the best model that will allocate obser vations to the groups existing in the data. Since survey sampling data generally contain discrete variables, the normality assumption will not be satisfied and logistic regression analysis is used as an alternative to discri minant analysis. In the First Section, discrimination problem in analysing the multivariate data and the purpose of the study have been pointed out. In the Second Section, dichotomous and polytomous logistic regression analysis, estimation techniques, residual diag nostics and comparison of logistic and discriminant analysis have been examined. In the Third Section, two applications of logistic regres sion analysis, to cardiologic and to student election exam data have been examined and results have been sum marized. In the first application, it has been seen that, multiple group logistic model with main effects which has been fitted to cardiologic data is preferred to two group logistic models for the purpose of estimation and gives better discrimination than discriminant function. In the second application, it has been mentioned that, logistic model is not suitable for student election exam data and this insufficiency is due to the data set. In the last Section, depending on the results, it has been pointed out that, despite multiple group logistic models don't have better discrimination power, they should be preferred to Begg and Gray's individualized logistic mo dels approximation for the purpose of estimation. On the other hand it has been seen that since the assumptions are not satisfied because of discrete variables, discri minant analysis doesn't have sufficient discrimination power.Lastly it has been emphasized that logistic regression analy sis gives good results for data a e 1 3 which deoands on concrete measurements n hile it is insufficient for social applications.