Detection of multiple outliers in logistic regression and examination of effectiveness

Regression is a method of analysis performed to examine the relationship between a dependent variable and one or more independent variables. In regression analysis, dependent variable is generally held continuous and measurable. But in some cases, dependent variable is binary and it takes only discrete values. Since dependent variable has binominal probability distribution in such cases, regression model should be set up as a probability model. In that sense, regression model can be designed in accordance with logistic distribution function. Like the linear regression models, there may be points at the end of the data set space also in logistic regressions. These observations which are known as outlier observations need to be determined and it should be elaborated whether they have any impact on goodness of fit of the regression functions and/or parameter estimates. Some single observation methods have been developed to identify outliers. The most frequent approach in practice is using such kind of methods to examine the outliers. However, in case of masking and swamping, single observation methods fail. Therefore, a method which will not be affected by these problems is required. In this study, the method of Generalised Standard Pearson Residuals (GSPR) which can be used effectively for identification of multiple outliers in logistic regression has been applied to the data concerning Turkey 2009 Elite Women Volleyball League and firm bankrupties and it has been proved that the method works effectively and eliminates the effect of masking.