Diagnostics of fitted binary logistic regression model based on individual subjects and covariate patterns

In logistic regression, before concluding that the model fits well, it is crucial that other measures be examined to see if goodness-of-fit is supported over the entire set of covariate patterns. This is accomplished through a series of specialized measures known as logistic regression diagnostics. In this study, one-step approximation diagnostics for logistic regression are computed on the basis of individual subjects as well as covariate patterns. The plots suggest that the outliers and influential observations are more clearly visualized and detected whether the diagnostics are computed on the basis of covariate patterns than individual subjects. So, diagnostic statistics should be computed taking into account covariate patterns specially when the number of covariate patterns is much smaller than the sample size and the number of subjects within any covariate patterns is larger than five. Finally, it may be concluded that one should not proceed to presenting the results from a fitted logistic regression model until the fit of the model has been thoroughly assessed using both summary measures and diagnostic statistics. The diagnostic statistics should be computed on the basis of covariate patterns, if necessary.

Actions (login required)

View Item