This study aims to apply a systematic statistical approach, including several plot indexes, to diagnose the goodness of fit of a logistic regression model, and then to detect the outliers and influential observations of the data from experimental data.
The proposed statistical approach is applied to analyze some experimental data on internal solitary wave propagation.
A suitable logistic regression model in which the relationship between the response variable and the explanatory variables is found. The problem of multicollinearity is tested. It was found that certain observations would not have the problem of multicollinearity. The P‐values for both the Pearson and deviance χ2 tests are greater than 0.05. However, the Pearson χ2 value is larger than the degrees of freedom. This finding indicates that although this model fits the data, it has a slight overdispersion. After three outliers and influential observations (cases 11, 27, and 49) are removed from the data, and the remaining observations are refitted the goodness‐of‐fit of the revised model to the data is improved.
A comparison of the four predictive powers: R2, max‐rescaled R2, the Somers' D, and the concordance index c, shows that the revised model has better predictive abilities than the original model.
The goodness‐of‐fit and prediction ability of the revised logistic regression model are more appropriate than those of the original model.
Chen, C., Yang, H.P., Chen, C. and Chen, T. (2008), "Diagnosing and revising logistic regression models: Effect on internal solitary wave propagation", Engineering Computations, Vol. 25 No. 2, pp. 121-139. https://doi.org/10.1108/02644400810855940
Emerald Group Publishing Limited
Copyright © 2008, Emerald Group Publishing Limited