3.5.2.2 Internal validity

The logistic regression model involving directors’ liability was analysed for residuals. SPSS identified three outliers.1 I discovered an interesting pattern. The outliers involved cases which did not score on one of the factors in the model (Table 7), but nevertheless involved determination of directors’ liability. To protect the data from being distorted by the outliers, I applied bootstrapping. It has been argued that bootstrapping is a very suitable validation method.2 Bootstrapping follows the idea of pooling but, by repeatedly resampling the data, it studies the uncertainty in the frequency estimate obtained.3 By resampling with replacement from the original sample, SPSS created one thousand alternate versions of the data set and in that way reduced the impact of the outliers. The aim of the bootstrapping was therefore to enhance the stability and reliability of the model by using the entire data set for validation purposes.4 I used the method for both regression models: ‘subjective bad faith’ and ‘directors’ liability’. In my case, after bootstrapping (with 95% percentile confidence interval), the models remained stable and did not lead to different conclusions.