Hi,

A higher Chi-squared value implies a lower p-value and a lower goodness of fit. Each time you build and fit the NB2 model using the instructions in my article, you will get a different goodness of fit for the fitted NB2 since your random chosen training set will be different each time . On the whole, the goodness of the NB2 model on the bicyclist counts data set is not going to be very good as the data set is highly dispersed. So the NB2 model, even though it will do a much better job than the Poisson model to explain te variance in 'y', will not be able to explain most of the variance in 'y'. You may want to calculate the Pseudo-R-squared of the fitted model. It will be able to give you a good feel for how much of the deviance in 'y' the NB2 model has been able to 'explain'. You can get pseudo-R-squared from Statsmodels for your fitted NB2 model.

Do let me know if I have addressed your questions.

'best