GLM influence goodness-of-fit sensitivity

[telferm57]

GLMInfluence does not yet support goodness of fit sensitivity around the deviance and chi-sqaured goodness of fit statistics (as stated in the documentation)

It should be fairly straightforward to include these, but I cannot see an enhancement story for it (perhaps I'm wrong?)

[josef-pkt]

I don't see a specific issue for this. There are several general issues to extend influence measures to more models.

Some of the problems in adding and extending methods is in finding numbers from other packages for unit tests.

Do you have code, examples, reference to other packages, ...?
Anything would be helpful.

Essentially I ran out of energy when I added GLMInfluence and similar, and was theoretically more interested in figuring out generic solutions for MLE, QMLE and M-estimators, so I didn't finish adding GLM specific measures.

[telferm57]

Would R be the best place to start for this?

[josef-pkt]

I'm not sure, I don't remember without searching again.
The SAS manual is also a good source for some of the statistics.

R has a lot of packages, but often it takes me a lot of time to find the relevant things, and if it's not one of the big packages, then I don't always trust the results.
Overall, including (Q)MLE, statsmodels has more outlier, influence measures than R and I didn't find many of those functions in it.
Pure GLM support in R is better than MLE/QMLE.

[telferm57]

I'll have a dig around sas and R ... QMLE is beyond my current level of understanding, btw!

[josef-pkt]

It would be good to get any GLM specific measures that are missing.

QMLE, M-estimators are difficult topics. I like general and generic solutions so we can use it across many models. That's why I spent time digging into that literature, which is not easy to figure out and translate into code.