[Rd] Wrongly converging glm()

[Harm-Jan Westra]

Dear Joris,

I’ll be more careful in my wording next time; thanks for the pointer, and thanks for the discussion. This whole process has been quite educational! 😉. I think we’ve reached a consensus here, where the situation as it is right now has been chosen to allow for flexibility of R’s glm() function.

With kind regards,

Harm-Jan



From: Joris Meys<mailto:jorismeys at gmail.com>
Sent: Thursday, July 20, 2017 16:06
To: Harm-Jan Westra<mailto:westra.harmjan at outlook.com>
Cc: r-devel at r-project.org<mailto:r-devel at r-project.org>
Subject: Re: [Rd] Wrongly converging glm()




On Thu, Jul 20, 2017 at 8:32 PM, Harm-Jan Westra <westra.harmjan at outlook.com<mailto:westra.harmjan at outlook.com>> wrote:
My apologies if I seemed to ‘blame R’. This was in no way my intention. I get the feeling that you’re missing my point as well.

I get that now. But you're on R-devel and you started with the claim that R "falsely reports...". That looks like a bug report, and that's why I initially answered that R correctly reports it converged. Maybe to the wrong value, but it converged.


What strikes me as odd is that R would warn you when your data is faulty for a function such as cor(), but not for glm(). I don’t see why you wouldn’t want to check both convergence criteria if you know multiple of such criteria exist. It would make the software more user friendly in the end.

The unfitness of the data bears no relation to the convergence criterium and vice versa. These data checks should be done before the convergence algorithm is even started, and as Mark Leeds also indicated, that's one hell of a job to do. That said, the glm function has an argument "method" by which you can provide an alternative version of glm.fit().  Adapting that one to use another convergence criterium is rather trivial, so technically R even allows you to do that out of the box. No patches needed.


I agree ‘that everybody using R should first do the effort of learning what they're doing’, but it is a bit of a non-argument, because we all know that, the world just doesn’t work that way, plus this is one of the arguments that has held for example the Linux community back for quite a while (i.e. let’s not make the software more user friendly because the user should be more knowledgeable).

That's a wrong analogy imho. You can expect Linux to be user friendly, but not "I will detect every logical fallacy in the article you're writing in this text editor" friendly. And honestly, that's a bit what you're asking R to do here. I understand why, but there's always cases that will be missed. And I wouldn't dare to speak in the name of the R core team, but I can imagine they have a little more urgent issues than helping my students to pass their statistics course ;-)
Cheers
Joris


--
Joris Meys
Statistical consultant

Ghent University
Faculty of Bioscience Engineering
Department of Mathematical Modelling, Statistics and Bio-Informatics

tel :  +32 (0)9 264 61 79
Joris.Meys at Ugent.be
-------------------------------
Disclaimer : http://helpdesk.ugent.be/e-maildisclaimer.php


	[[alternative HTML version deleted]]