[R-sig-ME] nAGQ = 0
bachlaw01 at outlook.com
Sun Sep 3 17:48:19 CEST 2017
I have not studied this extensively with smaller datasets, but with larger datasets --- five-figure and especially six-figure n --- I have found that it often makes no difference.
When uncertain, I have used a likelihood ratio test to see if the differences are likely to be material.
My overall suggestion would be that if the dataset is small enough for this choice to matter, it is probably also small enough to solve the model through MCMC, in which case I would recommending using that, because the incorporated uncertainty often gives you better parameter estimates than any increased level of quadrature.
Sent from my iPhone
> On Sep 3, 2017, at 3:15 AM, Rolf Turner <r.turner at auckland.ac.nz> wrote:
> Greetings mixed models gurus.
> Some time ago I asked a question of this list concerning problems that I was having getting a glmer() model to fit to a data set. (Using the binomial family with the cloglog link.)
> It was suggested to me, in the first instance by Tony Ives (thanks again, Tony), that I should include the "nAGQ=0" option in my call to glmer(). I did so, and it worked like a charm.
> I do not however really understand what "nAGQ=0" actually does. I gather (vaguely) that it has something to do with the numerical integrals needed when evaluating the log like, and I (even more vaguely) gather that with nAGQ=0 this integration is somehow entirely dispensed with. Perhaps I am miss-stating things here.
> Be that as it may, it worries me slightly that I (apparently) have to use a somewhat less precise method than I otherwise might in order to
> get any answer at all.
> What risks am I running by setting nAGQ=0? What perils and pitfalls lurk? Surely there must be a downside to using this (???) short cut.
> Although the upside, that I actually get an answer, pretty clearly trumps (apologies for the use of that word :-) ) the downside.
> I would like some advice, pearls of wisdom, whatever from someone who understands what is going on in the underpinnings of fitting mixed models.
> Thanks for any wise counsel that you can provide.
> Rolf Turner
> Technical Editor ANZJS
> Department of Statistics
> University of Auckland
> Phone: +64-9-373-7599 ext. 88276
> R-sig-mixed-models at r-project.org mailing list
More information about the R-sig-mixed-models