[R-sig-ME] Binomial Temporal GAMM does not converge (R::mgcv)

[Ken Beath]

First rule of statistics is start with the simplest reasonable model and
then try to build something more complex if it is necessary.

I would start with a standard mixed effects with random effects for
intercept and slope, judging from what you have already. This can be done
in lmer, you will probably need to set the nAGQ value to something greater
than 1, increase it until nothing seems to change. This is due to the high
variance for the intercept random effect, and is probably causing the
problem with your other analysis, as it uses PQL which just won't work well
in this case.

Then look at residuals versus fitted values, with a lowess curve as they
will be very noisy, and see if you need something more complex.  If so a
regression spline is easiest using ns() and a small number of knots.

Ken

On 5 June 2015 at 09:05, Benjamin Kellman <bkellman at eng.ucsd.edu> wrote:

> Hello R sig,
>
> I'm new to mixed effect models and I would really appreciate some help.
> I've posted by question on cross validated:
>
>
> http://stats.stackexchange.com/questions/155524/binomial-temporal-gamm-does-not-converge-rmgcv
>
> If any of you would care to weigh in, it would be greatly appreciated.
>
> Thank you
>
> --
> Benjamin P. Kellman
>
> PhD Student
> Bioinformatics and Systems Biology
> UC, San Diego
>
>         [[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>



-- 

*Ken Beath*
Lecturer
Statistics Department
MACQUARIE UNIVERSITY NSW 2109, Australia

Phone: +61 (0)2 9850 8516

Building E4A, room 526
http://stat.mq.edu.au/our_staff/staff_-_alphabetical/staff/beath,_ken/

CRICOS Provider No 00002J
This message is intended for the addressee named and may...{{dropped:9}}