[R-sig-ME] Plotting best fit lines binomial GLMM

[Fox, John]

Dear M.,

The effects package does work with GLMMs fit with glmer() in the lme4 package. See ?Effect. Here's an example adapted from ?glmer:

	library(effects)
	library(lme4)
	library("HSAUR2")
	gm2 <- glmer(outcome~treatment*visit+(1|patientID),
            	 data=toenail, family=binomial, nAGQ=20)
	Effect(c("treatment", "visit"), gm2)

producing

treatment*visit effect
              visit
treatment              1         2          3          4           5           6            7
  itraconazole 0.2236820 0.1155113 0.05588527 0.02612852 0.012014461 0.005481597 0.0024920184
  terbinafine  0.2104865 0.0871212 0.03303451 0.01208159 0.004358651 0.001564643 0.0005606578

I hope this helps,
 John

-----------------------------
John Fox, Professor
McMaster University
Hamilton, Ontario
Canada L8S 4M4
Web: socserv.mcmaster.ca/jfox




> -----Original Message-----
> From: R-sig-mixed-models [mailto:r-sig-mixed-models-bounces at r-
> project.org] On Behalf Of M West
> Sent: January 31, 2016 11:29 PM
> To: Phillip Alday <Phillip.Alday at unisa.edu.au>
> Cc: r-sig-mixed-models at r-project.org
> Subject: Re: [R-sig-ME] Plotting best fit lines binomial GLMM
> 
> Thanks for this suggestions Philip  - it looks like the effects package doesn't
> work for GLMMs - it works with glms.....
> 
> On Sun, Jan 31, 2016 at 1:05 AM, Phillip Alday <Phillip.Alday at unisa.edu.au>
> wrote:
> 
> > Addressing the plotting issue: look at the effects package. You can
> > directly plot effects objects (which will yield lattice plots) or you
> > can convert them to data frames and plot by hand (e.g. if you want
> > more control and/or ggplot).
> >
> > Best,
> > Phillip
> >
> > On 30/01/16 08:18, M West wrote:
> > > Main questions:
> > > (1) How to extract coefficients from GLMM to plot best fit lines to data?
> > > (2) Are there other options for dealing with these sorts of data
> > > besides mixed effects models (or RM ANOVA)?
> > >
> > >
> > > Specifics: I have a short time series data across 12 sites over 8 years.
> > > I'd like an omnibus plot that summarizes the main pattern interest
> > > in
> > these
> > > data.
> > >
> > > The dependent variable is frequency females (data are # smokers out
> > > of
> > the
> > > total population). The independent variable is also a frequency (#
> > infected
> > > out of the total population).
> > >
> > > Plotting each year separately it's easy to see the positive
> > > correlation between smokers and infection. However, given the
> > > variation among years, plotting all the data on a single plot
> > > obscures the overall pattern....I need to fit regression lines to
> > > each year.
> > >
> > > I know how to do this with lme....but I can't quite find how to do
> > > this with GLMM and I've analyzed the data with a GLMM with a
> > > binomial distribution (following Crawley) [While the data are
> > > binomial, they are not binary (i.e., not 0 and 1)so a logistic curve
> > > doesn't work].
> > >
> > >
> > > I found this thread on inspecting the residuals but I haven't been
> > > able
> > to
> > > find anything on plotting a best fit line for these type of data.
> > >
> > >
> > http://stats.stackexchange.com/questions/70783/how-to-assess-the-fit-o
> > f-a-binomial-glmm-fitted-with-lme4-1-0
> > >
> > >
> > > I would *much prefer* to use something other than mixed effects
> > > models (I think the results are not straightforward to interpret and
> > > every book or blog recommends a different approach) for this
> > > analysis so if there are other suggestions they are also welcome!
> > >
> > > Thanks,
> > > M.
> > >
> > >       [[alternative HTML version deleted]]
> > >
> > > _______________________________________________
> > > R-sig-mixed-models at r-project.org mailing list
> > > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
> > >
> >
> 
> 	[[alternative HTML version deleted]]
> 
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models