[R-sig-ME] Error bars for treatment interactions in Gaussian and Binomial mixed models

[Tom Wenseleers]

Dear Sarah,
The effect package I believe would give confidence intervals as if they would be calculated from the SE on the Wald statistics, so this is not totally ideal. 
Would the lsmeans package perhaps work better for you? That I believe can also work with glmer objects, so should cover both the Gaussian and binomial case...
(The plotting you'll have to do yourself though, unlike in the effects package.)

Cheers,
Tom

-----Original Message-----
From: r-sig-mixed-models-bounces at r-project.org [mailto:r-sig-mixed-models-bounces at r-project.org] On Behalf Of John Fox
Sent: 26 August 2013 19:27
To: 'Sarah Dryhurst'
Cc: r-sig-mixed-models at r-project.org
Subject: Re: [R-sig-ME] Error bars for treatment interactions in Gaussian and Binomial mixed models

Dear Sarah,

You might try effect(), Effect(), or allEffects() in the effects package.

I hope this helps,
 John

> -----Original Message-----
> From: r-sig-mixed-models-bounces at r-project.org [mailto:r-sig-mixed- 
> models-bounces at r-project.org] On Behalf Of Sarah Dryhurst
> Sent: Monday, August 26, 2013 12:03 PM
> To: r-sig-mixed-models at r-project.org
> Subject: [R-sig-ME] Error bars for treatment interactions in Gaussian 
> and Binomial mixed models
> 
> I have a model that I am running via lme4:
> 
> m1 <- lmer (Dependent ~ Tmt1 * Tmt2 + (1 | Block / Tmt1))
> 
> I am running this model for each of the three years of my experiment 
> in turn.  Each treatment is a two level factor (treatment and control).
> It is
> a split plot design (with Treatment 2 embedded within Treatment 1 
> embedded within block).
> 
> I'm interested in plotting each level of the interaction between the 
> two treatments (i.e. 4 means), and some measure of error around each 
> of these treatment means, for each year in turn.  This first set of 
> models are Gaussian, but I have another set which use a binomial error 
> distribution.
> 
> I am wondering if there is any package available that calculates 
> interaction treatment means and their error bars for such models?
>  Everything I come across completely ignores the random effect 
> variation in error bar calculation (e.g.the effects package)...
> 
> I tried plotLMER.fnc in language R which was useful for my Gaussian 
> model, but doesn't work for the Binomial one (as the error 
> calculations are based on an mcmc simulation).  Further I cannot 
> actually extract the HPD 95% CI values for each treatment that 
> plotLMER gives me!
> 
> Using the ez package I have been able to predict the treatment means 
> and calculate the variance for these means, but I am unsure how to 
> convert these variances to appropriate error estimations by hand (as I 
> am assuming I need to take account of the random effect variance 
> too?).  Again this is only for the Gaussian models...
> 
> If anyone has any suggestions it would be much appreciated!  Sorry if 
> this is a simple question - I am new to these models :-)
> 
> Thanks in advance,
> 
> Sarah
> 
> 	[[alternative HTML version deleted]]
> 
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list 
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

_______________________________________________
R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models