Hi Judith,
You might want to check out the GLMM FAQ http://glmm.wikidot.com/faq
Your questions are pretty basic stuff which has been covered a number of
times on the list i think.
1) check the FAQ for the P-Value issue. Options include pamer.fnc in
LMERConvenienceFunctions or pvals.fnc in the LanguageR packages for easy
computations. The FAQ has other comments on this though. Your fourth
question is also a method of determining P-Vals.
2) What do you mean by an "overall model statistic"? R2 or similar? The FAQ
covers this too.
Your model summary does just that - it summarises the model: reports
coefficients etc
By comparing the models you can test to see if spec improves the model (it
doesnt for your minimised dataset). Look up likelihood ratio tests.
Check the multcomp package for post hoc tests. It does Tukey tests, not
sure about LSD though
Re your random plots - 5 is a very small number (in fact its often regarded
as the absolute minimum of levels for a RE). It could produce quite bad
estimates of variance (as you can see for your model with 2 plots : the
random effects part of the summary output).
HTH
Alan
--------------------------------------------------
Email: aghaynes@gmail.com
Mobile: +41794385586
Skype: aghaynes
On 4 October 2012 11:41, Riedel Judith wrote:
> Dear people of the help list
>
> I am drying to analyze my data using the 'lmer' function and I keep having
> problems.
>
> This is the model:
> > fm1<-lmer(dbh~spec+scheme+(1|Plot),data=d, REML=FALSE).
>
> I analyse tree size (dbh) of 3 different species (spec) and 3 planting
> schemes (scheme). I have 5 plots, which I hope to model as a random factor.
> (However, the subsequent output is based on some simplified dummy data,
> which is based on only two plots and ha only few observations).
>
> No I do:
> > anova(fm1)
> and I get some output, which I don't understand. Looks like this:
>
> Analysis of Variance Table
> Df Sum Sq Mean Sq F value
> spec 2 6.098 3.0490 0.6142
> scheme 2 13.161 6.5803 1.3255
>
> The problems I have are:
> (1) How can I get the P-values?
> (2) How can I get the overall model statistic?
>
> Than I do:
> > summary(fm1)
>
> and get:
> Linear mixed model fit by maximum likelihood
> Formula: dbh ~ spec + scheme + (1 | Plot)
> Data: d
> AIC BIC logLik deviance REMLdev
> 147.2 157 -66.6 133.2 125.8
> Random effects:
> Groups Name Variance Std.Dev.
> Plot (Intercept) 0.0000 0.0000
> Residual 4.9644 2.2281
> Number of obs: 30, groups: Plot, 2
>
> Fixed effects:
> Estimate Std. Error t value
> (Intercept) 6.9074 0.9424 7.329
> specCED 0.3859 1.0265 0.376
> specTAB 0.8585 0.9828 0.874
> schemeMON 0.6572 0.9554 0.688
> schemePRO -1.0344 1.1259 -0.919
>
> Correlation of Fixed Effects:
> (Intr) spcCED spcTAB schMON
> specCED -0.537
> specTAB -0.529 0.500
> schemeMON -0.588 0.002 -0.072
> schemePRO -0.565 0.064 0.063 0.510
>
> What is this? What does it tell me?
>
> The statistics help advised me to do a second model, like this:
> > fm2<-lmer(dbh~scheme+(1|Plot),data=d,REML=FALSE)
> > anova(fm1,fm2)
>
> But why would I compare the two models?
>
> What I get is:
> Data: d
> Models:
> fm2: dbh ~ scheme + (1 | Plot)
> fm1: dbh ~ spec + scheme + (1 | Plot)
> Df AIC BIC logLik Chisq Chi Df Pr(>Chisq)
> fm2 5 143.96 150.97 -66.982
> fm1 7 147.21 157.01 -66.602 0.7584 2 0.6844
>
> What does this mean? Why Chi?
>
> Finally I would like to do some LSD post hoc tests, but I have no idea how
> to do it.
>
> In the end I would like to be able to report something like: 'DBH differed
> significantly between, species, planting schemes, and plots (Fx,xx = X; P =
> X). DBH of species 1 was significantly larger than DBH of species 2 (LSD
> post hoc test, P = X)'.
>
> I greatly appreciate any suggestions! Thank You a lot for Your help!
>
> Kind regards,
>
> Judith
> XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
> Judith Riedel
> ETH Zurich
> Institute of Agricultural Sciences
> Applied Entomology
> Schmelzbergstrasse 9/LFO
> 8092 Zurich
> Switzerland
>
> Tel: ++41 44 632 3923
> Fax: ++41 44 632 1171
> judith.riedel@ipw.agrl.ethz.ch
> http://www.em.ipw.agrl.ethz.ch
>
> _______________________________________________
> R-sig-mixed-models@r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
[[alternative HTML version deleted]]