[R-sig-ME] A little help on lmer format and interpretation
Thierry.ONKELINX at inbo.be
Fri Feb 11 16:29:16 CET 2011
Do you have repeated measurements on the trees? In that case the maximal model would be lmer(Photo ~ B * T * W * S + (1|TreeID)) or assuming that the site has no effect on any of the treatments: lmer(Photo ~ B + T * W * S + (1|TreeID)).
Since you have only two site, you can't get good variance estimates of the site effect. Hence it is better to keep that in the fixed effects. With a large(r) number of sites you could go for a more complex model: lmer(Photo ~ T * W * S + (T * W * S|B) + (1|TreeID))
Without repeated measurements on the trees you have to drop the (1|TreeID) random effect. Hence you no longer have a mixed model if the number of sites is only 2.
ir. Thierry Onkelinx
Instituut voor natuur- en bosonderzoek
team Biometrie & Kwaliteitszorg
Research Institute for Nature and Forest
team Biometrics & Quality Assurance
tel. + 32 54/436 185
Thierry.Onkelinx at inbo.be
To call in the statistician after the experiment is done may be no more than asking him to perform a post-mortem examination: he may be able to say what the experiment died of.
~ Sir Ronald Aylmer Fisher
The plural of anecdote is not data.
~ Roger Brinner
The combination of some data and an aching desire for an answer does not ensure that a reasonable answer can be extracted from a given body of data.
~ John Tukey
> -----Oorspronkelijk bericht-----
> Van: r-sig-mixed-models-bounces at r-project.org
> [mailto:r-sig-mixed-models-bounces at r-project.org] Namens
> Saunders, Michael R
> Verzonden: vrijdag 11 februari 2011 15:51
> Aan: r-sig-mixed-models at r-project.org
> Onderwerp: [R-sig-ME] A little help on lmer format and interpretation
> I am currently trying to fix an unbalanced split-split plot
> design using lmer(). The design is:
> B = Block (site) in two levels
> T = Trenching (whole-plot treatment) in two levels
> (untrenched vs trenched), replicated 4 times per site
> W = Weeding (split-plot treatment) in two levels (weeded vs
> unweeded), applied to each whole-plot
> S = Species (split-split plot treatment) in two levels
> (chestnut vs oak), with 1-2 trees of each in each split plot
> treatment (we had mortality, otherwise it would be balanced)
> My response variable is Photo (photosynthesis levels).
> I specified my full model as:
> fullmod<-lmer(Photo ~ B + T + W + S + B:T + B:W + B:S + T:W +
> T:S + W:S + B:T:W + B:T:S + B:W:S + T:W:S + B:T:W:S + (1|B) +
> (1|B:T) + (1|B:T:W),data=light,REML=F)
> QUESTION 1: Is the code correct?
> In my output, I have no variance or standard deviation
> associated with the B, B:T, B:T:W random effects. I am
> assuming that this is because I have only two levels at each split.
> QUESTION 2: Is that assumption correct?
> Lastly, my plans were to drop each higher-order interaction
> in succession and use the anova comparison between the
> successively simplified models (i.e., full vs. no 4-way; no
> 4-way vs. no 3-way; etc.) to get p-values for the fixed
> effects (obviously not using REML).
> QUESTION 3: Is this a proper approach?
> I am self taught in mixed-models and R, so any help is
> appreciated. lme4 is new to me.
> Thanks in advance,
> Mike Saunders
> Asst. Professor of Silviculture
> Purdue University
> [[alternative HTML version deleted]]
> R-sig-mixed-models at r-project.org mailing list
More information about the R-sig-mixed-models