Dear all, I am quite new to R so apologies if I fail to ask properly. I have done a test comparing bat species richness in five habitats as assessed by three methods. I used a linear mixed model in lme4 and got habitat, method and the interaction between the two as significant, with the random effects explaining little variation. I then ran Tukey's post hoc tests as pairwise comparisons in three ways: Firstly in lsmeans: lsmeans(LMM.richness, pairwise~Habitat*Method, adjust="tukey") Then in ‘agricolae’: tx <- with(diversity, interaction(Method, Habitat)) amod <- aov(Richness ~ tx, data=diversity) library(agricolae) interaction <-HSD.test(amod, "tx", group=TRUE) interaction Then in ghlt 'multcomp': summary(glht(LMM.richness, linfct=mcp(Habitat="Tukey"))) summary(glht(LMM.richness, linfct=mcp(Method="Tukey"))) tuk <- glht(amod, linfct = mcp(tx = "Tukey")) summary(tuk) # standard display tuk.cld <- cld(tuk) # letter-based display opar <- par(mai=c(1,1,1.5,1)) par(mfrow=c(1,1)) plot(tuk.cld) par(opar) I got somewhat different levels of significance from each method, with ghlt giving me the greatest number of significant results and lsmeans the least. All the results from all packages make sense based on the graphs of the data. Can anyone tell me if there are underlying reasons why these tests might be more or less conservative, whether in any case I have failed to specify anything correctly or whether any of these post-hoc tests are not suitable for linear mixed models? Thankyou for your time, Claire [[alternative HTML version deleted]]
Bert Gunter
2014-May-22 16:29 UTC
[R] Post-hoc tests on linear mixed model give mixed results.
Wrong list! This does not concern R programming. Post on the r-sig-mixed-models list instead in **PLAIN TEXT** rather than html. Cheers, Bert Bert Gunter Genentech Nonclinical Biostatistics (650) 467-7374 "Data is not information. Information is not knowledge. And knowledge is certainly not wisdom." H. Gilbert Welch On Thu, May 22, 2014 at 6:52 AM, Claire <c.wordley at live.com> wrote:> Dear all, > > I am quite new to R so apologies if I fail to ask properly. I have done a test comparing bat species richness in five habitats as assessed by three methods. I used a linear mixed model in lme4 and got habitat, method and the interaction between the two as significant, with the random effects explaining little variation. > > I then ran Tukey's post hoc tests as pairwise comparisons in three ways: > > Firstly in lsmeans: > lsmeans(LMM.richness, pairwise~Habitat*Method, adjust="tukey") > > Then in ?agricolae?: > > tx <- with(diversity, interaction(Method, Habitat)) > amod <- aov(Richness ~ tx, data=diversity) > library(agricolae) > interaction <-HSD.test(amod, "tx", group=TRUE) > interaction > > Then in ghlt 'multcomp': > summary(glht(LMM.richness, linfct=mcp(Habitat="Tukey"))) > > summary(glht(LMM.richness, linfct=mcp(Method="Tukey"))) > > tuk <- glht(amod, linfct = mcp(tx = "Tukey")) > summary(tuk) # standard display > tuk.cld <- cld(tuk) # letter-based display > opar <- par(mai=c(1,1,1.5,1)) > par(mfrow=c(1,1)) > plot(tuk.cld) > par(opar) > > I got somewhat different levels of significance from each method, with ghlt giving me the greatest number of significant results and lsmeans the least. All the results from all packages make sense based on the graphs of the data. > > Can anyone tell me if there are underlying reasons why these tests might be more or less conservative, whether in any case I have failed to specify anything correctly or whether any of these post-hoc tests are not suitable for linear mixed models? > > Thankyou for your time, > Claire > > [[alternative HTML version deleted]] > > > ______________________________________________ > R-help at r-project.org mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code. >