[R-sig-ME] LME and TukeyHSD

[Christoph Scherber]

Dear Sibylle,

You could also try to remove the NA´s like this:

mydata<-mydata[is.na(mydata$sinmortality)==F,]

or:

mydata<-na.exclude(mydata) # this may be a bit too strong, especially if many other variables contain NAs

Best wishes
Christoph




Sibylle Stöckli wrote:

> Dear Christoph
> 
> Thanks a lot! I still have some trouble with the is.na function
> 
> #
> mydata<-subset(Kaltenborn,ADDspecies!=1)
> mydata$sinmortality<-asin(sqrt(Kaltenborn$PropMortality))
> mydata<-mydata[!is.na(mydata$sinmortality),]
> error: is.na() applied to non-(list or vector)of type "NULL".
> 
> I tried to use as.list
> #
> mydata<-mydata[!is.na(as.list(mydata$sinmortality)),]
> 
> but this function produced some error within the lme model (varialbe lenght of sinmortality).
> 
> Do you have some idea, how to change the !is.na function for non-list data?
> 
> Thanks Sibylle
> 
> 
> 
> Dear Sibylle,
> 
> The reason for the error message may be that you transform your response
> variable inside the model formula. This frequently causes errors when
> working with components of lme models. In addition, you should try "ML"
> instead of "REML" (when working with the fixed effects).
> 
> You may want to try:
> 
> ##
> 
> mydata<-subset(Kaltenborn,ADDspecies!=1)
> mydata$sinmortality<-asin(sqrt(Kaltenborn$PropMortality))
> mydata<-mydata[!is.na(mydata$sinmortality),]
> 
> modelF<-lme(sinmortality~Diversity+Management+Species
> +Height+Year+Diversity*Year, data=mydata, random=~1|Plot/SubPlot,
> weights=varPower(form=~Diversity), method="ML")
> 
> summary(glht(modelF,linfct=mcp(Species="Tukey")))
> 
> ##
> 
> Be reminded that it may not be sensible to test for Species main effects
> when interactions are present; have you tested for interactions between
> Species and other terms in the model?
> 
> Best wishes,
> Christoph
> 
> 
> 
> 
> 
> Dear R users
> 
> I used lme to fit a linear mixed model inlcuding weights=varPower()
> and subset. Additionally I wanted to use glht to calculate Tukey-
> Kramer multiple comparisions.
> 
> I think that the error in glht() is probably caused by the subset and
> varPower function within the lme model? I would very much appreciate
> an input.
> 
> Thanks
> Sibylle
> 
> LME MODEL
> modelF<-lme(asin(sqrt(PropMortality))~Diversity+Management+Species
> +Height+Year+Diversity*Year, data=Kaltenborn, random=~1|Plot/SubPlot,
> na.action=na.omit, weights=varPower(form=~Diversity), subset=Kaltenborn
> $ADDspecies!=1, method="REML")
> 
> anova(modelF, type="marginal")
>                numDF denDF  F-value p-value
> (Intercept)        1   162  7.12789  0.0084
> Diversity          1    14 12.89284  0.0030
> Management         2    30  5.52544  0.0091
> Species            3   162 41.81003  <.0001
> Height             1   162  2.59134  0.1094
> Year               1   162  7.07660  0.0086
> Diversity:Year     1   162 12.88095  0.0004
> 
> TukeyHSD
> library(multcomp)
> Loading required package: mvtnorm
> Warning messages:
> 1: package 'multcomp' was built under R version 2.7.2
> 2: package 'mvtnorm' was built under R version 2.7.2
> summary(glht(modelF,linfct=mcp(Species="Tukey")))
> 
> error
> Error in glht.matrix(model = list(modelStruct = list(reStruct =
> list(SubPlot = -0.305856275920955,  :
>   ‘ncol(linfct)’ is not equal to ‘length(coef(model))’
> 
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
> 
> 
> 
>