Riedel Judith
2012-Oct-03 15:54 UTC
[R] Difficulties in trying to do a mixed effects model using the lmer function
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 PS. the complete output is attached. 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 at ipw.agrl.ethz.ch<mailto:judith.riedel at ipw.agrl.ethz.ch> http://www.em.ipw.agrl.ethz.ch
Bert Gunter
2012-Oct-03 17:08 UTC
[R] Difficulties in trying to do a mixed effects model using the lmer function
Please post this on the r-sig-mixed-models list, not here. You are waayyy more likely to get useful help there. Also this is primarily a statistics, not an R issue. You may wish to consider consulting with your local statistician to help you understand the statistics. Cheers, Bert On Wed, Oct 3, 2012 at 8:54 AM, Riedel Judith <judith.riedel at ipw.agrl.ethz.ch> 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 > > PS. the complete output is attached. > > > > 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 at ipw.agrl.ethz.ch<mailto:judith.riedel at ipw.agrl.ethz.ch> > http://www.em.ipw.agrl.ethz.ch > > > ______________________________________________ > 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. >-- Bert Gunter Genentech Nonclinical Biostatistics Internal Contact Info: Phone: 467-7374 Website: http://pharmadevelopment.roche.com/index/pdb/pdb-functional-groups/pdb-biostatistics/pdb-ncb-home.htm