Re: [R] Mixed effects model with binomial errors - problem
anyone? RFTW wrote: ok... the model now runs properly (say, without errors). Now about the result. These are the averages per treatments tapply(VecesArbolCo.VecesCo.C1,T2,mean) a b c d 0.49 0.56 0.45 0.58 I run this very simple model summary(model1-lmer(cbind(VisitsExpTree,TotalVisits-VisitsExpTree)~ treatment +(1|Individual), family=binomial, data=r)) Generalized linear mixed model fit by the Laplace approximation Formula: cbind(VisitsExpTree,TotalVisits-VisitsExpTree)~ treatment +(1|Individual) Data: r AIC BIC logLik deviance 242.3 255.9 -116.2232.3 Random effects: GroupsNameVariance Std.Dev. Individuo (Intercept) 0.14075 0.37517 Number of obs: 112, groups: Individuo, 37 Fixed effects: Estimate Std. Error z value Pr(|z|) (Intercept) 0.372280.19031 1.9562 0.05044 . treatmentb 0.033670.24520 0.1373 0.89079 treatmentc -0.606060.23330 -2.5978 0.00938 ** treatmentd -0.255040.22790 -1.1191 0.26311 --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Correlation of Fixed Effects: (Intr) T2bT2c T2b -0.675 T2c -0.697 0.543 T2d -0.720 0.544 0.581 wouldnt we expect the intercept to be roughtly the mean of treatment a? and thus the estimate of treatmentb to be +0.07, c: -0.04 and d: +0.09 roughly? Is this model just completely not estimating well, or are the estimates not the 'real values'. I tried to get teh predict function to give me the 4 predicted values based on the model, but i havent succeeded in doing so. maybe someone can help me on that one too (predict(model1,type=response) doesnt work) thnx -- View this message in context: http://www.nabble.com/Mixed-effects-model-with-binomial-errorsproblem-tp19413327p19566778.html Sent from the R help mailing list archive at Nabble.com. __ R-help@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.
Re: [R] Mixed effects model with binomial errors - problem
First of all I'm forwarding this mail to the R-SIG-mixed-models, which is more appropriate to your question. Remember that family = bionomial uses by default the logit link. Hence all parameters will be on the logit scale. So you will need to backtransform them for comparison. Then you'll see that the parameters are much closer to the averages. They still differ, but that is due to the difference in model. Your averages are essentially something like summary(model1-glm(cbind(VisitsExpTree,TotalVisits-VisitsExpTree)~ treatment +(1|Individual), family=binomial, data=r)) library(boot) intercept - 0.37228 treatmentb - 0.03367 treatmentc - -0.60606 treatmentd - -0.25504 inv.logit(intercept) [1] 0.5920098 inv.logit(intercept + treatmentb) [1] 0.6001164 inv.logit(intercept + treatmentc) [1] 0.4418197 inv.logit(intercept + treatmentd) [1] 0.5292765 HTH, Thierry ir. Thierry Onkelinx Instituut voor natuur- en bosonderzoek / Research Institute for Nature and Forest Cel biometrie, methodologie en kwaliteitszorg / Section biometrics, methodology and quality assurance Gaverstraat 4 9500 Geraardsbergen Belgium tel. + 32 54/436 185 [EMAIL PROTECTED] www.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: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] Namens RFTW Verzonden: vrijdag 19 september 2008 8:16 Aan: r-help@r-project.org Onderwerp: Re: [R] Mixed effects model with binomial errors - problem anyone? RFTW wrote: ok... the model now runs properly (say, without errors). Now about the result. These are the averages per treatments tapply(VecesArbolCo.VecesCo.C1,T2,mean) a b c d 0.49 0.56 0.45 0.58 I run this very simple model summary(model1-lmer(cbind(VisitsExpTree,TotalVisits-VisitsExpTree)~ treatment +(1|Individual), family=binomial, data=r)) Generalized linear mixed model fit by the Laplace approximation Formula: cbind(VisitsExpTree,TotalVisits-VisitsExpTree)~ treatment +(1|Individual) Data: r AIC BIC logLik deviance 242.3 255.9 -116.2232.3 Random effects: GroupsNameVariance Std.Dev. Individuo (Intercept) 0.14075 0.37517 Number of obs: 112, groups: Individuo, 37 Fixed effects: Estimate Std. Error z value Pr(|z|) (Intercept) 0.372280.19031 1.9562 0.05044 . treatmentb 0.033670.24520 0.1373 0.89079 treatmentc -0.606060.23330 -2.5978 0.00938 ** treatmentd -0.255040.22790 -1.1191 0.26311 --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 Correlation of Fixed Effects: (Intr) T2bT2c T2b -0.675 T2c -0.697 0.543 T2d -0.720 0.544 0.581 wouldnt we expect the intercept to be roughtly the mean of treatment a? and thus the estimate of treatmentb to be +0.07, c: -0.04 and d: +0.09 roughly? Is this model just completely not estimating well, or are the estimates not the 'real values'. I tried to get teh predict function to give me the 4 predicted values based on the model, but i havent succeeded in doing so. maybe someone can help me on that one too (predict(model1,type=response) doesnt work) thnx -- View this message in context: http://www.nabble.com/Mixed-effects-model-with-binomial-errorsproble m-tp19413327p19566778.html Sent from the R help mailing list archive at Nabble.com. __ R-help@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. Dit bericht en eventuele bijlagen geven enkel de visie van de schrijver weer en binden het INBO onder geen enkel beding, zolang dit bericht niet bevestigd is door een geldig ondertekend document. The views expressed in this message and any annex are purely those of the writer and may not be regarded as stating an official position of INBO, as long as the message is not confirmed by a duly signed document. __ R-help@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.
Re: [R] Mixed effects model with binomial errors - problem
ok... the model now runs properly (say, without errors). Now about the result. These are the averages per treatments tapply(VecesArbolCo.VecesCo.C1,T2,mean) a b c d 0.49 0.56 0.45 0.58 I run this very simple model summary(model1-lmer(cbind(VisitsExpTree,TotalVisits-VisitsExpTree)~ treatment +(1|Individual), family=binomial, data=r)) Generalized linear mixed model fit by the Laplace approximation Formula: cbind(VisitsExpTree,TotalVisits-VisitsExpTree)~ treatment +(1|Individual) Data: r AIC BIC logLik deviance 242.3 255.9 -116.2232.3 Random effects: GroupsNameVariance Std.Dev. Individuo (Intercept) 0.14075 0.37517 Number of obs: 112, groups: Individuo, 37 Fixed effects: Estimate Std. Error z value Pr(|z|) (Intercept) 0.372280.19031 1.9562 0.05044 . treatmentb 0.033670.24520 0.1373 0.89079 treatmentc -0.606060.23330 -2.5978 0.00938 ** treatmentd -0.255040.22790 -1.1191 0.26311 --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Correlation of Fixed Effects: (Intr) T2bT2c T2b -0.675 T2c -0.697 0.543 T2d -0.720 0.544 0.581 wouldnt we expect the intercept to be roughtly the mean of treatment a? and thus the estimate of treatmentb to be +0.07, c: -0.04 and d: +0.09 roughly? Is this model just completely not estimating well, or are the estimates not the 'real values'. I tried to get teh predict function to give me the 4 predicted values based on the model, but i havent succeeded in doing so. maybe someone can help me on that one too (predict(model1,type=response) doesnt work) thnx -- View this message in context: http://www.nabble.com/Mixed-effects-model-with-binomial-errorsproblem-tp19413327p19436083.html Sent from the R help mailing list archive at Nabble.com. __ R-help@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.
[R] Mixed effects model with binomial errors - problem
Hi, We released individual birds into a room with 2 trees. We counted the number of visits to each of the 2 tree. One of the trees is always a control tree and the other tree is either treatment 1, treatment 2 or treatment3 or treatment 4. Ind Treat ContrTree ExpTree Total visits 1 1 11 16 27 1 2 6 9 15 1 3 5 13 18 1 4 11 25 36 2 1 2 3 5 4 1 6 7 13 4 3 4 4 8 4 4 2 5 7 6 1 1 1 2 6 4 5 16 21 etc etc (as you see, not all treatments are included for all individuals) Our question is if the proportion of visits to the experimental tree, in relation to the total number of visits to both trees differs between treatments. We have made treatment and individual into a factor All individuals were subjected to a maximum of 4 treatments, so 'individual' is a random factor We came up with this model: model1-lmer(cbind(ExpTree,Total visits-ExpTree)~ Treat +(1|Ind),method=ML , family=binomial, data=r)) However, the error we get is this: Error in match.arg(method, c(Laplace, AGQ)) : 'arg' should be one of “Laplace”, “AGQ” HELP! -- View this message in context: http://www.nabble.com/Mixed-effects-model-with-binomial-errorsproblem-tp19413327p19413327.html Sent from the R help mailing list archive at Nabble.com. __ R-help@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.
Re: [R] Mixed effects model with binomial errors - problem
ok, the model does run now! but, dont i need the method=ML when i want to compare this model with a reduced model using anova(model1, model2)? The R-Book tells me that REML is not good for that (p.635) So, besides that... how do i now do a sort of posthoc test to see 1) estimates of all treatments and 2) which treatments are different from which. Luc. PS. the names are not the original names in the file Ben Bolker wrote: RFTW l.temarvelde at nioo.knaw.nl writes: Our question is if the proportion of visits to the experimental tree, in relation to the total number of visits to both trees differs between treatments. We have made treatment and individual into a factor All individuals were subjected to a maximum of 4 treatments, so 'individual' is a random factor We came up with this model: model1-lmer(cbind(ExpTree,Total visits-ExpTree)~ Treat +(1|Ind),method=ML , family=binomial, data=r)) why not leave out method=ML and see what happens? for the current iteration of lmer, REML is not a possibility in any case. The default Laplace method should work OK. I'd be slightly worried about your variable name with a space in it (`Total visits`) -- are you sure that is working as expected? For further questions along these lines I would suggest e-mailing [EMAIL PROTECTED] instead ... Ben Bolker __ R-help@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. -- View this message in context: http://www.nabble.com/Mixed-effects-model-with-binomial-errorsproblem-tp19413327p19414516.html Sent from the R help mailing list archive at Nabble.com. __ R-help@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.
Re: [R] Mixed effects model with binomial errors - problem
RFTW l.temarvelde at nioo.knaw.nl writes: ok, the model does run now! but, dont i need the method=ML when i want to compare this model with a reduced model using anova(model1, model2)? The R-Book tells me that REML is not good for that (p.635) maximum likelihood (ML) is the default [and only!] estimation method for generalized linear mixed models (GLMMs, or in other words lmer with a family argument) in lmer. It is correct that you shouldn't compare REML fits with different fixed effects, but unfortunately it's almost as bad (not theoretically wrong, but very unreliable for small to moderate sample sizes) to use likelihood ratio tests (as in anova(model1,model2)) for these tests. See http://www.zoo.ufl.edu/bolker/glmm_rev-26aug.pdf for more information ... Ben Bolker __ R-help@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.
