Re: [R] mixed model MANCOVA
Dear Erika et al., I'm not sure exactly what you want to do, but you might take a look at the Anova() function in the car package, which will compute "type-II" or "type-III" tests for multivariate linear models, including models with repeated measures; you can also get the traditional univariate tests. Be very careful of Type-III tests, especially if the covariate is involved in interactions, since it's very easy to test strange hypotheses (as is true even in a univariate context). There are some examples of repeated-measures ANOVAs and MANOVAs in ?Anova. I hope this helps, John -- John Fox, Professor Department of Sociology McMaster University Hamilton, Ontario, Canada web: socserv.mcmaster.ca/jfox > -Original Message- > From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On > Behalf Of hadley wickham > Sent: September-10-08 8:53 PM > To: Erika Crispo > Cc: r-help@r-project.org > Subject: Re: [R] mixed model MANCOVA > > On Mon, Sep 8, 2008 at 1:06 PM, Erika Crispo > <[EMAIL PROTECTED]> wrote: > > Hello, > > > > I need to perform a mixed-model (with nesting) MANCOVA, using Type III sums > of squares. I know how to perform each of these types of tests individually, > but I am not sure if performing a mixed-model MANCOVA is possible. Please let > me know. > > Why? What is the purpose of your analysis? Why do you think a > mixed-effects MANCOVA is appropriate? What have simpler models told > you about the data? What exploratory graphics have you done? > > Hadley > > > -- > http://had.co.nz/ > > __ > 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-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 model MANCOVA
On Mon, Sep 8, 2008 at 1:06 PM, Erika Crispo <[EMAIL PROTECTED]> wrote: > Hello, > > I need to perform a mixed-model (with nesting) MANCOVA, using Type III sums > of squares. I know how to perform each of these types of tests individually, > but I am not sure if performing a mixed-model MANCOVA is possible. Please let > me know. Why? What is the purpose of your analysis? Why do you think a mixed-effects MANCOVA is appropriate? What have simpler models told you about the data? What exploratory graphics have you done? Hadley -- http://had.co.nz/ __ 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 model MANCOVA
Hi Erika, As mentioned, I haven't run the model before and I don't have access to your data set, so you might want to post your reply to the list as well (cc'd again). As another guessing-without-trying-anything, I'd first make sure that in fact pop and family are factor variables with the same length as a,b,c,treat,centroidsize. Also having not used glmer before, I'm not sure how to get the p values...since estimates and std. errors and t-values are reported, the df's are likely known and so they probably exist in the "model" object you created. Of course, your highest t-value is 1.92, so none of your fixed effects would be significant at the .05 level (the two-tailed z-score cutoff is 1.93, which is the limit for t). --Adam On Wed, 10 Sep 2008, Erika Crispo wrote: Thanks! I am still having some problems. I have tried the following: model=glmer(cbind(a,b,c)~pop*treat+centroidsize+(1|pop/family)) Error: Matrices must have same number of columns in rbind2(..1, r) In addition: Warning messages: 1: In family:pop : numerical expression has 104 elements: only the first used 2: In family:pop : numerical expression has 104 elements: only the first used I don't get the error messages if I exclude the nesting (i.e. exclude pop on the RHS). But even then, I don't know how to interpret the output. How can I get P values for pop and treat? I've attached my data file. summary(model) Linear mixed model fit by REML Formula: cbind(a, b, c) ~ pop * treat + centroidsize + (1 | family) AICBIC logLik deviance REMLdev -685.3 -656.2 353.6 -822.8 -707.3 Random effects: Groups NameVariance Std.Dev. family (Intercept) 9.8877e-13 9.9437e-07 Residual 2.3502e-05 4.8478e-03 Number of obs: 104, groups: family, 28 Fixed effects: Estimate Std. Error t value (Intercept) 4.518e-03 4.329e-03 1.0438 popkah -2.338e-03 1.902e-03 -1.2297 popkant-2.328e-03 1.881e-03 -1.2380 poprwe -3.728e-03 1.941e-03 -1.9204 treatn -8.703e-04 1.957e-03 -0.4448 centroidsize-1.886e-06 2.464e-06 -0.7656 popkah:treatn 3.440e-03 2.695e-03 1.2765 popkant:treatn 1.198e-03 2.699e-03 0.4439 poprwe:treatn 4.662e-03 2.746e-03 1.6976 Correlation of Fixed Effects: (Intr) popkah popknt poprwe treatn cntdsz ppkh:t ppknt: popkah -0.228 popkant -0.335 0.507 poprwe -0.193 0.490 0.492 treatn -0.092 0.485 0.476 0.479 centroidsize -0.951 0.009 0.119 -0.023 -0.128 popkah:trtn 0.121 -0.705 -0.352 -0.346 -0.719 0.036 popknt:trtn 0.095 -0.352 -0.680 -0.347 -0.721 0.063 0.520 poprwe:trtn 0.117 -0.346 -0.346 -0.707 -0.706 0.037 0.510 0.511 <>< <>< <>< <>< <>< <>< <>< Erika Crispo, PhD candidate McGill University, Department of Biology http://www.biology.mcgill.ca/grad/erika/index.htm <> ><> ><> ><> ><> ><> ><> ----- Original Message - From: "Adam D. I. Kramer" <[EMAIL PROTECTED]> To: "Erika Crispo" <[EMAIL PROTECTED]> Cc: Sent: Wednesday, September 10, 2008 5:47 PM Subject: Re: [R] mixed model MANCOVA Hi Erika, I have not tried this before, and I hope that somebody will correct me if I'm wrong, but the glmer function in the lme4 library appears to do what you want. From examples(lmer): lmer> (gm1 <- glmer(cbind(incidence, size - incidence) ~ period + (1 | herd), family = binomial, data = cbpp)) ...I guess that this will do what you want it to because it has multiple variables on the LHS and both continuous and categorical variables on the RHS, along with an explicit grouping structure. In your case, you probably want to leave the family= argument out, as noted in ?glmer, "If 'family' is missing then a linear mixed model is fit; otherwise a generalized linear mixed model is fit." ...MANCOVA tend to be generalized linear models. Once again, though, I have not used this system personally, haven't seen your data, and don't know what output to expect. Hopefully somebody else can confirm or deny this solution's efficacy. --Adam On Mon, 8 Sep 2008, Erika Crispo wrote: Hello, I need to perform a mixed-model (with nesting) MANCOVA, using Type III sums of squares. I know how to perform each of these types of tests individually, but I am not sure if performing a mixed-model MANCOVA is possible. Please let me know. Erika <>< <>< <>< <>< <>< <>< <>< Erika Crispo, PhD candidate McGill University, Department of Biology http://www.biology.mcgill.ca/grad/erika/index.htm <> ><> ><>
Re: [R] mixed model MANCOVA
Hi Erika, I have not tried this before, and I hope that somebody will correct me if I'm wrong, but the glmer function in the lme4 library appears to do what you want. From examples(lmer): lmer> (gm1 <- glmer(cbind(incidence, size - incidence) ~ period + (1 | herd), family = binomial, data = cbpp)) ...I guess that this will do what you want it to because it has multiple variables on the LHS and both continuous and categorical variables on the RHS, along with an explicit grouping structure. In your case, you probably want to leave the family= argument out, as noted in ?glmer, "If 'family' is missing then a linear mixed model is fit; otherwise a generalized linear mixed model is fit." ...MANCOVA tend to be generalized linear models. Once again, though, I have not used this system personally, haven't seen your data, and don't know what output to expect. Hopefully somebody else can confirm or deny this solution's efficacy. --Adam On Mon, 8 Sep 2008, Erika Crispo wrote: Hello, I need to perform a mixed-model (with nesting) MANCOVA, using Type III sums of squares. I know how to perform each of these types of tests individually, but I am not sure if performing a mixed-model MANCOVA is possible. Please let me know. Erika <>< <>< <>< <>< <>< <>< <>< Erika Crispo, PhD candidate McGill University, Department of Biology http://www.biology.mcgill.ca/grad/erika/index.htm <> ><> ><> ><> ><> ><> ><> [[alternative HTML version deleted]] __ 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-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 model MANCOVA
Hello, I need to perform a mixed-model (with nesting) MANCOVA, using Type III sums of squares. I know how to perform each of these types of tests individually, but I am not sure if performing a mixed-model MANCOVA is possible. Please let me know. Erika <>< <>< <>< <>< <>< <>< <>< Erika Crispo, PhD candidate McGill University, Department of Biology http://www.biology.mcgill.ca/grad/erika/index.htm ><> ><> ><> ><> ><> ><> ><> [[alternative HTML version deleted]] __ 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 model MANCOVA
Hello, I need to perform a mixed-model (with nesting) MANCOVA, using Type III sums of squares. I know how to perform each of these types of tests individually, but I am not sure if performing a mixed-model MANCOVA is possible. Please let me know. Erika <>< <>< <>< <>< <>< <>< <>< Erika Crispo, PhD candidate McGill University, Department of Biology http://www.biology.mcgill.ca/grad/erika/index.htm ><> ><> ><> ><> ><> ><> ><> [[alternative HTML version deleted]] __ 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.
