Dear Dimitris, Thanks for your prompt reply.
I agree that the specification of the correlation structure in theory only affects the efficiency. But I noticed that the fixed effects change dramatically when comparing the unstructured correlation with my other attempts. With most models the effect of Year is about -0.12. But with the unstructured correlation structure it switches sign. So the effect of Year becomes 0.10. In all models the effect of Year is strongly significant. Therefore I think I will stick to the model with PlotID as id and an exchangeable correlation structure. Regards, 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 thierry.onkel...@inbo.be 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: Dimitris Rizopoulos [mailto:d.rizopou...@erasmusmc.nl] Verzonden: woensdag 22 april 2009 11:48 Aan: ONKELINX, Thierry CC: r-help@r-project.org Onderwerp: Re: [R] Gee with nested desgin Hi Thierry, in geeglm() and in connection with the GEE theory, argument "id" should identify the independent sample units in your data set, which seem to be the plots. GEE will produce consistent estimates even if you misspecify the correlation structure in your data (i.e., ignore nested design). If you specify it correctly, you gain in efficiency. So, in your case you may gain a little (or a lot) if you specify your own nested working correlation matrix. In geeglm(), argument 'corstr' gives the option for a user-defined correlation matrix, but I don't know how exactly this needs to be defined. Regarding the 'waves' argument and AFAIK, this used in connection with the AR1 working correlation structure in order to identify the order and spacing of the observations (i.e., if subject 1 has visits at time 1 and time 2, whereas subject 2 has visits at time 1 and time 3). A final thing: you said that you have some missing values. In this case note that the GEE approach will be valid (i.e., produce unbiased estimates) if the missing data mechanism is Missing Completely At Random, that is when the reasons for missingness do not depend on Y. I hope it helps. Best, Dimitris ONKELINX, Thierry wrote: > Dear all, > > Is it possible to incorporate a nested design in GEE? I have > measurements on trees that where measured in two years. The trees are > nested in plots. Each plot contains 24 trees. The number of plots is 72. > Hence we would expect 2 * 24 * 72 = 3456 data points. A few are missing, > so we end up wih 3431 data points. > > This is what I have tried until now. > > #assuming independence between trees and thus ignoring the plot level. > library(geepack) > geeglm(formula = Y ~ Year, id = TreeID, family = binomial, corstr = > "exchangeable") > > #using waves. But I'm wondering if this is correct. > library(geepack) > geeglm(formula = Y ~ Year, id = PlotID, waves = TreeID, family = > binomial, corstr = "exchangeable") > > #using a unstructured correlation on the plot level. geeglm with > unstructured correlation resulted in an out of memory error. > library(Zelig) > zelig(formula = Y ~ Year, model = "logit.gee", id = "PlotID", corstr = > "unstructured") > > #Ideally I think I need a correlation matrix structured like below > (given for a plot with 3 trees). Here a1 is the correlation within a > tree, a2 the correlation between trees from the same plot and the same > year and a3 the correlation between trees from the same plot but a > different year. Does it make sense to run the model with a unstructured > correlation, calculate the average a1, a2 and a3 and use that as a fixed > working correlation? > > matrix(c( > 1, "a1", "a2", "a3", "a2", "a3", > "a1", 1, "a3", "a2", "a3", "a2", > "a2", "a3", 1, "a1", "a2", "a3", > "a3", "a2", "a1", 1, "a3", "a2", > "a2", "a3", "a2", "a3", 1, "a1", > "a3", "a2", "a3", "a2", "a1", 1 > ), nrow = 6) > > Best regards, > > 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 > thierry.onkel...@inbo.be > 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 > > 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. > -- Dimitris Rizopoulos Assistant Professor Department of Biostatistics Erasmus University Medical Center Address: PO Box 2040, 3000 CA Rotterdam, the Netherlands Tel: +31/(0)10/7043478 Fax: +31/(0)10/7043014 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.
