> On Mar 28, 2017, at 9:32 AM, Leon Lee <bhamlio...@gmail.com> wrote: > > Hi, R experts > > I am new to R & GAM toolbox and would like to get inputs from you all on my > models. The question I have is as follows: > I have 30 subjects with each subject being scanned from one to three times > in the first year of life. The brain volume from each scan was measured. > The scan time was randomly distributed from birth to 1 year. > Each subject has different gestational age ranging from 38 to 41 weeks > Each subject has chronological age from birth to 1 year old > Each subject has gender category. > Now, I want to look at how predictors, such as subject's chronological age, > gestational age and gender will explain the changes in brain volume. I also > want to include interactions between gender and age, gestational and > chronological age. Random effects are also included in the model to account > for subject variability. My model looks like the follows: > > gam=gam(brainVolume~ s(age) + ti(age, gestationalAge) + gestationalAge + > sex + s(age, by=sex) + s(subjIndexF, bs="re"), method="REML", data=mydata) > > Are there any obvious mistakes in the model? Any suggestions will be > greatly appreciated!
I'm not seeing mistakes in the syntax but I would question whether 30 subjects is sufficient to adequately support estimates in a a model of this complexity. I would also think that the 's(age)' and 'sex' terms would get aliased out in a model with "+ s(age, by=sex)". Most R regression functions handle removal of over-parametrization automatically. You also have a variable number of measurements per subject. I am unable to comment on the effort to account for the implicit and variably measured correlation and auto-correlation of values within subjects using a "smooth" on subjIndexF, since that is not an approach I was familiar with. But I am getting concerned whether you are also new to statistical modeling in addition to your use of R and GAM being "new to you"? (Perhaps Simon or one of the mixed-effects experts can correct the gaps in my understanding of how to model repeated measures in the context of small numbers of subjects and irregular emasurements.) Please read the Posting Guide and the pages of candidate mailing lists. Rhelp is not really the place to go when you need statistical advice. I'm not sure if this is really in the center of concerns that get discussed on the Mixed Models list, but to my eyes it would be a better fit there. -- David. > > L > > [[alternative HTML version deleted]] > > ______________________________________________ > R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see > 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. David Winsemius Alameda, CA, USA ______________________________________________ R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see 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.
