Thanks for your comments, David and Bert. The best would be to provide an example. Let's say we have a dataset like this one: IDEmployee Company OU CountViewPortal CountLogin TimeOnTask Performance 1 Company1 Company1.OU1 21 33 627.8 4.3 2 Company1 Company1.OU2 45 54 34.8 2.3 3 Company2 Company1.OU1 23 33 3.8 1.0 4 Company2 Company1.OU1 34 12 44.8 2.3 5 Company2 Company1.OU2 55 22 55.8 4.5 6 Company2 Company1.OU3 45 44 34.8 3
I want to see if there is correlation between CountViewPortal and Performance. Moreover, I'd like to reveal the influence of CountViewPortal+TimeOnTask on Performance. However, I expect that employees within a OU, and than a Company have similar behavior. Thus, I'll have 3 levels -> employee, OU, Company. In R, I would do something like this: randomInterceptCount <- lme(Performance ~ CountViewPortal, data=analysis, random=~1|OU/Company1, method="ML") But, then the point is that CountViewPortal, CountLogin and TimeOnTask are non-normally distributed. I guess that my question is, what should I do in case of non-normal distribution? I really appreciate your help. Thanks again! Srecko On Mon, Sep 30, 2013 at 5:14 PM, David Winsemius <dwinsem...@comcast.net>wrote: > > On Sep 30, 2013, at 3:22 PM, srecko joksimovic wrote: > > > I thought so, but then I found this: > > "Normality > > The assumption of normality states that the error terms at every level > of the model are normally distributed" > > maybe I misinterpreted something. > > Notice that it is the _error_terms_ that are to be normally distributed, > not the data itself. One might even infer that "normally distrited data > might be suspect because the "correct distribution should be a mixture of > normals. Since the errors never are going to fit on a straight line on a QQ > plot, the real question is "how far from Normal" and what the impact might > be on the quantities being estimated. > > -- > David. > > > > > > On Mon, Sep 30, 2013 at 3:06 PM, David Winsemius <dwinsem...@comcast.net> > wrote: > > > > On Sep 30, 2013, at 2:50 PM, srecko joksimovic wrote: > > > > > I have an example of multilevel analysis with 3 levels, but data are > > > non-normally distributed. In case of normal distribution, I would > perform > > > multilevel linear analysis using lme function, but what should I do in > case > > > of non-normal distribution? > > > > > > > But normal distribution is not a requirement for linear models. Please > review your theory. > > > > > thanks, > > > Srecko > > > > > > [[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. > > > > David Winsemius > > Alameda, CA, USA > > > > > > David Winsemius > Alameda, CA, USA > > [[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.
