Re: [R] NLME: Problem with plotting ranef vs a factor

From: Spencer Graves <spencer.graves_at_pdf.com>
Date: Wed 09 Aug 2006 - 15:31:49 EST

Your question is entirely too complex for me to try to answer in a reasonable amount of time, especially since your example in not self contained.

If you would still like help on this, I suggest you try to generate a self contained example that is as simple as you can make it that illustrates your problem, as suggested in the posting guide "www.R-project.org/posting-guide.html". With only a modest amount of luck, the things you try to simplify your example will lead to enlightenment. If that fails, please submit another question that is self contained, simple and clear. Doing so should substantially increase your chances of getting a quick, useful reply.

I know this doesn't answer your question, but I hope it helps.

Spencer Graves

Greg Distiller wrote:
> Hi
> I am following the model building strategy that is outlined in the Pinheiro and Bates book wrt including covariates but am having a problem with the plot. Basically I am using 4 covariates (1 of them is continuous) and 3 of them are fine but the 4th one is being shown as a scatterplot despite the fact that it is a factor. I have explicitly declared this to be a factor (pcat<-as.factor(pcat)) and have also checked by using the "is.factor" and the "levels" command that it is a factor. Yet despite this the plot command is not recognising it as a factor.
>
>
> I am reading in the data by:
>
> attach(Data)
> Data1_93<-transform(Data,log2game=log2(gamedens+1))
> pcat<-as.factor(pcat)
> Data1_93<-groupedData(log2game ~ day | subjectno, data=Data1_93)
> detach(Data)
>
> Here is the code to check that the covariate called pcat is indeed a factor:

```>> levels(pcat)
```

> [1] "1" "2" "3"
>
```>> is.factor(pcat)
```

> [1] TRUE
>
> and then after the model is fitted I extract the random effects:
>
> D1C2.ran <- ranef(mod11.103nlme,augFrame=T)
>
> and here is an extract from the object:
>
> C R day gamedens pcat site mutcat1 pdens0 log2game
> NA02_259 -1.016987007 0.0162825099 15.75000 23.50000 1 Namaacha Mixed 15018 3.761099
> NA02_073 -0.939355374 0.0132589702 10.50000 23.75000 1 Namaacha Resistant 6170 3.675543
> M00_12 -0.775048474 0.0047124742 10.50000 25.00000 1 Mpumulanga Sensitive 17525 3.768326
> M00_93 -0.555801118 0.0053872868 14.00000 37.50000 2 Mpumulanga Sensitive 332000 4.254319
> NA02_053 -0.327990343 -0.0037659864 6.00000 39.25000 1 Namaacha Resistant 65529 4.292481
>
> Note that this output also seems to indicate that pcat is a factor as it is summarised correctly.
>
> I then generate plots for my random effects:
>
> plot(D1C2.ran,form= C ~site+mutcat2+pcat+pdens0)
>
> and the problem is that the panel for my random effects vs pcat is displayed as a scatterplot rather than as a boxplot.
> I am getting told to check warnings and these warnings look like:
>
> Warning messages:
> 1: at 0.99
> 3: all data on boundary of neighborhood. make span bigger
> 4: pseudoinverse used at 0.99
> 6: reciprocal condition number -1.#IND
> 7: zero-width neighborhood. make span bigger
>
> I do not get these warnings if I exclude the problematic variable pcat so must be something to do with this. Any ideas?
>
> Many thanks
>
> Greg
>
> [[alternative HTML version deleted]]
>
> ______________________________________________
> R-help@stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help