[R] glm - change offset to avoid NA?
dwinsemius at comcast.net
Tue Jul 16 18:20:45 CEST 2013
On Jul 16, 2013, at 8:52 AM, Hermann Norpois wrote:
> I did not think of something like "try". I thouht that there should always
> be a value if I do a logistic regression but sometimes the values are far
> from being meaningful. So there is a cut-off.
That seems implausilbe. I think you are just seeing the effect of a random selection creating a data situation where the interaction term is aliased with the others, in which case `glm` will set the coefficient to NA.
> My plan was to change the
You said there was an "error", but you should have said that the results were unexpected. You can create an "error-condition" when this occurs with the `stop` function. And then do whatever remedial work is needed.
> 2013/7/15 Bert Gunter <gunter.berton at gene.com>
>> I think what you want is
>> ?try ##or
>> ## The second is more flexible but slightly more complicated.
>> to trap the error and perhaps refit the model without interaction?
>> On Mon, Jul 15, 2013 at 10:45 AM, Hermann Norpois <hnorpois at gmail.com>
>>> I use glm within a function testing for the appearence of the coexistence
>>> of (minor allels in a subset of) snps. And then I extract the
>>> Pr(>|z|)-value for the interaction. Principally it works but sometimes
>>> function stops because this "value for the interaction" is NA. For
>>> instance, this is the case in the following example:
>>> lz <- glm(trait~rs7572685*rs10520302, data=mus, family=binomial)
>>>> summary (lz)
>>> Coefficients: (1 not defined because of singularities)
>>> Estimate Std. Error z value Pr(>|z|)
>>> (Intercept) 0.05614 0.16782 0.335 0.738
>>> rs7572685 0.49041 0.41437 1.183 0.237
>>> rs10520302 0.49269 0.43514 1.132 0.258
>>> rs7572685:rs10520302 NA NA NA NA
>>> I would prefer some values instead of NA (though it does not make any
>>> in terms of interpretation) for the sake of the smooth running of my
>>> function. How is this done? I guess I have to change the offset but I
>>> understand how.
>>> [[alternative HTML version deleted]]
>>> R-help at r-project.org mailing list
>>> PLEASE do read the posting guide
>>> and provide commented, minimal, self-contained, reproducible code.
>> Bert Gunter
>> Genentech Nonclinical Biostatistics
>> Internal Contact Info:
>> Phone: 467-7374
> [[alternative HTML version deleted]]
> R-help at r-project.org mailing list
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
Alameda, CA, USA
More information about the R-help