Re: [Rd] standard errors for glm

From: christophe dutang <dutangc_at_gmail.com>
Date: Thu, 28 Aug 2008 09:58:34 +0200

Hi,

Thanks for answering so quickly.

Actually, I do not have the John Fox's book. On this webpage, there is a handout for logistic regression (GLM I'm interested in) (cf. http://socserv.socsci.mcmaster.ca/jfox/Courses/SPIDA/logistic-regression-handout.pdf).

The summary function return standard errors for coefficient estimates but not for predictions. According to
http://www.nag.co.uk/numeric/cl/Manual/pdf/G02/g02gbc.pdf<http://www.nag.co.uk/numeric/cl/Manual/pdf/G02/g02gbc.pdfprovides> the standard errors for the linear predictors X \hat{\beta} is X C X^t where C is the variance/covariance matrix of coefficient estimate \hat{\beta} . But I do not know standard errors for a response.

Christophe

2008/8/27 <markleeds_at_verizon.net>

> Hi: you can check John Fox's CAR book if you have it. I don't remember for
> sure but I may have standard error related calculations for some of
> his graphics in the GLM section of the book ? But, Can you not get the
> sigma^2 hat the summary of a GLM ? i would do
>
> glmsum<-summary("yourglmmodel") and check there.
>
> I'd be really surprised if the variance of the error term wasn't there.
> Then, if you have that I think you can use that to calculate the prediction
> standard error as long as you assume that the parameters of the model are
> known with certainty. The formula for that is in the regular regression (
> non GLM ) textbooks but i don't remember it off the top of my head. Good
> luck and hopefully someone else will reply with more exact info.
>
>
>
>
>
>
> On Wed, Aug 27, 2008 at 1:25 PM, christophe dutang wrote:
>
> Hi,
>>
>> I'm currently using biglm package to compute GLM outputs on a very large
>> dataset. But no function computes standard erros of predictions. I look in
>> what is done in R, namely in the function predict.glm.R in stats package.
>> In this function, we call predict.lm to compute the standard errors (line
>> 51). The code of predict.lm (in lm.R) is very hard to understand.
>>
>> I wonder if there is any good reference and / documentation on this topic?
>> the manual at
>> http://www.nag.co.uk/numeric/cl/Manual/pdf/G02/g02gbc.pdfprovides
>> a good overview of the method used in R, but there is no reference
>> to standard errors...
>>
>> I suppose this topic have already raised in the past, but I found only
>> this
>> http://tolstoy.newcastle.edu.au/R/help/04/08/1762.html
>>
>> Thanks in advance
>>
>> Christophe
>>
>> [[alternative HTML version deleted]]
>>
>> ______________________________________________
>> R-devel_at_r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-devel
>>
>

        [[alternative HTML version deleted]]



R-devel_at_r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel Received on Thu 28 Aug 2008 - 17:23:01 GMT

Archive maintained by Robert King, hosted by the discipline of statistics at the University of Newcastle, Australia.
Archive generated by hypermail 2.2.0, at Thu 28 Aug 2008 - 21:37:51 GMT.

Mailing list information is available at https://stat.ethz.ch/mailman/listinfo/r-devel. Please read the posting guide before posting to the list.

list of date sections of archive