[R-sig-ME] GLMM linearity checking

Shujuan Feng fengsj at mail.utexas.edu
Thu Jun 3 21:40:27 CEST 2010

Thanks Ben, it works.

Then I feel there may be problems with the model by looking at the residul 

When I just plot the residuals, I get two big group values (the plots of 
residual~predicotr are also like this). See the attach file: residaul.jpeg.

I have difficulty in imaging the residuals in GLM with the original 
dependent are 0s or 1s. Residuals should be the difference between the 
observed and the predicted.  I can understand the predicted in terms of 
transformed scale(logit), but I don't know how the observed 0s and 1s are 
transformed. Are the two group residual values from the 0s and 1s 
respectively?  Have anyone see such residuals?


PS: In my model, I have three continuous predictors, the dependent are 0s or 
1s and I use binomial(link = "logit"). I tried GLMM and also just GLM, I got 
similar residual plots.

----- Original Message ----- 
From: "Ben Bolker" <bolker at ufl.edu>
To: "Shujuan Feng" <fengsj at mail.utexas.edu>
Cc: "R Mixed Models" <r-sig-mixed-models at r-project.org>
Sent: Thursday, June 03, 2010 1:37 PM
Subject: Re: [R-sig-ME] GLMM linearity checking

>  [cc'ing back to r-sig-mixed]
>  That was going to be my suggestion.
>  Try omitting rows of the data set with NA predictors or responses
> (na.omit() will work if your data frame does not have *other* columns
> with NAs in them beyond those used in the model) before you start.
> Shujuan Feng wrote:
>> or maybe I can just delete the missing rows.
>> ----- Original Message ----- 
>> From: "Shujuan Feng" <fengsj at mail.utexas.edu>
>> To: "Ben Bolker" <bolker at ufl.edu>
>> Cc: <r-sig-mixed-models at r-project.org>
>> Sent: Thursday, June 03, 2010 12:10 PM
>> Subject: Re: [R-sig-ME] GLMM linearity checking
>>> Thanks so much!
>>> I read about Graphical checking for GLMM (transformed by the link
>>> Function) before fitting the model from a paper. I have difficulty in
>>> imaging how the 0s and 1s are transformed by the ink. .....
>>> I tried your suggestions and this way should give me more valuable
>>> checkings for the model. But because I have a lot of missing data, I 
>>> could
>>> not put the residuals into the data. I got errors:
>>> Error in `$<-.data.frame`(`*tmp*`, "resid", value = c(-0.776415  :
>>> replacement has 13580 rows, data has 68158
>>> Is there any way to match residuals and the predictor?
>>> I tried just plot(model), but it doesn't work for GLMM.
>>> Thanks!!
>>> ----- Original Message ----- 
>>> From: "Ben Bolker" <bolker at ufl.edu>
>>> To: <fengsj at mail.utexas.edu>
>>> Cc: <r-sig-mixed-models at r-project.org>
>>> Sent: Thursday, June 03, 2010 11:25 AM
>>> Subject: Re: [R-sig-ME] GLMM linearity checking
>>>> fengsj at mail.utexas.edu wrote:
>>>>> I am sorry for asking this question here. It is more related to
>>>>> logistic regression.
>>>>> I need to use GLMM (binomial(link = "logit")) )to fit my data. The
>>>>> dependent variable is 0 or 1 and  I'd like to do some roughly
>>>>> graphical checkings for my data to see if the responses of
>>>>> transformaed data are linear with respect to continuous predictors in
>>>>> general. How should I do this?
>>>>> Thanks,
>>>>   how about
>>>> m <- glmer(...,data=d)
>>>> d$resid <- residuals(m)
>>>> xyplot(resid~continuous_predictor_1,type=c("p","smooth"),data=d)
>>>> ...
>>>>  Non-linearity on the transformed scale will appear as a (non-flat)
>>>> pattern of the (smoothed line fitted to the) residuals as a function of
>>>> the continuous predictors ...
>>>>  Ben Bolker
>>> _______________________________________________
>>> R-sig-mixed-models at r-project.org mailing list
>>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
> -- 
> Ben Bolker
> Associate professor, Biology Dep't, Univ. of Florida
> ***   bbolker at gmail.com , bolker at math.mcmaster.ca
> bolker at ufl.edu / people.biology.ufl.edu/bolker
> GPG key: people.biology.ufl.edu/bolker/benbolker-publickey.asc 

More information about the R-sig-mixed-models mailing list