Skip to content

nested mixed effects logistic regression binomial glm) results differ by function.

5 messages · David Duffy, Thierry Onkelinx, Linus Holtermann +1 more

Original
Lize van der Merwe
# 
Please advise:

I have a dichotomous outcome on 2500 individuals. From 18 geographical
areas, and many households nested within areas. I need to assess the
association between various predictors and my outcome, adjusting for the
correlation within households, as well as within areas. The following R
functions provide dramatically different results.

glmer(CC~predictor+1|area/household,family=binomial)

and

glmmPQL(CC~predictor, random=~1|area/household),family=binomial)

Why? Which is correct?

Thanks in advance.  (I posted this on another site too.)

Lize
David Duffy
#
On Fri, 24 Apr 2015, Lize van der Merwe wrote:

            

      
      
      
    

        
PQL is known to be biased, the amount depending on a few things including 
the proportion CC in the sample, and number of levels for the REs. You 
could try hglm (package hglm, using EQL) and see how different the results 
are from that ;)  It is also possible one or both programs encountered 
numerical problems because of features of your data. If you can send your 
original data, or simulated data of the same structure (that gives a 
similar problem!), we could have a look.

Cheers, David.

| David Duffy (MBBS PhD)
| email: David.Duffy at qimrberghofer.edu.au  ph: INT+61+7+3362-0217 fax: -0101
| Genetic Epidemiology, QIMR Berghofer Institute of Medical Research
| 300 Herston Rd, Brisbane, Queensland 4006, Australia  GPG 4D0B994A

  
  


  


      

    

    

      
      

        


  

        

      Thierry Onkelinx
    

    

      
      #
    

  


  

      
Dear Lize,

glmmPQL() uses Penalized Quasi-Likelihood and glmer() uses the likelihood
in case of a binomial family. I prefer methods that uses the likelihood.

Best regards,

ir. Thierry Onkelinx
Instituut voor natuur- en bosonderzoek / Research Institute for Nature and
Forest
team Biometrie & Kwaliteitszorg / team Biometrics & Quality Assurance
Kliniekstraat 25
1070 Anderlecht
Belgium

To call in the statistician after the experiment is done may be no more
than asking him to perform a post-mortem examination: he may be able to say
what the experiment died of. ~ Sir Ronald Aylmer Fisher
The plural of anecdote is not data. ~ Roger Brinner
The combination of some data and an aching desire for an answer does not
ensure that a reasonable answer can be extracted from a given body of data.
~ John Tukey

2015-04-23 18:16 GMT+02:00 Lize van der Merwe <lizestats at gmail.com>:

            

      
      
      
    

        

      
      
    

        

  
  


  


      

    

    

      
      

        


  

        

      Linus Holtermann
    

    

      
      #
    

  


  

      
Dear Lize,

maybe you give Bayesian methods a try. The excellent MCMCglmm package should be able to handle your model. Often MCMC provides more reliable results when a wide range of variation in group size and relative small number of observations per group are present in the data.

Best regards,

Linus Holtermann
Hamburgisches WeltWirtschaftsInstitut gemeinn?tzige GmbH (HWWI)
Heimhuder Stra?e 71
20148 Hamburg
Tel +49-(0)40-340576-336
Fax+49-(0)40-340576-776
Internet: www.hwwi.org
Email: holtermann at hwwi.org
?
Amtsgericht Hamburg HRB 94303
Gesch?ftsf?hrer: PD Dr. Christian Growitsch | Prof. Dr. Henning V?pel
Prokura: Dipl. Kauffrau Alexis Malchin
Umsatzsteuer-ID: DE 241849425

-----Urspr?ngliche Nachricht-----
Von: R-sig-mixed-models [mailto:r-sig-mixed-models-bounces at r-project.org] Im Auftrag von Thierry Onkelinx
Gesendet: Freitag, 24. April 2015 09:43
An: Lize van der Merwe
Cc: r-sig-mixed-models at r-project.org
Betreff: Re: [R-sig-ME] nested mixed effects logistic regression binomial glm) results differ by function.

Dear Lize,

glmmPQL() uses Penalized Quasi-Likelihood and glmer() uses the likelihood in case of a binomial family. I prefer methods that uses the likelihood.

Best regards,

ir. Thierry Onkelinx
Instituut voor natuur- en bosonderzoek / Research Institute for Nature and Forest team Biometrie & Kwaliteitszorg / team Biometrics & Quality Assurance Kliniekstraat 25
1070 Anderlecht
Belgium

To call in the statistician after the experiment is done may be no more than asking him to perform a post-mortem examination: he may be able to say what the experiment died of. ~ Sir Ronald Aylmer Fisher The plural of anecdote is not data. ~ Roger Brinner The combination of some data and an aching desire for an answer does not ensure that a reasonable answer can be extracted from a given body of data.
~ John Tukey

2015-04-23 18:16 GMT+02:00 Lize van der Merwe <lizestats at gmail.com>:

            

      
      
      
    

        

      
      
    

        
_______________________________________________
R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

  
  


  


      

    

    

      
      

        


  

        

      Lize van der Merwe
    

    

      
      #
    

  


  

      
Thank you so very much to everyone responding to my request.  I learnt a
lot.  You helped me figure out my mistake.  I wanted to adjust for the
correlation inside households.  Most of the households, however, contained a
single individual.  When I combined them into a single cluster, the answers
were exactly what I needed.
Regards
Lize


-----Original Message-----
From: David Duffy [mailto:David.Duffy at qimr.edu.au] 
Sent: 24 April 2015 05:12
To: Lize van der Merwe
Cc: r-sig-mixed-models at r-project.org
Subject: Re: [R-sig-ME] nested mixed effects logistic regression binomial
glm)results differ by function.

        
On Fri, 24 Apr 2015, Lize van der Merwe wrote:

        

            

      
      
      
    

        
PQL is known to be biased, the amount depending on a few things including
the proportion CC in the sample, and number of levels for the REs. You could
try hglm (package hglm, using EQL) and see how different the results are
from that ;)  It is also possible one or both programs encountered numerical
problems because of features of your data. If you can send your original
data, or simulated data of the same structure (that gives a similar
problem!), we could have a look.

Cheers, David.

| David Duffy (MBBS PhD)
| email: David.Duffy at qimrberghofer.edu.au  ph: INT+61+7+3362-0217 fax: 
| -0101 Genetic Epidemiology, QIMR Berghofer Institute of Medical 
| Research
| 300 Herston Rd, Brisbane, Queensland 4006, Australia  GPG 4D0B994A