Skip to content

Mixed model vs GEE

7 messages · Tahsin Ferdous, Ben Bolker, Milani Chaloupka +3 more

Original
Tahsin Ferdous
# 
Hi all,


I am analyzing repeated measures data. Both the mixed model and generalized
estimating equation are appropriate for my data. In this case, how can I
decide that which one is better (LMM or GEE)? I know that GEE is a *marginal
model*. It seeks to model a population average. Mixed-effect/Multilevel
models are *subject-specific*, or *conditional*, models. Thanks.


Best,

Tahsin
Ben Bolker
# 
I think that depends on what kind of questions you are asking ... ?? 
(If anyone wants to point to a great resource on marginal vs conditional 
models and when each type is appropriate, that would be great.  I know 
this distinction is discussed in Agresti's _Categorical Data Analysis_ 
book but I don't know if it goes into detail / gives examples about when 
one would want either one ...)
On 11/4/21 8:22 PM, Tahsin Ferdous wrote:
Milani Chaloupka
# 
One useful reference is :

Muff S, Held L, Keller L (2016) Marginal or conditional regression models for correlated non-normal data? Methods in Ecology and Evolution 7: 1514?1524. doi: 10.1111/2041-210X.12623

Milani
Tahsin Ferdous
# 
Dear all,

Thanks a lot.

Best,

Tahsin

On Thu, Nov 4, 2021 at 10:02 PM Milani Chaloupka <m.chaloupka at uq.edu.au>
wrote:

  
  


  


      

    

    

      
      

        


  

        

      Paul Johnson
    

    

      
      #
    

  


  

      
This might be useful, although it's focussed on non-normal GLMMs, not LMMs. I read it a while ago, but I remember it being excellent:

Marginal or conditional regression models for correlated non-normal data?
Stefanie Muff, Leonhard Held, Lukas F. Keller
https://doi.org/10.1111/2041-210X.12623

"For normally distributed response variables, that is in linear regression, the choice between a marginal and a conditional formulation is not particularly delicate, because the interpretation of conditional and marginal linear regression models turns out to be equivalent. On the other hand, the choice is relevant for non-normal data, as the interpretation of conditional and marginal regression models is usually different."

        
On 05/11/2021, 00:38, "R-sig-mixed-models on behalf of Ben Bolker" <r-sig-mixed-models-bounces at r-project.org on behalf of bbolker at gmail.com> wrote:

        
I think that depends on what kind of questions you are asking ... ?? 
    (If anyone wants to point to a great resource on marginal vs conditional 
    models and when each type is appropriate, that would be great.  I know 
    this distinction is discussed in Agresti's _Categorical Data Analysis_ 
    book but I don't know if it goes into detail / gives examples about when 
    one would want either one ...)

        
On 11/4/21 8:22 PM, Tahsin Ferdous wrote:

        
> Hi all,
    > 
    > 
    > I am analyzing repeated measures data. Both the mixed model and generalized
    > estimating equation are appropriate for my data. In this case, how can I
    > decide that which one is better (LMM or GEE)? I know that GEE is a *marginal
    > model*. It seeks to model a population average. Mixed-effect/Multilevel
    > models are *subject-specific*, or *conditional*, models. Thanks.
    > 
    > 
    > Best,
    > 
    > Tahsin
    > 
    > 	[[alternative HTML version deleted]]
    > 
    > _______________________________________________
    > R-sig-mixed-models at r-project.org mailing list
    > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
    >

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

  
  


  


      

    

    

      
      

        


  

        

      Phillip Alday
    

    

      
      #
    

  


  

      
Dimitris Rizopoulos covers this in his course slides:

http://www.drizopoulos.com/courses/EMC/CE08.pdf

The slides might be a bit math heavy for end users, but big important
assumptions and intuitions are called out in clear language.

        
On 4/11/21 7:38 pm, Ben Bolker wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Muldoon, Ariel
    

    

      
      #
    

  


  

      
I remember enjoying the discussion of marginal vs conditional inference in Stroup's book Generalized Linear Mixed Models: Modern Concepts, Methods and Applications . It is not in front of me at the moment but I think this is covered chapter 3. He specifically discusses how the issue affects non-normal GLMM's and I believe he first shows an example of the non-issue in (normal) LMM vs GEE and then goes on to a binomial example. There is also a brief overview of when/why we might want conditional vs marginal inference (he seems to come down hard on the side of conditional).