Skip to content

Question about random effects

4 messages · Adriana Maldonado Chaparro, Thierry Onkelinx, Ben Bolker +1 more

Original
Adriana Maldonado Chaparro
# 
Greetings,

I want to ask for advise on the following issue:
I fitted a mixed model where I'm trying to explain variation in Litter Sex
Ratio as a function of social network position. In this model the random
effect, individual identity, explained none of the variance, and one of the
reviewers argued that I should exclude it from my model because of these
reason. I think I should keep it because I have repeated measures. What are
your thoughts on this matter?

Thanks in advance,

Adriana Maldonado
Postdoctoral Researcher
Thierry Onkelinx
# 
If the random effect reflects the design of the study then it should remain
in the model.

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

2016-05-23 17:13 GMT+02:00 Adriana Maldonado Chaparro <
maldonado.aa at gmail.com>:

  
  


  


      

    

    

      
      

        


  

        

      Ben Bolker
    

    

      
      #
    

  


  

      
I agree, although I'll also say that if you are faced with a power
imbalance (reviewer/supervisor/etc. insists that it should be removed),
in the case where the random effect variance is estimated as zero there
is really very little (no?) *practical* difference in this case between
keeping or removing the random effect. In particular, the estimates of
any other variance components in the model, as well as all of the
contents of summary() [point estimates and Wald standard errors of
fixed-effect of coefficients] should be identical (try it and see).

  cheers
    Ben Bolker

        
On 16-05-23 11:42 AM, Thierry Onkelinx wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Reinhold Kliegl
    

    

      
      #
    

  


  

      
There is another kind of power issue involved as well:  Keeping spurious
variance components in the model leads to significant loss in statistical
power.

Stroup (2012, Generalized linear mixed models: Modern concepts, methods and
applications, p. 185):
"Neither the [maximal] nor the [minimal] linear mixed models are
appropriate for most repeated measures analysis. Using the [maximal] model
is generally wasteful and costly in terms of statistical power for testing
hypotheses. On the other hand, the [minimal] model fails to account for
nontrivial correlation among repeated measurements. This results in
inflated [T]ype I error rates when non-negligible correlation does in fact
exist. We can usually find middle ground, a covariance model that
adequately accounts for correlation but is more parsimonious than the
[maximal] model. Doing so allows us full control over [T]ype I error rates
without needlessly sacrificing power."

See also:  http://arxiv.org/abs/1511.01864

        
On Mon, May 23, 2016 at 5:57 PM, Ben Bolker <bbolker at gmail.com> wrote: