Skip to content

sample size in glmer model

3 messages · Souheyla GHEBGHOUB, Thierry Onkelinx, Ben Bolker

Original
Souheyla GHEBGHOUB
# 
Hi,
In Field (2012) we need 10-15 *participants* per variable for regression,
in Levshina (2015) we need 10-15 *observations* per variable.

Is it participant or observation? I am so confused as I have 53
participants and 1484 observations.

Thank you,
Sou
PhD in Education
University of York
Thierry Onkelinx
# 
Dear Sou,

I'd suggest at least 10-15 observation per **parameter**. Categorical
variables with more than two levels require more than one parameter. An
interaction between two continuous variable requires 3 parameters (2 main
effect + 1 interaction). Don't forget to count the hyperparameters of the
random effects.

Best regards,

Thierry

ir. Thierry Onkelinx
Statisticus / Statistician

Vlaamse Overheid / Government of Flanders
INSTITUUT VOOR NATUUR- EN BOSONDERZOEK / RESEARCH INSTITUTE FOR NATURE AND
FOREST
Team Biometrie & Kwaliteitszorg / Team Biometrics & Quality Assurance
thierry.onkelinx at inbo.be
Havenlaan 88 bus 73, 1000 Brussel
www.inbo.be

///////////////////////////////////////////////////////////////////////////////////////////
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
///////////////////////////////////////////////////////////////////////////////////////////

<https://www.inbo.be>


Op wo 22 jan. 2020 om 14:55 schreef Souheyla GHEBGHOUB <
souheyla.ghebghoub at gmail.com>:

  
  


  


      

    

    

      
      

        


  

        

      Ben Bolker
    

    

      
      #
    

  


  

      
It's going to depend on whether the parameters you're trying to
estimate are capturing processes that vary at the level of
participants or at the level of observations.  For example, if you
wanted to compare trends over time among groups of participants, each
measured multiple times (i.e. a random-slopes model with a
fixed-effect interaction between treatment and slope) you'd need to
consider the number of participants. (In a sense the answer to this
question goes back to the classical experimental design bestiaries of
nested vs randomized block vs split-plot vs ... and deciding what the
"denominator degrees of freedom" are supposed to be in each case/for
each test of interest).

On Wed, Jan 22, 2020 at 9:45 AM Thierry Onkelinx via

        
R-sig-mixed-models <r-sig-mixed-models at r-project.org> wrote: