Skip to content

Mixed model parameterization

8 messages · Thierry Onkelinx, Salvador Sánchez-Colón, Manuel Spínola +1 more

Original
Manuel Spínola
# 
Dear list members,

I am measuring an index in plots, in different hours, different days, and
different months in 3 different "habitats".

12 plots by "habitats".

I would like to estimate the mean index in the 3 environments.

Is this model appropriate to achieve my goal?

model <- lmer(index ~  habitat + (1 | plot/hour/day/month), data = mydata,
REML = FALSE)

Best,

Manuel
2 days later
Thierry Onkelinx
# 
Dear Manuel,

You'll need to think about the structure of your random effects. Your
current random effect structure is (1|plot) + (1|plot:hour) +
(1|plot:hour:day) + (1|plot:hour:day:month). Which might not be what you
had in mind.

Best regards,

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 za 18 mei 2019 om 01:56 schreef Manuel Sp?nola <mspinola10 at gmail.com>:

  
  


  


      

    

    

      
      

        


  

        

      Manuel Spínola
    

    

      
      #
    

  


  

      
Thank you very much Thierry.

I just only want to accomodate the nested repeated measure of my sampling.

I measure the index at the same plot, several times in an hour, several
days and during several month.

3 "habitats", within each habitat 12 plots, and on each plot I measured the
index several times, within hours, several days and several months.

I am not interested in the evolution of the index in time, just to account
for the repeated measure of my design (I measured the index hundreds of
time on each plot).

Maybe I don't need to worry about hour, day and month?

Manuel

El lun., 20 may. 2019 a las 2:00, Thierry Onkelinx (<
thierry.onkelinx at inbo.be>) escribi?:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Salvador Sánchez-Colón
    

    

      
      #
    

  


  

      
Hola Manuel:

If you are not interested in examining the between-hours or between-days variability, then you might consider analyzing the average of your hourly values and thus simplify the model. 

Best regards,

Salvador 

Salvador S?NCHEZ-COL?N

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Manuel Spínola
    

    

      
      #
    

  


  

      
Thank you Salvador,

So, doing this:

model <- lmer(index ~  habitat + (1 | plot), data = mydata, REML = FALSE)

Manuel

El lun., 20 may. 2019 a las 8:08, Salvador S?nchez-Col?n (<
salvadorsanchezcolon at prodigy.net.mx>) escribi?:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Thierry Onkelinx
    

    

      
      #
    

  


  

      
Aggregating the data will remove al lot of information. And might introduce
bias in case of an unbalanced design.

You need to think about the (combination of) variables which have a common
effect. (1|plot) + (1|plot:hour) assumes a plot effect and an effect of
each combination of plot and hour, but no common hour effect.

Days are nested in months. The corresponding random effect is (1|month) +
(1|month:day). You have (1|day) + (1|day:month) which is nonsense.

You really need to do some reading on mixed models. I recommend Zuur et al
(2009) Mixed Effects Models and Extensions in Ecology with R

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 ma 20 mei 2019 om 16:07 schreef Salvador S?nchez-Col?n <
salvadorsanchezcolon at prodigy.net.mx>:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

  
  


  


      

    

    
1 day later

    

      
      

        


  

        

      Manuel Spínola
    

    

      
      #
    

  


  

      
Thank you very much Thierry.

Manuel

El lun., 20 may. 2019 a las 8:59, Thierry Onkelinx (<
thierry.onkelinx at inbo.be>) escribi?:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Joaquín Aldabe
    

    

      
      #
    

  


  

      
I agree that using the mean you miss a lot of information. The simplest
case would be no interaction between random effects. Say,
(1|month)+(1|day)+(1|hour)+(1|plot). As far as I understand, if you think
that, for example, hour effect depends on the day, then you should consider
a nested structure of randoms effects.

Let us know your progress!

joaquin

<https://www.avast.com/sig-email?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=webmail>
Libre
de virus. www.avast.com
<https://www.avast.com/sig-email?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=webmail>
<#DAB4FAD8-2DD7-40BB-A1B8-4E2AA1F9FDF2>

El mar., 21 may. 2019 a las 14:01, Manuel Sp?nola (<mspinola10 at gmail.com>)
escribi?: