Skip to content

random effects in a linear mixed model

4 messages · Elizabeth Green, Christian Parker, Dunbar, Michael J. +1 more

Original
Elizabeth Green
#
On Aug 19, 2011, at 2:00 AM, r-sig-ecology-request at r-project.org wrote:

            

      
      
      
    

        
Edgar, the random effects represent the variable that has an effect,  
but not an effect that is likely to be biologically interesting, in  
this case, your fishing vessel. The effects you want to test are the  
fixed effects, in this case time, because you want to see how your  
variable varies by year, accounting for the random effects of the  
different vessels fishing.

bessie

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Bessie Green
  907-388-4496
http://mercury.bio.uaf.edu/~elizabeth_green/

  
  


  


    

    

      
      

        
      

    

    
1 day later

    

      
      

        


  

        

      Dunbar, Michael J.
    

    

      
      #
    

  


  

      
Edgar

I agree with what Chris says

(1 + BOAT | fYEAR) makes no sense

It is probably 

(1|BOAT) + (1|fYEAR) that you want

Providing you have enough boats and years to estimate variance components for these two factors: please feel free to post any further queries but you will need to provide more information on the data if people are to give useful help.

Clearly boat is random, making year a factor and then random is a useful simplification to compare variances providing it makes sense to your research question.

Regards
Mike


-----Original Message-----
From: r-sig-ecology-bounces at r-project.org [mailto:r-sig-ecology-bounces at r-project.org] On Behalf Of Christian Parker
Sent: 20 August 2011 17:44
To: Elizabeth Green
Cc: r-sig-ecology at r-project.org
Subject: Re: [R-sig-eco] random effects in a linear mixed model

Edgar,

Based on the information provided its difficult to thoroughly answer your
question. But I will assume that you have samples within each year, say
weekly or monthly. Based on this I would say that you can treat the years as
a random effect but I dont think that I would do it the way that you have
specified

LMM2 <- lmer(Op_Obj ~ BOAT
+ (1 + BOAT | fYEAR), data=YM, family=gaussian, na.action=na.omit,
verbose=T)

Which tests for a difference in Op_Obj across boats and allowing that
difference to vary from year to year. Instead I think you are interested in
whether Op_Obj (either average or as a function of other predictors
measured) and you want to allow that relationship to vary by both boat and
year. So I would suggest that if you want to allow the relationship to vary
from year to year that you also include boat as a random variable. I believe
the syntax is like this:


LMM2 <- lmer(Op_Obj ~ 1 +(1|BOAT)+(1| fYEAR), data=YM)

Which is basically an anova design where you allow the intercept to vary by
boat and year. If you are interested in covariates you would include them in
place of the 1 on the "fixed" effects side of the model. And if you wanted
to allow their slopes to vary you would include them in place of the 1's on
the "random" effects side of the model.

There always seems to be a good discussion going in the mixed effects
discussion group as to the definition of fixed and random effects and how to
model them (e.g http://andrewgelman.com/2010/12/so-called_fixed/) but I am
curious as to what advice others on this list have.

Cheers,
Chris Parker

        
On Fri, Aug 19, 2011 at 11:05 AM, Elizabeth Green <bessie.bug at gmail.com>wrote:

        

            

      
      
      
    

        

      
      
    

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