[R] Help with logistic model with random effects in R

[Ben Bolker]

LAURA WEIR <0611792W <at> student.gla.ac.uk> writes:

> 
> Hello everyone, 
> 
> I'm quite new to R and am trying to run a logistic model 
> to look at how various measures of boldness in
> individual animals influences probability of capture,
>  however I also want to include random effects and
> I'm not sure how to construct a model that incorporates both
> of these things. 
> 
> Data was collected from 6 different groups of 6 individuals
> with 10 replicates for each group. Whether an
> individual was caught in a trial or not was called "Catchability"
> and the outcome is binomial (caught = 1,
> not caught = 0). "Catchability" is my response variable that
>  I want to see if I can explain by the other
> variables. "Individual" and "Group" would be my random effects.
> I have already installed lme4 but not
> sure how to code for the model I need, can anyone offer any help please?
> 
> The column headings in my data table are as follows.
> 
>  [1] "Trial"          "Group"          "Individual"     "Mark"          
>  [5] "Catchability"   "Mboldness1"     "Mboldness2"     "Nboldness1"    
>  [9] "Nboldness2"     "Standardlength"
> 
> This is the string of my data:
> 
> str(data)
> 'data.frame':   36 obs. of  10 variables:
>  $ Trial         : int  1 2 3 4 5 6 7 8 9 10 ...
>  $ Group         : int  1 1 1 1 1 1 2 2 2 2 ...
>  $ Individual    : int  1 2 3 4 5 6 7 8 9 10 ...
>  $ Mark          : Factor w/ 8 levels " - O"," - P",..:
>      8 2 4 6 3 1 6 3 1 5 ...
>  $ Catchability  : int  1 1 1 1 0 1 1 1 1 1 ...
>  $ Mboldness1    : int  7 10 0 11 15 5 15 4 15 1 ...
>  $ Mboldness2    : int  1 270 600 1 10 203 10 230 1 580 ...
>  $ Nboldness1    : int  0 0 0 0 0 0 1 0 2 0 ...
>  $ Nboldness2    : int  270 110 50 50 90 70 130 90 260 220 ...
>  $ Standardlength: num  40.5 37.4 38.6 41.1 39.1 40.5 50.2 60.3 53.9 55 ...
> 

  I'm a little confused about your experimental design.  You say you
have 10 replicates per group, but I only see 36 observations in your
data set -- from your description I would have expected 360?

If you had 10 replicates per group I would say should do something like

glmer(Catchability~[fixed effect predictors]+(1|Group/Individual),
   family=binomial,data=data)

  A couple of other points:
* it's considered bad practice to name your data set 'data' (which is
also the name of a built-in R function).  Usually it's OK, but occasionally
it could cause problems.
* A rule of thumb is that you should not try to fit a model with
more than k=N/10 parameters, where for binary data 'N' is the minimum
of (number of successes, number of failures).
* 6 groups is at the lower edge of feasibility for fitting a random
effect -- you may find that glmer estimates the group-level variance
as zero. (Don't panic.)

  I would suggest that you send follow-ups to r-sig-mixed-models at r-project.org
and possibly take a look at <http://glmm.wikidot.com/faq>

  cheers
    Ben Bolker