st: How to specify random factor in xtmelogit (LogReg w/ random factor)

[Susan Lingle <susan.lingle@uleth.ca>]

Dear Stata-listers

Apologies in advance if my question is too rudimentary. I am new to 
Stata and have been figuring out how to deal with categorical factors as 
independent variables the last couple of days (by turning them into 
indicative variables). I am now wondering whether I need to do anything 
special when including a random factor in an analysis.

I am looking at the effect of three fixed factors including species (2 
species), year (7 years), and area (4 areas) on the survival of deer 
fawns. I have 147 deer fawns from 127 mothers and want to use the 
mother's identity as random factor to control for family effects (the 
survival of twin fawns is unlikely to be independent).

Do I simply 'tell' Stata to treat mother identity as a random factor (a 
grouping factor)? Or do I need to somewhere specify that this is a 
grouping or categorical factor. Clearly I do not want to create 126 
indicative new variables from the mother's identity, the process one 
would follow when using categorical fixed factors!

I will paste in the commands I used and also the output. I am only 
interested in the effect of the fixed factors. I only include mother's 
identity to control for that variable (not to assess its effects).

I was surprised to see that the effect of species is far weaker in this 
analysis than it was when I ran a logistic regression that excluded the 
20 twins (i.e. only using 1 fawn per mother), for which I did not need 
to include a random factor.

Thanks so much for any insight you can provide.

Susan


xi: xtmelogit WinterSurv speciesno i.year i.winterareano, || 
motheridno:, covariance(independent)

i.year            _Iyear_1994-2005    (naturally coded; _Iyear_1994 omitted)
i.winterareano    _Iwinterare_1-4     (naturally coded; _Iwinterare_1 
omitted)
Note: single-variable random-effects specification; covariance structure 
set to identity

Refining starting values:

Iteration 0:   log likelihood = -72.506965 
Iteration 1:   log likelihood = -71.738377 
Iteration 2:   log likelihood = -71.639325 
Performing gradient-based optimization:

Iteration 0:   log likelihood = -71.639325 
Iteration 1:   log likelihood = -71.638967 
Iteration 2:   log likelihood = -71.638967 
Mixed-effects logistic regression               Number of obs      
=       147
Group variable: motheridno                      Number of groups   
=       127

                                               Obs per group: min 
=         1
                                                              avg 
=       1.2
                                                              max 
=         2

Integration points =   7                        Wald chi2(10)      
=      6.62
Log likelihood = -71.638967                     Prob > chi2        =    
0.7608

------------------------------------------------------------------------------
 WinterSurv |      Coef.   Std. Err.      z    P>|z|     [95% Conf. 
Interval]
-------------+----------------------------------------------------------------
  speciesno |  -3.383566   1.626545    -2.08   0.038    -6.571536   
-.1955957
_Iyear_1995 |   2.469029   1.420471     1.74   0.082    -.3150429    
5.253102
_Iyear_2000 |   2.711117   1.730716     1.57   0.117    -.6810241    
6.103259
_Iyear_2001 |   2.751563   1.862746     1.48   0.140    -.8993519    
6.402478
_Iyear_2003 |  -.3852582   1.448119    -0.27   0.790    -3.223519    
2.453003
_Iyear_2004 |   4.209779   2.282672     1.84   0.065    -.2641753    
8.683733
_Iyear_2005 |   4.307069   2.250629     1.91   0.056    -.1040832    
8.718222
_Iwinterar~2 |   1.477567   1.264967     1.17   0.243    -1.001722    
3.956857
_Iwinterar~3 |   3.386436    1.83254     1.85   0.065    -.2052769    
6.978149
_Iwinterar~4 |   .5591356   1.529517     0.37   0.715    -2.438663    
3.556934
      _cons |    3.08273   2.351341     1.31   0.190    -1.525813    
7.691273
------------------------------------------------------------------------------

------------------------------------------------------------------------------
 Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. 
Interval]
-----------------------------+------------------------------------------------
motheridno: Identity         |
                  sd(_cons) |       1.88   1.239339      .5164477    
6.843673
------------------------------------------------------------------------------
LR test vs. logistic regression: chibar2(01) =     2.08 Prob>=chibar2 = 
0.0748



*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/