R help - Jul 2010 - logistic regression with repeated measures for species-habitat associations

Hi,
     I have three years of species presence/absence data for a set of ~100
ponds, with a list of associated habitat characteristics (park, vegetation,
hydroperiod, etc.).  The datasets differ slightly by year because some ponds
were dry in some years and not in others.  My goal is to look at habitat
associations for a couple focal species.

For each year, I have a data table that looks like (an expanded version of)
this:

pond park  hydroperiod  elevation  veg.cover  fish  odonate  beetle
 bullfrog  newt  tiger.sal . . . etc.
1       DV    perennial        536           25         1         1
 1           1        0         0
2       DV    perennial        722           0           0         1
 1           0        1         1
3       SN    perennial        643           30         0         1
 1           0        1         0
4       SN    seasonal       1238           0          1         1
 1           0        0         0
5       SN    seasonal       2406          90         0         1          1
          1        0         1
.
.
.
etc.

So far I've analyzed each year individually using (example for 2004):

newts2004 <- glm(newts ~ ., data=ponds2004, family="binomial")
step(newts2004)

I'd like to do a holistic analysis for all three years but am unclear how to
do logistic regression with repeated measures in R.  Also not sure if
missing data (from dry ponds) will be a problem.

Thanks very much for any advice.
Cheers, Maureen

	[[alternative HTML version deleted]]

Maureen Ryan <meryan <at> ucdavis.edu> writes:
> 
> Hi,
>      I have three years of species presence/absence data for a set of ~100
> ponds, with a list of associated habitat characteristics (park, vegetation,
> hydroperiod, etc.).  The datasets differ slightly by year because some
ponds
> were dry in some years and not in others.  My goal is to look at habitat
> associations for a couple focal species.
> 
> For each year, I have a data table that looks like (an expanded version of)
> this:
> 
 [snip]

  Maureen,
  
  I think this is a generalized linear mixed model with binary
outcomes, with fixed effects for the habitat variables, year, and
(possibly) the interaction of habitat by year, and with ponds
as random effects.  I would recommend either lmer (in the lme4 package)
or MCMCglmm (in the package of the same name), and I might recommend
also that you visit the r-sig-mixed-models mailing list with additional
questions ...
   Missing data should not be a big problem.

  Ben Bolker