R help - Apr 2012 - random effects using lmer

Hi,

I am trying to run a logistic regression to look at the risk of malaria
infection in individuals. I want to account for intra household correlation
and so want to include a household level random effect. I have been using
the lmer command in lme4 package but am getting some strange results that
are completely different to those I get using STATA.

Can I just check that this is the correct code

lmer(IsPos ~  Dist + (1 | HouseID), family=binomial)

Where IsPos is a binary vector of positives and negatives, Dist is the
variable of interest and HouseID is the household ID number.

For those STATA users, the equivalent I'm using is

xtlogit IsPos Dist, i(HouseID)

Any help would be much appreciated..!

Thanks!
Hugh

--
View this message in context:
http://r.789695.n4.nabble.com/random-effects-using-lmer-tp4566030p4566030.html
Sent from the R help mailing list archive at Nabble.com.

HughSt <hughsturrock <at> hotmail.com> writes:
> I am trying to run a logistic regression to look at the risk of malaria
> infection in individuals. I want to account for intra household correlation
> and so want to include a household level random effect. I have been using
> the lmer command in lme4 package but am getting some strange results that
> are completely different to those I get using STATA.
> 
> Can I just check that this is the correct code
> 
> lmer(IsPos ~  Dist + (1 | HouseID), family=binomial)
  This does seem reasonable, although I would probably use
glmer() rather than lmer() for clarity [lmer() automatically
calls glmer() when 'family' is specified]
 
> Where IsPos is a binary vector of positives and negatives, Dist is the
> variable of interest and HouseID is the household ID number.
> 
> For those STATA users, the equivalent I'm using is
> 
> xtlogit IsPos Dist, i(HouseID)
> 
  So you're not doing anything obviously silly.  It seems that
STATA uses gauss-hermite quadrature by default -- glmer uses
Laplace, so you might try something like nAGQ=10 and see if
anything changes.

   Follow-ups should go to r-sig-mixed-models <at> r-project.org