Thanks very much for the advice. I have thought about
the effect of time. The mailings were sent out
periodically to provide continual education, so I
don't think the effect waned as much due to time.
There is expected to be a large change from time 0 to
1 (3 months out) and probably not so large of a change
after the 3 month average of antibiotic prescriptions.
I will use xtgee with probably an unstructured
(possibly auto-regressive) correlation assumption.
Greg
--- "FEIVESON, ALAN H. (AL) (JSC-SD) (NASA)"
<[email protected]> wrote:
> Mike - You should also consider the effect of time -
> i.e. as time goes on,
> the effect of the intervention might wane. Suppose
> you have a logistic
> regression model something like
>
> logit p(prescribe antibiotics)=f(time,intervention)
> + e
>
> In GEE, (with an equicorrelated assumption), you are
> only assuming equal
> correlation of e within dentists - no other
> distributional assumptions on e.
> With xtlogit, e is split into two random effects,
> between and within
> dentists, with each effect normally distributed and
> independent. So the
> equicorrelation arises because the between-dentist
> random effect is the same
> for all observations pertaining to that dentist.
>
> As in most statistical modelling problems, the more
> sophisticated model
> (xtlogit), if correct, gives greater power. However
> there is a risk of bias
> if the model is not correct. GEE makes fewer
> assumptions, so there is less
> risk of bias, but it would in general be less
> powerful.
>
>
> Also, in GEE you have the option of specifying other
> types of correlation
> within dentists - see manual.
>
>
> Hope this helps-
>
> Al Feiveson
>
>
>
>
>
> -----Original Message-----
> From: Mike Schmader [mailto:[email protected]]
> Sent: Tuesday, November 12, 2002 8:17 AM
> To: [email protected]
> Subject: st: Mixed effects or GEE?
>
>
> Dear listers,
>
> I am confused about the differences between random
> effects mixed effects and GEE models.
>
> I am looking at prescribing patterens of dentists.
> My
> outcome variable will be % of prescriptions written
> for antibiotics at baseline (t=0), 3 months (t=1), 6
> months (t=2) and 12 months (t=3) post initiation of
> intervention mailings (educational material).
> Dentists were randomly assigned to intervention
> groups: group 1- "no intervention"; group 2- "low
> level of intervention"; group 3 -"high level of
> intervention".
>
> I will be analyzing this data with panel techniques
> but I am not sure whether I should use GEE, random
> effects, or mixed effects models. I will be using
> dentist ID as my clustering variable and indicator
> variables for level of intervention. Can anyone
> point
> me in the right direction as to the main differences
> in these techniques as I have not used these methods
> before?
>
> Thank you in advance.
>
> Mike
>
>
>
> __________________________________________________
> Do you Yahoo!?
> U2 on LAUNCH - Exclusive greatest hits videos
> http://launch.yahoo.com/u2
> *
> * 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/
> *
> * 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/
__________________________________________________
Do you Yahoo!?
U2 on LAUNCH - Exclusive greatest hits videos
http://launch.yahoo.com/u2
*
* 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/
