Joseph, thanks much-- I have tried the identity link and it does
produce nearly identical results as the logit model, although the
logit tends to fit the data *slightly* better. But I think the glm
approach is a better way to go to get the ARs. Thanks, Tim
On Fri, 04 Mar 2005 13:12:10 +0900, Joseph Coveney
<[email protected]> wrote:
> Tim Wade wrote:
>
> I have estimated probabilities for two groups (e.g., exposed and
> unexposed) using the "adjust" command after a logistic regression,
> holding several other covariates to their mean or other specified
> values. The main exposure varies among individuals and is continuous.
> I would like to estimate the risk difference for these each individual
> (p_exposed-p_unexposed) and the confidence limit for this estimate.
> P_unexposed is constant for all unexposed. The risk difference is no
> problem, simply p_exposed-p_unexposed but I can't figure out how to
> get the confidence limits or standard error. A partial line listing of
> the predicted values follows, xse is the standard error of the linear
> predictor (not of the probability). The only other option I could
> think of would be to manually input the mean and other values get the
> estimates and their ci's using predictnl. Does anyone have any other
> suggestions?
>
> ----------------------------------------------------------------------------
>
> Why not get them directly by changing the link function?
>
> Currently: -glm . . . , family(binomial) link(logit)-
>
> Recommendation: -glm . . . , family(binomial) link(identity)-
>
> Joseph Coveney
>
>
> *
> * 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/
