Manon,

I am not sure how you group the data. I hope the following information helpful.

In Greene's Book 19.4.6 (page 835, 4th edition) Analysis of Proportions Data. He gives an approach to deal with the groupd data.

Note: The estimated coefs maybe much smaller from grouped data than those from individual(ungrouped) data because the variations reduce due to grouping.

Manon Girard wrote: > Hello members, > > I am stuck in a basic problem. > > My response variable is a proportion (A/B) and I have to modelized it in a > logistic regression, then I have to estimate the treatment effects using > predicted percents from the model. > > It my proc below would do it ? > > proc logistic data=xxx; > class trt_grp covariates; > model A/B=trt_grp covariates; > output out=x predicted=p; > run; > > When I tested it with my knowledge of logistic regression, I assumed that > with the intercept only, this should come as the same proportion of the > actual proportion. > As an example, if I use a 0/1 response variable (say C), a proc like "model > C=;", would give an estimated proportion as being the same as the proportion > of 0/1 in my dataset. > > I followed this strategy but this doesn't seem to work this way. > > If I put the question in another way, I calculated "result=A/B" for each > individual in my dataset. I want to see if there is a treatment effect in > "result", while taking into account all the possible covariate. Is Logistic > the best way to do such analysis ? > > Thanks > Manon