Propensity Scores

[SD29]

Hi all,

So I have observational data where my treatment variable has three possible values (high dose, low dose, and no dose). I am calculating propensity scores for my sample and have a question about how propensity scores are calculated/work. After using PROC LOGISTIC to calculate PS, I realized that my output data has 3x the sample size of my initial dataset and for each person, I now see three rows reflecting their PS for being in each of the three treatment groups. My understanding is that SAS is calculating the probability of each person being in each of the groups but I am really confused about how to move ahead. Do I continue working with all the three PS or just use the PS for the treatment group that an individual is actually in? I am going to use the inverse of PS to weight my models.

Sorry if this is a stupid question-- I have now spent hours reading about this but I am still confused.

Thank you!

[StatDave · Accepted solution]

You are probably using the PRED= (or P=) option in the OUTPUT statement. Since you have a multi-level response, you should be using the LINK=GLOGIT option in the MODEL statement and the PREDPROBS=INDIVIDUAL option in the OUTPUT statement and remove the PRED= option. The PREDPROBS option will produce you 3 variables for that give you the predicted probabilities that each observation belongs in the 3 response levels.