I am measuring the effect of a new in-store promotion technique. This technique will be applied to 1000+ products over a period of time, in the store. I obtain the uplift from each product promotions based on a regression model (actual minus fitted value in promotion period). I want to summarize these 1000+ results. I have a median. Is it reasonable to do a point estimate and build a confidence around it for these 1000+ results? I am not sure I can do that mainly because the uplift is not a random out comes, it are based on my model. And since all uplift calculation are based on the same model, they are probably not independent either because they may be effected similarly by whatever model bias that there may be in the model.

I would appreciate any guidance in this matter. Thank you for your time.