The logit model is the most popular tool for estimating demand in differentiated products markets. However, in its aggregate version, practitioners have to “guess” the size outside good. We propose a way to remove the bias created by an inaccurate guess in simpler versions of the model.