Discrete choice models' ρ2: A reintroduction to an old friend

We first review the intuition behind ρ2, and its conceptual interpretation. We then comment on the choice of benchmark (typically either the equally-likely, EL, or the market-share, MS, model), together with discussion of what is considered a “good” value. After a brief mention of the adjusted ρ2 and of statistical distributions associated with ρ2, we close with a description of its computation under three special circumstances: repeated observations, unequal choice sets, and deterministically-segmented models.