“…In the logit model, we assume that each bundle σ 2 fxy, x, y, ?g has a 'systematic utility' u : fxy, x, y, ?g ! R ++ , and that σ is chosen with logit probability, namely p (u,λ) logit (σ) = e u σ /λ ∑ τ2fxy,x,y,?g e u τ /λ (1) where λ > 0 is a scaling factor (measuring the variance of the underlying Gumbel errors, see McFadden [21]). In this specification, purely random behaviour (i.e.…”