A major concern with the derivation of willingness to pay (WTP) distributions from mixed logit models is the incidence of values over a range that are deemed ‘behaviourally questionable’, with respect to the sign and magnitude. Recent research in redefining the ‘space’ within which a choice model is estimated as WTP space, instead of preference-space, has offered encouraging evidence in reducing the range of behavioural implausibility. This paper develops models to compare the empirical evidence when a mixed logit model is estimated with unconstrained distributions on random parameters in WTP and preference space, taking into account scale and the correlation between the attributes. The findings suggest that the gap between the evidence in WTP and preference space narrows significantly when both scale and preference heterogeneity are identified.