Two properties of the nested logit model

This paper presents simple formulae for the utility covariances of the nested logit (NL) model, and based on these defines a ‘scaled tree’ that can be used as an aid for the interpretation of estimation results. The paper also shows that the full information log-likelihood function of linear-in-the-parameters NL models is concave in the utility parameters. Thus, conditional on the scaling parameters, full information maximum likelihood searches cannot get trapped in local maxima.