The performance of nested logit models when welfare estimation is the goal

In this paper we examine the performance of nested logit models in the face of two specification errore. The first specification error arises when a nested logit model is appropriate, but the wrong nesting structure is chosen. The second specification error occurs when the underlying stochastic process is not consistent with nested logit. Particular attention is placed upon the impact that these errors can hake on welfare predictions. Monte Carlo experiments are used, together with analytical results, to examine the resulting bias to welfare estimates. In addition, we explore the value of alternative model selection criteria in choosing nesting structures.