Can mixed logit reveal the actual data generating process? Some implications for environmental assessment

Recent advances in the specification of the utility function of mixed logit models allow the analyst, in principle, to consider a vast variety of individual heterogeneity. Nevertheless, when estimating the model it is common practice to experience severe difficulties in discriminating between different specifications to infer the “true” data generating process. We investigate possible sources for this difficulty focusing on the confounding effects inherent in two basic assumptions of discrete choice model utilities: linearity in the parameters and added error terms. We analyse the role of these assumptions in giving rise to confounding effects and why this increases the difficulty of discriminating among different structures. Finally, we investigate how these problems may affect benefit appraisal using these models. Empirical evidence is provided for two different environmental contexts and a more typical transport context using various kinds of data.