A Reassessment of Likelihood Approximation by Integration on Sparse Grids

ABSTRACT

This paper revisits sparse grid integration proposed in the literature for approximating integrals that occur as choice probabilities in random coefficient discrete choice models. First, we successfully replicate their main findings for the panel mixed logit. Second, for higher variances and for a different structure of the variances of the random coefficients, in certain cases, we fail to replicate the original results. Third, for the important special case of cross‐sectional mixed logit, replication of the original results is successful when the number of alternatives is moderate but fails otherwise.