The Nested Multinomial Logit (NMNL) model is used extensively in modeling consumer choices among discrete alternatives when the number of alternatives is large. Unfortunately, applied researchers often find that estimated NMNL models fail to meet the Daly-ZacharyMcFadden (DZM) sufficient conditions for consistency with stochastic utility maximization. Borsch-Supan (1990) provides a relaxed set of conditions to test for consistency. While these conditions are increasingly cited, they are seldom tested. This paper corrects and extends BorschSupan's Theorem 2, providing simple necessary conditions on first, second, and third derivatives of choice probabilities and a graph oft he bounds they place on dissimilarity parameters.
This report is published in Economic Letters, Vol 50, No. 1, January 1996, pp. 33-39
Herriges, Joseph A. and Kling, Catherine L., "Testing The Consistency of Nested Logit Models with Utility Maximization" (1994). Economic Staff Paper Series. 268.