Traffic Engineering and Control, 38(11), 593-7 (1997) ON THE ERROR STRUCTURE OF DISCRETE CHOICE MODELS Marcela A. Munizaga Departamento de Ingeniería Civil, Universidad de Chile, Casilla 228-3, Santiago, Chile, Benjamin G. Heydecker Centre for Transport Studies, University College London, Gower Street London WC1E 6BT, UK and Juan de Dios Ortúzar Departamento Ingeniería de Transporte, Pontificia Universidad Católica de Chile, Casilla 306 Santiago 22, Chile ABSTRACT The increasing popularity of the stated preference (SP) approach, together with the strong recommendation to practitioners that whenever possible SP data should be mixed with revealed preference (RP) data, has revived old doubts about the applicability of logit models in travel demand forecasting. We consider here the error structure of the most common practical cases and use Monte Carlo simulation to examine the performance of various functions including the powerful multinomial probit model. Our results confirm the traditional view that logit models are remarkably robust and should perform reasonably well in most practical cases. 1. INTRODUCTION Discrete choice models have become firmly established as the primary travel demand forecasting tool in the last quarter of a century. The simplest version, the multinomial logit model (MNL), has remained the most popular one in practice; this has been aided by the fact that a close relative, the hierarchical or nested logit model (HL), helped to solve its most serious deficiency -