Unlike a traditional fixed parameters logit model, the computation of marginal effects (MEs) for random parameters logit models is much more complex. In this study, a random parameters logit with heterogeneity in means and variances model was estimated using crash-severity data. Three computing methods for MEs based on global means, individual estimates and Monte Carlo simulation of random parameters were proposed and the results, along with the software-reported MEs, were comprehensively compared. Results indicate that: 1) enormous bias was detected in software-reported MEs; 2) simply using means of random parameters also produced bias; 3) the Monte Carlo simulation was most likely an effective way to computing MEs; 4) the individual estimates method may also be reliable as the random parameters distribution has been well captured. Methods provided by this study can prompt the proper application of random parameters approach for not only road safety but also other traffic scenarios.

random parameters model;marginal effects;logit model;crash injury