ENBIS-17 in Naples

9 – 14 September 2017; Naples (Italy) Abstract submission: 21 November 2016 – 10 May 2017

Estimating Mixed Logit Models

11 September 2017, 10:30 – 10:50


Submitted by
Martina Vandebroek
Martina Vandebroek (KU Leuven), Deniz Akinc (KU Leuven)
The mixed logit model is often used to model the results of a discrete choice experiment. It allows to estimate the mean preference in the population as well as the heterogeneity in these preferences. The model can be estimated with maximum simulated likelihood and by hierarchical Bayesian methods and both approaches have their advantages and disadvantages. Both approaches require that several choices are made. For maximum simulated likelihood, the number and type of random draws have to be chosen as well as the starting values and the optimization algorithm. For hierarchical Bayes estimation priors have to be chosen for the mean and covariance matrix of the parameters. We present the results of a simulation study in which we investigated the impact of these choices on the results. We report on the root mean squared error of the estimates for the mean, the covariance matrix as well as the individual preferences. We focus mainly on the number of quasi-random draws and on the prior for the covariance matrix as these are found to have a large impact on the results.

Return to programme