Abstract | Keywords | Export | Availability | Bookmark

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

This thesis focuses on two types of discrete choice models, the multinomial logit model and the subset conjunctive specification. We are trying to understand how they relate to each other by simulation with several diverse settings manipulating a number of subjects and choice sets. We take into account the different ways how each model explains opt-out response. Their performance is evaluated based on multiple criteria such as goodness model fit, estimation accuracy and forecasting ability. Furthermore, we compare them on an empirical data set regarding coffee types. Based on our simulation, we conclude that the model used to simulate certain data is always the one that fitted the best for such a data set. In general, the MNL model performs better in terms of recovery of the true parameters. The settings with a lower number of participants usually lead to worse overall performance. The model used to simulate certain data has also the best prediction ability for that data set. According to the results of our empirical analysis, the MNL model fits the data slightly more than the SC model. However, the interpretation of the coefficient estimates is nearly the same for both models so we can conclude both specifications produce reliable outcomes.