choice-metrics.com

Hi there,

I would like to implement a mode choice experiment with many modes (walking, cycling, car, car-sharing, and public transport). However, I would like to distinguish between three PT alternatives with different main modes: bus, tram, and train. For complexity and response burden reasons, it would be better to only show one out of these three PT main modes in each choice task. I see two options to approach this problem:

1. Something along the line of chapter 8.11 in the NGene manual: Showing only a subset of alternatives, where one basically has one utility for each of bus, tram, and train that would be shown to respondents. That would require a partial choice set design and constructed candidate set. Since we have many different distance classes, the effort needed to implement it is huge but would be clean(er) from a modeling perspective than option 2.

2. In NGene, only one PT utility where the main mode is a dummy attribute with three levels corresponding to bus, tram, and train. That would not require a partial choice set approach, if I am not mistaken. Would it still be possible to model three utilities, say in an MNL model, based on which main mode attribute was shown?

Is there another approach more suitable in this case?

Thank you in advance for your thoughts.

Best regards, Thomas

The second method I refer to as "labelled-to-unlabelled reformulation" for generating partial choice set designs, and I presented this at a conference in 2018. I can send you my presentation slides where you can also see an Ngene example script. It works by creating interactions with indicator variables so that you can still estimate the same parameters as you would with having separate utility functions. Please send me your email address in a private message here and I will send you the slides.

Best wishes,
Michiel

Done.

Thank you, Michiel!