I recently had a philosophical question regarding the random utility theory: Does discrete choice model make sense for every individual?
Suppose a person lives 20 min drive from campus and owns a vehicle. If the transit service gets a little better for her (say transfer reduced by 1), she will not choose transit due to many other reasons. It seems unreasonable to assume her choice probability of transit will increase as the transfer gets reduced by 1.
Suppose another person lives close to campus and is currently driving. When the transit gets a little better, the person decides to switch to a combination of two modes (transit + walking). When the weather is good and time permits, she will walk; otherwise, she will take a bus. In this case, improving transit will increase the choice probabilities of both transit and walking, which violates the assumption of the random utility theory.
Does anyone know there is any work like this in the field to answer my questions above? Thanks in advance!
A follow-up question:
Nowadays, the activity-based model (ABM) has been widely used around the world, and an important step in the ABM is to use the discrete choice model to predict the travel mode for each individual/household. However, as many people discussed before, the discrete choice model is a probabilistic model and aims to predict at the aggregate level. Also, a lot of papers have suggested that the predictive accuracy of the discrete choice model is relatively low at the individual level. So, how can we ensure applying the discrete choice model to ABM can produce meaningful predictions for individuals' activities? Thank you very much!
Thanks, Marcus. Let me ask a follow-up question. If discrete choice model is so powerful, why it produces low predictive accuracy at the individual level (many research studies have this kind of results)? How can we trust the logit model's interpretation based on such low predictive capabilities? After all, the major goal is to forecast the travel demand using logit models.
Thank you, Marcus. Nowadays, the activity-based model (ABM) has been widely used around the world, and an important step in the ABM is to use the discrete choice model to predict the travel mode for each individual/household. However, the discrete choice model is a probabilistic model and aims to predict at the aggregate level. As you mentioned above, "By the very nature of the models, it is not possible to measure predictive accuracy at the individual level." So, how can we ensure applying the discrete choice model to ABM can produce meaningful predictions for individuals' activities?
Thank you, Juan. I know this paper very well. I'm currently working on this research topic, and recently wrote a paper about it: https://arxiv.org/abs/1811.01315. If you are interested, we could talk more about it.
- Badges
- Science topic
- Similar topics
- Computer Communications (Networks)
- Simulators