ICLV model scenario (simulation) analysis

[metrokim217]

Dear Prof. Hess and colleagues,

Thanks for your efforts and help on Apollo modeling. My model is a simple ICLV model that has 4 latent variables (pro-car, pro-bus, env-awareness, and safety) to explain commuting mode choice behavior (car, bus, bike, and walk). In this regard, I have a question regarding ICLV model scenario (simulation) analysis: Based on the manual, I acknowledge how I can simulate change of mode split when, for example, if bus travel time is reduced by 10%.

database$BUS_TIME=(0.9)*database$BUS_TIME
forecast <- apollo_prediction(model, apollo_probabilities, apollo_inputs,
prediction_settings=list(modelComponent='choice'))

However, I'd like to conduct the similar simulation analysis for a latent variable, such as environmental-awareness. (Similar to Section 6 of Hess et al. [2018] paper published in TRA) For example, I'd like to see how the mode choice estimation results may be different if environmental-awareness level of participants in my sample are changed to a certain degree that I'd like to examine.

Could you please provide me with a clue or something how I can conduct this kind of simulation analysis?

Best regards,
Junghwan

[stephanehess]

Hi Junghwan

as discussed by Chorus & Kroesen, it is not advisable to make a change to the latent attitude in forecasting as the scale of this arbitrary. So e.g. a 10% increase is not a meaningful concept. What we did in the 2018 paper is to look at a situation where everyone adopted the attitudes of a given socio-demographic segment. So you can look for example what would happen if all men took on the attitudes of women. You can do this by after estimation changing the definition of the latent variable inside apollo_randCoeff. So e.g. if you had:

randCoeff[["LV"]] = gamma_female * female + eta

where female is a column in the data, you could then change this to:

randCoeff[["LV"]] = gamma_female * 1 + eta

meaning that the gamma_female shift is used for all people.

This is a better approach than making the change in your database as the latter would also change the role of gender in the utility functions net of the latent variable

Best wishes

Stephane

[metrokim217]

Dear Professor Hess,

Thanks a lot for your kind reply! I tried it based on your suggestions, and it worked! Thanks again for your help!

Best wishes,
Junghwan

[FuZheng]

Hi! Dear Prof. Hess and colleagues,
I have a similar problem. My model is MDCEV model that has several variables to explain activity choice bahavior(Work,Study,Leisure,Personal Care and Unpaid Work).How I can simulate change of activity mode split , for example, if "Work" time is reduced by 10%?

[dpalma]

Hi,

In an MDCEV model the amount consumed of each alternative (activity) is your dependent variable, so you cannot calculate the impact of a change in it.I understand you want to understand how a change in consumption in one alternative influences the others, but it cannot be done directly. The only thing you can do is make a change in your explanatory variables that leads to a change in the consumption of one activity, and see how that affects the consumption of the others.

For example, if age positively influences the amount of time assigned to work, you can try increasing the age of individuals in your database, and see how that changes time allocation. I would expect an increase in work and a decrease in other activities.

Hope this helps.

Cheers
David