I had a breakfast discussion about this simulation-for-transportation-policies with two friends of mine who have been looking into similar topics for years: Nadav Levy and Itzhak Benenson of Tel Aviv Univ.
This was triggered by my post about a related topic. Here are the main things I learned during breakfast:
They are interested in policies like:
- Pricing of parking so as to minimize car cruising (until you find parking)
- Where and at what frequency should buses run
- Where to put your light train for best results, taking the future into consideration
Obviously, those have different time-frames (building the light train track will take years, so longer-term simulations are needed).
So how do you predict stuff? There are various solutions:
- Asking people: Good, but you get stated preferences, not observed preferences, and they almost never actually know what they will want in some imagined situation
- Measuring current behavior: Good for getting preferences in the current situation, but very hard to interpolate into future situations (e.g. what would happen if you had a train near you).
- White boarding: Good but hard to verify – what if you did not think of something? How do the various ideas combine?
- Simulation: Good, but has various issues:
Issues with simulations for transportation policies:
- Hard to construct
- The input world is, some sense, infinite
- Related: Simulation can only combine basic events you give it. But it will not dream up new basic events
- Often it is the same modelers who construct both the problems and the agents’ reaction to the problems. This can cause biases, especially when modeling human behavior.
Here are some examples of new basic events that somebody must supply:
- The Bridge-to-Malmo example from the original post: Nobody thought of considering the effect of real estate prices
- A wait-at-intersection parking example: Most people would not consider the fact that some people, when faced with no parking space, will simply wait at an intersection until a parking space appears (or until somebody honks behind them)
So how do you dream up new basic events? This is a big topic – expect several future posts about it.
My friends have an idea which may (partially) solve this problem, as well as the “stated preferences vs. observed preferences” problem mentioned above. Can’t talk about it yet, unfortunately ;-).