Analysis of Braess's Paradox in Boston

[cozzyd]

I just saw http://arxiv.org/abs/0712.1598, which discusses some empirical examples of Braess's Paradox in Boston (and London and NYC).

It would be interesting to integrate over all interesting pairs to help figure out what roads are essentially useless to through traffic.

 

[vanshnookenraggen]

That hurts my head.

 

[CSTH]

My hunch is that this would apply to all urban interstates / freeways everywhere....

 

[cden4]

It definitely seems like Braess Paradox is a big factor in the Sagamore/Bourne Bridge congestion

 

[Arlington]

That hurts my head.

The examples and explanations are easier than the math:

In a nutshell: everyone crowds onto the slow-moving segment that is part of their shortest trip. It remains optimal for everyone to crush that one segment (that it'd be better to just remove)

In the real (traffic) world you also get these re-inforcing Braess:
- segment produces conflicting flows (congestion at intersections)
- segment induces demand (technically, not part of the paradox, as I understand it)

My favorite (hunch) is the segment of Rt 16 that connects Mass Ave Camb/Arl to Route 2. If you nipped it out, Some would move to Broadway or other streets, far bypassing, and the remaining Mass Ave corridor traffic would spread itself all up and down Mass Ave and Rt 2, and the 16/2 junction would go away (greatly simplifying & speeding flows around Alewife T)

Further (but not strictly part of the paradox) you'd probably drive a lot more people to just park at Alewife (assuming you expanded parking) or shift to other modes.

 

[tangent]

I don't think it is counter-intuitive at all to think that removing some segments in street network with traffic lights at intersections could improve overall driving times. This paper doesn't appear to address traffic on non-intersecting roadways such as express ways.

This reminds me, does DoT have a good computer model of the entire transportation network where they can model the effects of changes?

Just enter in the parameters of the change, how much it will cost and the result would be how much it improved overall transportation times in the computer model. The cost could then be used to prioritize projects that give the most bang for the buck.

 

[CSTH]

This paper doesn't appear to address traffic on non-intersecting roadways such as express ways.

Just because there aren't traffic lights doesn't mean there aren't intersections.

...How many of the local highway pain points we know and love come from overloaded on and off ramps???

 

[cozzyd]

I don't think it is counter-intuitive at all to think that removing some segments in street network with traffic lights at intersections could improve overall driving times. This paper doesn't appear to address traffic on non-intersecting roadways such as express ways.

I believe all that is necessary is that travel time gets worse with congestion for the paradox to be possible.

This reminds me, does DoT have a good computer model of the entire transportation network where they can model the effects of changes?

Just enter in the parameters of the change, how much it will cost and the result would be how much it improved overall transportation times in the computer model. The cost could then be used to prioritize projects that give the most bang for the buck.

It would be awesome if this was public (if it exists). What would be interesting is if something like Google directions would not necessarily give you the best route all the time, but rather give people different routes in order to decrease total congestion.

 

[Matthew]

CTPS does have a big comprehensive transportation model. They use it all the time in attempting to perform +10, +20 or +30 year projections for various projects.

The question is: can it be believed?

I don't think so. Models are human creations, suffering from all the assumptions and foibles of the people who program them. Using a model for a projection is a fancy way of saying "trying to predict the future" and we all know how poorly that usually turns out.

None of the models used by Federal or State DOTs predicted the peak and subsequent downturn in VMT/person. Most of them are still having difficulty with the concept, ten years later.