While I'm at it, I'll share a few other critiques that I made note of while reading the paper:
I was suspicious of where this was going in Section 2.1 where he claims public transit carries 14% of commutes in the U.S. That number is way above anything I’ve heard (5% being the usual number I see), so I checked the footnote and it’s proprietary, not public data. But then I noticed the modes listed add up to more than 100%. 107% to be precise, and it may still be omitting some modes like carpool. So we started in a weird place with inaccurate data that assumes almost 3X more transit usage than even the industry lobbying group.
The paper concludes that doubling parking charges will decrease miles driven, presumably because you have to pay to park near your office, and if parking costs more, you might choose transit. It doesn’t consider that people with AVs aren’t stuck paying the high price of parking downtown. They just have the vehicles drive back out to where parking is cheap. At 18 cents per mile marginal cost, you can drive a long way and still save money. This could double VMT if everyone’s car just drives home for the day.
It talks about modest changes to public transportation being enough to make a difference. Such as a 40% reduction in travel time. “40%” doesn’t sound like much, but that’s the same as saying making public transportation 67% faster. This is a massive challenge. Each moving segment has to be 67% faster. Each dwell time has to be 67% faster. These are not easy, simple changes. Especially during a cycle when transit usage (even under the optimistic model) is going down.
It’s a good paper that furthers the argument that AVs will increase sprawl and congestion. I still think he underestimates the impacts, especially on transit. Despite a lot of detail on how the model is set up and calibrated, it just skips to the conclusion that somehow transit is hardly affected by the introduction of AVs.
Given the setup of the model, that doesn’t make sense. They cut the disutility of driving to be equal transit, but now with an AV you get to take a direct trip, without out-of-the-way travel and transfers and a general mode preference that prefers a personal space. How does this not take away ridership from transit?
Figure 4 is the one I can’t figure out. It shows a 40% increase in driving, with no change in home locations, but transit usage drops by only <5%?? Where do these new drivers come from? At first I assumed I was reading it wrong, but in the appendix, there is a simlar Figure A.2. It sums to zero as you expect it would – the shift to driving is matched by a corresponding drop in transit.
This part fails the common sense test – having your own SDV beats transit in time, comfort, transfers, everything except cost, but even that is fairly close … and somehow people still choose transit at about the same rate as today? Something’s off.
One thing I think this analysis overlooks is that widespread deployment or adoption of self-driving vehicles is not some future scenario but in a sense already here since every owner of a 2020 or later model tesla can already pay the $100 per month to get the capability, and the price of a tesla - especially a used one - is much lower than a new Waymo-type vehicle. Indeed Tesla is selling robotaxi rides in Austin right now, and software updates combined with their next generation cameras and processing chips (such as should be running in the CyberCabs) should bring the newest Teslas to the frontier. So in a way the roads are already full of millions of vehicles with these capabilities, like mine which I use every day. One of Musk's proposals is that one day every Tesla owner could rent their vehicle out like the Airbnb version of Uber and it could even drive itself to automated charging stations instead of parking it. As a regulatory or insurance matter, that settlement is probably further out than the tech itself. Still, a point I've made previously elsewhere is that despite local political opposition and legal roadblocks to Wayno, one can already pay a fee to have a robot car drive one around DC, New York, Boston, or Toronto right now, it's just that instead of being allowed to hail a Waymo, one has to be in the privileged position of being able to afford to own and park a Tesla, a case of what one might call "AV-ride-inequality".
As personal testimony, having self-driving in my tesla has absolutely shifted my transportation-options indifference curve sharply in the direction of more driving. It's not just for local moves and commutes, where traffic jams and parking costs can still weigh in favor of public transit. It's even more stark for medium-distance long trips, say between 200-400 miles, when I might have been tempted to take a flight, long distance train, or bus ride. No longer! At last so long as I expect to have easy or cheap parking at my destination. In my opinion and recent experience, a self-driving car for such trips has an absolute advantage over the alternatives in the clear majority of dimensions.
The simulation for AVs generally follows spatial models (e.g., Muth, Cities and Housing , 1969) that indicate that more mobility = sprawl. The idea that transport can be a setting (e.g., for doing email or reading) rather than just a link applies as well to transit (oh how I miss DC metro for my commuting days). Then what this suggests is the race between AVs and transit. All depends on public policy of whether we get more rail or more highway.
I find it hard to imagine that will happen, but if it does, autonomous cars could conservatively double the capacity of freeways by maintaining smaller gaps in traffic
Of course Congress should change the mass transit law to not require keeping redundant transit workers, but another option for a commuter rail operator that wanted to go to full automation with smaller, more frequent trains, would be to reassign the now unnecessary drivers to be like rail marshalls who police behavior on the new self driving trains. In this way the trains would be more frequent, less crowded and well behaved. Since every train wouldn’t require a marshall, not new employees would be needed, and labor costs can be reduced by attrition.
While I'm at it, I'll share a few other critiques that I made note of while reading the paper:
I was suspicious of where this was going in Section 2.1 where he claims public transit carries 14% of commutes in the U.S. That number is way above anything I’ve heard (5% being the usual number I see), so I checked the footnote and it’s proprietary, not public data. But then I noticed the modes listed add up to more than 100%. 107% to be precise, and it may still be omitting some modes like carpool. So we started in a weird place with inaccurate data that assumes almost 3X more transit usage than even the industry lobbying group.
The paper concludes that doubling parking charges will decrease miles driven, presumably because you have to pay to park near your office, and if parking costs more, you might choose transit. It doesn’t consider that people with AVs aren’t stuck paying the high price of parking downtown. They just have the vehicles drive back out to where parking is cheap. At 18 cents per mile marginal cost, you can drive a long way and still save money. This could double VMT if everyone’s car just drives home for the day.
It talks about modest changes to public transportation being enough to make a difference. Such as a 40% reduction in travel time. “40%” doesn’t sound like much, but that’s the same as saying making public transportation 67% faster. This is a massive challenge. Each moving segment has to be 67% faster. Each dwell time has to be 67% faster. These are not easy, simple changes. Especially during a cycle when transit usage (even under the optimistic model) is going down.
It’s a good paper that furthers the argument that AVs will increase sprawl and congestion. I still think he underestimates the impacts, especially on transit. Despite a lot of detail on how the model is set up and calibrated, it just skips to the conclusion that somehow transit is hardly affected by the introduction of AVs.
Given the setup of the model, that doesn’t make sense. They cut the disutility of driving to be equal transit, but now with an AV you get to take a direct trip, without out-of-the-way travel and transfers and a general mode preference that prefers a personal space. How does this not take away ridership from transit?
Figure 4 is the one I can’t figure out. It shows a 40% increase in driving, with no change in home locations, but transit usage drops by only <5%?? Where do these new drivers come from? At first I assumed I was reading it wrong, but in the appendix, there is a simlar Figure A.2. It sums to zero as you expect it would – the shift to driving is matched by a corresponding drop in transit.
This part fails the common sense test – having your own SDV beats transit in time, comfort, transfers, everything except cost, but even that is fairly close … and somehow people still choose transit at about the same rate as today? Something’s off.
One thing I think this analysis overlooks is that widespread deployment or adoption of self-driving vehicles is not some future scenario but in a sense already here since every owner of a 2020 or later model tesla can already pay the $100 per month to get the capability, and the price of a tesla - especially a used one - is much lower than a new Waymo-type vehicle. Indeed Tesla is selling robotaxi rides in Austin right now, and software updates combined with their next generation cameras and processing chips (such as should be running in the CyberCabs) should bring the newest Teslas to the frontier. So in a way the roads are already full of millions of vehicles with these capabilities, like mine which I use every day. One of Musk's proposals is that one day every Tesla owner could rent their vehicle out like the Airbnb version of Uber and it could even drive itself to automated charging stations instead of parking it. As a regulatory or insurance matter, that settlement is probably further out than the tech itself. Still, a point I've made previously elsewhere is that despite local political opposition and legal roadblocks to Wayno, one can already pay a fee to have a robot car drive one around DC, New York, Boston, or Toronto right now, it's just that instead of being allowed to hail a Waymo, one has to be in the privileged position of being able to afford to own and park a Tesla, a case of what one might call "AV-ride-inequality".
As personal testimony, having self-driving in my tesla has absolutely shifted my transportation-options indifference curve sharply in the direction of more driving. It's not just for local moves and commutes, where traffic jams and parking costs can still weigh in favor of public transit. It's even more stark for medium-distance long trips, say between 200-400 miles, when I might have been tempted to take a flight, long distance train, or bus ride. No longer! At last so long as I expect to have easy or cheap parking at my destination. In my opinion and recent experience, a self-driving car for such trips has an absolute advantage over the alternatives in the clear majority of dimensions.
The simulation for AVs generally follows spatial models (e.g., Muth, Cities and Housing , 1969) that indicate that more mobility = sprawl. The idea that transport can be a setting (e.g., for doing email or reading) rather than just a link applies as well to transit (oh how I miss DC metro for my commuting days). Then what this suggests is the race between AVs and transit. All depends on public policy of whether we get more rail or more highway.
I find it hard to imagine that will happen, but if it does, autonomous cars could conservatively double the capacity of freeways by maintaining smaller gaps in traffic
Of course Congress should change the mass transit law to not require keeping redundant transit workers, but another option for a commuter rail operator that wanted to go to full automation with smaller, more frequent trains, would be to reassign the now unnecessary drivers to be like rail marshalls who police behavior on the new self driving trains. In this way the trains would be more frequent, less crowded and well behaved. Since every train wouldn’t require a marshall, not new employees would be needed, and labor costs can be reduced by attrition.
“Less crowded and well behaved”? It’s not a foregone conclusion.