Transportation For America The traffic forecast used to justify your road widening is bogus

The traffic forecast used to justify your road widening is bogus

29 Jun 2023 | Posted by Chris Rall | 8 Comments | congestion, state DOTs, traffic, usdot

The predicted traffic levels on which transportation planners base their decisions are erroneous and rooted in obsolete methods. Here’s how transportation models fail to accurately predict future traffic, and how you can call out their misuse.

Highway lanes crisscross across an otherwise barren landscape. Rows of tightly clustered cars dot the lanes — The 26-lane Katy Freeway in Houston had worse traffic after its widening than before. Were the traffic models wrong? Photo: Wikimedia Commons

You’ve seen it before. A state DOT claims they must widen a highway through your community to reduce congestion and accommodate future traffic. The transportation agency points to traffic projections that we all take at face value. They might even claim that widening the highway will improve traffic flow thereby reducing emissions. You don’t want the highway widening in your community, but what can you do in the face of experts saying it is necessary and pointing to data that “proves” their case?

Transportation agencies use transportation models to predict future traffic and plan the roadway system accordingly. But the underlying algorithm for these models was developed in the 1980s when the computers in use were less powerful than today’s smartphones. Due to this past limitation in computing power, travel demand models use a simplified approach that doesn’t accurately represent how people make travel decisions.

T4America experts collaborated with our partners to look inside the black box of transportation models (also sometimes called travel demand models or traffic models). We submitted a memo to the US Department of Transportation asking them to apply more accountability to agencies using these models to correct them.

Some of the transportation models’ specific flaws

The proof that transportation models are failing us is plain to see in the long term trends. Over the last 20 years, congestion has increased in every single U.S. metropolitan area regardless of how much they’ve expanded their highways and regardless of whether their population grew or shrank.

Graphic showing increase in population, lane-miles, and delay from the Congestion Con report — *From the Congestion Con*

In what way have transportation models misled us? It largely has to do with the underlying approach which is too simple, chosen because of limited 1980s computing power. Transportation models use a Static Traffic Assignment (STA) algorithm which is a sort of snapshot in time of how much traffic is on each roadway in a region at a given moment. This static algorithm is problematic, since people make decisions on different factors every day, often in the moment. People are dynamic not static.

What’s more, STAs do not properly account for bottlenecks, or constrain forecasts based on roadway capacity. No roadway can ever carry more cars than its maximum capacity, any more than a coffee mug can hold 110% of its coffee capacity. Yet agencies routinely and confidently make claims like, “without this expansion, the roadway will be at 110% capacity.” If you point out that a roadway can’t handle more cars than it has capacity for, they say that extra 10 percent is “latent demand.” In other words, they are certain that there’s exactly 10 percent more cars and trips out there that must be served.

We call this induced demand—demand created by the new road itself—a concept those same agencies often claim doesn’t exist. (But which the public absolutely understands, as our brand new national polling shows.) By trying to sell the project on all that “latent” demand, they can claim a traffic nightmare if nothing is done without admitting that the project will actually create more traffic—and more greenhouse gas emissions, fine particulates, etc. [USDOT and the Environmental Protection Agency support that approach for some unfathomable reason, never asking if the models used to justify federally funded projects have been right.]

In reality, as congestion increases toward that 100% capacity mark, people make different travel decisions, change their routes, choose to travel at a different time, use a different mode or choose a closer destination to fulfill the same need. If there is a crash, people delay their trip or consult Google maps and choose a different route. But transportation models using the STA approach unrealistically assume people will blindly keep driving a congested roadway, no matter what is happening or how long their trip will take.

Not only does the model assume no changes in behavior, but it will also output results that show drivers stuck at one bottleneck, while simultaneously allowing them to magically pass through that one to also be stuck at another bottleneck downstream.

Compounding these issues, planners rarely, if ever, look back at their past work to see if their predictions were correct. Did the traffic materialize? We’re stuck with decades-old models that are never tested or upgraded to reflect reality, as shown here:

graphic combining 20+ increasing projections of VMT — This graphic from the Frontier Group combines past federal projections of future growth in vehicle miles traveled. Every year a new optimistic projection was made that ultimately didn’t pan out, but they kept on predicting the same thing.

How to question your region’s model

We’re hopeful that USDOT will eventually provide accountability to upgrade the state of practice on transportation modeling, but you can also ask questions about the transportation models used to promote road widenings in your community. Here are some things you can ask your local transportation planners to illustrate the flaws of using transportation model results to justify road widenings:

Does your model use Static Traffic Assignment?
What is the maximum volume to capacity in your model runs, and how is that realistic? (If they give a volume over 100%, ask how a road can carry more than its capacity. And ask if latent demand will fill the new capacity they are building then what good will this investment do?)
How does your model account for dynamic changes in commuting patterns, responses to crashes, or the threshold at which people shift to other modes?
What is your protocol for evaluating the accuracy of your past traffic projections and using that to improve upon the model? Where is it published?

Getting our transportation models to better reflect reality will help planners make better decisions about where to invest our tax dollars. Calling on USDOT to upgrade standards for transportation models, and calling out their misuse locally in the meantime will help us turn the corner to more sensible improvements to transportation in our nation.

8 Comments

willpoundstone
10 months ago

The fact that VMT has been largely flat since 2000 despite countless highway expansion projects since kind of debunks your belief that new roads always fill up.
Allen Muchnick
10 months ago

@willpoundstone

The graph above shows that VMT increased about 18% (from about 2.75 trillion to about 3.25 trillion) from 2001 and 2019. That may not greatly exceed the rate of U.S. population growth, but it’s not “largely flat”.
- willpoundstone
  10 months ago
  
  It’s a lot slower than the growth before then.
Tom
10 months ago

It’s not the models that are wrong, it’s the decisions that are made. Blaming the modeling technique is like blaming air quality sampling device for climate change. Stop attacking tools and professionals, they are not the reason for congestion. It’s policy and decision makers that form this problem.
Harry Grouchman
10 months ago

Ironically, you’ll have to get in line behind the people who want to build wider roads. They will also say “the model is wrong”.

They’ll say it underestimate the scale of traffic congestion, overestimates speeds on highways, doesn’t accurately capture vehicle flow breakdowns, traffic queues and spillbacks for miles, nor does it accurately project all the “diversion” onto alternative arterial roadways that occurs even today, let alone in the future. All of this is valid BTW.

They’ll also, in instances where “they” are savvy enough to know, say it underestimate true demand and ask for an “unconstrained” model run because the off the shelf version includes “punitive” mode shift away from the majority’s natural and preferred mode of travel – driving alone. What you called induced demand they call true demand – from the fantasy world where there is no traffic congestion ever.

Their conclusion: We’re not building fast enough to accommodate all that demand that the model isn’t showing and underestimating all that traffic congestion the model isn’t accurately showing. They blame the model for telling a falsely rosy story. You and they will agree the bad model leads to bad decisions – you’ll just agree to disagree on whether we are under or over investing in roads.

So where does that get us? Nobody proposes a technical solution to a better model. Dynamic assignments don’t solve the fundamental problems inherent to STAs. Those will get either get shot to pieces for making trips disappear illegitimately or spill car trips all over local roadways.

Most of the technical people who run these models are very well aware of the tools limitations, share them openly in documentation and in person, encourage cautious interpretation to planners and decision-makers, work hard trying to make them better in their conferences and spare time, and beg for modest sums of money to make incremental advancements. They’ll tell you “all models are wrong” but cheerfully follow that up with “some are useful”. A common refrain about people who use models to make long term predictions, in every industry. I’m not sure what good yelling at these poor souls (or about them) will do.

After all the rocks get thrown by “both sides” and everyone solemnly acknowledges that the model is wrong — what happens next? Answer: You land where you started, with opinions and anecdotes and individual biases. For most people and most decision-makers in America, the popular opinion is still … build more roads.

So if the battle for scare investment dollars comes down to opinions, the mode that most people use for most trips is most likely to be most popular. You’re fighting a noble but uphill battle (alongside many others but certainly not a majority).

You’ll need a better argument than “believe me, the trips will just disappear if we just do nothing.” and you’ll also need a better argument than “people will take the bus instead”. There’s no evidence that transit investments will pull Americans out of their cars effectively. Buses also have induced demand and we aren’t anywhere near the urban scale of Paris or Amsterdam or Buenos Aires or whatever in most places in America to support financially viable investments to make the sort of changes we need to really drive down total VMT and head off continued increases in traffic delays. The next bus project that solves traffic forever is just like the next freeway widening project that solves traffic forever — the first.

Most people don’t believe that traffic won’t get any worse if we don’t build any roads – even those who understand induced demand is real. Because it WILL get worse. People that understand induced demand also understand that many trips (commutes, school, daycare, emergencies, etc) are not discretionary, that population is growing, and people aren’t just going to sit at home if we want a functioning society and economy. They might even know that that old people and handicapped people will soon be zooming around in robo-taxis. The simultaneous increases in mobility and age of population is going to be a double whammy on the demand front.

So with all that demand for travel, the VMT reduction strategies are great in theory but too often they come off as Person Miles Traveled reduction strategies. Acceptable when it’s other people being told to stay at home, but the minute it applies to them and not somebody else there is trouble.

I don’t have a solution to this problem and am just venting here, I admit. But so is this article ultimately. We need a comprehensive and dramatic shift in thinking towards more progressive strategies of first and foremost land use and second for sustainable transportation investments. Carrot and stick for motor vehicles (tolls and reallocation of road space for better alternatives) are absolutely necessary. We need bold funding ideas and bold and brave investment strategies both.

I am very much with you in the desired outcomes but not in the tactics implied here. I very much doubt yelling about models or at modelers is going to help. But maybe I am wrong – perhaps maybe in some places it will. Godspeed.
Joe
9 months ago

“But transportation models using the STA approach unrealistically assume people will blindly keep driving a congested roadway, no matter what is happening or how long their trip will take.” This hasn’t been true for decades. STA runs iteratively towards a theoretical equilibrium where the travel time from an origin to a destination won’t be faster if a traveler chooses a different route based on the level of congestion. There are many algorithms that work towards finding this solution. Some of the most common are described here: https://tfresource.org/topics/Network_assignment.html
Chris Rall
9 months ago

Joe, Good point, and thanks for the link to this resource. The models might optimize the driving route, but the origin and destination remain static, the time of the trip stays static and the mode stays static. Which is why they don’t accurately reflect travel demand in a congested urban environment. For example, if traffic is bad, I’m going to walk to the taco place in my own neighborhood, not drive all the way across town to the place that has slightly better al pastor. Models don’t reflect these kinds of decisions very well, which is why they tend to overestimate traffic Armageddon if a road is removed, or not expanded.
D Wieb
9 months ago

Good example, walking to the local taco place instead of creeping along the freeway. A focus on pedestrian safety and improving neighborhood services would lower VMT and vastly enhance quality of life.