The Projections Fallacy

Image via Unsplash.

On September 8, Wiley & Sons will release the second book in the Strong Towns series: Confessions of a Recovering Engineer: A Strong Towns Approach to Transportation. This book pulls back the curtain from the lack of rigor inherent in many of the standard practices of the engineering profession. Nowhere is this more apparent than in the use of traffic projections. More accurately, the veneer of technical rigor that traffic projections provide to projects.

I wrote this article back in 2012. I’m sharing it today with only light edits so you can see how I struggled to explain the fallacies involved in traffic projections. This is me working through the insights of Nassim Taleb, who is perhaps the person whose thoughts and writings have had the greatest impact on me. The main conclusion of this article is one I struggled with: “We don't need better projections, we need a system that is robust to modeling error.” What does that mean in practice?

I give an answer in Confessions of a Recovering Engineer, one that my fellow engineers will find challenging, but one that is inescapable when you acknowledge that it is better to plan systems for a world where we can’t predict the future, than to pretend that we can.

And, by the way, the 2012 predictions made by local engineers that I cite in this article were all wrong. In this case, they were a huge overestimate. We never returned to the robust growth of the pre-2008 housing bubble as they anticipated and, of course, they also never anticipated a pandemic. I know if we asked them they will say that they did the best they could with the information that had. That’s a pathetic answer, one that may serve their interests but do nothing to serve the interests of the public.

---

We spend billions every year in this country on our transportation network, large percentages of it based on traffic projections—despite the fact that we have a long record of not being able to accurately project traffic. The answer isn't better projections, but a better transportation system, one that is robust to modeling error.

My home town newspaper recently ran the standard repeat-what-the-engineer-says article on traffic projections. Essentially, the report indicated that we're going to be inundated with traffic. As things continue to "full build out" (it was in quotes so I'm assuming it is an engineering term), traffic is going to increase by 75%, an astounding amount since most locals will attest we are already drowning in traffic (we're not, but most would attest that we are). The recommendation for dealing with all this traffic seems sensible: make some prudent investments today to acquire more land for future road expansion and then, as they are built, oversize the roads to meet this future demand.

A lot of the rationale for these projections, as well as the public's acceptance of them, comes from the fact that growth has been robust. In fact, if you go back decades and look at the projections that were made for the present time, they are laughable in how dramatically they underestimated the amount of traffic. We projected traffic based on what our experience taught us to anticipate, but we were wrong, and it cost the city a lot of money to retrofit all of the places that were inundated with cars.

This reality fits a national trend. My experience is backed up by studies demonstrating that, the higher the functional classification and the larger the traffic volumes, the greater the degree of underestimate. This correlates with work by the Patron Saint of Strong Towns Thinking, Nassim Taleb, who has made the same observations of economic systems, governments, etc. (For one example, go to the 5:10 mark of this video.)

Amazingly, the fact that we have been so consistently wrong doesn't make us any less confident today, either in my hometown or nationwide. We've "enhanced" our models now and believe we have it figured out this time, revising the data upward to reflect what we have experienced in the "real" world. This is the essence of modeling, and what else could be more rational?

Or more foolish. In these models, we've taken something that is unpredictable (driver behavior) and treated it as if it were actuarial science, akin to estimating life expectancy or your odds of drawing a face card when the dealer is showing fifteen. The idea behind our hubris is that, while one driver may be unpredictable, the average driver will react in a predictable way and, thus, we can model based on a normal distribution. These models are failing to account for things like consumer preference, the ability to access financing, overall market growth, cost of construction materials, gas prices, government employment levels, and so on and so on and so on… We assume all drivers make predictable traffic decisions. They don't.

Image via Piqsels.

I'll take my hometown as an example. Our old traffic projection models assumed that cars multiply like people (I guess that's what's going on in all those two-car garages), so we projected traffic based on historical growth rates the same way we projected population. Then what happened? I live in a resort/tourist area with cheap, abundant land and lots of natural resources, just the kind of place retirees and near-retirees wanted to move to, starting roughly in 1990. Fuel that desire with a stock market bubble (I wrote a lot of permits for people who paid cash for their new lake home after selling stocks) and then with a cheap credit housing bubble, and—unexpectedly—we are inundated with traffic.

Today we already have Super Walmart, Super Target, Home Depot, Menards, Fleet Farm, Kohls, JC Penny's, Best Buy, Costco, and a myriad of chain restaurants, gas stations, and other highway parasites sucking off of the hundreds of millions we've invested in the adjacent high-capacity highway system stroad (street/road hybrid). The baby boomers are stuck in their existing homes, which are now worth much less than they had hoped, while many are also stuck in their job as their retirement savings suffers the same fate. The cheap shoreline is gone, sold off decades ago to people who are now reaching the age where maintaining a lake home may not be worth the effort. The cheap gas is gone, too, and it doesn't seem too likely that the state is going to throw hundreds of millions more into shortening travel times from the Twin Cities.

Has any of this new reality informed our projections? Of course not. Like a mad scientist adding random chemicals to a potion they just don't understand (and then testing it on unwilling human guinea pigs), we've tinkered with our models and convinced ourselves that this time will be different. And you can bet that there is no way we'll be caught underestimating. We've "seen" what happens and so we've "fixed" that problem. The new answer is simple: estimate high (which is, in the perverse vernacular of the trade, called being "conservative" with the estimate).

We’ve had many discussions in the Strong Towns network about traffic projections, including the following comment from one of my friends: 

We're working on a traffic study for a small development. According to their comp plan (from 2007) we need to offer recommendations for a county road equivalent assuming 4 percent annual growth. We looked up the numbers and found that VMT has (of course) declined steadily in the area since 2003/4. Someone at our firm contacted the county about the growth rate (a 26 yr old traffic engineer by the way) and, from the message that was relayed back to me, we were instructed to just proceed with the Comp Plan's assumptions. Long story short, we're probably going to have to recommend a turning lane. 

In another conversation, one of our members from California was knee deep in the debate over widening I-710 to 14 lanes. In a note to the spokesman for the Los Angeles County Metropolitan Transit Authority, he asked:

I know that the analysis of vehicle trips and container traffic for this project were done prior to the global financial system collapse in 2008, and so the idea of traffic volumes or container shipment volumes decreasing by 2035 were not considered.

 However, when the Panama Canal is open, it looks like our region will be receiving less cargo simply because going through Panama is much cheaper than shipping to LA or Long Beach and moving via trucks, trains, etc.

 Is anyone at Metro thinking about this project differently now? Shouldn't the "No Build" alternative for this project (and all the alternatives) be revised?

Here was the answer he received:

The Ports re-evaluated their growth projections after the recession and made adjustments. The revised projections show slower growth in Port activity in the next decade, but the 2035 total TEU estimate remains the same: 42.7 Million. In other words, the Ports anticipate to handle 42.7 Million TEUs by 2035, even if growth is slower in the next few years.

In other words, it might slow down for a while, but we're confident it will go back up ultimately because our projections say it will.

There is a certain contingent out there that is already pounding out their email to me demanding that, if I'm so smart, I provide a better way of estimating. There's the fatal flaw in our current system. I'm not so smart, but neither is anybody else. The big difference here is that I'm not pretending to be able to predict the future.

Inherent within the auto-centric pattern of development that defines the Suburban Experiment is the hierarchical road network. Like a river swelling during a steady rain, changes on the periphery have an enormous impact on the trunk components of the network, as do many other things that have nothing to do with driver behavior. The reason we put so much time and effort into projections that we know will be wrong—that we have no consistent history of getting right—is because the cost of being wrong is so great. When the projection is off by even 10%, the level of service on major roadways can plummet. Since nearly every trip is funneled into this network, failure is catastrophic. This is an incredibly fragile approach.

We don't need better projections, we need a system that is robust to modeling error. We need a system of growth and development where we don't need to project correctly in order to succeed. We need a system where we build incrementally in a replicable and evolving pattern, one where each fractal evolves continually and naturally over time. We need a system where we're not required to place huge bets on the future, oversizing infrastructure in service of projections, but instead can invest in high return endeavors where the likelihood of success is great.

We need a Strong Towns approach which, if you stop and look, is a lot like the pre-automobile approach that served us well for thousands of years. I'm not saying we get rid of the automobile, but when we build our entire environment around its propagation, we are slave to our own hubris and lack of understanding.

If you are interested in learning more about how we can comfortably live in a world where we don’t pretend to predict the future, we have created some pre-order bonuses for people who purchase Confessions of a Recovering Engineer before September 8. This includes “30 Days of Confessions,” a video series we have now begun rolling out to those who have pre-ordered the book. You can also immediately check out Aligning Transportation with a Strong Towns Approach at Strong Towns Academy, or the shorter Local-Motive session on “Establishing a Street Design Team.“

---

---