How good are we at prediction?
Transportation planning is full of
projections -- a
euphemism meaning
predictions. Generally, when we need a euphemism, it means we may be accommodating a bit of denial about something.
Predicting the future, at a time when so many things seem to be changing in nonlinear ways, is a pretty audacious thing to do. There are professions whose job it is to do this, and we pay them a lot to give us predictions that sound like facts. I have the highest respect for them (all the more because what they do is nearly impossible) but only when they speak in ways that honor the limitations of their tools.
Good transportation planning does this. at the very least, it talks about future scenarios rather than predictions, often carrying multiple scenarios of how the future could vary. Scenarios are still predictions, though; they're just hedged predictions, where we place several bets in hope that one will be right.
I will never forget the first time that I presented a proposed transit plan and was told: "that's an interesting idea; we'll have to see how it performs." The speaker didn't mean "let's implement it and see what happens." He meant, "let's see what our predictive model says." You know you're inside a silo when people talk about prediction algorithms as though they
are the outcome, not just a prediction of the outcome that is only as good as the assumptions on which it's built.
What's more, we seem to be
really bad at predicting curves, or even acknowledging them as they happen.
Something really important happened in the US around 2004, which experts call the "
VMT Inflection." Vehicle Miles Traveled in the US -- the total volume driving -- departed from a linear growth path that it had followed for decades, and went flat. Here's the same curve looking further back. Around 2003, you could be forgiven for thinking that this steady slope was something we could count on.
![](http://urbanist.typepad.com/.a/6a00d83454714d69e201a3fd31e91c970b-800wi)
(At this point an ecologist or economist will point out that the VMT inflection shouldn't have been a surprise at all. This graph looks like what a lot of systems do when their growth runs into a capacity or resource limit. The VMT inflection is a crowdsourced signal that the single-occupant car is hitting a limit of that kind.)
So reality changed, but the Federal projections didn't. Even as late as 2008, when the new horizontal path had been going for four years, Federal projections claimed that the growth in driving would immediately return to the previous fast-rising slope. Again:
![](http://urbanist.typepad.com/.a/6a00d83454714d69e201a73dec3648970d-500wi)
This isn't prediction or projection. This is denial.
All predictions rest on the assumption that the future is like the past. Professional modelers assume their predictive algorithms are accurate if they accurately predict past or current events -- a process called
calibration. This means that all such prediction rests on a bedrock idea that human behavior in the future, and the background conditions against which decisions are made, will all be pretty much unchanged, except for the variables that are under study.
In other words, as I like to say to Millennials:
the foundation of orthodox transportation planning is our certainty that when you're the same age as your parents are now, you'll behave exactly the way they do.
We describe historical periods as "dark" or "static" when that assumption is true. Everyone acts like their parents did, so nothing ever seems to change except the accidents of war and the name of the king or pharaoh .
Historical progress arises from people making different choices than their parents did, and there seems to be a lot of this happening now.
What we urgently need, in this business, are predictions that try to quantify how the future is not like the past; for example, by studying Millennial behavior and preferences and exploring what can reasonably be asserted about a world in which Millennials are in their 50s and are in the position to define what is normal, just as their parents and grandparents do today.
We already know that the future is curved. (With rare exceptions like the growth of VMT from 1970 to 2004, the past has been curvy as well). Millennials are not like their parents were at the same age. There will be major unpredictable shocks. There are many possible valid predictions for such a future. The one that we can be sure is wrong is the straight line.
My work on
Abundant Access -- part of the emerging world of
accessibility studies -- is precisely about providing a different way to talk about transportation outcomes that people can believe in and care about. It means carefully distinguishing facts from predictions, and valuing things that people have always cared about -- like getting places on time and having the freedom to go many places -- from human tastes that change more rapidly -- such as preferences and attitudes about transit technologies. It's a
Socratic process of gently challenging assumptions. Ultimately, it's part of the emerging science of
resilience thinking, extending that ecological metaphor to human societies. It posits that while the future can't be predicted there are still ways of acting rightly in the face of the range of likely possibilities.
Imagine planning without projections. What would that look like? How would we begin?