You can follow Chuck’s whole excellent podcast series here.
Induced demand is the observed fact that if you make something easier to do, people will do it more. For example, if you create new capacity for cars in a place where travel demand is high, the result is more cars. If you build more capacity to “fix congestion”, you end up back near the same level of congestion you had before.
After decades of observing this pattern, most people, including many road-building authorities, are still reluctant to face what this means. Part of the problem, surely, is that we’re presenting induced demand as an observed discovery, allowing us to perform quarrels over data, research methods etc.
But induced demand isn’t just an observed fact. It’s also an axiom of biology, so we are as sure about it as we are of the facts of math. This means we don’t really need to be doing this experiment over and over, just as we don’t need to keep measuring circles to be sure of the value of pi.
In this context an axiom is a statement that can be taken as true because it is part of the definition of a concept you are using, or follows logically from that definition. The value of pi is axiomatic because it follows from the definition of a circle, in standard Euclidean space that describes our everyday world.
Now consider the concept of an organism. It implies that:
- The thing consumes some resource from its environment, in order to have enough energy.
- It will expend energy getting this resource.
- Therefore, it must run a positive balance sheet: The energy it spends getting the resource must be less than the energy the resource will provide.
We humans are organisms, so we do what they do. We seek resources in the easiest possible way (i.e. with the lowest possible expenditure of energy.) So:
- If driving suddenly becomes easier than taking transit (due to a road widening, for example), more people will shift to driving, increasing congestion.
- If a road widening makes it possible for developers to save money (i.e. energy) by building in more distant places where land is cheaper, they will do that.
- This process changes the shape of the urban area so that people travel longer distances (due to sprawl) at slower speeds (due to congestion).
- Therefore the average organism will need to expend more energy to reach the same resources it reached before. (Your job flees from downtown to a distant business park where taxes are lower. Your grocery store closes because a WalMart opened two miles away where you can’t walk to it, or even walk from the nearest point that a bus could get to.)
- On average, organisms in this system end up in a weakened state, with a worse balance sheet of energy expended vs energy gained.
The organisms in this parable are all trying to harvest more energy than they spend in the act of harvesting. Even unimaginable aliens on distant planets would do this in the same situation. So it’s axiomatic that, in the absence of other pressures, road widening in a high-demand area will induce more traffic and more sprawl.
So although road-building departments keep doing the induced demand experiment many times every year, and getting the same results, you don’t need to do more experiments, just as you don’t need to keep measuring circles to be sure of the value of pi. You can add complexity by taking this into the human sciences and trying to model subtleties of human behavior, but all the resulting insights will be marginal compared to the axiomatic fact that above all, we’re organisms, so we’ll do what organisms do.
Recently, I asked whether we should build transit infrastructure projects with the goal of expanding access to opportunity, as opposed to existing measures of success that depend on ridership prediction. (If you’re not sure of what I mean by access, read this first.) This stimulated a great conversation in the comments with Alex Karner, an Assistant Professor of Architecture at the University of Texas, Austin, with a useful interjection from Willem Klumpenhouwer, a postdoctoral fellow at the University of Toronto Transportation Research Institute. It’s very lightly edited.
Here is the part of my piece that set this off, followed by our exchange:
[Most Federal criteria for transit funding] are built on the same shaky foundation: a prediction of ridership well into the future. Access analysis may help to shore up those foundations, because an access calculation is much more certain than a prediction.
That should be especially obvious during the Covid-19 pandemic. The utterly unpredicted ridership trends of 2020 are just an extreme example of the kind of unpredictability that we must learn to accept as normal. As I argued in the Journal of Public Transportation, we can’t possibly know with certainty what urban transportation will be like in 10-20 years, or how our cities will function, or what goals and values will animate people’s lives.
Ridership prediction models generally begin with something like an assessment of access. If a project improves travel time for a lot of possible trips, that’s the starting point for a high ridership prediction. But then, predictive modeling mixes in a bunch of emotional factors that amount to assuming that how humans have behaved in the recent past tells us how they will behave in the future. This is equivalent to telling your children that “when you’re my age, I know you’ll behave exactly the way I do.”
Of course some human behavior is predictable. We’ll still need to eat. But the world is changing in non-linear ways, which means that the recent past is becoming less reliable as a guide to the future. So if we measured access, we’d still be measuring ridership potential, but without all the uncertainty that comes from extrapolating about human behavior, or telling people that you know how they’ll behave 20 years from now.
I agree with your criticisms of long-range forecasting but think there’s a substantial role for near-term (current year or opening-day) forecasts when thinking about the impacts of a proposed service change or the inequities inherent in the current system. Because these rely on more timely data about travel behavior, land use, and levels of service, we have more confidence that their results are meaningful. These forecasts can provide information about how travel times (walk, wait, in vehicle, and transfer), number of transfers, and out-of-pocket costs change or differ across people and places. (E.g., the proposed service change will increase origin-destination travel times for low-income transit riders on average by 8 minutes).
To be sure, accessibility (access/freedom) is super important. But we should also look simultaneously at expected impacts on transit riders today. Activists and advocates always request these types of measures when service changes are being proposed.
Access is undoubtedly an important measure, and is definitely under-used as a metric of success or value in many transit system evaluations.
I do have one quibble with your argument, however. Measuring access now and using it for long-term projects into the future is *still* making static assumptions about human behaviour (that the destinations you measure access to are important and will be into the future), as well as assumptions about future land use. While I do think it’s not as explicitly modeled as with a standard econometric behavioral model, it’s still implied.
One way to potentially fuse some of that together and add flexibility is to get into the habit of bundling destinations into a weighted “basket of destinations”, something I argued in a JTG paper. Then the trick becomes figuring out ways to determine what those bundles should be.
Ultimately, I think there needs to be more research done on how access translates into use (of which ridership is one metric), and how that happens.
Alex and Willem
Alex: I agree completely that there’s role for understanding current needs and maybe even near-term forecasting, but the kind of transit infrastructure we’re talking about here has to be useful for decades if not centuries, and the present is just too brief a time to be the the only consideration or even the main one. Since the narcissism of the present will dominate the project politics anyway, I want to highlight measures that push against the present bias, because there’s literally no other way to give our unborn grandchildren a place at the table.
We all live inside our grandparents’ bad infrastructure design decisions — decisions that made sense at the time, in the culture of the time, for the people who were being listened to at the time. You can work to make that present-oriented conversation smarter, more fact-based, and more inclusive/equitable, and I support that 100%, but your approach still leaves us saying that our grandchildren will be just like us in all kinds of ways that we have no right to assume.
How do we know what our grandchildren will want? Two ways: (1) We can assume that they will want what history and biology tell us that humans have always wanted and (2) beyond that, we can focus on giving them the freedom to be whoever they turn out to be and want whatever they turn out to want.
For example, under (1) we have the biological needs that drive a lot of our daily activity, but also historical insights like Marchetti’s Constant, which help us set useful travel time budgets for a daily trip or “commute.” We can do some philosophical work to delineate the boundary of those two categories. This is where I was going with the Bortworld thought experiment in my Journal of Public Transportation paper.
Willem raises a good point about how we can know, on behalf of our grandchildren, what the relative importance of different kinds of destinations will be. Biological and historical knowledge can take us a long way. When it comes to human motivations, the longer something’s been true in the past, the more likely it is to be true in the future. Finally, yes, that weighting will still be a judgment. But if we could get to the point where we were arguing mainly about that, I think we’d have made transformative progress in how we think about infrastructure. More on that here soon.
I think Willem’s point about having to make assumptions about future land use–either how it changes or assuming it stays constant–clarifies that accessibility or freedom analyses are still subject to at least some of the limitations of long-range ridership forecasts.
In terms of infrastructure, I’m fine with having different standards for fixed-guideway projects with high capital costs and longer-lasting impacts on urban form as compared to bus network redesigns or other tweaks to the bus network. Although I wonder how much land use and density is already baked in and how that differs by urban areas. Will the locations of high land use intensity look much different in 50-75 years, in terms of their locations, than they do today? If not, then holding land use constant and also using current/near-term forecasting can both provide important insights about impacts now and in the future. I don’t think we have to pick one set of metrics or approaches over the other. Both are important.
One key area where I think we differ is in our relative weighting of future vs. past impacts. I definitely appreciate your future orientation–this is important from a climate, health, sustainability and resilience perspective. But I think to make traction on these issues and to get residents to buy into any specific public transit vision, we (academics, practitioners) have to acknowledge that transportation infrastructure development (highways *and* transit) has historically had baleful effects on low-income people and people of color in the US. Black people were especially negatively affected throughout the 20th century and continue to bear the brunt of many of the transportation system’s most direct impacts while not sharing fairly in the system’s benefits.
I see looking at impacts on current riders and near-term forecasting as at least partially atoning–or at least acknowledging–these historical impacts. In conducting a current/near-term analysis, we’re saying that we value the experiences of current public transit riders and want to understand how our proposed changes will affect them. To be sure, there’s a lot more than can and needs to be done in this regard (this pending TCRP project will help to suss out exactly what a reparative approach for public transit planning/policy could look like). But jumping straight to the future without acknowledging the past seems like a surefire way to alienate the essential riders upon which public transit depends.
The folks at the Untokening Collective have written about this in their “Principles of Mobility Justice,” one of which is the following: “Mobility Justice demands that we fully excavate, recognize, and reconcile the historical and current injustices experienced by communities — with impacted communities given space and resources to envision and implement planning models and political advocacy on streets and mobility that actively work to address [the] historical and current injustices [they experience].”
Current/near-term forecasting doesn’t live up to this high standard on its own, but providing resources to communities to vision future transportation systems and to understand how their travel outcomes (in terms of performance, not necessarily choices) will differ in those futures might get us moving in the right direction.
We are in complete agreement about the need to show the impacts of proposals on the present, especially relatively short term work like bus network design. That’s what we do in all our projects.
Access analysis honors the future but is also an important way to talk about the present. For example, we can talk about the impact of a service change on the access to opportunity of existing riders based on their boarding location, so that we are specifically addressing the benefits and disbenefits that each such rider will experience. This can help riders see beyond an understandable initial assumption that all change is going to be bad for them.
There’s also a space for access analysis in giving elected officials another way to think about what they are hearing from the public, and to relate a service plan debate to larger goals that they care about, because expanding access supports so many of those goals.
But you’ll have to explain how forecasting serves the goal of “excavating … historical and current injustices.” How does predicting human behavior the near future help us understand the past, or our moral options for rectifying the injustices of the past? That one I just can’t follow.
It seems like we also differ in terms of whether we think access analyses are enough on their own to demonstrate present-day impacts. The analysis you describe based on boarding location sounds helpful. I’m arguing that in addition to evaluating those quantities, we should also look at impacts on *current trips* and *current riders,* summarized by place or for specific groups (e.g., low-income people, Black people, equity-priority neighborhoods, etc.). Access it great, but it does not tell us how people are using the system today and how a proposed change will affect the trips they currently need to undertake.
Our knowledge about the injustices of the past informs the places and groups that we think it important to analyze. A high-quality transit rider survey will capture the travel behavior of a sample of transit riders. These results need to be carefully weighted and expanded to represent all transit travel. This weighted and expanded sample is our best representation of the full range of travel being undertaken on a particular system. Trip characteristics can be modeled to assess how they change from a base (no-build) to a build scenario. These changes represent the real impacts that will be experienced by actual travelers today and can be used to understand differences between groups. If done well, this analysis will provide insight into current injustices, if any exist (e.g., wide disparities between the trip characteristics between places or groups).
If desired, appropriately crafted simulation models can also be used to understand how behavior will change in response to changing levels of service (or demographics or land uses). Model results can also be used to assess current injustices and to help us understand whether we are making progress towards redressing historical wrongs.
We certainly don’t disagree about the value of studying how the system is being used now, and evaluating impacts of changes on existing riders. We see the value in using rider surveys for this purpose, alongside access analyses that show how a network proposal changes what people *could* do (but aren’t doing now because the transit system doesn’t let them).
But I’m still puzzled about how models that predict “how behavior will change” are helpful in understanding or rectifying past injustice — unless you just mean really safe predictions such as “if we make high-demand trips possible that aren’t possible now, people will begin making those trips.” Is that all you mean?
That’s not all that I mean. If we have a “good” rider survey collected recently we can use that survey to estimate a ridership model that will help us understand how changes in level of service and land use will affect transit use. The changes need not be limited to “high-demand trips.”
If we constrain our forecasts to use near-term or current year data, then we will have more confidence in the outputs we’re generating than if we use a 30- or 50-year horizon.
And if we examine outcomes for groups that have historically experienced injustice, our results can speak to how their experience of using public transit will change.
We seem to end where we began, Alex is arguing that to understand people’s experience, we have to predict what they’ll do. And I’m wondering if it’s better to just talk about what they’re free to do. We debate, you decide.
What if we planned public transit with the goal of freedom? Well, it’s hard to improve things that you can’t measure, but now it’s becoming possible to measure freedom, or as we call it in transport planning, access.
Access is your ability to go places so that you can do things. Over the last few years, I’ve come to believe that may be the single most important thing we should be measuring about our transport systems — but that we usually don’t. Access isn’t a new idea, but as our data gets better it’s becoming easier to measure, and it could potentially replace many other measures that are groping toward the idea but not quite getting there.
We calculate access, for anyone anywhere, like this:
Whoever you are, and wherever you are, there’s an area you could get to in an amount of time that’s available in your day. That limit defines a wall around your life. Outside that wall are places you can’t work, places you can’t shop, schools you can’t attend, clubs you can’t belong do, people you can’t hang out with, and a whole world of things you can’t do.
We chose 45 minutes travel time for this example, but of course you can study many travel time budgets suitable for different kinds of trips. A 45 minute travel time one way might be right for commutes. For other kinds of trips, like quick errands or going out to lunch, the travel time budget is less. For a trip you make rarely it might be more.
But the key idea is that we have only so much time. There is a limit to how long we can spend doing anything, and that limit defines a wall. We can draw the map of that wall, and count up the opportunities inside it, and say: This is what someone could do, if they lived here.
Access is a combined impact of land use planning and transport planning. We can expand your access by moving your wall outward (transport) or by putting more useful stuff inside your current wall (land use). We can use the tool to identify how much of a place’s access problem lies in the transport as opposed to the development pattern.
We can calculate access for any location, as in this example, but we can also calculate the average access for the whole population of any area. In the first draft of our bus network redesign for Dublin, Ireland, for example, we found that the average Dubliner could reach 20% more jobs (and other useful destinations) in 30 minutes. To discuss equity, we can also calculate access for any subgroup of the population: low income people, older or younger people, ethnic or racial groups, and so on.
Why Access Matters
People come to public transit with many goals that seem to be in conflict, but it turns out that a lot of different things get better when we make access better:
- Ridership tends to be higher, because access captures the likelihood that any particular person, when they check the travel time for a trip, will find that the transit trip time is reasonable. Ridership goes up and down for all kinds of other reasons, but access captures how network design and operations affect ridership.  In our firm’s bus network redesigns, we’ve been using access as a key measure of success for about five years now, and it consistently leads us to ridership-improving network designs.
- Emissions and congestion benefits all improve, because they depend on ridership, which depends on access.
- Economically, the whole point of a city is to connect people to abundant opportunities. People come together in cities so that more stuff will be inside the wall around their lives. When we measure access we’re measuring how well the city functions at its defining purpose.
- As for equity or racial justice in transit, well, isn’t equal access to opportunity at the core of what these movements are fighting for? Access describes the essence of what has been denied to some groups through exclusionary development planning and exclusionary transport planning, so it helps us quantify what it would mean to fix those things. This, in turn, could help justice struggles avoid a lot of distractions. Because in the end, access is …
- Freedom. Where you can go limits what you can do. If we increase your access, we’ve expanded the options that you have in your life. Isn’t that what freedom is?
When we improve access, with attention to who is benefiting most, we improve all of those things. It’s this remarkable sweep of relevance that makes access analysis so interesting and potentially transformative as a way to think about transportation.
Access Compared to Common Measures
Most methods for studying or improving transit assume that we should care about (a) what people are doing or (b) what people want to do.
Data about what people are doing includes travel behavior data, which are the foundation of much of the accepted methods of transport planning. In public transit, ridership data is in this category. Ridership is the basis for transit’s benefits in the areas of congestion and emissions, and also of fare revenue.
However, what people are doing isn’t necessarily what people want to do, or what they would do if the transport network were better. Much of what people do may just be the least-bad of their options given the city and transport network as it is. This problem leads to various methods of public surveying to “find out what people want,” in some sense. But there are lots of problems with that, mostly lying in the fact that people are not very good at knowing what they’d do if the world were different in some major way.
Access takes us outside of both of those frames. Instead of asking “what do people do?” or “what do people want to do?” it asks “what if we expanded what people can do?”
Access analysis does not try to predict what you’ll do. In fact, it doesn’t need to predict human behavior at all, which is a good thing because human behavior is less predictable than we’d like to think. Access calculations are vastly more certain than almost anything emerging from social science research, because they are based almost entirely on the geometric patterns of transport and development. 
Instead, access starts with one insight about what everybody wants, even if they don’t use the same words to describe it. People want to be free. They want more choices of all kinds so that they can choose what’s best for themselves. Access measures how we deliver those options so that everybody is more free to do whatever they want, and be whoever they are.
What Access Analysis Can’t Do
Will access analysis of transit put the social sciences and market research out of business? Of course not.
- We need to understand how different users experience public transit, and how the experience can be better designed to meet those various needs.
- We need to know exactly who won’t be served by access based network design so that we can decide what actions to take for those people, if any.
- We need to keep exploring the relationship between access and ridership so that we can identify the factors that sit outside that relationship and must be considered.
- Access analysis would also become more powerful if we had better data on the locations — to within 1/4 mile (400m) or so — of various non-work destinations: retail, groceries, medical, and so on — so that we could better assess people’s ability to get to such places.
But in 30 years of listening to public comment, I’ve heard enough times that people want to go places so that they can do things. So let’s measure how well we’re delivering that, and let’s ask ourselves if that’s more important that some of the things we measure now.
This post could have been much longer; in fact, I hope it will become a book. Meanwhile, here are some great resources:
- The 2020 Transport Access Manual is the first comprehensive explanation of access and how it can be applied to various questions. It’s the work of a team led by professors David Levinson (University of Sydney) and David King (Arizona State University). Full disclosure: I had a role and wrote some snippets.
- The University of Minnesota’s Accessibility Observatory, founded by Levinson and now led by Andrew Owen, is one of the main research centers on the topic. For several years they’ve been publishing Access across America, an atlas showing where people can get to from various places by car, transit, etc..
- On the philosophical issues about freedom vs. prediction, and why it’s important to separate physical knowledge from social science knowledge, see my fun Journal of Public Transportation paper, “To Predict with Confidence, Plan for Freedom.” Seriously, it’s fun.
- On what high-access public transit tends to look like, here’s a fairly evergreen 2013 post of mine, with downloadable handout, on how some of the big debates of transit planning line up with a goal of high access for a community.
I will update this post with further links.
 In the academic literature, what I’m calling access is usually called accessibility. Both of these words have contested meanings, because both have been used specifically to refer to the needs and rights of people with disabilities. I follow the recent Transport Access Manual in using access as the less confusing of these two words. Of course, we are talking here specifically about spatial access — the ability to do things that require going places — which is not the only kind. However, a lot of the ways that people are cut off from opportunity do turn out to be spatial. Transportation (i.e. access) is a major barrier to employment in the US, for example.
 This paper, for example, establishes a relationship between transit access and public transit’s mode share, one that is especially strong for lower income people.
 There are exceptions. Traffic congestion, for example, is a human behavior that affects the access calculation.
We’ve all been trained to view the confident prediction as evidence of expertise. The expert commits to a prediction — “Blazers win by two, “Biden wins New Hampshire,” “We’ll all be riding driverless cars by 2019” — and we’re supposed to be impressed. “If he’s so confident, he must know what he’s talking about” we are supposed to think.
He doesn’t. The only statements about the future worth considering are those hedged with uncertainty and margins of error, where certainty is approached gradually through many people studying the facts. That’s the long, slow, misunderstood process by which we got to the consensus on climate change. But most practitioners of that craft don’t call this work prediction. They speak more humbly (and accurately) of projections and scenarios. They tell us that things are moving in a direction, or that some outcomes are more likely than another, or that “if nothing changes” it will look something like this in 2050.
Prediction isn’t humble in this way. Often it’s just a sales pitch: “Buy this product and you will be happy.” “Thanks to our product, public transit will soon be obsolete.” Ignore these claims utterly. They are not trying to make you smarter. As always when you hear any statement about a patented new thing, lean into the wind. The more you detect self-interest behind the prediction, the more you should doubt it.
When I say prediction-like things in my role as an expert, they are of two kinds. Either I am predicting the continued existence of physical facts, (“In 2100, an elephant still won’t fit inside a wineglass”1) or I’m offering if-then statements that point to the listener’s power: “If you do this, it will have this effect”. I’m careful to stay in those bounds, where I’m certain. When journalists ask me “what will cities be like in 2030?” I decline.
Here’s the thing: Prediction — by which I mean any non-trivial assertion about the future — is the opposite of moral thinking, because it implies we are passive receivers of the future instead of creators of it.
Predictions tell us that we will happen anyway if accept the future passively, doing nothing to change it. But all credible, properly hedged projections about that future are dire. So we will act, and our action will disrupt all the models and assumptions and prejudices that make prediction possible.
To feel powerful, then, you must resolve to reject all confident predictions that you hear. Honor the projections and scenarios that reflect decades of humble work. But don’t let anyone tell you they know what the future will be. Nobody knows, and it would be cause for despair if they did.
(A much expanded version of this argument is in my Journal of Public Transportation paper here.)
1 A more relevant insight about urban planning than you might think, as I explain near the beginning of most of my public speeches — this one, for example.
The Journal of Public Transportation has a special issue out consisting of thinkpieces by a range of figures in the business. I’m honored to be there alongside industry leaders like Susan Shaheen of UC Berkeley, Graham Currie of Australia’s Monash University, Kari Watkins of Georgia Tech and Brian Taylor of UCLA, as well as our favorite operations and scheduling consultant, Dan Boyle.
My contribution is called “To Predict with Confidence, Plan for Freedom.” It basically outlines the argument of my next book, so this would be a great time to hear some critiques of it. Here’s the opening:
What will urban transportation be like in 10-20 years? How will automated vehicles interact with social and cultural trends to define the city of tomorrow? Will the vehicles of the future be owned or shared? How will pricing evolve to motivate behavior? What will happen to public mass transit? What other innovations can we expect that will transform the landscape? This paper, which is merely the outline of a larger argument, suggests three interconnected answers.
- We can’t possibly know. History has always been unpredictable, punctuated with shocks, but if the pace of change is accelerating, then unpredictability may be increasing too.
- We can reach many strong conclusions without knowing. A surprising number of facts about transportation, including some fairly counterintuitive insights that would be transformative if widely understood, can be described and justified solidly with little or no empirical ground, because they are matters of geometry and physics or of nearly axiomatic principles of biology.
- Prediction may not be what matters anyway. If we abandoned hope of predicting the future, we could still describe a compelling outcome of transportation investment, one that motivates many people who will never care about a ridership prediction or economic impact analysis. We could also predict it in the sense that we can predict the continued value of pi. That idea is freedom, as transportation expands or reduces it.
So if that catches your interest, read the whole thing, and share your comments below!
For the Congress for the New Urbanism conference in Seattle last month, I tried out a new angle on my usual stump speech. I asked: Can we live without predictions? What would it mean to approach a city planning problem — say, transit planning, which I do — without needing to know the future?
I’m pretty happy with how it came out. It’s embedded below, but it seems to be slightly sharper here.