Claim: Induced Demand is a Myth

TheNZThrower

Active Member
According to a Cato Institute article by Randal O'Toole, the effects of induced demand are exaggerated and insignificant.

He beings by claiming that a Wired magazine article he is responding to misrepresents a 2009 study by economists Turner and Duranton - published in the peer-reviewed journal American Economic Review - they cited.
“Building bigger roads actually makes traffic worse,” asserts Wired magazine. “The reason you’re stuck in traffic isn’t all these jerks around you who don’t know how to drive,” says writer Adam Mann...

In support of the induced‐demand claim, Mann cites research by economists Matthew Turner of the University of Toronto and Gilles Duranton of the University of Pennsylvania. “We found that there’s this perfect one‐to‐one relationship,” Mann quotes Turner as saying.

However, this is simply not true. Nor is it what Duranton & Turner’s paper actually said. The paper compared daily kilometers of interstate highway driving with lane kilometers of interstates in the urbanized portions of 228 metropolitan areas. In the average metropolitan area, it found that between 1983 and 1993 lane miles grew by 32 percent while driving grew by 77 percent. Between 1993 and 2003, lane miles grew by 18 percent, and driving grew by 46 percent...

The paper also calculated the elasticities of driving in relationship to lane kilometers. An elasticity of 2 would mean a 10 percent increase in lane miles would correspond with a 20 percent growth in driving; an elasticity of 1 would mean that lane miles and driving would track closely together. The paper found that elasticities were very close to 1 with standard errors of around 0.05. Even though this is contradicted by the previously cited data showing that driving grew much faster than lane miles, this is the source of Turner’s “perfect one‐to‐one relationship.”
However, according to the abstract, there is a 1:1 relationship:
We investigate the effect of lane kilometers of roads on vehicle-kilometers traveled (VKT) in US cities. VKT increases proportionately to roadway lane kilometers for interstate highways and probably slightly less rapidly for other types of roads. The sources for this extra VKT are increases in driving by current residents, increases in commercial traffic, and migration. Increasing lane kilometers for one type of road diverts little traffic from other types of road. We find no evidence that the provision of public transportation affects VKT. We conclude that increased provision of roads or public transit is unlikely to relieve congestion.
And Randal is likely referring to this table for his figures:
Screen Shot 2022-09-28 at 7.59.51 pm.png

The increase in VKT and lane km on the Interstate Highway (AKA: IH) is as follows:

Change in IH VKT 1983-1993Change in IH VKT 1993-2003
4,128,000 (7,627,000)4,056,000 (7,328,000)
Change in IH lane km 1983-1993Change in IH lane km 1993-2003
68 (79)72 (129)

If we adjust those figures respectively for the percentage of VKT and lane km that's urbanised, we get 2,955,260 (6,317,120) km for 1983, 5,238,200 km (10,670,440) for 1993, and 7,661,000 (15,157,920) km for 2003; and 330.6 (478.5) km for 1983, 434.8 (622.4) km for 1993, and 512 (743.2) km for 2003. Now we can calculate the change in VKT in urbanised areas:

Change in IH VKT 1983-1993 (Urban)
Change in IH VKT 1993-2003 (Urban)
2,282,940 (4,353,320)2,422,800 (4,487,480)
Change in IH lane km (Urban) 1983-1993Change in IH lane km (Urban) 1993-2003
104.2 (143.9)77.2 (120.8)

Now we can calculate the lane km growth percentage vs VKT growth percentage:

Change in IH VKT % [urban] 1983-1993
Change in IH VKT % [urban] 1993-2003
53% (46%) [77% (69%)]34% (30%) [46% (42%)]
Change in IH lane km % [urban] 1983-1993Change in IH lane km % [urban] 1993-2003
4.1% (4.7%) [31% (30%)]5.9% (7.8%) [18% (19%)]

So Randal's initial statement that:
In the average metropolitan area, it found that between 1983 and 1993 lane miles grew by 32 percent while driving grew by 77 percent. Between 1993 and 2003, lane miles grew by 18 percent, and driving grew by 46 percent.
Seems to be true based on the aforementioned figures derived from Giles & Duranton's table. This seems to contradict the 1:1 relationship suggested in the abstract:
VKT increases proportionately to roadway lane kilometers for interstate highways and probably slightly less rapidly for other types of roads.
So are Giles & Duranton that idiotic or dishonest? Or did Randal miss out on some aspect of their paper, some extra data that might make sense of those figures, intentionally or not?

In fact, Randal did leave out the following data from Giles & Duranton:
Table 2 reports estimates of the elasticity of MSA VKT to lane kilometers from univariate OLS regressions. Each panel considers a different type of road: MSA interstates in panel A, urbanized MSA interstates in panel B, major urban roads in panel C, and nonurban MSA interstates in panel D. Columns 1 to 3 consider the 1983, 1993, and 2003 cross sections in turn. Depending on the decade, the elasticity of MSA interstate highway VKT with respect to lane kilometers is between 1.23 and 1.25. Focusing only on interstate highways in the urbanized part of MSAs yields similar results. For major urban roads and nonurban MSA interstates, we obtain slightly lower estimates between 1.00 and 1.14.
Here is Table 2:
Screen Shot 2022-09-30 at 10.38.53 pm.png
So evidently Randal used a childishly simplistic comparison between the % change in VKT & lane km to determine the elasticity between the two variables, when an Ordinary Least Squares (OLS) regression is more appropriate in this context, and gives much different results.

Now I don't know what OLS is, and the math in Giles & Duranton's paper is way above my head, but I think that two economists trained in their field probably know quite a bit more than is required to be capable of such a brazen contradiction that Randal alleges. Ergo, if you spot a contradiction - real or not - in a peer reviewed paper written by experts in the relevant field, chances are they would have done the same, as it is highly unusual for them to miss such a simple contradiction.
 
So evidently Randal used a childishly simplistic comparison between the % change in VKT & lane km to determine the elasticity between the two variables, when an Ordinary Least Squares (OLS) regression is more appropriate in this context, and gives much different results
yeah, if you ignore the method a researcher uses to arrive at a result, your crititism is likely to be meaningless

According to a Cato Institute article by Randal O'Toole, the effects of induced demand are exaggerated and insignificant.
From your excerpts, an elasticity greater 1.0 means more lanes generate disproportionally more traffic, i.e. traffic get worse than by 1:1 when you build more roads? have I got that right? so why does O'Toole then claim 1:1 is "exaggerated" when it's an understatement?
 
Last edited:
Side note:

From the abstract: "We find no evidence that the provision of public transportation affects VKT."

Germany had local public transport essentially free from June to August 2022, and preliminary data shows a congestion reduction of 20-30% in some areas. I believe this topic warrants further study.
 
This is the abstract:

We investigate the effect of lane kilometers of roads on vehicle-kilometers traveled (VKT) in US cities. VKT increases proportionately to roadway lane kilometers for interstate highways and probably slightly less rapidly for other types of roads. The sources for this extra VKT are increases in driving by current residents, increases in commercial traffic, and migration. Increasing lane kilometers for one type of road diverts little traffic from other types of road. We find no evidence that the provision of public transportation affects VKT. We conclude that increased provision of roads or public transit is unlikely to relieve congestion.
Content from External Source
I'm not that bright, but having lived my life in California, it seems that the supply of lanes often lags the demand for the lanes. I wonder if this is accounted for in the study.

In other words by the time roadway lane kilometers are added the demand for vehicle-kilometers traveled is already built in. People aren't necessarily then driving more, they were already trying to drive from one point to another, they just have more lanes to do it on.

Last spring on a trip, it took me 1 1/2 hours to go ~15 miles on I210 in Pasadena. Likewise for the vehicles around me. We were moving at around 10 MPH. So would that be calculated as VKT in that time period for me and the cars around me? Say it was me and 9 other cars, so 10. We each traveled ~ 24 km in that 1 1/2 hour, or 240 VKT.

Now if there were more lanes and we could travel the speed limit of 65 mph I would have traveled ~95 miles in that same amount of time as would the other 9 cars around me. So we each traveled 150 km in that amount of time, or 1500 VKT. It went from 240VKT to 1500VKT with he same amount of cars on the road, just more lanes.

Maybe I confused myself.
 
To continue with Randal's article, he claims that Duranton and Turner's definition of an urbanised area is flawed:
One source of error in Duranton & Turner’s paper may be in their definition of urbanized area. They first collected data for metropolitan areas (which include entire counties that contain urban areas), then estimated the share of driving within those metropolitan areas that takes place in the urbanized portions. Their estimate that well under half of all driving is in the urbanized areas does not seem credible considering the non‐urbanzed portions are, by definition, rural and house few people and businesses. By comparison, the data I used are based on state highway bureau estimates of driving and lane miles within urbanized areas.
However, Randal's summary of Turner and Duranton's methodology is oversimplistic enough to eliminate any nuance as to how they actually arrived at their figures, and he does nothing more than to dismiss their data on the amount of VKT in urban areas without any adequate counter. Even assuming his data is valid and refers to lane km and VKT in urban areas, it does not quantify the amount of driving in urban vs non-urban areas, nor does Randal point out any discrepancies between his data on the VKT and lane km in urban areas and those of Turner and Duranton, and argue why his data is therefore more valid.

But anyways, Turner and Duranton didn't just ''estimate'' the share of VKT in urban areas. This is demonstrated by their calculation of the percentage of VKT in urbanised areas is as follows:
We take the (consolidated) MSA [metropolitan state area] drawn to 1999 boundaries as our unit of observation. Since each MSA aggregates one or more counties, MSA boundaries often encompass much land that is not “urban” in the common sense of the word. MSAs are generally organized around one or more “urbanized areas,” however, which make up the core(s) of the MSA and typically occupy only a fraction of an MSA’s land area. By using data collected at the level of “urbanized areas” we can distinguish more from less densely developed parts of each metropolitan area.

To measure each MSA’s stock of interstate highways and traffic, we use the US HPMS [Highway Performance Monitoring System) “universe” and “sample” data for 1983, 1993, and 2003... For each year, for the entire universe of the interstate highway system within their boundaries, states must report the length, number of lanes, and the number of vehicles per lane per day passing any point. This last quantity is referred to as the average annual daily traffic (AADT). We use a county identifier to match every segment of interstate highway to an MSA. We then calculate lane kilometers, VKT, and AADT per lane km for interstate highways within each MSA.

In the sample data states report the same information (and more) for every segment of interstate highway within urbanized areas. By merging the sample with the universe data we distinguish urban from non-urban interstates within MSAs.

The sample data also report information about a sample of other roads within urbanized areas. This sample is intended to represent all major roads in urbanized areas within the state. From the sample data we calculate road length, location, AADT, and share of truck traffic for all major roads in the urbanized area. The HPMS sample data also assign each segment to one of six functional classes, described in DOT (1989). One of these classes is “interstate highway.” We group four of the remaining five classes; “collector,” “minor arterial,” “principal arterial,” and “other highway” into a measure of major urban roads, omitting the last class, “local roads.” Our definition of “major urban road” thus includes all nonlocal roads that are not interstate highways. Within urbanized areas, interstates represent about 1.5 percent of all road kilometers and 24 percent of VKT, while major urban roads represent 27 percent of road kilometers and another 62 percent of VKT (DOT 2005a).
Giles & Duranton have some supplementary info on their methodology:
The HPMS consists of two parts. The universe data are supplied for most road segments in the interstate highway system and some other major roads, and provide a description of each segment. The sample data provide additional information about all segments in the universe data, including an urbanized area code for segments falling in urbanized areas. For a sample of smaller urbanized area roads, the sample data also provide all data fields that occur in the universe and sample data.

In general, each segment reported in the HPMS represents a larger set of similar segments (typically of the same road), called a sample... For urbanized-area roads in the relevant classes, reporting rules require that the union of all samples be the set of all urbanized-area roads. Loosely, urbanized area road segments are partitioned into sets of similar segments, and one segment from each set is reported in the HPMS sample data. In this sense, sample data represents all urbanized road segments subject to reporting requirements.

For the interstate highway system, the HPMS records number of lanes, length, AADT, and county. By construction, road segments do not cross county borders. For segments in urbanized areas, the HPMS also provides an urbanized area code. Since MSAs are county-based units, these data allow us to calculate VKT for the urbanized and nonurbanized area interstate systems by MSA.

Now what are Turner and Duranton saying?
  1. They're taking MSA traffic data from the US HPMS, divided into universe data (the data encompassing all interstate highway lanes within all given MSA boundaries) and the sample data (the supplemental data on the universe interstate segments)
  2. They then merge the HPMS sample data with the universe data of all the MSAs surveyed to obtain the amount of urbanised Interstates; or they take the urbanised area code in the sample data, and then calculate the share of total interstate traffic in the MSA within those urban areas.
Hence what they're doing is far from some arbitrary estimation that O'Toole suggests. Rather, it is a methodical calculation of the share of VKT in urban areas as distinguished by their code. O'Toole is either dishonest, or puerile in his understanding of the complex econometrics used to arrive at the conclusion suggested by Turner and Duranton.

I have used ellipses to remove some of the more superfluous terminology from quotes from Turner and Duranton, so if you think I'm omitting important context, please point it out to me.
 
To try to simplify this, let me parse out the original post. I think you’re asking for a debunk on the statement “Induced demand is a myth”.

As evidence, there is an article by O’Toole, critiquing an article by Mann, and also critiquing a study by Turner & Duranton (TD). The article by O’Toole is linked and quoted and the TD study is linked and quoted as well. Unfortunately, the Mann article is not linked but Google let me find 2 Mann articles on induced demand. I read the one from 2016, but that’s the wrong article as O’Toole’s article is written in 2014. Mann’s 2014 article is paywalled for me.

O’Tool’s main objection seems to be that Mann is mischaracterizing the TD paper. There are issues with the O’Toole article, like O’Toole quoting Mann quoting Turner, but his first argument is the 1:1 ratio mistake, now citing the TD paper. His argument in regards to looking at change in lane KM over change in KVT vs. calculated elasticity has value. Elasticity is used by economists to simplify complex relationships and OLS may be a step above a simple ratio, but is simple itself. Statisticians look for correlation coefficients and other statistical validations to understand the quality of the output and compare those in studies. No idea what O’Toole’s background is, but I’ve seen similar arguments between math and numbers in other sciences including pharma and meteorology.

The topic of using the questionable math to support a position is a pet peeve of mine.

My summary is that O’Toole presents criticisms which are potentially valid. Too bad he wasn’t selected for the peer review process. I have no idea what Mann really said, so cannot render an opinion there, but it sounds like he was supportive of the TD claim.I read another of Mann’s articles on the same topic and found him somewhat biased in several areas but generally supportive of Induced Demand.

I read quite a bit of the TD study and their conclusions were very clear in claiming Induced Demand is real. There may be other ways to slice and dice their numbers, which may be valid, but it doesn’t change their conclusions.

Based on the evidence presented, I have to conclude that Induced Demand is indeed real and the claim in the thread title is debunked.

Edit: Should not use IOS for long posts.
 
Last edited:
To continue, Randal proclaims that induced demand is good:
But let’s say it were true that there is a perfect one‐to‐one relationship between driving and highway capacity, and further stipulate that increasing capacity leads to increased driving (rather than the other way around, which is equally credible if highway engineers are trying to keep up with demand). Is that a bad thing?

We know that every car on the road has someone in it who is going somewhere that is important to them. Increasing the number of cars on the road means more people are getting to do things that are important to them. Provided we aren’t subsidizing that travel... then increasing highway capacity leads to net economic benefits because it generates travel that wouldn’t have taken place otherwise.

By comparison, building expensive transit systems aimed at getting people out of their less‐expensive cars generates zero economic benefits if it generates no new travel. Only new travel generates economic benefits, so people who argue that new roads induce new travel are actually arguing that new roads create economic benefits.
Randal misses the point behind the critiques of induced demand. The reason why induced demand isn't good in relation to congestion is because it doesn't do an effective job of solving it in the long run. It only relieves congestion in the short term while increasing it in the long term as the reduced time cost of driving will encourage not only more trips in total to be made especially during peak time, but it will also encourage transferring from alternative modes (e.g. transit) into driving, and change land use patterns by encouraging everyone to live further away from their jobs and errand locations, while encouraging businesses to also utilise the roads to conduct freight hauling operations more often. This is explained in the Wired article Randal responded to:
Intuitively, I would expect the opposite: that expanding a road network works like replacing a small pipe with a bigger one, allowing the water (or cars) to flow better. Instead, it's like the larger pipe is drawing more water into itself...

As it turns out, we humans love moving around. And if you expand people’s ability to travel, they will do it more, living farther away from where they work and therefore being forced to drive into town. Making driving easier also means that people take more trips in the car than they otherwise would. Finally, businesses that rely on roads will swoop into cities with many of them, bringing trucking and shipments. The problem is that all these things together erode any extra capacity you’ve built into your street network, meaning traffic levels stay pretty much constant. As long as driving on the roads remains easy and cheap, people have an almost unlimited desire to use them.
To use one example asides from Turner and Duranton, City Observatory noted that even though the travel times after the Katy Freeway expansion in 2008 improved for a short period of time, it was cancelled out by the subsequent increase on the route from Pine Oak to Downtown Houston:
congestion on the Katy has actually gotten worse since its expansion.

Sure, right after the project opened, travel times at rush hour declined, and the AHUA [American Highway Users Alliance] cites a three-year old article in the Houston Chronicle as evidence that the $2.8 billion investment paid off. But it hasn’t been 2012 for a while, so we were curious about what had happened since then. Why didn’t the AHUA find more recent data?

Well, because it turns out that more recent data turns their “success story” on its head.
Dashboard 1.png
Now this is only one route in one city, but Turner and Duranton's data as mentioned before backs this up.

As induced demand can and does increase the total number of cars on the road through generating extra trips made via the reductions in time cost and redistributing trips made by other modes (e.g. transit), it will increase congestion as well as the negative externalities inherent to car travel such as air pollution and CO2 emissions. Randal's point about inducing demand increasing the economic benefits of people travelling to destinations that generate economic growth such as workplaces or stores is moot when you understand that induced demand applies to transit and other modes as well. This means that as transit can carry more people per lane and even more people in total than cars while using less energy and producing less pollution and emissions per passenger, it can replicate those benefits when combined with walkable city districts.
 
To continue, Randal proclaims that induced demand is good
two side notes:
• "Provided we aren’t subsidizing that travel..."—but we do, financially, in terms of population health, and ecologic impact
• "By comparison, building expensive transit systems aimed at getting people out of their less‐expensive cars generates zero economic benefits if it generates no new travel. "—another outcome of Germany's 3-month "9 euro ticket" for public transport this summer is that it generated a lot of new travel (and congested railway cars).

I agree with the argument that more roads means that people organize their lives differently, not that they do "more"; and that organising lives with less car travel in it is more sustainable and may be better for society.
 
To use one example asides from Turner and Duranton, City Observatory noted that even though the travel times after the Katy Freeway expansion in 2008 improved for a short period of time, it was cancelled out by the subsequent increase on the route from Pine Oak to Downtown Houston:

Do any of these control for population growth? In your example the highway was expanded in 2008 and by 2014 travel times were worse, presumably because of induced demand. But the population of the Huston metro area also increased. From 2007, just before the project was completed to 2014, by just under a million. I would think some of those people are using the highway. It's up another million since then and I would imagine the travel time is up also. Because there were more lanes added in 2008? Or there is just a lot more people?

1665160715657.png
EDIT: Forgot my source. https://www.macrotrends.net/cities/23014/houston/population
 
Last edited:
Do any of these control for population growth? In your example the highway was expanded in 2008 and by 2014 travel times were worse, presumably because of induced demand. But the population of the Huston metro area also increased. From 2007, just before the project was completed to 2014, by just under a million. I would think some of those people are using the highway. It's up another million since then and I would imagine the travel time is up also. Because there were more lanes added in 2008? Or there is just a lot more people?

1665160715657.png
EDIT: Forgot my source. https://www.macrotrends.net/cities/23014/houston/population
That's a good point. Honestly I was in too much of a rush to carefully consider this variable. My bad.
 
That's a good point. Honestly I was in too much of a rush to carefully consider this variable.
It was just my first reaction once I understood what they were getting at with "induced demand". In California, it seems lane construction appears to be far behind the demand. But, that's just my reaction, not anything from a study or published paper.

On first thought, I think they would have controlled for population and if they didn't the Cato guy would have pointed it out. But maybe they were as guilty as Cato in pushing their own argument. Cato is very "libertarian" and is going to attack, maybe inaccurately, what they see as "government encouraging people to live in a particular way". In this case they view the concept of induced demand as being used to have the government encourage people to live a high density urban life style.

Likewise the induced demand folks see building more lanes as encouraging people to live a suburban life style, something I think they oppose. So maybe to make their argument look better, they ignored or just forgot population growth. I guess we'd have to really break down the study to see.
 
Induced demand is just simple gradient-following, and I suspect people think it's easy to argue against because it leads to apparent paradoxes like Braess's paradox (technically they're not logical paradoxes, but merely unexpected conclusions that seem to contradict the wrong thing that you would expect from naive expectations - and I think that's a valid use of the term "paradox").

And, given that we're simple gradient followers, that should be the default expectation. Anything that deviates from that should be considered the strange thing and require a high power test and convincing evidence.
 
On first thought, I think they would have controlled for population and if they didn't the Cato guy would have pointed it out.
You assumed correctly. See tables 3 and after. Note the ln(population) in these tables. They tested several different specifications of their model to see how their elasticity measurements change. The models all generally agree with the elasticity around 1.
 
You assumed correctly.
That happens occasionally.

They tested several different specifications of their model to see how their elasticity measurements change. The models all generally agree with the elasticity around 1.
I'm not an economist so I looked up "elasticity":

In differential calculus, elasticity is a tool for measuring the responsiveness of one variable to changes in another causative variable. Elasticity can be quantified as the ratio of the percentage change in one variable to the percentage change in another variable when the latter variable has a causal influence on the former and all other conditions remain the same. For example, the factors that determine consumers' choice of goods mentioned in consumer theory include the price of the goods, the consumer's disposable budget for such goods, and the substitutes of the goods.[2]

An elastic variable (with an absolute elasticity value greater than 1) responds more than proportionally to changes in other variables. A unit elastic variable (with an absolute elasticity value equal to 1) responds proportionally to changes in other variables.
Content from External Source
So, if the government of Huston builds more lanes, more people will drive. The variable KL (kilometers of lanes) is causative to the variable VKD (vehicle kilometers driven).

Is the opposite then also true? If Huston had not built additional lanes, fewer people would have been driving. Despite population growth?

Seems counter-intuitive. Some might have stopped driving, but most would have just spent a lot more time sitting in traffic I'm thinking. But I could be wrong.
 
Is the opposite then also true? If Huston had not built additional lanes, fewer people would have been driving. Despite population growth?
No.

In your quoted example, if you lower the price, more people will buy.
But that doesn't tell you anything about what happens when people have more disposable income—it's not "the opposite", it's a different factor influencing the variable.

If you have f(x,y)=x*y, then the elasticity of f is 1 with regard to both x and y. That doesn't mean f stops changing with y when you hold x constant.
 
Back
Top