Vanishing Automobile update #14

Do More Highways Cause Congestion?


Highway data show that building new freeways increases per capita freeway driving. However, it does not increase total per capita driving. Instead, it shifts driving from ordinary streets to the freeways. Since freeways are safer, and ordinary street driving is particularly dangerous for pedestrians, new freeways are the ultimate pedestrian-friendly design.

This update also shows that reductions in density ("sprawl") has no effect on driving. Nor does increased congestion, suggesting that planners' hopes that more congestion will lead people to shift from driving to public transit are unfounded.


If we've heard it once, we've heard it a hundred times: There is no point in building new road capacity because added capacity simply leads people to drive more. We can find one test of the truth of this myth in the data gathered by the Texas Transportation Institute for their annual mobility reports.

The institute's 2001 raw data cover 68 urban areas and include the following information for each year from 1982 through 1999:
* The number of lane miles of freeway;
* The number of lane miles of other arterials;
* The total number of miles of roads;
* The number of miles driven on freeways;
* The number of miles driven on other arterials;
* The total number of miles driven on all roads;
* Population; and
* Density.

Comparing Data

If building more roads leads people to drive more, then per capita driving will increase faster in urban areas that rapidly expand their road system than in urban areas that build few new roads. This can be tested with a simple statistical calculation known as the correlation coefficient, or r-squared.

The r-squared of two sets of data is a number between zero and one. If it is one, then the two data sets perfectly correlate with one another. For example, if one data set were 1, 2, 3, 4, and another were 2, 4, 6, 8, then they would perfectly correlate and the r-squared would be 1.0. 1, 2, 3, 4 is also perfectly correlated with 8, 6, 4, 2. But if the second data set were 4, 8, 6, 2, they would not be well correlated. and the r-squared would be close to zero.

Using my spreadsheet's random number function for two sets of 68 numbers, I get r-squareds from less than 0.001 to 0.066. So any r-squareds in the Texas data sets that are less than 0.066 are no better than random.

Correlation, of course, does not imply causation. If data set A correlates with data set B, it could mean that A causes B, B causes A, or both A and B are influenced by some other factor C. The way to ferret out causation is to compare lots of data sets measuring many different factors, as well as applying a little common sense.

The raw data used by the Texas Transportation Institute for major urban areas are supplied by individual state transportation offices to the Federal Highway Administration. For smaller urban areas, the institute went directly to the states. Only some states cooperated, which is why most of the institute's smaller urban areas are in Texas, Oregon, and a few other states.

How reliable are these data? The states know to the hundredth of a mile how many roads they have, so road miles and lane miles should be pretty reliable. Miles of driving (vehicle miles traveled, or VMTs) is not quite so reliable, but the states monitor traffic at scores of locations in every urban area so they should have a good idea about trends. Population data are estimated each year by the Census Bureau and while the estimates aren't perfect they are at least as good as miles traveled.

Perhaps the most questionable data provided by the states is the land area of each urban area. Some states update this every year, so that Atlanta's land area increases by about twenty square miles each year. Other states aren't as meticulous, so that the land areas of some regions remain the same for several years then suddenly grow a huge amount, then stay constant for a few more years. It is likely that these problems average out over the seventeen years from 1982 to 1999.

Population, Density, and Driving

If the data are reliable, what do they tell us about the rate of growth in driving? First, table one shows there is a strong correlation between changes in population and changes in driving (VMTs).

As a matter of fact, per capita driving increased in all urban areas except Colorado Springs and Oklahoma City. Per capita driving in those two regions declined by 4 percent. However, both of these urban areas registered large increases in per capita driving on freeways and arterials.

Table One
Correlations Between Driving and Demographics
(r-squareds of changes from 1982 to 1999)
Population vs. total VMTs                      0.57
Density vs. per capita VMTs                    0.0006

Smart-growth advocates argue that sprawl leads people to drive more. If true, then there should be a strong correlation between changes in population density and per capita driving. Yet the data show that there is absolutely no correlation. Thus, driving is independent of density or sprawl.

New Highways and Highway Driving

Table two shows that there are also strong correlations between the growth in lane miles of freeways and arterials and the miles driven on those freeways and arterials. The correlations are a little higher when comparing lane miles with total VMTs than with per capita VMTs. This suggests that some of the correlation between lane miles and total VMTs is due to more lane miles being built in fast-growing regions.

Table Two
Correlations Between Freeway/Arterial Driving & Miles
(r-squareds of changes from 1982 to 1999)
Freeway lane miles vs. total freeway VMTs       0.60
Freeway lane miles vs. per capita freeway VMTs  0.55
Arterial lane miles vs. total arterial VMTs     0.70
Arterial lane miles vs. per capita art. VMTs    0.62

In short, table two suggests that smart-growth advocates are right: Building freeways and arterials does lead to more driving on those highways.

New Highways and Total Driving

Table three, however, provides a new insight on this claim. The correlation between per capita driving on all roads and total road miles turns out to be no better than random. Moreover, the correlation between per capita driving and freeway lane miles is also no better than random, while the correlation between per capita driving and arterial lane miles is only slightly better than random.

Table Three
Correlations Between Total Driving and Road Miles
(r-squareds of changes from 1982 to 1999)
Total road miles vs. per capita VMTs           0.01
Freeway lane miles vs. per capita VMTs         0.04
Arterial lane miles vs. per capita VMTs        0.09

This means that building more freeways leads to more freeway driving but NOT to more total driving. In other words, new freeways and arterials lead people to change routes from ordinary streets to the freeways and arterials. The result is less traffic on the collectors and local streets. Many of the studies that claim to have proven that more freeway capacity merely stimulates increased driving look only at freeway driving and ignore the big picture.

Accident data show that freeways are the safest highways in urban areas, followed by arterials, then collectors, then local streets. Freeways see few pedestrian accidents because pedestrians are generally excluded. But freeways also make cities safer for pedestrians because they take cars off of collectors and local streets, where most pedestrian accidents occur.

Freeways are thus the ultimate pedestrian-friendly design: If you want to make your city safer for pedestrians, then build more freeways.

Congestion and Driving

Table four refutes another smart-growth assumption: that increased congestion discourages driving and leads people to seek alternatives such as public transit. If this were true, then the growth in driving or per capita driving would correlate with the growth in congestion.

Table Four
Correlations Between Congestion and Driving
(r-squareds of changes from 1982 to 1999)
Travel time index vs. VMTs                      0.02
Travel time index vs. per capita VMTs           0.0001

In fact, the relationships are no better than random. The travel time index is the Texas Institute's estimate of the ratio of the duration of trips taken at rush hour vs. other times of the day. Thus, if a trip takes ten minutes in uncongested traffic, a travel time index of 1.5 suggests that it will take 15 minutes at rush hour. Table four says that there is no indication that increasing congestion has any influence on driving.

Thus, when urban planners say "We are going to increase livability by building rail transit instead of highways," what they really mean is, "We are going to increase congestion by building rail transit instead of highways, and everyone who drives is going to suffer."

Thoreau Institute | Vanishing Automobile | Vanishing Automobile Updates