Failed Attempts to Predict Bicycle Traffic Volume

HOME                             SOCIAL

Over the years there have been attempts to predict bicycle traffic volume. The close connection between interest in increasing bicycle traffic volume and interest in bikeways means that many of these are attempts to measure the extent to which building facilities increases bicycle traffic volume, most particularly of bicycle commuting volume.

Stated Preference Surveys

To my knowledge, all of the early attempts to predict bicycle traffic volume relied largely on stated preference surveys, in which people indicated whether or not they would cycle for any particular purpose if certain bikeways were constructed. The simplest such question would be: "Would you cycle to work if safe bike lanes were provided?" Naturally, the number of people carrying on that activity after the facility was provided proved to be only a very small fraction of the number of favorable answers that had been elicited by the survey. One trouble could have been that the wrong people had been asked the question, for instance people who lived in the area served by the bikeway but worked in an entirely different area that was not served by that bikeway. However, the more serious error was asking people who greatly exaggerated the dangers of cycling to work while simultaneously greatly minimizing, through ignorance or idealism, the inconveniences involved in that activity. Such people are incapable of providing accurate predictions of their behavior regarding bicycle transportation.

A more complicated version of this error involved urban planning, in that it asked people whether or not they would cycle if neighborhood shops existed, or if they lived closer to work, or if the schools were nearer, etc. Again, people were asked these questions who had no idea of the complications inherent in producing this different urban plan. For example, closeness to work often implies a much narrower choice of employers, so that the employment available within a near distance might not suit the employee, or, for that matter, the employees locally available might not suit the employer.

An entirely reasonable stated preference question would be to ask, of an informed person, "If you had a typical residence and a typical job in Amsterdam, would you be likely to cycle to work?" A large proportion of yes answers would be reasonable, because the bicycle mode share in Amsterdam is high. However, such an answer would be completely irrelevant if assumed to apply to any U. S. city, because of the enormous differences between Amsterdam and any U. S. city. But that question is often asked in the sense of, "If your city had a bikeway system like Amsterdam's, would you cycle to work?" That is erroneous because it ignores all the differences between Amsterdam and Rock Island, for example, except the difference in bikeways.  However, that question is often asked implicitly when arguing that building bikeways like Amsterdam's will produce a similar amount of bicycle transportation. Again, that error does not appear to be significant to those in America who believe that there is a great volume of potential bicyclists who are held back largely through fear of same-direction motor traffic, so that removing that fear will unleash torrents of bicyclists. The fact that this has not occurred even where bikeways are provided shows that the whole hypothesis is erroneous.

It is reasonable to say that no system for predicting bicycle traffic volumes that is based on stated preference surveys has produced predictions that have been supported by the actual bicycle traffic measurements.

Catchment Area Surveys

A different analysis would be based on catchment area. About 1985, there was a proposal for a bicycle freeway running westward from UCLA and Westwood, crossing over a freeway and through a cemetery to residential areas beyond. This was planned to be a miniature freeway for bicycle traffic only, with two lanes in each direction and access by ramps, without any intersections. This bicycle freeway was planned to attract the bicycle commuters going to UCLA and the Westwood professional area. I estimated the average speeds for using city streets and using this bicycle freeway to determine the area for which it would be quicker to divert to the bicycle freeway rather than riding directly on city streets. I found that the catchment area based on travel time to be only quite close to the access points. However, the project did not get beyond the early planning stage, so no empirical evidence is available.

It would appear that a good catchment area survey of the users of some existing bikeway would provide information that could be used as the basis for predicting the volume of traffic for a proposed bikeway of similar type. Such a survey would have to determine two kinds of questions concerning the bikeway itself, concerning diversion of existing traffic and creation of new traffic, plus consideration of the attractive power of the traffic generators at the ends. This would require careful questioning of actual users. The diversion question would have to measure the various reasonable routes between that particular individual's origin and his destination and ask why the bikeway user chose to use the bikeway rather than any of the other reasonable routes. The new traffic issue would require some different lines of questioning, trying to understand what modes the user had used before the bikeway, or would use if the bikeway were closed, and the like, to see whether the bikeway itself strongly affected the decision to use a bicycle. Preferably there should also be a similar study of those with similar origins and destinations who use modes other than bicycling, for comparison.

I know of no such detailed study of the users of any bikeway, and hence I conclude that there are no data sufficient to produce reasonable estimates of future bicycle traffic volume on a proposed bikeway.


Some people have denied the supposed claim that vehicular cycling training of cyclists should be used to increase the volume of bicycle traffic. That is a projection upon the vehicular-cycling program (Effective Cycling) of the concern, uppermost in the minds of these deniers, for increasing bicycle traffic. That was never my concern. I intended that the Effective Cycling Program increase the number of competent cyclists and never thought about any effect upon the causes for the production of the many incompetent cyclists who were its raw material.

Correlation Studies

America's foremost producer of correlation studies in bicycle transportation is John Pucher, professor of planning at Rutgers University. Three of his papers are reviewed herein: Bicycle Renaissance and Bicycling Boom and Lessons from Europe. In each of these papers Pucher presents an array of European statistics and practices regarding bicycle transportation (taken from the official publications of the high-cycling nations). Pucher then argues that if we in America were to copy certain practices that Pucher likes (he is particularly fond of bikeways) we would see similar results here. There are three troubles with this procedure. Pucher has never traced a causal connection between any practice and any statistical result. There is no evidence that Pucher's list of practices covers the actual causes of the results he quotes; there is strong evidence that Pucher does not list the most powerful causal factors. There is no evidence that Pucher uses scientific judgment in deciding which among his list of practices he chooses to recommend.

Correlation does not demonstrate causation, except in the mind of an ideologist. 

horizontal rule

Dill, J. and T. Carr. Bicycle commuting and facilities in Major US cities: If you build them, Commuters will use them. In Transportation Research Record: Journal of the Transportation Research Board. No. 1828, Transportation Research Board of the National Academies, Washington, D.C., 2003: p. 116-123. <>

I had seen this paper before.

This paper studies the relationship between bicycle commuting mode and a multitude of socio-economic and similar variables. The authors conclude that by far the most significant relationship is that between density of bike lanes and bicycle commuting mode. However, the authors conclude that there is no evidence as which direction the causal relationship exists, if any. That is, bike lanes may create bicycle commuters, or bicycle commuters may create bike lanes. Despite admitting this, however, the authors have subtitled their paper: If you build them, Commuters will use them. I suggest that the existence of many bike lanes that are not used by commuters proves this title false.

Just one more silly attempt to try to measure the ability of bicycling facilities to change our urban world, and with only the usual negative results.

horizontal rule

The following four studies are contained in:

Tools for Predicting Usage and Benefits of Urban Bicycle Network Improvements: Final Report

By: Gary Barnes, Kevin Krizek, of the Humphrey Institute of Public Affairs, University of Minnesota; Minnesota Department of Transportation; December 2005

horizontal rule

The Effect of Neighborhood Trails and Retail on Cycling and Walking in an Urban Environment: by Kevin Krizek, Pamela J. Johnson

This study takes data from a region-wide travel behavior survey and compares the frequencies of trips taken by bicycle for different distances from bike trails and from bike lanes. People who live less than 1/4 mile from a bike lane are more likely to take a trip by bicycle than people who live more than 1 mile from a bike lane, with a maximum odds ratio of 2.3 in the final model. However, the study data do not indicate whether or not the nearest bike lane was used for the trips counted. And, again, there is no line of causality; the bike lanes might be where they are because more cyclists live in that area, or have chosen to move to an area that, for whatever reason, has more bike lanes. One cannot conclude that the presence of a bike lane caused the individual to make a trip by bicycle instead of making it by automobile, or not making it at all.

There is a great difference between studying bicycling as an end in itself, as, say, a means of reducing obesity or of recreation, and studying the transportational effect of bicycle transportation as part of the trips necessary for the proper functioning of a city. This study covers only the first purpose, and provides no data regarding the second.

horizontal rule

Appendix B: Valuing Bicycle Facilities with an Adaptive Stated Preference Survey: by Kevin Krizek, Nebiyou Tilahun, David Levinson

This study presented viewers with a pair of 10-second videos of traveling along different types of facility (bike trail, bike lane with no parking, bike lane with parking, street with no parking, street with parking), and asked the viewer to state which he preferred and how much increased travel time he would give up to use his preferred facility. The authors state their conclusions badly, and, I think, incorrectly. They give bike lane without parking as the best value, without really considering the issue. However, the paper includes a plot of additional travel time the viewers said they would choose to spend to make the trip on the facility they preferred. The base was a 40 minute commuting trip on normal streets with parking. I show the plotted data as a table (values estimated from the plotted points).

Trip times with equal preference, minutes    
  Total trip time Time over base time
Normal street with parking 40.0 0.0
Normal street without parking 58.1 18.1
Bike lane with parking 60.0 20.0
Bike lane without parking 60.6 20.6
Bike trail 62.5 22.5

Of course, this is nothing more than an exercise in taste, about which discussion is pointless, similar to considering a menu in which veal cutlets are offered for $12.95 and leg of lamb at $15.20. However, if a city is interested in best serving the taste of its citizens, then a response table like that above makes sense, together with the cost of the ingredients, which is never mentioned in a restaurant menu but which both the chef and the city need to consider.

Aside from the above criticism, there is more to say about the research method. Approximately one fifth of the subjects had not cycled in the previous year, and produced results indistinguishable from those produced by those who had cycled in the previous year. The authors think this good, on the grounds that this eliminates consideration of having to serve experienced cyclists differently from others. I say the opposite: if those ignorant of cycling produce the same results as those supposedly informed, then one should conclude that those supposedly informed know as little as those ignorant, which is typical of American knowledge about bicycle transportation.

There is no verification of the value of time which the viewers placed on what they saw, and I consider this to be inaccurate. I think that it would be unlikely for a cyclist to choose to increase his commuting time by almost 50% just to avoid parked cars. This might vary according to the type of area, being different for Cambridge MA than for the area around the University of Minnesota (all the subjects being non-academic employees of that institution), but from my knowledge of Minneapolis I think the problem would be much less in the study area than in Cambridge.

The report shows only one frame from each of the 10-second videos shown, and the horizontal speed of the camera, equal to the speed of the supposed cyclist, was not given. Well, 10 mph for 10 seconds is about 150 feet of travel. It is not possible to evaluate the characteristics of a given 40-minute cycling route by viewing only 150 feet of it. It is extremely unlikely that such a sample, not even when selected at random (which it certainly was not) can convey the multitude of traffic interactions along the route. In short, all that the video showed was an image to which the viewer attached his own preconceptions about cycling. I repeat, this whole study is one of taste, and de gustibus non disputandum.  

horizontal rule

Appendix C: Longitudinal Approaches To Examining The Effects Of Bicycle Facilities On Mode Share: by Gary Barnes, Kristin Thompson, Kevin Krizek

This paper presents comparisons of bicycle commuting mode share between years 1990 and 2000 along several routes with bicycle facilities improvements made in that period. The commuting mode usage is taken from personal transportation census data according to Traffic Analysis Zones, for areas near the route and near the end of the route. That is, for TAZs with centroid 1 mile or less from route, or 1.5 mile or less from end of route. These data generally show statistically significant increases for the TAZs used. Using mode share rather than traffic volume does tend to prevent changes being observed due to external changes, such as, for instance, changes in enrollment at the University of Minnesota, the source of much traffic.

I make two criticisms. The first is that we really wish to know the attractiveness of these bicycle facilities relative to normal streets. Increases in bicycle mode share along the routes would mean nothing if similar increases had occurred elsewhere. In short, there is no control sample in this study, short of some minor notes that elsewhere the bicycle mode share either did not increase or decreased.

The second criticism is that while the changes may be statistically significant, they do not constitute a significant share of actual traffic. The change from 1.701 to 2.000 mode share cited for central city areas is meaningless when considering what needs to be done to provide for the area's transportation. A change of 0.3% in mode share is meaningless.

horizontal rule

Appendix D: A Survey of Residents Near Three Minneapolis-Area Bike Trails: by Kevin Krizek, Gary Barnes, Kristin Thompson

This is the result of a mail-out, mail-in survey of people living within a mile of each of three trails. 3,000 questionnaires sent out, 1,057 usable returns. I generally agree with the conclusions quoted below.

"The results of this survey generally support the findings of other surveys. Various demographic factors are associated with a higher likelihood of bicycling, for example, higher incomes and education, and being male. We found some expected attitudinal correlations, such as shopping at natural food stores, donating to environmental causes, and listening to National Public Radio. However, the impact of these factors was limited, and interestingly, political party affiliation was not at all correlated with bicycling behavior. Ultimately, about 50% of the population bikes at least sometimes, and this is a large enough number to encompass a wide range of opinions on other issues. And given that we found (as have others) that most cycling is recreational, perhaps it is natural that politics is not a major factor in predicting who will ride.

"In terms of better understanding the relationship between facilities and frequency of cycling, our results here were somewhat consistent with the other studies described in this report. We generally found a relationship in that people who lived closer to facilities tended to ride more, but the difference between the closest and farthest groups (less than a quarter mile versus more than one mile) was not that big. However, most of our respondents lived within about a mile of a trail, so impacts at a larger scale than this would not have been captured by our survey. This indicates that perhaps the impacts of a trail are not confined to the short distances indicated by one of our other studies; and is more consistent with the commuting study, in which even people a mile and a half from a trail still seemed to show a higher likelihood of biking."

The questionnaire apparently failed to ask how the respondent reached the trail. For walking, the falloff in frequency of use with distance is little more than that for bicycling. One would expect that the falloff would be more rapid for walkers than for cyclists. Does this indicate that walkers, or maybe both walkers and bicyclists, frequently reach the trail by motoring to it?

This is a study of trails that are largely used for recreational purposes. Its data may be usable to make reasonable estimates of usage volumes for trails that are proposed in equally attractive locations. However, the study does not concern itself with the reasons why users used the trail (beyond the vague categories of recreation, work or school, etc.). The only utility that is implicit in the study is the utility of simply being on a trail. Therefore, there is no way to tell  how useful the trail is, and, therefore, there are no data to compare the utility of any of these trails against the utility of a proposed trail.

Specifically, therefore, this study provides no means of estimating the transportational effect produced by the existence of these, or any other, trails.

horizontal rule

The above collection of studies is titled: "Tools for Predicting Usage and Benefits of Urban Bicycle Network Improvements". I see no means disclosed for making reasonable predictions of either usage or benefits.

horizontal rule

horizontal rule