Review of a paper presented at Transportation Research Board 2006
Sidepaths in typical urban settings have long been recognized as dangerous cycling facilities. However, it is well known that some politically powerful groups wish to preserve bicycle sidepaths for a variety of motives; their pressures produce studies of this type. This study, one of a series, was funded by the Florida Department of Transportation to discover those design features that might make sidepaths safer than roads. The characteristics that make sidepaths much more dangerous than roads were identified thirty years ago and have been frequently described since; there is no mystery about them.
In this study, its authors have managed, through what they consider to be a scientifically sound procedure, to identify what they call the four most important factors for the safety of sidepath cycling, not one of which has any connection with the five most important danger factors that have been recognized for decades on the bases of experience, test, and analysis.
This result raises three important issues.
Was there ever any reasonable probability that any investigation of this type would produce positive results? This is discussed in the next section.
Does the process by which such an astonishing result was produced conform to standards? This will be discussed in this review.
By what process or reasoning did the Transportation Research Board accept this paper? This can be only hinted at in this review.
This study is one of a long series of studies, going back thirty-five years, that have attempted to demonstrate the hypothesis that cycling on bikeways according to bikeway operating procedures (the bikeway hypothesis) is safer than cycling on normal streets according to normal traffic operating procedures. The contrary hypothesis is that cycling on normal streets according to normal operating procedures is safer than cycling on bikeways according to bikeway operating procedures (the vehicular cycling hypothesis).
All persons with traffic-engineering expertise should by now have recognized the controversy over these two hypotheses. A short history of this controversy is given in Forester, 2001. Despite an enormous amount of officially supported investigations, there have been no results that have either favored the bikeway hypothesis or disproved the vehicular cycling hypothesis. The relatively small amount of investigations by well-informed cyclists have demonstrated the clear superiority of the vehicular cycling hypothesis over the bikeway hypothesis.
In short, ever since the start of the bikeway controversy, the great weight of the evidence has been in favor of the vehicular cycling hypothesis. Considering both the amount of effort put in by each side and the results produced, it is highly unlikely that any evidence will ever be discovered that will reverse the balance in favor of the bikeway hypothesis.
This situation leads to two conclusions.
There is little likelihood that any study proposing to support the bikeway hypothesis will ever produce positive results.
If any study claims to support the bikeway hypothesis, it is likely to do so on the basis of scientific errors, because it is based on an hypothesis that has vanishingly small probability of being correct.
Any administrator of research projects around the subject of bikeways should seriously consider this scientific situation. Failure to do so will involve the administrator in failed investigations with wasted money, whose only result is propaganda for the faithful. The Transportation Research Board should never be party to such unscientific procedures.
The scope of the study concerns features that would typically be known to a highway design engineer. That is, the study assumes that the characteristics that make sidepath cycling safer than roadway cycling are options available to the highway designer for designing sidepaths that will result in the desired effect that cycling on them is safer than cycling on the road. The authors do not admit to any such assumptions; they probably were unaware that they had made them.
Florida DOT’s interest in trying to develop a scientific rationale for sidepaths is shown by previous work done for it by Sprinkle Consulting. This work had developed an agreed methodology consisting of the following steps:
[Determine] "the level of accommodation for bicyclists on the adjacent roadway, paired with the potential bicycle travel along the roadway,
[Determine] potential safety of a sidepath facility,
[Determine] the availability of alternative routes,
[Determine the] adequacy of right-of-way to accommodate a 'safely functioning sidepath',
[Determine the] access to probable destinations,
[Determine the] appropriateness of sidepath length and the design of termini, and
[Determine] the level of comfort and safety the proposed sidepath would provide to cyclists."
The authors are vague on the purpose of this particular study; presumably it is to support item 2, to determine the potential safety of a sidepath facility, and, possibly, item 4, in that the design characteristics of a safer sidepath, as determined by item 2, may require some amount of space that may, or may not, be available. (The authors suggest a further investigation to develop "a conceptual level of service framework for sidepaths.")
The authors report that the following factors have the greatest significance regarding the safety of cycling on sidepaths:
"the width of the sidepath,
the effective distance between the sidepath and the roadway,
the posted speed limit on the adjacent roadway, and
the number of lanes on the adjacent roadway."
This is truly astonishing. Safety, as far as these authors are concerned, refers only to car-bike collisions. It takes only one moment to recognize that a cyclist riding on a sidepath route can incur a car-bike collision only if the car crosses the sidepath route or the cyclist rides or turns off the sidepath across the car’s path. Besides that, true safety also considers that sidepaths are facilities with chaotic traffic patterns that threaten cyclists with collisions from unpredictable and undisciplined users.
All persons qualified to investigate bicycle traffic issues should be familiar with the issues regarding sidepath safety that have been well published to the profession since the experiences on the Palo Alto sidepaths in 1972. The most important factors concern:
The amount of motor traffic crossing the sidepath and how it crosses
The extent to which cyclists using the sidepath need to cross into traffic
The quantity and nature of the undisciplined traffic using the sidepath
The speed of the cyclists using the sidepath
The skill and knowledge that these cyclists can apply to avoiding collisions with all this traffic.
The authors have managed, through what they consider to be a scientifically sound procedure, to identify what they call the four most important factors for the safety of sidepath cycling, not one of which has any connection with the five most important factors that have been recognized for decades, on the bases of experience, test, and analysis.
The authors surveyed some of the literature regarding sidepath accident rates (Wachtel & Lewiston; Pasanen and Rasenen), while including other material that did not (Kaplan). While these reports generally indicate that paths have high accident rates, the authors stated no conclusion regarding these reports. The authors surveyed the design recommendations made in various design standards, and concluded that these recommendations "have not been validated based upon actual crash data and the relative safety of in-street and sidepath facilities." The authors did not survey any of the operational analyses regarding the methods, or behaviors, used by drivers operating on or around sidepaths, such as Forester’s.
The authors therefore decided to collect much information of the type that would typically be available to a highway designer faced with the task of determining the suitability of a sidepath, and to correlate this information with car-bike collision rates to determine which of these characteristics had a significant effect on the ratio of car-bike collision rates on the roadway and on the sidepath. The authors studied 21 segments of roadway associated with sidepaths. The study used the car-bike collision rates that occurred for cyclists cycling along the roadway and the car-bike collision rates for cyclists cycling along the sidepath.
Determining car-bike collision rates requires determining the number of car-bike collisions, and the authors report that they had difficulties in determining which of the reported car-bike collisions were relevant to their analysis. They do not report on the criteria they used to select from all the reported car-bike collisions those that they used in calculating car-bike collision rates. I suggest one difficulty. Consider a collision between a car traveling on the road to which the sidepath is adjacent and a cyclist crossing that road. Was that cyclist crossing that road because he had approached that road on the intersecting street, or because he had turned from the sidepath? How do you tell? If you could tell, what ought to be the significance of the knowledge?
The other value used in such determinations is the traffic volume. Those for motor traffic were obtained from the standard traffic surveys; those for bicycle traffic were obtained from special counts on several days.
Having collected all these data, the authors then analyzed them for significant correlations using Pearson correlation and factor analysis "with respect to difference in crash rates". The four factors that they found to be significant are:
"the width of the sidepath,
the effective distance between the sidepath and the roadway,
the posted speed limit on the adjacent roadway, and
the number of lanes on the adjacent roadway."
The authors then subjected the data to stepwise regression analysis to determine the appropriate coefficients for each factor. I give the coefficients for each of the above factors, where W is width of sidepath, D is separation between roadway and sidepath, S is speed limit, L is number of lanes. (ln is natural logarithm of the specified number) The coefficients are arranged so that positive result means that the sidepath has a lower car-bike collision rate than the roadway, and vice versa. I then describe the effect of the factor.
Sidepath width: 6.311W - 0.465W^2
The contribution of sidepath width is positive for narrow paths, peaks at a
width of 7 feet and a maximum contribution of about 21 points, and declines
to become negative at a width between 13 and 14 feet.
Sidepath distance (D) and speed limit (S): 0.015SxD -
0.685D
The contribution of sidepath separation is increasingly negative, ranging
from -0.04 for separation = 4 feet to -0.18 for separation = 18 feet.
The contribution of motor vehicle speed limit is uniformly positive and
increasing with speed, ranging from +2.4 for 16 mph to +8.4 for 56 mph.
The contribution of separation and mv speed per the formula is, when
calculated for speed in mph = 4 x separation distance in feet (which ratio
covers a reasonable range of values), is a wobbly negative variable with a
most negative value of about -2 for low speed and close separation to zero
about 48 mph and 12 feet separation, to +3 for 60 mph and 15 feet
separation.
Number of lanes: -1.528ln(L)
The contribution of the number of lanes is positive and increasing, ranging
from +1 for a 2-lane road to +3 for an 8-lane road.
On top of all these contributions is an arbitrarily determined correction term of -17.555, just to make the formula come out better.
The first criticism to be noted is that the study completely ignored all crashes that are not car-bike collisions. Considering the chaotic traffic conditions typical of bicycle paths of any type, these are significant, both in quantity and as hindrances to transportationally useful cycling.
To proceed to the study itself, note the relative maximum sizes of these contributions:
Sidepath width: +21, peaking at width about 7 feet and decreasing on each side
Arbitrary constant: -17.555
Motor vehicle speed: +8.4 and safety increasing with speed.
Number of lanes: +3, and safety increasing with number of lanes
Separation and speed combined: -2, +3, with safety generally increasing with higher values
Sidepath separation distance: -0.18, with safety decreasing with increasing separation distance
According to this list, the most influential factor in car-bike collision rate is sidepath width, with a maximum in sidepath safety at about 7 feet. The next most influential factor is the arbitrary constant, inserted into the formula to make everything else come out more accurately, and, therefore, representing unknown factors. The negative sign of the arbitrary constant indicates that these unknown factors include very significant sidepath dangers. Motor vehicle speed is less than 1/2 as powerful as either of the two more powerful factors, while everything else is less than 1/2 as powerful as motor vehicle speed.
So far, all this has been merely an exercise to tease out mathematical correlations. As is commonly stated, correlation does not demonstrate causation, and, without recognizing the causative mechanism, there is no credible assurance that correlations demonstrated in this set of data would appear in another set of data from other locations. That is, because we have no knowledge of why the correlations have appeared in this set of data, we can make no reasonable prediction that they would appear in any other sets of data.
It should be obvious that, for a correlation to represent a cause in this analysis, there must be some physical causal connection between the factor and the car-bike collision rate. Consider the factor with the supposedly most powerful effect, sidepath width. I cannot imagine any really significant relationship between the width of the sidepath and the probability of collision with a car that is necessarily crossing the path, and is, therefore, in either a driveway or an intersecting street. If the car is in a driveway, then a wider path might give the cyclist a greater chance of avoiding the collision, but this would not be true if the car is in a street, where the width of the path has become effectively infinite.
The factor of motor-vehicle speed on the adjacent street might be partially explained by the idea that the streets that attract higher-speed traffic are also arterial streets, which generate longer trips and, therefore, fewer turning movements per vehicle passing through, and at which crossing traffic is more cautious and is more often controlled by traffic signals, which materially reduce collisions between crossing streams of traffic. However, these possible explanations have little connection with the speed of the traffic itself, but are merely correlates with traffic speed.
The number of lanes in the street which the sidepath parallels might also be associated, in the same way, with the prevalence of traffic controls and traffic signals.
As for the separation distance, if the formula is really valid for the full range of separations (which is unknown), then it says that no separation, and hence no sidepath, is best of all.
In short, this paper is merely the report of fanciful imaginary circumstances, teased out of a collection of data by mathematical means that ignore both causality and previous knowledge, surrounding the problem of car-bike collisions produced by cycling on sidepaths.
It is easy to see that the authors failed to recognize the most important factors contributing to car-bike collisions occurring on sidepath routes.
One physical factor and two operational factors powerfully affect the probability of car-bike collision when cycling on sidepaths. The physical factor is the presence of motor traffic crossing the path. Without that, there will be no car-bike collisions during cycling on sidepaths (though there may be as the result of intending to use, or having used, sidepath routes). Therefore, the rate of such collisions when cycling on sidepaths will be approximately proportional to the frequency of such crossing traffic. One would have thought that some such measurement should have been made, but it was not. Neither was there any study of the presence of traffic controls that would limit the danger of such crossing traffic.
The two operational factors are those intertwined factors of cyclist speed and cyclist skill. With sufficient skill both to recognize the dangers and be motivated to slow down to a safe speed, practically any sidepath can be cycled at a reasonably low car-bike collision rate. It is said that anecdotes do not make data. I differ in saying that the extreme anecdote may produce valid data. When I tested the safety of urban sidepath cycling in 1974, to eliminate the effect of different speeds I deliberately rode at the same speed I normally used on the adjacent roadway. My skill in recognizing the dangers and avoiding them prevented five or so car-bike collisions in a short distance, until the one at which I recognized that only luck had saved my life, at which time I stopped the test and have never resumed it. In the intervening thirty years, no other person has reported being sufficiently brave to repeat this test.
In summary, the authors ignored the factors that are obviously of greatest importance and produced a mathematical fantasy by calculating from the factors that are of little importance. One more failure in the attempts to produce a scientific rationale for bikeways.
This paper was accepted by the Transportation Research Board for presentation at its annual meeting of 2006. Presumably the responsible committee was ANF20 Bicycle Transportation. Many years ago, when I was a referee for that committee (which then had a different number), the TRB announced that its standard for refereeing was Bernard K. Forscher’s Rules for Referees (Science 150 319-321, 1965). The following quotations are taken from that document.
"2) Bibliography. Does the manuscript contain a complete or representative set of references? (If an important relevant report has been omitted, it should be cited in the comments. Authorship by the referee is not necessarily a valid criterion of relevancy.) In other words, has the author given a fair description of the knowledge available at the time the manuscript was written? Furthermore, does the manuscript accurately report the statements in the references cited?"
The dangers of sidepath cycling have been widely discussed in the field of bicycle transportation. Much information about bicycle transportation circulates informally, but its circuits are known to the authors. However, formal statements of these dangers were published as early as 1977, now available in Forester; Bicycle Transportation; The MIT Press, 1994. A discussion of these, with a lengthy bibliography, was published in Forester; The Bicycle Transportation Controversy, Transportation Quarterly V55 No2, Spring 2001, p 7-17. Knowledge of the information presented in these would have directed the authors toward the kinds of data that would have best assisted their investigation.
The most significant defect in the paper is addressed by the following questions suggested by Forscher:
"Specific questions ...
"3) Reliability of methods. Are the methods used in the work under consideration adequate to support the conclusions? Bitter differences of opinion can develop over this point. The referee is expected to support his opinion by evidence from the literature if he disagrees with the author."
"General questions ...
"2) Validity of the logic. Is there a defect in the reasoning used for deriving the conclusions from the observations? If the referee believes there is, he should specify the step he thinks incorrect and say why he believes it is faulty. The referee should consider only the conclusions the author has presented; he should not extend or redirect them.
"3) Alternate interpretations. Are there other valid interpretations of the observations, in addition to the interpretation offered by the author? The existence of such alternatives does not in itself invalidate the author's interpretation, but the editor should be aware of them and should consider the extent to which they should be recognized in the paper."
These three issues are all relevant to this paper: reliability of methods, validity of the logic, and alternative interpretations, relevant because the basic defect spreads so widely.
The basic defect is expressed as failure to consider the elementary principle of statistics that correlation does not demonstrate causation. The paper implies one conclusion with two uses: that where a sidepath complies with the suggested design characteristics, cycling on it is safer than cycling on the roadway. This conclusion may be used both to survey sidepaths and to design them.
The method chosen was to collect data thought to be relevant so that mathematical correlation analysis could be performed. The data chosen to be collected did not include the data whose relevance was given by the pre-existing literature, which is defect enough. No matter how mathematically valid the methods of data analysis (which validity is not discussed herein), without the relevant information no valid and relevant correlations could be established.
However, the paper’s most significant defect involves all three issues. Drawing conclusions of accident causation requires not only statistical correlations but also demonstration of the causal links that underlie the correlations. The method is inappropriate because it does not do that, the logic is invalid because it does not include such a demonstration, and, with no demonstration of causation, a great many alternative explanations become acceptable.
The evidence is very strong that the Bicycle Transportation Committee of the Transportation Research Board, an arm of the National Academy of Sciences, should, by its own standards, have rejected this paper. However, this unwarranted acceptance of an unacceptable paper should not have been unexpected. During my term of service on that committee twenty years ago, that committee frequently accepted papers that Forscher’s Rules would reject, and rejected papers on grounds that Forscher’s Rules excluded from consideration. In each such case, the decision favored the bikeway program. What more needs to be said?
Forester, John; The Bicycle Transportation Controversy, Transportation Quarterly V55 No2, Spring 2001, p 7-17
Forester, John; Bicycle Transportation; The MIT Press, 1994, first edition 1977
Forscher, Bernard K.; Rules for Referees (Science 150 319-321, 1965).
Kaplan, J. Characteristics of the Regular Adult Bicycle User. Petrisch et al give the source as: FHWA, U.S. Department of Transportation, 1975. However, this is a Master’s Thesis, University of Maryland, said to be available from the National Technical Information Service, Springfield, VA.
Pasenen E. & M. Rasenen; Cycling Risks in the City of Helsinki, English summar; Helsinki City Planning Department, 1999
Wachtel, Alan, & D. Lewiston; Risk Factors for Bicycle-Motor Vehicle Collisions at Intersections; ITE Journal, September 1994