Top: John S. Allen's Home Page
Up: Table of Contents
Previous: Calculation of miles/hour figure
Next: John S. Allen's response

The CPSC Study of Bike Use and Hazard Patterns:
Goals, Methodology, and Response to Criticisms

by Gregory B. Rodgers, PhD

Gregory B. Rodgers, PhD., was the Bicycle Project Manager for the CPSC's Bicycle Project. Views expressed in this article are those of the author and do not necessarily reflect the position of the CPSC or other members of its staff. Since the article was written in the author's official capacity, it is in the public domain and may be freely copied or reprinted.

This article responds to the major points raised by John S. Allen in his recent review of the U.S. Consumer Product Safety Commission's (CPSC) national study of bicycle use and hazard patterns (Allen, 1995). The review, entitled "Analysis of Bike Use and Hazard Patterns," appeared in issue number 37 of Bicycle Forum. Before addressing the review's major points, however, I will provide a brief description of the general methodology and goals of the CPSC study (CPSC, 1994). This is necessary because Mr. Allen provided little description of the study design.

In addition, it should be stressed at the outset that the CPSC bicycle study was intended to be an analysis of the use and hazard patterns of the general population of bicycle riders, not just of avid adult bicyclists. The study found that there were more than 65 million bicycle riders nationwide in 1991 and that the great majority were children and occasional adult riders. Consequently, the results of the study are not necessarily representative of the experience of avid adult bicyclists, such as those who might be members of the League of American Bicyclists or who might have taken the League's training course in Effective Cycling.

Reasons for the Study

The main impetus for the CPSC study was the lack of national data on bicycle use and hazard patterns, in the face of the large number of injuries and deaths involving bicyclists every year. According to the CPSC's National Electronic Injury Surveillance System (NEISS), an injury reporting system consisting of a statistical sample of the nation's hospital emergency rooms, there are more than 500,000 bicycle-related injuries treated in hospital emergency rooms every year. In addition, there are on the order of about 900 bicycle-related deaths annually.

Despite the large number of injuries and deaths, little information has been gathered on the characteristics and use patterns of the general population of bicyclists. Moreover, while there has been some detailed analysis of injuries resulting from bicycle-motor vehicle collisions (see, e.g., Cross and Fisher, 1977 and Roland et al., 1979), little attention has been given to the great majority of bicycle-related injuries which do not involve motor vehicles.

This lack of data on bicycle use and hazard patterns has been well documented. See, for example, comments in The National Bicycling and Walking Study recently completed by the Federal Highway Administration (FHWA) of the U.S. Department of Transportation (FHWA, 1994). The League of American Wheelmen (now the League of American Bicyclists) made this same point in comments urging the CPSC to include a "comprehensive study of bicycling" as a priority project several years ago (Skonecki,1985).

The lack of available data is also amply clear from Mr. Allen's review itself. With the exception of a recent study of bicycle risks at several sites in one city (Wachtel and Lewiston, 1994), all of Mr. Allen's references date from the 1970s. Moreover each referenced study evaluates only one segment of the bicycle riding public, such as League of American Wheelmen members, college age bicyclists, elementary school children, or a sample of bicyclists in Palo Alto, California. None can be considered representative of the general population of U.S. bicyclists, a goal of the CPSC bicycle study.

Study Design

The study's focus was on two national telephone surveys - an injury survey and an exposure survey - that were conducted by the CPSC in 1991. The injury survey gathered information on a statistical sample of 463 bicycle riders who were injured and treated in NEISS hospital emergency rooms during 1991. This survey consisted of about 70 questions that gathered information on the characteristics of the injured bicyclists (e.g. age, gender, and riding experience), their use patterns (e.g. where and how much they ride), and the type of bicycle used. In addition, detailed information about the injury scenario and circumstances surrounding the accident (e.g. where, when, how, and what happened just before, during, and after the accident) was collected by means of a series of both closed and open-ended questions.

The exposure survey was a state-of-the-art national probability telephone survey that gathered information on the characteristics and use patterns of the general population of bicyclists. This survey was conducted for the CPSC by a respected survey firm, Abt Associates, Inc., during June and July, 1991. The 53 question survey collected information from a representative sample of 1,254 bicyclists from around the nation.

Because the injury and exposure surveys were designed to gather parallel information on injured and noninjured bicyclists, CPSC staff were able to conduct a formal risk analysis to determine and quantify the factors associated with risk. In addition to the risk analysis, the study included an analysis of possible bicycle mechanical hazards reported from the injury survey, a human factors assessment of children's injuries, an analysis of bicycle helmet usage patterns, and an analysis of bicyclist deaths and fatality risk patterns.

The study also evaluated the results of a complementary (but completely independent) survey of adult bicyclists conducted by National Family Opinion, Inc. for the Rodale Press, the publishers of Bicycling magazine. Although the Rodale Press survey did not gather information on one of the CPSC's major focuses, bicycle use by children, it did gather data on a wide range of topics relevant to the risk and safety behavior of adult bicyclists. These data included accident and bicycle use information, allowing CPSC staff to conduct a risk analysis based on the Rodale Press survey database.

Technical Issues Raised in the Allen Article

The detailed results of the CPSC analysis are contained in the study itself, which is entitled Bicycle Use and Hazard Patterns in the United States (CPSC, 1994) and is available from the CPSC upon request.[1] In addition, substantial segments of the study have already been published in the professional literature. See, for example, Rodgers (1994, 1995a, and 1995b) and Tinsworth, Polen, and Cassidy (1994). The Allen article raises several issues, however, which are discussed below.

Measures of Bicycle Use

The first point Allen raises is that the study used an inaccurate measure of riding exposure. Some background is necessary here. After discussions with bicycle user and industry groups, CPSC staff decided to estimate riding exposure in terms of hours of bicycle use rather than riding distances. Contrary to Mr. Allen's suggestion that bicycle experts were not involved when the survey was being developed (Allen, p. 15), the CPSC solicited and received extensive comments on the questionnaire from a number of organizations and individuals. In addition, The Bicycle Federation of America (BFA) organized a meeting of interested parties on December 12, 1990, to discuss these very issues. Attendees included representatives from the BFA, the Bicycle Manufacturers' Association of America, the National Highway Traffic Safety Administration, the Federal Highway Administration, and the National Safe Kids Coalition. The consensus of this meeting was that hours of bicycle use would be the more useful way to measure riding exposure in large part because of the difficulties associated with estimating mileage for children and occasional riders. On the other hand, the Rodale Press survey, which interviewed only adults age 18 and older, estimated exposure in terms of miles ridden...[2]

Mr. Allen's comparisons of riding exposure are not very meaningful. First, since it is well known that survey responses are sensitive to the ways questions are asked, there is no reason to expect exposure estimates to be directly comparable across distinct surveys. In this case, the CPSC and Rodale Press surreys used two entirely different sets of questions to gather information on two different measures of riding exposure. Riding time estimates from the CPSC surveys were based on responses to several questions intended to determine the number of hours each bicyclist spent riding in an average month. In contrast, the Rodale Press survey simply asked "Approximately how many miles do you ride your bicycle in an average warm weather month?" Second, the estimates of exposure from both the CPSC and Rodale Press surveys were positively skewed. (The Pearson coefficients of skewness were 0.98 and 0.77, respectively.) This indicates that the means were larger than the median exposure estimates, and suggests (accurately in this case) the presence of outliers. This commonly occurs when asking about variables like hours of use or miles ridden, and can be adjusted for in the technical analysis (as was done in the CPSC study). Moreover, it suggests that the median estimates of riding exposure provide a better measure of central tendency than do the mean estimates. Table 1 provides the median estimates of monthly riding exposure, from both surveys, for various age groups.

Table 1. Median Estimates of Monthly Riding Exposure (by Age of Bicyclist)

Age (yrs.) Mi/mo
(Rodale Press)
(CPSC Survey).
18-30 25 10
31-40 20 9
41-50 20 10
51-60 20 10
>60 18 8
Total 20 9

Estimates of individual riding exposure obtained from surveys and based on recall certainly cannot be considered precise. Many individuals probably did estimate the total time of rides, including the time they were not actually moving. In addition, as Mr. Allen suggests, conducting the survey in the summer may also have raised estimates of riding exposure. There are also problems in estimating mileage. Nevertheless, the median exposure estimates described in Table 1 do not seem unreasonable as a first approximation for the general population of bicyclists. More over, if there are any biases, they appear to be systematic - that is, the median exposure estimates of one survey are roughly a constant multiple of the other. And if that is the case, the exposure estimates can still do what they were intended to do - they allow for the evaluation of risks while statistically controlling for riding exposure. It is also worth noting that the risk findings relating to riding exposure were quite similar in both the CPSC and Rodale Press databases, a finding that adds additional credibility to their use as relative measures of exposure in the risk analyses. Both analyses indicate that risk increases with riding exposure, but that it increases at a decreasing rate for those who ride more. This finding reinforces Forester's conclusion that, on a per mile basis, risk declines for those who ride more (Forester, 1983).

In his concluding remarks on the CPSC's method of measuring riding exposure, Allen asserted that "the [CPSC] researchers failed to hone their investigative technique on the lessons of earlier studies of crash exposure... " (p. 14). Unfortunately, while each of the earlier studies referenced by Allen was useful in its own right, none serves as a methodological model for improving the CPSC's technique on this point.

Consider how the earlier studies estimated riding exposure. Kaplan (1975), in a survey of League of American Wheelmen members, simply asked respondents "How many total miles did you ride in 1974? (best estimate)." Schupack and Driessen (1976), in a survey of college students, estimated annual mileage from two questions. The first asked the number of months a bicycle was regularly ridden (possible responses: 1-2, 34, 5-6, 7-8, 9-10, 11-12 months); the second got at the number of miles ridden during those months (possible responses: <16, 16-24, 50-99, 100-199, 200-399, 400-799, 800 miles or more). Finally, the survey questions in Chlapecka et al. (1975), were even more general. In a survey of elementary school children, this study made only a rough estimate of annual mileage based on questions asking when the most recent bicycle ride took place (possible responses: yesterday, 2 days ago, 3 days ago, 4 days ago, 1 week ago, 1 month ago, or more than 1 month ago), the number of hours the children estimated they were on their bikes on that most recent ride (possible responses: 1/2 hour, 1 hour, 2 hours, 3-5 hours, or more than 5 hours), an assumption that the children rode 5 miles per hour when on the bike, and an assumption that 14 percent of the children's annual riding took place during the month of the survey (the survey was conducted in May, 1972)...[3]

In both the Schupack and Driessen (1976, p: 15) and the Chlapecka et al. (1975, p. 17) studies, the authors caution that the mileage estimates were not intended to be absolute exposure estimates comparable to estimates from other studies. Nevertheless, Mr. Allen does just that at p. 13; he compares the mileage estimates from all of these studies to the results of the CPSC study.

Various Extraordinary Findings

Mr. Allen also mentioned a number of what he called extraordinary findings. Only a few can be dealt with because of limited space. However, I believe these examples illustrate Mr. Allen's approach to analysis in the review.

The Issue of Age and Risk

Mr. Allen says (p. 13):[4]

"Children ride more, and their fatality rate is lower than for older bicyclists, though their injury rate is higher. These conclusions reflect not only the inaccuracy of the CPSC's riding time data, but also the dilution of its definition of "bicyclist" by including the many millions of Americans, mostly adults, who ride as infrequently as once per year. "

Based on the results of our study, 71.4% of the nonfatal injuries involved children under the age of 15, but only about 37.0% of the fatalities did. This suggests that children and adults have different risk patterns, and that adults are more likely than children to be fatally injured when they are involved in accidents. Moreover, since these injury and fatality statistics are independent of any exposure measure, they cannot reflect the inaccuracy of the "time" data or the definition of a bicyclist.

The Issue of Gender and Risk

Take another example (p. 13):

"Fatality risk is five times as high for males as for females. National Safety Council figures cited in the CPSC report do show 720 male and 121 female fatalities for 1991. But risk also depends upon exposure. Though both CPSC and Rodale data show about as many male as female riders, neither report shows riding time/mileage or location by gender. A CPSC analysis of Rodale data finds no independent statistical relationship between gender and risk, as does Wachtel and Lewiston 1994. Chlapecka 1975 and Schupack and Driessen 1976 show a 60 percent higher crash rate for females. Though these are not fatality rates, they call the CPSC's conclusion into question. "

Contrary to Mr. Allen's statement, the higher relative fatality risk estimate for males in the CPSC study does account for exposure, as described in detail on pages 104 and 106 of the CPSC report. That is, in estimating the relative fatality risks associated with the gender of bicyclists, the risks were adjusted to take into account male and female riding times. However, even if they hadn't, the fact that almost six times as many male as female bicyclists died in 1991 suggests the likelihood, or at least the strong possibility, that the fatality risk for males is higher than that for females.

Moreover, the CPSC finding of no statistically significant difference in the nonfatal injury risks of males and females does not call into question the fatality risk conclusions, which were quite strong. Rather, it suggests very different fatal and nonfatal injury patterns, and different riding patterns for males and females. This issue is discussed extensively in Rodgers (1995b). In brief, however, while only a small proportion of nonfatal injuries involve motor vehicles or occur after dark, about 90% of the fatalities involve collisions with motor vehicles on public roadways and about 35% occur after dark. The higher fatality risk for men is therefore most likely associated with the finding from the CPSC exposure survey that men are more likely than women to ride on roadways and at night.

Roadways, Bike Paths, and Risk

Another issue that concerned Mr. Allen is the study's finding that risk was significantly higher on roadways than on bike paths. This finding apparently contradicts the conventional wisdom in the bicycling community (see, e.g., Forester, 1983 and 1993; Schubert, 1994). However, the study's results were quite clear.

While about 17% of bicyclists said they rode a large proportion of the time on bike paths, less than one percent of the injuries occurred on bike paths. And this is not because bike path injuries that occurred at intersections with roadways were miscoded as roadway injuries, as Mr. Allen suspects (p. 14).[5] Nor does the finding depend on the survey estimates of riding time.

One can speculate that this risk finding was related to the sample of injuries collected in the CPSC injury survey - surveys are subject to sampling errors, and need to be replicated. However, this result was also consistent with the risk analysis of the Rodale Press survey data. Although the specific relative risk estimates varied (as would be expected), both risk analyses found that the risks on roadways were significantly higher than on bike paths.

Now, having said this, what does it mean? It certainly does not mean that bike paths are always safer than roadways. Schubert (1994) correctly points out a number of potential safety problems with bike paths. The statistical results only suggest that for the average bicyclist (who presumably is different from the average member of the League of American Bicyclists) the risks on bike paths are lower than on roadways.

Most importantly, the CPSC study did not find or infer that the construction of bicycle paths was the sole or even the primary answer to bicycle risks. Nor was there any suggestion whatsoever that the access of bicyclists to roadways should be limited.

Nor did the CPSC recommend "equipping a million or so miles of neighborhood streets ... with bike lanes or supplanting them with bike paths," as suggested by Mr. Allen (p. 14). In fact, the CPSC study (1994, p. 15) states only that its risk findings:

"suggest that the riding environment should be an important focus of efforts to reduce bicycle injuries and deaths. Such efforts might focus on improvements in roadway design aimed at reducing many of the serious injuries involving collisions with automobiles every year. The development of bike paths (i.e., paths that separate bicycles from parallel motor vehicle traffic) and bike lanes (i.e., designated lanes on roadways which are off-limits to motor vehicles) should also be considered."

This statement certainly does not support Mr. Allen's characterization of the CPSC as engaging in "pie-in-the-sky planning" for bike lanes and bike paths (p. 14).


This paper responded to some of the major criticisms raised in the Allen review. All studies have inherent limitations which must be recognized and kept in mind -- and the CPSC study is no exception. However, as described above, the CPSC study was carefully designed with input from both the bicycling and safety communities.

Moreover, the study succeeded in providing a wealth of previously unavailable and vital information on bicycle use and hazard patterns. The CPSC exposure survey, for example, provided extensive information on the characteristics and use patterns of bicycle riders in the United States. Unlike earlier studies, which only evaluated data on individual segments of the bicycle riding public (e.g., League of American Wheelmen members, college age bicyclists, elementary school children), the exposure survey covered the entire population of bicyclists. In addition, the injury survey provided detailed information on the full range of nonfatal bicycle-related injuries and hazard patterns - including the great majority of injuries which do not involve motor vehicles and had never been evaluated on any systematic basis. And together, the combined results of the injury and exposure surveys allowed the CPSC to evaluate bicycle risk patterns.

The CPSC bicycle study has also served as a springboard for discussions on ways to enhance bicycle safety. Mr. Allen's review of the bicycle study and the response in this paper, for example, at least indirectly promote bicycle safety by getting some of the important issues on the table. Perhaps more directly, the CPSC study findings also led to a conference on nighttime bicycle safety, held at CPSC headquarters on November 9, 1994[6]. The CPSC-sponsored conference evaluated ways of reducing nighttime bicycle risks and included a wide range of interested participants, among them representatives of bicycle advocacy and user organizations (such as the Bicycle Federation of America and the League of American Bicyclists), industry groups, researchers, and state and local bicycle program coordinators.


End Notes

1. To obtain a copy of the study, write to: Bicycle Study (OS), U.S. Consumer Product Safety Commission, Washington DC 20207.

2. I have here omitted Dr. Rodgers's discussion of a mathematical error in my first publication of my analysis, which I have corrected in the current republication -- JSA.

3. If, for example, a child reported that the last time he rode was for two hours 'yesterday, " the annual mileage estimated would have been 1,429 miles. However, if he had reported the ride took place two to four days ago, rather than yesterday, the estimate would have been 571 miles.

4. 1 have left out Mr. Allen's references to the page numbers of the CPSC study.

[Omitted here: footnote about my incorrect attribution of a data source. I have corrected the attribution in the current republication -- JSA].

5. As mentioned above, information about the injury scenarios and the circumstances surrounding the accidents was collected by means of a series of both closed and open-ended questions. Consequently, it would have been apparent if an accident had occurred at a bike path/roadway intersection.

6. For a description of the conference, see the January/February 1996 issue of Bicycle USA.


Allen, John S. "CPSC Analysis of Bike Use and Hazard Patterns" Bicycle Forum 37: 12-15; February 1995.

Chlapecka, Thomas W, Stuart A. Schupack, Thomas W. Planek, Nancy Klecka, Gerald J. Driessen. "Bicycle Accidents and Usage Among Elementary School Children in the United States" (Technical Report). Chicago, IL: National Safety Council; 1975.

Consumer Product Safety Commission. Bicycle Use and Hazard Patterns in the United States. Washington, DC: Author; June 1994.

Cross, Kenneth D., Gary Fisher. 'A study of bicycle/motor vehicle accidents: Identification of problem types and countermeasure approaches" (Technical Report DOT-HS-803 315). Washington, DC: National Highway Traffic Safety Administration; 1977.

Federal Highway Administration. National Bicycling and Walking Study. (Technical Report FHWA-PD-94-023), Washington, DC: Author; 1994.

Forester, John. Bicycle Transportation. Cambridge, MA: MIT Press; 1983.

Forester, John. Effective Cycling. Cambridge, MA: MIT Press; 1993.

Kaplan, Gerald A. "Characteristics of the Regular Adult Bicycle User. " Masters Thesis, University of Maryland, 1975.

Rodgers, Gregory B. "The Characteristics and Use Patterns of Bicyclists in the United States." Journal of Safety Research 25(2): 83-96; 1994.

Rodgers, Gregory B. "Bicycle Helmet Use Patterns in the United States, A Description and Analysis of National Survey Data. " Accident Analysis and Prevention 27(1): 43-56; 1995a.

Rodgers, Gregory B. "Bicyclist Deaths and Fatality Risk Patterns. " Accident Analysis and Prevention; 27(2): 215-223; 19956.

Roland, H.E., W.W Hunter, J.R. Stewart, and B.J. Campbell. "Investigation of motor vehicle/bicycle collision parameters" (Technical Report DOT-HS-804 840). Washington, DC: National Highway Traffic Safety Administration; 1979.

Schubert, John. "Studies Show Bike Paths Have Dismal Accident Record." Bicycle Retailer and Industry News, pp. 1 and 22-23; October 1994.

Schupack, Stuart A., and Gerald J. Driessen. "Bicycle Accidents and Usage Among Young Adults: A Preliminary Study" (Technical Report). Chicago, IL: National Safety Council; 1976.

Skonecki, Peggy. "Comments on the Consumer Product Safety Commission Priorities Under Consideration for Fiscal Year 1987. " (mimeo); 1985.

Tinsworth, Debra K, Curtis Polen, and Suzanne Cassidy. "Bicycle-Related Injuries: Injury, Hazard, and Risk Patterns." International Journal for Consumer Safety 1(4): 207-220; 1994.

Wachtel, Alan, and Diana Lewiston, "Risk Factors for Bicycle-Motor Vehicle Collisions at Intersections." ITE Journal: 30-35; September 1994.

Top: John S. Allen's Home Page
Up: Table of Contents
Previous: Calculation of miles/hour figure
Next: John S. Allen's response

Contents 2001, John S. Allen
Last modified April 19, 2001