The "Bicycle Compatibility Index" - critique of implementation manual and workbook

John S. Allen

The University of North Carolina Highway Safety Research Center has conducted research[1] leading to publication by the Federal Highway Administration (USA) of an implementation manual for a so-called "Bicycle Compatibility Index" ("BCI").[2] This critique addresses mainly the implementation manual.

Really only a comfort index, and less than satisfactory as one

Level of service as conventionally defined describes traffic flow: unimpeded, congested, gridlocked. In a modern GIS database containing traffic data, an area-wide evaluation can be carried out automatically once the parameters of a formula are included in the database software. However, the BCI does not meet the requirement for such analysis, for several reasons.

• Though the BCI is described as a level-of-service concept, it was developed using regression analysis of ratings of bicyclists' comfort level on roads shared with motor traffic, paying no attention to bicyclists' mobility (and by implication, safe travel speed), as do conventional level of service concepts.
• More people want to travel at some times than at others. Weather and other factors also affect traffic flow. The usual application of the level of service concept is to quantify whether a road is congested, or might become congested, at one location and time or another. Specifying level of service only in terms of location is insufficient.
• The BCI formula uses measures of motorist speed and volume, and of travel lane and bike lane/shoulder width. Mobility of bicyclists depends to a great degree on traffic signal timing, directness of routing, terrain and congestion. Of these, only congestion is addressed in the BCI, indirectly in terms of motor traffic volume. There is no input reflecting bicycle or pedestrian traffic volume.
• The BCI formula is based on faulty assumptions about bicyclists' use of the roadway. For example, the formula does not consider that bicyclists may overtake in a lane to the left of stopped or slow traffic (for example, vehicles waiting to turn right, double-parked vehicles, or a bus picking up or discharging passengers). It only considers overtaking in the rightmost travel lane, parking lane or bike lane. This implies overtaking the slower or stopped traffic on its right, a move which is often hazardous, and in many cases illegal.
• In developing the BCI, bicyclists were asked to rate their comfort with overtaking motor vehicles displayed on a video screen. Wayne Pein has published a critique of this research method[3]. Viewing videos rather than real traffic does indeed, as the research report indicates, allow for more uniformity of data presented to experimental subjects, and for more convenient and faster testing. But there was no actual bicycle riding -- and worse yet, the camera was at the curb. There was no object in the road at the location where a bicyclist would be riding. For this reason, the locations of the motor vehicles were not the same as they would have been with a bicyclist present.
• Bicyclists' speed affects their safety, mobility, and comfort -- their level of service. The video clips used in the BCI development were made with a stationary camera, skewing the perception of motorists' speed relative to that of the prospective bicyclist subjects, and providing no input on the effects of bicyclists' speed on their need to process information and maneuver accordingly.
• There are mathematical, programming and descriptive errors in the BCI formula as embodied in a Microsoft Excel workbook included with the report.
• There is a variable for parking occupancy, but only a true/false choice as to whether there is a parking lane, and none to indicate whether there is parallel or angle parking. Available travel width, therefore, is undefined where there is a parking lane.
• The BCI rates a street with a bike lane or shoulder higher than one with a wide outside lane, though no difference in the safety and performance of these configurations has been demonstrated, and bike lanes adjacent to parking may encourage bicyclists to ride within range of opening doors of parked cars.[4]
• Most car-bike collisions involve turning and crossing movements, and most impediments to bicyclists' travel by motor vehicles also involve turning and crossing movements. The Index accounts for only one such movement, the motorist right turn.

Effect of linear model on the contribution of motorist speed

Increased speed of motorists' overtaking bicyclists may adversely affect the bicyclists' comfort and safety, but has little effect on bicyclists' mobility except as it affects their making left turns. On the other hand, when motorists are stopped or traveling slower than bicyclists, bicyclists' mobility, a real-world measure of level of service, can be seriously reduced.

The BCI makes a deduction in its rating linearly in proportion to motorist speed. The BCI therefore fails to measure the difference between bicyclist speed and motorist speed, as already mentioned, but it also rates streets higher as the motor traffic becomes more congested and increasingly impedes bicycle travel.

The implementation paper indicates that the BCI is supposed to be used only with outside lane motor vehicle traffic flow between 90 and 900 vehicles per hour; outside lane width between 3.0 and 5.6 meters; bike lane/shoulder width between 0.9 and 2.4 meters; and motorist speeds between 40 and 89 km/h (25 and 55 mph). The speed and flow ranges exclude many conditions typical of congested urban traffic.

If the person applying the BCI does not keep the data range limits in mind, it is very likely that out-of-range data will be input. If the formula is applied automatically to GIS data, results based on out-of-range data are almost certain to go undetected.

The following examples (for which data are given in a table, below), illustrate the inaccurate results the BCI gives under some typical urban traffic conditions that produce out-of-range data.

• In a right lane too narrow for motorists and bicyclists to share side by side, and with no other lane available for bicycle travel, the most favorable possible motorist speed, according to the BCI, is zero -- gridlock.
• In a wide outside lane, or an outside lane with a shoulder or bike lane to its right, the BCI also favors stopped motor traffic over moving traffic, even though bicyclists overtaking on the right may get "doored" from the left, or collide with jaywalking pedestrians and crossing vehicles hidden by the stopped traffic. These risks require that a cyclist travel very slowly and sometimes avoid moving forward at all, a limitation that does not occur if the motor traffic is moving faster than the bicycle traffic or if bicyclists overtake on the left.
• If the right lane has zero traffic flow, because it is gridlocked (queue waiting to turn right, etc.) but the next lane is flowing freely at a nice, comfortable speed for cyclists, 30 km/h, the level of service rating is D or lower. Clearly the authors never even addressed bicyclists' using any lane to the left of the rightmost travel lane.
• And if there is no motor traffic at all, the quantity in the formula for motor traffic speed is undefined and the formula fails.

The table below shows results which the formula gives under several conditions with data in range and out of range. Only the data fields which change are shown. The calculations including all data fields are included in a Microsoft Excel workbook[5] made available in connection with this article.

In the upper rows of the table, gridlocked traffic rates a higher LOS than moving traffic, whether or not there is room on either side of the gridlocked travel lane for bicyclists to overtake freely.

The bottom few rows illustrate the failure to account for overtaking on the left of stopped traffic. Note that a bike lane increases the rated level of service, despite the issues raised earlier about overtaking on the right.

Reformatted workbook allows comparisons

The Excel workbook provided as part of the BCI package[6] allows entry of data for only one location. This is a very inefficient and cumbersome use of computer spreadsheet software. I have reformatted the workbook to allow comparison of data for many locations. My revision does not change the formulas, and so it does not solve the problems with the BCI, but rather, helps to reveal them.

I have referenced cells in later workbook sheets to the equivalent cell in the first sheet -- so it is not necessary to type in the name of the location on all three sheets. I also have made cells fit more compactly around the text. Each worksheet will now fit nicely on a 1024 x 768 pixel computer screen at 100% display size. I have also placed the formulas on the new rows inside conditional statements, so error messages will not display unless a location is named.

I have highlighted obligatory data entry cells in yellow, cells in which data may be entered either as a number or as a formula in green, and results in blue. There is one formatting problem I have not corrected: there are four data entry variables in the second worksheet, though the names of the worksheets indicate that only the first sheet is for data entry.

Programming and descriptive errors

In the authors' workbook, one of the data entry cells in the second sheet, for curb lane truck volume, is supposed to default to a different value depending on the number of lanes in the street, but the workbook does not automatically set the default value.

Several header cells describe data entry in percentages where the data must be input as decimal fractions.

There is a serious programming error in the third sheet. The cell for speed (in red) refers to a header cell in the previous row. I have corrected this problem in the rows I have added.

The formula, as I have corrected it, carries an assumption that the 85th percentile speed of traffic is 15 km/hr faster than the speed limit, if an actual 85th percentile speed is not given. Though the report indicates that the 15 km/h adjustment can be altered, it must be altered in a formula rather than in a table cell. If an 85th percentile speed of zero (out of range but common in urban areas)  is entered, the formula will "correct" this to 15 km/h.

As already noted, the workbook does not post error messages if numbers are posted for conditions outside the range which the BCI is supposed to account for. As many common traffic conditions are outside its range, I have not attempted to correct this problem in my revised workbook, because I want users to be able to test the BCI with out-of-range data.

There are data cells with range limitations imposed by hard mathematical limits -- for example, parking occupancy can not be higher than 1.00 or lower than 0.00. These cells, too, would benefit from error trapping, to prevent input of invalid data.

I have highlighted the cells with outright programming and descriptive errors in red, and I have highlighted cells with range limitations that are within the range of normal traffic data in purple.

Conclusions and recommendations

While a level of service rating is a useful tool, the Bicycle Compatibility Index is not a level of service rating, and it is flawed as a comfort rating. It is remarkable that the FHWA has sanctioned this work in the light of other highly advanced research into traffic flow that it has generated.[7]

How could a better bicyclist level-of-service rating be developed? That question is difficult to answer. Let us consider the three primary elements of a rating: mobility, safety and comfort.

• An objective formula for mobility could be developed. though it would be considerably more complicated than for motor traffic. To give an example, the degree of mobility under congested conditions depends on lane width and on the composition of the motor traffic. Even in a completely stalled traffic jam, a bicyclist usually can move forward slowly. And as described above, traffic signal timing, directness of routing and terrain also strongly affect bicyclists' mobility, and in ways they do not affect mobility of motorists. On the other hand, speed limits rarely affect bicyclists, who rarely travel as fast as the posted speed limit.
• Safety of streets also can be at least roughly evaluated based on the results of the research record. Safety, however, depends very heavily on the level of skill of the bicyclist, and so no single safety rating is possible. Roadway conditions that are reasonably safe for an adult bicycle commuter, tourist or fitness rider -- particularly, one with vehicular cycling skills -- are far less safe for children and casual adult bicyclists. Conversely, conditions which are reasonably safe for casual cyclists and children may not be so for faster, experienced adult cyclists. This is particularly so on multi-use paths, where the presence of pedestrians makes it unsafe to travel at speeds that fit adult bicyclists easily can attain.
• Comfort is the most troublesome variable. Comfort depends primarily on the perceived level of safety, however, perceptions correlate poorly with actual levels of safety. Perceptions often are at odds with reality, as evidenced by many examples of bicyclist behavior such as ducking between parked cars, riding opposite the flow of traffic, and riding at high speeds on trails that are less suitable than roads for such speeds.

Workable ratings of bicycle routes have been produced by experienced bicyclists. Though objective factors such as traffic counts, motor vehicle speeds, lane widths and steepness of climbs play an important part in such ratings, there are other important factors which, though objective, do not show up in the usual highway department databases -- for example, quality of pavement near the edge of the road, or the availability of roadside amenities. A good route surveyor will be familiar with the type of bicyclist who is to use a route -- indeed, must have experience with the type of cycling that is to be accommodated. The surveyor also must travel the route to experience it firsthand. In the end, a rating may be based to a large degree on factors that were not even considered before surveying the route, and which would require an impractical, complicated data structure to quantify. The rating therefore becomes to some degree subjective. Despite these problems, the success of ratings developed by and for bicyclists is shown, for example, by the low crash rate on the Bikecentennial TransAmerica route.[8] Such ratings are, in the author's opinion, the best ones available.

Footnotes

[1] University of North Carolina Highway Safety Research Center (1998) Development of the Bicycle Compatibility Index: A Level of Service Concept, Final Report, FHWA-RD-98-092. http://safety.fhwa.dot.gov/fourthlevel/pdf/bcifinalrpt.pdf or http://www.hsrc.unc.edu/research/pedbike/98072/index.html

[2] University of North Carolina Highway Safety Research Center (1998) The Bicycle Compatibility Index: A Level of Service Concept, Implementation Manual, FHWA-RD-98-095. http://safety.fhwa.dot.gov/fourthlevel/pdf/bci.pdf or http://www.hsrc.unc.edu/research/pedbike/98095/index.html

[3] Pein, Wayne. (2003) Critique of FHWA Bicycle compatibility Index. http://humantransport.org/bicycledriving/library/critique_BCI.pdf

[4] See article summarizing dooring research.

[5] .zip compressed revised Microsoft Excel workbook with examples, on this site.

[6] Microsoft Excel workbook in the Bicycle Compatibility Index: A Level of Service Concept, Implementation Manual, http://www.hsrc.unc.edu/research/pedbike/98095/BCILOS/BCIEQUTN.XLS

[7] See, for example, U.S. Federal Highway Administration Turner-Fairbank Highway Research Center. Traffic Flow Theory, a State of the Art Report, Chapter 2. http://www.tfhrc.gov/its/tft/chap2.pdf

[8] See Burden and Burgess (1977), Bicycle Safety Highway Users Information Report, republished on this Web site.

Acknowledgements: I thank Wayne Pein and John Forester for their helpful comments.