Published in transpo.philica.com
In a recent study (Walker, in press) we demonstrated using an instrumented bicycle that various factors including riding position and helmet use influenced the passing proximities of drivers as they overtook a bicyclist. He we present a short extra analysis of these data revealing clear time of day effects. Overtaking data were collected at various times between 0700 and 1800 during May and June in the United Kingdom. We show an increasing linear relationship between time of day and mean overtaking proximity. An implication of this is that there are clear differences between the morning rush-hour—where drivers tended to give the rider less space—and the evening rush-hour where they left more leeway. The results are discussed in terms of their implications for drivers, town planners and employers.
Understanding factors which affect the amount of space drivers leave as they pass bicyclists has the scope to inform our understanding of the likelihood of overtaking collisions occuring. In a recent study (Walker, in press) we collected data from over 2500 vehicles as they overtook an instrumented bicycle. This revealed effects of various factors on the proximity of these vehicles, including the bicycle's position in its lane and whether the rider wore a helmet or not. Overall, these factors accounted for 8% of variance in raw overtaking proximities and raised the ability to predict particularly close overtaking events (operationally defined as the bottom quartile) to substantially above chance.
One factor not directly assessed in Walker (in press) was the time of day of the overtaking events. As well as the inherent interest in knowing whether there are time of day effects on how much space drivers leave, if an effect were seen such that drivers left less space at some times of day than others this might have relevance for designing interventions. For example, if peaks were seen at the times children are taken to and from school, this would suggest natural groups of people who could be targetted in advertising.
This is a secondary analysis of data from Walker (in press) — the reader is referred there for information on how these were collected. This analysis included data from 2355 overtaking vehicles which included a mixture of vehicle types. The overtaking events were collected at various times of day between 0700 and 1800 on various days in May and June 2006. The riding position (distance from the edge of the road) was changing frequently, so each time band in these data should contain proximities from several different riding positions, thus largely cancelling out the effects of this factor. The same goes for the other factors manipulated in the main study, including helmet-wearing.
The mean overtaking proximities (calculated as the minimum proximity from the outermost part of the bicycle) for each for one-hour bands between 0700 and 1800 are shown in Figure 1. From this it can be seen that overtaking proximity tended to increase as the day progressed such that drivers tended to pass closer early in the morning but leave more room in the afternoon. Overall, the data shown in Figure 1 fit relatively well to a linear function with R2=.59.
Figure 1 — Relationship between time of day (one-hour bands) and mean overtaking proximity
The equation describing the regression line is y=1.20+0.017x, where x is the beginning of the time slot (in hours, so 7 = 0700) and y is overtaking proximity in metres. This formula predicts that the difference between the mean overtaking proximity for the 0700-0800 time slot and the 1700-1800 time slot is 17 cm.
The data from this experiment show a tendency for drivers to give more leeway when overtaking a bicyclist later in the day than early in the day. Perhaps the most important implication of this is that people show somewhat different behaviour in the two rush-hours: drivers seem to get closer in the morning rush-hour than in the evening. This is not particularly surprising, as many workers and people taking children to school have much more fixed deadlines for arrival in the morning than in the evening, and so are likely to exhibit more impatience and more risk-taking as a consequence. Nevertheless, even though this result is not surprising, it is useful to have definite data on the behaviour as this can inform drivers and traffic planners of how people are behaving. The information should also be of interest to employers, who may as a result see benefits to allowing flexible working arrangements.
Walker, I. (in press). Drivers overtaking bicyclists: Objective data on the effects of riding position, helmet use, vehicle type and apparent gender. Accident Analysis and Prevention.
This research was supported by a grant from the Engineering and Physical Sciences Research Council of the UK. The instrumented bicycle was designed and built by Jeff Brewster of the University of Bath Mechanical Engineering department.
Information about this Article
Peer-review ratings (from 5 reviews, where a score of 100 represents the ‘average’ level):
Originality = 217.84, importance = 229.33, overall quality = 219.23
This Article was published on 30th September, 2006 at 13:53:52 and has been viewed 16328 times.
This work is licensed under a Creative Commons Attribution 2.5 License.
The full citation for this Article is:|
Walker, I. (2006). Time of day effects in drivers’ overtaking of bicycles. PHILICA.COM Article number 24.
1 Peer review [reviewer #9049] added 3rd October, 2006 at 13:06:25
R^2=.59 indicates that the data dose not fit a linear function very well. To fit a linear function R^2=(something close to 1).
Also it seams from the graph that a very limited amount of data was taken. 11 trails is a farley small quantity to make good recommendations and predictions from. If there wear more trials perhaps it could be indicated some how.
Originality: 7, Importance: 5, Overall quality: 5
2 added 3rd October, 2006 at 14:42:04
As a psychologist, I can affirm that R^2=.59 is really a pretty good fit when discussing human behaviour.
It’s important to realise that there aren’t just 11 trials here as the above review suggests — there are actually 2355 trials and this is indicated in the text; each data point is the average of many trials, thereby cancelling out quite a lot of the “random” variation between events to reveal typical proximities for each time slot.
3 Peer review [reviewer #296] added 3rd October, 2006 at 18:14:32
the data are very interesting. The R2 of .59 is certainly a useful approximation of linearity. As the author indicates, this value with behavioural research of this kind is unusually high. An interesting paper.
Originality: 7, Importance: 7, Overall quality: 7
4 Additional peer comment [reviewer #9049] added 4th October, 2006 at 17:05:18
Reading over the report again, I found the quantity of trials. It is indicated, though I some how missed it on my first reading.
Second, In my studies though limited and in the Civil Engineering field, liner regression data needs be closer to 1 than .59, though in other fields, or brooder studies that my very well be a good linear approximation.
It is a very interesting topic, and well written.
5 Peer review [reviewer #61593] added 8th October, 2006 at 19:02:51
From the plot is it not clear whether the regression was fitted to the mean data points or to all the data (i.e. is the line fit to 11 points or to 2355 points?) It seems like it really ought to be fit to all the data points individually. The plot could also use error bars, as it is unclear just how much variance was actually present.
Have you done any other statistics? Is there a significant difference between the passing distances at, say, 7:00 and 17:00? A t-test would be a quick enough way to tell.
Finally, it would be interesting to include (if the data was recorded) some measure of the number of cars on the road at each point. It could be that it is not time, per se, that matters but that there are an increasing number of cars on the road later in the day, leaving less room for cars to clear bicycle traffic. It’s unclear whether that was implied in the discussion.
Originality: 5, Importance: 5, Overall quality: 5
6 Peer review [reviewer #997] added 9th October, 2006 at 17:39:47
I wonder whether the linear regression is an adequate method for estimation of these results. Maybe the curve is “wavy” withsome period and changes during the day. It would get along with the observation that rush hours make the distance shorter between the bicycle and the car.Thus, drivers being out of rush hours would be more careful of bicyclist. That would suggested the change in the school leaving by the children - out of the rush hours. Nice result.
Originality: 7, Importance: 7, Overall quality: 7
7 added 10th October, 2006 at 11:05:50
I have just read a new paper:
Kim, J., Kim, S., Ulfarsson, G.F. & Porello, L.A. (in press). Bicyclist injury severities in bicycle-motor vehicle accidents. Accident Analysis and Prevention
which, based on analysis of police accident records from North Carolina, identified the period from 0600-0959 as a significant predictor of collision severity (collisions in this period were more likely to be fatal than in other periods). As the present study shows drivers get closer in this period, this lends some support to the idea that drivers getting closer will be a risk factor for collisions (NB it is known that collisions during overtaking frequently have severe outcomes - see Walker, in press).
Re comment #6 above, I did try other regression models up to a fourth-order polynomial, and although the more complex models fitted slightly better (as one would expect), the increase in fit was rather trivial (linear R^2 = .59, quadratic R^2 = .64, cubic R^2 = .64, fourth-order R^2 = .65). None of the other models’ patterns made any real sense as an explanation of the traffic behaviour (they still looked largely linear on the graph). As such, the linear model was the most parsimonious and sensible fit.
8 Peer review [reviewer #7116] added 20th October, 2006 at 16:03:38
This is an interesting idea for an analysis, but I was hoping for a little more discussion. In particular, it would seem that traffic density varies considerably throughout the day, as does traffic makeup (I would assume that between 0900 and 1500 there is a much higher proportion of commercial vehicles, for example, and fewer commuters.) This article does not state whether or not each hourly observation is composed of the same number of overtaking events—I would imagine that they are not, with the mid-day observations being sparser. Standard deviations on the observations themselves would probably help ally some of the sense that this data is “patternless.”
Originality: 5, Importance: 4, Overall quality: 4