The effect of a yellow bicycle jacket on cyclist accidents by Harry Lahrmann et al.
Aalborg Uni, Denmark, published in Safety Science. (August 2017).
This paper’s conclusion, which stirred much media interest at the time, is that: “This
randomised controlled study delivered strong evidence that cyclists are protected against
multiparty accidents when wearing a bright-coloured jacket.”
The authors deduced this by analysing reports from 6,793 volunteer, regular cyclists
recruited across Denmark, who were randomly assigned either to a test group (with
jackets = HVZ) or control group (no jackets = NJK). (My highlighting).
The test and control group shared similar characteristics, the typical
destination was work/education and the study lasted from 1
November 2012 to 31 October 2013.
The jacket was fluorescent yellow, with reflective strips.
The researchers collected monthly reports from their participants on
both single ‘personal injury accidents’ (PIAs, where only the cyclist
was involved) and on multi-party PIAs (involving a motor vehicle).
The incidents had to be on public roads and meet at least one of the following criteria:
• the cyclist had to be in physical contact with a counterpart;
• toppled and/or injured as a consequence of the counterpart’s behaviour (including
damage to the cyclist’s belongings, even if no physical contact had occurred);
• toppled/injured whilst riding without the involvement of others.
To calculate the ‘accident rate per month’ (AR), the researchers divided the number of
PIAs by the total number of ‘person months’ cycled by the participants – NJK cycled
slightly more person months than HVZ (around 38,500 to 37,500). From this, they
worked out the accident rate ratio (ARR).
This led the researchers to state that:
• The AR for all PIAs (single and multi-party) was 47% lower for people wearing hi-vis
jackets than those not wearing jackets; and 55% lower for incidents involving cyclists
and motor vehicles.
• They had identified a ‘response bias’ (see below) and adjusted for it. This reduced the
47% to 38%.
Note: this study’s hypothesis was clearly pro-hi-vis: “… that the use of high-visibility
clothing on the upper body of a cyclist would improve cyclists' visibility and consequently
lead to a reduction in the number of multiparty PIAs.” Also, the trial was funded by an
organisation who strongly advocates hi-vis and ordinarily supply jackets to cyclists,
according to its website at the time).
The following weaknesses and limitations are also worth noting:
• Although this was a large, year-long study, representing over 76,000 ‘person months’,
the sample of PIAs that met the researchers’ criteria was very small - just 302:
Test group Control group
Multiparty PIAs Single PIAs Multiparty PIAs Single PIAs
43 (35%) 80 (65%) 83 (46%) 96 (54%)
(Apart from anything else, this implies that cycling in Denmark is not unduly risky!).
• This was a ‘non-blind’ study. As a result, it could well have suffered from ‘response
bias’ because the participants knew whether they were in a test or control group and
what they were testing. As such, the authors admit that “… it is possible that the test group [HVZ] reported slightly fewer PIAs than they should because they wanted to
prove the safety effect of the bicycle jacket. This source of bias is well-known both in
psychology…”. In fact, they point to the weaknesses of a non-blind trial several times
(e.g. “The internal validity of the trial is affected by the fact that the study is non-
blinded.”)
• The authors speculate: “It is likely that risk adaptation compensates for the effect of
the increased visibility, i.e. cyclists become less careful when they feel more
protected.” They suggest that this helps cancel out the participants’ response bias
(along with the finding, from a British study, that drivers pass helmeted cyclists more
closely). (Looked at another way, this seems tantamount to claiming that wearing a
hi-vis jacket saves people from the adverse effects of wearing a hi-vis jacket; and that
these added hazards simply gave HVZ yet more incidents not to report).
• Yet the authors argue elsewhere that all their volunteers could well have been more
likely than the general population to be risk averse because they had signed up to
use a jacket expected to improve their road safety (NJK were promised a free jacket
at the end of the trial). This line of thought allows them to contend that the jacket
could have “… a higher effect for the average cyclist, compared to the effect on the
group in this study”, i.e. because the average cyclist in the external world may be less
risk averse than the people recruited for the trial. (Although it could equally well be
argued that anyone who buys themselves a hi-vis jacket is as likely to be as risk
adverse – or have been encouraged to become as risk adverse – as someone who
signs up for a hi-vis jacket trial; and that recommending hi-vis to the wider, less risk
adverse population (if it is indeed less risk adverse) could lead to more people ‘risk
adapting’ and putting themselves in greater jeopardy. ‘Risk adaptation’ is, after all, a
known phenomenon often used to caution against putting too much faith in other
protective accessories such as helmets).
• The authors did not account for cycle mileage, despite acknowledging that: “Apart
from the jacket use, the mileage driven is an important factor affecting the number of
accidents. It is generally expected that the higher the mileage (i.e. exposure), the
higher the accident number. Although a recording of the mileage could provide insight
into this correlation, the study did not record the participants’ cycling mileage in the
monthly questionnaires.”
• Although the participants agreed to cycle 3x a month, nowhere do the authors
confirm that this is what most of them did. Nonetheless, they base their calculations
on whole months in which a participant cycled, so do not factor in how many trips
they did (another good way of determining exposure). (Note that NJK cycled a little
more than HVZ, at least in terms of months).
• There were no months when 100% of HVZ riders wore their jackets “on a random
day”: this ranged from 84% in November 2012, to 25% in July 2013. To get round
this, the researchers asked all HVZ riders to specify whether they wore other
yellow/bright clothing instead from April 2013, but did not give the control group this
option. It is not impossible, therefore that some NJK riders were sometimes ‘brightly’
clothed too. (Whether they were wearing their jacket/bright clothing or not, all
qualifying HVZ PIAs were always logged against HVZ).
• In numeric terms, HVZ reported significantly fewer single PIAs than NJK (no motor
vehicle involved). The authors say: “the bicycle jacket was not expected to affect the
number of single accidents” and put this anomaly down to ‘response bias’. The
authors therefore adjusted for this.Oddly, though, they say that the proportion of single PIAs was higher amongst HVZ
than for NJK (65% v 54%). What’s more, they state: “Among respondents who
reported a high jacket use, the proportion of single PIAs was higher (72%) than
among respondents with low jacket use (59%). Finally, among those who stated that
they wore the jacket during the accident, the proportion of single PIAs was higher
(69%) compared to the proportion among those who reported that they did not (56%).
However, the two latter comparisons did not reach statistical significance.” ‘Risk
adaptation’ is a possible explanation for this (see above), as is response bias
(perhaps because HVZ, keen to prove the value of jackets, might have been tempted
to downplay multi-party incidents and (overly) willing to report those where driver
detection was irrelevant).
• Although the authors do not highlight this, their figures suggest that HVZ’s incidents
tended to be more serious than NJK: not only was a higher proportion of HVZ multi-
party PIAs “reported by [sic] the police” and to insurance companies, but the victims
were more likely to seek treatment from the emergency services, rather than just a
doctor alone. Clearly, these were incidents they couldn’t gloss over:
• Finally, the authors finish with a couple of caveats of their own:
“… the effect will most likely decrease if an increasing number of cyclists start using a
bright-coloured bicycle jacket because the jacket will not attract as much attention when
more cyclists use it.”
“… other road users’ risk may increase when attention is directed to cyclists with bright-
coloured jackets at the expense of other cyclists”.
This study is the first randomised controlled trial (RCT) of the safety effect of high-visibility bicycle clothing. The hypothesis was that the number…
www.sciencedirect.com
So yeah, open and closed case...
Be careful of using AI on research.
pmc.ncbi.nlm.nih.gov
