Bicycle DRLs - Actually reading the paper, actually looking at the

Another thread mentions Madsen et. al., "Safety Effects of Permanent
Running Lights for Bicycles", Accident Analysis & Prevention 50 (2013)
820-829. Scharf said "Read the study," But I'm pretty sure he's read
only the abstract. I downloaded the study and skimmed it a few days
ago. Last night I took the time to read it in detail. Here's what I found.

One might say the study is a promotion of the Reelight flashing light
system, powered by magnets on wheel spokes.
https://www.reelight.com/en/ Those are the lights "tested" in the
study. And there are enough weirdnesses in the results to indicate that
this is more a promotion than a serious study. Here are details and
(warning!) data analysis.

They advertized they wanted to do a study and asked who in Odense wanted
to participate in a light study, luring participants by promising free
lights. They got over 18000 applicants, about 10% of the population.
That's the first "self-selection" clue. Those who already had
sufficient lighting, those who didn't care about lights and those who
simply didn't hear about this would be excluded.

They used a random process to select 1845 of those people to get lights
immediately and 2000 people to act as a control group, promising the
controls free lights if they stayed in the study for a full year.
Obviously, those who got the lights knew they had them, so there's
nothing "blind" about this study. Placebo effects are certainly
possible, and certainly did occur - see below. Some people dropped out
in the course of the 1-year time period. 3306 completed the entire year.

The study scheme echoed one "Danger! Danger!" study from Portland, in
that they emailed the paricipants every two months (Portland used one
month) to ask about any bike "accidents" and "injuries." This is so
participants wouldn't forget any of those. It seems to me that if an
"accident" or "injury" is forgettable, it should be below the radar, so
to speak. But they didn't want to miss even the most minor "incident."

They defined "accident" as "an incident where you are forced off your
bike and/or crashed either by yourself or due to collision OR
NEAR-COLLISION with others." So actually, a near miss counted as an
"accident"! And I suppose "forced off you bike" could be coming to a
stop and having the bike topple as you stand over it. IOW, they are
exaggerating dangers, as usual. More on that later.

The paper does not define "injury," but recall that the Portland paper
counted _any_ injury. I assume this one did too. Broken fingernail?
Injury! But the injuries were still extremely rare.

Overall, there were 39672 "man-months" of bike use recorded. There was
no data on number of trips by bike, km traveled or hours exposure;
therefore all "rates" are "per man-month." And the numbers are very low
indeed.

Exactly how low is a bit difficult to tell. Examining data from three
of the paper's tables and doing some minor number crunching gave three
different answers for injuries per man-month. Table 6 works out to
0.00103 injuries per man-month; table 8 claims 0.00489 instead; and
table 9 implies 0.00318 instead. There may be something I'm missing;
but whichever is correct, all those figures are extremely low! The
worst yields an average of 204 months or 17 years of riding per
"injury," including (I assume) very minor and forgettable injuries.
Only 41 injuries were seen by medical people, and apparently none were
serious. Does that sound dangerous?

Similarly, accident rates per man-month were very low - only 261
"accidents" (including near-misses!) from 39672 man-months exposure.
That's less than 0.007 "accidents" per man-month, including (I assume)
putting one's foot down if one drops his bike.

Regarding the type of "accident": It would certainly be good to collect
data on what specific type of "accident" occurred or were "prevented."
But there's absolutely no mention of that in the paper. They give
separate numbers for winter vs. summer, daylight vs. darkness, solo vs.
"multiparty," etc. but nothing at all about whether they were cars
pulling out, left crosses, right hooks, cyclists running stop signs,
pedestrians jumping out, bike-bike collisions or anything else. This
makes actual analysis of causes impossible.

However, since "solo" accidents are listed, those must be simple falls
due to skidding, running into objects or pavement defects, simply losing
one's balance, etc. REELIGHTS SUPPOSEDLY PREVENTED ABOUT 25% OF SOLO
CRASHES! The authors acknowledge that this is flat impossible, and
evidence that the riders with the lights are cooking their reports, i.e.
not reporting some accidents.

The authors attempt a crude correction for this effect, but I see no
justification for its accuracy - certainly not down to the three
significant figure reporting of results. Still, they tried using their
correction (i.e. fudge factor) to improve their results, and (bottom of
page 827) said "... a result of the applied correction... has the effect
that NONE OF THE ESTIMATED DIFFERENCES IN THE INCIDENCE RATES ARE
[STATISTICALLY] SIGNIFICANT, the incidence rates for multiparty
accidents being the only but very important exception."

OK, so what's the rate of those multiparty (bike-ped, bike-bike,
bike-car) "accidents" (including near misses, which are not really
accidents? It's 0.00178 "accidents" per man-month. That means an
average of 562 months or 47 years of riding between each of those
"accidents" (including near misses).

How does this relate to Scharf's and Joerg's claims that DRLs make a
huge, obvious difference? As usual, the actual numbers indicate a very
strong probability of bull****. And the bull**** applies to the paper's
conclusions regarding Reelights' magic effectiveness as well.

This paper, like so many others, emphasizes changes in extremely rare
hazards. Even if Reelights or other DRLs reduce "accidents" (including
near misses) from - say - once every 47 years to once every 88 years on
average, which is what they claim - is it really worth yelling "Danger!
Danger! Buy our lights!"

Well, I suppose if you own the Reelight company, or if you get a
commission by selling Chinese flashlights on your web pages, you'll say
it's worth it. Less biased people probably have other opinions. Which
is why, contrary to Scharf's statements, only a minuscule percentage of
cyclists use daytime running lights.

--
- Frank Krygowski

Hell, Franki-boy, your complaint is longer than the paper. And you tell us only what we know already, you hate all studies that don't blow your trumpet. When will you get a new tune (or learn to sing your old tune in tune)?

Andre Jute

On Monday, July 13, 2015 at 4:53:57 PM UTC+1, Frank Krygowski wrote:
Another thread mentions Madsen et. al., "Safety Effects of Permanent
Running Lights for Bicycles", Accident Analysis & Prevention 50 (2013)
820-829. Scharf said "Read the study," But I'm pretty sure he's read
only the abstract. I downloaded the study and skimmed it a few days
ago. Last night I took the time to read it in detail. Here's what I found.

One might say the study is a promotion of the Reelight flashing light
system, powered by magnets on wheel spokes.
https://www.reelight.com/en/ Those are the lights "tested" in the
study. And there are enough weirdnesses in the results to indicate that
this is more a promotion than a serious study. Here are details and
(warning!) data analysis.

They advertized they wanted to do a study and asked who in Odense wanted
to participate in a light study, luring participants by promising free
lights. They got over 18000 applicants, about 10% of the population.
That's the first "self-selection" clue. Those who already had
sufficient lighting, those who didn't care about lights and those who
simply didn't hear about this would be excluded.

They used a random process to select 1845 of those people to get lights
immediately and 2000 people to act as a control group, promising the
controls free lights if they stayed in the study for a full year.
Obviously, those who got the lights knew they had them, so there's
nothing "blind" about this study. Placebo effects are certainly
possible, and certainly did occur - see below. Some people dropped out
in the course of the 1-year time period. 3306 completed the entire year.

The study scheme echoed one "Danger! Danger!" study from Portland, in
that they emailed the paricipants every two months (Portland used one
month) to ask about any bike "accidents" and "injuries." This is so
participants wouldn't forget any of those. It seems to me that if an
"accident" or "injury" is forgettable, it should be below the radar, so
to speak. But they didn't want to miss even the most minor "incident."

They defined "accident" as "an incident where you are forced off your
bike and/or crashed either by yourself or due to collision OR
NEAR-COLLISION with others." So actually, a near miss counted as an
"accident"! And I suppose "forced off you bike" could be coming to a
stop and having the bike topple as you stand over it. IOW, they are
exaggerating dangers, as usual. More on that later.

The paper does not define "injury," but recall that the Portland paper
counted _any_ injury. I assume this one did too. Broken fingernail?
Injury! But the injuries were still extremely rare.

Overall, there were 39672 "man-months" of bike use recorded. There was
no data on number of trips by bike, km traveled or hours exposure;
therefore all "rates" are "per man-month." And the numbers are very low
indeed.

Exactly how low is a bit difficult to tell. Examining data from three
of the paper's tables and doing some minor number crunching gave three
different answers for injuries per man-month. Table 6 works out to
0.00103 injuries per man-month; table 8 claims 0.00489 instead; and
table 9 implies 0.00318 instead. There may be something I'm missing;
but whichever is correct, all those figures are extremely low! The
worst yields an average of 204 months or 17 years of riding per
"injury," including (I assume) very minor and forgettable injuries.
Only 41 injuries were seen by medical people, and apparently none were
serious. Does that sound dangerous?

Similarly, accident rates per man-month were very low - only 261
"accidents" (including near-misses!) from 39672 man-months exposure.
That's less than 0.007 "accidents" per man-month, including (I assume)
putting one's foot down if one drops his bike.

Regarding the type of "accident": It would certainly be good to collect
data on what specific type of "accident" occurred or were "prevented."
But there's absolutely no mention of that in the paper. They give
separate numbers for winter vs. summer, daylight vs. darkness, solo vs.
"multiparty," etc. but nothing at all about whether they were cars
pulling out, left crosses, right hooks, cyclists running stop signs,
pedestrians jumping out, bike-bike collisions or anything else. This
makes actual analysis of causes impossible.

However, since "solo" accidents are listed, those must be simple falls
due to skidding, running into objects or pavement defects, simply losing
one's balance, etc. REELIGHTS SUPPOSEDLY PREVENTED ABOUT 25% OF SOLO
CRASHES! The authors acknowledge that this is flat impossible, and
evidence that the riders with the lights are cooking their reports, i.e.
not reporting some accidents.

The authors attempt a crude correction for this effect, but I see no
justification for its accuracy - certainly not down to the three
significant figure reporting of results. Still, they tried using their
correction (i.e. fudge factor) to improve their results, and (bottom of
page 827) said "... a result of the applied correction... has the effect
that NONE OF THE ESTIMATED DIFFERENCES IN THE INCIDENCE RATES ARE
[STATISTICALLY] SIGNIFICANT, the incidence rates for multiparty
accidents being the only but very important exception."

OK, so what's the rate of those multiparty (bike-ped, bike-bike,
bike-car) "accidents" (including near misses, which are not really
accidents? It's 0.00178 "accidents" per man-month. That means an
average of 562 months or 47 years of riding between each of those
"accidents" (including near misses).

How does this relate to Scharf's and Joerg's claims that DRLs make a
huge, obvious difference? As usual, the actual numbers indicate a very
strong probability of bull****. And the bull**** applies to the paper's
conclusions regarding Reelights' magic effectiveness as well.

This paper, like so many others, emphasizes changes in extremely rare
hazards. Even if Reelights or other DRLs reduce "accidents" (including
near misses) from - say - once every 47 years to once every 88 years on
average, which is what they claim - is it really worth yelling "Danger!
Danger! Buy our lights!"

Well, I suppose if you own the Reelight company, or if you get a
commission by selling Chinese flashlights on your web pages, you'll say
it's worth it. Less biased people probably have other opinions. Which
is why, contrary to Scharf's statements, only a minuscule percentage of
cyclists use daytime running lights.

--
- Frank Krygowski

Frank Krygowski wrote:

:They defined "accident" as "an incident where you are forced off your
:bike and/or crashed either by yourself or due to collision OR
:NEAR-COLLISION with others." So actually, a near miss counted as an
:"accident"! And I suppose "forced off you bike" could be coming to a
:stop and having the bike topple as you stand over it. IOW, they are
:exaggerating dangers, as usual. More on that later.

That's a perfectly reasaonble definition. If a car pulls out in front
of you, and you avoid the car, but hit the curb and crash, the car has
caused an accident, even though you didn't collide with it. (That,
exactly, has happened to me.)


--
This is not a randomly numbered sig.

On 13/07/2015 2:55 PM, David Scheidt wrote:
Frank Krygowski wrote:

:They defined "accident" as "an incident where you are forced off your
:bike and/or crashed either by yourself or due to collision OR
:NEAR-COLLISION with others." So actually, a near miss counted as an
:"accident"! And I suppose "forced off you bike" could be coming to a
:stop and having the bike topple as you stand over it. IOW, they are
:exaggerating dangers, as usual. More on that later.

That's a perfectly reasaonble definition. If a car pulls out in front
of you, and you avoid the car, but hit the curb and crash, the car has
caused an accident, even though you didn't collide with it. (That,
exactly, has happened to me.)



That has exactly happened to me too and isn't that uncommon IMO. It
seems that when you have a point to prove, nothing short of death will
be counted as dangerous.

On 7/13/2015 11:55 AM, David Scheidt wrote:
Frank Krygowski wrote:

:They defined "accident" as "an incident where you are forced off your
:bike and/or crashed either by yourself or due to collision OR
:NEAR-COLLISION with others." So actually, a near miss counted as an
:"accident"! And I suppose "forced off you bike" could be coming to a
:stop and having the bike topple as you stand over it. IOW, they are
:exaggerating dangers, as usual. More on that later.

That's a perfectly reasaonble definition. If a car pulls out in front
of you, and you avoid the car, but hit the curb and crash, the car has
caused an accident, even though you didn't collide with it. (That,
exactly, has happened to me.)

In San Francisco I see that kind of "near miss" often, though I prefer
to call it a "near hit."

With a DRL the driver behavior is visibly better. You can literally see
them hesitate, then yield the right of way. Does it always work? No.
Today a delivery truck did a right hook in front of me in Cupertino as
it turned into the Apple campus. It really wasn't a "near hit" but it
was still annoying.

The bottom line here is this. Frank doesn't use a DRL. Hence he believes
that DRLs are ineffective. Any study that shows that belief to be false
must have some terrible flaw in how it was conducted.

Or change "DRL" to "helmet." The schtick is always the same.

On 7/13/2015 12:03 PM, Duane wrote:
On 13/07/2015 2:55 PM, David Scheidt wrote:
Frank Krygowski wrote:

:They defined "accident" as "an incident where you are forced off your
:bike and/or crashed either by yourself or due to collision OR
:NEAR-COLLISION with others." So actually, a near miss counted as an
:"accident"! And I suppose "forced off you bike" could be coming to a
:stop and having the bike topple as you stand over it. IOW, they are
:exaggerating dangers, as usual. More on that later.

That's a perfectly reasaonble definition. If a car pulls out in front
of you, and you avoid the car, but hit the curb and crash, the car has
caused an accident, even though you didn't collide with it. (That,
exactly, has happened to me.)



That has exactly happened to me too and isn't that uncommon IMO. It
seems that when you have a point to prove, nothing short of death will
be counted as dangerous.

What I don't understand is what is the upside of the denial.

Everyone is aware of the benefits of DRLs for motorcycles.

Everyone is aware of why some countries mandate DRLs on cars.

Is there any reason why someone would think that a study that also
showed DRLs to also be effective for bicycles must somehow be so bogus
that it's necessary to desperately search for some fatal flaw?

No one is forcing anyone to use DRLs on their bicycle (though for
motorcycles and cars they are sometimes mandatory). It's just one fairly
easy way to make yourself much more conspicuous to motor vehicles in
certain riding environments. In most cases the flashing DRL capability
is already present in whatever lights you use at night (except for most
dynamo lights) so there is no extra expense other than charging the
batteries a little more often.

On 7/13/2015 2:55 PM, David Scheidt wrote:
Frank Krygowski wrote:

:They defined "accident" as "an incident where you are forced off your
:bike and/or crashed either by yourself or due to collision OR
:NEAR-COLLISION with others." So actually, a near miss counted as an
:"accident"! And I suppose "forced off you bike" could be coming to a
:stop and having the bike topple as you stand over it. IOW, they are
:exaggerating dangers, as usual. More on that later.

That's a perfectly reasaonble definition. If a car pulls out in front
of you, and you avoid the car, but hit the curb and crash, the car has
caused an accident, even though you didn't collide with it. (That,
exactly, has happened to me.)

Their sentence said "...forced off your bike and/or crashed..."
That "or" says that one could be forced off without crashing. If a near
miss causes a dismount but no crash, is it really an accident?

I have a Bike Friday, a folding bike with no top tube. That means when
I stand astride the bike, it's somewhat prone to tip over because it
lacks the stabilizing effect of a top tube between my legs; and it has
tipped over several times that way. I would never report that as an
accident.

--
- Frank Krygowski

On 7/13/2015 2:28 PM, Andre Jute wrote:
Hell, Franki-boy, your complaint is longer than the paper.

Wrong, as usual.

--
- Frank Krygowski

On 7/13/2015 3:36 PM, sms wrote:

What I don't understand is what is the upside of the denial.

Some people value accuracy and truth, whether or not you do.


--
- Frank Krygowski

On 7/13/2015 3:29 PM, sms wrote:

The bottom line here is this. Frank doesn't use a DRL. Hence he believes
that DRLs are ineffective. Any study that shows that belief to be false
must have some terrible flaw in how it was conducted.

Address the points I noted in the paper. I gave numbers and statements
quoted from the paper, plus some further data analysis. If you see
errors, why not post corrections? (Yes, I know... to do that, you'd
have to actually read the paper!)

The authors themselves accept that DRLs cannot possibly reduce solo
accidents; yet the DRL users claimed the lights magically reduced solo
accidents by 24%. The authors realized that indicated bias, and
therefore they attempted a (somewhat clumsy) mathematical correction to
their raw data to account for DRL users obvious bias.

That takes the purported DRL benefit down to a mere 19%, and turns a
negligible "accident" (not crash) rate of one per 47 years into a more
negligible rate of one per 88 years of riding.

What part of that did you not understand? Perhaps the fact that they
observed so few instances of the problem you so greatly fear?

--
- Frank Krygowski