Over the years of keeping a steady eye on motorcycle related research, I have noticed that the number 37 keeps cropping up. This could just be confirmation bias... or a cosmic coincidence... or if the tinfoil hat is cutting off circulation, a conspiracy... but it's seems just too prevalent. Below are a smattering of examples... there are many more where 37 features as a key quanta in motorcycle statistics. Many of the examples come from the US, so it might be something related to them... Anyway, just thought I'd share... http://www.spokes.com.au/assets/protective-clothing/Brochure_ProtectiveClothing.pdf Nationally, riders are over 37 times more likely than car drivers to be seriously injured… http://www.spokes.com.au/assets/whats-on/FULL%20REPORT%20Motorcycle%20Tracking%20Report%202010.pdf Regional riders are less likely to believe there is a high risk of being detected for drink riding (33% vs. 37% for metro riders. http://www.spokes.com.au/assets/whats-on/TAC-MM-2012-Report-of-findings-28092012-full.pdf …Taught by family members/friends 37% …Although the sample size of this group was small (n=37), http://www.spokes.com.au/assets/whats-on/TAC-Motorcycle-Tracker-Wave-4-24SEPT2012.pdf …The proportion of riders who believe there is a ‘low’ risk of being detected for speeding (37 per cent)… …the main reasons for being pulled over were for a routine licence check (77 per cent) and breath testing (37 per cent)… http://www.spokes.com.au/assets/whats-on/EURAP.pdf found that 47 per cent of impacts occurred when the rider was still on their motorcycle, with only 37 per cent sliding across the carriageway. http://www.upi.com/Health_News/2012/07/13/Helmets-cut-motorcycle-deaths-37-percent/UPI-78551342237939/ "We've seen that helmets reduce the likelihood of death in a motorcycle crash by about 37 percent," http://saferoads.org/motorcycle-helmets Motorcyclists are about 37 percent more likely as passenger car occupants to die in a traffic crash. (NHTSA, 2009) http://trafficsafety.org/safety/sharing/motorcycle/motor-facts/motor-injuries-fatalities In 2006, 37% of all motorcyclists involved in fatal crashes were speeding, compared… http://www.motorcyclecruiser.com/streetsurvival/maids_motorcycle_accident_study/viewall.html In 37 percent of cases, the primary accident contributing factor was a human error on the part of the motorcyclist. http://hellforleathermagazine.com/2010/04/motorcycles-with-abs-37-safer/ According to a study conducted by the Insurance Institute for Highway Safety, motorcycles equipped with ABS are 37 percent less likely to be involved in a fatal crash than those without. http://www.ridesmartflorida.com/dataandstatistics.htm 37 percent of motorcyclists were not wearing a safety helmet before a fatal crash during 2010. http://www.ghsa.org/html/media/pressreleases/2012/20120522_motorcycle.html …motorcycle deaths dropped 37 percent,… http://safetrec.berkeley.edu/research/motorcycle.html …37 percent said they had received no formal training,… http://sheriff.lacounty.gov/wps/portal/lasd/!ut/p/b0/04_Sj9CPykssy0xPLMnMz0vMAfGjzOItDCDAyN3M083A09_d0y3A2NzIIMhEvyDbUREAB3KcTQ!!/?1dmy&page=dept.lac.lasd.home.newsroom.detail.hidden&urile=wcm%3Apath%3A/lasd+content/lasd+site/home/home+top+stories/traffic+services+detail+conducts+motorcycle+safety+enforcement+operation …have resulted in a 37 percent decrease since the 2008 peak… http://www.northjersey.com/news/186121952_Road_Warrior__Training_--_the_antidote_for_motorcycles_crashes_training__can_boost_bike_safety.html?page=all …Motorcycle deaths in 1997 amounted to 44 – nearly 37 percent fewer than last year –…

I think you are finding a number thatis an artifact of the mathematical methods used to study data. 37%, or more accurately, 0.36787944... is 1/e. this is important in statistics calculations, but at the moment I can't remember why. I will look further into it and get back to you.

The number comes up from assuming a poisson distribution for the data being analysed, which would make sense when looking at something like motorcycle accidents, and having a mean number of "incidents" at one, and a you interrogate the data for the probability of one occurrence to happen over a given period of time, you get 1/e out as an answer. If you see this as a number in future, maybe you should take those statistics with a grain of salt. It looks like they have just put the numbers into a Poisson distribution and used inappropriate time intervals, and found an "accurate" answer that everyone assumes must be correct because of the values derived.

If you take the last number of the axial tilt of Earth and take the last number of the axial tilt of Uranus.... you get 37. In conclusion when earthlings are talking out of their arse ....you get 37

Oh this is good. Next step is to explicitly demand TAC and MUARC explain exactly how they arrived at this figure and if they cannot prove they didn't make the assumptions Middo has identified, then everything they claim can be dismissed as unproven. You sir are a knight among men.

No Rob, no leg pulling. While a portion of the data may be accurately 37%, a significant amount is just caused by the method of data analysis. Lets look at one claim: Without being able to see the original data for this, there is a significant chance that the 37% came about because on average per time period there was one less crash on ABS bikes, than non-ABS bikes. The problem is that we don't get the original data. And if analysed differently we may get answers of 20% or 40%. There will be a reduction in those involved in fatal crashes, but the 37% is what is suspect. Poisson distributions used to be on the year 12 maths course in WA before it got dumbed down.

...but that would mean that the 37% stat should be coming out in many other epidimeological / population type studies. Are you aware of any non motorcycling examples?

this is gold. pulling apart one of these studies based on misinterpreting data SO badly would be awesome... at the same time it's nothing new to see these bodies misrepresent figures, this is an interesting new way of HOW they do it... but the problem remains.. how are we going to get it across to gen pop that it's happening and what they're being told is a lie?

It is all about horse power... "A practical application of this distribution was made by Ladislaus Bortkiewicz in 1898 when he was given the task of investigating the number of soldiers in the Prussian army killed accidentally by horse kick; this experiment introduced the Poisson distribution to the field of reliability engineering.[6]"

Just a couple of instances that may be affected by the sample sizing issue: An analysis of project failure rates comes up with the magic 37% http://www.zdnet.com/blog/projectfailures/cio-analysis-why-37-percent-of-projects-fail/12565 There will be a 37% rise in the number of audiologists employed: http://www.bls.gov/ooh/community-and-social-service/health-educators.htm **** is undereported by37% http://www.feminist.com/antiviolence/facts.html Note that this issue is caused by the relatively small number of data points. For a really large population, such as all car accidents, you would not find 37%, but bike deaths, at maybe one per week? Or accidents requiring hospital admission at one per day? The mean and the time periodised are crucial.

I believe that the assumptions they have to make are actually the right ones as they are using a poisson distribution rather than a normal distribution for a small number of discrete events. I've never used this stuff (don't like dealing with small data sets). Another distribution that is used for discrete stuff (that I also haven't used!) http://en.wikipedia.org/wiki/Logistic_distribution ...but the lack of error bars on those conclusions is a drama. too slow

smileedude, I think you'd better take your tinfoil hat back. I don't know what to think about this situation... I think I might have a chat with a statistic professor I know...

I dunno, this sounds like a conspiracy going all the way back to the laws of mathematics and the universe. Government must be involved.

Hi middo, While you're right that 37% can turn up in probabilities from the Poisson distribution (and its hairy mate, the exponential distribution), I don't think that's what we're seeing here. Specifically a Poisson process gives you a probability of 1/e (37%) for "what is the probability that exactly one event [e.g. crash] occurs in the time interval where, on average, one event occurs", and "what is the probability that the time between events occurring is greater than the average". Looking over robsalvv's list, the 37s are appearing mostly in contexts where you would never ever use a Poisson distribution: - the majority were referring to sample proportion, nothing to do with probability ("33% vs. 37% for metro riders", "Taught by family members/friends 37%", "The proportion of riders who believe there is a ‘low’ risk of being detected for speeding (37 per cent)", etc) - one sample size ("n=37") - one "37 times more likely" - and a few comparing likelihoods: "We've seen that helmets reduce the likelihood of death in a motorcycle crash by about 37 percent", "Motorcyclists are about 37 percent more likely as passenger car occupants to die in a traffic crash", "motorcycles equipped with ABS are 37 percent less likely to be involved in a fatal crash than those without". While you could use a Poisson process to model crash rates - and in fact, insurance companies do that, because insurance claims come in more or less as a Poisson process - you wouldn't specifically expect to see any kind of "37% more likely" as the output of a model like that. In fact those are the kinds of figures that you'd be putting into your model to figure out how much to charge people for their insurance. Personally I'd put my bets on "this is all a complete coincidence".