A new study from VT on crashes

PGlaves said:

Maybe with one variable. But with as many variables as might present themselves in motorcycle crashes I would suspect a lot more cases than that would be highly desirable in order to draw actual conclusions.

Click to expand...

No argument on the multi-variable aspect.

BTW - Don't kill this thread

36654 said:

BTW - Don't kill this thread

Click to expand...

Plenty of blame to go around for the recent thread closures. A smart person wouldn't drag the "poison" around.
OM

I find VT's use of odds ratio analysis somewhat troubling, as it can give a very skewed answer. Here's an example.

Suppose that in a sample of 100 men, 90 develop a certain disease, while in a sample of 100 women only 20 develop the same disease. The odds of a man developing the disease are 90 to 10, or 9:1, while the odds for women are only 20 to 80, or 1:4 = 0.25:1. The odds ratio is thus 9:1/0.25:1 = 36, showing that men are much more likely to develop the disease than women.

Now lets look at the probability ratio. In this same example, the probabilities of developing the disease are 0.9 for men and 0.2 for women. The probability ratio is 0.9/0.2 = 4.5, a significantly different number than the odds ratio. In this example men are 4.5 times as likely to develop the disease than women, but have 36 times the odds. Using the odds ratio gives a much more alarming answer than using the probability ratio.

kthutchinson said:

I find VT's use of odds ratio analysis somewhat troubling, as it can give a very skewed answer. Here's an example.

Suppose that in a sample of 100 men, 90 develop a certain disease, while in a sample of 100 women only 20 develop the same disease. The odds of a man developing the disease are 90 to 10, or 9:1, while the odds for women are only 20 to 80, or 1:4 = 0.25:1. The odds ratio is thus 9:1/0.25:1 = 36, showing that men are much more likely to develop the disease than women.

Now lets look at the probability ratio. In this same example, the probabilities of developing the disease are 0.9 for men and 0.2 for women. The probability ratio is 0.9/0.2 = 4.5, a significantly different number than the odds ratio. In this example men are 4.5 times as likely to develop the disease than women, but have 36 times the odds. Using the odds ratio gives a much more alarming answer than using the probability ratio.

Click to expand...

Thanks for the clarification and education. I'm still struggling with a world where the Median rarely equals the Mean

Good comments. I can't wait for the new study to come out.

I am amazed that about 50% of the accidents are low speed, or no speed, tip-over.

E.

easy said:

Good comments. I can't wait for the new study to come out.

I am amazed that about 50% of the accidents are low speed, or no speed, tip-over.

E.

Click to expand...

Typically, I don't think those events would be called accidents. More, along the lines of embarrassments.

easy said:

Good comments. I can't wait for the new study to come out.

I am amazed that about 50% of the accidents are low speed, or no speed, tip-over.

E.

Click to expand...

Wonder why they would include useless data like that?

martyhill said:

Wonder why they would include useless data like that?

Click to expand...

It is due to the study design. Put gizmos on bikes and report everything that happens. At some point humans need to decide what is relevant and useful and what isn't. My problem is that the number of events that are relevant and useful - 15 - is way too small to provide much information. Get it up to a few hundred and I will start to believe their conclusions. But that would be a much larger study than their grant would fund I am willing to bet.

Much research has been done in Europe, with the MAID study being the most recent big one:

https://en.wikipedia.org/wiki/MAIDS_report

There is also a motorcycle collision literature survey that was done out of Oxford, in 2004:

http://www.mosac.eu/public/file/Huang%20e_a_%20A%20literature%20review.pdf

akbeemer said:

...Hmmmm, I'm a insurance company CEO. I commission a "study" on motorcycle risks, and lo and behold it turns out we've been under charging for m/c insurance.... who knew.

Click to expand...

Which happened in 2007 when the IIHS released a study on highway crashes involving "supersport" bikes.

kthutchinson said:

I find VT's use of odds ratio analysis somewhat troubling, as it can give a very skewed answer... Using the odds ratio gives a much more alarming answer than using the probability ratio.

Click to expand...

As Mark Twain used to say (attributing it to British Prime Minister Benjamin Disraeli)
"There are three kinds of lies: lies, damned lies, and statistics." One has to be extremely careful, detailed and scientific when manipulating and trying to understand what a series of events actually represents, and if the sample size is under 1,000 with no control group (not that you could have one for this), lumping dropping your bike at virtual stand-still in with running off the road, etc.,and with the virtually unlimited variables, it makes the problems insurmountable in my mind.

To further invalidate the study all you need to do is read it and see what they are counting! A full 37.8% of all their CNC (Crash or Near-Crash) events are identified as

Subject over left/right lane line Negotiating a curve

Click to expand...

Naturally the vast majority of these were "Subject over left lane line" as far too many people cut their corners. Cutting a corner, however, especially if there is a clear line of sight with no oncoming traffic, while inappropriate, is a far cry from what should in a rational world be considered a CNC incident.

As I mentioned, earlier, there is an outlier here that completely skews the results of this study to the point of not being able to have useful statistics. One can draw useful information, but definitely not statistics.

One individual rider (1% of the group) accounted for 13 CNC incidents (8.5%). Without knowing what these were, it's really impossible to the meaning of these figures. If they rode on isolated open roads and had poor lane discipline when there wasn't any traffic, than that is an issue but not a "Near-Crash" incident in my book.

Watch the video on page 14 of the PDF under the "Crash Descriptions" headline. Rider comes up to a stop sign to turn right. At a virtual stop the rider drops the bike. It is not a crash nor is it a near-crash. It is a drop and I strongly feel it has no business being included in the CNC portion of the study. Certainly it can be used to identify that some of us have poor control of our bikes and virtually all of us will have a brain-fade and drop a bike (I've done it a couple of times), but do not lump that into a CNC statistic.

57% of the 30 crashes were low-speed “capsizes”

Click to expand...

Which that video is listed as an example of.

The real problem here is that the sample size is absolutely way too small to build any credible statistics on and including things like a simple bike drop at 0.5 mph or cutting a corner on a wide-open full view deserted road are considered to be CNC incidents.

There is good info to be gleaned but, there are also, for me, significant issues which make me scratch my head.