First of all, I'd like to say thank-you for all the pointers.
Well, the numbers I had were 46 and 17, and I understood that there are about the same number of long-EZs and vari-EZs flying. I hadn't done the maths, but it seemed fairly clear there was a case to answer... or at least to discuss. Because I'm not just asking about the Rutan designs. My question is more general... Are there some designs that are substantially safer than others, and how can I be guided in making a decision?
OK... The NTSB holds data on 14 fatalities in 11 accidents for the Long-EZ and 3 for the long-EZ between 1978 and 2009, and 21 fatalities for the Vari-EZ and 43 for the Varieze. In other words about 17 for the Long-EZ and 64 for the Vari-EZ. I haven't checked every record, but the databases don't seem to be duplicates (i.e. 2 entries for the same plane, under different names).
63/17 is 3.7.
There are 798 long-ez registrations in the US and 797 variezes (this database seems more tolerant to different naming conventions).
Let's assume for the sake of argument that VariEZs, being a slightly older design and cheaper to fly, average twice the flying hours of the LongEZs. When I put this into Fisher's exact test and gradually increase the number of hours, the p-value asymptotes to around 0.02. In other words... statistically significant. Even when I swing the odds substantially in favour of the Vari-EZ.
What I did was to use a matrix of
17 63
500 1000
and then try ever increasing numbers of hours - all the way up to 700000 and 1400000. After a certain number of flying hours the p-value settles down and only changing the ratio between the Long-EZ hours and the Vari-EZ hours seems to make a difference. Or of course changing the numbers of fatalities.
So you're right, I hadn't done my sums and I was operating on imperfect information (I didn't know to look in the NTSB databases). But I had looked for information, and not finding it I chose to ask here. Having done the sums... it still seems that we have something to discuss.
My metric isn't a meaningless number. It's just a different metric from the usual one, which can just as validly be dismissed as meaningless. And more to the point, it was the only meaningful statistic I could calculate with the data I had available to me.
Some of the 'dangerous' planes - such as the Lancair - are very quick. So you might well argue that a Piper Cub is safer because there are fewer fatalities per hour. If you're flying for fun, you should get a Cub. But if you regard flying as a means of transport - perhaps you like to visit your parents 500 miles away on a regular basis - you might still be safer choosing the Lancair. Here I really haven't done any sums or looked to see whether the cub is a deathtrap and my contrived example is completely back to front. So don't shoot me down on these scores. I'm just trying to make a qualitative point that the relevant metric depends on what question you're asking.
Assuming I'm typical of the sort of person who might want to build and fly a plane (remember at this stage I have no real concept of how many hours I'm likely to fly between my first training flight and handing in my licence and how that compares to the 'average'...), it seems my chances of killing myself might be 1/20 or worse. That should be, and to me it is, quite chastening. And that's all I'm asking of this metric. Are my chances of killing myself substantial enough to worry about?
If only 1/1,000,000 people died, I would say 'so what'? 1/1,000,000 isn't much more than 1/3,000,000. But if you're talking odds of 1/20 v. 1/60 then this is very much within the range where it makes sense to discuss the relative risks of different designs. Which is where the accident incidence/100,000 hours starts to be useful.
