Steve Spencer
Offline
• • • • • • •
Upload & Sell: On
|
 p.7 #11 · 5DII failures on LL Antarctic trip | |
I wrote:
As a scientist who teaches statistics to PhD students,
jamesf99 responded:
Uh oh....Now here is a point that I say is conclusively, unequivocally, and absolutely insignificant. 
Sorry, but I spent enough years in graduate school with fellow PhD students and masters degree candidates, and some of the worst instructors an Ivy league school could provide to know that your point is easily "challengable"........
My response to James:
It is fine with me if you think my teaching of statistics is irrelevant to this question, but then you should also agree that Mark's claim that he is a scientist is also irrelevant to the question at hand. It was his statement "as a scientist" that led me to respond as I did.
By the way, what point of mine is challengable; that I teach statistics?--probably not; that the statistical significance of the failures can't really be tested?--sure this is challengable and by all means feel free to do so. The way to challenge my point is to actually do a statistical test of significance. I would be happy to learn from you, just as I said I would be happy to learn from Mark. I never said my point that the results here can't really be tested isn't challengable, but I would say that it would take someone actually doing a test of statistical significance to challenge my point. So far no one has done that, but I would genuinely be interested if someone has a way to do so.
I wrote:
I can say I know of no test that would let one make an inference about statistical significance in this case. If there is such a test I would be happy to have Mark correct me. Part of the problem here is that it is not even clear what is being tested. Is the test whether there are more failures than zero? If that is the test then it is not a matter of statistics, but simply a matter of observation. Yes there were more failures than zero--obviously. Is the test, however, whether there were significantly more failures among the Canons (or specifically among the 5DMKIIs) than among the Nikons or Hasselblads or whatever? Now this question could possibly be addressed with statistics and one wouldn't even need to know confidence intervals. It could be tested with a simply chi-square test, if there were enough observations of Nikon cameras failing. With the current data there just isn't a big enough sample of Nikon cameras to do the comparison. A chi-square test would require more observations of other cameras failing. I don't see a way to test for statistical significance here, and that makes the report here an anecdote. A compelling one, but unless someone makes it clear what comparison they are testing and how they carried out the test, then they should not claim that the reported incidence of failures is statistically significant.
Best wishes, Steve ...Show more →
jamesf99 responded:
Of course we don't have all the information and what we do have is mostly - but not all - anecdotal, but 6 out of 26 is bad news. You can spin it anyway you want, and spend 6 weeks designing your test/population/confidence levels, but it's bad news. If you take 26 cameras at random, use them in an ordinary way, and 23% of them fail, you are talking about a serious issue and you don't need to design a complete statistical analysis to know that. No Nikon died under the same circumstances.
EX: 26 new, randomly selected, undamaged Toyota's drive from Boston to NY. 6 die along the way for one reason or another, but in no case were they subjected to improper or even deliberate misuse. Surly you must ask why, but in this case if you want the most basic analysis you could start with "what is the chance your new 5d will go belly up if you look at it wrong"? You may or may not want a larger sample size, but I think we're on to something here.
We don't need a chi-square test, any more regression testing, we need common sense. This shouldn't happen. Period....Show more →
My response to James:
It is fine by me if you think and say that 6 out of 26 cameras failing is important, meaningful, and a significant number of failures. I think it is noteworthy too. Just don't say it is "statistically significant," unless you test the significance statistically. I am all for measuring significance in other ways than with statistics. I genuinely believe that is a good thing, but it bugs me when people make a "scientific" claim to statistical significance without backing up such a claim. So, I am not saying that it isn't meaningful because you can't test it statistically. I am just saying it isn't statistically significant unless you can test it statistically.
By the way, I was not troubled at all by your original post as you were straightforward about not knowing whether it was indeed a statistically significant finding and were simply speculating about the statistical significance. In contrast, Mark claimed that he knew it was statistically significant, and that did bug me because he seemed to be saying he knew this without actually conducting a statistical test. Maybe I am wrong and he has done such a statistical test and he can show me I am wrong. I am perfectly fine with being wrong here, but if he hasn't done such a test, then he shouldn't be saying that, "as a scientist" he knows it is statistically significant.
jamesf99 also responded:
Unfortunately, common sense isn't all that common.... M. Twain
My response: I definitely agree. This is one of my favorite quotes. And to me it is just common sense that for people to claim something is 'statistically' significant they might have to examine the issue statistically.
Best wishes,
Steve
|
Have any picts you would like to share. 
