Re: M6 problem survey

[ABreull@aol.com]

Ok, ok, Patrick. 

I never wanted to discredit you, neither personally nor in front of the
group. 

I've always appreciated your ideas, informations, comments and support, and
I'm really sorry, that I stepped on your toes.

- ------------------------------------------------------------- Now the stuff:

Any new M6 from which parts fall into your lap while opening the package,  is
close to a catastrophy to every person as highly engaged in photography as
LUGlers. So, no matter, whether there's a probability of .50, .37, .09, or
.001, you are definitely not interested arguments on probabilities, as long
as you're the one with the broken M6. 

Maybe the Tschernobyl example was tasteless, but there also persons were
absolutelly not interested, how small the probability might have been for the
catastrophy, as long as they were the victims. And coming to your
computations - they didn't change anything in the personal disappointment of
a certain LUGler with a broken M6 in his lap. Or, do you suggest, that 9% (vs
50%) might make those LUGler more happy now?

You used student's t, computed the .90 confidence intervall, and you pointed
at several, severe possible biases, indicating that all the results were
probably errorous. Additionally, you argued (recentyl), that you used the t
distribution, because it's used most, but that the binomial distribution
would be correct. 

Now, the argument, that you do something because it's done most, is a little
clumsy - similar to having caries, because it's usual and most people have
it. 

And, since a broken M6 belongs to the group of "rare events", neither the t
distribution nor the binomial distribution are correct, but the Poisson
distribution, certainly the non central Poisson distribution in this case,
since Ferdinand's viewpoint is the event of broken M6s.  

And, since broken Ms belong to "rare events", the computation of the
asymetrical confidence intervall (if at all, you could also reflect on one or
other type of congregation test) is indicated, on which Marascuillo/
McSweeney (thanks for correction of the names) give a somehow nice, empirical
approximation.

Further, you can't use the Microsoft Fortran in windows 95, since the lastest
F77 is an 8-bit DOS application, which is not accepted in windows 95; and the
DOS C uses upper memory, which doesn't agree to windows 95 also.

But, that's history, and not the point, at least not for me.

My argument has always been a different one. 

Alf