Re: Geiger, not binomial ?

[brian whatcott <inet@INTELLISYS.NET>]

At 11:36 3/24/00 -0800, Bernard C wrote:
> ...it's the sample mean.  the mean is found when all have decayed.
> However, it's a good estimate, as can be determined by a number of tests.

Preaching to the choir, I realise: but no, the mean is not found
when all have decayed. The errors, like the poor, are always with us.

 You recall someone noted the undercounting of multiple decays
(Or if they didn't, they should have  :-)
   the G-M tube has a finite recovery time so coincident pulses are
not represented. Ludwik, and anyone else, may sum the counts to a
few less than those measured, and the instrument itself will skip and
 add a few counts of its own, as well as misplacing some counts in the
wrong 'bucket'

> .... use (early and often) the Chi Square test.

Advocating the use of a statistical test which helps to assign some
degree of confidence to ones stated precision, taking into account
the number of fitting parameters one used, seems like a very helpful
objective.
 In this particular case, there seems to be just one parameter to fit,
the mean count per time interval which satisfies a known distribution,
so simpler statistics could work.

In particular, forcing reference to statistical tables can be an
unfortunate 'mumbo-jumbo' factor in my view.

This makes for a useful argument in favor of
routine use of statistical computer codes for every experimental
 purpose - and a step back from those alienating statistical
 derivations. (I never tire of praising NLREG for regressions,
for example, even though it quotes Student's t, not chi squared)

In my view, we see, in this list, the results of many bright kids
being forced through ab initio derivations of statistical
distributions - a more than healthy distaste for the entire topic.




brian whatcott <inet@intellisys.net>
Altus OK