|
|
From: Alexandre Eremenko (no email)
Date: Tue Oct 26 2004 - 20:34:35 EDT
Dear Fred,
On Tue, 26 Oct 2004, Fred Hebard wrote:
> But if you reject observations often, especially more than
> one per session, I would say this is not a good idea
Agreed. I was talking of that particular series only.
> So it was a borderline case,
> but one where conventional wisdom says rejection is OK.
Borderline by the standard deviation criterion.
But there are other criteria for rejection (besides
convential wisdom).
In a series where the distance SHOULD INCREASE (this I knew
in advance, of course),
and 5 of the 6 observations clearly follow increasing pattern
but one does not (and substantially!) this one had to be rejected.
If one accepts this one
(I am always talking of my particular series of numbers!)
one had to reject the TWO previous ones.
This was my (almost unconscious) argument.
I did not compute any standard deviations. This was just
plane common sense.
On the other hand (as a scientist:-) I carefully recorded it
to be able to reduce it separately when needed.
Maybe this also has some psychological explanation.
When I said that some people "never make blunders",
this was an exaggerration of course. I don't think
there are such people indeed:-) But every reasonable
person probably knows somehow (maybe unconsciously)
HOW LIKELY s/he is to commit a blunder in such and such situation.
My own feeling of this is based on my practice,
and I know how frequently I misread the scale, copy a digit
wrongly, add a correction instead of subtracting it, make a misprint,
and
so on.
My second series of Lunars
Mon Oct 25 2004 - 22:46:20 EDT also shows that I estimate the
likely magnitude of my random measurement errors correctly.
Alex.
|