Re: OMEGA HAS A NONZERO BLOCK
From:"Lewis B. Sheiner"
Subject:Re: [NMusers] OMEGA HAS A NONZERO BLOCK
Date:Mon, 07 Oct 2002 11:16:33 -0700
So, we converge: 'Science' is to be used to supply what the current data do not,
but science is to be based on empirical evidence. Unlike Ken, I'm not generally
inclined to demand the data on which the domain expert's opinion rests (I'd not
likely understand it's implications fully, not being a domain expert myself),
but I'm glad someone is checking ...
I, too, can't leave this without 2 last quibbles:
1. I was not defending fixing the corr to .5, although I do agree with Nick that
if you have to fix it to something (e.g., because it's too technically demanding
to use a full Bayesian framework), for CL and V that is, I'd bet .5 is closer to
truth than either 0 or 1 (that's me being a domain expert).
2. More importantly, though, it is, in my view, MORE 'hyper-subjective' to fix a
parameter to a sharp value (ANY sharp value, 0, 1, .5, ...other) lacking
empirical support than it is to specify a non-degenerate distribution for it
(also lacking such support).
I'll repeat my main message another way: THERE IS NO WAY OUT OF THE ILL-POSED
INVERSE PROBLEM-- REPEAT, NO WAY OUT -- THAT DOESN'T INVOLVE MAKING ASSUMPTIONS
THAT ARE UNTESTABLE ON THE CURRENT DATA. Which, to reassure Ken, is not to say
that this releases us from an obligation to be careful about those assumptions,
question the data on which they are based, etc., etc.
LBS.