[extropy-chat] silent night

[Adrian Tymes]

--- Eliezer Yudkowsky <sentience at pobox.com> wrote:
> >> On Dec 26, 2004, at 1:44 PM, Adrian Tymes wrote:
> >>> Absence of evidence is not evidence of absence.
> 
> Yes it is.  See the "Law of Conservation of
> Probability" in the _Intuitive 
> Explanation of Bayesian Reasoning_.  If A is
> evidence for B, not-A is 
> necessarily evidence for not-B.

Technically true, but not quite the way I meant it.
One can prove that there has not been much proof that
idea futures markets can successfully predict things.
But this lack of proof can back more than one
scenario:
1. I.F. markets don't successfully predict things.
2. No one has spent the time to thoroughly study this
   either way, or at least the results of such studies
   fail to show up when we search for them.  (A study
   locked away for nobody to ever see or know about
   has about the same practical effect as if the study
   was never done.)
3. Possibly others, which I'll ignore for now for
   sake of simplicity.

We can further distinguish between cases 1 and 2 by
whether anyone has published disproof of I.F. markets.
Again, we find none.  Therefore we can conclude that
the lack of evidence is neither because I.F. markets
prove things nor because I.F. markets disprove things,
but merely points to a lack of studies.

In practical English, many questions do not have
simple yes-no answers.  (Granted, most of these can be
reduced to equivalent series of yes-no questions, but
"can be" and "are" are two very different concepts.)
This has several consequences that can be viewed as
unfortunate, for example the inability of strict
Bayesian reasoning to deal with these questions
(unless care is taken to translate them into yes-no
domains first, which was not done in this case).

> If a supposed effect is tested using an experiment
> of high statistical 
> power (that is, high probability of discovering an
> effect of the given 
> magnitude supposing one exists), then the failure of
> that experiment to 
> produce statistically significant results is
> evidence against the effect. 

Again: we only see that there are no/few reports of
successful experiments.  But we also see that there
are no/few reports of unsuccessful experiments.  The
conclusion is that no/few experiments - successful or
not - have been reported on, not that the experiments
were necessarily successful or not.

One might wonder why those running the experiment, if
successful, did not report.  But one might also wonder
why no one else (say, those competing for research
funding with the I.F. markets) reported on their
experiements if they were not successful.