[extropy-chat] aargh

[Eliezer Yudkowsky]

Damien Broderick wrote:
> At 11:23 AM 12/18/2004 -0600, I wrote:
> 
>> Using pre-screened `star' Ss, Ertel claims a robust success rate 
>> averaging 70% where 50% +/- is expected by chance.
> 
> Sorry, I hadn't been awake long when I wrote that. Ertel's set-up 
> contains balls numbered 1 through 5, so the chance likelihood is 0.2, 
> not 0.5. Which is why his Ss' results are so impressive.
> 
> Here's the abstract of a recent poster (I hope the superscripts etc make 
> it through the emailer):

Damien, I see where he says that results under control are statistically 
significantly less good than results obtained at home.  (What a surprise!) 
  I don't see where it says 70% hits (20% expected) effect size.

Say, remember that research on the power of prayer that claimed an effect 
size of 50% conceptions versus 25% conceptions?  Remember how I said that 
this was a nice non-marginal effect size, and how the claim was much 
healthier than all these alleged marginal effects, because it was much 
easier to test?  A nice, clear, claim - even though I expected it to turn 
out false?

And then remember how that research turned out to be totally bogus?

If this guy is getting an effect size of 70% hits on a 20% target, I like 
that.  That's a nice, clear claim.  Needless to say, it will also turn out 
to be totally bogus.

Here's the sad thing about psi research that I discovered during my own 
investigation:  The interesting results turn out to be just FAKE.  You 
don't usually think of that when you're trying to explain a reported 
experimental result... but with psi, so it goes.

-- 
Eliezer S. Yudkowsky                          http://singinst.org/
Research Fellow, Singularity Institute for Artificial Intelligence