[ExI] vet my Bayesian logic, please

William Flynn Wallace foozler83 at gmail.com
Sat May 20 20:20:52 UTC 2017


A book by Levitin, a Field Guide to Lies, includes a discussion of illusory
correlation.  He cites an example of the identical twins who shared quite a
few uncommon traits and then calls the idea that something genetic was
involved an example of illusory correlation.  Here is what I want to write
to him:

Now I know about illusory correlation as well as anyone, but let's think
Bayesian for a moment.

Take that picture of the identical twins raised apart sitting on their
stools by the nearly identical workshops.  OK, so there's really a limited
number of ways to set up a workshop, so that's not really meaningful.

But the kicker was that both made miniature furniture!  What are the odds
of that?  Let's come back to that.

If I were informed that those twins both ate at McDonalds. favorite color
was blue, drove a Ford, had only a high school degree, and had another
hundred things in common, I would sniff at any attempt to call it more than
coincidence - illusory correlation.

But - what about flushing before using?  I've never seen that, though I've
done it a few times when it had not been flushed.  Taping pencils and pens.
Scratching their heads with their middle finger.  Never seen that.  (There
were more similarities in the original article.)

So, I propose that we have to think Bayesian.  Just what are the prior
probabilities of any two people *picked at random* sharing *all* of those
highly uncommon behaviors?  Not just one or two, but all of them.

Unlikely to the extreme, I think.  I am left not knowing what to think, but
I do think I'd keep the idea that such shared behaviors, seemingly distant
from the kind of behaviors we know have some genetic influence, may not be
random, filed for future reference.

bill w  (rewrite the last sentence for clarity?)
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20170520/0886f613/attachment.html>


More information about the extropy-chat mailing list