From: CurtAdams@aol.com
Date: Wed May 29 2002 - 17:57:40 MDT
In a message dated 5/28/02 16:29:10, mail@HarveyNewstrom.com writes:
>Assuming group identity is not informative, most people will be treated
>less accurately most of the time if others use group identity to inform
>actions. Or in other words, using group identity to inform actions only
>works if it is proven that group identity is directly and consistently
>informative for the trait being questioned.
Sort of right. It doesn't have to be consistent; probabilistic is enough.
And
it certainly needn't be direct. For example, if I find somebody has the
DNA microsatellite profile of a serial killer the profile didn't make him
a killer and being a killer didn't give him that profile; but even though
you would expect a dozen or so other humans will have the same profile
you don't want to go home with him alone. The association of the two
traits is pure happenstance, but still useful.
>> "People who have been nice to you in the past", after all, is
>> just another group.
>Incorrect. "People who have been nice to you in the past" is a group
>derived from specific measurements of individuals. Only individuals who
>were nice are in the group. Any individuals who were not nice are
>excluded. Group membership derives from the judgment. The group is an
>accurate demarkation of the determinate trait.
Right. So are groups demarkated by skin color, eye color, religious
preference,
ice cream flavor preference, or whatever. Group identity is useful to
actors when it correlates with some behavioral/ability characteristic of
note - for whatever reason.
>Using group membership to inform actions is the exact opposite of this
>approach. People are assigned to groups first without specific
>measurements being made of the individuals.
No, you have to measure *something* or you can't assign a group.
>These are not the same thing at all, but are in fact polar opposites!
>One way works and is a mathematically valid according to set theory.
>One way does NOT work and is mathematically INVALID according to set
>theory.
Anders posted the Bayesian math; it *is* valid under the right
circumstances.
This archive was generated by hypermail 2.1.5 : Sat Nov 02 2002 - 09:14:29 MST