[EM] election-utility strategy

[Richard Moore]

Forest Simmons wrote:
> What if the polls could tell us (for each i and j) what percentage of the
> voters approve both candidates i and j.  If that percentage is not close
> to the product of the percentages of approval for i and approval for j, it
> would tell us that that approval for i and j are statistically related;
> perhaps the nature of this relationship might be useful information for
> approval strategy.
> 
> This information wouldn't require additional questionaires, only summing
> n*(n-1)/2 combinations from each existing questionaire (where n is the
> number of candidates).

See my April 12 post. I defined Bij(X) as the probability that i will beat
j if i has exactly X votes. If we know nothing about the relationship
between i and j votes, then for this value we can substitute the cumulative
probability that I called Cj(X).

A correlation (or an anti-correlation) of i and j votes would skew Bij(X).
So if you had the n*(n-1)/2 sums, then perhaps you could determine how
to skew the Cj(X) values to get accurate Bij(X) values.

I don't know what that calculation would look like, though. How good are
you at statistical formulas?

  -- Richard

----
For more information about this list (subscribe, unsubscribe, FAQ, etc), 
please see http://www.eskimo.com/~robla/em