[EM] election-utility strategy

[Richard Moore]

MIKE OSSIPOFF wrote:
> Though election-utility depends on an assumption that might be
> more approximate than the assumptions that Weber-Tideman uses,
> all the inputs for these methods are approximate anyway, and so
> it's questionable whether any serious loss of accuracy results when
> using approximate methods.

That's a fair enough statement. All the inputs are *inferred* from
some known data points, such as the results of a poll, or the
results of a previous election. The variations due to the assumptions
behind the inference can outweigh the variations between the different
approximations and the exact method. For example, in the zero-info
example I gave a while back with 12 voters, I assumed a binomial
distribution with each candidate having a 50% chance of getting
approval from each voter. But if the other voters are aware that they
should be using a few-voters strategy, perhaps a 40% chance would have
been a better assumption. Also, my assumption gave equal probability
to a voter voting for no candidates, all candidates, or any given
combination of candidates, so I know it's going to be a little off.
And I didn't account for the possibility of no-shows, or the
possibility that there may be more voters than the 12 I'm aware of.

I think in non-zero-info cases, if the inference is based on each
candidate's probable vote totals and margin of error from a reliable
Approval-style poll, then the inference is supported by statistical
evidence, and therefore the resulting strategy is more reliable than
the zero-info, few-voters strategy. Whether it's important to a voter
to actually go through the math, rather than use an approximate
strategy (or even just rely on his/her instincts), is up to that
voter. But it wouldn't be difficult to write an open-source computer
program that would be available to everyone who wanted to go that
route.

And for zero-info, many-voters cases, the above-average utility
method is always accurate.

  -- Richard

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