# [EM] Idea I had- "The Psephocrina"

Bart Ingles bartman at netgate.net
Sun May 28 12:18:02 PDT 2000

```About all I have right now is a sort of advocacy piece written to
promote Approval Voting for use in county-level government in the U.S.,
particularly for minor single-seat positions such as Tax Assessor,
Public Defender, etc.  Most of this would likely be relegated to your
"manifesto" section, if you could use it at all.

I do have some ideas for simulation testing based on Norm Petry's
VoteSim software, if I ever get time to play around with it.  If so, the
results might be interesting enough to post on a web site.

(1) One is to see how often (what percentage of elections) a given
method produces winners with below-average social utility (below the
mean utility for the candidates in a particular race).  One could also
look at other thresholds besides mean utility; another possibility is to
calculate "area below the threshold".  The general idea is that worst
case behavior is worth considering, not just expected average results.

(2) It also seems to me that one could verify the existence (or lack) of
zero-info strategy in various methods by plugging strategies similar to
Approval's into simulation runs for each method -- at least if it is
valid to assume that a true zero-strategic-info voter should adopt a
strategy which maximizes utility for the entire group, since he doesn't
know which part of the group he represents.  Although trivially,
applying the normal Approval strategy of ranking all candidates whom the
voter rates above the mean equally in first place, and truncating the
rest, should produce results identical to Approval.  It might be more
interesting to see what happens if the voter ranks all above-the-mean
candidates sincerely, and truncates the rest, in pairwise methods as
well as in IRV.

(3) Finally, the robustness of Approval Voting could be tested by
biasing the normal "above the mean" strategy.  After calculating each
voter's mean utility, which would be the normal cutoff, you could then
vary that cutoff by calculating a weighted average of the normal cutoff
and a bias value, to find a biased cutoff which would be used in the
simulation.  Assuming the utility range is 0..1, the bias value could be
either 0, 1, or rand(1.0) where a new random value is obtained for each
voter.  Something like an 80:20 weighting (80% ideal value, 20% bias)
might be a reasonable place to start.

Bart

David Catchpole wrote:
>
> At last count there were 5 actual members. I still occasionally check on
> them on the web. Suffice to say it's basically Craig and Craig alone.
>
> Anyways, on the issue of the Psephocrina (swell of occult music ala Lost
> Ark), first thing I'm going to make a submission on is on hedonic
> formalisms on behaviour. I've been trying to work out for myself, or at
> least look up, an extensive declarations of principles that covers a
> probabilistic-hedonic formalism. It's not easy because it looks as though
> it requires a cycle of definitions (i is an actor where A and B, j is an
> action by i if B and C,... etc.)
>
> Does anyone else feel like writing an introduction to or an article on
> some issue that relates to theory of voting?
>
> On Thu, 18 May 2000, Bart Ingles wrote:
>
> > You mean there's more than one?  Jeez.
> >
> > David Catchpole wrote:
> > > I've sent them / him a copy of the message.
> > >
> > > On Wed, 17 May 2000, Bart Ingles wrote:
> > > > What about Polytopes?
> >
> >
>
> --------------------------------------------------------------------
> -When I makes tea I makes tea, as old mother Grogan said. And when I makes
> water I makes water.
>                                                 James Joyce
>                                                 "Ulysses"

```