On May 7, 2010, at 12:20 AM, Jameson Quinn wrote:
Finally, I come to the end of this series. To conclude, let me
return to the initial question which started me down this path: what
do the values and heuristics have to say about voting system criteria?
Utility (that is, outcome social utility) is a unique value in that
we have a single clear tool for analysis: Bayesian regret
simulations. The only two things we need are a model for voter
utilities - which should have only weak, linear effects on the
performance of a given system - and a model for voter strategy. I
hope that my analysis of strategy helps us get the latter. And
meanwhile, I hope that the outstanding issues of strategy, along
with the issues of expressivity, legitimacy, and cost, can help the
partisans of Bayesian regret analysis (a group in which I weakly
include myself) be humble about the results so far.
Since the world of measuring personal and group utilities is very
complex I find it useful to sometimes simplify things by fist naming
the targets and assumptions and then comparing the candidate methods
against this simplified model.
When trying to measure the true utilities one often sets some limiting
assumptions like utilities that are normalized to some given range of
values. One may also have different targets in different elections.
Sometimes one wants to maximize the sum of utilities, sometimes
maximize the smallest personal utility. One could also have some more
sophisticated functions that e.g. try to lift the utility of 90% of
the voters/citizens to some agreed/ideal level.
Models that are based on rankings are simplified and practical models
of the world. It is interesting that often such models are considered
to be ideal. I mean that for example the majority principle is taken
as granted although it can be seen to violate some underlying utility
principles (the minority might have stronger opinions). But models
agreed by the society and well working models (that can e.g. handle
strategic temptations) have a value of their own and therefore may be
adopted as "values".
For these reasons (complexity and debatability of the true utilities,
different needs in different elections, assumptions on the society
(competitiveness, morale, traditions), influence of models that we
have created ourselves, need to meet also some (irrational)
marketability ad perception related needs) it may often be enough to
just state the targets and assumptions (the model to be used) and then
try to implement that using some concrete method. The value of the
model (targets, assumptions) and the properties of the method (that
tries to implement the targets) can then be discussed separately.
The criteria that relate to expressivity are mainly participation,
consistency, and related criteria. There is a certain fundamental
tension between expressive freedom and outcome utility in a decisive
system. Ballots which give a large degree of decisive freedom to
voters inevitably involve strategic tradeoffs between utility and
expressivity. Perhaps one way to resolve this is to make a hybrid
system, which is part election and part nonbinding poll. The
election part would strictly limit voter freedom in order to
eliminate strategic concerns; two-rank Bucklin is one possible
example of the limited freedom I'm imagining. And the poll part -
probably based on Range, for maximum freedom - would allow free
expressiveness; while it wouldn't be binding, to keep stratgic
concerns from taking over, it would certainly affect a candidate's
mandate. The poll would be optional; a simple Bucklin vote would be
counted with some default assumptions about what that means in Range
terms. Maybe you could even give marginal impact to the poll without
making it too strategic. For instance, if you were building a polity
from scratch, the rule could be that a Bucklin winner is elected,
and if they're also a Range winner they get an extra year on their
term; or that the Range winner gets to appoint the attorney general;
or some such marginal importance for the poll.
Note that in competitive environments also pure polls may get
strategic. If we have first a poll and then an Approval election it
might make sense to give strategic answers in the poll to make my
Approval favourite look stronger than some other candidate in order to
make the supporters of that other candidate approve also my favourite
(since the strategic advice in Approval will be to approve at least
one of the leading candidates). Also in other elections people tend to
vote for strong candidates / potential winners.
The criteria that relate to legitimacy are participation;
consistency; and clone-, strategy-, and fraud- resistance criteria.
The main criterion that relates to cost is one-round decisiveness.
(Maybe also time and effort to count and verify the results, and need
to buy voting machines, and maybe also need to debate about the
results after the elections.)
Juho
The heuristics relate to various criteria, but in another sense they
function as vague criteria themselves.
I'll repeat my conclusion to the long section on strategy here, too.
Strategy's biggest effects are not on outcome social utility, but on
legitimacy and expressivity. I believe that these effects are
serious and worth avoiding. Cabal strategy seems to me the best
model for strategy analysis, but if it isn't going to get hung up on
all Condorcet ties being strategic, it needs to include factors
which affect the likeliness of strategy actually being used. These
factors include motivation, implied dishonesty, and necessary
participation.
Finally, for those who have read through this nearly endless
treatise, thank you for your patience. I hope that I've said enough
to make it worthwhile. I also hope that I start more than one
productive discussion - productive enough to change my mind about
some aspects of what I said. My ultimate hope is that this kind of
discussion will help us have the perspective to recognize, to
design, to evaluate, and finally to begin to agree on the best
possible voting systems. The more big-picture perspective we have
and share, the more we will become an activist force with which
nations must reckon.
----
Election-Methods mailing list - see http://electorama.com/em for
list info
----
Election-Methods mailing list - see http://electorama.com/em for list info
