[EM] Problems with finding the probable best governor

[Bart Ingles]

MIKE OSSIPOFF wrote:
> 
> >I guess I view vN-M utilities as the "ultimate sincere utility", since
> >it has an external frame of reference (risk).  By simply rating some
> >alternatives 0 through 10, different voters could have different ideas
> >about what 5 means, given identical circumstances.
> 
> But the voter only needs his own ratings. But, as you say
> below, maybe it's more difficult to estimate sincere ratings.
> 
> >  Being able to
> >compare benefit against some known (or assumed) risk provides a form of
> >calibration.  And over the long run, over many elections, vN-M utilities
> >SHOULD represent the best deal, unless the voter is wrong about the
> >levels of risk.
> 
> But the voter's vN-M utilities before a particular election,
> based on risk-aversion and how he feels about the outcomes
> over the past decades, those two views might involve
> conflicting kinds of utility. And after those decades of
> elections, the voter will most likely agree that, on the average,
> over a long time, he'd have been better off if he'd used
> sincere ratings instead of his risk-averse vN-M utilities.

But his vN-M utilities before an election are influenced by past
experience.


> Of course I admit that before a particular election is when
> the decision has to be made, when the risk-skittishness influences
> his expectation in a valid way. But because he has such a small
> chance of influencing the election, how much is fear, as opposed
> to objective, rational principle, going to influence his vote?
> One an afford to be especially principled when it's unlikely to
> matter materially which choice one makes.
> 
> >
> >Sincere ratings are unobtainable anyway in real life (vN-M utilities
> >could probably be inferred from voter decisions, at least in some
> >cases), and simulations don't distinguish between the two definitions of
> >utility.
> 
> Yes, that hadn't really occurred to me, that sincere ratings
> might be more difficult to guess than hypothetical lottery
> choices. I'm not completely sure. But is estimating what you'd
> do in a hypothetical lottery any easier than just estimating
> the best way to vote, directly? If not, then there'd be no
> point in guessing the lottery choices, and calculating
> strategy based on those, rather than just intuitively guessing
> the best way to vote. Both are perfectly valild. Isn't the
> latter just as valid, & just as easy, as the lottery guess?

YES!  Ratings and probability estimates are just a way of putting what
you know in one place -- you ultimately have to decide how to weigh the
information.  And the 'lottery' part mostly just consists of identifying
frontrunners.


> But I'm not saying for sure that I believe that you aren't
> right about that. When I determined how I'd vote in the SF
> mayor race, among 5 candidates, I estimated sincere ratings,
> and win probabilities, and used Tideman's suggestion for
> using win probabilities to get Pij. I felt that guessing the
> sincere ratings was easier for me than guessing the best way
> to vote. Maybe that's because one's guess about the best way
> to vote depends on (without really considering them) one's
> ratings of the candidates. I felt that I might as well estimate
> that separately, and use it, rather than estimate something that's
> based on those ratings plus the probabilities. I felt that, the
> way I did it, I was estimating less. On the other hand, though,
> maybe it's better to guess the best way to vote, rather than
> address the sincere ratings directly. I don't know. But the
> vN-M utility approach seems in the middle, without the
> arguments of either extreme. Isn't it just as valid, & just as
> easy, to guess what feels like the best way to vote in the actual
> election, as to guess what feels like the best choice in those
> hypothetical lotteries? If so, then the vN-M approach merely
> adds unnecessary complication. This is a subjective matter, and
> I'm not entirely sure that your position isn't more right, for
> all strategizing voters. But what I've said above sounds convincing
> to me.

I'm not sure we have the same exact definition of vN-M utilities.

My view:  If you start by assigning ratings to the candidates, and
optionally consider the probabilities if you know them (generally by
simply acknowledging the front runners), and you feel good about voting
based on your calculations, then the ratings and the vN-M utilities are
probably in close agreement.

If instead you decide the mathematical result just doesn't look quite
right, and overrule your initial calculations in order to vote what your
gut tells you, then you are really choosing vN-M utilities over sincere
ratings (provided the voting system doesn't have some sort of bizarre
optimal strategy).

In the second case, you are in effect weighing the value of additional,
perhaps intangible information, and using it to correct your raw
calculations in order arrive at the vN-M utilities implied by your final
vote.



> If the voters weren't fooled, then, for the Democrats, Perot
> really is better than the Bush, and for the Republicans, Perot
> really is better than Clinton.

Over the long haul the Democrat and Republican voters may not be better
off, if they are likely to win at least some elections and they think
that Perot is only slightly better than the main adversary.

Additionally, society may not be better off, if it winds up with someone
that most people hate only marginally less than a worst choice, instead
of a government with support of at least a sizable minority.