[EM] "Weighted Webster", Apportionment, Moral rightness etc

[Michael Ossipoff]

Warren wrote:

Moral rightness is murky.

I reply:

For apportionment, there's only one kind of moral wrong: systematically 
giving more representation per person to residents of small states, as 
compared to large states (or vice-versa).

>
>If it is really true (as Ossipoff now says) that this moral rightness 
>conclusion
>was based on a "fallacy,"

I reply:

My justifiation for my claim that Webster is intrinsically large-biased 
(which was also my justification for BF and Weighted BF), and your 
justification for exponentially-weighed BF, were different. We preferred 
some form of weighted BF for different reasons. My fallacy was that, in 
judging expecatation, I neglected the fact that you or I have less chance of 
being born in a small state.
There may be a different fallacy involved with your reason for preferring 
weighted BF. For instance, you said it was about unfairness per person, 
which justified dividing by q. But unfairness sometimes isn't something that 
has to be divided among its recipients. When we're unfair to a state, all of 
its residents get all of the unfairness, and they don't have to divide it 
up. That may be a fallacy in your reason for preferring weighted BF.

Warren wrote:

and the simplest method also happens to be the most morally
>right, that'd be great - win-win scenario.

I reply:

Yes, that's so. The simplest of the methods at your website is the more 
morally right, fairest, true divisor method. The one that gives everyone 
equal representation expectation, by giving each cycle a number of seats 
equal to its number of quotas (based on the current quota being tried). Or, 
as I've worded it so far, giving each cycle an s/q = 1.  To the extent that 
they're all 1, they're all equal.

Warren continued:

(I don't currently understand what Ossipoff's
>moral reasoning is, though.)

I reply:

What I said in my previous reply paragraph in this reply. Equal 
representation expectation for everyone is my goal. I should add that that 
can only be possible if we don't have detailed information about what part 
of a cycle the various states are in. Obviously if you somehow know that 
your state is in the most unfavorable part of its cycle, then your 
representation expectation based on your information is less than that of 
residents of other states.

Equal representation expectation per person--isn't that the goal for all of 
us?  By the way, with Cycle-Webster and Adjusted-Rounding, that comes closer 
to being equal representation per person, in every apportionment, rather 
than just equal expectation.

>
>One way Ossipoff could explain, is he could write an alternate writeup to 
>my
>webpage's section titled "Which one is the most 'morally right'?" which 
>could be stuck
>there in its place, or at least compared side by side.

Quote my fairness goal at your website, and my claim that WW is the true 
divisor method that meets that goal.

You might want to give me a little recognition there, since I derived WW 
with the goal of equal representation expectatioon per person, achieved by 
making the cycles' s/q equal to 1, and therefore to eachother, as nearly as 
possible.

If you agree about WW being the fairest, why not take out the part about BF 
and exponentially-weighted BF. Or at least the part about it appearing to be 
fairest.

Warren continued:

>But frankly, it seems to me that eggheaded reasoning about moral rightness 
>is
>not worth as much as computer simulations measuring how well the methods 
>work in
>reality (or simulated reality).   I would like to see more of those.

I reply:

Yes. But I claim that WW, Webster, AR & CW can be shown to have their unbias 
just from arguments without empirical testing. But the testing is desirable 
too.

I've been looking at 1990 with Pearson correlation, just a little. WW, using 
Warren's d=.495..., had 4 times less bias than ordinary Webster.

I didn't test correlation for cycles, or do AR or CW, because that takes 
longer to program, and I have had very little time to work on it. I tested 
Pearson correlation of individual states.

Warren continued:

That is why my web page suggested choosing the parameter d
>to optimize historical performance - as opposed to getting d from a 
>theoretical
>oversimplified model of reality - actually get it from reality!

I reply:

By finding what d value minimizes measured bias. But using least-squares to 
find A, and thereby d, would be good too.

Mike Ossipoff