[EM] IIAC. Juho: Census re-districting instead of PR for allocating seats to districts.

[Juho Laatu]

On 17.6.2012, at 0.41, Michael Ossipoff wrote:

> And I'll repeat what I mean by SL's optimal proportionality:
>  
> SL minimizes differences in seats per person (whether seats are being allocated to districts or to parties).

Ok, the Sainte-Laguë side of the claim seems to be clear now. This is the same point that you mentioned in your previous mail. I commented it by comparing it to how Largest Remainder optimizes the results it in my previous reply.

> > (Btw, was this the key property that makes LR unacceptable to you, when
> > compared to SL?)
>  
> The property that makes LR unacceptable compared to SL is that LR allocations have greater differences in seats per person.

Ok, this clarifies the LR side of the claim a lot. You consider the SL method to be optimal, and LR naturally uses another approach.

> That, and the paradoxes, which are avoidable in party list PR and district seat apportionment.

Ok, that is the Alabama paradox and related paradoxes. Highest averages methods are based on sequential seat allocation, but that property is not always needed (can cause problems too, depending on what properties one wants).

> Those, and also the fact that LR is a two-part method, involving two entirely distinct successively-applied rules, the 2nd of which has nothing whatsoever to do with proportionality, and only spoils the proportionality achieved in the first part.

You could give the two part definition of LR and explain what goes wrong. But if you mean just that LR does not use the "minimal difference of seats per person" approach, then there is no need to explain this point again.

> > The purpose of that question was just to check that
> > you don't claim that SL would always give optimal results or better results
> > than other methods.
>  
> I claim that SL will always give optimal results, better results than other methods.

Ok, that said it clearly. I consider SL to be one of the best, but not a method that would be always better than the other alternatives. Also SL has its problematic cases.



I think I now got an answer to why you like SL and don't like LR. (You consier SL allocation method to be ideal, and LR differs from this ideal, and has in addition some paradoxes that you don't like.) If you are interested in why I consider the LR approach to be a good alternative to the SL approach for many needs, see the short analysis in my previous mail.

Juho