[EM] Testing 1 2 3

Dgamble997 at aol.com Dgamble997 at aol.com
Fri Jan 2 13:57:02 PST 2004

James Gilmour wrote:

>David Gamble wrote:

>> I would like volunteers to test this model and give their
>> comments on it. If you would like a copy of the model to test

>No test is possible because you have not specified the criteria against 
>you wish to test the systems.

Basically the idea was to attempt to generate a realistic set of votes (one 
where voters vote principally but not entirely on a single ( left-right ) 
dimension) and then calculate the results under a variety of different methods to 
see how they differ. The model can generate other sets of votes based on other 
realistic scenarios and also (IMHO) completely unrealistic scenarios. I find 
how results differ under different systems interesting, as surely the entire 
purpose of reforming an electoral system is to get different and better ( though 
what constitutes better is highly subjective) results. You can address a 
number of questions using the model not just proportionality of party. For 

1/ Does using a good Approval strategy generally elect the Condorcet winner 
if there is one? ( answer yes as far as I can tell so far).

2/ is a winning votes cycle resolution method superior to a winning margins 
cycle resolution method? ( answer yes ).

By test I meant does it work ?, are the results in people's opinion 
realistic?, how could it be improved?

One of the things I hoped to get was people's opinion's on what it should 
show if it were working correctly you wrote:

>If your criterion is proportionality of party, all the multi-member systems 
will perform >better than any of the single member systems, except by chance.

Yes It shows that

>The degree of deviation in the single-member systems will depend on just how 
you >allocate the votes to the single-member districts.

Yes It shows that as well.

>If you want the highest possible degree of proportionality of party, D'Hondt 
Highest >Average ( 1 fifty member district) will give a better result than 
D'Hondt Highest >Average (5 member districts) or STV (5 member districts). 

Yes It shows this also.

>If you maximum voter choice among all candidates, STV (5 member districts) 
will >allow that in a way than none of the other systems will. But that feature 
will not be >demonstrated in your model.

STV was the most problematic system to model and the one I'm least happy 

>Incidentally, I was surprised to see Borda included in your list. Borda is 
>fundamentally flawed that I don't think anyone will ever use it for public 
elections - at >least, I hope not. Whether the flaws will be obvious in your 
model will depend on the >datasets you have chosen and on the allocation of 
votes to districts.  

That second preferences count against first preferences under Borda can be 
seen virtually whatever data set you put in it. It's flaws, though very 
noticeable, are not quite as bad as some people make out. It is in there because some 
people ( Donald Saari, Steve Barney) do take it seriously and write books and 
papers telling us how good it is.

David Gamble






-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20040102/7d47bfc4/attachment-0003.htm>

More information about the Election-Methods mailing list