Interesting Links for 12-06-2014

[andrewducker]

• It's Really Hard to Be a Good Guy With a Gun
  (tags: guns usa )
• Using "hero rats" to save lives by finding mines. (And the rats don't get hurt!)
  (tags: rats minesweeper )
• An Open Letter To J.K. Rowling
  (tags: jkrowling Scotland independence )
• No more IPv4 addresses in Latin America and the Caribbean
  (tags: internet ipv6 ipv4 )
• Response by Ray

Comments

[steer]

Going to put this here before I forget: Election in ExampleVille D'Hondt method with no modifications, 10 regions, 4 parties (ABCD), 4 seats a region 600 voters a region -- numbers chosen to make the example simple but doesn't depend on numbers, You can construct a similar result with 10 seats a region and 100,000 voters -- differences in voting will be less startling

[danieldwilliam]

Are you sure about d'Hondt not favouring large (in the electoral region) parties. If I put exactly the same votes into a Sainte-Lange it comes out with 10 seats each. My observation of comparing D'hondt and S-L is that S-L tends to see smaller parties just winning one extra seat and larger parties just losing that final seat.

[danieldwilliam]

I think any non-preferential system with regional constitutencies is going to be prone to this sort of effect. You get proportionality within the region, not necessarily across the whole chamber

[steer]

Yes, it is an arbitrary example finely balanced to show De Hondt favouring small parties. I think dh minimises the mean error between proportion of seats and proportion of votes. Got a long plane trip so might try to prove it. Can't remember the circumstances under which that is true, I think it is more obvious in Jefferson's formulation. Dh is pretty much what I would come up with I think. All systems have their flaws and any regional system you could come up with similar crazy.

[danieldwilliam]

If you fancy spending the flight on this I'd be very interested in the outcome of your cogitations.

[steer]

Hahaha... so I spent a bit of the flight on this and it turns out that D'Hondt is nearly but not quite unbiased. It looks like it's going to be but then you need to add a term to the quotient which makes it 1/2s+1 rather than 1/s+1 (where s is the number of seats). This, in fact, makes it the Sainte Lague method.

I was so convinced as well... didn't see that factor without working it all through.

My apologies.

[steer]

Bugger and no, there's still a mistake in my maths somewhere. I will keep trying.

[danieldwilliam]

Small parties nationally - they happen, in this example, to be large parties regionally.

[steer]

Incidentally a friend of mine has come up with a non pr system where your MPs vote in Westminster is related to their majority. It has interesting properties... All votes count equally, all regions get their most popular MP. Everyone I ever described it to hates it.

[danieldwilliam]

Yes - I hated it when it was described to me.

[andrewducker]

I think the issue there (and there are always issues with any system) is that you could end up with no representatives for a given party over a large area because they're excellent at coming a close second. So I might find that the nearest representative for me is 300 miles North, with a tiny majority.

[steer]

Yes, but if they come a close second then they have made the winning party ineffective. It does fall down if that party gets no elected MPs of course.

[andrewducker]

True, it would definitely knock down the others around them. You'd end up with some MPS worth only a couple of votes!

[andrewducker]

I think it's because plenty of papers have been published agreeing that D'Hondt is less fair than other methods, and generally prejudiced towards larger parties?

http://www.ajol.info/index.php/orion/article/viewFile/34253/6251 for example.

To me, the ideal method would be to divide number of voters by number of available seats to give a threshhold. For each party, divide number of votes by this number. Allocated each party a number of seats equivalent to the integer part of the result. Allocate remaining seats to parties in order of the size of the remainder.

So if party A has 2.7x the threshold, and party B has 0.85x the threshold then party A gets two seats and party B gets 1 seat. As opposed to D'Hondt, which would give party A seats at 2.7x, 1.8x and 0.9x, leaving party B with nothing.

[andrewducker]

I agree that it _can_ be done as you show - but I think the point of the article was that we're probably going to have an election next year where a party gets _no_ seats on a sizeable chunk of the vote - and it's entirely possible for them to get no seats while getting the largest share of the vote.

At which point the government isn't even slightly a reflection of the will of the people, and that's not a tenable situation to be in.

(Much though I loathe the party in question, I loathe this kind of failure of democracy more.)