[EM] General PR question (from Andy Jennings in 2011)

Richard Fobes ElectionMethods at VoteFair.org
Sat Oct 4 15:54:35 PDT 2014


On 10/3/2014 3:54 PM, Toby Pereira wrote:
> When I do ranked votes I normally put greater than signs in, so A>B>C.

Ah, that makes sense.

When there is a ranked ballot with some same-level preferences, the 
equal sign is useful:

A > B = C > D

Alas, the equal sign doesn't work so well for approval ballots:

A = B

In the message I replied to I don't recall seeing a reference to 
approval ballots, but then I'm juggling multiple projects and don't have 
time to read all the posts in a thread.

I apologize for the distraction.

Admittedly I could not imagine someone promoting approval ballots for PR 
(proportional representation).  Now, after reading your explanation 
below I see the point you are making.  Thanks for taking the time to 
explain it.

 > On approval v rank for proportional voting, there's always the problem
 > with ranked voting that you don't know the strength of the preference.

Someday when there is a better way to count score ballots it will become 
possible to use them for proportional representation, and that may 
overcome this issue.

 > 3: A>B>C>D>E>F
 > 3: A>C>D>E>F>B
 > 3: A>D>E>F>B>C
 > 3: A>E>F>B>C>D
 > 3: A>F>B>C>D>E
 > 12: B
 > 12: C
 > 12: D
 > 12: E
 > 12: F
 > 14: G
 > 14: H
 > 14: I
 > 14: J
 > 14: K
 > 14: L

In this example your approval-based PR approach would be vulnerable to 
strategic voting because it is a "tipping point" or a "tie-like" case. 
I regard it as troublesome that a change in one vote can cause such a 
dramatic change in outcome.

Aside from that, you make an interesting point that for a special set of 
ballots and preferences, approval ballots can cause a PR outcome that 
_might_ be fairer than the "equivalent" preferences on ranked ballots.

Thanks for your patience!

Richard Fobes


On 10/3/2014 3:54 PM, Toby Pereira wrote:
> When I do ranked votes I normally put greater than signs in, so A>B>C.
 >
> Also, with approval voting, I'd normally specify above the vote table
> that it's approval voting, although sometimes it doesn't happen. But I
> suppose I neglected to do so, because it had followed on from previous
> examples, which were all approval cases. In any case, I'll try to be
> explicit in future.
>
> On approval v rank for proportional voting, there's always the problem
> with ranked voting that you don't know the strength of the preference.
> With single-winner Condorcet methods, there's the argument that it
> doesn't matter because every candidate is compared in a head-to-head
> manner so degrees don't matter. But with multiple-winner methods, it
> becomes more important. Which would you rather - your single first
> choice, or your second and third together? You can't say. Also with STV
> methods, your unused preferences are ignored anyway, so you might have
> got lucky and got your next choices elected or you might not. But the
> voting system doesn't take it into account at all. With approval voting,
> you at least get to say which candidates you like. And if you prefer
> more information, score voting would be preferable to ranks. It's worth
> looking at this example from Warren Smith http://rangevoting.org/PRcond.html
>
> It's also worth looking at this slightly more complex example with six
> to elect. Some voters have bullet voted, some have ranked several
> candidates.
>
> 3: A>B>C>D>E>F
> 3: A>C>D>E>F>B
> 3: A>D>E>F>B>C
> 3: A>E>F>B>C>D
> 3: A>F>B>C>D>E
> 12: B
> 12: C
> 12: D
> 12: E
> 12: F
> 14: G
> 14: H
> 14: I
> 14: J
> 14: K
> 14: L
>
> I'm not sure what your method would do, but STV methods would tend to
> elect A and five of G to L - e.g. AGHIJK. But it could be that the A
> voters, while they prefer A, actually wouldn't mind that much about
> which of their ranked candidates get elected - they might all be from
> the same party. If they knew what would happen, they could forget A, and
> this would be enough to get all of BCDEF elected - five candidates
> instead of just one. This is a case where approval voting would
> outperform ranked ballots. Yes, it's a contrived example, but it's
> contrived to show how these voters could go from having just one
> candidate to having five elected. It wouldn't need to be anywhere near
> as contrived to make a one or two candidate difference. And when several
> candidates are being elected, it is less about my single favourite
> candidate but about the candidates I wanted elected, so it makes sense
> to reflect that in the type of ballot used.
>
> Toby




More information about the Election-Methods mailing list