When are "Real World Failure Rates" Useful?

[Marylander]

I have seen a lot of IRV advocates claim that Condorcet Failure isn't a problem because it only happens about one percent of the time, and that the mathematical models predicting much higher rates of failure and generally poor performance from IRV therefore don't apply to the real world. (Most of them link to this article: https://www.fairvote.org/every_rcv_election_in_the_bay_area_so_far_has_produced_condorcet_winners). One problem with this argument is that some pathologies don't need to actually occur to have a negative effect. For example, if a candidate who would have been more popular than either of the two frontrunners declines to run out of fear of being a spoiler, the race will be recorded as not having been affected by the spoiler effect even though damage has clearly been done. So the real world rate of the spoiler effect is going to be deflated, since it can affect races by discouraging candidacies, but discouraging candidacies prevents it from appearing to have occurred. And since the spoiler effect and condorcet failure are closely linked (the latter can only occur if the former does), I'd anticipate that real-world condorcet failure rate is also likely to be deflated. It seems like the usefulness of 'real world' failure rates is very limited. Does it have any value?

[Brian Olson]

For no additional cost we could have a system where that failure is not possible.

That should be an obvious win.
Why do they prefer the system that can fail?

On Thu, Sep 6, 2018 at 8:42 PM, Marylander <deh...@gmail.com> wrote:

I have seen a lot of IRV advocates claim that Condorcet Failure isn't a problem because it only happens about one percent of the time, and that the mathematical models predicting much higher rates of failure and generally poor performance from IRV therefore don't apply to the real world. (Most of them link to this article: https://www.fairvote.org/every_rcv_election_in_the_bay_area_so_far_has_produced_condorcet_winners). One problem with this argument is that some pathologies don't need to actually occur to have a negative effect. For example, if a candidate who would have been more popular than either of the two frontrunners declines to run out of fear of being a spoiler, the race will be recorded as not having been affected by the spoiler effect even though damage has clearly been done. So the real world rate of the spoiler effect is going to be deflated, since it can affect races by discouraging candidacies, but discouraging candidacies prevents it from appearing to have occurred. And since the spoiler effect and condorcet failure are closely linked (the latter can only occur if the former does), I'd anticipate that real-world condorcet failure rate is also likely to be deflated. It seems like the usefulness of 'real world' failure rates is very limited. Does it have any value?

--
You received this message because you are subscribed to the Google Groups "The Center for Election Science" group.
To unsubscribe from this group and stop receiving emails from it, send an email to electionscience+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Kevin Baas]

I've discussed how STV/IRV can be fixed to not have this issue, in an OpAVote blog post:

https://blog.opavote.com/2016/06/guest-post-rethinking-stv-fundamentals.html?m=1

[Ciaran Dougherty]

...you can't fix the Condorcet failure of STV by readmitting previously eliminated candidates, because if it's going to happen in STV, it will happen in the last seat, which is functionally IRV.

Yes, reviving eliminated candidates helps with the problem, but Burlington didn't have any eliminated candidates to revive, and had that problem.

On Sep 8, 2018 13:12, "Kevin Baas" <happy...@gmail.com> wrote:

I've discussed how STV/IRV can be fixed to not have this issue, in an OpAVote blog post:

https://blog.opavote.com/2016/06/guest-post-rethinking-stv-fundamentals.html?m=1

[NoIRV]

Can we resume this discussion on forum.electology.org please?

[Marylander]

OK. https://forum.electology.org/t/when-are-real-world-failure-rates-useful/

[Kevin Baas]

I never said you could.

there are three differences in my method from traditional Stv, not one.

in the single winner case my method reduces to a broda count, which doesn't have condorcet failures.

in the multi winner case my method also doesn't have condorcet failures, while multi winner borda would because it doesn't transfer unused votes.

[NoIRV]

On Monday, September 10, 2018 at 4:29:40 PM UTC-4, Kevin Baas wrote:
> I never said you could.
>
> there are three differences in my method from traditional Stv, not one.
>
> in the single winner case my method reduces to a broda count, which doesn't have condorcet failures.

Uh yes it does...?
5 A>B>C
3 A>C>B
7 B>C>A
Condorcet winner is A.
Borda: A = 16, B = 19, C = 10.

> in the multi winner case my method also doesn't have condorcet failures, while multi winner borda would because it doesn't transfer unused votes.

Whaaaaaaaaaat?! Please clarify.