<div dir="ltr"><div dir="ltr"><br></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Thu, Feb 7, 2019 at 5:01 PM Kristofer Munsterhjelm <<a href="mailto:km_elmet@t-online.de">km_elmet@t-online.de</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">On 04/02/2019 01.02, John wrote:<br>
> [Not subscribed, so please CC me replies]<br>
> <br><br></blockquote><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
As I said to Rob Lanphier when he was asking about MAF, I would suggest<br>
that the candidates who go on to the general always include the winner<br>
of the single-winner election based on the primary ballots. That is, if<br>
you're using Tideman's Alternative Smith, then it would be better to<br>
have an even number of candidates plus the Alternative Smith winner,<br>
than an odd number of candidates.<br>
<br>
Two reasons for this: First, this makes the primary+general system no<br>
worse than if there were no primary to begin with. If the voters behave<br>
identically in both rounds, then the single-winner method's choice is<br>
preserved through both rounds.<br>
<br></blockquote><div><br></div><div>Doesn't actually work.</div><div><br></div><div>We've found that voters reliably rank up to six, then drop off rapidly. If you have 7 candidates, just have your general election and no primary; but what if you have, say, 30?</div><div><br></div><div>AAAAAA BBBBBB CCCCCC DDDDDD EEEEEE<br></div><div><br></div><div>Thirty candidates in groups of six. Assuming they're evenly distributed ideologically (they're not), voters will elect the center of these groups, roughly, under STV.</div><div><br></div><div>What about under a Condorcet system?</div><div><br></div><div>You wind up with voters grouping around A, B, C, D, or E span. The largest group elects the Condorcet candidate among that group. If your largest group is the D-span voters, you get the condorcet for D. You may get something between C-D or D-E as an actual candidate span.</div><div><br></div><div>While STV will fail this way if near-exclusively A and A-B voters and only E and E-D voters show up to the primary, it will still tend to elect from the span A, B, C, D, and E if 1/6 of the voters are in each span. Moreover, you're more-likely to get overlap toward the center, and there's a greater population in the mean and so 2% lower turn-out among that population doesn't correlate to 2% lower total votes in that area.</div><div><br></div><div>So if your span is 22, 21, 17, 19, 21, your Condorcet candidate will tend to be someone around A to A-B.</div><div><br></div><div>Your five STV candidates will be somewhere around the middle of A, B, C, D, and E.</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
Second, the logic of STV is ultimately the same logic as that of IRV. I<br>
imagine the reason you propose an odd number is that if the left and<br>
right wings are roughly equally strong, then there will be a center<br>
candidate for that last spot; but STV will use IRV to populate the last<br>
spot (after the deweighting for each wing roughly cancels out). And IRV<br>
has a problem with center squeeze in such situations.<br></blockquote><div><br></div><div>In a field with a large number of candidates, the span across the center will be full of similar candidates.</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
You could get the same effect by using your more convoluted measure, but<br>
just augmenting an even group with the single-method winner is easier<br>
and gives the same result when the wings are balanced.<br>
<br></blockquote><div><br></div><div>Only if you assume all voters rank all the way to the Condorcet candidate. If voters tend to rank substantially-fewer, then you get what is conceptually-similar to a Plurality winner among the usual STV results.</div><div><br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><br>
> <br>
> My preferred STV system is Meek-STV, although I'm uncertain how that<br>
> really compares to Warren-STV. My experiments suggest Meek-STV will<br>
> always elect a candidate representing the Droop quota: if electing 3<br>
> and 25% of the population ranks some subset of candidates above all<br>
> candidates ranked first, second, and third by the other 75% of the<br>
> population, then the 25% will determine their own candidate.<br>
<br>
STV in general meets the Droop proportionality criterion, which states<br>
that if more than k Droop quotas' worth of voters rank m candidates<br>
ahead of everybody else (not necessarily in the same order), then min(k,<br>
m) of these must be elected.<br>
<br></blockquote><div><br></div><div>Good to know. </div><div><br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
As for Meek and Warren, when I did some proportionality simulations long<br>
ago, those two methods were very close, so I'd say that there's little<br>
difference between the two. The main reason for choosing one over the<br>
other is if you prefer additive weights over multiplicative ones (in<br>
which case choose Warren), or the other way around (in which case choose<br>
Meek).<br></blockquote><div><br></div><div>Nice. Which is simpler to implement? ;)</div><div><br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
> I have considered a more-convoluted measure:<br>
> <br>
> - Run STV;<br>
> - Replace the final winner with a Condorcet candidate computed from all<br>
> ballots with their final weights (including weight at exhaustion),<br>
> excluding the winning candidates.<br>
> <br>
> Largely, I have problems with the early-elimination issue in STV, and<br>
> have considered tweaks like restarting the count from the top after each<br>
> win, accounting for the weights of the ballots and excluding the winning<br>
> candidates (this will work with Warren, but not Meek).<br>
<br>
You could use Bucklin transferable vote instead of STV. Basically, you<br>
do a Bucklin election and whenever a candidate who's not yet elected<br>
exceeds the quota, you elect him and reweight the ballots.<br>
<br>
This will mitigate the early-elimination problem because BTV doesn't<br>
eliminate any candidates. On the other hand, it doesn't have the kind of<br>
record of former use that STV has.<br>
<br></blockquote><div><br></div><div>Bucklin is also highly-susceptible to strategy and can be manipulated by bullet voting.</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
A Condorcet-STV hybrid like CPO-STV or Schulze STV would entirely<br>
eliminate the problem. Unfortunately, they're way too convoluted and<br>
generally take time proportional to some power of the number of possible<br>
outcomes, which is rather hairy when that number is 42 choose 7.<br>
<br></blockquote><div><br></div><div>Schulze STV would be awesome.</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
> Unified Majority with Meek-STV/Tideman's Alternative has a few<br>
> interesting features:<br>
> <br>
> - It represents the span of the electorate's ideology, rather than<br>
> parties, in the General;<br>
> - It is immune to gerrymandering;<br>
> - It is immune to tactical nomination for MANY reasons;<br>
<br>
Strictly speaking, STV is not independent of clones; see<br>
<a href="http://www.votingmatters.org.uk/ISSUE3/P5.HTM" rel="noreferrer" target="_blank">http://www.votingmatters.org.uk/ISSUE3/P5.HTM</a>. So there is one way that<br>
it's not immune to tactical nomination, although it is probably fairly<br>
resistant since the single-winner method is independent of clones.<br>
<br></blockquote><div><br></div><div>True, although Hare is fairly resistant in that it's more likely to select one of the clones than Plurality. Adding a clone can eliminate a third candidate and elect a fourth, rather than eliminating the candidate being cloned.</div><div><br></div><div>These things become less-likely with larger primaries, and you can just skip the primary with 7 candidates. The 2016 election was half a dozen Republicans, three Democrats, a Libertarian, a Green, an Independent, and a few others who showed up somehow. That's over a dozen.</div><div><br></div><div>That's a practical matter; it's always technically-vulnerable.</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">> ----<br>
> Election-Methods mailing list - see <a href="http://electorama.com/em" rel="noreferrer" target="_blank">http://electorama.com/em</a> for list info<br>
> <br>
<br>
</blockquote></div></div>