[EM] Plurality with Condorcet polling is effectively Condorcet. Condorcet for 2012!

[Juho Laatu]

I think this could be an efficient way to promote ranked methods (primarily Condorcet methods). Physical presence in different plazas may attract also media interest. The results could be also good material for the media.

The first phase of this approach is to collect the ballots. That information (or all the local matrixes) should be made available to all interested parties. The second phase could consist of various approaches to collect national level conclusions from that information. The approach that you described should be one of them, but one should probably not claim that it is the only correct one. The reason to separate the first phase is that first phase information is very solid material (of course not worth actual election ballots but good anyway), but the second phase is subject to choices of those who count the results. Openness adds credibility here. Third phase would be use of various methods to count the results based on the second phase results. Fourth phase would consist of well planned publication of the results (media, own web site).

The geographic weighting approach is one way to correct the bias in the collected ballots. There are however also kind of bias that requires balancing. One key problem is that the results could be very different depending on the plaza where you collect the ballots, time of day etc. In some places you could get democratic results and in some other places republican results. One approach to balancing the results could be to weight them so that republicans and democrats will get about the expected level of support in each state (or at national level). I'll leave the details open since these questions belong to the "free" second phase anyway. There sure are already good methods available in statistics / polling industry.

Material that will be collected at one plaza may not represent people of the whole state - maybe only people within 10 miles from that plaza. For that reason balancing of the results of different plazas based on political bias may be even more important than generalizing them to cover also the nearby areas.

In the third phase all common Condorcet methods are likely to give the same results (maybe the difference between margins and winning votes becomes visible first, but probably also they will give the same results). Also IRV results should be counted for comparison (and will be if first and/or second phase information is available).

One interesting question is how to organize the polls and find all the people that should implement the polls in practice. I think a good approach would be to first plan how the polls will be carried out, including ballot design, actual polling procedure, name and logo for the campaign, and how to record the ballots (and other poll related data like place and time) in digital format. The ballot format could be simply a pdf file that all the volunteers can print and copy. Volunteers will need also a ballot box. I guess any reasonable box will do. You have to trust the volunteers in any case to conduct the polls and do vote counting honestly.

The best way to collect sufficient number of volunteers might be to simply advertise this activity in some appropriate mailing lists and forums / organizations. If you do that properly you might end up having a large number of volunteers. EM list members certainly won't be enough.

In any case the second phase will not be easy, and the results can not be claimed to be very reliable. They could become more reliable if one could compare them to some truly extensive poll or actual ranked election ballots. But that luxury is not available, and therefore the results will unavoidably be just one interesting piece of news, or maybe a best guess and best estimate on what the whole country might have answered.

But on the other hand, if one collects such poll information in a well standardized way (phase 1) that procedure can be repeated, and the results can be compared to the previous results. This means that it might make sense to arrange not only one poll but several polls. Changes between the polls could be measured in a quite reliable way, assuming that many enough of the polling stations would stay the same and would have about the same number of voters at the same day of week. It is thus possible to collect also scientifically meaningful data using multiple polls during one season. Also the next elections could be compared to the previous election.

Good luck with this idea. I think it would be worth trying (and a lot cheaper than asking some gallup company to do it). If you do it well, people might get interested also in the used method in addition to the interesting results.

Juho




On 21.10.2011, at 3.47, MIKE OSSIPOFF wrote:

>  
>  
> (I emphasize the obvious fact that some of the most useful and helpful participants in this mailing list have been people who reside
> in countries other than the one that I reside in. The project that I'm proposing is polling for the U.S. presidential election of 2012)
>  
> I've briefly mentioned this idea in a previous post.
>  
> I suggest that all here who want a better voting system for the U.S., and who reside in the U.S., so that they can conduct polling here,
> work together on the following project:
>  
> All of us, in our own counties &/or cities, conduct a public poll. We can do that polling in places such as public
> plazas, etc.
>  
> The poll consists of a rank-ballot, of the candidates in the 2012 U.S. presidential election.
>  
> The reason why I suggest polling in person, in public places, is to avoid the ballot-stuffing possible in Internet voting, and
> to, avoid possible selection biases on the Internet.
>  
> Of course, one should only do one day's session of polling, because, if one polls for several days, it will be difficult to
> recognize people who have already voted.
>  
> Each poll-conductor, after his/her polling session, will post his/her rankings to the EM mailing list. (We'll know what the
> names of the polling volulnteers are, so only their rankings will be counted).
>  
> When every volunteer has done his/her polling session, and all of their ballots have been posted at EM, anyone, including
> me, can count the ballots to find a CW.
>  
> But, instead of just counting the raw ballots, I suggest weighting them according to the number of ballots in each local poll, and the
> population of the U.S. region in which that particluar local poll was conducted.
>  
> Here's how I'll do that (unless someone has a better suggestion):
>  
> On a U.S. map (conic projection or locally-centered azimuthal equidistant), I'll draw a line between each pair of
> neighboring polling cities. Then I'll draw the perpendicular bisector of that line.
>  
> The set of perpendicular bisectors, together, will form a set of irregular polygons. I'll refer to those irregular polygons as
> "regions".
>  
> For each local poll, each of its ballots will be weighted by multiplying it by the the population of the region in which that local
> poll is conducted, divided by the number of ballots in that local poll.
>  
> How to find the population of a region? 
>  
> Of course if a state is entirely in that region, then its population is simply added into the region's population.
>  
> What if a state is partly in that region?
>  
> Sum the population of the major cities in that region, and add that sum into the region's population.
>  
> Assume that the state's population outside the major cities is uniformly-distributed.
>  
> Determine the area of that state that is in the region.
>  
> Multiply that area by the state's population density (adjusted by subtracting the populations of the major cities
> that have already been added in)
>  
> Add, into the region's population, the result of that multiplication.
>  
> How to determin the area of a state that's in the region?
>  
> Unless someone has a better suggestion, I'll do as follows:
>  
> I'll use the method of transects, using, as the numerical integration method, either Simpson's rule, or another
> closed Newton-Cotes formula.
>  
> In the method of transects, a line is drawn across the area to be measured, more or less through the region's center. 
> The area's width, measured perpendicular to that line, are measured at regular intervals along the line. My measuring
> interval will be the millilmeter marks on a ruler. The Newton-Cotes forumulas, including Simpson's rule, use regularly-spaced interval-divisions,
> such as those on a ruler. Such forumulas give an area estimate.
>  
> Of course, the area-measurements needn't be exact, because the overall project will involve approximating assumptions less accurate than
> the area-determinations.
>  
> Anyway, thereby will be gotten an estimate of how the ballots should be weighted, to simulate a national vote.
>  
> Using the weighted ballots, we find the Condorcet winner. That candidate has a win. We announce that CW to
> various small parties, alternative candidates, political organizations, and progressive media (or of course any media you want to announce it to).
>  
> These people and organizations can make the CW known around the country, if they want to. Looking it it from the point of view of
> a Progressive, I point out that, if the CW is a Progressive candidate, then the everyone who prefers a Progressive to the
> Democrats will know that they can probably safely vote for that Progressive CW, because s/he has a win, even in Plurality.
>  
> At least, to the extent that voters and candidates are distributed on a one-dimensional political spectrum, the CW can win in Plurality
> if everyone to one side of hir votes for hir.
>  
> Mike Ossipoff
>  
>  
>  
>  
>  
>  
>  
>  
>  
>  
>  
> ----
> Election-Methods mailing list - see http://electorama.com/em for list info

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20111021/d58ae834/attachment-0002.htm>