[EM] IRV vs Plurality

Dave Ketchum davek at clarityconnect.com
Fri Jan 15 19:05:58 PST 2010


On Jan 15, 2010, at 7:46 PM, Juho wrote:
> On Jan 14, 2010, at 2:13 AM, Dave Ketchum wrote:
>> On Jan 13, 2010, at 4:49 AM, Juho wrote:
>>> On Jan 13, 2010, at 9:14 AM, robert bristow-johnson wrote:
>>>
>>>> it still is a curiosity to me how, historically, some leaders and  
>>>> proponents of election reform thunked up the idea to have a  
>>>> ranked-order ballot and then took that good idea and married it  
>>>> to the IRV protocol.  with the 200 year old Condorcet idea in  
>>>> existence, why would they do that?
>>>
>>> 1) The basic idea of IRV is in some sense natural. It is like a  
>>> street fight. The weakest players are regularly kicked out and  
>>> they must give up. I'm not saying that this would lead to good  
>>> results but at least this game is understandable to most people.  
>>> Condorcet on the other hand is more like a mathematical equation,  
>>> and the details of the most complex Condorcet variants may be too  
>>> much for most voters. Here I'm not saying that each voter (and not  
>>> even each legislator) should understand all the details of their  
>>> voting system. The basic Condorcet winner rule is however a simple  
>>> enough principle to be explained to all. But it may be that IRV is  
>>> easier to market (to the legislators and voters) from this point  
>>> of view.
>>
>> When there is a CW in Condorcet, the CW has won in comparison with  
>> each other candidate.  While a few may like X or Z enough better to  
>> have given such top ranking, the fact that all the voters together  
>> prefer the CW over each other should count, and does with Condorcet.
>>
>> Else there is a cycle in Condorcet.  Perhaps the following Minimum  
>> Margins Method Condorcet variant should be used to establish  
>> Condorcet's preferability over other methods.  Then let other  
>> variants compete with this one before finally deciding which to use.
>>
>> Minimum Margins Method:  Consider the cycle, such as A>B>C>A, and  
>> the margins that create it, such as 60A>30B, 40B>20C, 21C>20A.   
>> Delete the weakest margins as many times as needed to destroy the  
>> cycle - in this case A becoming the CW (note that if one C>A voter  
>> had voted A>C in  this election, A would have become CW with no  
>> cycle).
>
> When I see this kind of scenarios I'm always tempted to ask the  
> question if it is necessary to limit the scope to the top cycle  
> members or if one can allow also the others win (when the cyclic  
> opinions in the top cycle are strong). I find also that approach to  
> be a working solution for many election types (although many have  
> indicated that they disagree with this).
>
Note that this method breaks the cycle at the point where the smallest  
number of ballots being voted differently would have broken the cycle.

Note that weaker candidates are unlikely to get enough votes to be  
part of a cycle - being weak they get few high rank votes.
>>>
>>> 2) IRV is easier to count manually. Condorcet gets quite tedious  
>>> to count manually when the number of candidates and voters goes  
>>> up. One can use some tricks and shortcuts to speed up manual  
>>> Condorcet counting but IRV probably still beats it from this point  
>>> of view. Manual counting was the only way to count for a long  
>>> time. Nowadays we have computers and Condorcet tabulation should  
>>> thus be no problem at all (at least in places where computers are  
>>> available). But this is one reason why IRV has taken an early lead.
>>
>> When an election district has only one polling place, life is simple.
>>
>> When the district is a state or city, life is more complex for each  
>> method.
>>    With IRV you first want top ranks from all the ballots in the  
>> district.  If there is no majority winner all the ballots for the  
>> worst loser must be scanned for all the polling places for whom  
>> those voters ranked next.  Repeat until winner gets decided.
>>    With Condorcet each ballot gets scanned one time and its content  
>> added into an N*N array, with such arrays summed for the whole  
>> district.
>
> Maybe scanning and other ballot checking can be done only once.  
> After that IRV requires either central counting or central control  
> of the distributed counting process.

How the counters go about their job is hard to define.  My point was  
that after a losing candidate is identified, ballots ranking that  
candidate high are the ones whose content must be findable in IRV -  
matters not if you must go back to the actual ballot, or someplace  
else which can provide THAT information.
>>>
>>> 3) Large parties are typically in a key role when electoral  
>>> reforms are made. Election method experts within those parties may  
>>> well have found out that IRV tends to favour large parties. In  
>>> addition to trying to improve the society the best way they can,  
>>> political parties and people within them also tend to think that  
>>> they are the ones who are right and therefore the society would  
>>> benefit of just them being in power and getting more votes and  
>>> more seats. The parties and their representatives may also have  
>>> other more selfish drivers behind their interest to grab as large  
>>> share of the power as possible :-). IRV thus seems to maintain the  
>>> power of the current strongest players better than Condorcet does,  
>>> and that may mean some bias towards IRV.
>>>
>>> 4) The problems of different election methods may appear only  
>>> later. A superficial understanding of IRV reveals first its  
>>> positive features. Like in Burlington the negative features may be  
>>> understood only after something negative happens in real  
>>> elections. This applies also to Condorcet. On that side one may  
>>> however live in the hope that the problems are rare enough and not  
>>> easy to take advantage of so that sincere voting and good results  
>>> would be dominant. The point is that IRV may be taken into use  
>>> first (see other points above and below) without understanding  
>>> what problems might emerge later. And once it has been taken into  
>>> use it may well stay in use for a long time (electoral reforms are  
>>> not made every year, people have already gotten used to the  
>>> method, having to change the method could be seen by the society/ 
>>> legislators as a failure/embarrassment, and people/parties who  
>>> were elected based on those rules and are strong in that system  
>>> may be reluctant to change the rules).
>>>
>>> 5) Both IRV and Condorcet have some weak spots that can be  
>>> attacked. As you point out the weak spots of IRV may well be worse  
>>> than those of Condorcet methods (for most typical use cases in  
>>> politics). Different problems may have different weight in  
>>> different political environments. For example in countries with  
>>> strong two-party tradition and single party government some  
>>> Condorcet properties like the possibility of electing candidates  
>>> that do not have strong first preference support in the ballots  
>>> may work against it (both in the case that one does not want the  
>>> system to change and in the case that one wants to renew the  
>>> system). Also strategic voting and fraud related problems (like  
>>> later no harm, burial, precinct counting) may be seen in different  
>>> light in different societies, e.g. in countries where strategic  
>>> voting is the norm vs. in ones where sincere voting is the norm.  
>>> One may thus have/develop points of view where Condorcet looks  
>>> worse than IRV (I guess it could also be worse for some uses in  
>>> some societies from some points of view).
>>
>> So some whose backers tend to rank them top object when they lose  
>> because more other voters rank them below others?  I like better  
>> Condorcet's having A win over B when more voters prefer A>B to B>A  
>> without caring which is top ranked by more.
>
> There could be some special cases where you really want first  
> preference support to exist. For example you could let the soldiers  
> vote on which country to attack. Eager front row fighters might be  
> needed. Maybe there are also other such environments (maybe more  
> realistic too :-).
>
If a voter ranks A=5, that has no strength in the A vs B competition  
if that ballot ranks B higher.  BUT, if that ballot ranks B lower, A  
should get credit for being liked better than B without having to be  
ranked top.
>>
>> single party government?  If some voters rank B>A and some rank  
>> C>A, A still wins if the total is A>B and A>C.  Good to get vote  
>> counts that identify what A is seen as doing well, and what it had  
>> better improve.
>
> I guess many of the fears of Condorcet electing too weak leaders are  
> based on the assumption that a party/candidate with 20% first  
> preference support would less capable of forming or leading a  
> government and a country than a party/candidate with 40% support. In  
> multiparty countries/governments it is more common to think that  
> whoever is the biggest of the coalition parties or who is the most  
> trusted leader among the coalition members may take the lead.

BUT, if the elected leader only had 20%, then in the collection who  
got 80%, each of those candidates must have ranked way low on many  
ballots, and thus have less claim to leadership ability (the one with  
40% was liked by many, but must have been more disliked by many to  
fail to get elected here).

Dave Ketchum
>
> Juho
>>
>>> Juho
>>>
>>> P.S. One more reason is that Condorcet promoters seem to be lazier  
>>> that IRV promoters :-). Condorcet has made some progress in the  
>>> academic circles but not yet in politics.
>>
>> Agreed, though this is not a reason why IRV deserves to win this  
>> race.
>>
>> Dave Ketchum





More information about the Election-Methods mailing list