Tactical Voting

In response to a blog conversation on Planet Debian, Wouter Verhelst writes about voting tactics in which he claims that Instant-Runoff (wikipedia) (the method used in Australia to elect members of the house of representatives) is broken.

I have read the Wikipedia review and neither it nor my previous understanding of Instant-Runoff leads me to believe that it is broken or prone to tactical voting problems.

The main tactical voting issue in Australian elections is to give first preference to a small party (such as The Greens) in a seat where there is almost no chance of the small party winning. The idea is to have the vote effectively be a vote for whichever of the two major parties the voter prefers while also sending a message (by the first-preference votes which are widely publicised) as to their desires. If a major party wins a seat while 20% of the first preferences went to the Greens then they might find it an incentive to try and get some Green policies adopted if they want to win the next election (whether this actually works is debatable but it is a fact that people try it).

There are many people who don’t understand how the Australian voting system works and believe that they have to vote for one of the major parties to avoid a “wasted vote”. This is not due to the voting system but due to American media – and ill-educated Australians who can’t work out which parts of the news apply to them. Wouter has an advantage in living in a country where English is not the primary language – English news does not directly concern him and his country-men can probably work out that English-language voting instructions should not be followed!

The site has a useful analysis which demonstrates one theoretical flaw in IR, but realistically getting the average voter to use Condorcet would essentially be an IQ test as a pre-requisite to voting (numbering all the candidates in order for IR is difficult enough).

The fact that an election system is not theoretically fair is not a valid criticism unless there is a viable alternative which is better. If the lower house is to actually represent the people of each electorate (which is rarely the case nowadays) then IR seems to be the best option. If members of the lower house are not expected to represent geographic regions then Wouter’s advocacy of proportional representation makes sense.

As for the Australian Senate, the voting system for that is far too complex and I think that as a matter of principle any voting system which can not be understood by the majority of voters is wrong. A good case could be made for proportional representation in the senate.

Update: In response to Wouter’s post – tactical voting that doesn’t change the outcome of the election and merely sends a message to a political party is OK IMHO. But if you convince enough people that tactical voting is the only option in the US then the result may change.

4 comments to Tactical Voting

  • Adam

    “[…] getting the average voter to use Condorcet would essentially be an IQ test as a pre-requisite to voting (numbering all the candidates in order for IR is difficult enough).”

    Um, but all a voter has to do for Concordet is number all the candidates in order. How would using Concordet to figure out the results of the vote make things any more difficult on the voters?

    And the basic idea behind Concordet isn’t that hard to figure out. The only difficulty arises when there’s a circular tie, at which point one option is to just use IRV for the Smith set (the smallest set of candidates such that every candidate in the set can beat all candidates outside the set).

  • Anonymous

    IRV and Condorcet require exactly the same information from the electorate. You say that “getting the average voter to use Condorcet would essentially be an IQ test as a pre-requisite to voting (numbering all the candidates in order for IR is difficult enough)”, but Condorcet requires precisely the same thing: numbering the candidates in order.

    The only difference lies in how it uses the information: IRV ignores any preferences you have other than your first choice, until it eliminates your first choice, while Condorcet always takes all of your preferences into account. For instance, if you vote A, B, C, in order of preference, IRV will treat that as “A” unless it eliminates A, and IRV will ignore that you prefer B over C. This can then allow IRV to eliminate B before C, even if a majority of people prefer B over C, if most of those people don’t list B as their first choice. Thus, if you care about having B win over C, you do have to decide whether you really want to rank A first or B first, rather than listing your real preferences.

    For the record, I wrote that explanation of a flaw in IRV entirely from memory, without consulting any references. I don’t think the average voter would have trouble understanding that explanation either.

  • Jackson Boyd

    You’re right. There’s this manic fringe in the U.S. that thinks alternatives like approval voting and range voting are The Answer to reformers’ problems, but can’t get their minds around the fact that it’s transparently obvious how to game their systems (e.g, you should bullet vote, pure and simple, or you may well hurt your first choice), while it’s quite opaque how to try game IRV. That’s why it has such a long history of success in major national elections and major non-governmental elections without complaints of tactical voting, while approval voting doesn’t work very well where it’s used in non-governmental elections and has zero support among elected officials and grounded reformers pretty much everywhere in the world.

    They will even say “runoffs work just fine,” not being aware apparently that the math of their alleged “spoiler” problem with instant runoff voting is just the same as it with runoffs.

  • Please see (particularly for comparisons of the mathematical properties of election methods. I agree with posters 1 and 2 who say that Condorcet is better.

    Additionally, Concorcet can be computed faster than IRV (in a country of 200 million people, this is no small thing), and ballots can be aggregated into a matrix of counts, an operation that is associative, communtative, and distributive, which means it can be performed during the election as ballots are coming in, even as they are cast. And it can be done by map-reduce. You can’t do any of that with IRV.