Friday, October 19, 2018

Oh Yes We Do Have Strategic Voting In Australia (Sometimes)

On Wednesday Alex Turnbull, who has been campaigning for voters to evict the Liberal Party from his father's former seat of Wentworth, switched his support from Labor to independent Kerryn Phelps on strategic grounds.  Amusingly, Turnbull jnr justified his support by reference to a popular American text called "Gaming the Vote" by William Poundstone, and posted a colourful excerpt explaining how the squeezing out of a centrist candidate who finishes in third places can lead to "unpalatable, Wizard-or-Lizard dilemmas".  After Clinton-vs-Trump, or even the utter farce that has been the present term of Australian parliament, wizards and lizards are both sounding pretty good at the moment.

The idea here is very simple: if Kerryn Phelps makes the final two she is more likely to beat Dave Sharma (Lib) on preferences from Labor (which will include some votes originally for the Greens and minor candidates), than Tim Murray (Labor) is to beat Sharma on Phelps' preferences should he make the final two.  Phelps is (mostly) seen as the more centrist candidate in an electorate that has never elected a Labor MP.  There will be voters who want to send a protest vote against the Liberals for disposing of that other Turnbull or whatever other reason, but who cannot bring themselves to vote Labor.

And yet, I have seen a few politically active people on Twitter claim (and persist in claiming after being refuted) that strategic voting is a non-issue in a preferential system, that we don't ever have it in this country and never need to, and that all a voter need ever do is put the candidates in order of preference and the system will process that as best for that voter as it can.

Now, normally it doesn't bother me if people who seem to be smart enough to know better insist on damaging their own voting power.  Also I have no interest in promoting a Phelps vote myself and I'm not sure I'd put her above either major party's candidate if I had the fortune to be a Wentworth voter.  But above all else, this goes to the question of our electoral system, what it is good for, what are its strengths and weaknesses, is it prone to gaming, and does it sometimes leave voters with difficult strategic choices that they do not have the information needed to make.  These questions need answers and the fact is that an incentive for strategic voting is a real, if rare, thing in our House of Reps system.

Here are some examples of strategic voting scenarios in Australian single-seat elections with preferencing (either compulsory or optional, the same points apply in each case).  The extent to which strategic voting actually occurred in each case is not that relevant; the point is that there was clear room for using it if voters were concerned with a specific outcome.

* In the 2009 Frome by-election in South Australia, independent Geoff Brock won after coming third on primaries.  After the preferences of minor candidates Brock reached the top two by a whisker (30 votes).  80% of Labor's preferences flowed to Brock and he won by 665 votes.  But had Brock been excluded, only 68% of his preferences would have flowed to Labor, and Labor would have lost by 643 votes.  Therefore, if someone's sole concern was beating the Liberals, their best strategic vote was to put Brock ahead of Labor and the Liberals last.  Those who voted Labor-Brock-Liberal actually nearly caused the Liberals to win, and indeed had 31 Liberal voters voted 1 Labor, the Liberals would have won the seat with Brock eliminated.

* In Denison in 2010, Andrew Wilkie also won from third.  After he narrowly held off the Greens for third place, Wilkie received a flood of Greens preferences and jumped the Liberals to easily make the final two.  He was then narrowly elected over Labor on Liberal preferences.  But had the Liberals made the final two, Wilkie's preferences would have elected Labor with a large margin.  The Liberals always knew they would never win Denison so it made strategic sense for Liberal voters wanting to kick Labor out of the seat to vote for Wilkie ahead of their own party.  And quite a few probably did.

* In the recent Wagga Wagga by-election there was a close three-way split between independent Joe McGirr, the Liberals' Julia Ham and Country Labor's Dan Hayes.  Hayes was eliminated in third and his preferences flowed massively to McGirr who scored an enormous win with nearly 60% two-party preferred.  Had McGirr been eliminated at this stage (which he avoided by 3.4%), Hayes would still have won on his preferences, but it would have been extremely close (just 106 votes in it).  So for those who only wanted to beat the Liberals, McGirr was easily the safer strategic vote, while a vote for Hayes was a much bigger risk.

* In the seat of Rockhampton in Queensland 2017, the ex-Labor independent Margaret Strelow was second on primary votes and appeared likely to win the seat from Labor on LNP and One Nation preferences.  But there was a problem - Strelow didn't stay second.  Preferences from the LNP candidate in fourth put One Nation into second and Strelow was eliminated, with her preferences causing Labor to win the seat.  Had just 372 of 3076 voters whose voted flowed from LNP to One Nation instead strategically preferenced Strelow, Strelow would have made the final two and beaten Labor.  Instead, these LNP-One Nation voters, many of them following the LNP card, caused Labor to win the seat moving Labor one more seat away from the embarrassment of minority government.  Admittedly for a conservative voter there was probably little to choose between Labor and Strelow, but such gifts to endorsed Labor candidates made the LNP's attempt to blame One Nation for Labor's majority hypocritical (not to mention factually wrong).

Of course, the vast majority of seats are not like this.  In nearly all seats everyone knows who the final two will be and it is only how the voter orders those two that makes a difference (except for public funding).  There are also other cases, like the three-cornered seat of Prahran (Vic) 2014, where strategic voting could have made a difference, but there was no way to predict how to do it (whether Greens votes will flow more strongly to Labor than vice-versa varies a lot between seats).  However, the case of Wentworth is the stock-standard case in which strategic voting is an issue: the two major parties against a prominent independent who is likely to be preferred by each party's support base over the other.

In any such case, if someone's sole concern is beating one of the major parties, then the correct strategic move is to preference the indie above both major parties, putting the party they want to lose last.  If the indie ends up being uncompetitive, it doesn't matter anyway as their preference flows to the preferred major party at full value.

It's not so simple for a voter in this position: They would ideally like Labor to win the seat. Phelps is their second preference and they definitely don't want Sharma to win.  This one, as with the choice facing Labor supporters in Wagga Wagga, is a tricky case of how the voter might balance risk and reward.  Is it worth voting for Murray and increasing his chance of winning when by doing so you also increase the chance of Sharma winning?  Or is it better to play safe, sacrificing Murray's chances in order to reduce Sharma's?

Ideally, to make this decision, the voter would like to have answers to these questions: If Phelps makes the top two, how likely is she to beat Sharma?  If Murray makes the top two, how likely is he to beat Sharma?  If the answers are, say, 80% and 50%, but you think a Murray win would be four times as good as a Phelps win, you might well still preference Murray over Phelps.  However, polling is nowhere near accurate enough to answer the probability questions.

As it happens only two of the ten (!) Wentworth seat polls have looked at this question.  The Refugee Council ReachTEL found Phelps winning 53-47 if in the final two but Murray scoring 50-50 if he's in the final two. The Voter Choice opt-in experimental poll actually found Murray winning by a slightly larger margin than Phelps (55.6 vs 55.4) but that should be treated with extra caution as it is contrary to the past history of these things and to Phelps' history of all-over-the-place political positioning.  Both results would be based on trivially small sample sizes.  If the first poll is accurate then the risk difference in voting for the two candidates is large.  But if there has been a swing away from Sharma to both opponents since the poll was taken, the risk difference reduces.  The problem is that seat polling in Australia is such a load of rubbish that if the polls say Labor is marginally ahead or marginally behind the Liberals on a 2PP basis, that might be completely wrong.

There is also a strategic voting issue for Liberals (see the Frome discussion above).  If they believe they are definitely going to lose to Phelps should Phelps be their opponent, then a small number of them may as well vote 1 Murray and attempt to knock Phelps out.  But in doing so (especially if too many do it) they increase the risk that if Murray does make the final two, he wins (which is an even worse result for the party, except that it increases their chance of winning the seat back at the election.)

Strategic voting scenarios are common in our various Upper House systems too, though that is beside the point of the current article and has been explored elsewhere on this site (see "tactical voting" section).

How could this be fixed?

The incentive to vote strategically in this way (in these rare cases) could be removed if Australia used one of many possible Condorcet systems.  In a Condorcet system, if a candidate is preferred by the voters head-to-head over every other candidate, then that candidate (in this case potentially Phelps) always wins. In such systems, there will always be a pairwise comparison between Phelps and Sharma, so there is no reason for someone who prefers Murray over the others to put Phelps ahead of Murray.

However,  Condorcet systems have practical drawbacks.  At least one candidate has to be compared pairwise against every other candidate to find the winner, so in practice I'd expect full data entry of all ballots (as in NSW) to be required.  Secondly in rare cases there isn't a Condorcet winner, so some form of consensus would have to be reached on the most appropriate tie-breaking method (and what risks that created of strategic voting too).  Discussion of this tends to quickly reach a level of nerdiness that the average voter has no hope of following.

There are very few cases based on actual vote data where the Condorcet winner would have been different to the person who actually won (Rockhampton mentioned above is probably one). My perception is that the Condorcet winners have a very high strike rate in these three and four candidate contests.  But what impact a Condorcet system would have on voting over time is hard to predict.  If it led to more focus on intermediate candidates and further eroded the primary vote of major parties, it might lead to a system where several candidates were competitive for certain seats.  That sounds great but I am just a little wary of the potential for it to unlock the door to how-to-vote-card preference trading, a scourge we just (largely) got rid of from our Senate system.

Anyway, as the strategic voting issue is a rare novelty in Australia, rather than a frequent necessity as in first-past-the-post systems like the UK's, I doubt there will be a serious push for change on this front any time soon.

"But Strategic Voting Is Wrong"!

There is a view in some circles that one should never advocate strategic voting and one should always put the candidates solely in one's preferred order and let the system sort out the mess.  Here I take no view on the morality of strategic voting - it is up to any voter to decide whether they wish to do it or not.  However it is simply a fact that our system has some weakenesses and that it is sometimes possible, in rare cases, to cause the candidate you least want to win to succeed by voting 1 for the candidate you most want to win.  I think that knowledge that this can be the case should be out there, and people can make their own decisions on what they do about it.  I only wish the polling data had the quality needed to provide better information about the risks in given cases!


  1. Good analysis. I think however that strategic voting can only be done through how to vote card given out by candidates. Individually the voter in a rush to put his vote and avoid the potential fine would not remember all the candidates names and what they stand for and more crucially how those candidates are directing there preferences. If greens are the benchmark for informed candidates then they make up only 7 to 8 percent of the voting population. I am not sure wether you have done any study on the percentage of informed voters.

    1. The fine is only if a voter doesn't turn up at all; once they have turned up and been ticked off they are no longer at risk of being fined for not voting. I think what we're actually seeing here is the use of social media to counteract how-to-vote cards that make strategically risky suggestions. How well it works we'll have to see.

    2. wow the mathematics of voting is interesting.... is there anyway people can tell strategic liberal voting from the election figures

  2. According to the Gibbard-Satterthwaite Theorem, even the Condorcet criterion is strategically manipulable. More generally, it's quite plausible that in many marginal seats in the late 70s and 80s, the Australian Democrat candidate would have been the Condorcet winner. Simulations assuming all preference orderings are equally likely show that if the number of voters is large - as in a public election - and the number of candidates is 6 or greater, there is about one chance in three that there will be no Condorcet winner. See William H Riker and Peter C Ordeshook, An Introduction to Positive Political Theory, Prentice-Hall, Inc., Englewood Cliffs, New Jersey, 1973, pp. 100-9.

  3. It seems as if my soundest strategy is to always vote independent, with the opposition and government candidates last. Our former prime minister has shown that no matter how good the candidate or grand the title they're given, their wishes are nought against the mob mentality of their party. Any human, free of the shackles of organised democracy will surely work harder for their electorate and their country than one whose primary goal is the reelection of their party.

  4. maybe it possible for scrutineers to watch for lab 1 lib 2 votes

  5. Thanks for this Kevin - a good outline of the issue (including recognising that even in the unlikely event that a Condercet system is introduced, it still wouldn’t fully solve the problems inherent in any single member electorate system).

    I think the key part of your piece is
    “If Phelps makes the top two, how likely is she to beat Sharma? If Murray makes the top two, how likely is he to beat Sharma? If the answers are, say, 80% and 50%, but you think a Murray win would be four times as good as a Phelps win, you might well still preference Murray over Phelps.”

    This is an important reminder that strategic voting isn’t just about who is most likely to win, but also how strong an impact does the voter believe differing outcomes will be.

    My guess would be 80-90% of Greens voters in Wentworth would prefer a Phelps win to a Liberal, but the political impact of a (definitely less-likely) Labor win would be immensely greater.

    Ironically, there is also the counter-point you mentioned, which is that Phelps would be more likely to hang on to the seat should she win it in the by-election than the Labor candidate would. (Which in itself reinforces the fact the she is much more Liberal leaning, hence giving Greens and other progressive voters less reason to preference her).

    Should progressive voters preference Phelps ahead of Labor for the much higher likelihood that it would give the government a slap in the face, or should they preference Labor for lower likelihood but much greater reward of kicking them in the nuts, ripping out kidney and leaving them much closer to the brink of actually imploding.

    Short term versus long term impacts, as well as the impossible task of assessing the odds of the chances of Phelps versus Labor.

    Not sure what I’d do myself if I was a Wentworth voter.

    1. I'd like to see retrospective studies on elections about who the condorcet Winner would be in each seat. Every election there's results on what the MMP or FPTP system would have looked like.

      Getting into nutty territory but I think the ideal electoral system uses 2 different methods (eg IRV and Schulze/Condorcet) and has a runoff election if those 2 winners are different.

    2. Interestingly the NSW Electoral Commission provides pairwise comparisons based on which you could find the Condorcet winner in each seat if so inclined. However NSW has optional preferencing, which is much less interesting for this purpose.

  6. Could tactical voting/preferencing for SA Best from ALP and Greens voters have prevented a Marshall getting a majority in SA?

    1. The average preference flows in SA on my estimates were SAB to Labor vs Liberal 50-50 but Labor to SAB vs Liberal 69-31. That sounds promising but I cannot find a single seat where tactical voting by those ALP voters (or preferencers from the Greens) who preferred SAB to Liberal would have changed the result. Generally the SAB primary is so far behind the Lib primary that reallocating Labor votes puts the Libs so close to 50% that they would not be caught.

  7. I'm a month late commenting on this, but just thought I'd note a major risk with moving to a Condorcet system. It is natural for supporters of one major party candidate to rank their chief opponent, usually the other major party candidate, last, regardless of whether they are their true last choice. The result of this voting behavior could be the election of very weak candidates that voters known little about.

    Consider a race with a candidate for major party A, one from major party B, and little-known independent candidate M. Supporters of A preference A>M>B, not because they genuinely know and prefer M second, but under the assumption that it does the most damage to B. Similarly, supporters of B preference B>M>A. The only voter to preference M first is M himself. Every Condorcet system elects M in this case, yet it remains unclear whether M is truly capable of beating either of the two major party candidates head-to-head. Under optional preferential, we may see an even higher number of 1st-preference-only votes to avoid such outcomes.

    A key strength of the preferential system today is that a vote for a later choice never hurts an earlier choice, and there is no advantage gained from ranking a formidable opponent last. This is why Nanson concluded that it was "extremely suitable for political elections."

    Anyway, I wanted to note some additional risks to Condorcet adoption.