Friday, September 25, 2020

Could Just 2000 Shifting Votes Swing The ACT Election?

 Advance Summary


(This article is rated 5/5 on the Wonk Factor scale.  It is extremely mathsy and technical.)


Yesterday the ABC published an article that claimed that the Liberals could win the 2020 ACT election if just 2,000 ACT voters switched their vote compared to how they voted in 2016.  The article is still up and the author continued to defend it after both Tim Colebatch and I independently pointed out on Twitter why it was incorrect, so here is an article to explain in detail why this claim is not correct.  In the process I hope to highlight that interpreting Hare-Clark spreadsheets really is rocket science and that a simple question like "how close was the election?" can have a very complex answer.  

The article's core claims are as follows:

* In Murrumbidgee, the Greens' Caroline Le Couteur was 800 votes ahead of the Liberals' Peter Hosking, so that difference would have been erased had 400 voters for Le Couteur switched to the Liberals.

* In Yerrabi:

"In 2016, sorting through this electorate's mess of preferences was more complex than elsewhere.

But when it mattered, Liberal prospect Jacob Vadakkedathu was 2,789 votes short of becoming an MLA.

If 1,395 voters had preferred him over Labor's Michael Pettersson or Suzanne Orr, he might have had their seat in the Assembly."


It is correct that had 401 voters for Caroline Le Couteur (this could include votes that preferenced her and reached her at full value) instead voted for Peter Hosking, or preferenced Hosking in such a way that their votes reached him at full value, Hosking would have won.  But this is not the same as saying that if 401 Greens voters switched to the Liberals at full value, Hosking would have won.  The reason is that a vote that switched from Le Couteur to Giulia Jones (Lib) would have decreased Le Couteur's tally, but not increased Hosking's.  A vote that switched from Le Couteur to become a primary vote for Jeremy Hanson, even if it was 1 Hanson 2 Hosking, would have decreased Le Couteur's tally, but would not have flowed through as a full vote benefit to Hosking (because it would have increased the value of all Hanson's surplus votes, many of which favoured Jones).  Thus the idea that 401 votes swinging would change the seat result is artificial, because it relies on all those votes moving to a specific candidate (which would never happen in real life).  A more realistic estimate is 570 votes.


In the case of Yerrabi, the article makes an error concerning the point at which Vadakkedathu's position mattered.  Vadakkedathu was 2789 votes short of quota after Count 28, at which point he had 5754 votes; quota was 8543.  But this is not the point at which it mattered, and this is a product of the ACT's habit of continuing to throw preferences past the point at which a contest has been mathematically decided.  (Markus Mannheim is in good company when it comes to getting tripped up by this quirk - even the Tasmanian Electoral Commission has been led astray here, in a different way.)  

In the contest for Yerrabi, nobody polled a quota on first preferences.  At count 16 (see Yerrabi table 2 here) Megan Fitzharris (ALP) was elected on the preferences of the first excluded Labor candidate.  At count 18 Alastair Coe (Lib) was elected on the preferences of the second excluded Liberal candidate.  Exclusions continued, reaching this point:

Milligan (Lib) 6899

Vadakkedathu (Lib) 5358

Petterson (ALP) 8086

Orr (ALP) 6855

Wensing (Green) 4968

At this point Wensing was excluded, and Count 24 consisted of Wensing's full-value votes (4900 votes).  Of these 62.3% flowed to Labor, just 5.8% to the Liberals, and the rest exhausted.

After this throw the numbers were:

Milligan (Lib) 7028

Vadakkedathu (Lib) 5511

Petterson (ALP) 9309 

Orr (ALP) 8683

Wensing (Green) 68 (part excluded)

At this point the election is already decided.  Petterson and Orr are over quota and have won so Labor has three seats.  There are only 974 votes left to throw and Milligan is 1517 ahead, so Milligan will certainly win the fifth seat.  But the ACT system continues distributing preferences anyway.  

What is important here is that Petterson and Orr can no longer receive those preferences as they have reached quota.  The two Liberals can receive them, but since they are mostly votes from the Greens that have already preferenced Labor, it's likely most of them would have actually gone to Labor had that still been possible.

Milligan reaches 7279 and Vadakkedathu 5754 at the point where the latter is irrelevantly excluded, but they only got most of those last few hundred votes because  Labor was blocked from getting them.  

Furthermore, the number of votes that need to be switched to cause the Liberals to win cannot be calculated simply by looking at how short Vadakkedathu is of quota, because just taking enough votes off Labor to put Vadakkedathu over one Labor candidate only causes him to win instead of the other Liberal, Milligan.

On the assumption that the 68 leftover votes for Wensing would not assist the Liberals more than Labor, let's see how many votes we have to shift to put a Labor candidate into last place.  Orr and the two Liberals between them have 21222 votes, so we need to bring Orr down to a third of that and the two Liberals up to the same figure.  This takes Orr down to 7074, removing 1609 votes from Orr.

However, we have a further problem: Petterson is still over quota by 766 votes.  We already know Wensing's votes flowed strongly to Labor, but those that flowed to one Labor candidate are much more likely to flow to the other, because this means the voter hasn't exhausted their vote at the end of the Greens' ticket and is also more likely to preference the Labor ticket than if all we knew was that they had continued preferencing.  On the other hand, they only have one Labor candidate left to go to, not two.  Let's say, conservatively, that Orr gets 70% of the surplus, the Liberals get 5% each, and the rest exhausts (though since it's the ACT, it may well be that nothing would exhaust; see below).  Now after Petterson's surplus, Orr is 498 votes ahead of the two Liberals.  We can fix this by taking another 332 votes from Orr and splitting them between the two Liberals, so we have now taken 1941 votes from Orr.  It might be a little bit more than that if no votes are actually exhausting.

But again, this is an extremely artificial scenario that relies not just on taking votes from Labor and giving them to the Liberals, but on taking votes from a given Labor candidate and giving them to two specific Liberal candidates in a very specific proportion.  It's unrealistic in practice and so the number of votes that would have to be moved in reality is much larger. It's actually extremely difficult to model, but a possible approach is: how many votes would we have to move at the key point if votes left Labor in proportion to the two candidates' totals, and arrived in the same proportion?

Here the answer is that Orr has 8683 votes (48.3% of Labor's total) and Vadakkedathu has 5511 (43.9% of the Liberals).  The gap is 3172 votes, and every vote that comes off the Labor ticket and goes onto the Liberals on average reduces that gap by 0.922 votes.  So we must move 3440 votes - far larger than the 1941 votes for a targeted artificial manipulation that would never happen in reality, which in turn is more than the 1395 votes specified in the article.  So the article's claim that:

"Four years ago, just 1,795 votes, across two electorates, prevented the Liberals from governing with a majority."

is false.  

(To further highlight that 1395 is incorrect, 1395 is the number of votes you would need to take from one of Petterson and Orr at the end and give to Vadakkedathu to put him in front of one of them.  But the fact that Petterson and Orr earlier in the cutup had more votes than they finished with should give some pause for thought that this is not the way to do it.)


I've already shown that the number of votes that must be moved to make the Liberals win, given realistic assumptions, is more like 4000 than 2000.   But there's more, because the electorates have been redistributed.  

As calculated by Ben Raue, in Murrumbidgee Labor loses 1.5%, the Liberals gain 1.73%, the Greens gain 0.09%.  The total vote in the seat was 50055, so that's +866 votes to the Liberals and +45 to the Greens.  Splitting the votes between the two remaining Liberals proportionally gives Hosking a 358 vote leg-up in the fight with Le Couteur.  Taking votes from Labor also gets rid of 99% of Labor's surplus of 760 votes, which benefited Le Couteur over Hocking by 603 votes.  To observers of Tasmanian Hare-Clark distributions, that may appear staggeringly high, but the ACT has a system that exhausts votes without continuing preferences and preferentially keeps votes with continuing preferences in the count.

All this means that the actual number of votes the Liberals need to be transferred to win in Murrumbidgee is zero (all else being equal); the seat is notionally Liberal by a very small margin as a result of the redistribution.  However, that assumes there are no transfers between other parties.  The Greens can retain the seat by taking votes from Labor, even if the Liberal vote stays the same, but if they take too many votes (around, say, 700) they can cause the Liberals to take Labor's seat instead.

In Yerrabi, it's a different matter.  Labor gains 0.63%, the Liberals lose 0.19% and the Greens lose 0.33% (323 votes, 97 votes, 169 votes).  That makes Labor's position roughly 320 votes stronger than indicated, pushing the number of votes that need to be shifted to probably over 3500 (approaching a 7% swing) in this particular electorate.

This shouldn't be surprising if we look at the 2016 primary totals in Yerrabi.  Labor beat the Liberals by 8.1% on primaries.  That alone would normally need a 4% swing to overturn unless the Liberals had something go right for them on the candidate breakdown, which they didn't (actually their 2nd and 3rd candidates are further apart than Labor's.)  But there is more, because the Greens, who polled 7.1% in this electorate, were eliminated, and their preferences favoured Labor.  

And this leads further to the generic problem with articles of the ABC type: if there is a big swing in one electorate, there won't be nothing happening in the others.  If thousands of votes are swinging in Yerrabi that probably means a massive ACT-wide swing with tens of thousands moving overall.


Hang on, who said anything about Brindabella?  Well, the point is seats shifting cuts both ways.  The Liberals could win government by gaining seats in Yerrabi and Murrumbidgee, but they are also vulnerable in Brindabella, which they would need to hold.  In 2016, Labor's Angie Drake lost by being 553 votes behind the Liberals' Nicole Lawder.  So if 277 of the voters who voted for Lawder instead voted for Drake, then Drake would have won.  But again, this is an artificial scenario because it relies on being able to take votes from one candidate of one party and shift them to a specific candidate of another party, distorting the proportion of votes within that party.  The real number of votes to shift would be larger.

But not so much larger anymore, because the redistribution has taken an effective 80 votes from the Liberals and handed them to Labor, so if those are redistributed proportionally, the minimum number to move to change the outcome comes down from 277 to about 245.  But by the proportional method I used for Yerrabi, the number of votes that need to shift without changing the breakdown within each party is more like 620, a 1.3% swing.  This makes sense because the Liberals beat Labor by 8.3%, which would normally require a 4.2% swing to shift.  However, Labor benefited from Greens and Sex Party preferences, and also from a more even split between its candidates.  

When the flag drops and counting starts many of these swing scenarios will go out the window, because of different exclusion orders, different fourth party candidates, different breakdowns between the parties and so on.  But overall, the key point is that the Liberals were not close to winning the 2016 ACT Election.  In particular, in Yerrabi, all else being equal they need a swing of several percent.  

Absent of a large swing, what would really spice up the ACT election is if a fourth-party candidate willing to work with the Liberals snatched a seat from Labor and the Greens in either Ginninderra (Belco Party, this is your life) or Yerrabi.  Given the candidate list the former seems more interesting. The pathway for the Liberals to victory then is to take Murrumbidgee from the Greens (which requires no swing at all) and to then form minority government with 12 seats.  This seems a more promising pathway than picking up the swing required to win 13 - but can any minor parties step up to the plate?

A Side Note On Exhaust

Finally, can we please not have headlines like "More than 15,000 exhausted voices weren't heard"?  This isn't in the article and plays into the same furphy spread by clueless opponents of Senate reform in 2016, the bizarre claim that exhaust is the same as disenfranchisement.  Every vote that is formal is heard.  The full value of every formal vote is counted as a primary vote, even if the voter only votes 1-5 (or just votes 1 and has their vote saved by the savings provision.)  The remaining value of the vote, which will be only part of it anyway if the vote has already helped elect somebody, leaves the count when the voter chooses to stop numbering boxes (or in rare cases makes a mistake).  This is (mostly) the voter's choice.   I generally advise voters to always number all the boxes (see How To Best Use Your Vote In ACT Elections) but if voters choose not to that is their right, and it does not mean the system has ignored their voice.  What the system has done is followed their instructions.

I will have live coverage of the ACT election on the night and of the counting over subsequent days. 


  1. Hi Kevin

    Excellent work as always.

    In the ACT do the savings provisions work for a vote if there's a numbering error? For example, if a voter marks '12' twice will the vote be coubted to preference 11 and then exhaust? Or is the whole ballot informal?

    And does it work the same way in Tas?

    1. In the ACT, so far as numbers are concerned, any ballot with a unique 1 is formal, no matter what happens after that. A ballot with two candidates ranked 1 or no candidates ranked 1 is informal. A ballot with a 1 and two 2s is formal but exhausts as soon as the candidate marked 1 is elected or excluded.

      In Tasmania each vote must contain the numbers 1-5 each once and once only. Duplication or omission errors after 5 are OK; it exhausts at the point of the first error.