Thursday, August 27, 2020

Preference Flows And The Eden-Monaro Result

 The much-awaited Eden-Monaro preference flows are finally out, following some time after the distribution of preferences, and it's useful to make some comments about these.  (Also see Antony Green's comments.) 

Labor won the by-election by 735 votes (50.39% 2PP).  This means that had 368 voters who preferenced Kristy McBain instead preferenced Fiona Kotvojs, Kotvojs would have won.  This is relevant for checking various claims about preferences of particular parties deciding the result.  In my Eden-Monaro live thread I set some markers at the time for particular preference shifts causing or not causing the result.  

Did Nationals preference flows weakening cause Labor to win?

Answer: Yes - but this isn't necessarily what it looks like.  In Eden-Monaro in 2019 an unusually strong 87.16% of Nationals preferences flowed to Fiona Kotvojs.  At the 2020 by-election the flow was only 77.73%, making a difference of 571 votes to each side's total, and a difference of double that to the margin.  While it was unrealistic for the very high flow from 2019 to have been maintained, any flow above 81.08% would have resulted in the Liberals winning the seat.  That figure is almost exactly the 2019 national average in contests where the Nationals were excluded, so the switch from an above-average to a below-average Nationals to Liberals preference flow can be said to have cost the Liberals the seat.  

This is significant because of claims that some Nationals voters were deliberately preferencing Labor in order to keep the path clear for a run by John Barilaro in 2022.  However, it's too simplistic to consider that the only factor, or even necessarily a significant factor at all.  For instance, I get that the preference flow in Queanbeyan booths weakened by 12.5 points, compared with 7.6 points elsewhere.  This would have been at least partly caused by Nationals' candidate Trevor Hicks' profile as a local councillor, which would have attracted voters who might not have voted for a generic Nationals candidate, with these voters then being more likely to preference Labor.  There were substantial swings to the Nationals in the Queanbeyan booths.  All this suggests that at least some of the Nationals "leakers" were not voters switching from National-Liberal-Labor to National-Labor-Liberal in their ordering of these three parties, but rather, they were voters who had preferenced Labor ahead of the Liberals in 2019 anyway, and in this case chose to vote 1 for the Nationals candidate.

Did the Shooters preferences flowing to Labor cause Labor to win?

Answer: No.  The Shooters, Fishers and Farmers preferences flowed 55.98% to Labor, favouring Labor by 606 votes.  But this is less than Labor won by, so even had the Shooters preferences split evenly, Labor would still have won.  For Labor to have lost, the Shooters preferences would have had to flow 51.27% to Kotvojs, a 7.25% difference.

It should also be noted here that Labor's advantage on Shooters preferences was 0.64%.  Around half of this, if not more, would have been the donkey vote, so the split of deliberate Shooters preferences to Labor would have been very weak.

When the distribution of preferences was first released, it looked like (as Antony comments) the Shooters had caused Labor to win, because Labor jumped from second to first on that final exclusion.  However it turns out that the votes received by the Shooters from other excluded candidates flowed more strongly to Labor (57.3%) than those that were 1 Shooters (56.0%).  This might have been caused partly by partial donkey voting (numbering a few boxes then filling in blank squares from the top) but there's actually not a lot of evidence for this.  In particular, a simple model for predicting the preference flow of the votes received by the Shooters (multiply the tally received from each party by that party's 2PP split) predicts a 57.1% flow of those preferences to Labor, almost exactly what happened.  The Shooters did receive a lot of preferences from HEMP and substantial numbers from Porter and the Greens, after all.

Did the Shooters' decision to preference Labor rather than the Liberals on their how to vote card cause Labor to win?

Answer: Possibly.  As noted above, the Shooters' flow to Kotvojs would have had to be just 7.25 points stronger for her to have won.  It is plausible a minor party's card could have made this much difference, but it's also plausible it didn't.  The past evidence re Shooters flows is messy because of the different nature of seats in which they have made certain preferencing decisions, but really there isn't enough evidence to say.  A close examination of the ballot papers would determine what percentage followed the card exactly, but this is unlikely to be conducted.

Did any other parties' preferences cause Labor to win?

Answer: Well, The Greens, obviously.  We know that Labor frequently gets strong flows from the Greens so this is hardly anything they wouldn't have depended on - but there is more to it than normal.  In fact, the flow of Greens preferences in Eden-Monaro increased from an already high 87% to a super-high 90.9%, which added 420 votes to Labor's margin.  This flow was stronger than the Greens to Labor flows in all bar two seats (Canberra and an estimate for Cooper) in 2019.  Had the flow from the Greens fallen by just three points to 84%, Labor would have lost.  

But there was more.  Preferences from HEMP split 67.27% to Labor, benefiting Labor by 744 votes (nine more than their margin of victory) so had the HEMP preferences split evenly Labor would also have probably lost (by a handful of votes), though in that case the result would have been so close that a recount might have overturned it.

Also, Labor gained 547 votes net from the Science Party, 300 from Karen Porter (unregistered "New Liberals"), 248 from Sustainable Australia and 218 from independent James Holgate, so various combinations of these (Science plus one of the others, or all three excluding Science) can be credited with Labor's victory on preferences, in the sense that had their preferences split evenly Labor would have lost.

That's not to say these parties running caused the result - they didn't.  Most likely they just took votes from both major parties, more from Labor than the Liberals, and these votes returned to their source in the form of preferences.

The major cause of Labor's win

I believe that for all the diving one can do into the mechanics of preferences, the biggest thing that got Labor home at this election was candidate quality.  That is not to say Fiona Kotvojs was a bad candidate - I think her performance in increasing her primary vote in a packed field in a by-election says otherwise.  But rather, the regionally concentrated swing in Kristy McBain's Bega Shire saved Labor from defeat, and is the single easiest factor to explain the variation in swings that has been seen.  A popular local mayor whose performance during the bushfire disaster had been praised got Labor over the line.  A generic party hack would have lost. 


  1. "However it turns out that the votes received by the Shooters from other excluded candidates flowed more strongly to Labor (57.3%) than those that were 1 Shooters (56.0%)"
    Does this mean that it's possible to access the full preference flows for all ballots (i.e. Including the flows from elected candidates)? I haven't seen it available on the AEC website, but I'd be interested to explore it for my own interest.

    1. No, that's not possible for House of Reps. What you get is a distribution of preferences (which shows how votes pass from one candidate to the next as candidates are excluded until the end) and a two-candidate preferred breakdown of the primary votes for each candidate, between the final two candidates. Each seat also gets a two-party preferred result between Labor and the Coalition, and in non-classic contests it's possible to use this to infer broadly about the 2PP split among voters for the non-major party candidate who made the final two.

      Complete data entry with all votes available occurs for the Senate and also for NSW elections.

      The Shooters finished third in Eden-Monaro and had accumulated quite a lot of votes from other parties at the point where they were cut out. From the distribution of the Shooters' total votes, and the 2PP results for votes that were 1 Shooters, it is possible to determine the combined 2PP for votes that passed to the Shooters from other candidates.

    2. Ah thank you, I didn't see the 2PP table at first. That makes more sense now. Shame it's not possible to get the full data entry!