Forecasting the 2016 ACT election
The Australian Capital Territory will elect it’s 9th Legislative Assembly in October this year. The Assembly will be expanded for the first time to 25 seats.
This means new electoral boundaries, more candidates, and new nomination strategies for the parties.
The chamber of the ACT Legislative Assembly
What impact will those new boundaries have on who wins the election?
Obviously the parties and the media will undertake polling later this year to see what vote the parties might win in each of the five divisions.
To prepare for such forecasting, the most useful thing we can do is make up an illustration of what result would have come out of the new boundaries at the last election in 2012.
The ACT Electoral Commission publishes election results broken down to the level of the 80 local polling booths, as well as the 6 pre-poll stations.
We can take the 80 local booths and identify them with the 5 new divisions fairly easily, and create new tallies of the party vote totals. (Individual candidate tallies become a bit meaningless here because different candidates ran in different old divisions, but we can at least tally party vote totals).
The ABC’s Antony Green did such an aggregation when the new boundaries were first published in April 2015.
One snag – which Green flagged – is that a very significant number of votes weren’t cast in local booths, but at one of the 6 pre-poll stations oven for 10 days or so prior to polling day. Other votes came in as postal votes, were cast interstate, or used other methods. In fact only 67% of votes were cast locally on polling day; the remainder were cast at pre-poll stations or were postal votes or other votes handled specially.
(By the way, it’s worth remembering that any ACT voter can vote anywhere they like; all ACT polling booths carry stocks of ballot papers for all electoral divisions. So doing an analysis using booth data is only a proxy for the ‘true’ voting profile of the suburbs around each booth. If course, it’s the best estimate it’s ever possible to get.)
It would be good to incorporate the 33% of votes that weren’t cast locally into our estimation. But we only know which of the old divisions these voters came from, not their local booth. To allocating them to new divisions we can use the 2012 booth total vote figures to create an estimate of what proportion of each old division’s voters have gone into each new division, and divide up the pre-poll and postal votes on that basis. It’s a bit rough, but obviously gives a much bigger sample.
Each of these two methods – using specific booth data for only 67% of the electorate, or using 100% of votes but resorting to an estimate of what location 33% of them came from – has its strengths and weaknesses. We’ll see in a moment how much their results actually differ.
I went about the problem in these steps:
- Allocate all the 80 local polling booths to the 5 new electoral divisions
- Identify all votes that were cast ‘at home’ by separating votes in polling booths which were clearly cast for out-of-divisions candidates
- Aggregate all the ‘at-home’ votes
- Pool together all of the following
- The remaining ‘not-at-home’ votes cast in polling booths
- All the votes cast at pre-poll stations
- All the other special votes such as postal votes, interstate votes etc
- Based on which old electoral division they were cast for, distribute all these ‘away-from-local-booth’ votes among the two (or 3) new electoral divisions covered, in proportion to the respective total number of ‘at-home’ votes cast in those regions.
- Sum the ‘at-home’ and the total ‘away’ votes by candidate in the new electoral divisions
- Sum the party totals of all those tallies of votes.
This gives us a matrix of vote totals for each party in each division, which in turn quickly gives us a matrix of quotas won (party totals) in each division.
And here are the results:
| Ginninderra | Yerrabi | Kurrajong | Murrumbidgee | Brindabella | total | |
| Quotas: | 7,493 | 6,213 | 8,469 | 9,673 | 5,020 | |
| Labor | 2.43 | 2.32 | 2.38 | 2.32 | 2.15 | |
| Liberal | 1.95 | 2.33 | 2.08 | 2.44 | 2.84 | |
| Green | 0.63 | 0.56 | 0.94 | 0.65 | 0.45 | |
| Motorists | 0.43 | 0.33 | 0.11 | 0.15 | 0.24 | |
| all other | 0.55 | 0.46 | 0.44 | 0.36 | 0.36 | |
| Seats: | ||||||
| Labor | 2 | 2 | 2 | 2 | 2 | 10 |
| Liberal | 2 | 3 | 2 | 2 | 3 | 12 |
| Green | 1 | 1 | 1 | 3 |
In fact, Antony Green’s calculations come to the same results, other than he rates the 5th seat in Yerrabi as “difficult to allocate” and leaves it open; I’ve given it to the Liberals.
If you compare the table of quotas won above to Green’s table, the main impact of leaving out the 33% of the votes which weren’t cast ‘at home’ in local booths seems to generally decrease the Liberals’ quotas and increase the Greens quotas compared to calculations based on all the data. That’s a noticeable and potentially significant difference, even if the distribution of ‘away’ votes across the divisions used above involves an assumption that they should be allocated in proportion to the booth sizes.
Overall, I think the Liberals in 2012 should be regarded as having been closer to winning a 13th seat on this hypothetical scenario than Antony Green’s analysis suggested, particularly because of the 2nd/3rd-placed candidate ratio issue outlined in this other post. And one more detail: the projected tally of 2012 votes onto the new seat of Murrumbidgee shows the Liberals’ 2nd and 3rd candidates were fairly close to each other in vote, which is the optimal position.
It follows that winning a 3rd seat in Murrumbidgee it the Liberal’s main – perhaps only – path to majority in 2016 (they must also secure a 3rd seat in Yerrabi, which might prove challenging).
Extrapolating Labor’s candidate configuration in Molonglo-2012 onto Murrumbidgee-2016 is pretty dodgy. In Molonglo in 2012 Labor’s 2nd candidate was well above the 3rd, but that was because of the unusual event that both the leader Katy Gallagher and deputy leader Andrew Barr ran in that seat; Labor’s deputy in 2nd place was therefore well ahead of their 3rd candidate. (In 2012 Molonglo was a 7-seat division, so the third candidate did actually get elected, but that’s irrelevant to 2016 forecasting.)
Labor probably won’t have the same candidate configuration in any seat in 2016, but in any case the moral of the story is: don’t nominate two very strong candidates in the same seat in front of a third candidate who will come a long way behind – the later will almost certainly get eliminated.
– 21 January 2016
On Elections 