Friday, July 1, 2016

Aggregated polling update

On the face of it, the polls are close, albeit with the Coalition slightly ahead. In two-party preferred (TPP) terms:

  • The latest Galaxy poll (28-29 June) has the Coalition on 51 to 49 for Labor.
  • Ipsos (26-29 June) has it as a tie: with 50 per cent each to Labor and the Coalition
  • Newspoll (24-26 June) has the Coalition on 51 to 49 for Labor
  •  ReachTEL (23 June) also has the Coalition on 51 to Labor's 49

In aggregate, and assuming any house effects sum to zero, I get an aggregated poll result of 50.6 to the Coalition to 49.4 to Labor. If I anchor the TPP aggregation to the 2013 election result, I get 51.6 to the Coalition and 48.4 to Labor.

Notwithstanding the long campaign, the Coalition has improved its position in the final weeks. If the Coalition wins the election, this will challenge the conventional wisdom that governments do poorly in long campaigns (based on the experience of the 1984 election campaign).

Moving from a TPP estimate to the number of seats each party will win is extraordinary complicated this election. The biggest confounding factor is the Nick Xenophon Team (NXT). Will they win a swag of seats, a couple of seats or none? And will they change the way preferences flow? The other confounding factors are the re-emergence of the independents Windsor and Oakeshott in rural New South Wales and the good showing for the Greens in the seat of Batman.

To address these confounding factors I will model 1,000,000 election outcomes using a mix of seat swings in 138 seats and longshot bias adjusted seat odds in 12 seats (namely, Melbourne (VIC), Denison (TAS), Indi (VIC), Kennedy (QLD), Fairfax (QLD), Cowper (NSW), New England (NSW), Barker (SA), Boothby (SA), Grey (SA), Mayo (SA), and Batman (VIC)). From this Monte Carlo Simulation, I will use the most common results to forecast an election outcome.

I start with the anchored poll result above (51.6 per cent for the Coalition; which in round terms is a 1.9 percentage point swing to Labor on the 2013 election). I make adjustments for different state swings based on Newspoll data as follows (where a negative number is a greater swing to Labor and a positive number is a lesser swing to Labor).

0 -0.1 1.7 -2.5 -0.8 2.5 0 0 0

The resulting analysis identifies these seats as the most likely to change hands.

TPP 2013 TCP 2013 2013 Outcome 2016 Favourite
Bonner (QLD) LNP 3.7% - Coalition Labor
Brisbane (QLD) LNP 4.3% - Coalition Labor
Capricornia (QLD) LNP 0.8% - Coalition Labor
Fairfax (QLD) - PUP 0.03% v LNP Palmer United Coalition
Forde (QLD) LNP 4.4% - Coalition Labor
Lyons (TAS) LIB 1.2% - Coalition Labor
Mayo (SA) LIB 12.5% - Coalition NXT
Petrie (QLD) LNP 0.5% - Coalition Labor
Solomon (NT) CLP 1.4% - Coalition Labor

But this analysis ignores close seats like Batman, where the incumbent (Labor) is only just ahead of the Greens according to betting markets, and the result could easily go either way. We can address this problem with a Monte Carlo simulation of the probabilities for each seat. The results of the simulation are:

My election prediction: the Monte Carlo simulation of the polls suggests the most likely outcome is 79 Coalition, 65 Labor and 6 others (comprising independents, Greens, Katter and NXT representatives).

Turning to Primary votes: the campaign has been kind to the Coalition and unkind to the Greens. Labor declined for much of the campaign, but has managed a very late and small recovery. The independent vote looks like being particularly high, and will make this an that could see some surprise upsets.

Finally, a comparison of all my models.

1 comment:

  1. This and your betting analysis is simply outstanding. Well done sir