Making sense of the polls

Making sense of the polls

Final House results and a polling critique

At the election held on 2 July, the Coalition won a bare majority of 76 of the 150 House seats, to 69 for Labor and five crossbenchers, representing a net loss of 14 seats for the Coalition and a net 14 seat gain for Labor from the 2013 election. The crossbenchers are Independents Cathy McGowan and Andrew Wilkie, the Greens’ Adam Bandt, Katter Party MP Bob Katter and Rebekha Sharkie of the Nick Xenophon Team (NXT).

The national Two Party Preferred (2PP) vote was 50.4-49.6 to the Coalition, a 3.1 point gain for Labor since the 2013 election. Primary votes were 42.0% for the Coalition (down 3.5), 34.7% for Labor (up 1.3), 10.2% for the Greens (up 1.6), 1.8% for the NXT and 11.1% for all Others. In SA, the NXT won 21.3% of the vote.

Conservatives have jibed that Labor’s primary vote was its second lowest since 1903, but the last election was only the second time since the beginning of the two party system in 1910 that the total major party vote was less than 80%, and this election has continued that downward trend. The Greens take substantial support on Labor’s left, and this is why Labor came close at this election despite a low primary vote.

Although they won only 76.7% of the overall vote between them, the major parties combined won 96.7% of House seats, showing how favourable the single member system used in the House is for the big parties.

If 2013 election preferences had been used, the Coalition would have won 50.5% of the 2PP, so actual preferences gave Labor a slight advantage over 2013 preferences. We will not know how minor party preferences flowed to Labor or the Coalition until the end of August.

The only close seat was Herbert, which Labor won by just 37 votes or a 0.02% margin. All other seats were won by at least 0.5%. Assuming the five current crossbenchers hold their seats and that all swings are uniform, Labor requires a 0.7 point swing to cost the Coalition its majority, a 1.1 point swing to win more seats than the Coalition, and a 1.5 point swing for a majority Labor government.

The swings required on the pendulum are now about right because sophomore surges are factored into the Coalition seats. That is, it is unlikely that the Coalition will be boosted by additional personal votes in seats they retained at this election after defeating Labor incumbents in 2013. The “sophomore surge” effect applies at the election after winning the seat, and after that it is factored into the seat’s margin.

Turnout at this election was 91.0%, down 2.2 points on 2013 and the lowest turnout at a Federal election since the introduction of compulsory voting in 1924. However, Peter Brent says this is because the electoral roll expanded to 95.1% of eligible voters, up from 92.4% in 2013. Turnout as a percentage of total eligible voters was 86.4%, up 0.2 points.

The informal rate at this election was 5.1% for the House, down 0.8 points. Brent says that the formal rate as a percentage of total eligible voters increased 0.9 points to 82.0% at this election.

State results

In the following table, I am using the post-redistribution pendulum to determine gains and losses. In 2013, the Coalition won 30 of NSW’s 48 seats to Labor’s 18, but this became 27 Coalition to 20 Labor, out of 47, post-redistribution. In WA, the Coalition won 12 out of 15, to Labor’s 3 in 2013, and this became 13 Coalition to 3 Labor, out of 16, post-redistribution.

Since all states swung against the Coalition, all swings recorded are negative. The Coalition gained a seat in Victoria, and this is recorded as a negative loss. Clive Palmer’s retirement meant his seat of Fairfax was assumed to return to the Coalition, and this is not generally regarded as a Coalition gain. All seats lost by the Coalition were to Labor, except Mayo in SA, where they lost to the NXT.

House final results.

The table shows that Labor made respectable gains in every state except Queensland and Victoria. A key reason for Labor’s underperformance in these states is that both now have Labor state governments. State Labor and Coalition governments tend to hinder their respective party’s Federal performance. Very unpopular state governments can drag the Federal party further down.

In Queensland in particular, had Campbell Newmann still been Premier, it is very likely that Federal Labor would have performed better, probably winning enough seats to at least cost Turnbull his majority.

Election result not a reaction to super reforms

There has been much noise from the hard right about how the superannuation reforms announced in the May budget cost the Liberals donations and votes. I cannot speak for donations, but the results show that super was not a factor in the unexpectedly close election.

Regions where Labor made its biggest gains were western Sydney and Tasmania, with Labor making six of its 13 total seat gains in these regions, which have relatively low incomes. On the other hand, wealthy inner city seats such as Kooyong in Melbourne and Curtin in Perth recorded small swings towards the Liberals.

The Poll Bludger has conducted a multiple regression analysis of the election results, and he has concluded that higher levels of education and income were associated with better swing results for the Coalition. If super had been a factor, we would expect better swing results for Labor in high-income seats.

There was a sophomore effect favouring the Coalition, but it was swamped by massive swings to Labor in western Sydney and Tasmania. However, in Queensland the Coalition held Capricornia and Petrie, which had been gained from Labor in 2013, while losing Longman and Herbert. The Coalition also held its post-redistribution regional NSW seats other than Eden-Monaro, and comfortably retained the three Victorian seats gained in 2013.

Polling critique

All of the final polls came within a point of the actual 2PP. Here is the final poll table, sorted by first date of fieldwork, with actual election results below the poll estimates. Bold in the table denotes a poll estimate that was within one point of the actual result.

Fed polls vs election.

Newspoll was clearly the best poll, getting all primary votes and the 2PP accurate to within 0.3 points, ReachTEL slightly overestimated the Coalition, but was good otherwise. Galaxy overestimated both major parties, and Ipsos overestimated the Greens by 3 points, and underestimated both major parties by about 2 points.

The poll results were too close to each other, and this implies herding, where all polls artificially move towards the same conclusion. In this case, herding gave the correct result, but at the 2015 UK general election all final polls were within 1% of showing a Labour-Conservative tie. The Conservatives won that election by 6.5%. In general, we should be wary of polls that are too close together.

The biggest example of herding at this election was Essential, which moved from 51-49 to Labor in the week before the election to its final estimate of 50.5-49.5 to the Coalition. This was accurate, but other polls taken at that time did not show a 1.5 point gain for the Coalition.

Ipsos overestimated the Greens, both relative to the election and other pollsters. This was not just a problem with Ipsos’ final poll; its polls have always had the Greens higher than other polls. This clear bias towards the Greens in Ipsos’ polls is a problem for Ipsos.

There were fewer polls during the campaignn than I would have expected. Other than Ipsos, only Essential, ReachTEL and Newspoll polled regularly. Newspoll was best at tracking the vote during the campaign, though its early campaign polls had Labor a little too high. Essential had Labor ahead until the final poll, and ReachTEL had some results that were too friendly to the Coalition.

Ipsos is currently the only live phone pollster, and this used to be the gold standard of opinion polling. However, at this election, ReachTEL (robopolling) and Newspoll (online panel and robopolling) were clearly superior to Ipsos. Methods other than live phone polling are improving.

Seat polls were very hit and miss, with some big misses. The Poll Bludger has graphs showing that seat polls had a large range of results vs actual outcomes, and were slightly biased towards the Coalition. An example of poor seat polls is Macarthur, which Labor won 58-42, but two seat polls had Labor only ahead 51-49 and a third was tied.