Source: The Conversation (Au and NZ) – By Adrian Beaumont, Honorary Associate, School of Mathematics and Statistics, University of Melbourne
Since Scott Morrison became PM in August 2018, Newspoll has usually been the best pollster for Labor. This was particularly so in late 2018, but these strong polls for Labor should now be questioned. In November 2018, Labor had two 52-48 lead results from Essential and Ipsos, but two Newspoll results the week before and the week after those polls gave Labor a 55-45 lead.
In late 2018, Essential usually gave Labor a lower two party vote than Newspoll, even though Essential’s preferencing method was better for Labor than Newspoll. Essential used last-election preferences, while Newspoll adjusted One Nation’s preferences to 60-40 to the Coalition – from the results, this adjustment was justified.
At the time, poll analysts assumed that Newspoll, not Essential, was correct, but the election results suggest Essential was more accurate in consistently having Labor’s primary vote lower than Newspoll had it.Analyst Kevin Bonham said there is no evidence of a late swing; pre-poll booths had a greater swing to the Coalition than Election Day booths (2% vs 0.8%). That suggests that the polling has been wrong for a long time, and that Newspoll was flawed. During the campaign, other pollsters herded their results towards where Newspoll had it, but this was wrong.
The four active pollsters at this election were YouGov Galaxy, which conducts Newspoll, Ipsos, Essential and Morgan. Galaxy uses online methods and robopolling, Essential uses online methods, Ipsos uses live phone polling and Morgan uses face-to-face interviews. No pollster does only landline polling – Ipsos calls mobiles.
Australian pollsters have inadequate documentation of their methods. For example, we do not know what portion of Galaxy’s surveys are robocalls, and what portion use online methods. This lack of documentation should change after this poll failure.
Polls have not been accurate in recent state elections
Labor led in the final federal election polls by about 51.5-48.5, but lost the election by about the same margin. We won’t know final figures for at least a few weeks, but this miss was about three points. At recent elections, polls have not been accurate, also missing badly at the November 2018 Victorian election. Polls appear to have have become less accurate since the 2016 federal election.
This is the table of polls vs the election outcome for the July 2016 federal election. As with the 2019 election, polls appeared to “herd” too close together, but in 2016 they herded to the correct result. Bold numbers for poll estimates in the tables below denote cases where the poll was within 1% of the actual result.
At the March 2017 WA election, polls were too high on One Nation’s vote, caused by One Nation not contesting all WA seats. The Greens and Labor did a bit better than expected.
At the November 2017 Queensland election, polls were good – unlike Queensland state breakdowns or polls at the federal election, which suggested 51-49 to the LNP or a 50-50 tie. The LNP currently leads in Queensland by 57.5-42.5.
At the March 2018 Tasmanian election, two polls, taken about a week before the election, understated the Liberal vote. But Tasmania has a bandwagon effect where people opt for the major party that can govern without needing the Greens.
At the March 2018 SA election, both major parties, particularly the Liberals, had a greater primary vote than the polls estimated, and SA-Best did worse. SA-Best support had been falling.
At the November 2018 Victorian election, Labor led in the final polls by about 53.5-46.5, and won by 57.6-42.4. Few people care when the party expected to win wins by a bigger than expected margin, so polls are not criticised for these mistakes as much as they should be. When the expected winner loses, polls are heavily criticised.
Only one poll at the March 2019 NSW election could be thought of as a final poll, with Newspoll at 51-49 to the Coalition (actual result 52.0-48.0). Movement to the Coalition was explained by the revelation of a video of Labor leader Michael Daley that could be construed as anti-Asian.
Late counting updates
Much counting of Liberal-friendly postal votes in Chisholm has confirmed it will be retained by the Liberals. That gives the Coalition an overall majority with 76 of the 151 seats. The Coalition is also likely to gain Bass, while Macquarie is still uncertain, but with the Liberals currently ahead.
The Coalition has improved its position in the Victorian and Queensland Senate races. With 58% counted in Victoria, the Coalition has 2.47 quotas, Labor 2.24 and the Greens 0.72. With 60% counted in Queensland, the LNP has 2.70 quotas, Labor 1.62, the Greens 0.72 and One Nation 0.70.
If these results hold up, the Coalition is well-placed to win the final Victorian Senate seat. In Queensland, whoever is last after preferences from LNP, Labor, Greens and One Nation misses out, and that looks likely to be Labor. That would mean Queensland would split 4-2 to the right in the Senate.
If current counting in Victoria and Queensland holds, there will be 38 right-wing senators out of 76 (35 Coalition, two One Nation and Cory Bernardi), 35 left-wing senators (26 Labor and nine Greens), two Centre Alliance and Jacqui Lambie. The Coalition’s easiest path to passing legislation opposed by the left would be with other right-wing senators, plus either Lambie or Centre Alliance.
AEC’s two party count does not yet include all electorates
The Electoral Commission has a two party preferred count on the home page of its results, currently showing the Coalition ahead by 51.3-48.7. However, this two party count only includes seats where the Coalition and Labor are expected to be the final two candidates. There are currently 15 seats out of 151 with “non-classic” contests, where the final two candidates were not from the Coalition and Labor.
These non-classic contests are added to the two party count via a special count between the Coalition and Labor candidates, but this will not happen until the seat count has been nearly finalised. Ten of these seats are very likely to favour the Coalition when added to the count, and only five will favour Labor. That means the current two party count is biased to Labor.
In addition, there are three seats – Calare, Grey and Barker – where the Electoral Commission thought Labor would not make the final two. Labor made the final two in those seats, so they are slowly being recounted between the Coalition and Labor. These conservative seats still have plenty of votes that haven’t been added to the two party count.
With these distortions factored in, the two party count is probably close to 52-48 to the Coalition, though Labor should improve as absent votes, which favour Labor, have yet to be included.
UK’s European Union elections are today
I wrote for The Poll Bludger about the UK’s European Union elections, which will be held today. No results will be released until all EU countries finish voting early on May 27 Australian Eastern Standard Time. Nigel Farage’s Brexit party is expected to win the UK’s EU elections, with the Conservatives crashing to their lowest ever national vote share. There has been much recent speculation that Theresa May will resign soon, so Boris Johnson could be the next PM.
– ref. Newspoll probably wrong since Morrison became PM; polling has been less accurate at recent elections – http://theconversation.com/newspoll-probably-wrong-since-morrison-became-pm-polling-has-been-less-accurate-at-recent-elections-117400