Autopsy turvy

Amid a generally predictable set of recriminations and recommendations, some points of genuine psephological interest emerge from Labor’s election post-mortem.

The public release of Craig Emerson and Jay Weatherill’s report into Labor’s federal election campaign has inspired a run of commentary about the way ahead for the party after its third successive defeat, to which nothing need be added here. From the perspective of this website, the following details are of specific interest:

• Labor’s own efforts to use area-based regression modelling to identify demographic characteristics associated with swings against Labor identifies five problem areas: voters aged 25-34 in outer urban or regional areas; Christians; coal mining communities; Chinese Australians; and the state of Queensland. The variable that best explained swings in favour of Labor was higher education. However, as has been discussed here previously, this sort of analysis is prey to the ecological fallacy. On this basis, I am particularly dubious about the report’s suggestion that Labor did not lose votes from beneficiaries of franking credits and negative gearing, based on the fact that affluent areas swung to Labor. There is perhaps more to the corresponding assertion that the Liberals were able to persuade low-income non-beneficiaries that Labor’s policies would “crash the economy and risk their jobs”.

• Among Labor’s campaign research tools was a multi-level regression and post-stratification analysis, such as YouGov used with notable success to predict seat outcomes at the 2017 election in the UK. Presumably the results were less spectacular on this occasion, as the report says it is “arguable that this simply added another data point to a messy picture”. The tracking polling conducted for Labor by YouGov showed a favourable swing of between 0.5% and 1.5% for most of the campaign, and finally proved about three points off the mark. YouGov suggested to Labor the problem may have been in its use of respondents’ reported vote at the 2016 election as a weighting factor, but the error was in line with that of the published polling, which to the best of my knowledge isn’t typically weighted for past vote in Australia.

• An analysis of Clive Palmer’s advertising found that 40% was expressly anti-Labor in the hectic final week, compared with only 10% in the earlier part of the campaign. The report notes that the Palmer onslaught caused Labor’s “share of voice” out of the sum of all campaign advertising fell from around 40% in 2016 to 25%, and fell as low as 10% in “some regional markets such as Townsville and Rockhampton”, which respectively delivered disastrous results for Labor in the seats of Herbert and Capricornia.

• It is noted that the gap between Labor’s House and Senate votes, which has progressively swollen from 1% to 4.6% since 1990, is most pronounced in areas where Labor is particularly strong.

Other news:

• The challenge against the election results in Chisholm and Kooyong has been heard in the Federal Court this week. The highlight of proceedings has been an admission from Simon Frost, acting director of the Liberal Party in Victoria at the time of the election, that the polling booth advertising at the centre of the dispute was “intended to convey the impression” that they were Australian Electoral Commission signage. The Australian Electoral Commission has weighed in against the challenge with surprising vehemence, telling the court that voters clearly understood that anything importuning for a particular party would not be its own work.

• The ABC reports there is a move in the Tasmanian Liberal Party to drop Eric Abetz from his accustomed position at the top of the Senate ticket at the next election to make way for rising youngester Jonathan Duniam. The Liberals won four seats at the 2016 double dissolution, which initially resulted in six-year terms being granted to Eric Abetz and Stephen Parry, and three-year terms to Duniam and David Bushby. However, the recount that followed the dual disqualifications of Jacqui Lambie and Stephen Parry in November 2017 resulted in the party gaining three rather than two six-year terms, leaving one each for Abetz, Duniam and Bushby. Bushby resigned in January and was replaced by his sister, Wendy Askew, who appears likely only to secure third place on the ticket, which has not been a winning proposition for the Liberals at a half-Senate election since 2004.

Andrew Clennell of The Australian ($) reports that Jim Molan is likely to win a Liberal preselection vote on Saturday to fill Arthur Sinodinos’s New South Wales Senate vacancy. The decisive factor would appear to be support from Scott Morrison and centre right faction powerbroker Alex Hawke, overcoming lingering hostility towards Molan over his campaign to win re-election by exhorting Liberal supporters to vote for him below the line, in defiance of a party ticket that had placed him in the unwinnable fourth position. He is nonetheless facing determined opposition from Richard Shields, Woollahra deputy mayor and Insurance Council of Australia executive, who was runner-up to Dave Sharma in the party’s hotly contested preselection for Wentworth last year.

Call of the board: Western Australia

Another deep dive into the result of the May federal election – this time focusing on Western Australia, which disappointed Labor yet again.

The Call of the Board wheel now turns to Western Australia, after previous instalments that probed into the federal election results for Sydney (here and here), regional New South Wales, Melbourne, regional Victoria, south-east Queensland and regional Queensland.

Western Australia has been disappointing federal Labor ever since Kim Beazley elevated the party’s vote in his home state in 1998 and 2001, and this time was no exception. After an unprecedented Labor landslide at the 2017 state election and expectations the state’s economic malaise would sour voters on the government, the May election in fact produced a statewide two-party swing of 0.9% to the Coalition, and no change on the existing configuration of 11 seats for the Liberals and five for Labor.

As illustrated by the maps below (click on the images to enlarge), which record the two-party swings at booth level, Perth typified the national trend in that Labor gained in inner urban areas, regardless of their political complexion, while copping a hit in the outer suburbs. This will be reflected in the seat-by-seat commentary below, which regularly invokes the shorthand of “inner urban” and “outer urban” effects. The map on the left is limited to seats that are clearly within the Perth metropolitan area, while the second adds the fringe seats of Pearce (north), Hasluck (east) and Canning (south).

For further illustration, the table below compares each electorate’s two-party result (the numbers shown are Labor’s) with a corresponding two-party Senate measure, which was derived from the AEC’s files recording the preference order of each ballot paper (with votes that did not preference either Labor or Liberal excluded). This potentially offers a pointer as to how much candidate factors affected the lower house results.

Continue reading “Call of the board: Western Australia”

Call of the board: regional Queensland

A deep dive into the darkest corner of Labor’s federal election failure.

Welcome to the latest instalment of Call of the Board, which probes into every seat result from the May federal election region by region. Earlier instalments covered Sydney, here and here; regional New South Wales; Melbourne; regional Victoria and south-east Queensland. Today we look at the electorates of Queensland outside of Brisbane, the Gold Coast and the Sunshine Coast.

The posts dealing with the big cities have featured colour-coded seat maps and the results of a model estimating how the results would have looked if determined by demographic factors alone. Unfortunately, colour-coding doesn’t get you very far when zooming out to vast and unevenly populated regional terrain, and the model hasn’t proved to be much use in producing plausible results for regional seats, in which elusive factors of local political culture appear to loom large. However, I can at least offer for purposes of comparison Labor two-party estimates derived from the Senate results, potentially offering a pointer to how much candidate factors affected the lower house results.

Seat by seat alphabetically:

Capricornia (LNP 12.4%; 11.7% swing to LNP): Labor held this Rockhampton region seat for all but one term from 1977 to 2013, but history may record that it has now reached a tipping point akin to those that have excluded the party from former regional strongholds including Kennedy (Labor-held for all but two terms from federation to 1966, but only once thereafter), Grey in South Australia (Labor-held for all but one term from 1943 to 1993, but never again since) and Kalgoorlie in Western Australia (Labor-held for all but three terms from 1922 until Graeme Campbell quit the party in 1995, and now divided between the safely conservative seats of O’Connor and Durack). The 11.7% swing to Michelle Landry, who has held the seat since 2013, was the biggest in the country, shading the 11.2% swing to the beloved George Christensen in Dawson. Landry’s primary vote was actually little changed, reflecting the entry of One Nation, who accounted for most of Labor’s 14.3% collapse. The rest came from a halving of the Katter’s Australian Party vote from 7.1% to 3.7% and the absence of Family First.

Dawson (LNP 14.6%; 11.2% swing to LNP): Dawson behaved almost identically in swing terms to its southern neighbour, Capricornia, as voters showed themselves to be a great deal more concerned about Adani and its symbolism than George Christensen’s enthusiasm for life in the Philippines. As in Capricornia, the LNP primary vote was little changed from 2016, but the arrival of One Nation soaked up 13.1% which neatly matched Labor’s 12.5% decline. Katter’s Australian Party held up better here than in Capricornia, their 6.3% being only slightly down on 2016.

Flynn (LNP 8.7%; 7.6% swing to LNP): Labor narrowly won this Gladstone-based seat on its creation at their 2007 high-water mark and sliced the margin back to 1.0% in 2016, but hopes of going one better this time fell foul of the party’s region-wide disaster. The swing in this case was fairly typical of those suffered by Labor outside the immediate range of proposed Adani mine, though in this case One Nation were not a new feature, their 19.6% being slightly higher than their 2016 result. The seat was a bit unusual in that Labor’s score on the two-party Senate estimate was 2.8% stronger than their House result.

Groom (LNP 20.5%; 5.2% swing to LNP): The 5.2% swing to John McVeigh was a bit below the regional Queensland par, despite him being a sophomore of sorts – although he may have arrived in 2016 with a ready-made personal vote due to his background as a state member. Nonetheless, it was sufficient to catapult the seat from fifteenth to second on the national ranking of seats by Coalition-versus-Labor margin, reflecting the narrowing of margins in many blue-ribbon city seats. The 2016 result was remarkable in that Family First polled 10.0% in the absence of right-wing minor party competition – this time the newly arrived One Nation polled 13.1% in their absence. The LNP primary vote was little changed and Labor was down 3.5%, the rest of the swing bespeaking a more right-wing minor party preference pool.

Herbert (LNP GAIN 8.4%; 8.4% swing to LNP): Labor’s most marginal seat pre-election, following Labor member Cathy O’Toole’s 37 vote win in 2016, the Townsville seat of Herbert was one of five seats across the country and two in Queensland that were gained by the Coalition (balanced to an extent by Labor’s gains in Gilmore and, with help from redistribution, Corangamite and Dunkley). While the swing was lower than in the Adani epicentre electorates of Dawson and Capricornia immediately to the south, it was sufficient to produce the most decisive result the seat has seen since 1954. O’Toole’s primary vote was down 5.0% to 25.5%, while LNP victor Phillip Thompson added 1.6% to the party’s 2016 result to score 37.1%. High-profile Palmer candidate Greg Dowling did relatively well in polling 5.7%, and One Nation were down from 13.5% to 11.1%.

Hinkler (LNP 14.5%; 6.1% swing to LNP): Keith Pitt, who has held this Bundaberg-based seat since 2013, picked up a swing well in line with the regional Queensland norm. He was up 2.2% on the primary vote, while Labor was down 3.8%; One Nation fell from 19.2% to 14.8%, mostly due to an expansion in the field from seven candidates to ten, including three independents, none of whom did particularly well individually.

Kennedy (KAP 13.3% versus LNP; 2.3% swing to KAP): Bob Katter had a near death experience at the 2013 election, at which time he was presumably tarred with the minority government brush despite being the only cross-bencher who backed the Coalition after the inconclusive 2010 result. However, he’s roared back to dominance since, picking up successive two-party swings of 8.9% and 2.3%, and primary vote swings of 10.5% and 2.6%. On the latter count at least, he’s been assisted by the fact that One Nation have declined to challenge him. In Coalition-versus-Labor terms, the seat participated in the regional Queensland trend in swinging 7.8% against Labor.

Leichhardt (LNP 4.2%; 0.2% swing to LNP): The negligible swing in favour of LNP veteran Warren Entsch was an exception to the regional Queensland rule, and was generally attributed to the centrality of tourism to the economy of Cairns, giving the region a very different outlook on issues like Adani. The result was generally status quo in all respects, but the seat had the distinction of being one of only three in the state where the Labor primary vote very slightly increased, along with Ryan and Fairfax. With Entsch’s primary vote down slightly, the two-party swing, such as it was, came down to an improved flow of preferences.

Maranoa (LNP 22.5% versus One Nation; 6.6% swing to LNP): For the second election in a row, Maranoa emerged with the distinction of being the only seat in the country where One Nation made the final preference count. One Nation and Labor were down on the primary vote by 3.2% and 2.7% respectively; at the last preference exclusion, One Nation led Labor 21.3% to 19.0%, compared with 23.6% to 22.9% in 2016. The other story here was the strong sophomore showing for David Littleproud, who was up 6.8% on the primary vote and by similar amounts on two-party preferred against both One Nation and Labor. The 25.4% margin versus Labor is now by some distance the biggest in the country, compared with the electorate’s ninth ranking on this score in 2016. Equally impressive for Littleproud is the distinction between his 25.4% margin and the 20.4% recorded by the two-party Senate measure.

Wide Bay (LNP 13.1%; 5.0% swing to LNP): Llew O’Brien may also have enjoyed a sophomore effect after succeeding Warren Truss in 2016, as his primary vote was up 3.2% while One Nation fell from 15.6% to 10.8%. However, the Labor primary vote held up unusually well, and the two-party swing was at the lower end of the regional Queensland scale.

Wright (LNP 14.6%; 5.0% swing to LNP): So far as the major parties were concerned, the result here was typical of regional Queensland, with LNP member Scott Buchholz up 3.1% on the primary vote and Labor down 4.0%. Independent Innes Larkin, who appears to have made his name locally campaigning against coal seam gas, scored a respectable 5.3%, which presumably helps explains the drop in the One Nation vote from 21.8% to 14.0%.

Sins of commission

Kooyong and Chisholm legal challenge latest; by-election rumblings in Isaacs; Jim Molan strikes back; and the Victorian Liberals gearing up already for federal preselections.

Possible (or possibly not) federal by-election news:

• The Australian Electoral Commission has petitioned the Federal Court to reject challenges against the federal election results in Chisholm and Kooyong. The challenges relate to Chinese-language Liberal Party signage that appeared to mimic the AEC’s branding, and advised voters that giving a first preference to the Liberal candidates was “the correct voting method”. As reported by The Guardian, the AEC argues that “the petition fails to set out at all, let alone with sufficient particularity, any facts or matters on the basis of which it might be concluded that it was likely that on polling day, electors able to read Chinese characters, upon seeing and reading the corflute, cast their vote in a manner different from what they had previously intended”. This seems rather puzzling to my mind, unless it should be taken to mean that no individuals have been identified who are ready to confirm that they were indeed so deceived. Academic electoral law expert Graeme Orr argued on Twitter that the AEC had “no need to intervene on the substance of a case where partisan litigants are well represented”.

• Talk of a by-election elsewhere in Melbourne was stimulated by Monday’s column ($) from acerbic Financial Review columnist Joe Aston, which related “positively feverish speculation” that Labor’s Shadow Attorney-General, Mark Dreyfus, would shortly quit his Melbourne bayside seat of Isaacs with an eye to a position on Victoria’s Court of Appeal. Aston further reported that Dreyfus hoped to be succeeded by Fiona McLeod, the prominent barrister who gained a 6.1% swing as Labor’s candidate for Higgins in May. Dreyfus emphatically rejected such “ridiculous suggestions” in late August, saying he was “absolutely committed to serving out this term of parliament”, and again took to Twitter on Monday to say he would be “staying and fighting the next election”. Aston remains unconvinced, writing in Tuesday’s column ($) that the suggestions derived from “high-level discussions Dreyfus has held on Spring Street with everyone from Premier Daniel Andrews, former Attorney-General Martin Pakula, his successor Jill Hennessy and his caucus colleagues”, along with his “indiscreet utterances around the traps”.

Federal preselection news:

• Jim Molan has won the endorsement of both Scott Morrison and the conservative faction of the New South Wales Liberal Party to fill the Senate vacancy created by Arthur Sinodinos’s departure to become ambassador to the United States. However, the Sydney Morning Herald reports this is not dissuading rival nominee Richard Shields, former deputy state party director and Insurance Council of Australia manager, and the runner-up to Dave Sharma in last year’s keenly fought Wentworth preselection. Shields’ backers are said to include Helen Coonan, former Senator and Howard government minister, and Mark Neeham, a former state party director. Earlier reports suggested the moderate faction had been reconciled to Molan’s ascendancy by a pledge that he would only serve out the remainder of Sinodinos’s two-year term, and would not seek re-election in 2022.

Rob Harris of The Age reports the Victorian Liberals are considering a plan to complete their preselections for the 2022 election much earlier than usual – and especially soon for Liberal-held seats. The idea in the latter case is for challengers to incumbents to declare their hands by January 15, with the matter to be wrapped up by late February or early March. This comes after the party’s administrative committee warded off threats to members ahead of the last election, most notably factional conservative Kevin Andrews in Menzies, by rubber-stamping the preselections of all incumbents, much to the displeasure of party members. Other preselections are to be held from April through to October. Also proposed is a toughening of candidate vetting procedures, after no fewer than seven candidates in Labor-held seats were disendorsed during the period of the campaign.

Self-promotion corner:

• I had a paywalled piece in Crikey yesterday which noted the stances adopted of late by James McGrath, ideological warror extraordinaire and scourge of the cockatoo, in his capacity as chair of the Joint Standing Committee on Electoral Matters, which is presently conducting its broad-ranging inquiry into the May federal election. These include the end of proportional representation in the Senate, the notion that parliamentarians who quit their parties should be required to forfeit their seats, and — more plausibly — the need to curtail pre-poll voting.

Call of the board: South-East Queensland

How good was Queensland? The Poll Bludger reports – you decide.

The Poll Bludger’s popular Call of the Board series, in which results for each individual electorate at the May 18 federal election are being broken down region by region, underwent a bit of a hiatus over the past month or so after a laptop theft deprived me of my collection of geospatial files. However, it now returns in fine style by reviewing the business end of the state which, once again, proved to be the crucible of the entire election. Earlier instalments covered Sydney, here and here; regional New South Wales; Melbourne; and regional Victoria.

First up, the colour-coded maps below show the pattern of the two-party swing by allocating to each polling booth a geographic catchment area through a method that was described here (click for enlarged images). The first focuses on metropolitan Brisbane, while the second zooms out to further include the seats of the Gold Coast and Sunshine Coast. As was the case in Sydney and Melbourne, these maps show a clear pattern in which Labor had its best results (in swing terms) in wealthy inner urban areas (for which I will henceforth use the shorthand of the “inner urban effect”, occasionally contrasted with an “outer urban effect” that went the other way). However, they are also bluer overall, reflecting Labor’s generally poor show across Queensland (albeit not as poor in the south-east as in central Queensland).

The seat-by-seat analysis is guided by comparison of the actual results with those estimated by two alternative metrics, which are laid out in the table below (using the two-party measure for Labor). The first of these, which I employ here for the first time, is a two-party estimate based on Senate rather than House of Representatives results. This is achieved using party vote totals for the Senate and allocating Greens, One Nation and “others” preferences using the flows recorded for the House. These results are of particular value in identifying the extent to which results reflected the popularity or otherwise of the sitting member.

The other metric consists of estimates derived from a linear regression model, in which relationships were measured between booths results and a range of demographic and geographic variables. This allows for observation of the extent to which results differed from what might have been expected of a given electorate based on its demography. Such a model was previously employed in the previous Call of the Board posts for Sydney and Melbourne. However, it may be less robust on this occasion as its estimates consistently landed on the high side for Labor. I have dealt with this by applying an across-the-board adjustment to bring the overall average in line with the actual results. Results for the Gold Coast and Sunshine Coast seats are not shown, owing to the difficulty involved in classifying them as metropolitan or regional (and I have found the model to be of limited value in regional electorates). The coefficients underlying the model can be viewed here.

And now to review each seat in turn:

Continue reading “Call of the board: South-East Queensland”

ANU post-election survey and Essential Research poll

Comprehensive new research suggests a telling shift from the “others” column to the Coalition through the campaign period, while Labor were either consistently overrated by pollsters or fell off a cliff at the end.

Some particularly interesting post-election research has emerged in the shape of a paper from Nicholas Biddle at the Australian National University’s Centre for Social Research and Methods. This draws from the centre’s regular online panel surveys on social attitudes, which encompasses a question on voting intention for reasons unrelated to prediction of election results. The study compares results for 1692 respondents who completed both its pre- and post-election surveys, which were respectively conducted from April 8 to 26 (encompassing the start of the campaign on April 11) and June 3 to 17 (commencing a fortnight after the election). Respondents were excluded altogether if they were either ineligible to vote or failed to answer the voting intention question.

The results are, to a point, consistent with the possibility that pollsters were confounded by a last minute shift to the Coalition, particularly among those who had earlier been in the “others” column. The changes can be summarised as follows, keeping in mind that a “don’t know” response for the April survey was at 2.9%, and 6.5% in the June survey said they did not vote. Since the disparity leaves a net 3.6% of the total vote unaccounted for, the shifts identified below will err on the low side.

The Coalition vote increased an estimated 2.6% from the time of the April survey, suggesting the polls were right to be recording them at around 38% at that time, if not later. However, no movement at all was recorded in the Labor vote, suggesting they were always about four points short of the 37% most polls were crediting them with. The exception here was Ipsos, which had Labor at 33% or 34% in all four of the polls from the start of the year. The Greens fell very slightly, suggesting a poll rounding to whole numbers should have had them at 11% early in the campaign. Newspoll consistently had it at 9%, Ipsos at 13% or 14%, and Essential fluctuated between 9% and 12%.

The biggest move was the 5.9% drop in support for “others”, although a fair bit of this wound up in the “did not vote” column. Even so, it can conservatively be said that pollsters in April should have been rating “others” at around four points higher than their actual election result of 15%, when they were actually coming in only one point higher. This three point gap is reflected in the size of the overestimation of support for Labor.

The results also point to a remarkably high degree of churn — an estimated 28.5% did not stick with the voting intention expressed in April, albeit that a little more than a fifth of this subset did so by not voting at all. The sub-sample of vote changers is small, but it offers little to suggest voters shifted from Labor to the Coalition in particularly large numbers. The Coalition recorded the lowest rate of defection, although the difference with Labor was not statistically significant (I presume it’s normal for major party supporters to be more constant than minor). Conversely, 49.4% of those who left the “others” column went to the Coalition (which comes with a 9% margin of error), and most of the remainder did not vote.

The survey also features statistical analysis to determine the demographic characteristics of vote changers. These find that older voters were generally less likely to be vote changers, and that young vote changers tended not to do so in favour of the Coalition, presumably switching for the most part between Labor and the Greens. Also particularly unlikely to budge were Coalition voters who lived in areas of socio-economic advantage. Those at the other end of this scale, regardless of party support, were most volatile.

Also out this week was the regular fortnightly Essential Research survey, which is still yet to resume its voting intention series but will do so soon. A question on the anticipated impact of government policies over the next three years produces encouraging numbers for the government, with 41% positive and 23% negative. A question on racist sentiments finds 36% agreeing that Australia is a racist country, and 50% saying it is less racist than it was in the past. Breakdowns record no significant differences between those of migrant and non-migrant backgrounds, although the former may include too many of British origin for the results to be particularly revealing.

A question on political interest finds only 15% professing no interest in federal politics, with 53% saying they follow it closely or “enough to know what’s happening”. A big question though is whether polling has gone astray because too many such people are included in their samples. The poll was conducted Thursday to Sunday from a sample of 1075 respondents drawn from an online panel.