Wednesday, March 12, 2014

Poll Roundup: National Polling All Over The Place

Poll Aggregate: 51.2% to Labor (update: 50.4% on 18 March)
Labor maintains narrow lead held since December

Nielsen 52-48 to Coalition? Newspoll 54-46 to Labor?  These two results just one week apart - what could it mean?

It's about time to have another of my irregularly scheduled roundups of the national polling picture and how the not quite so new Abbott government is travelling.  My last federal roundup of sorts was a month back in the Griffith article.  At the time Labor seemed to be starting to drop off a little, and the lukewarm by-election result seemed consistent with that.

Shortly afterwards we had the 52-48 to Coalition from Nielsen.  At the time some dismissed it as a rogue, but it also seemed it could be a harbinger of a return to a Coalition lead.  There was some support for this from a Morgan which gave Labor a wafer-thin 50.5:49.5 lead by last-election preferences (equivalent to a thin Coalition lead given the apparent house effect of Morgan's methods).  Also, two weeks of 51:49 to Labor from Essential proved little, since we know Essential can be contrary to, or lag, the trend of other polling.

This was all enough to put the Coalition into the lead on my aggregate but it lasted only a week before the 54:46 Newspoll put an end to that.  As Adrian Beaumont notes, there is no way there was a six point swing in a week!  Later the same week ReachTEL came in at 53 to Labor (I make it 52.7 prior to rounding, as ReachTEL publish decimals for primaries.) Again Essential went the other way, going to 51 to Coalition before coming back to 50 this week.  This week's other new polls have been a 51 to Labor from Newspoll, and a Morgan reported as 53.5 by respondent preferences, for which my estimate of last-election preferences is 52.5 to Labor.

It's notable that the last few ReachTELs have been about as strong for the ALP as Morgan and this is something that a few psephs are keeping an eye on.

In all, the Nielsen result with the Coalition ahead was a result seen in none of the last four Newspolls, three ReachTELs and seven Morgans since the start of December.  However one of the Morgans showed a Coalition lead and one a tie once they were adjusted by a point for house effect.  Essential did show six 51s for the Coalition to two 51s for Labor and two ties, but we know from the 2013 campaign that Essential is a beast apart (and in that case not a very effective one). 

This is what my basic 2PP-vote tracking graph looks like for the 23 weeks since the aggregate was turned on in October:

The gap in the middle is because Aggedor the aggregator (named after a Dr Who monster) goes to sleep unless it has at least three recent polls in its belly, but the tracking in those weeks would have been about stable anyway.  So basically since week 11 (the second week in December) the aggregate has showed a stable ALP lead with a 2PP of 51 to 51.5 excepting the week of the Nielsen and the week before that.  The current reading is 51.2% to Labor.

Aggedor is a rather rough and spiky beast when provoked by unusually bouncy polling, and if I smooth down its fur with a very simple rolling average method I get something that looks like this:

On that basis the one-week Coalition lead wasn't real but was caused by the single outlying Nielsen, which carried a heavy weighting in the aggregate.  Most likely there was a small dip towards 50:50 around this time (both the previous week's data points and the Morgan in the same week support this, and Bludgertrack has had something similar albeit off slightly higher 2PP estimates for Labor) but it was really just a brief wobble in Labor's modest lead.  Even the smoothing method used in this rolling average exaggerates the impact of the Nielsen by virtue of it coming before the Newspoll.

With the Newspoll three points off the aggregate that included it, and the Nielsen over two points below the rolling average, it's safe to say these two polls were inaccurate samples and should be deemed to more or less cancel each other out against the backdrop of a slim Labor lead.  The Newspoll moves from 51 to 54 and back to 51 against were most likely nothing but random bouncing but this did not stop Dennis Shanahan relating the week's events as if they could have caused Labor to "lose" support it never actually had in the first place.  Both the Nielsen and the 54:46 Newspoll were on the edges of statistical respectability (calling them rogue without knowing the rounding is a little bit harsh, but they weren't far from that status at best).  So far as the national 2PP picture goes the best we can do is carry on as if they never occurred.

Nielsen Indie Nuisance

I add a little gripe re Nielsen.  The two polls released by Nielsen since the election have both, probably largely by bad luck, been serious outliers in opposite directions.  But there is one thing they are doing that I do not like at all, and that is their inclusion of "independent" as an option.  It's blatantly obvious that this indicator (running at 6% in the most recent poll) is polling way too high given that only 1.4% voted independent at the election.  It's being chosen as an ambit claim by people who at best might vote for an indie if they can find one they like (but probably won't.) From time to time those picking the "independent" option might lean to a specific side of politics, either systematically or intermittently, and assuming their preferences flow similarly to those of independent voters at the 2013 election could produce very misleading results.

One concern I have here is that Independent voters preferenced Labor 57:43 last election, but if the strongly ALP-leaning preferences of the successful independents Andrew Wilkie and Cathy McGowan are excluded, that drops to about 50:50, and the Wilkie and McGowan preferences had no bearing on the result.  If those thinking of voting independent (but who probably won't actually do so) are likewise split down the middle but are treated as similar to the greater pool of indie voters (including the Wilkie and McGowan supporters) then that alone could introduce a Labor bias of as much as 0.3 points into the 2PP.  If the "indies" come mainly from the Coalition side for some reason, it will be more. Not something we'll pick up from a couple of polls, but worth keeping an eye on over a full term. 

(Peter Brent here also covers indie voting patterns more broadly and suggests some potential for preference allocations to overpredict Labor's preference success at the next election.  At this stage my own concern is confined to the Independent option in Nielsen.)

So Labor's Ahead: Should They Be Happy?

Well, it's certainly better to be up 51:49 than down 41:59, but beyond that I'd say Labor has much to do before they have all that much to celebrate.  For starters a 51:49-ish lead for a first-term Opposition that was thumped last time is probably not even election-winning if produced on the day, thanks to the new personal votes of all the first-term Opposition MHRs, and given that a government with a large majority has a lot of tactical flexibility.  An election held right now could well be 1998 all over again - a little closer maybe, but not much. In that election Labor won the two-party vote but got nothing like the pendulum prediction in seat terms.

The second problem for Labor is the usual one of government resilience, especially on the Coalition side.  Governments that are on their way to going down in a screaming heap will usually start losing big quite early in their term.  The historically quick loss of the lead by the Coalition (see Abbott Fastest Ever To Lose Poll Lead) was a positive sign for the new Opposition, but it hasn't yet been followed by a blowout into regular 54s and 55s.  Perhaps a nasty Budget will fix that, but we are not yet seeing the rapid development of overwhelming "buyer's remorse" that tends to characterise doomed governments.  I'd say therefore that these slim leads for Labor add very little indeed to their chances of pulling off an unlikely first-term victory.

Neither Leader Popular

Leadership ratings show that both Tony Abbott and Bill Shorten aren't that popular.  Newspoll gave Abbott a -16 netsat in the previous sample, but in light of the 54:46 2PP result, that figure can't be trusted.  In the current Newspoll, however, Abbott is still on -12, with a 50% dissatisfaction rate already.  Of previous PMs since Gorton, only Keating (four months in) was recording such high dissatisfactions faster (excluding Fraser's tenure as caretaker in 1975); Gillard took eight months and all others at least a year.

A -12 result in a 49:51 poll was actually slightly good news for the PM, however.  It weakened the relationship between the 2PP and his personal ratings (now explaining a mere 91% of variation!) and provided more evidence that the Coalition can poll competitively even with Abbott polling indifferent personal ratings.  His break-even point (the netsat at which his party is expected to get a 50% two-party preferred vote) is now at -3. (See Early Abbott Era Polling Roundup for scores for earlier PMs). Given Abbott's history of bad ratings the Coalition would want it to drop further over time.

Essential Report (-6 in February and -7 this month) and Nielsen (-2) also gave Abbott a mild thumbs down, while ReachTEL's figures come out as -9 on my ReachTEL conversion method (treating half the Satisfactorys as positive and the other half as neutral.)

Preferred/Better Prime Minister ratings have been giving Abbott the sort of  lukewarm leads to be expected from a not especially liked PM whose party is a little bit behind.  Newspoll had a lead of 1 point in the suspect sample followed by 6 points in the more believable one, Nielsen had Abbott up by 10 and Essential currently has an Abbott lead of 6.  ReachTEL was also the first to venture back into the waters of a Coalition-only preferred-PM poll, predictably showing Malcolm Turnbull (53.2%) still streets ahead of Abbott (27.7%) and Hockey (19.1%).  However, Abbott is well ahead of the others among Coalition voters, with 58.6%.  The unpublished breakdown among non-major-party voters appears to be about Turnbull 59%, Hockey 32%, Abbott 9%.

It was an eye-opener that Bill Shorten was on the skids at -4 in the last Newspoll despite its ALP lean.  His latest netsat is -7. Among opposition leaders only Peacock (second term - immediate) and Howard (first term - three months) and Downer (three months) went negative faster, while Beazley (second term) and Nelson went negative at around the same time as Shorten.  This is not the list of predecessors that one wants to have.  Confirmation that Shorten was at best lukewarm in net rating terms came from Essential (-2 then -7), Nielsen (zero) and ReachTEL (also zero by the method given above). 

Here, furthermore, is a rough comparison of early Shorten and Nelson netsats by number of fortnights in the job:

A lot of data for Shorten are missing because of Newspoll's long vacation over Xmas/New Year but the point is that Shorten's ratings aren't any better for the same time in office as those of Nelson, who is regarded as having been a failure.  In truth it was mostly not Nelson's fault; he was on a hiding to nothing in the face of a very popular new government, which was recording huge 2PP leads and consequently huge preferred prime minister leads and would have done so over whoever was in office.

Another unappealling comparison between Nelson and Shorten lies in the exact way in which each dropped from initially decent ratings to poor ones.  In comparison with his very first rating four and a half months ago, Shorten has now lost eighteen netsat points.  This however arose by the positive side of his rating rising from 32 to 44 then dropping back to 33 while his dissatisfaction has climbed from 24 to 43.  So the net change has been voters moving from undecided to negative, but the worse news is that either some of those were positive along the way, or else some initially undecided voters have liked Shorten and then returned to undecided.  Nelson did a similar thing:  his first rating came out at 36-19; a couple of months later he was at 35-37.  This sort of thing isn't that unusual, but most often the Opposition Leaders who quickly regress to around their starting satisfaction rating while gaining 20 or so points of dissatisfactions are those, like Abbott and Turnbull, who came into office as well known commodities with relatively low don't-know rates (and hence less undecided voters to turn into positives.) 

All the same, it's not like Abbott and Shorten are all that unpopular yet.  Their combined netsats total -22, but in the previous term of parliament the same figure at one stage was below -50 for six months in a row.

There was also some attribute polling out from Newspoll that is well worth a look, but it seems this was taken in the same sample as the 54:46, so results that look friendly to Labor (like the recapture of the lead on the handling of unemployment) should be treated with caution.   Of more concern to Labor should be that even in a friendly sample, it seems to have lost ground on asylum seekers to "someone else", though whether this is PUP or the Greens is not made clear.   Of other notable findings, Nielsen has found that voters generally support a royal commission into union corruption (and even Labor voters want the mess cleaned up), ReachTEL has found that even Coalition voters are not that opposed to cutting defence spending to get the budget back to surplus, and Essential finds foreign aid, the arts, private schools and business subsidies to be voters' favourite targets for spending cuts.

Essential also showed voters equivocal about whether the government is handling the asylum seeker issue well, but found 25% thought it was being too soft, to 28% thinking it was too harsh.  (34% said it was taking the right approach).  Here 65% of Labor supporters thought the government was being either too harsh or too soft, but these 65% couldn't make up their minds which, splitting only 37-28 in favour of too harsh.  Essential found the government rated poorly on all issues offered bar the economy and asylum seekers.

The government now faces a significant test of its popular support in the form of the Western Australian Senate by-election, which will hopefully be the last to be held under the current group ticket voting system.


Note re Redcliffe Those with an interest in the predictive abilities of different pollsters should note that Galaxy has pretty much bullseyed the difficult Redcliffe (Queensland) by-election (which I lacked the time to cover here).  The exact final official 2PP result is not easily determined online but Galaxy's 57:43 was dead right accounting for rounding in the accounts of final-ish results I've seen (57.1 being the most commonly reported value).  The poll did overpredict Labor and the Greens by a few points of primary vote, but it seems all of this went to minor candidates who then preferenced Labor (or to the extent they didn't under optional preferencing, other preferences were stronger than expected).  The Galaxy poll's prediction of the LNP vote was spot on.  Underprediction of very minor candidates who are down the list seems to be a minor hazard with automated polling. 

A union-commissioned ReachTEL conducted two months before the poll scrubbed up rather well in the circumstances, with a 2PP error of about three points, not bad considering the time out from the poll and that when the poll was conducted the Palmer United Party were expected to contest (but didn't).  A Lonergan conducted just over a month prior however had a way higher ALP primary than eventuated.  While the pollster predicted the party gap would narrow I think it's more likely that the poll was incorrect in the first place.


Update 18 March: Both Nielsen and Essential came in 51:49 to Coalition this week, cutting Labor's lead in my aggregate to a wafer-thin .4%.  The apparent 0.8-point shift is possibly overstated because of the mix of polls most prominent in the aggregate this week compared to last, with Essential and Nielsen now the most Coalition-leaning on average (albeit slightly and in Nielsen's case based on only three polls.)    However, I'm aiming to avoid applying house-effect corrections to polls when the apparent lean of the poll is well below a point.

The Nielsen ratings for both leaders were lukewarm with Abbott a net -4 (45-49) and Shorten a net zero again (42-42). Surprisingly Abbott's preferred Prime Minister lead was cut to an unimpressive five (48-43).

20 March:  This week's Bludgertrack has come in with a Coalition lead (albeit 50.1% only) for the first time this year.  Often Bludgertrack has been more ALP-sympathetic than my aggregate this year, but not in this case. That said, in the absence of new data my aggregate will become 50:50 when the weekly reset is applied on Saturday.

No comments:

Post a Comment