Tuesday, September 9, 2014

Poll Roundup: One Year On / The Magic Of Newspoll Rounding

2PP Aggregate: 51.7 to ALP (+0.1 since last week after no change in previous week)

The Government's First Year In Polling

This week's polling marks the end of the Abbott Government's first year in power.  In summary, the Government has spent most of that year behind in the polls.  Of the seven first-term governments for which polling exists, it is the first to spend most of its first year trailing in the 2PP vote, and only the second known to have been behind at all.  However, it ends its first year in arguably better shape than the Whitlam government did.  At the end of its first year the Whitlam government trailed 43-50 on primaries (equivalent to about 46:54 2PP) in a single Morgan poll, and while there was probably a bit of bouncing in that particular poll, it had been behind for a couple of months prior to that too.  Less than six months later, Whitlam's government went to an early election and was returned.

The end of a government's first year is a convenient but arbitrary milestone.  One government that demonstrates this very well is Fraser's, which as its first year ended was in the middle of a slide in voting intention that soon after briefly put the ALP ahead.  This was accompanied by a slide in Fraser's approval ratings from the high 40s into the mid to high 30s, where they then stayed for most of 1977.  Whether you talk about the state of Fraser's polling after 11, 12 or 13 months (or even whether you measure it from his appointment as caretaker PM or the start of his government) you get quite different answers, but the end result is that at the end of '77 Fraser won re-election with his massive majority barely dented.  There's no reason why a government's standing after exactly one year, or any other time period, should be especially predictive.

It may however be more useful to look at how performance over a first year has done as a predictor of the fates of all governments for which polling exists.  I have multiple first-year poll results for all governments back to the Menzies government formed at the 1949 election.

Here's the contingency table for governments from 1949 on:

(Note added Oct 2014: having found some missing data, "Trailed for some of year" should read Won 7 - Lost 2 Win 78%)

The one losing government that is "not known to have trailed" in its first year is Gorton's/McMahon's (elected 1969) and that assessment is based on a meagre supply of four polls.  That government may well have trailed briefly during its first year, but the same is also true of some governments that were re-elected.

Of the two governments trailing for some of the year that lost, one (Fraser's third, elected 1980) was trailing for almost half of its first year.  The general pattern seems to be - from limited data - that it is an electoral health hazard for a government to trail for most of its first year, but it's by no means fatal.

Abbott's First Year

We're also at around the anniversary point for Tony Abbott as Prime Minister, and this opens up comparisons with not just PMs elected from Opposition but also those who took over mid-term.  The Australian has a graphical thingy for recent PMs but that lacks 2PPs for Howard and Keating (and also over-compresses Abbott's year because there have been fewer Newspolls during it); Peter Brent fills in the gaps for Howard.  Keating's first year in converted 2PPs saw everything from 43:57 deficits early on to a 56:44 lead near the end.

Abbott has recorded marginally the worst average first-year 2PP polling for any new Prime Minister (from Menzies' 1949 return to the job on), but this statistic is slightly misleading as Labor's drawn-out leadership process delayed the resumption of polling.  Abbott has recorded the second-worst first-year personal netsat average after Keating.  In terms of end-of-first-year position for the PMs for whom both figures exist, however, he ranks 7th out of 10 on both, ahead of Gillard and McMahon on both voting intention and personal rating, ahead of Whitlam on voting intention and ahead of Keating on personal rating.  Whitlam and Keating were both re-elected after more than a year in the job, so it is hard to argue that Abbott's current position makes him un-reelectable.

This week's polls

The last two weeks have seen no net movement from Essential, which has stayed at 52:48 to Labor.  Newspoll had a 1% swing back to Labor from 52% to 51% and Morgan went the other way, from 54% to 53% (respondent-allocated, deduct at least 1.5 points from each for house effect.)  In all, my 2PP aggregate sees its first gain to Labor in nearly two months (to 51.7%), but it's only a tenth of a point, and it's too early to be confident the drift back to the Coalition has stopped.  Also, note that BludgerTrack has lately been giving the Coalition most of a point more than I have, so anyone aggregating the aggregators might feel that my figure flatters Labor slightly [Edit 11/9: Bludgertrack has now gone to 52.0 for Labor, so scrub that bit] - and even if it doesn't, Labor have some history of underachievement compared to leadup polling on the day.  So this is by no means a solid lead for the Opposition; it's just better than being caned 55:45 or worse, which is a common fate of deposed ex-governments.

Much is being made of the three-point Newspoll gain for the Greens, but there was no such gain in Morgan and only one point in Essential.  The figure of 14% is very likely excessive even by normal polling standards.  It's possible that khaki bipartisanship over Iraq is driving voters away from the majors and to the Greens, but its also possible that PUP (down one in Essential, still very low in Morgan and apparently down in Newspoll too) are shedding protest votes and some of these are landing in the Greens column. A Queensland state ReachTEL also shows decline in PUP support, as did a Tasmanian state poll, and for all the commentariat views about Jacqui Lambie, it actually looks like some voters are over the party's various antics and incidents.

Here's the smoothed tracking graph, with a tiny uptick at the end that may or may not still be there next time around:

The Strange Magic Of Newspoll Rounding

This week's Newspoll results were mildly controversial because they showed the Coalition down one (39), Labor up one (35), the Greens up three (14) and yet, the Labor 2PP was only up by a point.  How, many asked, could "the left" gain four points of primary vote and only one point of two-party preferred?

The answer comes from rounding.  Newspoll figures are published as integers, but these represent figures originally calculated as decimals and then rounded to the nearest whole number.  So a 39 could be anything from 38.5 to 39.49.  An apparent change of three points for a given party from X to (X+3) could really be just over two points of change ((X+0.49) changes to (X+2.5)) or it could be close to four points ((X-0.5) changes to (X+3.49)).  An apparent change of one point for a given party on either primary or 2PP could really be anything from next to nothing ((X+0.49) rounds down, while (X+0.5) rounds up) to nearly two points ((X-0.49) changes to (X+1.5)).

Also, Labor loses nearly half a point of 2PP for every point by which the Others vote goes down, so a one-point swing from Others to Greens will normally give Labor only just over a third of a point of 2PP gain.

So we don't really know that "the left" gained four points of primary vote. Rather, they may in theory (not looking at the 2PPs for the moment) have gained anything from just over two points to just under six, while Others may have gone down by anything from just over two to just under four.

And we don't really know Labor gained one point of 2PP; it could on paper be anything from just above zero to just below two. If the primaries were exactly as stated (with no rounding) then Labor would have gained just over two points 2PP and a one-point shift would be impossible, but the primaries are frequently rounded.

Consider the following deliberately extreme and artificial examples of raw results prior to rounding:

Case 1: L-NP 39.5 ALP 34.499 Green 11.499 Other 14.5
Case 2 (two weeks later): L-NP 39.499 ALP 34.5 Green 13.5 Other 12.499

Between these two cases the Green vote is up a little over two points, the Others vote is down a little over two points and the major parties barely move.  The actual 2PP for Labor increases in the second case by 0.72 points.  Yet the first case rounds to last Newspoll's results (L-NP 40 ALP 34 Green 11 Other 15) with a 2PP of 51-49 (50.96 rounded up) and the second rounds to this time's (39-35-14-12) with a 2PP of 52-48 (51.68 rounded up).  (Note: my 2PP model includes a small adjustment for Liberal-National contests.  Newspoll's method may differ very slightly).

In fact it isn't necessary to consider rounding of this week's results at all since they come out to about 52.4% 2PP and round down to 52% without making any changes - which is not to say that Labor's 2PP was necessarily that high, just that it may have been.  And it's not necessary to make such implausibly artificial changes to last time's either.  The previous poll would have come out at 49.7% 2PP and rounded up to 50-50 if the numbers were exactly as published.  So either the Labor vote or the Greens vote must have been a few tenths of a point higher than published (perhaps both), and the Coalition vote was a few tenths lower.  That's all that's needed to explain how four points of apparent primary gain over a fortnight is only worth one point of apparent 2PP.

(There are also rare cases where one party's primary must be rounded by more than half a point, to make the totals sum to 100. I don't think these increase the range over which the 2PP result might vary for a given set of primaries, but I haven't looked at this angle exhaustively.)

I have seen various Newspoll 2PPs disbelieved for this sort of reason, and I am yet to find one that fails to check out as possible once rounding is taken into account.

Leadership results

Newspoll this week saw Tony Abbott with an unchanged netsat of -19 and Essential also had Abbott's netsat unchanged in one month, at -17.  Both polls had respondents less committed on the PM's qualities with one point going from each end to undecided in Newspoll, and two points from each end doing so in Essential.  Essential also saw substantial declines in both Strongly Approved (down 4) and Strongly Disapprove (down 5) for Abbott.

Bill Shorten lost six of the seven netsat points he gained last fortnight in Newspoll (which I attributed to an expression of sympathy following his handling of old accusations, a story that so far has blown over quickly) and is back at -7.  Essential's data has him at -1, compared to -6 a month ago, with the gain mainly coming from a reduction in disapprovals.  Essential continues to show that even fewer voters strongly approve of Shorten (5%) than Abbott (9%).  On preferred leader scores, Newspoll has a tie at 37-all and Essential has a one-point Shorten lead, 36-35.

Other results

A good example of how you ask the question mattering was the Iraq/Islamic State issue.  Newspoll asked respondents:

Now, thinking about Iraq.  Currently Australia is providing humanitarian aid and weapons to forces opposing Islamic State militants, otherwise known as ISIS.  Are you in personally in favour or against Australia providing such assistance?

Essential asked:

Do you approve or disapprove of the Australian Government's decision to supply arms and weapons to Kurdish forces fighting in Iraq?

The former got the thumbs up, 62-25, while the latter was a 38-39 near-draw.  The questions are chalk and cheese, and the truth on support for current anti-IS actions is probably somewhere in between.  The first question would have better used the expression "weapons and humanitarian aid" rather than the other way around, since the military aspect of assistance is the most contentious.  The Essential question doesn't mention who the Kurdish forces are fighting and doesn't mention humanitarian aid, while stressing that it is a Government decision is an open invitation to a polarised response.

Essential gets very similar results for Abbott and his government on international relations as they get for Abbott overall, but there is a mildly positive response (44:31) for the demise of Labor's mining tax.  The fomer result supports the evidence that the current focus on national security isn't massively boosting the Coalition's vote, and indeed we don't know how much of the Coalition's recovery since the Budget would have happened anyway.


NB: I was always intending to look at PM and government performance over the first year in this fortnight's piece anyway, but the impetus for looking at older polling specifically after one year as well came from an info request in connection with this Michelle Grattan lecture.


  1. Interesting article in lieu of our discussion on poll bludger.

    it would be interesting to see what happened in the 4 governments that were behind first year and won. Howard was one of them of course.

    It seems that you generally have a 75 percent chance of winning as and incumbent. Abbots might be around 50 or even below

    I have a question. Do 1st term governments recover less or more slowly than 2nd and 3rd term governments.

    1st term governments may be worse at managing public perception without the practice. In this measure gillard would be a 1st term as would rudds 1st term

  2. The governments behind most of their first year that won were:

    * Menzies 51-54: Government was probably already trailing when it introduced a notoriously scary budget five months in. Government trailed for almost its whole term, at times massively, but became competitive very late in its term and was helped over the line by the Petrov affair.

    * Menzies 1961-63: Menzies had nearly lost the 1961 election, because of a credit squeeze. His government trailed for all the following year then became more competitive in early '63 at around the time of the "36 faceless men" episode.

    * Hawke/Keating 1990-93: Labor were losing after a few months of Hawke against Hewson and remained behind the rest of the first year, all of the second and half of the third. They recaptured the lead under Keating in November 1992 when there was a very sudden polling flip probably caused by growing doubts about the GST and mass protests against the new Kennett government in Victoria.

    I did have Howard 1998-2001 in this category but there's been a recount and I've moved them up a notch as the lead was pretty much evenly shared in the first year of that govt.

    All the really notable examples of major recoveries in a great hurry do involve governments later than first term. There are some examples of first-term governments regaining, say, three lost 2PP points quickly (Howard in 1998 (One Nation bubble), Fraser in 1977 campaign, Whitlam in early 1974) so I think it's more that they don't tend to get into trouble often in the first place.

    My own current assessment of the government's re-election percentage chances is: low 60s.