Tuesday, December 11, 2012

Is the Silly Season Real?

(Admin note to readers: jumpbreaks poll closed - thanks to those who voted - outcome at bottom of post)

 Advance Summary

1. The view that a swing to the Coalition in the final Newspoll of 2012 was due to a general tendency for voters to "switch off" politics as Christmas approaches is disproven by historic evidence.

2. Comparisons between the December 2012 and December 2011 Newspolls on account of them having the same two-party preferred vote are simplistic and misleading because they fail to take into account surrounding trend data.

3. Although there has been a now clear and significant move to a substantial Coalition lead since early November, Labor's polling position in mid-December 2012 is not as bad as at the same time in 2011.

4. Leads similar to the Coalition's current lead have been held by many losing Oppositions five or six months prior to elections.

5. Whatever intuitive or subjective views different commentators hold, informed by data or otherwise, there is no known valid scientific basis for predicting the result of the next federal election at this stage, nor even for declaring a very strong favourite.  This applies no matter who the leaders are.



---------------------------------------------------------------------
Yesterday's Newspoll, showing a far from remarkable 3-point 2PP jump to the Coalition against the backdrop of a general trend back to the Coalition in other polls, has provoked much gnashing of teeth and even some conspiracy-theorising among Labor supporters, and a lot of triumphalist chestbeating from supporters of the Coalition.

A common response has been to blame it all on the Coalition's support of the AWU saga, a response which makes little sense given that the AWU saga is primarily about Julia Gillard, and her own Newspoll approval ratings have barely moved at all (one point).  Of course, it's possible there was a substantial change in voting intention that was confined to voters who disliked both Gillard and Abbott already, which would represent the Preferred Prime Minister factoid making a gallant effort to thumb its little nose at me after I dismissed it as rubbish yesterday.  But it's more likely that most of the three points is noise - that the previous 51 for the Coalition would have been a 52 with more sampling, that this 54 would have been a 53, and that random poll-to-poll bouncing is making it look like much more than it is.  None of the multi-poll aggregators are showing a genuine three-point shift to the Coalition yet; they are typically showing more like 1.5-2 points, and over a month, not a fortnight.

Additionally, people looking for Reasons (the capital R is a Mumblism) to explain the supposed shift could look at anything from humorless responses to the Gillard end-of-the-world speech to increasing gloom about the prospects of a budget surplus. 

But of all the theories that have been put about to explain something as prosaic as a three-point 2PP shift in a poll that has a long-term average poll-to-poll 2PP shift of 1.92 points (most of it caused by sample noise) the one that caught my eye - repeatedly - is the idea that we are in the silly season and that this poll should therefore be ignored.

According to this theory, at this time of year voters have their minds so full of tinsel that they are unable to focus on politics.  In the various guises I saw it in last night this means they may (i) favour the conservatives (ii) favour the party that's in front (iii) bounce around in voting intentions more (iv) consider the year as a whole not current developments (v) lose interest in politics.  I thought I'd have a look at this, to see if there is any evidence that voters behave any more differently in the last poll of the year to the one (typically) two weeks before.

For this I used Newspoll data back to 1985 and the null hypothesis that a poll being the last poll of the year makes no difference to any of the things it was supposed to, and set out to see if I could disprove the null hypothesis based on the following measures:

1. The 2PP standing of the Government at the time
2. The poll to poll change in 2PP.
3. The 2PP of the Coalition parties
4. The 2PP of the party that is ahead
5. As an indicator of voter disinterest levels, the sum of the "uncertain" figure for the PM's approval question and the Opposition Leader's.

In each case I compared the last Newspoll of December (usually but not always the only one in that month) with the most recent Newspoll prior to it, unless that one was taken only one week before, in which case I used the next most recent.

I excluded a number of years from the sample entirely because of factors likely to produce misleading results; these were:

2001 as a federal election had very recently been conducted
2003 as the Opposition Leader was replaced during the relevant period
2004 as a federal election had very recently been conducted
2006 as the Opposition Leader was replaced during the relevant period
2007 as there was no December Newspoll and a federal election had recently been conducted
2009 as the Opposition Leader was replaced during the relevant period

Here are the results for the four before-and-after type tests (mean score, standard deviation, standard error) :


The average poll to poll difference in 2PP was 1.69 points, which isn't even above the long-term average, let alone significantly.

For those not familiar with standard deviation and standard error, the first is a measure of how much individual data points tend to vary from the mean and the second is a measure of how far out the estimate of the mean is likely to be.  With more data, the standard error figure goes down and small differences in mean between the two groups become more likely to mean something.

I don't need to quote p-values to make the point that there is not a lot to see here - none of the differences are even close to statistically significant (especially not when running multiple comparisons at once) and even if they were they are very small average differences anyway.  There's nothing there that would suggest a 3-point swing to a Coalition Opposition in the last Newspoll of the year is in any way something that is produced by it being the final Newspoll. Null hypothesis 5 - silly season 0.

And indeed, those who thought there was some kind of pattern of the last poll of the year moving to the Coalition only had to look at 2011 to see that in that case the last Newspoll had a 3-point 2PP swing to Labor.

Is Everything Old New Again?

This brings me to another of the myths that has sprung up - that since last year's final Newspoll was 46:54 and this one's is too, that therefore we are back where we started. We are not.  Peter Brent's column today may be confused for this genre, but is actually (in context) a refutation of the idea of
"the Gillard government’s end-of-year recovery", and not an argument that Labor has really made no progress. Nonetheless, the similarity in stats between this year and last year is the basis for the myth that Labor has made no progress, so I'll use the stats Peter presented as examples, and place them in the context of previous Newspolls.

Yes, the Coalition 2PP in the current Newspoll is the same.  But the end-of-2011 Newspoll had been preceded by a 57, a 54, a 53 and a 57, while this 54 follows a 54, a 50 and two 51s and is therefore well over the rolling average.

Yes, Gillard's approval is 36 now just as it was then, but her disapproval is 4 points lower and her netsat of -16 follows a run of -14, -16, -15, -15, compared with last year's -20 coming after -33, -30, -30, -21 (and it soon went up again).

No, Abbott's approval is not 30 compared with 33 (it is actually now 28), and not only is his netsat actually seven points worse than in December 2011, but his -24 last December followed -17, -21, -23, -21.  His -31 this time follows -22, -28, -36 and -31.

And even on PPM, if you must: Gillard's 9-point lead now follows leads of 10, 11, 14 and 13, while Gillard's 7-point lead in December 2011 followed balances of -5, -3, -1 and 5.  (Indeed that Dec 2011 PPM score of 7 for Gillard was her best since May that year and she would not have a lead that large again til this September). 

The December 2011 Newspoll was thus a relatively lenient spot in a run of generally terrible polling for Labor, while the current poll, while having the same 2PP and only modestly better leadership figures for Labor, is a probably exaggerated (by Newspoll standards) reversal following a run of very competitive results.  It is not usually the case that a single poll describes the situation better than a rolling average of polls from the same company, or better still a multi-poll aggregator, and the aggregators show Labor's position now as being about 1-2 points better than then. On my own Newspoll-specific rolling averages, Labor's position now is 2.4 points better on 2PP, 8.6 points better on PM netsat, and 8.2 points better on net lead on that other indicator some people insist on taking seriously, while the Opposition Leader's netsat is 8.8 points worse.  The 2PP shift is exaggerated by a run of friendly recent Newspolls, but the idea that a single loosely matching poll shows that Labor has made no progress in a year is simply false.  Whether Labor has made sufficient progress remains to be seen.

Certainty In An Uncertain Environment

A characteristic of the current polling environment is that we have quite a few observers going "all in" on predictions of what will occur at the next election although we have never seen a federal election in circumstances much like it before. For instance Andrew Elder, a much more intelligent and original commentator than the vast majority of those who do that sort of thing for a full-time living, has carried the prediction that "Tony Abbott will never be Prime Minister of Australia" as a badge of honour on his site for some time now.  On the other hand,  Peter Brent has recently made unequivocal declarations that if Labor go to an election with a leadership team of Gillard and Swan they will lose, even against Abbott.  Brent's view seems to be supported mostly by a view that Gillard and Swan are hopeless political communicators, which in turn doesn't seem to rest on any empirical data in particular, while Elder's view that Abbott cannot win rests on extensive analysis of Abbott's personality and policy deficits.

It may well be those seemingly conflicting views will be resolved by Abbott not making it to the next election at all, allowing both to be right.  I don't think there is any good data basis for calling the election one way or another, or even being all that confident of the outcome. Subjective or intuitive arguments can be made, with or without data, in any direction because there is so much to be said against both sides.   Some of those making such arguments will be right, and some will be wrong, Looking at the 2PP situation, if an election is held in about August next year, we are eight months out, but similar leads were held by Oppositions five to seven months out in five successive election cycles (1993 to 2004) and only one of those Oppositions won.  Indeed, as Poliquant just noted (and 1993 is another example), 2PP changes of four points or more within campaigns are not unheard of. 

The next codicil often advanced goes like this (I've picked Pollbludger member and Rudd barracker ShowsOn's post as a good type specimen).  In essence, Labor's a minority government (and one that lost its own majority too), and therefore won't display the kind of resilience that majority governments have usually done - voters will throw it out so they don't have to listen to another Rob Oakeshott speech ever again.    Seems intuitively convincing.  But we have no federal standard of comparison to know if that is how voters will actually behave.  At state level, while governments that have lost their majorities then served a term in minority have often been cleaned out the election after (exceptions include SA 1977 and ACT 2012), some of those have lost against the headwind of having their own party in power federally.  It's also possible that the government's minority position is already factored into its polling trajectory, meaning that those who want to throw it out on that account are already on the Coalition's side, and hence that there is no "resilience" difference after all. 

Good luck to those who "know" the outcome of the next election already, some of whom will just be wrong, some of whom will get the result right without having really known, and some of whom may actually be on to something!

---------------------------------------------------------------------------------------------------------

 (Admin note to readers: the "jump breaks" poll has closed.  Thankyou very much to all who voted; I value your input greatly.  The result of the vote on primaries is Yes 13 Only for old articles not the current one 11 No 7 What's a jump break? 2 (one of them mine for monitoring purposes.)

I am treating the last option as an informal vote.  Candidate No is excluded in third place and its preferences are deemed to flow sufficiently heavily to "Only for old articles not the current one" to make that the winner.  Another way of looking at it is that 24 voters supported jump breaks for old articles cf 7 who opposed, while 18 voters opposed jump breaks for the present article whereas 13 supported them.  Therefore, I deliver the majority finding in both cases and most older articles except "About" will henceforth be jumpbreaked.) 


1 comment:

The comment system is unreliable. If you cannot submit comments you can email me a comment (via email link in profile) - email must be entitled: Comment for publication, followed by the name of the article you wish to comment on. Comments are accepted in full or not at all. Comments will be published under the name the email is sent from unless an alias is clearly requested and stated. If you submit a comment which is not accepted within a few days you can also email me and I will check if it has been received.