Sep 29, 2012

The Primary Dynamic

The extraordinary growth in the relationship between perceptions of the Prime Minister and the electoral fortunes of the government they lead. A statistical analysis of our new primary dynamic.

Possum Comitatus — Editor of Pollytics

Possum Comitatus

Editor of Pollytics

Before we get on to more serious business, which might sound like a bit of an odd thing to say considering the zeitgeist – here’s the latest Pollytrends.



Which are all kinds of self-explanatory.

Now for the comparatively serious business – the primary dynamic of voter perception on federal politics since the end of 2007.

We all know that there’s a relationship between the vote a government receives in the polls and the satisfaction with the Prime Minister of the day. It’s hardly ground breaking stuff – to show how it all plays out over the very long run (say, since 1986) we can simply chart the two party preferred vote of the government of the day against the Prime Minister’s satisfaction rating, and do it again with the net satisfaction rating (where it’s Satisfaction minus Dissatisfaction). We’ll use Newspoll monthly averages.


I’ve added a simple linear regression line there to make the point – but it’s one worth covering quickly. The relationship here is statistically significant to all the usual levels and if we just look at that last chart which compares the two party preferred vote of the government against the Prime Minister’s net satisfaction rating, we find that changes in net satisfaction explain a little over half of the variation in the two party preferred vote over the long term since 1986.

The stats just happen to play out in such a (fortuitous) way that if the net satisfaction of the Prime Minister was exactly zero – where there were as many people equally as satisfied with the PM as there were dissatisfied – we would expect the two party preferred vote of the government to be exactly 50% to the nearest whole percentage.  For every 10 point change in net satisfaction, we’d expect the two party preferred vote to change  by 1.3 points – towards the government with net satisfaction increases and away from the government with net satisfaction decreases.

So over the long term it’s been a solid relationship, but not an overwhelmingly dominant one – with the dynamic explaining 55% of the variation in the government’s two party preferred.

Before we go any further, it’s worth noting that the relationship between the net satisfaction of the leader of the opposition and the two party preferred vote of the opposition is completely non-existent over the long term.

A drunk bloke shooting paintballs on a chart comes to mind with that particular graphic.

So the vote of the government is intrinsically linked to satisfaction levels of the Prime Minister in the general case, and the public’s satisfaction or otherwise with the opposition leader of the day is really neither here nor there. It’s all about the government/PM dynamic as far as this goes, and how the opposition strategies and tactics operate within that constraint. So if you tear down the PM, the government’s vote will go with it – hardly rocket science – but it explains a lot.

However, what you may not know, is that while this relationship holds for all federal governments collectively since 1986 (as we’ve seen), it also holds for each individual government since 1986 –  but with significant differences between them.

Let’s have a look at how the relationship played out in the Hawke/Keating government.

Through the Hawke/Keating period between 1986 (which is as far back as Newspoll goes by the way) and its end in 1996, the dynamic was very similar to what we saw looking at all governments together. A 10 point change in net satisfaction of Hawke/Keating would be expected to come with a change in the two party preferred vote of the government of 1.3, with the dynamics explaining 52.5% of the variation on the government vote.

Moving on to Howard, we have:

Here the variation in Howard’s net satisfaction rating (between 1996 and 2007) only explained 41% of the variation in the Howard government’s two party preferred vote, significantly less than the proportion explained during the Hawke/Keating government it replaced.

Yet for every 10 point increase in Howard’s net satisfaction, we’d expect the two party preferred vote of the Howard government to move by 1.5 points – slightly more than for the Hawke/Keating government it replaced.

So while changes in Howard’s net satisfaction ratings resulted, on average, in larger changes to the Howard government vote than occurred under Hawke/Keating, there was much more variation in the size of those changes.

Now let’s move to the Rudd/Gillard era – it’s quite spectacular.

Changes in the net satisfaction rating of the Prime Minister explains 89% (!!) of the variation in the two party preferred vote – over twice that of the Howard government. For every 10 point change in the net satisfaction rating of the PM, the Labor government two party preferred vote changed, on average, by over 1.7 points – the highest of the three governments.

The dynamic between the PMs satisfaction and the government’s vote during the Rudd/Gillard era is utterly dominant. But wait! [nerd time!] It’s so dominant, that even that incredible explanatory power of 89% very likely underestimates the true strength of the relationship because of sampling error and rounding errors in the polls that we’re using for analysis.

We know that polls aren’t dot points, but are approximately normal probability distributions (think “bell curve”) with a mean of the published poll result and a standard deviation of some particular size that is a function of the sample size of the poll itself – that’s where our margin of error comes from for each poll. So the statistical relationship between the PM’s net satisfaction ratings and government’s TPP will have some noise in it simply from sampling error, plus a small additional noise component from rounding issues. How much on the rounding issues? Well, satisfaction and dissatisfaction are rounded to the nearest whole number when published, and we are subtracting one from the other to get net satisfaction (adding up to 0.9 points of rounding error on net satisfaction and up to 0.5 on the two party preferred results)

So let’s run a hypothetical – let’s assume that the exact quantitative relationship between net satisfaction and the government vote in the Rudd/Gillard era is the regression line in the above chart – where the two party preferred is 50.078 when net satisfaction is zero, and where a 1 point change in net satisfaction comes with a 0.1718 change in two party preferred. In this hypothetical world, the true underlying values of each poll would fall exactly on that line every time.

But we have noise in the polls. Because we can finely estimate the probability distributions of the components that make up the noise, we can model what that noise would look like using a bit of monte carlo simulation. So in our hypothetical world where we knew that linear relation between net satisfaction and TPP was exactly true, exactly all of the time, what would Newspoll charts look like if they tried to measure it, simply because of the statistical noise involved? Well here’s 4 examples (click on each chart to expand it)



It looks pretty close to what Newspoll has actually measured. After 100,000 iterations (or examples) of this simulation, the R-squared value (the one that tells us what proportion of the variation in the two party preferred results can be explained by variations in the net satisfaction ratings) averages 91.1%.

So even with a hypothetical “perfect relationship” between net satisfaction and government two party preferred, we would expect to see polling produce R squares of only around 91% – with the 9% unexplained residual being created purely through statistical and methodological issues associated with the underlying data generation and presentation process.

With the real world Newspoll results giving us 89%, the actual strength of the true underlying relationship is quite likely higher than that –  making it simply astonishing.

But correlation isn’t causation” I hear you say! And you’d be right – so let’s go a step further and look not just at correlation, but peer as best we can into causation.

One of the tests we have available is Granger Causality – which tests whether one time series can predict future values of another time series. It’s not a measure of perfect causality – for no such thing generically exists – but it’s a measure of a practical causality.

Running Granger Causality tests on our two series (which are called NETSATGOV and TPPGOV in my database – representing PM net satisfaction and the government two party preferred respectively), this is what we get on a month by month change basis for all of the government’s we’re looking at, as well as over the entire period since 1986.

So we can reject the null hypothesis that net satisfaction does not Granger Cause the two party preferred vote (suggesting that in actuality it *does* granger cause TPP), and we cannot reject the null hypothesis that TPP does not Granger Cause net satisfaction (telling us that TPP doesn’t Granger Cause net satisfaction).

So on a month by month basis, net satisfaction Granger Causes the government’s vote. That’s the effective direction of practical causality that we can measure.

Over periods of time longer than month-by-month, both series Granger Cause each other – suggesting what’s commonly known as a feedback loop. Using some factor analysis and Vector Autoregression Models, the feedback runs about 70/30 in favour of net satisfaction.

To summarise –  looking at all governments since 1986, over the short term, changes in net satisfaction cause changes in the government vote, while over the medium term, a feedback loop develops between net satisfaction and the two party preferred vote – but where net satisfaction is the primary driver of around 70% of that medium term relationship.

However, while the feedback loop is about 70/30 over the whole period, for Hawke/Keating and Howard it was closer to 60/40 while for Rudd/Gillard it is closer to 80/20.

Now we know the direction of the correlation between PM satisfaction and the vote, it’s probably a good time for a recap.

With nearly 90% of the change in the vote of the Rudd/Gillard governments being driven by changes in the public perception of the Prime Minister – twice that of the Howard government and astonishing in its own right as a number – this is quite simply the primary dynamic of federal politics over the last four and a half years.

Our public perceptions of leadership have become all encompassing of our politics . Change perceptions of that leadership, change the vote – drive perceptions of the PM into the dirt, drive the government’s vote into the dirt with it. Lift the public’s satisfaction with the PM up, the government vote gets dragged up too.. If you want to understand the dynamics of federal politics, you need to recognise the prism through which voters are seeing and behaving to politics and the actual nature of the battlefield – this is it.

The electoral fortunes of the government since 2008 have been effectively drifting on whatever currents of public perception about the Prime Minister were flowing at the time. The Howard and Hawke/Keating governments before them appeared to have developed some form of institutional anchor that ameliorated the voting consequences of the vagaries of Prime Ministerial perceptions (to about half of what is happening today according to the stats). Though I often wonder how Hawke, Keating and Howard would have operated in today’s media environment where facts are generally considered to be optional.

Whether it’s the government, or us, or the media environment (or some combination thereof) that has caused this dynamic, the one thing we do know is that leadership matters more today than it has at any time over the last 25 years. The Prime Minister today doesn’t just lead the government – as far as voter perceptions and voting intentions go, they *are* the government.

Small note: The sidebar charts have Gremlins – they’ll be updated with the latest data when it’s solved.

(Visited 100 times, 1 visits today)


Leave a comment

12 thoughts on “The Primary Dynamic

  1. Kevin Bonham

    Thought it was useful to focus on just the Gillard/Abbott dynamic without the distractions of Rudd, Turnbull and Nelson.

    * Gillard netsat strongly Grainger-causes 2PP.

    * 2PP does not Grainger-cause Gillard netsat.

    * Abbott netsat and 2PP weakly Grainger-cause each other.

    * Gillard netsat Grainger-causes Abbott netsat.

    * Abbott netsat weakly Grainger-causes Gillard netsat.

    No expert on Grainger, but a possible interpretation is that when Gillard behaves in an unpopular way this not only drags down the 2PP but also brings out the most negative in Abbott, causing him to sling mud some of which sticks to the PM but a lot of which sticks to him.

    We may now be approaching the mysterious point where we see if Abbott’s ability to damage Gillard reduces when the Gillard-fail factor subsides and whether this results in him shooting himself in the foot less often or whether he just floats off on his own.

  2. Kevin Bonham

    Fraser was actually a popular Opposition Leader for most of his time in the job. Mostly his netsats were in the +20s. He polled bad ratings in that role only for a few weeks during a major constitutional impass which, as it turned out, rebounded massively in his favour when Whitlam was sacked. (At the same time both Whitlam and his government had briefly bounced back in polling and become popular, having earlier been as far behind as 33-60 on primaries.) Even then, Fraser’s worst netsat was -24, which is bad, but not as bad as some.

    After Whitlam was sacked Fraser still had fairly bad netsats for the first two weeks of the campaign, but he then recovered to record two -2s in the last two weeks and won the election massively.

    Fraser’s momentary unpopularity during an extremely turbulent time is not the same thing as Abbott recording shockers for over a year in a row. And Fraser is a dubious example anyway because he had a big assist from the Governor-General.

    It is true that state opposition leaders have sometimes won with very bad approval ratings even at the time of the election, but there are very few examples, with Kennett and Court the most obvious. But state politics is not a model for federal politics in this regard anyway because the federal situation very strongly influences state election results. This can be seen on Antony Green’s graphs here: http://blogs.abc.net.au/antonygreen/2012/04/the-cycles-of-party-politics.html Once a party wins government federally it will typically shed seats at state elections until it is eventually kicked out federally. This means that where the party in opposition at state level is also in opposition federally, it has an advantage. Both Kennett and Court had that advantage, and had it in spades because the federal government at the time was unpopular, and were also up against ageing Labor state governments.

    Before Abbott, there was actually a strong link at federal level between large Opposition poll leads and good Opposition Leader netsats. The median Opposition Leader netsat for a pre-Abbott Newspoll with an Opposition 2PP of 54 or greater was +16.5. The Abbott tenure has trashed this all the way to -1 which is little different from the long term median netsat for Opposition Leaders irrespective of 2PP.

    The drunk-bloke-shooting-paintballs chart is confounded by the tendency of new Opposition Leaders to record positive netsats (because of the high number of don’t-knows who often later become negatives), and they account for a lot of the low-2PP-high-netsat dots, since new leaders are more likely to take over when their party is badly trailing. Nearly all the cases of netsats over 10 for an Opposition Leader whose 2PP was 45 or worse come from Beazley’s first eight months (Beazley had high dunno ratings for a long time, though not as long as Hayden) or from the early months of Turnbull or Nelson. This problem doesn’t exist so much with new PMs because they are usually known quantities; their don’t-know rate goes down much more quickly.

    For cases where the undecided figure is below 20 this is what the scatter looked like before Tony Abbott:


    (Not a strong relationship but it seems to sober up the drunk bloke, though I think he becomes drunk again if you also remove Rudd.)

    And this is what it looks like for Tony Abbott:


    That lonely little triangle at (50,-30) is the most recent reading.

    Even if Opposition Leader ratings don’t predict 2PP on an instantaneous basis, I think the track record of lastingly unpopular opposition leaders never winning federally provides good reason to keep an eye on them.

  3. Warren Joffe

    spur212 seems to have it about right. And, it might be added, it is probably a good thing that a government’s chances depend so much on the standing of the leader. It does suggest a certain willingness of the community to move together as a kind of faith community which implies more solidarity than one would sometimes think one saw in our country.
    As you PC say the standing of the government leader matters in a way the Leader of the Opposition’s standing doesn’t and it is not new news. I ran into Gary Morgan once in the 1980s when the Victorian state Liberal Party was about to attempt another coup against Jeffrey Kennett and he was quite hot under the collar about the stupidity of it because, as he noted, while the Premier’s approval rating mattered, the Opposition Leader’s didn’t. I remember him citing Malcolm Fraser’s abysmal approval rating in 1975.

  4. fredex

    Gotta do somethin’ about the media in this country!

  5. Outsider

    Absolutely fascinating analysis! It’s implications are profound, and I suspect similar analysis underpins the entire opposition strategy since the rise of Abbott. Sustained attacks on the PM of the day (both Rudd and Gillard), drive down the net satisfaction PM rating, and bingo! It’s been spectacular stuff, and with the damage done to the PM’s credibility, through fair means and foul, perhaps it’s all too late for Labor. What is of greater concerns is what this all augurs for the future of Australian politics, and Possum’s comments about how Hawke/Keating/Howard would have fared in the current media environment resonate.

  6. Possum Comitatus


    The “ALL” in the final table isn’t the average of the three separate periods, it’s the entire period taken as a whole.

    The whole doesn’t necessarily equal the sum of the parts.

  7. Michael Fink

    Way out of my depth here, but is there a typo in the last table?

    The average(?) TPP change from a ten point change in PM net satisfaction is listed as 1.31, which is less than the figures for all three periods under consideration.

  8. Possum Comitatus


    The number of Qld State wide public polls from Galaxy and ReachTEL etc are pretty small, but we also have seat level data and unpublished polling. So we can model it in a more complicated fashion that accounts for demographics and regional cluster effects, relational responses other than satisfaction and a bag of other things.

    I haven’t looked at other states other than Qld – I don’t think I actually have the spare brainspace to accomodate it at the moment!

    It will be interesting though to see whether the dynamic we’ve witnessed not only across the last two PMs, but the tight satisfaction/vote share path they’ve both taken, continue – because it might be stuctural (as in, the political reaction to media reaction to political reaction nexus that treats PMs predominantly as political celebrity)

    Kevin, the 1993 election result often confounds political models of any description!

    Interesting that the dynamic seems to hold as much at election time as it does in the more hypothetical mid term polling though

  9. Kevin Bonham

    I thought it would be interesting to plug the election-campaign netsats and election 2PP outcomes into the equations for each of the periods given and see how they scrubbed up. I’ve just used the average of all Newspolls in my data set that were less than a month out from the election. So …


    1987 (11 July) Hawke +11.7 Proj 52.6 Actual 50.8
    1990 (24 March) Hawke -11.3 Proj 49.6 Actual 49.9
    1993 (13 March) Keating -29.8 Proj 47.1 Actual 51.4
    1996 (2 March) Keating -19.2 Proj 48.5 Actual 46.4


    1998 (3 Oct) Howard -8.3* Proj 47.8 Actual 49.0
    2001 (10 Nov) Howard +22.2* Proj 52.3 Actual 51.0
    2004 (9 Oct) Howard +12.4 Proj 50.9 Actual 52.7
    2007 (24 Nov) Howard +2.2 Proj 49.4 Actual 47.3


    2010 (21 Aug) Gillard +3 Proj 50.6 Actual 50.1

    * Polls for Howard in 1998 and 2001 were worse closer to the election.

    It’s only nine data points but there’s no reason to think that elections are behaving much differently here, except for the big bad outlier 1993.

    1993 was also a big bad outlier on a regression I did that plotted 2PP election result against worst 2PP polling position for the party in government in that term (and using approximations from Gallup data to go way back to 1949). It’s an especially notable outlier on that test considering that Labor otherwise tended to markedly underperform their projection on that one and the Coalition to overperform it.

    One thing I am curious about is that while there is apparently no correlation between Opp netsat and Opp 2PP, lasting bad netsats for an Opposition leader is a very strong predictor of defeat assuming that leader makes it to an election at all (and this goes back to 1969).

  10. John64

    “Our public perceptions of leadership have become all encompassing of our politics.”

    … or is it a phenomenon that’s only seen when we have particularly good (Rudd?) and particularly bad (Gillard?) leaders?

  11. Laocoon

    Hey Possum

    Very nice work. One of your comments at PB was:
    [My research for the union picked up the identical dynamic playing out in Qld with Campbell Newman – it’s why I went and had a look for it Federally in the first place.]
    I was wondering. The number of observations on Campbell Newman must be relatively few, and also in a period of a rather roller coaster ride (moving from very high netsat to lower netsat very quickly indeed).

    And thinking about Rudd/Gillard, their PM tenures has also been marked by quite exceptional levels of netsat (both highs/lows). It is also a somewhat shorter period that the Hawke/Keating, Howard decades.

    I wonder whether the strength of relationship between netsat and 2PP could be a manifestation of the measurement in a period of extremes of the netsat measures (rather than any a change in the media/general body politic). And that if the measurement included periods when the ship of state was in more becalmed waters, the strength of the relationship would then decline.

    Have you seen a comparable change in strength of the relationship in other states (ex Qld)?

  12. spur212

    So lets see:

    1. Reputation and trust are everything

    2. If you’re a political leader in this country, you must lead with authority, respect, trust and have the ability to take people with you when making policy changes

    3. Once a leader’s reputation gets damaged, it’s pretty much impossible to turn it around (because of the unhinging dynamic of uncertainty, high expectations and the electorates need for simple explanations)

    4. If you take a stand on an issue that people believe in and want and you know how to communicate properly and have a clean reputation, you can reap a lot of rewards

Share this article with a friend

Just fill out the fields below and we'll send your friend a link to this article along with a message from you.

Your details

Your friend's details