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.
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.