Watching the Poll Watchers

Imre Salusinszky had an amusing little column in The Australian yesterday (which unfortunately isn’t online) calling on his fellow pundits:

to wake from your Rip Van Winkle-like slumbers and accept the reality that is staring you in the face. The Howard Government is gone.

His column, intriguingly tagged ‘Hoi Polloi,’ is apparently read by his ‘beloved colleague’ from Sydney’s Eastern Suburbs, Caroline Overington, but she’s not prepared to contemplate such a nightmare.

Advertisement

Interestingly, though, even Dennis Shanahan has avoided the temptation to spin a ‘bounce’ or a ‘fight back’ on the back of the latest Newspoll as had been his wont. At least in his news article (at the time of writing the accompanying opinion piece had yet to be posted), he implicitly acknowledges the margin of error issue when noting ‘little change’ from last week’s survey.

Indeed, there appears to have been a belated recognition of some of the points forcefully made in the blogosphere over the last year about psephology and the interpretation of polling data. As noted by QUT academics Jason Wilson, Axel Bruns and Barry Saunders at the group blog Gatewatching, some sympathetic journos have been openly drawing on and citing the psephological bloggers. And it’s fair to say that in a fairly patchy way, the hitherto unmoveable pundits at The Australian have discovered some basic statistical concepts such as margin of error despite all the bluster earlier in the year about ‘owning the polls.’

More broadly, Tony Wright in The Age has exemplified a switch in the ‘media narrative’ towards the truth of what the polls have been indicating (note not predicting) for a long time:

However, a wall of evidence is now in place showing that the election is now Rudd Labor’s to lose. Pick a poll any poll and trace it back seven months or so and you will find there has been no change or ‘narrowing’ worthy of the name: Howard’s Coalition has been trailing Rudd’s Labor Party by seven to 10 percentage points all along (give or take the occasional spike or dip or rogue poll) and the election campaign hasn’t shaken the trend.

But we’ve got a long way to go. And I don’t just mean the next 18 days before ‘the only poll that counts.’

One of the much touted advantages of the blogosphere (and, despite a lot of the boosterism that surrounds many things Web 2.0, this is a real one) is the phenomenon of aggregating ‘distributed knowledge.’ The big picture is that lots of people whose opinions and knowledge base wouldn’t otherwise come to the attention of the media get a chance to enter (and make an impact on) national conversations something that can build momentum and influence with time, as I think we’ve been seeing with the developments I’ve been discussing.

It’s striking, incidentally, how many Australian bloggers are PhD students and academics from disciplines as disparate as political science, sociology, economics, law and literary studies. The conversational and discursive register used when making a sustained and interactive argument through blogging differs greatly from the ‘expert’ model of the public intellectual where a journo phones someone from the local uni’s Poli-Sci Department for a ‘comment.’ At its best, the blog-based mode of discourse can also lead to genuine interaction and knowledge-sharing between journos and bloggers. And despite the ‘bloggers are destroying journalism’ meme, there are some media folk who are quite open to collaboration rather than competition.

But even more exciting, in a way, is the way an incisive comment on a blog thread can get at a truth rarely acknowledged.

‘Man of Ideas.’ Thanks to Paul Batey.

My main critique of those who’d argue that the polls are in some way predictive is twofold. I’ll leave aside the epistemological questions here, except to say that I see elections as unique and unrepeatable events and the political dynamic as just that. So, in my view, it’s foolhardy to build statistical models that purport to predict results off either past elections or off supposedly causal factors such as economic indicators.

More importantly, for the present purpose, is my view that most polls are sociologically blind. Let’s take a point from one of those commenters I was talking about as an illustration in this case, ‘Dinsdale Piranha’ on the current Newspoll thread at the Poll Bludger blog:

I have made this point before. I’ll say it again. The national 2pp [Two-Party Preferred ‘vote’] is a statistician’s nicety. It is meaningless because there isn’t one election happening, there are 150. The 2pp only means something in the seats themselves.

That’s an excerpt. Go read the whole thing. It’s well worth it.

Superficially this is a similar point to the ‘national polling vs seat polling’ narrative that Overington still clings to like a ‘safety blanket’ (Salusinszky’s phrase, not mine) and that we heard much of from Dennis Shanahan and David Flint earlier in the campaign. But it’s a lot more sophisticated and it goes to the correct insight that Salusinszky has pendulums notwithstanding, it’s not 16 specific seats Labor has to win but any 16.

Perhaps it’s some residue of the Rousseauian ‘general will‘ but we seem to think of elections very perversely in a parliamentary system as if there were some uniform national mind making itself up. There are patterns and trends to the process of electoral decision-making, but the aggregate is something (though only something) of a fiction.

In reality, in campaigns, there are a lot of sociologically distinct groups moving in different directions (though in this election year, it’s been more like moving in the same direction Labor’s at different speeds).

Mark Latham went down terribly with suburban women, for instance. The nature of a parliamentary system means that these groups are micro-aggregated geographically. So we can get quite different trends occurring in different types of seats and different trends in different States and Territories. As our friend ‘Dindsale Piranha’ suggests, it’s more than possible that voters deserting minor Parties and returning to the Coalition are those in the ‘safer’ Liberal seats.

Now, unfortunately, we don’t have the sort of polling that can tell us very much about that in the sort of detail we need. George Megalogenis has a good stab at it by matching census data to electorates, but that doesn’t tell us much in the absence of firstly some degree of tracking of those seats over time and secondly of understanding what actually influences votes. The Australian Elections Study gives us exactly that sort of data but only for past elections.

I’m arguing that partisan shifts which are significant, and I believe we’re seeing some of them in this election, can build up for a long period of time until a tipping point is reached. And time is a huge factor as well. Some of the seemingly inconsistent marginal seat and State figures we’ve been seeing relate to the churn in votes that happens when you observe movement on a smaller scale than the national polls, where a percentage point or two may mean hundreds of thousands of voters.

As I’ve argued before, you can’t extrapolate too much from one-off marginal-seat polling you need the tracking polls the Parties do privately to get a sense of the trend. But, as I’ve also argued before, the Parties only leak polling when it’s in their interest to do so, and you can’t necessarily rely on being told the truth.

So where does that leave us as election watchers?

Much as I don’t hold to the theory that the betting markets are ‘better predictors’ of election results than polls, there is something of the gambler in the political analyst’s method. You need to take what isolated data you can have and make sense of it, and try to weave it into a bigger pattern. And you should prefer macro-data like the ‘national polls’ over micro-data, but still try to understand with a keen sociological eye how that micro-data fits into the bigger picture.

Not dissimilar, in a way, to a punter studying the form before having a bet on the Melbourne Cup.

But you also need, for instance, to discount headline claims about ‘the strength of the economy’ and to look at the lived experience of voters in different demographics and regions. You need to understand how particular legislation or policy actually affects people and those effects can be in the realm of perceptions as well as in the material realm. You need to take an empathetic and imaginative leap in lieu of the focus groups the Parties have access to.

The advocates of ‘betting markets’ have often argued that less reporting of polls would make for better political debate. I doubt that, for a whole range of reasons. But, just as a really good form guide trumps conventional wisdom distilled into a Cup sweep, better reporting of polls, based on better analysis, would leave us much the wiser in the face of what really is an ultimate uncertainty about the eventual winner, no matter how good the favourite’s odds are.

This is an edited version of a post on NewMatilda.com‘s election blog PollieGraph.

What is kwoff?