Friday, August 30, 2013
And down the stretch they come
Silver's signature analysis for the NYT of the error rates of US poll companies in 2012 is not itself flawless, since there did appear to be a very late swing towards the incumbent of a point or so from the Hurricane Sandy aftermath which threw everybody out of whack even more than they already were, which is why all but four of the pollsters he listed showed a Republican lean. It showed that the methodology of a poll does carry its own problems - robopolls conducted via automated systems and/or those which exclude mobile phones both oversample conservative voters - but that in-house errors due to sampling techniques can be even more pronounced, up to a 7.2% gap between reality and the results of usually respectable Gallup.
We don't have a local version of Silver, a credible interpreter of the polls for a mass audience who has the statistical chops to back up his analysis. Antony Green is our closest, but he stays out of the daily hustle and bustle to concentrate on electoral process. William Bowe does his best at Poll Bludger over at Crikey with his BludgerTracker index updating for every new poll and hosting massive comment streams where partisans bludgeon each other with the latest figures, while Kevin Bonham goes more in depth in a weekly round up.
Kevin's article on the first week of the 2013 election campaign sums up the high point for Labor polling thus far, as it has been all pain all the time since then. Back then, even though they trailed slightly on the two-party-preferred, it was possible to posit a Labor victory given that they were going to gain more marginals in Queensland and WA than they would lose in NSW, Victoria and Tasmania. In the weeks after, every part of that theory has been progressively attacked by national, state and individual seat polls.
If Andrew Elder's theory that Tony Abbott is never going to be the Prime Minister is any chance of coming true, not only does the 2PP need to swing back to somewhere near the first week's levels, there has to be something wrong with the seat-level polling which in marginal Labor seats has shown as much as a nine per cent swing against the ALP. These are all robopolls and/or have very small samples, so they are weak evidence. One wonders if their marginals are indeed hosting such a swing, why the national 2PP is so strong for Labor. Where are their extra votes coming from, safe seats? It's a mystery, not really helped by a strong base of diverse, comprehensive and credible poll sources.
Then there are the betting markets, which started off as highly favouring a Coalition victory even in the first week when the 2PP was 50:50, and have blown out to unbackable odds with one outlet already announcing payouts to those gambling on the Libs. Not that that is a sure fire sign, either: bettors on wins by Tiger Woods and Chris Judd in various recent markets have enjoyed early payouts for competitions in which their ticket didn't end up winning. Politics is a novelty market for bookies, conducted for promotional purposes only.
If I was Nate Silver looking at all of this, I daresay I would be going against the prevailing trend as he did last year and declaring the race as still too close to call. He would probably not have as much faith in the quality and breadth of Australian pollsters to make as confident a call as he made last year, that Obama had a 90%+ chance of winning. That is not to say that Rudd has much of a chance, mind you. The margin of error for Labor is their only hope at this point, but as the days count down and the polls don't narrow by enough, they are running out of doubt from which to take the benefit.