An Australian Nate Silver?

- First published in the Australian, 23 November 2012.

Since Nate Silver’s fame was well and truly cemented with the US presidential election result, some have suggested that Australia “needs a Nate Silver”.
Silver, who blogs for the New York Times, rose to fame with PECOTA, a system for forecasting baseball players’ performance. (Wikipedia entry here.) I can’t claim to know much about that, but assuming it does what it’s supposed to it sounds like a work of great genius and originality.
What he did for the presidential election was more mundane: synthesise the voting intentions opinion polls and produce state-by-state and overall odds.
Silver wasn’t the only person to do this, just the most famous. Simon Jackman, an Australian academic at Stanford University (who sometimes writes for this newspaper) was another, and he came up with pretty well the same odds.
On election eve they gave Obama about an 80 per cent chance of victory and Romney about 20 per cent. That 80 per cent encompassed everything from a narrow win to a big win. What was unusual was that all the polls were right in all the states. I wonder what the odds of that were?
The fact that Obama won didn’t make Silver “right”, and if Mitt Romney had won it wouldn’t have proved him “wrong”.
If anyone was “right” it was the opinion polls. Silver no doubt crunched them in a very sophisticated way, digging beneath the surface by obtaining details from pollsters and doing some of the weighting and other manipulation that pollsters generally do themselves.
In 2008 Silver apparently got all the states right except one (Indiana) but what that really meant was the polls got all the states right except one.
It was surprising to learn afterwards that the Romney camp actually expected to win the election. One imagines a campaign team operates two mindsets concurrently: the realistic one and the hopeful one. That’s fair enough. On one level you know you’re not likely to win, but on the other you say “I’m feeling pretty good; we might just do it.”
And the polls, while great in number, all showed it being reasonably close. They weren’t pointing to a landslide.
But it seems the Republicans really expected Romney would be comfortably elected. For that belief to take hold they had to have hard-headed number-crunching types telling them it was so.
And people like Karl Rove and Dick Morris did do that. They too had convinced themselves; they thought most of the pollsters’ and Silver’s assumptions and measurements regarding turnout of different groups of people, such as Latinos and Blacks, were wrong. (Read Morris’ mea culpa here.)
There was also something about surveyed Independents (when Americans register to vote they usually have to describe themselves as “Republican” or “Democrat” or “Independent”) slightly favouring Romney. Usually whoever Independents vote for wins. But everyone knew Independents this time consisted of an extra group of people to the right of the Republicans who would favour Romney. People were saying that before the election; there was no excuse for not realising it.
Camp Romney created its reality, from the ground up.
But back to the original proposition of an Australian Nate Silver. For such a creature to exist would require two things. The first is lots and lots more published opinion polls. Americans don’t just have 14 times the poll data we have (14 being the approximate magnitude of their greater electorate size); they have more than that.
The second aspect relates to the electoral architecture and makes the task even harder. Americans particularly poll the “swing states”—those that tend to go to the winner and which have more than a handful of Electoral College votes. The Australian House of Representatives has three times as many “electorates”, 150 of them currently, and they are of equal size. To predict seat-by-seat, pollsters would have to survey individually all that have the slightest chance of being in play.
That would be a massive task. The economies of scale do not remotely exist.
Currently the closest we have are people like Pollytics and Poliquant. Both have a Nate-like understanding of stats and they average and weight and project the latest published polls to give a two-party-preferred number they then apply to the pendulum, or state-by-state bits of the pendulum. (Pollytics has other “secret” data but that would play only a small part.)
To me this is going a bit overboard given the meagre data available. (They both also make other calculations.) We can all see the recent published polls and can get the vibe ourselves. It will be more useful during the campaign, but even then there are so few polls it’s debatable whether a formula that takes into account last week’s polls is more useful than just looking at today’s.
Nate reckons he’s not that much into politics actually. With the election out of the way, sport is creeping into his blog.
Which is here.