Beyond the Poll

I'd like to talk a little bit about polling. You know, one of the traditional lynchpins of market and political research. The word poll derives from an ancient Nordic word that translates to head. So polling, in its most primordial form, is really just the act of counting heads. The act of enumerating the electorate seems to have developed as a way of countering the inherent noisiness of the old acclamation system, whereby a canditate was elected on the basis of the loudness of his supporters. Over several centuries, the poll evolved to the point where it is now one of the hallmarks of all opinion-centered research, with huge tracking samples of 10,000 people. One imagines this concatenation of polls (heads) as innumerable little grains of sugar, from whose opinions generalizations about the population are made.

But just how accurate are polls in the modern context? Let's look at a parochial example. There was a provincial election held here in Quebec a few days ago. An inexperienced, poorly funded, weakly organized right wing party shocked the province by garnering 31% of the popular vote, when every poll done in the 7 days prior to the election had pegged their support at no more than 26%. So unexpected was this result that a local commentator went so far as to opine that traditional polling paradigms may no longer hold true. To that end, new predictive approaches have emerged that attempt to forecast elections not on the basis of polling, but on the basis of online and offline media buzz.

But maybe it's not polling that's flawed, but the pollers themselves. Perhaps the pollers just aren't listening properly. It could be that the subtle shifts in public opinion that happen from night to night (even hour to hour) are insensible to the unwieldy weekly/bi-weekly polls done by the big firms in the run-up to an election. There is some evidence to support this. In the 2006 Canadian federal election, tiny SES Research ran circles around the established firms in their final pre-election poll by nailing the results for each party's share of the popular vote to within one-tenth of one percent of the actual figure. And they didn't use a massive sample to do it. Rather, they used a rolling nightly tracking poll; instead of listening periodically, they listened constantly. Thus, they were able to track the ebb and flow of opinion right up to the very last, and most decisive, moment.