By Tom McFeat Do gamblers and academics have the answer?
There's been a lot of talk about the polls this campaign. We love taking a peek into the minds of our fellow voters. On the surface, it seems astonishing that a random sample of 1,200 people can yield a reasonably accurate picture of the voting intentions in a country of 32 million. But polling's accuracy (within certain margins) is statistically provable and empirically verifiable.
Polls do have their limitations, of course. For one thing, they don't predict what will happen on election day. They are merely snapshots of stated voter intentions as of the date they're taken. Pollsters also assume that respondents are telling them the truth. And there's also that "undecided" category.
But for some, their biggest drawback is that polls don't directly convert voter intentions into seats. What could 38 per cent support nationally mean if it were to carry through to election day? Minority government? Majority? The answer, of course, is that the conversion of voting intentions into seat projections is an inexact science, at best – especially in Canada.
But there are gamblers, academics and others more than willing to give it a try. And they've got theories, position papers, and mathematical models to back up their efforts.
Election stock markets
Sometimes, the most reliable indicator of how an election will turn out has come from the futures market. The idea is that when people put their own money at risk, they'll do a better job at predicting who will win than polling. Polls ask who people want to win; futures markets look ahead and ask who they think will win.
The University of British Columbia's Sauder School of Business has operated futures markets for various elections in Canada and the U.S. since 1993. In the 2000 federal election, its market predictions for party vote shares were more accurate than many of the final polls from the major polling companies. But it underestimated the size of the Liberal majority by 18 seats. For this election, it's running several trading markets, including one that allows people to invest/bet on the percentage of total seats each party will get on election day. Bets run from $25 to $1,000. In the 2000 election, total investments were $113,000.
Prof. Werner Antweiler at UBC has also developed an election forecaster, which is separate from the UBC election stock market. Prof. Antweiler's forecaster is an interactive online program that requires you to enter the probability that people who voted for one party in the 2004 election will shift their votes to another party in 2006. His forecast program then translates your particular "voter migration matrix� to figure out how many seats each party will win this time around.
Prof. Antweiler is careful to note several caveats with his model – among them that it doesn't take into account unique local factors or changes in population. This forecast depends on the figures you input, so the UBC election forecaster doesn't come up with specific seat projections itself. But others use it as the basis for their projections. The website Jord.ca, for example, has used Prof. Antweiler's election forecaster and then applied its own subjective analysis in some ridings to come out with its projection.
The public relations firm, Hill & Knowlton, also has an interactive "election predictor" on its site.
It uses a mathematical formula to generate seat projections, based on data supplied by visitors. The predictor shows what might happen when a percentage of the vote swings from one party to another – either nationally or regionally. The company calls it "a fun yet informative tool."
Other seat projection models
The Laurier Institute for the Study of Public Opinion and Policy (LISPOP) has developed a seat projection estimate that's based on a blending of national polls and a "regional swing model� developed by Prof. Barry Kay, a political scientist at Wilfrid Laurier University. Prof. Kay's model refines something called the cube law. Simply put, the cube law is a mathematical formula that attempts to translate popular vote into seats for each party.
That cube law hasn't worked too well in Canada because of the highly regional nature of our politics and because we don't have a two-party system. Prof. Kay's model takes into account the swing in popular support that parties experience from election to election in five different regions in Canada. As with UBC's election forecaster, voter migration patterns are the key.
DemocraticSpace.com is another seat projection exercise. It uses a five-poll rolling average and a model developed by Prof. Greg Morrow of MIT. It uses a four-step process to adjust provincial and regional data to come up with provincial and national projections. In the 2004 election, its final projection was correct within 4 seats for each of the parties.
Several international betting sites also accept wagers on various political races in Canada. Nothing scientific about these sites. The Dublin-based Trade Exchange Network allows people to wager on the likelihood of a majority, the party that will win the most seats, and the number of seats won by each party on Jan. 23. Netherlands Antilles-based PinnacleSports.com offers an over-under betting pool on total seats won by each party.
Some seat projections
The following are seat projections as of 10 a.m. EST Jan. 19, 2006. Take them with a few grains of salt. After all, these projections have all changed dramatically (just like the polls) throughout the campaign.