Living in a First-Past-the-Post electoral system leads to a number of interesting intellectual exercises for those interested in politics. Consider, for example, the current Ontario provincial election. A recent poll (ending November 3) by Nanos Research shows that the Conservatives (36.4) are statistically tied with the incumbent Liberals (35.9). Does this mean that they will get the same number of the seats? How can we tell?
The Laurier Institute for the Study of Public Opinion (LISPOP) has an answer. Use aggregate poll results from near the end of the campaign in combination with the results from the previous elections to generate the number of seats for each party. The model was developed by respected political scientists. So what is the prediction?
On October 3, 2011 the Institute predicts that the seats will be distributed as:
- Liberals 47
- Conservative 39
- NDP 21
According to the projection, the model tends to be accurate (based on recent history) within 4 seats.
As mentioned in previous reports, the reason the Conservatives trail the Liberals in seats despite having the same popular support pertains to the small number of seats they lost by less than 10% of in the previous election (LISPOP release from October 3)
So this would suggest that a Liberal minority is fairly safe and a Liberal majority might even be possible — something that the aggregate result would not suggest. Could this influence voters if it was widely known?
The model clearly combines new information from that of the past and is therefore at serious risk if either one of two things occurs in the election. First, the polls do not catch a late shift in voting intentions or if people are unwilling to acknowledge that they are voting for a controversial leader. Second, if the election is disruptive. The last federal election would certainly fit the model of being disruptive and not surprisingly, the LISPOP projections were widely off (missing the Conservative seat count by 23).
I have always been uneasy with the idea that we should try to predict an outcome based on data, which fundamentally is not capable of supporting the prediction. Provincial polls can predict vote share not seat share. Voters don’t behave like trains and so the past is not a good indication of where they are going the future.
The question: Is this harmless fun? Is there a risk to democracy to release these educated guesses? Does it help voters in a positive way?