How to read polls? The failed 2014 early election strategy of the Parti Québécois

By Claire Durand, Professor at the University of Montreal
Blog [in French]

What is the story?

On March 5, 2014, the Parti Québecois (PQ, center-left sovereignist) called an early general election in the Canadian province of Quebec. The party was at that time leading a minority cabinet and was hoping to comfort its position in securing a majority of seats in parliament. The situation was (seemingly) good for the PQ, as some polls estimated that the party was somewhat ahead of its main competitor, the Parti Libéral du Québec (PLQ, center-right federalist) in the general population and several points ahead among French-speaking voters.

However, a month later, on April 7, this strategy appeared to have been a disaster. The PQ lost the election by more than 15 points. What is worse, the PLQ obtained a comfortable majority of seats. What happened? In this text, I show that the polls were not wrong; the incumbent party just misinterpreted them.

Polls in Canada and Quebec

Two ‘catastrophes’ recently occurred in Canadian polling. In the Alberta 2012 election, polls conducted at the end of the campaign predicted that the Wildrose Party would obtain about 10 points more than its main opponent, the incumbent Progressive-Conservative Party. On Election Day, the exact opposite occurred. The Progressive-Conservative Party won by 10 points over the Wild Rose Party. The second catastrophe occurred during the British Columbia 2013 election where polls conducted during the last week of the campaign predicted a victory of the New Democratic Party (NDP) over the incumbent Liberal Party by an average of 8 points. Ultimately, the Liberal Party won the election and ended up 4 points ahead of the NDP.

In Quebec, polls are known to be more reliable. However, they are also known to systematically underestimate the score of the PLQ. In the 2012 Quebec election for example, the party was underestimated by close to 4 points. A number of reasons may explain this phenomenon. Among them, one can mention the possibility that PLQ voters tend to hide their preferences to pollsters more than other voters and to less participate in elections.

Raw versus corrected scores in 2014 Quebec polls

By default, pollsters allocate non-disclosers (those who do not indicate a vote intention) among the parties proportionally to their strength among disclosers (those who do indicate a vote intention). Figure 1 shows the evolution of vote intentions in Quebec in 2014 according to polls published in the media. In February, three polls put the PQ ahead of the PLQ. In early March, at the time when the party called an early election, the party was leading by 2-3 points. According to this figure, the early election strategy of the incumbent party made sense.

durand1

Figure 1. Evolution of vote intentions in Quebec in 2014 (as published).

Until 2002, pollsters in Quebec used to correct for the overestimation of PLQ by attributing the vote of non-disclosers in the following way: 60% to the PLQ, 30% to the PQ, and 10% to the third party (nowadays represented by the CAQ). I have continued working with a non-proportional attribution of non-disclosers since then but I decided to modify the formula and attribute 50% of non-disclosers to the PLQ and 25% each to the PQ and the third party. This formula gave good estimations of the vote in recent elections (2003, 2007, 2008 and 2012). Figure 2 reports the corrected vote intentions using this formula. As we can see, the situation of the PQ in early March was far from being advantageous. The PLQ was ahead by 1-2 points. If the PQ had used this number, they would have been more careful before calling an early election.

Durand2

Figure 2. Evolution of vote intentions in Quebec in 2014 (corrected).

This entry was posted in Blog. Bookmark the permalink. Both comments and trackbacks are currently closed.