Saturday, May 18, 2013

2013 BC Election and the Pollsters

The dust is slowly settling after the surprising results of the 40th BC general election. The geographical divide between BC's different regions was very stark - even more so than in 2009:

Note: Updated to reflect changes due to the final count.
Island, Coast and Northwest: 15 NDP, 2 LIB, 1 GRN (14-4-0 in 2009)
Vancouver, Burnaby and New Westminster: 11 NDP, 5 LIB (8-8 in 2009)
Rest of Metro Vancouver: 19 LIB, 5 NDP, 1 IND (16-8-1 in 2009)
Fraser Valley and Interior: 23 LIB, 3 NDP (21-5 in 2009)
TOTAL: 49 LIB, 34 NDP, 1 GRN, 1 IND (49-35-0-1 in 2009)

But the salient feature of these results was, of course, its unexpectedness. Why did all the polls miss the boat? This seems to be a recurring theme in recent Canadian elections, but as we saw last fall, polls in the United States have remained quite accurate. Why are American pollsters doing so much better than their Canadian counterparts? Here are a few potential explanations, and how I feel about them.

1. Live Phone vs IVR vs Internet Polling
There is no evidence that traditional polls fare better than non-traditional ones in the United States or other countries. Some traditional Canadian pollsters suggest that new methodologies are behind the erratic polling in this country, but it's unclear why Canada would be different.

2. Resources for Data Analysis (Likely Voter Screen)
That said, methodology probably does matter, but at the stage of data analysis. Some US media organizations, unlike Canadian ones, actually pay substantial sums for their polls. This enables polling firms to do proper research and develop, for example, a reliable likely voter model. The availability of extensive exit polling data in the United States also helps.

3. Polarization and Undecided Voters
As we all know, the differences between Democrats and Republicans are stark, and even a month before the vote, there remain very few undecideds in a US presidential election, and in most US Senate elections. In Canada, the fraction of undecided voters is typically much higher. This produces a potential for sudden shifts that are hard for pollsters to capture.

4. Local Candidates
In the US, polls (except for "generic Congressional ballot" polls, which tend to be inaccurate) state the names that will actually appear on the ballot. In Canada, they typically only state the name of the party. A popular Canadian incumbent from an unpopular party may therefore enjoy more support than pollsters imply.

Hopefully, the Canadian pollsters figure things out before 2015, or the projections for the next federal election could be dead wrong...