Sunday, May 1, 2011

Pollsters and seat projectors should beware Monday's results

****UPDATED 12:15 AM MAY 2ND

"Polls don’t elect MPs," said Michael Ignatieff today. "Votes elect MPs. Let’s wait for the Canadian people to do what they want to do."

The pseudo-science of gauging public opinion in a world in which many voters (especially younger voters) no longer own land lines or ignore entreaties to take part in online surveys will be truly tested on Monday night in Canada.

Frankly, I look forward to re-assessing which pollsters still are able to accurately gauge public opinion in this modern world and which ones have little credibility.

We can today look at the final numbers predicted by pollsters:

Decima-Research: Cns 36%, NDP 30%, Libs 19%, Greens 6%, BQ 6%.

Nanos: Cns 37.1%, NDP 31.6%, Liberals 20.5%, Bloc 5.7%, Green 3.8%.

EKOS: Cns 33.9%, NDP 31.6, Libs 20.8%, Bloc 6.4%, Greens 5.9%.

Forum Research: Cns 35%, NDP 33%, Liberals 19%.

Abacus Data: Cns 37%, NDP 32%, Liberals 18%, Greens 7%, BQ 7%.

Angus-Reid online poll: Cns 37%, NDP 33%, Liberals 19%, BQ 6%, Greens 4%.

Ipsos-Reid: Cns 38%, NDP 33%, Liberals 18%, BQ 7%, Greens 4%.

I can't find any final numbers from Environics.

And, of course, the always laughable Compass, frequently happy to boost Conservative spirits when they need it most: Cns 46%, NDP 26%, Libs 17%, Bloc 7%, Greens 4%.

I look forward to comparing the final vote percentages with these guys' numbers on Monday night.

Even more interesting will be how accurate seat predictors will be.

Forum Research, despite putting the Conservatives and NDP in a dead heat in voter support, still predicts the Tories will gain seats and land at 147 by the end of the night. But they caution with the normal margins of error for their survey, the election still might produce a either a Conservative or NDP minority government.'s Eric Grenier has had a great run with regular columns in the Globe based on his projections. *****Tonight, he pegs Conservative support st 36.4%, NDP at 27.3%, Libs at 22.8%, Bloc 6.7% and Greens at 5.6%. Grenier estimates this will translate into a House of Commons made up of 143 Conservatives, 78 New Democrats, 60 Liberals, and 27 Bloc members. I think he pegs the NDP too low and the Liberals (sadly) and the Bloc too high.

But of course I can't forget Grenier's claim in November that the Liberals would win only 14% of the vote in the Winnipeg North by-election, based simply on his analysis around vote percentage changes. Of course, in that local race, the Liberals won the seat with 46% of the vote. Of course, we'll see how well his final predictions stack up against the real results tomorrow night.

Election Prediction Project is another great site and it's been reliably cautious is changing its predictions for the entire campaign. Only in the last few days have the NDP seat counts increased substantially. As of midnight ET, the site predicts 146 Conservatives, 65 Liberals, 61 NDP, 33 Bloc, 1 Independent in Quebec and 2 races too close to call (Dartmouth-Cole Harbour in NS and Gilles Duceppe's Laurier-Ste-Marie in Quebec, both of which will not be won by Conservatives).

And hasn't put out his final numbers yet. The last predictions are from April 29th, so I imagine he'll update these by tonight.

It's all about the numbers now.


Unknown said...

Just thought it should be pointed out that Ekos AND Nanos DO reach cell phones.

It's not like they call people out of the phone book. Some companies use random digit dialing (which by definition is equally likely to reach cell phones or land lines), others have numbers sold to them (which includes cell phones).

It really is amazing that this "pollsters aren't reaching cell phones" myth perpetuates. Kady O'Malley, Andrew Coyne and Paul Wells have tried to go to great lengths to dispel it. I guess it will endure nonetheless.

Matt Guerin said...

Thanks for the point about cell phones. I do still doubt that reliable cell phone numbers are easily contacted. Random dialing would produce so many wrong numbers, it seems you might dial 50 wrong numbers for one correct number, and then would that person actually pick up?

If my cell phone company has sold my phone number to a pollster, I'll be very pissed. Generally, I don't answer my cell unless I recognize the number.

I'm not denying that pollsters can't still find ways to be somewhat accurate. There is a definite trend picked up at the end of the campaign by all pollsters. We'll just see what the real vote turns out to be tomorrow.