Tuesday, February 05, 2008

Polling results often inaccurate due to small sample sizes - but pollsters not disclosing

Bill Tieleman’s 24 Hours Column
Tuesday February 5, 2008

Polls don't tell the whole truth


Polls are for dogs.

- Prime Minister John Diefenbaker

Across Canada the federal Liberal Party is ahead of the Stephen Harper Conservatives by three percentage points, 32 per cent to 29 per cent say pollsters Harris/Decima.

Or the Conservatives are ahead of Stephane Dion's Liberals by eight points, 37 per cent to 29 per cent, reports Ipsos-Reid.

In provincial politics, the Green Party is at 21 per cent, just seven points behind the B.C. New Democratic Party, which at 28 per cent is a massive 20 points behind the Gordon Campbell B.C. Liberals at 48 per cent, according to Environics.

Or the B.C. NDP is at 35 per cent, the B.C. Liberals at 45 per cent and the Greens at 16 per cent, Ipsos-Canada says.

Huh? Welcome to the wacky world of political polling, where you can take your pick of results. Just don't criticize the pollsters!

Polling problems were highlighted again in the United States Democratic Party primaries last month, where a Gallup poll in New Hampshire said Barack Obama was 13 per cent ahead of Hilary Clinton just before the vote. But Clinton beat Obama by 39 per cent to 37 per cent.

One big problem is small sample sizes, something pollsters are very loathe to publicize.

In fact, when I contacted Environics Research last week to ask what the B.C. sample size was in their provincial poll, vice-president Tony Coulson e-mailed me that it was just 200, with a margin of error of plus or minus seven per cent.

But their online information doesn't mention that and even says that the results were from a poll of 2,032 residents of Canada with an estimated accuracy of within 2.2 per cent. Coulson said the lack of B.C. information was a "quality control error" and would be corrected.

Still, Ipsos-Canada senior vice-president John Wright isn't impressed by such complaints.

"Forgive me, but I've heard the argument on small sample size for 19 years," he said in an interview. "Would you have us put the margin of error after every number?"

Well, yes.

Here's the problem - a January 27 Ipsos-Canada federal polling news release states: "In British Columbia, the Tories at 43 per cent are well ahead of the Grits (26 per cent), NDP (15 per cent) and Green Party (13 per cent)."

Only if you download online polling tables do you find out that just a miniscule 110 voters were polled in B.C. - and the margin of error is not reported. In italics as the bottom of the news release it states that: "The margin of error will be larger within regions."

Bob Penner isn't impressed.

"Anything less than a 300 sample for B.C. should be viewed with a great deal of skepticism," says the president of Strategic Communications, 24 hours newspaper's official pollsters. "The NDP is not at 15 per cent in B.C. - that's a reflection of the sample size."

Time to walk the dog.


Anonymous said...

"Anything less than a 300 sample for B.C. should be viewed with a great deal of skepticism," says the president of Strategic Communications, 24 hours newspaper's official pollsters. "The NDP is not at 15 per cent in B.C. - that's a reflection of the sample size."

Mustel, with its 851 sample size for BC's federal voting intentions, had the NDP at 21% in early November.

Budd Campbell said...

The polls have other problems besides small sample sizes. For one thing, there are few people willing to answer the phone these days, and that may twist the sample whatever its size.

Also, pollsters in Canada don't focus on likely voters as American ones do. One pollster told me that all the research showed those people who vote and those who have a clear voting intention are essentially the same or similar people, but I am not entirely convinced. My personal sense of the situation is that there are all kinds of blowhards who will be happy to express some opinion or preference, but cannot be bothered to vote, because it's really just entertainment for them.

Anonymous said...

I just ran through a number of US papers and some show Obama in the leader while others claim he is behind more than ten points.They all have polls toagree with what they are saying. A BC Political Science professor who's expertise is US Politics and a large number of US papers say Clinton is way ahead. Others tell us it's too close to call.
Some tell us that McCain will wrap it up today and now it seems the far right religious chap Huckalbee has won the first state today.

The poll that counts is the one the folks are in today. So lets all sit back and watch the results come in. Polls are places dog's use when doing their bodily function outside. I always enjoy it when some upset happens and the experts shift geats and try to tell us they knew it all along.

THe NDP are shown as beyond reach every time, but sometimes they get to run things. a large amount of folks simply don't bother following the political events until election day, if at all

Anonymous said...

Hey, Bill, you are right on about the polls! I joined both major polling groups. They wanta to kinow from the outset whom you voted for in the last few elections, so you bet I don't get consulted on any political things! They give me questions on shopping and investing habits. Even with the random phone calls, the pollsters ask a few questions, and when it's obvious that I'm Left (and I'm more Left than the NDP brass is willing to be), they drop the interview! So these pollsters, far from being objective, are trying to elicit a certain result from their polls, so that they can use it to influence the feeble-minded/the undecided voters who still read and watch the Asper media.