Pages

09 December 2005

SES and Gregg compared - some observations

Something has been bothering me all day about the Globe/CTV/Strategic Counsel claim that regional "weakness" in the Liberal support could make this a much closer race than it appears.

Later in the day, I saw Allan Gregg discussing the numbers for British Columbia. The Strategic Counsel poll shows the Liberals, Conservatives and New Democrats in a near tie for public support.

I checked the margin of error (MoE) for the BC results only to find out it was +/- 7%, 19 times out of 20. The whole report is here, at thestrategiccounsel.com

Ok.

Check SES and let's compare them:

Gregg has the Liberals at 30, the Conservatives at 29, the NDP at 31, the Greens at 10, with an MoE at 7%.

SES shows Liberals at 42, CPC at 29, NDP at 23 and the Greens at 5 with an MoE of +/- 6.4%.

Right off the bat, I have some problem with Gregg basing his analysis on numbers with MoE of anything more than 3% plus or minus. Look at these polls and you'll see why - Gregg shows a near tie. SES shows the Liberals well ahead.

But, if you adjust the figures within the margin of error for both, you can easily produce a result that has the Liberals at 36, CPC at 29, NDP at 27 and the Greens running around 7.5%. All I did was split the difference on the numbers. That produces a result which is actually closer to the SES result than the Gregg dead heat scenario. The MoE is so high for both BC results, though, that I could just as easily produce a bunch of numbers that have the Liberals sweeping the place or tanking and still be fine.

BC may be the most dramatic example of this problem with the Gregg analysis, but there are similar variations in the other numbers. For example, Gregg's national figures have been consistently out of line with other polling, not just SES, sometimes by as much as four or more percentage points.

Theoretically, pollsters using similar methodologies should produce results within the margins of error. Trust me, when these guys were all polling during the spring, I used their research reports to illustrate polling methods to a class of students. A simple chart showed just how similar the numbers were, given the same time frames, for four or five separate firms.

For Gregg and Strategic Counsel, though, I haven't been able to come up with a good explanation for the anomalous numbers. Undoubtedly there is some statistician's reasoning but since I am not one of those, it escapes me.

Another part of the Gregg report released today also caused some concern, but for a different reason. On page five there's a lovely colourful graph of the Strategic Counsel's momentum analysis beginning in May. It shows, pretty clearly that the Conservative momentum has dropped, by Gregg's figures from 36 to 25, while the Liberals have climbed from 22 to 27.

That looks pretty straightforward - but if you look at the way the graph is constructed, it appears that the Liberals and Conservatives are so close as to be almost tied. The digits say one thing; the picture says another. And that's my problem.

The decline for the Conservatives is twice as big as the gain for the Liberals. By whatever methodology you want to use, therein lies a telling story at this stage of the campaign. Rather than try to talk about close races, the SC analysis could have easily pointed to a stalling of the Conservative campaign.

That analysis would marry up with the national voter choice figures, that have a small margin of error, and help people get a broader sense of what has occurred at the end of the second week of campaigning. That approach would also give the Connie bloggers something to think about instead of all sorts of feeble excuses for why their party isn't showing in the polls as they wished for, all year.

I am left wondering why Gregg chose the approach he did when there is another story in his own numbers.

As for the seat projections, I am reasonably comfortable in my belief that Gregg's comments today were not based on his own polling data alone. If you look at democraticSpace.com, you can see a rather complex seat projection model built on everyone's public polling. Sure enough it supports the idea that there may be a small Liberal majority even given the national polling numbers. But notice - he is basing his analysis on everyone's numbers. Gregg is showing just his own and then somehow producing the same analysis. It doesn't add up.

This certainly doesn't mean that Strategic Counsel is cooking the books. Far from it. Allan Gregg is a competent, professional opinion researcher. What may be happening here is that the pressure for news that is different - what the client is looking for - may be affecting what Gregg can toss up to talk about. It's a real-world issue and no one can fault Gregg for dealing with it. The national exposure he is getting is worth its weight in billable hours gold.

For those of us who are political junkies, though, this election coverage and the access to some fairly sophisticated polling results - as public results go - is like having your own meth-lab. We can fiddle with numbers, run seat projections and basically develop a picture of the world using the work of some of the best in the opinion research business.

There's also a good warning in all this for people who are wedded to one set of polling numbers: get a divorce, or at least negotiate an open marriage. In order to get an accurate picture of what is going on out there among the Canadian electorate, you have to take in as much data you can from as many sources as possible and then make your own assessment.

Even after decades of public opinion research, there are still perils in polling, as my old professor, Hugh Whelan, used to say.