26 May 2005

The Perils of Polls

It's interesting to notice the Labrador result in light of the Telelink poll done for NTV in St. John's a few weeks ago.

Telelink reported a 42% undecided/don't know/no answer rate, but among decided voters, the Liberals were ahead with 29% and the Conservatives came in with 23%.

Telelink spokesperson Cindy Roma said the results were "too close to call". As I said when the poll was released the results seemed a little bit off to me.

If you look at the actual numbers - not the percentages of decideds, what Telelink actually got was something like this:

Liberal: 13.92%
Conservative: 11.04%
Und/DK/NR: 42%

Essentially, the answer with those numbers was not that it was too close to call. Nope. The answer actually was: we don't have a freakin' clue what will happen because our poll results don't give us data on which to make any conclusions.

Since Telelink didn't probe the undecideds at all, they couldn't even get a sense of which way people were leaning and try to pick some sense out of it on that basis.

If you look at the provincial district breakouts Telelink did, their confidence interval jumped up to something like 10 to 12%. Those results are completely useless and should never have hit the air.

As you will recall, I did contact NTV to get a copy of the research report. I was advised my e-mail had been passed on to higher management.

To date, I haven't received a reply from NTV. I haven't had the time to follow-up on it, but I will do so because the whole thing just gets me even more curious the more I think about it.

There are probably good answers to what questions I do have and the answers might also help people who aren't familiar with polls to appreciate why MUN political scientist Michael Temillini wound up making some off-the-wall comments on Canada AM yesterday based on the Telelink data.

I'll keep you posted.