13 September 2006

Shag the polls

Despite all the excellent commentary out there that we should all be highly suspicious of public opinion polls, people still report ones coming out of New Brunswick as if they were reliable.

Take for example the Canadian Press story on canadaeast.com, a site that bills itself as Eastern Canada's information source. It deals with a poll conduct by Bristol Group's research arm, Omnifacts.

According to CP, the poll results found that the Lord Tories are in front with 45% of decided voters with the Liberals at 38%. 37% reportedly were undecided.

Further down the story, you'll see a quote from Don Mills, president of Corporate Research Associates, the research company in the CCL Group. He dismisses undecideds claiming that they usually split like the decideds.

The CP story reports the margin of error on this poll at 4.4% at the 95th confidence interval meaning the results would fall with plus or minus 4.4% of the reported figure 95 times if you did 100 polls using exactly the same methodology.

Here's a couple of quick points:

1. CP is wrong. The small sample size doesn't make the margin of error high, the overall survey design does. You can accurately poll national public opinion on a sample only slightly larger than the one used here. You can do New Brunswick with fewer people and get a smaller m of e.

Dig deeper and you'll likely find problems with response rates, sampling methodology, demographics and so forth that are all understandable but all of which can render the results functionally useless.

No sane person would rely on these numbers alone to predict seat counts, for example.

2. Given that both the Omnifacts and CRA polls have such horrendous margins of error (both pushing 5%), simply toss them in the bin. They are useless as a way of telling anything at all about the election.

The polls are only good for the firms releasing them. They get what they want: media coverage to market themselves. Nothing wrong with it but we shouldn't pay any more attention to the results than we would to yet another MacDonald's spot.

Their secondary value is to the party in the lead which can add to the impression - potentially false - that they are heading for a comfortable win or to a trailing party that it is in contention, possibly equally false.

3. As for Mills' ongoing belief that undecideds either break like the decideds or don't count 'cause they won't vote, we can only say "bullshit".

There's no way he can sustain the latter conclusion unless he specifically determined likelihood of voting and then screened those people out from the start. Given his comments, Mills doesn't exclude those people so he gets them in both the undecideds and in the decideds.

He also can't sustain his former conclusion - namely that they break like decideds. They don't. In 1999, undecideds polled after the lone leadership debate in the Newfoundland general election broke heavily Tory. There are countless other examples of the same thing happening.

In interpreting poll results, some people often distribute the undecideds around like the decideds simply as a convenient way of guessing what might happen. But that's a dodgy game unless one has conducted far more in-depth polling than the stuff done by research companies and released like this.

Publicly released polling results during campaigns are fine as long as you understand what they are and what they aren't.

All things considered, the polling in Atlantic Canada elections give a bit of news coverage but people following the elections shouldn't put much stock in them, one way or the other.