Assessing opinion polls
Therese Arseneau a regular contributor to TVNZ has an article on opinion polls at the TVNZ site. The whole article is worth a read, as Ms Arseneau discusses the current crop of polls and the attempted disparagement of the results by Helen Clark. Amongst other matters Ms Arseneau writes:-
When assessing the reliability of polls much depends on the details of the survey. Three points are striking. First, the findings of the three polls are consistent. Roy Morgan, One News Colmar Brunton and Fairfax Media-AC Nielsen found support for National at 52.5%, 55%, and 54% and Labour at 31.5%, 29% and 30% respectively.
Second, the sample sizes for these polls are good: 775 people interviewed in the Morgan poll, 1000 (with 831 probed party supporters) by Colmar Brunton and 1101 by AC Nielsen. When the surveys are projected out to estimate support for the two major parties in the broader population, they have a margin of error, with 95% confidence, of roughly plus or minus 3.6% in the Morgan poll, 3.5% in Colmar Brunton and 3% in AC Nielsen.
Labour says a 95% confidence interval means one poll in 20 may be a “rogue poll”. This term is misleading. A 95% confidence interval simply means one poll in 20 may produce a result outside the stated margin of error – and it could be just 0.1% outside. But the chance of all three polls being outside at the same time is more like 1 in 8,000 – statistically possible but highly improbable.
Third, all three are part of a long running series of polls and, crucially, the June results are not out of line with the polling trends.
Later, in conclusion, she sounds a note of caution:-
At this point in the election cycle polls are best used as indicators of voting trends and vote flows. In MMP the source of a vote gain is as important as the gain itself. At the moment National is trending upwards and mainly at Labour’s expense, hence claims that a change of government is likely.
But if National starts to fall in the polls, focus turns to where the support goes. If it is to Labour or Labour’s likely support parties, then the election will be tighter. The consequences for National are not as dire if the support moves to non-vote. And if it moves to National’s allies, then it is not really lost at all.
Finally, polls should be used sparingly. Elections are too often treated as horse races, and polls used by the impatient to predict the winner before the race is even run. The focus now should be on the campaign itself and on providing voters with the information they need to make an informed decision.
As Adam noted earlier well worth reading the complete item.
