Dialogue: For political pollsters some voters don't exist

By KEITH RANKIN*

Political polls are important in a democracy. They help voters to know who the real contenders are and, therefore, how to make the most effective use of their votes.

More importantly, they enable our political representatives to gain ongoing feedback about how well they are representing - or being seen to represent - the public on whose behalf they govern.

Because of this strong and ongoing link between the people and their representatives, the people must take more responsibility for adverse political outcomes. One political outcome this year - the splitting of the Alliance that forced the early election - can in large part be attributed to its poor showing in the polls in 2000 and last year.

It has always been a puzzle why the Alliance support fell so much in polls that showed strong support for the Government, of which the Alliance was a part. The apparent loss of support for the Alliance has been the only source of instability that this Government has had to face.

Certainly, the impression given by the polls, as reported, was that Alliance voters were switching to Labour. If that interpretation was correct, it could hardly be argued that Alliance voters were punishing the Alliance in Government for being too close to Labour. Yet there was no other reason for Alliance voters to punish their party.

An alternative interpretation is that the polls were misreported in 2000 and last year and that they are still being misreported today. Indeed, in general, the misreporting of polls may be the largest single reason small parties in Government self-destruct.

The most important bit of information in political polls - the proportion of undecided voters - is the least reported. Further, this problem of failing to report the level of indecision has become more pervasive in the past few years, presumably because it is easier to translate poll results to seats in Parliament if only the preferences of decided respondents are reported.

It's a pity because the polling companies generally do a good job. Because of their professionalism, polls are capable of conveying accurate information about voters' opinions.

The proportion of undecided voters is normally high between elections. So the erosion of support in the polls for the Alliance in 2000 might have been, in the main, 1999 Alliance voters becoming undecided rather than shifting their allegiance to Labour.

Further, it is likely that many of the undecideds in mid-2000 who voted Alliance were also undecided in mid-1999. The pattern is that the undecided vote only firms during the election campaign, and disproportionately goes to parties other than National and Labour.

It follows, therefore, that support for parties other than Labour and National disproportionately reverts to "undecided" once the election (and the honeymoon) are over.

This process can only be monitored if the "undecided vote" is traced as a significant part of the poll result and not just as some irrelevance that can be ignored by the media.

The Alliance might not have been in anything like the trouble that caused it to panic last year. The flipside of the Alliance crisis was the apparent popularity of Labour. The reported polls led the pundits to believe that Labour made huge popularity gains after 1999.

But that might not be so. Rather, Labour's support increased only in proportion to the declining number of committed voters.

As this campaign progresses, the support for Labour appears to decline as the undecideds come out once again in favour of the smaller parties. It's not Labour losing support; it's simply that both Labour and National support is now expressed as a percentage of a larger pool of voters than in earlier polls.

We need a new way of reporting polls. When a news report says that the support for Labour is 50 per cent, readers, listeners and viewers should be able to assume that 50 per cent of the people polled said they would vote Labour. That's far from the case today.

For example, Wednesday's Herald-DigiPoll survey records that 46.7 per cent of those polled said they would vote Labour. That suggests 426 out of the sample of 912 opted for Labour. In fact, probably only about 350 said that.

If we had truth in reporting, very few commentators or politicians would have seriously considered that Labour ever had much chance of gaining an outright victory.

The election campaign is all about winning the hearts and minds of undecided voters. Why then are the polls, which are based on sound statistical principles, reported as if the critically important undecided voters do not matter or, worse, do not exist?

* Keith Rankin teaches statistics and economics at Unitec.

Full news coverage:
nzherald.co.nz/election

Election links:
The parties, policies, voting information, and more

Ask a politician:
Send us a question, on any topic, addressed to any party leader. We'll choose the best questions to put to the leaders, and publish the answers in our election coverage.