Polling lessons from Election ’07

Two lessons from the election result and the polls:

1. As a basis for understanding what’s going on, data are better than gut instinct.

2. Polls conducted closest to election day are likely to be more accurate than those conducted earlier.

On lesson 1:

All four main polls had been tracking with remarkable consistency the voting intention of the electorate for months.

Given this consistency, and the generally good record for accuracy built up by the main polls in Australia over many years, there was no basis for thinking the election would result in other than a clear Labor win.

Aside from the consistency of the national surveys, there were clear signs of what was likely to happen from Newspoll’s big marginal-seats survey in the first week of November,.

This survey, across the 18 most marginal Coalition seats, showed an average swing to Labor of 7%. Newspoll’s founder and former chairman, Sol Lebovic, noted that this swing, if repeated on election day, would deliver 25 seats to Labor. This will be very close to the actual result when the count is complete.

On lesson 2:

The Morgan poll, conducted on the eve of the election, was clearly the most accurate of the national surveys taken in the final week, as the table shows.

It is no coincidence that the most accurate of the final surveys was the one taken nearest to the election day.

Now it is undoubtedly true, as AC Nielsen’s John Stirton has acknowledged, that their final poll, showing a vast two-party-preferred lead of 14 points to Labor, was a “rogue” poll. That is, it was among the 5% of polls that produce results outside sampling variance. This is hard luck.

Even so, there is a lot of evidence to show that one way to minimise the risk of inaccuracy is to poll as close to election day as possible. This doesn’t eliminate the rogue-poll problem, but it cuts down on other risks, particularly those arising from potentially vote-changing events in the last few days.

For example, had there been a poll taken in the seat of Lindsay before the scandal over the racist anti-Labor propaganda there on Thursday, it is likely that the poll result would have been embarrassingly out of date by Saturday.

Our experience, when polling for The Age and the SMH, was that election-eve polls tend to provide the most reliable data. We did this for the 1987, 1990 and 1993 elections, and were able to get them published in the two newspapers the following morning, election day.

Because you have little time, you may have to compromise on sample size (we used a sample of 1000), but the results suggested that this was more than compensated for by the accuracy that came with reaching people as late as possible before they actually voted.

Aside from Nielsen’s rogue poll, the big four did well this time. Morgan’s result, to within 0.2 percentage points of the actual result so far, was excellent, but Newspoll and Galaxy, each within 1.3 percentage points, produced results that were well within sampling variance.

Never believe politicians when they say that they don’t watch the polls. They watch them all the time because they know that, plus or minus a bit, they reflect what is going on in the electorate.

The ordinary voter is almost as perceptive as the polls. A clear majority could smell a Labor victory.

The professional bookies could also smell it, the smell of money being the most potent portent of all.