Election 2004: opinion polls - part 2
I have mused on a few occasions about a structural pro-Labor bias in the Morgan two-party preferred opinion poll, when compared with Newspoll.
The crux of my argument has been that if there was no bias, probability theory suggests Morgan would only have Labor ahead of Newspoll for half the poll dates over the longer term. However, Morgan almost always had Labor ahead. This bias is depicted in the following graph of all Morgan and Newspoll polls for the period 1 January 2003 to 9 October 2004 (the date of the last Federal election).
A good question at this point is to ask why I label this phenomenon as Morgan’s pro-Labor bias rather than Newspoll’s pro-Coalition bias. The simple answer is that for the last two Federal elections Morgan has predicted a much stronger two-party preferred vote for Labor than either Newspoll predicted or the electorate delivered. I concede this is a contestable description, and as will become clear as you read on, at least some of the problem was a pro-Coalition artefact in the 2002 and 2003 Newspoll data.
On Thursday, I wondered whether this bias could be explained by the treatment of preferences. In the poll immediately prior to the Federal election, Morgan significantly overestimated the flow of preferences to Labor and under estimated the flow of preferences to the Coalition from voters who gave their first vote to a minor party or an independent. My preliminary observation was that the same trend was emerging in the four Morgan polls since the 2004 election, compared with Newspoll.
I have now looked at the preference flows from Morgan and Newspoll over the 21 months prior to the 2004 election. The graph for the preference flows to the Coalition (from those who initially voted for a minor party or independent) is as follows.
This graph suggests that for the period from January 2003 to February 2004, the different patterns of flow in preferences between Morgan and Newspoll goes a substantial way in explaining the different two-party preferred polls for Morgan and Newspoll. However, since February 2004, the difference in preference flows offers less of an explanation.
At this point it is worth noting that for most of the election cycle Newspoll actually calculates preference flows based on the flows at the previous election. It does this without polling preference intentions. It is only in the lead up to the next election that Newspoll begins asking people about their preference intentions. In the above data, Newspoll only began asking the second preference question from the poll of 16-18 January 2004.
When Newspoll began polling preference flows in 2004, it was getting broadly similar results to Morgan. This suggests some merit to Morgan’s claim that people actually cast preferences at the ballot box differently to how they say they would when polled. Newspoll’s practice of not polling on preferences for much of the electoral cycle also goes some way to explaining the pro-Labor bias in Morgan. Perhaps it might be more accurate to describe this as a pro-Coalition artifact in Newspoll during 2002 and 2003. Certainly since the 2004 election, a similar artifact has emerged in the Newspoll data when compared with the Morgan two-party preferred prediction based on the stated preferences of those polled. I think Morgan have wisely decided to issue two, two-party preferred predictions: one based on preference flows at the last election and the other based on people’s stated preferences.
(As an aside, for the three polls Newspoll has conducted since the 2004 Federal election, it has used preference flows at the 2004 election to calculate a two-party preferred prediction.)
However, this does not completely explain the pro-Labor bias in Morgan. Morgan also demonstrated a small pro-Labor bias throughout the 21 month period in respect of the proportion of the population giving their primary vote to the Coalition. As the next graph demonstrates, at most times Newspoll found a marginally higher Coalition primary vote than Morgan. Over the period the average Morgan poll result was a Coalition primary vote of 41.7 per cent. Newspoll found an average of 43.4 per cent.
Why this would be the case I am not sure. It suggests methodological differences in sampling and/or the way in which the question on primary voting intention is asked. It certainly begs for further examination
Having considered the above graphs, I have decided I was a bit harsh on Morgan in Thursday’s piece when I said alarm bells should have been ringing with its poll finding that only 22 per cent of the preferences would flow to the Coalition. I now realise that the last Morgan poll before the election had two anomalous factors. The first, as we noted on Thursday, was the lowest preference flow to the Coalition over the previous 21 months. The second was the highest Coalition primary vote over the previous 12 months. I can understand it if the pollsters at Morgan rationalised this result as a cohort deciding to vote directly for the Coalition rather than just preferencing the Coalition. I was similarly harsh on Newspoll. Its 33 per cent was completely consistent with previous poll findings.
Unfortunately, I do not have the data. Nonetheless it would be interesting to look at how ACNielsen and Galaxy allocated preferences to obtain their two-party preferred predictions over the months prior to the 2004 Federal election.
In conclusion, we have come some way in answering the question of a structural difference in the Morgan and Newspoll two-party preferred opinion polls in the lead up to the 2004 Federal election. In part, it stemmed from Newspoll’s use of preference flows from the previous election during 2002 and 2003 whereas Morgan based its preference flows on individual responses. And in part, Morgan consistently found marginally less primary voting intention for the Coalition than Newspoll. This second observation begs a number of methodological questions. Ah, grist for future blogs.