Something interesting has been popping up in the polling with the two party preferred vote – there is a statistically significant difference between the two party preferred estimate when poll respondents get to allocate preferences compared to the two party preferred estimate when preferences are allocated on the same basis that they were at the 2007 election.
We have two pollsters that publish these two different ways of allocating preferences to get a two party preferred estimate – Morgan and Nielsen – and in both cases the same phenomenon occurs. There’s been 72 Morgan polls and 13 Nielsen polls since Rudd was elected, so we have a decent set of data to work with – certainly enough to imply that the polls might actually be a little better for Labor than the published headline results suggest.
UPDATE: Nielsen’s headline numbers use respondent preferences even though they measure both , while Morgan and Newspoll use 2007 election preference distributions.
First up, if we run the ALP two party preferred results of each pollster using the two metrics, we get:
In both cases, the ALP two party vote is consistently higher when respondents get to allocate preferences than when we use the preference distribution of the 2007 election. It’s worth pointing out that the ‘headline number’ used when reporting polls is based on the 2007 preference distribution for Morgan and Newspoll, but not for Nielsen.
If we take the difference between the ALP two party results based on how voters say they will allocate preferences and the ALP two party vote based on the 2007 preference distribution for each pollster, it tells the basic story.
If we now transform those raw results into the percentage of time that the voter allocated result is higher than the 2007 preference distribution result, and by how much we get:
Only 12.5% of all Morgan Polls have the ALP TPP based on the 2007 preference distribution higher than when voters allocate preferences themselves (2.8% for one point difference, 9.7% for half a point difference). 19.4% of the time the two measures are the same 68.1% of all Morgan polls had the ALP TPP higher when voters allocate preferences compared to the 2007 preference distribution.
For Nielsen, not one of the 13 Nielsen polls has had the ALP TPP based on the 2007 preference distribution higher than what happens when voters get to allocate their own preferences. It’s also interesting that both pollsters have the same approximate proportion of their polls having a ALP TPP when voters get to allocate preferences rather than using the 2007 preference distribution.
Yet, if we run a scatter and regression on Morgan and Nielsen results where the left axis is the ALP two party preferred poll result based on how preferences were distributed at the 2007 election, and where the bottom axis is the result from the same poll but where poll respondents allocate their own preferences – a noticeable difference emerges between the two.
The dots are the actual poll results, the red lines are the linear regression line of best fit through those poll results and the black lines are where we’d expect the regression line to be were the differences between the 2007 preference distribution and the poll respondent preference allocation essentially random.
With Morgan – the higher the ALP two party preferred vote becomes according to the 2007 preference distribution, the more voters tend to give an even higher vote to the ALP when they allocate preferences themselves. Nielsen on the other hand has a consistent difference regardless of the size of the vote – with 9 polls out of their 13 giving the ALP a two party preferred vote 1 point higher than the published headline figure that’s based on the 2007 preference distributions.
The basic statistics for the difference between voter allocated preferences and the 2007 preference distribution for the ALP TPP come in like this:
The difference between the two TPP measures is, on average, half a point higher for the ALP with Morgan polls and over two thirds of a point higher with Nielsen polls.
The big question here is whether the ALP two party preferred is actually being undercooked a little by using the 2007 preference distribution. If we look at the way preference distributions have been changing at elections for the last 20 years, the ALP is currently on a bit of a preference harvesting upswing. The interesting bit is that the Greens vote basically stood still between the 2004 and 2007 elections (7.2% in 2004 compared to 7.8% in 2007), but the ALP preference allocation still increased – so it’s not just a higher Greens vote that’s causing this Labor preference growth.
Mayhap, the polls are actually a little better for the ALP than the headline figures for Morgan and Newspoll based on the preference allocation at the 2007 election suggest.