Analysing the Poll Bias: Morgan vs. Newspoll Part II
Posted by Possum Comitatus on July 8, 2007
Continuing on from Part I, where we found that Morgan does show higher primary vote estimates compared to Newspoll for the ALP and lower estimates for the Coalition, let’s move onto the question of “by how much?”
First up, let us compare the aggregate monthly primary vote estimates for the ALP and the Coalition that both Morgan and Newspoll have produced since 1996.
As we can see, Morgan tends to produce higher estimates for the ALP and lower ones for the Coalition, however there is an awful lot of variation between the two over the period from April 1996 to June 2007(which is the common sample I used for the two series).Since 2001, Morgan has consistently produced higher ALP primary vote estimations than Newspoll, but only since mid 2004 have Morgan consistently produced primary vote estimates of the Coalition that are lower than Newspoll.
If we take a three month moving average of both Newspoll and Morgan primary vote estimates for the ALP and the Coalition, we can not only remove some of that volatility, but it will even up those split Morgan polls a bit.
I had this theory that a large chunk of the differences between the two polling organisations could be explained in terms of the way each measures the minor party and undecided vote. While that may be the case, I couldn’t find a great deal of quantitative evidence to support it, with most analysis only showing that the minor party and undecided vote interaction could explain between about 15 and 20% of the difference in the two polls across the whole time series.
However, what did become apparent was that since 2005, the larger the size of the minor party and undecided vote as measured by Newspoll could account for about 40% of the size of the difference between the Morgan and Newspoll primary vote estimations. This isn’t that surprising considering the analysis of swinging voter movement over that period.
Now, getting to some numbers. If we subtract the monthly Newspoll primary vote series from the monthly Morgan primary vote series for each party we’ll end up with some idea of the size of the differences in estimation between the two polls for both parties over the period April 1996 through to June 2007.
From this we can say that over the sample period, Morgan has on average estimated the ALP primary vote to be about 1.5 points higher than Newspoll, and estimated the Coalition primary vote to be about 0.7 points lower than Newspoll.
However, if we take the period from January 2000 through to June 2007 and repeat the process, a different pattern emerges.
From this we can say that since 2000 Morgan has, on average, estimated the ALP primary vote to be about 2.3 points higher than Newspoll and estimated the Coalition vote to be about 1 point lower than Newspoll.