Nielsen came out this morning in Fairfax showing a primary vote running to Labor 44/43, for a two party preferred of 53/47. That’s quite a change since the last Nielsen in March ago which ran 47/37 to Labor for a TPP of 58/42. The poll comes from a sample size of 1400, giving an MoE around the 2.6% mark.
Movement or noise is the question – it could be either – and we’ll run the Pollytrend numbers tomorrow to get the answer. Essential Report comes out later today and Newspoll is delivered in the morning which will give us a good set of data to work with.
Meanwhile the primary and two party preferred charts for Nielsen this term have come is as:
Unlike some news organisations are reporting – that would be you Sky Nooz – this isnt the best result for the Coalition since the last election, for that award went to the Nielsen poll immediately after the Turnbull ascension in September last year, where the primaries were 41/42 in the Coalition’s favour, washing out into a two party preferred of 52/49 to Labor. So this is actually the second best polling result for the LibNats.
Nielsen also ran some supplementary budget questions, the full results of which look like this:
Finally, something we haven’t done in a while – the approvals and Preferred Prime Minister history for Nielsen.
Looking at the commentary around the traps, it’s good to see polls being reported more often with their margin or error being mentioned. However, while that is a marvelous first step and is to be applauded, there are still a few mistakes being made – mostly about how to intepret changes in polls. For instance, Peter Hartcher this morning went:
Today’s poll appears to confirm the expectation of a senior Liberal that the budget would mark the return to “normal politics.”
If so, power in Australia is once more contestable rather than being a Rudd monopoly.
It might be an aberration, but two factors give some confidence that the poll marks a threshold, not a detour.
First, most of the movement in the figures is far bigger than the poll’s 2.6 per cent margin of error.
In the case of Rudd’s approval rating it’s a walloping 10 per cent drop.
The margin or error between any two polls is, in fact, larger than the margin or error of each of those two polls since each poll has it’s own uncertainty distribution. Let’s say, for example, that there were two consecutive polls with MoE’s of 3%. The first was 56, the second was 52. Since the latter poll of 52 is outside of the plus/minus 3% of the first poll, that doesnt actually mean that the movement was statistically significant, because the latter poll also has a plus/minus 3% attached to it. The first poll was, with a 95% confidence interval, somewhere between 53 and 59 with a most likely value of 56. The second poll was somewhere between 55 and 49 with a most likely value of 52. Even though the second poll was different to the first by an amount more than the margin of error of the first poll, the two polls still shared the potential values of 53, 54 and 55 across the two periods. As a result, the movement wouldnt be statistically significant at our normally accepted 95% threshold.
The hows and whys of the math gets a little complicated, so I might try and build a widget for people to enter figures into for a Nerdy Sunday post down the track that gives you an estimate of the probability of the change between any two poll results being real, or being noise. Afterall, my name is Possum, I’m from Queensland and I’m here to help – as the saying goes 😀
I haven’t updated the polls page or the polling averages yet, I’ll do that later this afternoon when Essential Report comes out.