The world needs statisticians and economists, particularly in these troubled times. But on occasion they get so carried away with their graphs and formulae we find them lodged up their own orifices. This fairly describes Possum Comitatus’s position after yesterday’s spray at the “MSM” for believing the Queensland opinion polls got it wrong.
Possum wrote that “polls can’t get any more accurate than these [the two final week opinion polls by Galaxy and Newspoll] were”. Of course they can: they could have said 51.4% for the ALP (or whatever the final two party preferred vote turns out to be) instead of below 50, as both of them did.
The margin of error with a 95% confidence interval does not mean the “true” result is just anywhere within the range; it is much more likely to be closer to the centre than near the edges.
Okay, that sentence was hard going. It means that calling an opinion poll result that is 2.5% away from the actual result as “accurate” as one that got it exactly right is to favour theory over practicality. (Probably most statisticians agree with Possum — but see the first par above.)
Anyway, that 95 confidence interval is just something statisticians decided upon and is arbitrary. It means we want to be 95% sure of the result. If we wanted to only be only 60% sure the margin of error would be smaller.
If we were just talking about one opinion poll versus the result, Possum would have a point. We can’t expect too much from one survey. But it wasn’t one, and it wasn’t two, it was six polls over the four-week campaign. None had Labor getting over half the two party preferred vote. We can’t take each in isolation; the “accuracy” accumulates as we get more data.
Those six polls gave Labor two party preferred support of 50, 49, 49, 49, 49 and 50. If we put them together, we get about 49.5% from a total sample of around five thousand respondents, which according to my formula means a margin of error of about 1.4%. The actual ALP vote looks to be easily over 51%, QED the polls over the campaign were wrong.
Possum’s Crikey colleague Mark Banisch is probably more on the money in reckoning there was a swing to the government in the dying days (although citing the last Newspoll as evidence is a stretch).
Both the final week surveys indicated huge swings to the LNP in and around Brisbane and modest ones (in aggregate) elsewhere in the State. (And yes: adding these two surveys together gives decent-sized component samples). In the end it seems the modest non-Brisbane swing eventuated, but the capital barely moved. (This table roughly shows seats in order of swing.)
If Brisbanites told pollsters they would rush into the LNP’s arms but baulked at the last moment, that is the interesting story.
But in the meantime, it behoves all of us to help Possum out. Everyone ready? One two three — pull!