The O in OPV: estimating two party preferred

Some folks reckon the concept of two party preferred vote is of limited use in Queensland, because of optional preferential voting and, at least in the past, a large right of centre non-major force. I reckon talking about state-wide two party preferred still makes lots of sense, but the concept of swing less so.

As my university's Chancellor notes in the SMH, One Nation and Family First are now 'pretty much off the radar'. The return of right-wing votes to the LNP increases their primary (and two party preferred) support without them having to lift a finger. (Alot of this already occurred at the previous election. If Qld had had compulsory preferential voting, the movement to non-major conservative parties wouldn't have affected the coalition two party preferred vote much in the first place.)

Anyway ... a reader asks how I arrived at my conclusions about Tuesday's Qld Newspoll. Like this:

As you know, Newspoll estimates two party preferred support from their primary support numbers, using preference flows at the most recent election. But it is unclear that they are really handling the optional bit in OPV well.

Using Tuesday's Newspoll, which had Labor with 42 percent of primary vote, LNP 41, Greens 7 'others' 10.

Let us assume Greens flow 80% to Labor and 20% to LNP, and that 'others' flow 50 50. But this is OPV, and it is reasonable to believe that about half the non-major votes will exhaust. (This last assumption is rather rough; party by party exhaust rates would be better.)

So Labor after preferences is 42 + .5x.8x7 + .5x.5x10 = 47.3
and LNP after preferences is 41 + .5x.2x7 + .5x.5x10 = 44.2

But these don't add up to 100. We turn them into percentages: Labor's is 47.3x100/(47.3+44.2) = 51.7, which rounds to 52, and the LNP's is 48.

So this gives us 52 to 48.

But remember that Newspoll's published numbers are rounded. We don't know the unrounded numbers, but they may account for Newspoll's published 53 to 47. We give Labor maximum benefit of the doubt by assuming those unrounded numbers are 42.5 (really 42.49999), 40.5,7.5 (really 7.49999) and 9.5.

Then Labor after preferences = 42.5 + .5(.8x7.5 + .5x9.5) = 47.9
LNP after preferences = 40.5 + .5*(.2*7.5 + .5*9.5) = 43.6

Percentage wise, Labor has 52.3, which still rounds down to 52. We still get 52 to 48.

Now, of course it is silly to so emphasise a single poll, particular a smallish one. But if we are going to throw around a headline two party preferred, we might as well make it realistic as possible.

[See Antony Green on Hanson and preferences.]