Following on from previous work projecting the 2016 Senate results onto the new Queensland state seats, I thought I might try a non-uniform swing model.
This model predicts the primary votes in each seat D_p (or in this case, four-party-preferreds, which given the number of parties in Queensland amounts to almost the same thing) from:
- The primary vote for the party statewide at the last election S_e
- The primary vote for the party in that seat at the last election D_e
- The primary vote for the party statewide from a poll of your choice S_p
This is done by means of monotonic cubic spline interpolation, between the points (0, 0); (S_e, D_e); (1, 1).
Such an interpolation makes a great many simplifying assumptions about the relationship between the statewide primary vote for a party and the vote it will get in a seat, and I'm not claiming any serious predictive power!
Something I do which is comparatively novel in the Australian psephological space is the Senate-votes 4PP - although admittedly nobody's really had the opportunity to do one of those until last year!
Let's examine the model in more detail. Using the 4PPs we get a base-case for each district. We perform an interpolation for each district, and then evaluate the district primary votes D_p for each party with their new expected statewide primary vote.
From there, we can simulate an election, performing eliminations and distributing preferences in line with what we know from 2016.
Enter predicted four-party-preferreds as percentages:
|Two Party Preferred||—||—|
Be aware also of seat-specific stuff - for example, One Nation isn't standing candidates against some sitting MPs.