Sunday, 23 March 2014

Score sheet

With Bob Such's sudden illness, the pressure on Geoff Brock to deliver a stable government left only one realistic option. To side with the Libs would leave the state indefinately poised with a 23-23 seat split between the Labor party and the Liberal-Brock coalition until Such sould either declare his intention or remove himself from office.

Today's announcement that Geoff Brock will deliver governemnt to the Labor Party has finally resolved the political deadlock in South Australia. Additionally, with more than a week since the vote, most of the seats SA can be called with confidence enough for me to consider publishing the results of my predictions.

Although the Tasmanian Legislative Assembly relies on a progressive counting system that requires more time to calculate, I am prepared to announce my results there too.

South Australia:

Under the old scoring system -- requiring over 90% accuracy to qualify as a decent prediction -- we did well in both houses of the South Australian parliament. In the lower house we got 91.1%, and in the upper 90.9%.

However, after the atypical federal election last year we altered our methodology to account for the general predictability of the election -- speciffically my success must be measured against a baseline determined by the polling. If I am doind worse than consulting the general polling something is wrong. If I am doing better, the question is how much better.

interestingly, despite a clear Liberal lead early in the campaign, the TPP polling prior to election day was where it was prior to 2010 (ALP 48%, Lib 52%). Thus the baseline predicts no change in seats.

This table indicates the baseline, my predicion, and the results:

I called four electorates incorrectly. The baseline called four seats incorrectly. However, because the baseline did not have two tossups, giving it a 91.5% success to my 91.1%.

However, the polling data for my predictions were based on data two weeks older, under which system the Libs were polling two percentage points higher and the baseline would have given a 93.5% accuracy.

So that's not great.

In the senate I'm still trying to develop a baseline system, since upper house polling is rarely conducted.

It seems all of my predictions were spot-on, however, except for the last seat which is always a devilish one, since it is more to do with scraping together 10th and 20th hand votes. Basically every now and then another party keels over and, like any good fantasy adventurer, the other parties loot the body and eventually the last one standing is crowned king. Normally this is the annoying useless kid who was obviously concealing the unheard of, untapped magical reserves that reaveal his true identity as the chosen one of prophesy. The same is true of politics, except there are thirty harmless kids who are backed by thirty seperate prophecies.

This time, however, the mighty good/evil warrior looks posied to do the sensible thing and not die heroicly/idiotically to clear the way for the little guys.

Although Powerful Communities did well on the preference swaps, that is of no real value if you are the first grouped candidate to drop out, which they were. At the moment it looks like the last place will go to the ALP, but we won't know until the count concludes.

In the absence of any better measure of success, we'll have to settle for a result of greater than 90%.


The count is still continuing in Tasmania, wiht two seats still well in the unknown basket -- Lyons 5th is looking to be a close run between the ALP and the Greens, and Braddon 5th has Palmer united 115 votes ahead of the Greens and 375 ahead of the ALP on first preferences. The high ALP-Greens trade is likely to cost the PUP, but stranget things have happened.

Expected results
This is where things go really bad for us. Not only have I not developed any reliable system of predicting multi-candiadte election processes (the SA upper house was a lucky result), but the predictions were formulated over a couple of hours. Unsurprisingly, we lost a few here:

Predicted results
Bass: 5/5
Braddon: 4/4
Denison: 4/5
Franklin: 4/5
Lyons: 4/5 (expected)
TOTAL: 21/24 (87.5%)

Compare that with our baseline:

Bass: 5/5
Braddon: 4/5 (possibly 5/5)
Denison: 4/4
Franklin: 4/5
Lyons: 4/5 (possibly 5/5)
TOTAL: 21/24 (possibly 23/24 = 95.8%)

At best we equal the baseline and get a neutral score. At worst, (if Lyons and Braddon both elect a Green) the baseline scores over 95% -- a brilliant score in the old system and more than 8 percentage points ahead of us.

So, this new scoring system is givving us a pretty bad thrashing. however, it is important to realise that the simple application of polling is currently out-predicting us. This illustrates the usefullness of the Mackerras pendulum.

The aim here is not to create a tool that performs better than the pendulum. That would be a herculean feat for a professional and a near impossibility for an amateur. Instead, I am trying to develop a suite of tools, each less useful than the pendulum but providing a more accurate prediction when used in conjunction with each other.

Clearly there is still a way to go.

No comments:

Post a Comment