Blogging Senate forecasts and results in the WA Senate re-election until officially declared.

Twitter: @AU_Truth_Seeker


Friday 27 September 2013

WA Senate - Update and BTL vote estimation - 26-9-13

Tonight I provide an update on the close WA Senate count.

In a comment earlier, I promised to determine a "transition matrix" for BTL votes. Using imperfect data, I have done so.


Critical count 1: AJP vs HEMP
If AJP beat HEMP at this stage, GRN/SPORT is effectively locked in.
However, the HEMP lead has shot up to 884 votes. This means that an ALP/PUP outcome is possible

Critical count 2: Shooters vs Christians
If HEMP wins Critical count 1 and if SFP beats the Christians, then ALP/PUP is very likely. Otherwise, GRN/SPORT is very likely.
Ignoring BTL votes from excluded parties, the Christians currently hold a lead of 61 votes (compared to ABC calculator votes of trailing by 352). This means, on ticket and inter-party BTL transfers, GRN/SPORT is slightly ahead, and this is in opposition to the ABC Calculator.

Critical count 3: LDP vs PUP
I previously raised the importance of this stage of elimination and predicted that the calculator margin in this count would reduce. Although assuming all votes as tickets results in a paper thin PUP lead of 362 votes, it is my estimate that PUP actually leads by a highly significant 1662 votes.
Hypothetically, if LDP was to catch PUP, they'd beat PUP to the final spot (and ALP would also be elected)

Critical count 4: ALP vs GRN
There has been some speculation that the GRN candidate could actually overtake the ALP candidate, resulting in an outside chance of a PUP/GRN election. However, I find this highly unlikely. Unfortunately, I have yet to quantify this... yet.

BTL votes - how will these affect CC2?
From the Tasmanian Senate election, of all the eliminations where SFP and Christians were competing for votes, the Shooters out-polled the Christians by a convincing margin of 158 to 75. That is, the Shooters picked up more than double the BTL leakage of the Christians. But, there were a few special circumstances that let this happen:
- There were three religious parties in Tasmania continuing who were competing with Christians for votes - FF, RUA, DLP.
- There were no other parties competing for the recreational right vote in Tasmania

(On the contrary, however, there are probably more christians and less shooters in Tas than in WA).

This situation is reversed in WA, where the Sports Party is competing with the SFP for a similar voter disposition, whereas there are no other religious parties in contention by the time of CC2.

But qualitatively, the "Christians" Party brand is much weaker than the other established brands such as DLP and FF. Also, the Shooters and Fishers Party is a leading brand of recreational right support.

Assessment of BTLs: Preferences will flow more strongly to Shooters than to Christians. This means the Christians will likely require a lead in order to overcome BTL leakage. This leakage may be worth up to 5% (~400 votes) of total BTL votes distributed to this point.

Experimental approach (highly geeky... read on with caution...)

Using published data from the Tas Senate count, I have calculated estimated the likelihood of BTL votes transferring from one ideology to another. For ease, I have assigned all parties to either: Left, Labor, Central, Liberal, Right, Religious. Using known preference flows from 10 different Tasmanian preference distributions, it is possible to estimate the likelihood of votes going from one ideology to another at any point of exclusion. I am using a sample size of 10x10 to create a matrix of 6x6 - obviously this is not going to be without error but it makes a few neat points regardless.

I won't bore you with the details of transition matrices - look it up on wikipedia if you desire. Essentially, a transition matrix defines the probability of a change in state from one category to another. Each column in a transition matrix sums to 1. It is intuitive to apply this methodology to BTL votes.

The 6x6 transition matrix " T " I have estimated is as follows:

Left Labor Central Liberal Right Religious
Left 55% 37% 39% 22% 24% 12%
Labor 12% 0% 13% 12% 10% 9%
Central 15% 33% 16% 20% 19% 14%
Liberal 6% 14% 8% 0% 13% 24%
Right 8% 8% 11% 11% 17% 2%
Religious 4% 7% 12% 35% 16% 40%
(columns may not add to 1.00 due to rounding)

For example, the likelihood of a BTL vote going from a Left party to Labor is 12%. The probability of a vote going from the Liberal Party to a Central party is 20%.

Let N be the 6x1 matrix of BTL votes of excluded parties when broken down by ideology. It therefore follows that TN=V, where V is the outcome of multiple distributions of preferences.

In the WA example, 
4759
0
N =           658                 
0
2087
801

Undertaking this matrix multiplication leads to the following:
3476
969
V =           1357                 
785
799
919
This means, an additional 919 votes will flow to religious parties and 799 votes to Right parties. Given in our situation two right parties remain, this model, when applied alone, implies that the Christians will actually benefit from a net gain over the Christians shooters  of 919 - 799/2 = 519 votes.

However, this is significantly in opposition to my gut feel as outlined above. I will further refine this matrix, and perhaps a slimmed down transition matrix, over the coming days.

Overall WA assessment: Ridiculously close - at this stage I estimate the margin will be within a few hundred votes.

13 comments:

  1. 'implies that the Christians will actually benefit from a net gain over the christians'

    ReplyDelete
  2. in essence what you seem to be saying is that because the Christians are the only religious group left in the count whereas the shooters are competing with Sports then they will benefit more from the flow of preferences from the like of FF.....that makes sense to me on the basis of your matrix however as you point out that is diametrically opposed to your gut feel observation.......my uneducated gut feeling would trust the matrix.......but who knows!

    ReplyDelete
    Replies
    1. Yes - i prefer working with numbers than gut feels. I get nervous if there's no numbers upon which to base my forecasts... My gut differs to yours, but I think I have uncovered a new way to estimate BTL flows. :-) Hopefully I can post this tomorrow or Sunday...

      Delete
  3. TS, what about ballot paper position? Have you tried modelling that. As you know people tend to number parties close to where they just numbered. Could you model that?

    Once you get this going it'll be great for the state elections.

    ReplyDelete
    Replies
    1. Agree that ballot position is important, particularly with mid-high range forecasts. Even I, a quantitative electoral nerd, started with my preferred ~10 candidates, and then did a few "sweeps" Left to Right of increasingly offensive candidates until I'd filled most of the boxes.

      But trying to model this would be crazy! In a quantitative sense, if I was running a regression, I'd want to control for it, but I'd also want to control for another 10-20 factors, which would decrease the power of my regression and probably result in spurious parameter estimates. So, I'd love to, but not sure I could...

      Delete
    2. And, hopefully, we never again even have any state or federal elections just like this one!!

      Delete
  4. Love your work. So the election of two senators may well come down to haw many Fishing and Lifestyle's 196 BTL votes make it over to the Shooting and Fish, and how many of FFirst's 440 BTL votes make it over to the Aust Chrisitians

    ReplyDelete
    Replies
    1. Thanks! These would flow more strongly than others, but all would flow. A 5% capture of the 5207 eliminated BTL "left" votes (incl Socialist, Sex, Secular, Wikileaks, Democrats and Animal Justice) is worth more than a 100% capture of AFLP's (now) 211 votes.

      But as above, I think I have a new path to doing a crazy 10 party preferred BTL vote estimation, so keep watching my site the next couple of days :-)

      Delete
  5. I like the matrix method!

    Similar to the America's Cup, all these little experimental refinements can add up to a better model.

    ReplyDelete
    Replies
    1. Ha! Can't believe they lost. I should have derived a transition matrix for Kiwi choking. :-)

      Delete
  6. I think you have failed to take into account the distortion in the Liberal and NP ticket vote as a result of the method used to calculate the Surplus Transfer and the issue of segmentation in the redistribution of exclude candidates votes.

    I agree that the crucial pivot point is Christians verwus LDP

    I am still not convince that LDP will remain behind PUP, I think PUP have the advantage.

    The value of not notional Green vote is minimal. The Greens are more likely to be subject to BTL voter leakage then the ALP.

    The only way the greens can stay in the count is if Sports are dealt an unexpected time on and remain in the count to win. Without Sports winning the fold up is not in the Greens favour. Too many votes locked in and not enough vote value to spare. Segmentation will determine the extent of the distortion in the calculation of the surplus value and distribution of votes.



    ReplyDelete
  7. Sorry mean Christians vs Fisherman Shooters as the crcial pibot ooint in WA

    ReplyDelete