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

Twitter: @AU_Truth_Seeker


Saturday 21 September 2013

Flip!

One moment you're winning the next you're not. This short post outlines another example of an AEC error.


I'm glad I'm not a Senate candidate. I'm glad I'm not actually barracking for any.

Last night I posted about the increasing vote of the Bullet Train and how things were falling into line for a chance of Shooters or Democrats glory. However, this was apparently a mirage.

Today, the votes have reverted - Bullet Train is down about 500 votes. So how could this happen? Additional vote counting yesterday gave Bullet Train 0.8% of additional vote, increasing their overall percentage from 0.205% to 0.218%. This was significant enough to put them ahead of FF and possibly change the overall order of election.

But the AEC have seemingly gone back to the future, reversing the votes of yesterday and changing the space-time continuum of future eliminations. Great Scott!

In hindsight, Bullet Train taking 0.8% of all additional votes counted appears highly fanciful. Without having running snapshots on a booth-by-booth basis, it is hard to tell where this error occurred. However, I would be curious to know what internal controls the AEC has on fanciful voting patterns. To date in this election we have seen:
- 1003 votes miscoded in Indi (although eventually corrected)
- 700+ votes be accidentally missed from the Coolum Beach PPVC booth and recorded elsewhere (although eventually corrected)
- Other odd Senate misapportionments at various times, including this one (although seemingly eventually corrected?)

I am not of a mind to suggest the AEC is corrupt, a disgrace, flawed, or part of some bizarre military conspiracy. On the contrary, the AEC is a professional organisation that gets exceptional results out of temporary and casual staff, and the professionalism of their staff and transparency of information provision leaves the State and Territory electoral commissions streets behind.

But this is not to say there is no possible improvement. It is clear that the AEC does not have "exception checking" built into their processes. Senior AEC staff should be immediately alerted to any discrepancy between the number of:
- House votes
- Senate votes
- Names marked off the roll
in each and every polling booth.

Further intelligent checks should be conducted to ensure the vote % for parties between Senate and House in every polling booth is in alignment. Where there is a discrepancy outside, say, 2 Standard Deviations of the mean variation for each party, these booths should be rechecked and recounted automatically. It would not be hard to quarantine such votes from even making it into the electronic Virtual Tally Room until they have passed a sanity check.

This would ensure that published results at all times reflect the actual number of votes counted.

15 comments:

  1. Well said! I couldn't agree more. We certainly need "sanity" checks - when figures are first determined, before they are recorded. In fact, I think the AEC could fine-tune this even more than you suggest - examples are booths where either the House of Reps or Senate or both votes are swinging say more than 10% more than the other booths in that Electorate, or certain categories of votes (eg Postal) behaving in a significantly different manner to that same category of the same Party's candidate in nearby demographically similar seats. These types of considerations / formulae would need to be adjusted according to the size of the sample / booth (so that a much larger variation of swing would be tolerated at a very small polling place without triggering a check).

    ReplyDelete
    Replies
    1. Agree, Agree, Agree. I would use data crunching techniques and then enable a vote counting "spare team" of experienced staff in each state to be directed to recount votes that are dubious.

      You refer to a 10% threshold. I would instead use data analysis to "list" the booths by discrepancy and have a hit-list of most likely booths to contain a discrepancy, for both house and Senate. It's a case of diminishing marginal returns so although it would always be worth rechecking every booth with limitless resources, AEC would need to focus on booths most likely to:
      - Contain errors; and
      - Modify the overall

      And guess what - I've been looking at using such techniques in my day job too to improve business outcomes.

      Delete
    2. Excellent suggestion again. For example, the Coolum Beach pre-poll centre would have shown up as the largest discrepancy, in fact it would have gone off the chart entirely. (The figures posted showed more than 8% increase to LNP primary, when for the whole electorate the LNP suffered an 8% decrease, and this in one of the biggest booths - quite apart from the mismatch in Senate votes cast and a huge 19% House of Reps premium to the LNP). But clearly, very small differences could show up as strong outliers in the Senate count, particularly for Micros, which as you have shown can totally flip the whole Senate race on its head. This is really very concerning for the Senate count. And I am pleased you've proved the practical application of your suggestions in your job.

      Delete
    3. Currently having a look at one such electorate. .. Will post any notable results!

      Delete
    4. While not worthy of a standalone post, I'll make this as a comment. The booth of Forcett in Lyons looks unusual and would be the first one I'd recount (Senate and house).
      - 597 house votes
      - 626 senate votes
      - RUA house vote=1.3% (Lyons range: 0-2.2%)
      - RUA senate vote=5.1% (Rest of Lyons range: 0-1.7%)

      This is no smoking gun, it's nothing that will turn the election, but it's one obvious example of where a priority recount should have been done ASAP. Realistically, software should probably have prevented this data being entered into the system on election night two weeks ago.

      Delete
    5. You're absolutely right. The figure for Forcett in the Senate is not credible, only (at a stretch) if the Senate candidate for RUA lives at Forcett, which local AEC official should be aware of.

      Delete
    6. The RUA vote is ridiculous and if candidate lives there might be explainable. But the discrepancy in total votes between Senate and house is not. Are these two discrepancies related? Was it just a typo? 32 instead of 3? If so, then why didn't the software detect this was bogus?
      -TruthSeeker

      Delete
    7. I'm not aware of anything that would cause the Forcett oddity but it's interesting that those RUA votes in Forcett are mostly shown as unapportioned at a stage of the count at which almost all votes are apportioned.

      Delete
    8. With that mismatch between House and Senate votes it looks very much like this is just a data entry error. It should be caught when they attempt to apportion those votes and discover that they don't exist.

      Delete
    9. Wonder if there were any other data entry errors?

      Delete
  2. I have scrutineered numerous times in Senate counts. As a consequence I am very relaxed about the AEC getting things right. I have seen mistakes from time to time, but they always find them, maybe a day later as has happened this time. And maybe it's as simple as someone making a mistake when updating the website. BTW, what does this alteration do to your predictions about the Dems or Shooters & Fishers taking out the 6th seat in NSW?

    ReplyDelete
    Replies
    1. This flip very much increases the likelihood of Sinodinos winning the final seat. Although I think this outcome most accurately reflects the will of the voters, it is still annoying that pundits such as me make predictions based on the available data only to be gazumped when correct data is subsequently loaded.

      Why not have the system quarantine dubious data until such data can be manually rechecked?

      I have a lot of confidence in the aec. But I remain concerned that the two highest profile house seats have seen massive errors corrected - what about the 100 safe seats no pundits have even looked at? What about the Senate which is near impossible to accurately scrutineer?

      Delete
  3. There has been some kind of glitch in the Tasmanian Senate count that has seen new figures overwritten with old figures - sometimes for hours at a time - although it has always been fixed eventually. It seems to especially target postal votes from one or more specific electorates at a time.

    ReplyDelete
    Replies
    1. Yes, hasn't this been annoying! In this case though I'm pretty sure this was a data entry error. The liberal ticket was right next door and Bullet had picked up an additional 500 votes or 0.8% of the daily count, 4x their overall average. This was reversed today.

      Delete
    2. PS - by Truth Seeker.
      (I have trouble logging in and commenting from one of my devices)

      Delete