This afternoon I received a tip-off from an anonymous, and also politically independent, amateur data cruncher who works in a non-political field. His/her analysis was maximally good, so let's call him/her "Maxine"
Following analysis of the BTL file, we have reached a number of interesting conclusions. The batch numbers in the AEC BTL file represent booth numbers, and batch 1663 represents our Geraldton Waggrakine booth. Whereas most of the batches start from 1 and sequentially count upwards (perhaps with the odd gap) but to over 150 in some cases, the BTL batch for Waggrakine unusually starts and ends at 16 - a vote for Louise Pratt that stays there.
As advised previously, Waggrakine seems odd. As stated currently, Waggrakine is seeming odder.
Maxine's suggestion, which I tend to agree with is:
-All votes are numbered
-All votes are given a unique ID based on polling place and number.
-All votes are effectively coded (bar code? other unique ID? etc)
-All votes are entered into the system twice, or three times in the event of error.
Can anyone who's recently scrutineered such a process confirm this is what happens?
For some reason, there were seemingly at least 16 votes "coded" in Waggrakine (I suspect 50 in total) of which almost all ended up as informal. This is so counter to the State wide and electorate wide trend to be NOT caused by chance or random variation alone.
However the AEC data dump only shows one vote being actually counted. This vote has a few numerical errors (repeated 13 and a blank) so I wonder if this had anything to do with it? Unlikely but we don't know.
Missing Democrat vote
Last week, I encountered an odd situation where I thought my VBA data manipulation of the BTL CSV file had was failing to pick up the first or the last line in the CSV file. This was driving me batty because I just couldn't find where I was missing one Democrat BTL vote (for Fernandez). However, Maxine has independently verified that the BTL file is missing this Democrat vote. At the high level, my calculation of BTL votes, based on the BTL file, perfectly matches the figures reported by the AEC apart from the Fernandez discrepancy.
Both Maxine and I get 49429 votes from the BTL file but the AEC website reports 49430 votes with an extra Fernandez vote (507 instead of 506)
I acknowledge it's just one vote, but it does call into question the accuracy of the count. Is the AEC aware of this discrepancy? How could they publish a data file of a different size to the number of ballots figure on its website?
Alas, I could not locate which polling place the missing Fernandez vote was included in. If anyone wants to go looking for it, I'd love to know - leave a comment or alternatively send me an email. Batch numbers are usually alphabetical by polling place and electoratel
Unfortunately this continued data sloppiness is eating away at my confidence in the announced results. A recount would help fix the uncertainty.