Dear Phil,
As you may be aware, I run a popular and topical website (originaltruthseeker.blogspot,com.au) and twitter account (@AU_Truth_Seeker) that has been analysing and predicting Senate outcomes nationwide. I am not a member or employee of any political party, nor is any member of my family as far as I am aware. I do not have any financial interest in the outcomes of the WA Senate election.
I write to you today regarding the recently ordered partial recount of voting in the WA Senate. Firstly, congratulations to your organisation for reopening the count. While I am generally supportive of your decision to only reconsider Above The Line (ATL) Votes, I am somewhat concerned at numerous discrepancies within the Below The Line (BTL) votes that requires further consideration.
These issues are as identified by me personally, or as raised with me publicly and/or privately by of many of my readers.
1. Waggrakine Discrepancy
As I posted on my site, there is clear evidence of a number of BTL ballots from the Geraldton - Waggrakine polling place (Division of Durack) being misrecorded as informal. Essentially, the BTL % is impossibly low and the informal % is implausibly high. In total, I expect approximately 50 formal BTL votes from Waggrakine have been incorrectly classified as informal. A scatter plot on my site makes this patently clear.
Your BTL Data Download for batch 1663 (the batch which corresponds to the Geraldton-Waggrakine booth) has just one formal BTL vote, despite at least three people acknowledging they had voted BTL at that place. Further, the "Paper" numbering for this booth lists the only formal ballot as Paper #16, when for almost all other booths the paper numbers start at #1.
Will the BTL informal votes be recounted? Your press release says that informals will be recounted, but BTLs won't. I hope, for the sake of the voters of Waggrakine, that the AEC goes to every effort in an attempt to resolve this glaring anomoly
2. Fernandez additional vote
In your Distribution of Preferences and your overall "Votes by Candidates" summary, you point to Chris Fernandez (Democrats) having 3768 votes, including 507 votes. However, your BTL Data Download only accounts for 506 Fernandez votes. I have confirmed this with a number of other amateur sleuths who have been in contact with me.
Furthermore, your Data Download contains 49429 votes, whereas 49230 votes is required to add up to the total number as reported in various spots on your website.
Where the election is so close, every vote counts and should be available to the public,
3. Senate vs House
I have raised a number of discrepancies on my website between the House and Senate vote numbers by polling booth, While I expect the number to be small for wholly explainable reasons (such as someone stealing a ballot paper, etc) I was quite surprised that over 600 out of 900 polling places in WA had a ballot discrepancy, and 21 polling places had a difference in double figures. Kambalda West and Derby were particularly troublesome with a discrepancy of precisely 50. While this is almost certainly due to someone miscounting a bundled stack of 50 votes, I am concerned you were ready to declare the election without investigating this further. However, I am pleased that these issues will be picked up in the recount and I look forward to this issue self-resolving over the coming weeks.
Has the AEC ever thought of counting Senate ballots by weighing them with electronic kitchen scales? This would enable a more prompt double check of the number of ballots in a particular pile.
4. Recount mechanics
While I acknowledge the time taken for recount may be substantial, I would also support the AEC proceeding in the most transparent manner possible. Accordingly, I believe democracy would be best served by having a clear recount schedule by Division or, preferably, by polling place. There also needs to be a clear way for the voting public to follow the progress of the recount and to be able to track any changes to the count, rather than just by silent uploads to your website.
5. Below the Line non-numerical data
I am concerned, as with many other psephologists, about the quality of the data that has been entered into the system. In particular, I note 29 examples of non-numeric data as follows (the listing below is Batch#"-"Paper#": Non-numeric data type:
29-35: ??
41-129: ??
41-146: ??
111-23: ??
450-5: ??
595-83: ??
668-25: -
776-53: ??
967-107: ??
1006-31: ??
1099-19: ??, ??
1108-51: ??
1121-182: ?
1140-10: ??
1140-25: ??, ??
1198-25: ??
1204-12: ??
1285-2: ??
1309-63: ??
1311-37: ??
1340-79: ??
1357-22: ??, ??
1375-39: /
1455-83: *
1525-28: ??, ??
1542-26: ??, ??
1564-11: ??
1564-35: ??, ??
1778-20: ??
I note there are more ballots here than the difference between the Shooters Party and the Christian Party at the critical point of exclusion. If these discrepancies are neither explained nor resolved, then we will have no choice but to lose faith with the AEC when you expressed confidence that all BTL ballots are correct because they've been independently re-entered by a second officer.
6. Below the Line Zeros
I note in your BTL Data Download that there appears to be at least of zeros. I have used Visual Basic to convert your Download into a useable format, but I don't believe my data manipulation has lead to the introduction of rogue zeros. I would have thought any zero would have resulted in the ballot being treated as informal?
7. Software availability
Finally, there exists a black hole regarding the availability of the software you use to calculate outcomes. Can you please advise whether:
- Software and its code will be downloadable?
- Software code will be made available to official scrutineers
- Software will be made available to interested parties.
On behalf of my audience, thank you in advance for considering my questions and those I am asking on behalf of my readers.
I will be posting and tweeting this correspondence, and intend on posting your response for the benefit of my readers.
Regards,
Truth Seeker
Excellent! Let us please not allow the AEC to pretend these issues don't exist. Note a word was omitted above - "at least of zeros".
ReplyDeleteYou're allowed a few sequencing errors in BTL votes, and you only need to fill in 90% of the squares. Electoral Act s270: "in not less than 90% of the squares opposite to the names of candidates, numbers in a sequence of consecutive numbers commencing with the number 1 or numbers that with changes to no more than 3 of them would be in such a sequence"
ReplyDeleteSo you can leave a few squares blank, or write non-numerical characters in some squares, and the vote can still be formal. I think that these are the cause of zeroes or question marks in the BTL file. (There can also be duplicates.)
I know with certainty the cause of one such case: I deliberately wrote my last two preferences as 987 and 1002, just to see how they'd be recorded in the BTL file. They both ended up as "??".
There are a handful of cases highlighted by D@W in the Poll Bludger thread that I don't understand - "020" and the like. It surprises me that the double-data entry wouldn't catch these cases, but I guess when there are a few million numbers being entered, there'll be a few cases of the same typos.
To add to the list of decrpacince in the BTL data file..
ReplyDeleteI have noted... 020 050 how does the system interpret this and more importantly is this a true transcription of the voters ballot? I have noted that looking at the other preferences the 030 should be 02?
It needs to also be stated that this discrepancies in the ballot has no impact on the crucial pivot point Christians versus Fisherman Shooters
The impact of the distortion in counting the vote (Surplus Transfer value and distribution method of excluded candidate's votes) has a far greater impact
Yeah, we should totally ask them to tally up the votes using a few different methods while they are at it. Then we can all discuss the results over coffee and six-sevenths of a cake.
ReplyDeleteI just can't believe they threw the slice of cake out. Surely someone could have eaten it? But no-one's been talking much about cake recently.
DeleteSeriously though, we owe it to ourselves as a nation to get it right. Let me assure you that weighing Senate ballots is a very effective way to determine the number of ballots in a stack.
As ridiculous as it might sound, in the Idaho Republican caucuses in the USA, voters vote by depositing a special coin in the bucket of their preferred candidate, and then the buckets of coins are weighed.
DeleteRegarding point 5, these values indicate that the voter's entry for this box was illegible. These are still entered into the computer, and it is the computer (not the vote counting officer) that decides whether the vote meets the saving provisions outlined in the scrutineers handbook.
ReplyDeleteThanks Mark. So if a data enterer can't decide what to put in, they do a double question mark?
DeleteHad a look at some cases in the Tasmanian BTLs where a voter put a number that was way higher than the number required. I was especially interested in a strange spike in the number of 88s and thought this might result from other numbers looking like an 88.
ReplyDeleteI found that:
* three of four 88s resulted from the omission of a single number that might look like an 88 (twice 38 and once 28).
* all three 77s likewise (twice 27 and once 47).
It's a rare thing but I wonder if a close inspection of these sorts of ballots might reveal them to be incorrectly interpreted or transcribed, which could cause the votes in question to prematurely exhaust from the system. It's clear what the voter's intention is in such cases but if the number absolutely looks like a 77 and not a 27 it may be it has to be ruled a 77.
Haven't looked at the same for WA yet.
(There were also cases of high numbers that resulted from apparently whimsical voting or deliberate "stamping" by the voter to mark their vote so they could check it against the data file, and there were a few votes with high numbers that were a real mess.)
Interesting. In the AEC ballot formality guidelines, one of their overarching principles is to consider the ballot as a whole, and if the other numbers give clues to the identity of a difficult number to read then this principle should apply. Another overarching principle is to start with the assumption that the voter intended to return a formal vote.
DeleteIn the case where a 38 is missing and an out-of-range 88 is oddly present, these principles should have been invoked. The questionable number should have been interpreted as 38 and the ballot declared formal. Clearly this didn't happen.
I've been looking at related issues in the WA data.
DeleteOf the 49,429 papers in the data set (which I believe to be only formal votes), a surprising 6,238 do not include a complete 1 to 62 sequence. Of these, 2,016 have a single number duplicated and another number (not including 62) missing. The top 20 combinations (Number Duplicated, Number Missing, Ballots) are:
61,60,63
4,41,48
36,37,28
25,35,25
46,47,24
32,33,22
24,34,20
54,55,20
29,39,19
28,38,16
56,55,16
4,40,15
34,35,15
50,51,14
51,50,14
8,18,13
59,58,13
21,22,12
26,36,12
5,51,11
A lot of these are obviously the sort of thing you get from bad handwriting, but could also get from bad data entry, albeit you'd think not if they're being entered twice.
As to whether you could tell the voter's intent, I looked at the 48 ballots that had duplicated '4's and no '41'. For 43 of the 48, I had zero doubt which of the '4's was intended as a '4' (it was next to a '3' or a '5' for a candidate of the same party) AND which was intended as a '41' (it was next to a '40' or a '42' for a candidate of the same party). 47 of the 48 met one criteria or the other.
I'm not in favour of the data entry people trying to guess at the voter's intent. But if I was entering one those 43 papers, there wouldn't need to be more than evidence of a close encounter with a pencil next to the second '4' before I'm seeing the intended '41'.
As an aside, looking at the 43 papers on which intent is pretty obvious, I have 4 that would have exhausted due the multiple '4's prior to the critical Shooters-Christians pinch point. However none of those papers would have sat with one of those parties at the time.
But this is just looking at a single combination of duplicated number / missing number. It's obviously not impossible that there could be a vote or two exhausting early that could have influenced that elimination. However I'm going to say that I'm confident there wouldn't be anything like 14.
Re anon comment 8:46, these are formal ballots we're considering here as informal ballots are not included in the file. So the issue is not the formality of the ballot, but that, for instance, if a debatable 38/88 is read as an 88 then the vote exhausts after preference 37 (should it be distributed to that point.)
DeleteAnon 6:59 is a fantastic bit of work. To me that sort of evidence justifies the addition of a formal saving criterion for cases of apparent duplicate numbering where the natural assumption of locally sequential preferencing points one way rather than the other.
Further to my earlier reply (6:59pm)...
DeleteI realised that I COULD in fact look at all the BTL ballots that exhausted before the critical Shooters-Christians elimination to see if any of them might be simply poor handwriting interpretation and affecting the count.
Essentially I’ve identified the 175 ballots that account for the 152 exhausted votes prior to the elimination of AUC’s Jamie van Burgel commencing in count 140. There are more ballots than votes because some of them had valid preferences that were used to elect someone and have only been exhausted at fractional value.
Most of the exhausted ballots clearly have issues beyond poor handwriting. People skip numbers and end up at 63, or duplicate numbers and end up at 61 or 60. (For example there are a surprising number of 1-2-3-4-3-4-5-6-7-8 … 60 ballots).
For some where handwriting DOES appear a likely issue (one number missing, a similar-looking one duplicated) the numbering sequence doesn’t always offer help on what might have been intended, although a second look at the ballot might. Even where the issue is clear, applying the deduced sequence usually leads to the vote ending up with someone other than AUC or ASP.
However there are some exceptions which I list below in decreasing order of likelihood to be worth looking at:
1074 / 8: Two ‘4’s and no ’40’. One of the ‘4’s is immediately before a ‘5’ and the other is immediately after ‘39’. Making the obvious interpretation, the ballot would sit with the Shooters on preference ‘8’ after count 139.
373 /1: Two ‘3’s and no ‘53’. One of the ‘3’s is immediately before a ‘4’ and the other is between candidates numbered ‘52’ and ’54’. Making the obvious interpretation (albeit a ‘5’ is not the easiest thing to miss), the ballot would sit with the Shooters on preference ‘5’ after count 139.
1205 / 37: Two ’2’s and no ’27’. Complex numbering sequence but there is a loose left-to-right pattern making it reasonably clear that the first ‘2’ (for AUC’s van Burgel) is intended while the second is the ’27’. Would obviously be still be with the Christians at count 139 if interpreted this way.
1203 / 10: Two ‘4’s and no ‘2’. No strong indicator that one of them should be a ‘2’ but maybe worth a look as one of the alleged ‘4’s is for van Burgel and the ballot would sit with him at count 139 (as 0.458 of a vote) if EITHER of the ‘4’s was actually the missing ‘2’.
173 /20: Two ‘6’s and no ‘9’. Difficult to see as a handwriting issue and the numbering sequence is most consistent with both being ‘6’s but would be worth looking, given that one of them being a ‘9’ would see the ballot sit with the Shooters on preference ‘13’ at count 139 rather than exhausting.
808 / 22: Two ‘2’s and no ’60’. Looks like simple numbering stuff-up, but maybe worth a look considering that one of the ‘2’s being something else higher than ‘5’ would see the ballot with the Shooters at count 139.
1106 / 69: Two ‘7’s and no ‘33’. Not an obvious handwriting issue and the numbering sequence suggests one of the ‘7’s is more likely to be another ‘17’ than the missing ‘33’, but even that would allow the ballot to survive to preference ‘8’ and sit with the Christians at count 139.
1434 / 30. Two ‘3’s and no ‘62’. One of many that duplicates a number early and ends at ‘61’ but included in the list because the numbering is all over the place and losing one of the ‘3’s would see the ballot sit with the Shooters at count 139.
All up maybe only the first three are likely handwriting issues where a second look at the ballot paper might be expected to reveal a mark that could change the interpretation, although it could be as few as just the first one. The others are much less clear based on the numbering sequence and most likely represent voter screw-ups but maybe worth a look as they potentially impact the critical elimination. However the biggest conclusion is that there are more potential votes moving to the Shooters than to the Christians, so the margin would be more likely to widen than narrow.