Friday, June 28, 2019

Most Tasmanian Senate Votes Were Unique

Over the last week or so I've been looking at some statistics relating to the uniqueness (or not) of Senate votes in Tasmania, and some other aspects of Tasmanian Senate voting.  At the moment I'm only doing this for Tasmania, but it can be extended to other states if anyone else wants to do so.  This article has been rated 4/5 on the Wonk Factor scale - it is obviously out and out wonkcore but the maths is not as tricky as in some of the stuff on this site.

All Senate votes are scanned by optical character recognition and the scans are verified by human data operators.  The AEC publishes files of all formal Senate preference votes that can be used by outside observers to verify that the AEC is getting the right results and computing the count correctly.  This year's formatting of these files is a lot more user-friendly than in 2016.  On downloading the files one can find all the numbers recorded as entered in the system for any vote recorded as formal.  Sometimes this includes both above the line preferences and below the line preferences (if both are formal, below the line takes precedence, an issue I will come to later on.)
One minor change is that ticks and crosses are no longer indicated by special characters, an aspect that was the source of some confusion among the easily confused at the last election.

One of the issues I was interested in was how often below the line votes were unique, and in what circumstances below the line votes weren't unique.   Determining whether a vote is a valid below the line vote in the current Senate system is quite easy - the voter is asked to vote 1-12 but under the savings provisions 1-6 is accepted, so the vote must have each of the numbers 1,2,3,4,5,6 below the line exactly once.  That means a product of six COUNTIFs (or equivalent depending on what system you're using) will equal 1 if it's a valid BTL and won't equal 1 if it isn't.

The next task is to sort the BTLs, which I did simply by sorting smallest to largest with a level for each of the 44 candidates.  After this, any group of identical BTL votes will be together in the list, so the simple test for uniqueness is if a vote differs from both the vote above it and the vote below it in the sorted list, then it has to be unique.

When I did this I found a very high rate of uniqueness for the Tasmanian Senate BTL votes.  81614 Tasmanian Senate votes counted below the line had a unique BTL vote and only 13929 did not.  Add in the above the lines (114726 unique, 141719 not) and overall 55.8% of formal Tasmanian Senate votes were unique at the level they were counted at.  (Some others might have been unique if considering the combination of BTL and ATL numbers, but I didn't do that.)

Early uniqueness

It's also possible to look at what stage a vote became unique at.  Mine did so at 4, but 332 Tasmanians picked a unique 1-2 below the line combination.  For instance of the 19984 voters who validly voted 1 for Lisa Singh, just one voted 2 for the second Nationals candidate Wendy Hilditch.  Every candidate except Jacqui Lambie got at least one #1 vote that was a unique 1-2 combination (Lambie voters gave more than one #2 to every other candidate).

In contrast, there wasn't a single unique 1-2 above the line combination in Tasmania, but only two voters voted 1 Fraser Anning's Conservative Nationals and 2 Greens, and only two voted 1 Citizens Electoral Council 2 Group O (Garland).   In 2016, nobody in Tasmania had voted 1 CEC 2 Arts Party above the line, and there had been six unique 1-2 ATLs including 1 Christian Democrats 2 Flux and five starting with 1 CEC.  In 2019, every possible above the line 1-2 combination happened, which was far from the case for below the lines - for instance eight candidates were not preferenced 2 by any of Nick McKim's 17358 below the line voters. 

Non-unique BTL numberings

The following were the commonest non-unique BTL votes I found in the Tasmanian Senate count:

1. 299 times - 1-6 ALP down ticket, stop

2. 145 times - 1 Singh (ALP), 2-6 down rest of ALP ticket, 7-9 Green, 10-12 Jacqui Lambie Network

3. 133 times - 1-6 ALP, 7-9 Green, 10-12 JLN

4. 130 times - 1-3 Green, 4-9 ALP, 10-12 JLN

5. 96 times - 1-3 Green, 4-9 ALP, 10-12 Liberal

6. 93 times - 1-6 ALP, 7-9 JLN, 10-12 Green

7. 79 times - 1 Singh, 2 Bilyk (ALP), 3 Brown (ALP), 4-6 ALP, 7-9 Green, 10-12 JLN

8. 67 times - 1 Singh, 2-6 ALP, 7-9 JLN, 10-12 Green

9a. 62 times - 1-3 JLN, 4-9 ALP, 10-12 Green

9b. 62 times - 1-3 Green 4-9 ALP 10-11 Sustainable Australia, 12 Animal Justice #1

11. 56 times - 1-6 ALP, 7-12 Liberal

(If you look up these on David Barry's Senate calculator you will often get slightly higher numbers - this is because, for instance, I treat 1-2-3-4-5-6-7-7-stop as different from 1-2-3-4-5-6-stop, but in practice both exhaust at 6.)

Losing information

What is interesting about these 11 common votes is that most of them are inefficient.  Votes 2, 7 and 8 have re-ordered candidates, but the rest have not.  All of the others could have been cast by above the line voting using fewer characters and leaving the voter free to number more boxes and distribute their preferences further if they wanted to.  For instance vote 3 is a 1 in the Labor box, a 2 in the Greens box, a 3 in the Lambie Network box and the voter could now preference another nine parties for free while numbering the same number of squares as they numbered originally.

Some voters may be deliberately using below-the-line voting to cast a vote that exhausts more quickly than an ATL vote that complies with the instructions.  Others might be using it in the false belief that they still need to do so to stop their party sending their preferences somewhere.  But the fairly high number of these inefficient votes is interesting.

This is especially interesting with the 1-6 Labor-and-stop votes (vote 1 in the list above).  Labor ran six candidates in Tasmania, presumably in order to try to minimise informal votes.  In Tasmanian Hare-Clark elections, a voter can get away with voting for just their own major party and stopping, so there may have been a concern with running just four candidates that some voters would vote 1-4 below the line and stop.  However, of these 299 votes with just 1-6 below the line, most (187) had something above the line as well.  In 94 cases it was a lone 1 for Labor, making the ATL and BTL components identical.  In six cases it was an informal ATL vote.  But in 79 cases the voter had voted above the line for Labor and expressed preferences above the line.  Under the savings provisions, their formal 1-6 vote below the line overwrote their above the line, thereby taking away the voter's ATL preferences for other parties.  I haven't yet tried to work out how common this was with other kinds of both-ATL-and-BTL votes.

The rule that below the lines override above the lines is a legacy of the old system in which a voter afraid of invalidating their vote with an error below the line could cast a fallback 1 for their preferred party above the line.  That it can now allow a voter to overwrite their above the line preferences with a shorter subset of themselves is a weird unintended consequence.  A savings provision or BTL vs ATL priority rule that takes care of this fully and fairly to all cases would be quite difficult to formulate, and perhaps the number of votes involved isn't worth it.  For instance an obvious solution would be to use the ATL if the BTL is a pure subset of it, but this would advantage those voters for whom this was true over those who had wanted to juggle their preferences.

Non-unique 1-44 votes

Most of the non-unique votes above weren't all that surprising, but what about non-unique votes that number all the BTL boxes?  These are so rare that an across-the-page donkey (1-44 left to right) was one of the commonest examples, but it wasn't the commonest.  The commonest, occurring 15 times in Tasmania, looked like this:

(Click on it to read the text which is illegible otherwise).

This vote starting with 1-3 Greens, 4 Garland, 5 Singh etc, is clearly left-wing on the whole and has multiple breaks in party sequence and order.  What is also interesting is that it clusters by booth - there is one booth where this vote appeared three times, and three where it appeared twice.  It would be interesting to hear from anyone who knows the origin of this vote.  It may have been the recommendation of a lobby group, or just perhaps it might have been a common output of an online issues-based voting card generator.  (The latter explanation struggles to explain why there was such a degree of booth clustering of this vote.)

(Update: I have found out the answer to this.  It was a "this is how I'm intending to vote" type vote posted on a private Facebook page, starting with one voter who devised a rough order and workshopped it with his wife, and then fine-tuned through the campaign with input from others and based on new information during the campaign.  A further 15 votes, at least, followed the order with minor variations or stopping at 12 or 13.  All 30 similar votes were cast in Clark and Franklin.)

Aside from that one and donkey votes, I found about a dozen other cases of the same 1-44 vote appearing three times or more, and the pattern of vote clustering was fairly obvious there too (either at booth level or in the same postal set).  A specific Australian Conservatives vote occurred five times, all of them in two northern booths.  It is tempting to suspect that these are cases of families co-ordinating how all members of the family will vote, or groups of friends doing the same.  If they were more widespread one might suspect multiple voting by individual voters, but it's very unlikely a voter would vote twice on the day at the same booth.  (Nearly all multiple voting is likely to be caused by voters with memory impairments forgetting they have already voted.)

Hopeless Cases

Voting both above and below the line is a bit of a worry.  I found 148 cases of voters numbering every box above the line and either every box or every box but one, with the BTL overwriting the ATL if both were formal.  Quite a few of these were not even for the same party, and only about a dozen were fully consistent above and below. (I describe a vote as fully consistent if the orders are the same except for the reordering of the ungrouped candidates and of candidates within parties.)

Another interesting thing picked up by Ross Leedham is that some voters are recorded as casting above the line votes that match a party how to vote card but with one number missing.  What is going on here?  Did the voter fill that box in but do so in a hopelessly illegible way?  Did the voter actually skip that box because they did not like that party, unaware that this would cause their vote to exhaust without reaching the other parties behind it on the card?  Is it possible any such cases could be errors in the vote-capturing system?   In the absence of any record of whether squares contained illegible markings, or at this stage of any external audit of the accuracy of the whole vote-capturing process as recommended by the Joint Standing Committee on Electoral Matters in the previous session, it's hard to be certain.  (I was quite impressed with the process as a scrutineer in 2016, but didn't do any scrutineering this time with the Tasmanian result a foregone conclusion.)

Privacy and theoretical coercion risks

There is a minor privacy implication of the increased uniqueness of Senate votes under the new system.  In theory, coercion to vote a given way is useless in the Australian context outside of postal voting, because the ballot box is secret, so the coercer cannot establish how the person voted.  However, in the Senate it is not that difficult to establish whether or not anyone at all voted in a particular manner, so it is also often possible to confirm that there is a specific way that nobody at all (either in general or at least at a specific booth) voted.   So in theory a coercive person could direct a voter to vote in a certain way, and then check the files to verify whether anyone in a booth, or at all, had voted in that way. The coercer could even choose to "sign" the directed vote with an unlikely combination of candidates or an unlikely set of repetitions, not affecting the fate of the vote, to make it more likely that the coerced vote would be unique in the whole state and not just at booth level.  That way if the vote did not appear, they would know the voter had not voted as they directed.

I don't know if there's any way around this other than to hope that this sort of thing just wouldn't actually happen because the penalties for coercion outweigh the completely trivial chance of political gain, and also because if one was going to coerce another voter then doing it through the postal vote system would be easier.  Just publishing votes by state and not by booth wouldn't stop a sufficiently aware coercer from doing it, and chances are if they're aware enough to isolate a vote and verify that it is unique, they're also aware enough to know how to make it unique in the first place.  In theory this could have also happened in the old Senate system, except that there the coercer would have had to specify a full BTL vote for the coerced person to follow.

Also, because the rate of how-to-vote card following for minor parties is so low in the Senate, it might be possible for a party to use knowledge of the booth a voter voted at to verify that that voter did not follow the party's how-to-vote card.

No comments:

Post a Comment