Group Voting Ticket process

[Sly (Sly Gryphon)]

I really like that the Pirate Party has such an open process for their group voting ticket, however analysing the data does suggest that it is very heavily influenced by the initial order decided by the preferencing committee.

Of course this is still better than other parties where a committee selects the order and that’s final.

Issue:

This can be seen in the data, when tabulated in final order vs votes by position. There is a lot of spread across the top few parties, as people changed their order; lower down however there is a very high concentration of parties being placed at specific positions, with a saw tooth pattern around them.

Members re-ordered the top few parties, and then scan the rest, occasionally moving a party to the top, and less occasionally moving a party down.

This means the majority of parties remained at their original position, or moved down 1, 2 or 3 spots (from moving 1, 2 or 3 lower parties to the top); occasionally they moved up a spot or two (by moving a higher party down). The pattern is quite distinct.

Spreadsheet Pivot Table

Suggestion:

If there is a desire to remove any initial bias, so the order is entirely member-driven, then some interface changes could accommodate this. Instead of a starting order, the parties could start in an unallocated area, possibly in random order, or with buttons to put in alphabetical order (to easily find parties) vs randomising.

Further, as pair-wise comparison is used for the final ordering, it isn’t actually necessary for members to order every party. After they have selected rank 1, 2, 3, etc (as many as they want), all the remaining unallocated are given the same rank, i.e. below the others in pair-wise, but equivalent to each other.

Effectively, with the pair-wise comparison these parties would be roughly ordered by the number of times they were selected or not.

For a slight improvement, maybe also add a “Do not preference” box, where people can drag parties they don’t want preferenced. In the pair-wise comparison, these would rank equal with each other but below the unallocated parties. Often there are parties people simply don’t want voted for, but it isn’t really material if they are last or second last. The order of these would roughly be based on the number of people who selected them as “Do not preference”.

Effect:

This can only be a guess, but Wikileaks may have got top spot, with Secular Party, Sex Party, Labor, and maybe Future Party doing better, and Fred Nile, Family First and the Coalition doing worse.

[piecritic (Brendan Molloy)]

If there is a desire to remove any initial bias[…]

I’ve never once heard it suggested.

It was designed from the beginning to be a biased proposal from those best placed to make a good decision. We spent probably two weeks comparing every single party against us for policy overlap and real world representation of said policies, and ordered the parties based upon this.

The committee for organising the original order was also open to all members who were interested so it wasn’t a closed committee either. It is likely the National Council will run things the same way again this year.

Further, as pair-wise comparison is used for the final ordering, it isn’t actually necessary for members to order every party.

Yes it is, because a single GVT cannot equally rank parties, so risking ties by allowing equally ranked parties—even though the underlying vote counting is Schulze pair wise comparisons—doesn’t seem sensible.

Instead of a starting order, the parties could start in an unallocated area, possibly in random order, or with buttons to put in alphabetical order (to easily find parties) vs randomising.

Manually ordering around 30–40 parties you may not know anything about nor want to put a lot of research into can cause problems.

Alphabetical order is bad. People will just hit submit and it would introduce a literal donkey vote to the system. That’s not a good property of a well designed system.

I considered starting them in an unallocated area, or otherwise providing support for doing so, but in the end it was too much of a pain in the ass (this is all coded for free) and the tradeoff was minimal.

Instead, I focused on a user experience that made it just as easy to complete this task on your phone as at a desktop computer, as a lot of our members seem to almost exclusively use their mobile devices for web interaction these days.

For a slight improvement, maybe also add a “Do not preference” box, where people can drag parties they don’t want preferenced. In the pair-wise comparison, these would rank equal with each other but below the unallocated parties. Often there are parties people simply don’t want voted for, but it isn’t really material if they are last or second last. The order of these would roughly be based on the number of people who selected them as “Do not preference”.

This would be functionally the same as just leaving the puddle of shit parties in the lower half, without changing their order as originally proposed. If anything, putting them in a “dead pile” only makes the final count less transparent.

Basically, there is nothing in the current user interface that stops you from ordering everything how you please. There have obviously been many trade-offs to ensure the user experience is good for the primary use case, which is shuffling a few parties and mostly leaving the others where they are.

Any major changes can be proposed to the National Council, but I would argue leaving things as they are is the best course of action. It is biased by design and it is no secret that the initial proposal was designed by the committee dedicated to getting us the best outcomes, not by accident.

This can only be a guess, but Wikileaks may have got top spot[…]

I chuckled.

[Mozart (Mozart Olbrycht-Palmer)]

I really like that the Pirate Party has such an open process for their group voting ticket, however analysing the data does suggest that it is very heavily influenced by the initial order decided by the preferencing committee.

Yes, this is by design. The Election Committee (which was open to all members and meetings of which were publicised) used transparent correspondence and policy comparison to decide an order that reflected parties from most to least compatible with Pirate Party policies and principles.

Members re-ordered the top few parties, and then scan the rest, occasionally moving a party to the top, and less occasionally moving a party down.

This implies that apart from a specific party or parties, members do not care particularly about the order of parties at the lower end of the list — or, alternatively, that the order was reasonably within member expectations.

If there is a desire to remove any initial bias, so the order is entirely member-driven, then some interface changes could accommodate this. Instead of a starting order, the parties could start in an unallocated area, possibly in random order, or with buttons to put in alphabetical order (to easily find parties) vs randomising.

No, there was a definite desire to have an initial bias. This is why Pirate-friendly parties were toward the top, and neo-Nazi parties were at the bottom. There were — I believe — just more than 50 parties registered at the last federal election. It is more beneficial, in a pragmatic sense, to have an initial proposed order that has been arrived at in a reasoned way and allow modification of that order.

Further, as pair-wise comparison is used for the final ordering, it isn’t actually necessary for members to order every party. After they have selected rank 1, 2, 3, etc (as many as they want), all the remaining unallocated are given the same rank, i.e. below the others in pair-wise, but equivalent to each other.

Effectively, with the pair-wise comparison these parties would be roughly ordered by the number of times they were selected or not.

For a slight improvement, maybe also add a “Do not preference” box, where people can drag parties they don’t want preferenced. In the pair-wise comparison, these would rank equal with each other but below the unallocated parties. Often there are parties people simply don’t want voted for, but it isn’t really material if they are last or second last. The order of these would roughly be based on the number of people who selected them as “Do not preference”.

Given there can be no ties and every box on the GVT must be numbered, it is my view that equal ranking in this context would be unhelpful. From memory it was difficult for the Election Committee to decide who to rank last, One Nation or Australia First.

[Sly (Sly Gryphon)]

This implies that apart from a specific party or parties, members do not care particularly about the order of parties at the lower end of the list — or, alternatively, that the order was reasonably within member expectations.

Maybe. At the top, there was certainly plenty of reordering, and maybe that is all that really matters, although it was stll heavily influenced.

What I think the data shows, instead of the order being within member expectations, was that any attempt to change preferences was absolutely swamped by those who left the default.

e.g. 45 votes moved the ALP to somewhere in the top 10, however despite that it remained at exactly #34, as those votes were outnumbered by the 51 people who just kept it as is; that means 51 people who by default placed it after the AFLP (#33) and before UAP (#35).

If you compare the votes to parties like the Non-Custodial Parents, Australian Sports Party, etc, then it seems the preference of voting members to place it higher on the list was simply outnumbered by the default order, i.e. the whole vote was useless as you ended up in the order determined by the committee.

Given there can be no ties and every box on the GVT must be numbered, it is my view that equal ranking in this context would be unhelpful. From memory it was difficult for the Election Committee to decide who to rank last, One Nation or Australia First.

The risk of ties would be minimal. There were enough votes that having a varying number of main preferences (3rd, 4th, 5th, etc), and a bunch of equal “unranked”, then a bunch of equal “do not preference” would likely still produce a unique order.

e.g If four people ranked one of the unranked parties, but only three ranked another, then the one with four votes would be ahead on the pairwise comparison. Similarly, if 12 people put Fred Nile in the do not preference box and 9 put Family First, then the pairwise comparison would put FF ahead.

The fact that individual votes have a lot of ties won’t matter once you aggregate the results.

(In fact, even the current system has a chance of a tie, if equal numbers of people offset each other’s votes.)

If you do end up with a tie somewhere in the middle, say between the Country Alliance and the Uniting Australia Party, then it is easy enough to each randomise (sortition) or have the council decide the final order. You still only end up with one GVT, just it is driven by member preferences (which appear to have had little influence on the order from last time).

[Mozart (Mozart Olbrycht-Palmer)]

What I think the data shows, instead of the order being within member expectations, was that any attempt to change preferences was absolutely swamped by those who left the default.

If the vote in favour of leaving it as-is is greater than the vote in favour of changing it, I’m not sure what the issue is.

The risk of ties would be minimal. There were enough votes that having a varying number of main preferences (3rd, 4th, 5th, etc), and a bunch of equal “unranked”, then a bunch of equal “do not preference” would likely still produce a unique order.

This is assuming that people will bother ranking more than a handful of parties. Which may be true. Or it may not be. It would be difficult to know, and not worth the risk in my opinion.

Overall I am still not persuaded that there is an issue to be resolved.