This document described in details the rules options for ConcreteSTV. Allowed options are presented in bold in the text below.
The federal rules have surplus distributed among all ballots, continuing or not, with one shared transfer value. The legislation changed significantly between 2013 and 2016, removing formal party tickets and changing formality requirements, but those get dealt with in this system before these count rules apply. The legislation appears ambiguous in various places to me, in particular when rules 273(17) or 273(18) are applied. I interpret both as after all exclusions and surplus distributions are finished. I don't claim this is more reasonable than other interpretations. I also have a variety of interpretations of the rules the Australian Electoral Commission (AEC) actually used in recent elections. I cannot find out what rules they actually used as the source code of their program to count them is a tightly held secret, but one can make a good guess by looking at the provided distribution of preferences. See Our Report for details of the differences between the below specific rules.
The rules actually used in each year are a secret and the AEC has persecuted those asking for the source code of the counting algorithm under FOI, so the actual rules are my best guess.
- FederalPre2021 My interpretation of the Australian Federal Senate election system prior to the 2021 legislation changes. Used to be called Federal but renamed as the legislation changed.
- FederalPost2021 My interpretation of the Australian Federal Senate election system post the 2021 legislation changes. Changes are to tie resolution and the removal of bulk exclusion, making it essentially AEC2016.
- FederalPost2021Manual My interpretation of the Australian Federal Senate election system for manual counting post the 2021 legislation changes. Same as FederalPost2021 except bulk exclusion is still included, making it essentially AEC2013.
- AEC2013 This is my interpretation of the rules actually used by the AEC for the 2013 and 2014 election. It is very similar to FederalPre2021 except when resolving 3 way ties by looking at prior counts, any difference is used as a discriminator, instead of requiring that each has a different count. Evidence: Vic 2013, count 60. I assume 2014 is the same in this respect as 2013 and 2016, where there is a similar case.
- AEC2016 This is my interpretation of the rules actually used by the AEC for the 2016 election. It is very similar to AEC2013, except the Bulk Exclusion rules are not applied (evidence : this crops up frequently).
- AEC2019 This is my interpretation of the rules actually used by the AEC for the 2019 and 2022 election. It is very similar to AEC2016 or FederalPost2021, except rule (18) is applied after determining who to exclude but before transferring any votes (evidence 2019 NSW, count 429 and 2022 QLD count 266).
Update: A personal communication (reproduced with permission) from the Electoral Commissioner clarified some of these interpretations. Note that the personal response incorrectly claimed that our report "validates the accuracy of the Senate election outcomes", which is the opposite of our view, and "this is a more appropriate interpretation as acknowledged in your paper". I don't know exactly what that was intended to mean, but we certainly don't think that what they did was an appropriate interpretation of the legislation. I will try not to put words into his mouth in this interpretation of said letter. He stated something difficult to interpret about the tie elimination, which I believe confirms that they do it the way we believed, rather than the way specified by the legislation. He also verified our interpretation of the bulk exclusion rules (that they are compulsory), and that rule (18) should not cut short an exclusion. He did not comment on the fact that that was not what the AEC did for either in 2019, or for the bulk exclusion in 2016. We interpret his interpretation of rule (17) to be our interpretation for surplus distributions, but he did not mention the (very likely to change an election outcome) priority of exclusions with respect to rule (17). He did not comment on whether any of these would be fixed for the next election, or whether they would adopt any of our recommendations. We are seeking clarification.
Update 2: The resolution has largely been to change the legislation (Dec 2021) to match what the AEC did. The tie breaking for exclusions has been changed to what the AEC did (which is arguably better than what the legislation had required), and the horrible Bulk Exclusion rules no longer apply when the count is done by computer (but do apply when the count is done by hand, which could produce different results). The error in the AEC's handling of rule 18 is not addressed in the legislation. The legislation does include a clause requiring some checking of the listed votes against the paper ballots, which is a great thing as it will encourage the AEC to actually provide some meaningful evidence that the election result is correct. A more detailed discussion is available. This has cause the creation of ruleset FederalPost2021 and FederalPost2021Manual.
The ACT Legislative Assembly is elected by STV with a generally well written and minimal set of rules, with surplus distributed amongst continuing ballots in the last parcel. A rule for restricting the resulting transfer value from exceeding the transfer value in the last parcel can lead to votes effectively set aside; such votes are counted by ElectionsACT as lost to rounding, which is harmless other than being mildly confusing. I have emulated this behaviour. In 2020 the legislation changed to count votes to 6 decimal digits instead of as integers. This introduced a (probably unintended) problem in the legislation as a surplus was constrained to having at least 1 vote above quota. This made sense and was equivalent to greater than zero when counts were all integers, but was unsatisfying with fractional votes - what should one do with a candidate who got 0.5 votes above a quota? ElectionsACT investigated this question in depth and concluded (sensibly IMHO) that anything above zero was actually intended to be a surplus. This seems consistent with the spirit and implied intention if not the literal wording of the legislation, so I have adopted the same behaviour. They did however also introduce a variety of new bugs at the same time which we pointed out. In March 2021 ElectionsACT quietly changed the distribution of preferences on their website having fixed the bugs we reported. This leads to different rules needed for 2020 and 2021.
Prior to 2020, ElectionsACT made their counting code publicly available (and it was unusually good quality). In 2020 they made it a secret (just in time for the bugs).
- ACTPre2020 : This is my interpretation of the rules used by ElectionsACT for the 2008, 2012, and 2016 elections. It seems to match the legislation well.
- ACT2020 : This is my interpretation of the buggy set of rules used by ElectionsACT for the 2020 election.
Use this ruleset to match the now removed original 2020 election results.
It differs from ACT2021 by emulating the following bugs:
- Round votes to nearest instead of down. (Generally small effect, but it allows negative votes to be lost to rounding, and thus for more than the allowed number of candidates to achieve a quota. Acknowledged by ElectionsACT and fixed in 2021.)
- Round transfer values to six digits if rule 1C(4) applies. (Like previous, except larger effect. Acknowledged by ElectionsACT and fixed in 2021)
- Count transfer values computed in rule 1C(4) as having a different value to all other transfer values with the same value. (Big effect, as it can change which vote batch is the last parcel. Denied publicly by ElectionsACT but still fixed in 2021.)
- Round exhausted votes to an integer when doing exclusions (instead of 6 decimal places). This can't change who is elected, just the transcript.
- Surplus distribution is completed even after everyone is elected. This can't change who is elected, just the transcript.
- ACT2021 : This is my interpretation of the fixed set of rules used by ElectionsACT to recount the 2020 election in 2021. It differs from ACTPre2020 in counting votes to 6 decimal places. To match the results currently (as of March 2021) on the ElectionsACT website use ACT2021 ruleset rather than ACT2020.
The NSW local government elections use somewhat random selection for surplus distribution. This means that rerunning the count with different random choices can significantly change the outcome (often changing who is elected). The NSWEC does not provide their choices, so attempting to reproduce their exact outcomes is not practical.
The legislation is very ambiguous, and my implementation is instead based on a specification the NSWEC produces, Functional Requirements for Count Module.
- NSWECRandomLGE2012 My interpretation of the rules used by the NSWEC for the 2012 local government election. Same as NSWECRandomLGE2016 except it sometimes computes the last parcel incorrectly (see our report)
- NSWECRandomLGE2016 My interpretation of the rules used by the NSWEC for the 2016 local government election. Same as NSWECRandomLGE2017 except it sometimes gets tie resolutions for exclusions and fractions incorrect (see our report)
- NSWECRandomLGE2017 My interpretation of the rules used by the NSWEC for the 2017 local government election.
The NSW local government election count algorithm changed prior to the 2021 election as the old legislation was probabilistic, and frequently gave different results when counted multiple times.
The new legislation is very ambiguous, and while I have made an interpretation of it, I would not say that my interpretation precludes just as valid other interpretations which are likely to result in different candidates being elected. After writing my interpretation, the NSWEC published the transcript of distribution of preferences, from which I was able to work out my interpretation of their interpretation, which is now also in the above link.
Note that rule (7)(4)(a), ... The surplus fraction is equal to the resulting fraction or (if the fraction exceeds 1) to 1,
is unambiguous but is not reasonable, as the resulting fraction may be negative, which causes all sorts of problems.
This situation did not happen to come up in the 2021 election, so I cannot say whether the NSWEC chose to
obey the literal legislation or to set the surplus fraction to 1 if the resulting fraction were negative.
For this reason I have make rules that cover both cases.
- NSWLocalGov2021 My interpretation, for what it is worth
- NSWECLocalGov2021 My interpretation of the rules used by the NSWEC for the 2021 NSW local government elections, assuming they didn't take (7)(4)(a) literally
- NSWECLocalGov2021Literal My interpretation of the rules used by the NSWEC for the 2021 NSW local government elections, assuming they did take (7)(4)(a) literally. This can result in candidates getting negative tallies or more candidates than there are vacancies going over quota.
The NSW legislative council elections use somewhat random selection for surplus distribution, very similar to the old NSW local government legislation. This means that rerunning the count with different random choices can significantly change the transcript. The NSWEC does not provide their choices, so attempting to reproduce their exact outcomes is not practical.
The legislation is very ambiguous, and my implementation is instead based on a specification the NSWEC produces, Functional Requirements for Count Module.
- NSWECRandomLC2015 My interpretation of the rules used by the NSWEC for the 2015 local government election. Same as NSWECRandomLC2019 except it sometimes computes the last parcel incorrectly (see our report. The situation where this error occurs does not crop up in the official count, so it is not clear whether this bug was actually present in the program used by the NSWEC, but the bug was present in the pseudocode for the documentation for the specification of the program at the time, so my best guess is that it was.)
- NSWECRandomLC2019 My interpretation of the rules used by the NSWEC for the 2019 and 2023 legislative council election.
The Victorian rules in the Electoral Act 2022 are relatively well written and mostly unambiguous, although 114A(28)(c) conflicted with 114A(12)(b). The VEC seems to have (reasonably) come down on the side of 114A(12)(b), and indeed the legislation changed in 2018 changing 114A(28)(c), fixing the conflict and confirming the VEC's interpretation.
Unfortunately the VEC does not publish full vote data, and so it is difficult to verify their count. The 2014 data matches the output of the Vic2018 rule set for the regions not updated in 2015.
- Vic2018 My interpretation of the rules post the 2018 conflict resolution (and a reasonable if not literal interpretation of prior rules).
The WA rules are relatively straight forward, although there is some ambiguity about when the election ends, and the division of transfers during a surplus is very ambiguous. See the source code for more detailed comments.
Thanks to Yington Li for some conversation about this ambiguity; errors are my own.
Unfortunately the WAEC does not publish full vote data, and so it is difficult to verify their count. Even worse, while their website is quite user friendly, most years they only publish a summary of the distribution of preferences that is insufficient for me to work out what they did. However the 2008 distribution of preferences is sufficiently detailed to be able to check most of the ambiguities.
- WA2008 My interpretation of the Western Australian Legislative Council rules consistent with the 2008 published official distribution of preferences.