Instantly share code, notes, and snippets.

sandys/MOTA - Honest Voting.md

Last active January 27, 2023 09:08
Show Gist options
• Save sandys/d15e44c23fdfbe71c08308cb601eff8a to your computer and use it in GitHub Desktop.
MOTA (Many of the Above) or Honest Voting - an application of approval voting in India.

MOTA (Many of the Above) is Honest voting - you dont have to fear about vote wasting

This is very easy to describe to voters: vote for everyone you approve of.
MOTA does one thing very well - it removes any reason to vote against your favorite candidate.

Intuitive Explanation

In any voting system, if there's only ONE candidate - not voting and voting for the candidate are the same.
For TWO candidates - it is better to pick the one you like more.

With THREE (or more) candidates, the strategies become more complex. Assume the 3 viable candidates are A, B and C preferred in that order (A > B > C). If you go with approval voting, the 8 possible votes ({none},{A},{B},{C},{A,B}, {A,C}, {B,C}, {A,B,C}) As a voter you have 2 primary goals: that A win and that C lose. Any vote that benefits C or doesn't include A, decreases the chance of BOTH your primary goals.
So of all the vote combinations, only {A} and {A,B} would even be considered. In other words the only strategic choice you have to make is whether to vote for B or not - the fact that you would vote for A is not even in question (you will NEVER be dishonest to your primary choice).
The voter will consider this decision by evaluating pros and cons - the pro of voting for B is that, it increases the chance of C losing. The con of voting for B is that, it increases the change of B defeating A in an election that A would have otherwise won.

Let's take a concrete example - Biden, Sanders, Trump

Let's take an example (note: a>>b>c means i like a much more than b and i like b just a little more than c)
Trump 40 voters
Biden 30 voters
Sanders 30 voters

For 20 people, Trump >> Biden > Sanders (honest vote is {Trump})
For 20 prople Trump >> Sanders > Biden (honest vote is {Trump})
For 30 people Biden > Sanders >> Trump (honest and strategic vote is {Biden, Sanders} if Trump is viable, {Biden} if he isn't)
For 30 people Sanders > Biden >> Trump (honest and strategic vote is {Biden, Sanders} if Trump is viable, {Sanders} if he isn't)

There are two kinds of people that exist now:

1. Hard-lined Sanders voters say "We're not changing our votes, so you should switch your Biden votes to Sanders so Trump doesn't win" and the hard-lined Biden voters say "Cool. Well we're not changing our votes, so you should switch".

2. flexible Sanders voters say "you should also approve Sanders so Trump doesn't win." and flexible Biden voters say "you should also approve Biden so Trump doesn't win."

These are the only two situations that exist. There is fundamentally no situation where a Sanders voter will switch to Biden or vice versa. The condition for dishonesty does not exist. In FPTP 30% of the voters (of either Sanders or Biden) must be dishonest or Trump wins. The best strategic ballot is an honest ballot.

If the Sanders supports were actually more like Sanders >> Biden > Trump, then voting {Sanders} even when Trump is viable is honest. However voting {Sanders, Biden} is also honest and strategic.

So if Trump is viable, who wins between Biden and Sanders? What influences that outcome?

it essentially becomes random i.e. small numbers of voters decide. Trump voters that do have a strong preference. Biden voters that don't like Sanders and Sanders voters that don't like Biden. This drastically raises the stakes for Biden and Sanders voters to bullet vote. It comes down for them to the utility of Sanders vs. Biden over the utility of either vs. Trump. But even so, they have no incentive to be dishonest. The maximum this manifests is both {Biden, Sanders} votes.

The variable is - A group of stubborn, uncompromising voters have a high utility on getting their first choice and a low utility for compromise vs. losing outright. They will bullet vote and influence elections. However, there is zero reason to change vote. The other voters simply have to weigh the utilities and probabilities and give additional votes.

The property of never-dishonest voting is also called "safety against Favorite Betrayal (Also known as Compromise Acceptance Criteria)"

In MOTA, your ballot power can be thought of as: probability that one of the candidates you voted for wins over a candidate you didn't vote for

Dishonest voting - Plurality/FPTP strongly incentivizes Favorite Betrayal (Also known as Compromise Acceptance Criteria)

In traditional voting, for all voters whose favorite isn't polling in the top 2, in order to prevent a spoiler effect a savvy voter realizes that they are better off ranking their favorite candidate lower or just not voting for them at all. This can lead to the actual favorite of the majority losing and the people having no idea that the true favorite even had the needed level of support. Favorite Betrayal Effect can also cause viable parties to seem unviable and other negative consequences, even when the candidates in question didn't have the support required to actually win.

The India impact - traditional voting will INEVITABLY lead to a two party or one-party system

Election methods which fail the Favorite Betrayal Criterion (i.e. you vote for your "second choice" instead of your first choice, to protect yourself from getting a worse choice) are inherently conducive (usually strongly conducive) to two-party domination.

Later-no-Harm criterion.

Approval voting fails later-no-harm

Approving an additional, less-preferred candidate creates direct competition with a more-preferred candidate. Each candidate you approve is pushed forward by the exact same amount. This potential later-no-harm violation is tough to completely avoid because of how approval voting works.
This is the biggest criticism of approval voting (its the technical reason why the aforementioned Majority Criterion is seemingly violated)

That certainly sounds valid criticism, but consider that this will never happen accidentally since political parties will definitely campaign against the notion of useless additional votes. So the only situation where candidates will give additional votes to be safe against their hated candidate winning.

So in a system with a functioning media/internet/information dissemination, later-no-harm is not a concern

Honest/Approval Voting is a Utilitarian Election method - not a Majoritarian Election method. And this is a good thing

By the majority criterion, a candidate X should win if a majority of voters answers affirmatively to the question 'Do you prefer X to every other candidate?'.

It is possible that a candidate whom well over half of voters see as a top choice could lose to someone who nobody sees as their top choice.

The majority criterion applies to situations where a single candidate is preferred above all others by a majority of voters. Therefore, in elections with more than two major parties, the majority criterion is frequently irrelevant. In an election with three or more serious contenders, there is often no candidate ranked first by such a majority.

In fact, every election system in the world has a different view of what "majority means". You can read about this more here

Given that no candidate has an outright majority, a "majoritarian" can be satisfied with the election of any candidate who is not the lose-to-all candidate (i.e. who can be beaten by all other candidates in a 1-vs-1 match).

The probability of Approval Voting electing a lose-to-all candidate is significantly lower than vanilla Plurality (more info here). This is especially true since voters are often dependent on each other for voting. And so the loser situation does not arise.

Simulated example for India

Let's take 100 total voters which reflects voter population of India roughly. This will be comprising of:
30 Hindu OBC
20 Hindu SC
10 Hindu ST
20 General Category
15 Muslim
5 Sikh/Christian

take 4 candidates A B C D
A is loved by Hindus but hated by otherz
B is loved by Muslims but hated by Hindus
C is loved by Hindus but not disliked by others
D is not loved by any bit not disliked by any either

The voting simulator is setup - - https://www.smartvotesim.com/sandbox/?v=2.5&u=1753613628

The names are as such:
30 Hindu OBC - Group 1
20 Hindu SC - Group 2
10 Hindu ST - Group 3
20 General Category - Group 4
15 Muslim - Group 5
5 Sikh/Christian - Group 6

In Approval voting, the victory of C is super stable. . But the most interesting thing is the "POLL behavior" - F+/F. This basically is a pre-election mechanism , wherein a certain candidate polls ahead. So people may choose to get influenced by that. In that case, they may choose to NOT approve of people (but they will still be honest to their favorite candidate).

Interestingly, the simulation indicates that there is only one situation like that - when candidate C actually polls as the frontrunner, but candidate A does NOT. This is the only situation, where all the A-voters may state super-hating C and not approve of C in the polls...causing C to lose.

Monetary Cost

the Utah Association of Counties put together an analysis of ranked choice and approval, which includes an estimated cost analysis for Salt Lake City (slides 25-29). https://le.utah.gov/interim/2021/pdf/00002099.pdf Approval voting had nearly no additional costs compared to regular, plurality voting