One reason is efficiency, if you have the machines counting you get results faster and then you can audit only a random sample of machines records and get statistical guarantees of the election integrity. (The machine can't know before that it will be audited so if you test thousands of machines and find no discrepancy it's highly unlikely a significant portion of the others did cheat)
A second is that it allows one more level of trust. With a paper ballot you have to trust that the poll workers are going to notice/stop/not help ballot stuffing etc. In most cases that's a good enough guarantee, but with voting machines you can also trust the people who programmed audited it
If the poll workers are trustworthy, because of the paper trail, you don't have to trust the auditers of the machines, but if you don't trust the poll workers then you can gain a modicum of trust from the auditers
This link talks about the benefits in the indian election, specifically a software lock on the amount of votes per minute that can be cast
It also shows that in 2013 they audited "only" 20k machines out of almost millions yet found no discrepancy. Statistically that's probably good enough if the choice of machines audited was random
One last way it might be useful, though this is an abstract scenario, is if you want to deploy a more complicated to count voting system to large areas. In some voting systems counting can't be done in parallel, you need to do one round of counting, wait for everyone else and then do a second round and so on.
Similarly some voting systems might benefit from a digitized interface (for instance ballots in austria look like this: https://cdn1.vienna.at/2013/09/zettel.jpg and there are systems which would lead to even larger forms) which outputs a paper trail with only the actual information the voter inputed (in the case of the ballot I showed it'd pretty much just show a party name and a list of candidate names)
In these cases a machine outputting an auditable digital model of the votes cast would greatly simplify counting procedures. You could have every polling station just publish a signed file with the votes in their station and everyone could run the election algorithm
The votes can then be audited same as in the indian system
> The machine can't know before that it will be audited so if you test thousands of machines and find no discrepancy it's highly unlikely a significant portion of the others did cheat
> It also shows that in 2013 they audited "only" 20k machines out of almost millions yet found no discrepancy. Statistically that's probably good enough if the choice of machines audited was random
So how do you protect against the corrupt actor manipulating every machine except the percent that will be checked?
> With a paper ballot you have to trust that the poll workers are going to notice/stop/not help ballot stuffing etc
There's a simple and straight forward solution: Have people from opposing political parties count the vote and check each others result. If you have members of the far left, far right and everything in between sitting there, checking each others results, theres about a 0% chance any vote will be miscounted.
> In these cases a machine outputting an auditable digital model of the votes cast would greatly simplify counting procedures. You could have every polling station just publish a signed file with the votes in their station and everyone could run the election algorithm
There is no difference between a machine publishing the counted votes and the poll workers publishing the counted votes, is there?
India checks 5 machines randomly selected in each voting "segment", no clue what that is but it's smaller than a constituency so I guess it's a polling station
From what I understood that's done locally right away after polls close. So that means that you'd need a way for the conspirators to identify the non tampered machines and for them to randomly select exactly those machines and you'd need them to be present in almost all polling stations and no one to challenge the choice
At that point it's essentially the same level of trust needed for paper ballots
> Have people from opposing political parties count the vote and check each others result
Yeah, but apparently that particular attack vector was a problem in India and this feature helps mitigate it, electoral solutions need to account for the practical situation they're in. For instance I don't support any form of electronic voting machines in my polity because the electoral system works fine with paper ballots and there aren't issues of ballot stuffing
> There is no difference between a machine publishing the counted votes and the poll workers publishing the counted votes, is there?
Yeah, but I was talking, for lack of a better term, about non additive electoral systems, in which the information conveyed in multiple ballots cannot be easily compressed
For instance in the FPTP system you can compress all the information in a polling station by counting how many votes for X how many for Y etc. And you can simply add up these numbers to other stations
The complexity scales linearly with the number of candidates
In a system like, for instance, IRV, where the voter ranks the candidates and the rank is an important part of the ballot itself you can't easily transpose all this information into a single number
The most you could do is build a tree data structure where the first level nodes are the first preferences and each node points to other nodes based on the successive preferences.
In this case the complexity of the process is exponential/factorial because you need a field for every possible combination of preferences, including cases in which not all candidates are ranked
My point was that going from a ballot to that structure is probably more error prone and time consuming to do manually than having a machine do it and that could conceivably be a reasonable justification to use electronic voting machines
The auditing will still be time consuming but that's mitigated by the fact that you audit only a portion of voting machines
But to clarify, I'm of the opinion that paper ballots and manual counting are preferable in every electoral system used at the national level that I'm aware of
India's electronic system is just not worse than paper ballots, which for electronic voting systems is a huge success
One reason is efficiency, if you have the machines counting you get results faster and then you can audit only a random sample of machines records and get statistical guarantees of the election integrity. (The machine can't know before that it will be audited so if you test thousands of machines and find no discrepancy it's highly unlikely a significant portion of the others did cheat)
A second is that it allows one more level of trust. With a paper ballot you have to trust that the poll workers are going to notice/stop/not help ballot stuffing etc. In most cases that's a good enough guarantee, but with voting machines you can also trust the people who programmed audited it
If the poll workers are trustworthy, because of the paper trail, you don't have to trust the auditers of the machines, but if you don't trust the poll workers then you can gain a modicum of trust from the auditers
https://www.brookings.edu/blog/techtank/2019/12/06/how-elect...
This link talks about the benefits in the indian election, specifically a software lock on the amount of votes per minute that can be cast
It also shows that in 2013 they audited "only" 20k machines out of almost millions yet found no discrepancy. Statistically that's probably good enough if the choice of machines audited was random
One last way it might be useful, though this is an abstract scenario, is if you want to deploy a more complicated to count voting system to large areas. In some voting systems counting can't be done in parallel, you need to do one round of counting, wait for everyone else and then do a second round and so on.
Similarly some voting systems might benefit from a digitized interface (for instance ballots in austria look like this: https://cdn1.vienna.at/2013/09/zettel.jpg and there are systems which would lead to even larger forms) which outputs a paper trail with only the actual information the voter inputed (in the case of the ballot I showed it'd pretty much just show a party name and a list of candidate names)
In these cases a machine outputting an auditable digital model of the votes cast would greatly simplify counting procedures. You could have every polling station just publish a signed file with the votes in their station and everyone could run the election algorithm
The votes can then be audited same as in the indian system