A couple friends of mine published a paper several years ago covering a few approaches to detecting election fraud based on numerical analysis of the results. I wonder whether the election discussed in this article would pass their test.
EDIT: also, the posted article probably deserves a [2018], since it's a discussion of Putin's 2018 re-election, not the recent referendum on (among other things) term limits.
Interestingly Putin changed nothing for 2020. His most recent election is even more polluted with these artifacts. Plotting turnout vs winner fraction gives you a grid pattern, instead of a cloud like you’d expect.
That we're down to subtleties of this sort shows some progress.
In the US south, before https://en.wikipedia.org/wiki/Baker_v._Carr , it was apparently popular to commit voter fraud by weighting votes in some areas more strongly than votes in others.
After that case, those committing fraud were so naive that all the dead people voted in alphabetical order. (IIRC the story continues: after arriving in the state Senate, the first order of business for the properly-winning candidate was a debate on how long someone had to be dead before they couldn't vote anymore.)
(In my voting jurisdiction we always conduct elections by mail. The last time there was a fraud, not only did all the people whose ballots were stolen report it, but the ballots had all been returned filled in with the same pen and same handwriting. I am glad we don't have "voting machines.")
Voter fraud is when a person sells their vote, or impersonates a different voter.
Election fraud is when the people in charge of collecting the votes manipulate the vote total in some way, such discarding some of the votes of people from a known party, manipulating voting equipment to skew the tally.
Why does it matter? Because some people claim rampant voter fraud as an excuse for restricting voting rights, despite the fact it has been shown repeatedly to be inconsequential [1]. At the same time, election fraud is a more significant issue because it simply scales better, yet most people don't appreciate the difference between voter fraud and election fraud.
For a counter point see [2], the heritage foundation keeps a list of voter fraud cases. I'd argue it proves the opposite of their point. Their database goes back at least 15 years and 1000 cases out of hundreds of millions of votes cast in that period. Also, their list is maximalist -- including cases such as "Larry Reker, of Worthington, voted twice in a contentious Independent School District 518 bond referendum special election, once in person and once by absentee ballot".
More related to the headline than the article but I was once at an election count where a candidate received 10,000 votes.
The returning officer read out (something like):
"Candidate One, four thousand three hundred and seventy nine.
Candidate Two, ten thousand.
Candidate Three,..."
He was immediately interrupted by the buzz of people in the room trying to work out what just happened.
Like how Everest was measured to be exactly 29000ft (in 1850s) but they thought no-one would believe how accurately they had surveyed it, so they reported it as 29002ft
I think this study is little flawed, I would appreciate if someone with more statistics background can comment.
I think:
- monte carlo simulation is not good to compare with election results. At least you have households tend to vote similar. Or some region more likely to vote on one candidate.
- also some percentages should be more likely, given the size of N considering first point.
Integer percentages are not indications of anything but rounding. I'm pretty sure you could find percentage rounding anywhere there are humans involved. Seriously; the null here is just gibberish.
You can use Benford's law to detect made up counts like people did with Bernie Madoff's frauds; they're not doing that.
But the percentages weren’t what was reported. I was caught off guard by that, too. I would recommend carefully re-reading where they explain how the integer percentages come about.
Voter turnout is reported as an absolute number of voters, not a percentage of voters. Those numbers aren’t round. Conveniently, when you actually go to calculate the percentage, a lot of them come out the same. They go into more detail explaining how unlikely this is, and how it can’t really be a result of humans rounding real numbers.
Apart from the other points already raised, it's worth noticing the peaks for round numbers exist almost entirely on the right side. The against votes don't have similar effects. If it was rounding, they would be properly distributed.
Wouldn't you see more "mexican-hat" kind of structures everywhere if they were merely rounding to the nearest decimal/decade? In the >= 95% you can see them but otherwise they are absent or less pronounced. Then again I'm no expert.
A quick-ish search on Benford's law and election results argues that it is problematic at best to determine fraud in election results. Though every paper I see recommends more research.
No one can really be surprised that there's a certain amount of voter fraud in Russia.
And the Russian leadership is not unsophisticated: of course they have access to mathematicians who could advise on how to avoid leaving these artifacts in the manipulated data.
I'd conclude from this that they don't mind others knowing they manipulate the results. Perhaps they even want them to know, as a little fuck-you show of power.
So why have elections at all? Because it gives Russia's international allies something to point to: they can claim to support democracy and human rights, etc, and pretend not to be convinced by this kind of circumstantial evidence, to an extent they would feel uncomfortable with if Putin were to decree "right, no more elections, no more constitution, I'll just rule for life". If you're really pressed and have to admit you find the evidence convincing, it only really demonstrates vote totals being manipulated by up to 1%. A tiny amount of the margin of victory and comparable to the levels of election fraud in large democracies like India or the US.
This kind of plausible deniability is what international relations is built on, and the bigger a power you are, the less plausible your stories have to be. Russian troops fight in the Ukraine while on vacation. China needs airfields in the Spratly Islands to support local fishermen. Guantanamo Bay is run in accordance with the Geneva Convention and definitely no one gets tortured there.
the Russian leadership is not unsophisticated: of course they have access to mathematicians who could advise on how to avoid leaving these artifacts in the manipulated data.
I'd conclude from this that they don't mind others knowing they manipulate the results.
I think you can just as easily get to the opposite conclusion, that they SO MUCH want to keep it quiet that they don't trust anyone including their mathematicians. And because the leaders themselves are mathematically unsophisticated (even if they're smart in general), they didn't know that this could happen.
It's not a tiny conspiracy though. It seems like the data is being manipulated for every precinct, in subtly different ways in different regions. So the manipulation is being done by whoever is in charge of reporting for each city or region, or at least they're aware of it (since the final numbers wouldn't match their tallies).
The headline shouldn’t use the term “voter fraud” as it’s a concept invented after the passage of the 1964 Civil Rights Act [1] to (one presumes) restrict democratic participation.
“Election fraud”, “electoral fraud”, are all far better for many reasons.
If you must refer to the other concept, “voter impersonation fraud” is the appropriate term.
PDF: https://www.cambridge.org/core/services/aop-cambridge-core/c...
EDIT: also, the posted article probably deserves a [2018], since it's a discussion of Putin's 2018 re-election, not the recent referendum on (among other things) term limits.