Violating the Benford's Law
theredelephants.com
theredelephants.com
If you read the issues, someone already pointed out that the distribution is not wide enough to expect Benford's law to hold even in principle https://github.com/cjph8914/2020_benfords/issues/9
The underlying issue is that vote counts are tallied over regions whose size and population are chosen to fall into reasonable bounds, so they don't follow a power law like e.g. the size of cities.
The first part of analysis says that Wisconsin turnout percentage is up "5.5 standard deviations" - as if turnout percentage follows any type of random distribution. The number of voters in Idaho increased 30% as well - is their fraud there?
Clearly this is a result of a "virtual statistical impossibility" and not the fact that this was the most publicized election of all time.
I think it's a mistake to dismiss all of it completely out of hand, without at least examining some of the more interesting and serious stuff as well: https://twitter.com/statsguyphd/status/1324352213595181059
All of the newly minted statistics experts on the internet wildly missaplying methods to support their foregone conclusions are exactly the same as Johny C. Theorist getting an internet education in materials science so that he can understand for himself that jet fuel doesn't melt steel beams. Trump's polling numbers also don't follow Benford's law in some cases. That's because there is no direct causation between legitimate elections and a Benford's law style distribution.
The fixation on wildly cherry picked data and missaplied statistics is the conspiracy theory. There are already court cases with the best experts in the land looking at the counts. Guess what? They are starting to be laughed out of court because they have no evidence to support their claims. It is either supremely arrogant or delusional to claim that you have found this thing that you spent five minutes learning about on the internet that proves magically that the Democrats stole the election while world renowned experts aren't able to prove it to a judge.
There is no liberal conspiracy to steal an election. Trump made the same claim in 2016. He made it during the Republican primaries about his opponents (that they cheated him). He said the system was rigged when he settled the fraud lawsuite for trump university. Hell... he even said he was being cheated when his tv show wasn't awarded an Emmy. This is just what he does and instead of spreading lies he needs to accept that he lost (probably... we'll see soon) and stop dividing the country.
This doesn't make any sense to me. Why do we see it in physical constants and so many other places, but not election results?
I did look through that paper by Deckert, Myagkov and Ordeshook from the Wikipedia citation, and I find it extremely difficult to follow. (A literature search reveals a number of other papers with varying conclusions.) I haven't done any stats since college, and that was over a decade ago, so I don't claim to be an expert. Still, if you actually understand something, you should be able to explain it from first principles!
I did some more digging, and I did find this very well-written piece on Benford’s law. It is something I can actually follow: https://towardsdatascience.com/benfords-law-a-simple-explana...
In general, I think it is best to avoid argument from intimidation. I also don't think that linking conspiracy theorizing with someone that just wants to understand something better is a kind thing to do.
That's the whole point. This isn't my field of expertise and it certainly isn't yours. No offense, but the 45 minutes you spent between my post and this one is not a "literature review." I am a professional scientist and for my most recent publication it took me three weeks to cover the literature on my one narrow finding. That was already after spending the last five years in a PhD becoming an expert in my field! The three weeks was spent deeply reading almost every significant paper on the topic I was publishing on. That is what literature review means and you don't really understand the nuances of applying Benford's law to election fraud by spending 45 minutes browsing wikipedia and skimming abstracts.
If the political science experts who have spent decades studying the facts surrounding Benford's law disagree that it is a reliable indicator of fraud then what are we doing running sketchy analyses and then making huge leaps to claim it as "mathematical proof that the democrats stole the election." And yes, that is a conspiracy theory with all of the hallmarks of one. It is to the point that state election officials are having to waste their time to go on television and debunk literal lies and disinformation that are being broadcast by certain political figures. Judges are already tossing Trump's lawsuits because even his lawyers have to tell the judges that they have no evidence of fraud. The continued assertion that Democrats are "stealing the election" is the conspiracy theory here and it is literally dangerous. A bomb threat was called into a Philadelphia counting center. Vote counters are being escorted to and from counting locations by the police for their protection from armed mobs that are aggregating in some areas. You'll have to excuse me if I'm already tired of seeing these baseless claims show up over and over again when not even Trump himself can pull out evidence to support them when it comes time to do so in court.
If you want to learn about Benford's law then good on you. I think that it is a really interesting topic and there is some good entertainment to be had in learning about it. However, it's delusional for people like the author of this article to spend an evening learning about it and then act like it is the magic bullet from the Kennedy assassination.
https://www.cambridge.org/core/journals/political-analysis/a...
To engage im brow beating of the person when you have data in front of you that shows admmittedly interesting skews evidences to an impartial observer an entrenched interest in avoiding the topic. You're doing the same type of thing I came down on people for during the 737 MAX fiasco, and as we've all found out, many of our "armchair experts" from that time ended up being right.
I have no love for either candidate, or the implication made by this data has for the country at large. In fact, that this is the most closely watched election in decades is exactly why these statistical anomalies are so important to look at. Not for Trummp's sake mind, but because usually there's no point in sweating it because most of the time there isn't really that much choice in candidate, so everyone just moves on and doesn't look that hard in the interests of keeping things moving along. Further, you have had drastic operational changes that have effected the efficacy of traditional poll watching. These statistical anomalies are the outcome. Claims that no one has evidence of any fraud in this is a bit llike claiming gerrymandering doesn't happen. It absolutely does, it just doesn't have a well formulated legal test making a determination that it has tenable in the Courts. There's a few drips of precedent around it here and there, but nothing to the scale of being in the national toolkit.
In a slight sense I agree with you; because I'm not sure that anyone wants to approach the can of worms created by dealing with the issue of "Okay, lets say this was a fraud. What then? Do over?" However, if life has taught me anything though, it's that nothing worth keeping and depending on is worth being spared the rigor of really testing and working out the warts of it from time to time. There is something odd afoot if we're heavily departing from historicals in so many places at once and if there is any merit to statistics at all as an auditing tool, and we rely upon these techniques elsewhere to perform as a sniff test, then you're either arguing the technique is flawed (which I don't fully buy; there being academic controversy on something doesn't disclude it from still being useful) and calling into doubt everything based on it; or you're asserting that for some reason Benford's magically doesn't work in the United States, and we've just capriciously used it as a specious reason to classify other country's elections as fraudulent for diplomatic purposes. Either of those is enough to justify some scrutiny. I don't, however, think the Courts are necessarily the right avenue for this disagreement to take place in.
In short, sit back, relax, and chew on the data. Forget how close to home it hits. Definitely don't dogpile those who are out of curiosity. Interesting times indeed.
If I had to offer up an interim solution for how the country should be run until any controversy gets resolved, I'd say that Executive authority should in a limited fashion be distributed to the cabinet as overseen by a re-affirmed Speaker of the House and Senate, working as co-equal partners. Not President and Vice President, but both together carrying equal weight. If one disagrees, the answer is no, if unanmity is reached on a direction, the answer is yes. The House and Senate must reaffirm any extension of that arrangement with a two thirds vote and Unanimous agreement between both houses. If the Congress can't manage to even do that, everyone gets dismissed and barred from reelection, and it goes back to the People to re-prime the pump. An alternative is to simply re-run the Presidential election. The Congressional statistics don't seem to be a subject of contention anywhere near as much as the Presidential. That seems to me the most consistent way to handle things given by my reading of the Federalist Papers, and other source material from around the Founding, and also has the perk of just kind of making sense given the current political situation.
- "fixation on wildly cherry picked data and missaplied (sic) statistics is the conspiracy theory"
- "There are already court cases with the best experts in the land looking at the counts"
- "They are starting to be laughed out of court because they have no evidence to support their claims"
I don't find your arguments compelling because you are arguing to authority. You also use a case that was dismissed that has nothing to do with Benford's Law. Please address specifically what data is being cherry picked and why Benford's Law shouldn't apply in this election.
There is evidence that in PA and MI republicans were not allowed to observe the vote counting. Why are they not allowed to observe?
> There is no liberal conspiracy to steal an election. Trump made the same claim in 2016.
In WI, MI there hundreds of thousands of mail-in ballots all arriving after election day. And they almost entirely for one candidate. I don't remind this happening in 2016, or never in US history.
Your other claim is literal misinformation.
Philadelphia City Commissioner Al Schmidt, a Republican: "Observers from the Democratic Party and Republican Party, from the Biden campaign and the Trump campaign, have been in our counting area observing, right up against where the process is taking place, from the very beginning" (he said this in response to a similar lie by Ted Cruz)
Benford’s law is a very well-established approach in forensic accounting to determine patterns of fraud. Is it appropriate to use in these cases, and was it used properly?
Selective analysis/p-hacking/cherry picking is a strategy all-too-often used in even scientific publications to exaggerate the strength of claims. Mathematical statistics itself, however, is pretty well-established as an approach to analyze data and identify outliers to gain insights. Is the author biasing the analysis with selective focus in a way that any objective observer would believe was unreasonable? Exactly how, and what would a more balanced analysis show?
Responses that identify methodological concerns with challenges to reported voting numbers in a systematic and thorough way would be of value to the HN community. Angry words that veer off-topic and provide only anecdotal information make you sound overly defensive and do not serve your cause well in this forum. It’s like running away from a dog, it only motivates the dog to run faster.
Just chill and address the points systematically. There aren’t that many points to address. Maybe it will take more than one post or a pdf of your own.
But if you skip a point, assume that the audience will interpret it to means that the post is correct on that point. Acknowledge if it is and move on.
I do love voting data. It collapses the wave function of political rhetoric...
MI, WI, PA stopped counting votes on November 3. These states had the same candidate winning by 130K+, 300K+ and 700K+ votes respectively. It's remarkable that at about 4AM in Nov 4, WI and MI reported just enough votes to change the outcome of the election.
And there is evidence of people that was prevented from observing the vote count, and apparently some observers were kept so far away that they had to use binoculars to observe.
If you want to discuss application of Benford law to election process, maybe a different tone would help.