Hacker News Reader

Comment by guscost | Hacker News Reader

Here’s someone who did analysis of 2020 NYT time-series data, and is trying to make a case that the currently-contested states show odd patterns. Note that this would be circumstantial evidence at best, but still a little fishy when compared with the timeline:

https://threadreaderapp.com/thread/1325592112428163072.html

I made a GitHub repo of this analysis ported to Python3, with a quickstart and a matplotlib script, if anyone wants to try it. If nothing else it’s kinda fun to tinker with the data:

https://github.com/guscost/nyt-election-2020-analysis

View on HN

The premise of this analysis appears to be that mail-in ballots are a uniform sampling of Democrats and Republicans. But of course, in 2020, that's not at all the case; the Republicans made a campaign issue out of not voting absentee, and of attempting legally to reject absentee ballots under a number of conditions. As a result: mail-in ballots are sharply more Democratic than Republican this year.

Further, any analysis that compares returns over time and holds Florida up as the model state to match against needs to account for the fact that Florida was able to begin counting ballots early, because they passed state laws to do so†. PA and WI were legally prevented from counting ballots until election day. The returns over time are going to be much different in FL than in PA.

Here's another data point: you're about to see Biden's lead in New York open up and drive his popular vote victory even higher, because NY, which is a shitshow, is only now getting started counting all the mail-in votes. Obviously, the Democratic Party didn't launch a giant vote fraud operation in New York State, which they would have won even with a dead candidate.

† Every state should be like Florida, in this regard.

I collected Florida ballot returns and early voting by party registration by date.

I allocated independent registrations 50/50 to each party, and ignored 3rd party registrations.

Obviously this isn't the same as actual votes, but it is instructive. The Democratic party had a large lead at the start from postal votes, but that shrunk as more early voting happened.

  Date         Likely Democratic lead
  24 Oct 2020     386,908
  26 Oct 2020     354,654
  27 Oct 2020     299,021
  28 Oct 2020     245,912
  29 Oct 2020     204,277
  1 Nov 2020     94,325
  2 Nov 2020     108,123
  3 Nov 2020     114,889

To me this indicates that the narrative of postal votes being vastly more from the Democratic party holds up.

In states (like PA) where postal votes weren't counted until after polls closed we'd absolutely expect to see a large Republican lead which would shrink as postal votes were counted.

I've watched California's election results for decades. You totally see mail in ballots trend Democratic. And urban areas tend to be slow to report. The tallies always shift towards Republicans then shift back toward Democrats.

AK is even crazier. It doesn’t start counting its absentee ballots till a week after the election which is why it’s been stuck at 56% votes counted since Election Day.

I'm guessing that's why it hasn't been called yet, despite trending very red. It'll probably still end up in Trump's column, but with such a disparity in voting methods by party it's more uncertain than usual.

Alaska will vote for Trump, but Alyse Galvin told me the other day that internal polls show her still in the running for the at-large House seat. We'll know more tomorrow.

edit: OP expanded their post and my comment below is now inconsequential and even misleading.

While I do not find the analysis particularly strong, it does not have the premise you describe. It only has the premise that the mail-in ballots are randomly shuffled before their counting starts (which is also not a reliable claim).

I could be misreading it, but I don't think I am; for instance, the bit in the middle, about how the batches should be getting more Republican as time goes on and returns come in from rural counties (also: not how it works).

You are right. The author is assuming the premise that all mail in ballots go to some central location, get shuffled together and thus should have a uniform count. The mail in ballots actually go to local precincts which count them and report them at different times. In fact there were times where mail-in ballots and in-person provisional ballots were being dumped at the same time because different precincts were at different stages of processing ballots at a given time.

This analysis makes no sense and shows nothing unusual.

Yeah, that one particular tweet becomes contradictory with the larger point they are making, but at face value [1] the larger point is that "each new batch of mail-in votes being reported as counted should have the same D/R ratio because they were shuffled before the counting started". It does not make a claim about what that ratio should be.

[1]: Given the flimsiness of the arguments, taking it at face value might be giving it too much credit. My opinion is that this is nonsense.

edit: your sibling comment about mail-in counts in PA also being explicitly not shuffled, but by county throws another wrench in their argument.

Sure, but the analysis is supposedly showing even vote-by-mail trending toward R in other states, explained by the hypothesis that mail from more outlying areas tended to arrive later. The observed discontinuities in the contested states could be explained simply by the sequence in which these states processed their mail-in ballots (i.e. ballots mailed from cities were processed later), but the fact that so many of them changed in the middle of the night on election night, while not evidence of anything by itself, is what I referred to as being “a little fishy”.

As a counterexample, Kansas appears to have hit a batch from a metro area shortly before midnight on election night, too.

You know, if you had been following 538, none of this was either mysterious or unexpected.

The challenge is that the vote counting got behind. Specifically in most places it got behind in major cities. Which voted more D than R, even among mail-in ballots (which already lean more D than R than in person voting in most states - Arizona being a notable exception).

As soon as you correct for where the ballots are from rather than guessing that it is all rural, the discrepancy disappears. But that's inconvenient for a disinformation campaign so gets lost.

It's not at all fishy. You could --- and places like DDHQ did --- scrape the SOS sites to know almost exactly where the ballots were coming in. The central premise of this analysis is not just that the mail-in process was partisan-neutral, but that it had the effect of randomizing the returns statewide. Of course, it did nothing of the sort.

Again, what you really saw here was that the election day vote trended sharply more R, because R voters were encouraged not to use mail-in votes, while the entire D campaign apparatus was --- like it has been since Obama --- designed to get people to commit their votes early with mail-ins. The mail-ins took longer to count, because unlike in Florida, they couldn't be counted early. So what you were watching as time went on was large tranches of mail-in ballots that took forever to count coming in from D-heavy areas (or from areas where Biden outperformed Clinton, and where the mail returns were, like everywhere, more D than the election day returns).

> The mail-ins took longer to count, because unlike in Florida, they couldn't be counted early. So what you were watching as time went on was large tranches of mail-in ballots that took forever to count coming in from D-heavy areas (or from areas where Biden outperformed Clinton, and where the mail returns were, like everywhere, more D than the election day returns).

So if I understand correctly, you're saying that there's a gradual (in PA) or not-so-gradual-at-first-but-then-gradual (in MI) shift from in-person ballots to mail-in ballots, which explains these trends? That does seem possible.

Another possibility is that the USPS processed the mail-in ballots from big cities differently from those from towns with smaller mail volume.

Here's another thing to look at: the Florida time series starts on election day. But vote counting Florida didn't start on election day. The state was allowed (because they are smart about this) to start counting 23 days in advance of e-day. The author of this thread is unaware of that, and has been fooled by Florida's efficiency into believing that vote returns are uniform everywhere. But that wasn't true in PA, where there was no preprocessing time at all. The clock started in PA on e-day, and the busiest, largest, most Democratic counties naturally took longer to count.

Interesting, that's probably a big factor. Counterexample: Georgia, which started processing mail-in ballots 15 days before e-day (as far as I can tell): https://abcnews.go.com/Politics/battleground-states-process-...

Georgia does limited preprocessing of ballots before e-day, to catch and report invalid ballots, but can't count until the day.

Apparently not this time:

> While county election officials were able to verify signatures and voter registrations upon ballot receipt, normally, absentee ballots can't be further processed or opened until Election Day. But this year, beginning 15 days before the 2020 general election, officials were allowed to open returned absentee ballots, remove those ballots from both the outer and inner secrecy sleeve envelope, and scan the ballots using a ballot scanner machine.

They're scanned. They're not counted. They're not allowed to count until the poll close.

(If you Google this you'll see a lot of other stories about why the counts from D-heavy areas were delayed; there were glitches).

Sibling reply since I've gotta get to bed:

> At least three election workers must be present at all times during this processing and officials are prohibited from doing any tabulation of results until polls close at 7 p.m. on Nov. 3. At that point, tabulation is simply pressing a button.

Narrator: It wasn't simply pressing a button.

https://www.fox5atlanta.com/news/software-glitch-causes-dela...

Again: we watched the ballots come in from Georgia. They came in county-by-county, in partisan-leaning lumps. There was a running joke on Twitter on Wednesday about people who have never been to Georgia suddenly acting like they knew where Clayton County was. It was a whole shared cultural experience!

I'm still struck by the fact that you'd have had to not be paying attention at all to not understand that the results weren't arriving uniformly from across the state, or that different locations in the state have different partisan leans. From what I saw, the whole country was basically transfixed by those two facts. We got nothing done! For like a week!

It's fine if you're not paying attention, but way less fine to build a narrative about voter fraud based on it.

So you're saying one county, with ~80,000 delayed votes, ~60% of which were for Biden, completely eliminates Georgia as a counterexample here? I don't have a good spider-sense for this kind of number crunching so I'd have to do some math.

> It's fine if you're not paying attention, but way less fine to build a narrative about voter fraud based on it.

I have not been claiming there is anything to this theory, but it is fun to argue about. And I'm definitely not happy to see this argument get personal, but from my perspective, it is least fine of all to gaslight people with a nonsense fairy-tale that corruption and fraud do not exist.

The gaslighting happening here is the accusation of widespread fraud without any evidence, based on the kinds of intentional misunderstandings tptacek spent so much effort laying out for you.

For what it's worth, my "you" upthread was meant to be rhetorical, to refer to the batshit Twitter thread, but I can see how it would have been taken as personal. I mean, I still think we should all be able to quickly and decisively agree that the thread was, well, batshit. But I wasn't trying to confront anyone commenting on HN.

NC also counted (er, tabulated) well in advance of Election Day.

FL was allowed to process and count mail in votes early (they always deal with a lot of absentee votes b/c many people have their residence there to avoid state taxes). Once polls closed they were able to transfer those with the early batches of day of votes. Because absentee was such a big thing this year, many states were also telling the networks/reporters what each batch of counts was made up of (absentee vs. in person). Even in GOP areas, absentee went heavily towards Dem.

The GOP legislature in PA denied the SOS request to begin processing absentee early. It was known for weeks that the count would not be done until Friday or possibly the weekend. Because the pandemic has been politicized, many GOP voters saw in person voting as a statement. Dem voters just wanted to vote.

There also wasn't really a gradual shift. Day of gets counted more quickly because states use machines that take the the vote directly from the user at the time. Absentee takes longer because the votes are opened, verified (some states check signatures, but I think all make sure the it's not a double vote, etc...), cured, and then the ballot is ran through a flattener then vote counter. I saw a story that this took longer in PA, because many of the votes had been sitting folded up for almost a month by the time they were opened.

No, the premise is that mailin ballots should have a roughly uniform D/R ratio, no matter what timeslice of a state you look at. Of course they skew, but the idea is that the first mailins counted should have basically the same ratio as the last. Which appears true in most states, but not some, for unknown reasons.

(The skew it proposes as normal, R later in time due to receiving R ballots on a later distribution, exposes an assumption it makes: that counting is in order of ballots received.)

They're not all counted at once. You watched the returns come in drop-by-drop in PA, and each was attributed to counties and precincts. You are not going to see a uniform D/R ratio between a drop from Carbon County and a drop from Allegheny County.

This is so basic --- you had to have not watched the PA returns come in at all to not know it --- that I read this twice, figuring that I had to have misunderstood what the thread was saying. But I don't think I am; I think this person believes that the mail-in process randomizes the ballots statewide.

> I think this person believes that the mail-in process randomizes the ballots statewide.

Oh god you are right. I hadn't even contemplated this.

Interesting enough, the mail-in ballot returns on a per-precinct basis were actually shockingly consistent, even over time, likely proving by this method that there wasn't any voter fraud.

The thing is though, if you look at the returns on a state level over time, what you are actually seeing is the results from all sorts of different types of votes in different places over time, so you won't see any consistent pattern. But if you break down the votes to the specific places then you see exactly what you'd expect to see.

Do you have the per precinct D/R ratio data?

> No, the premise is that mailin ballots should have a roughly uniform D/R ratio, no matter what timeslice of a state you look at.

Which is based on a series of assumptions that simply do not generally hold (the most central being that the entire statewide collection of mail-in ballots is homogenized before being counted.)

Having not thought for more than a few seconds about this, it sounds like a clearly invalid premise. People decide to mail their ballot for all sorts of reasons including seeing an ad, being telephoned by a campaign worker, seeing a movie on tv, being well organized or lazy, and on and on. Those catalysts are obviously not likely to be uniformly distributed between parties.

Philadelphia, Detroit, and a number of other cities are notoriously slow in counting their ballots.

This happens every election that I can remember, going back to 2000.

If I had to guess, the complexity of counting may increase in a non-linear way after a certain number of precincts.

I'm unconvinced the proportions should be consistent at the state level or any level really. There have been GOTV events that have thousands of attendees, then there's the "souls to the polls" event every time, I can't think of a way to slice it where I would be convinced by unbalanced batches.

> If I had to guess, the complexity of counting may increase in a non-linear way after a certain number of precincts.

I'd bet that the larger counties, while they have more ballot counting machines, probably don't have proportionally more. Expensive, specialized, maintenance-intensive, etc. Every county needs at least one, but beyond that I doubt there is a huge priority on having excess capacity.

Ballot counting machines are probably a case of over-automation. The polls station clerks are volunteers and are maybe doing their first election. Setting up a manual process is just way easier then dealing with complex computer reporting systems. As soon as some machine or computer is not cooperating they are stuck. With pen and paper the worst that can happen is that the pen is out of ink.

If all the ballots were shuffled in the mail across the timespan in which they were counted, that would make sense.

However, they are not, so the assumptions underlying the analysis do not hold.

For one thing, many states did not begin counting these ballots until Election Day, per state law. For another, the entities performing the unofficial counts were not necessarily the entities receiving the ballots. The author clearly conflates these two points when he says that Milwaukee ballots heading to the county to be counted were received after the deadline. They were received by the precinct at the appropriate time, and then transported to the county for final tabulation - again, presumably, in compliance with state law.

(EDIT: Notably, the video clearly shows the ballots being transferred by law enforcement, so it's patently dishonest to suggest that was the time they were received through the mail.)

If a state performs constant counts and constant updates as they receive individual ballots from across the state, then you should get a more-or-less constant D:R ratio.

If you count mail ballots in batches of updates from different counties and within counties from different precincts, then you will see the D:R ratio generally shift, with precincts and counties with higher populations (ergo higher batch processing times) coming later.

> Around 3am Wisconsin time, a fresh batch of 169k new absentee ballots arrived. They were supposed to stop accepting new ballots, but eh, whatever I guess.

Anyone who knows anything knows Milwaukee (not just as a city but as Wisconsin's largest city) votes blue. So as soon as those votes start getting counted the votes will shift blue. You _should_ see this bimodal distribution in vote counts.

It seems reductively easy to see that the basic premise is incredibly flawed here: that ballots are uniformly shuffled in all states. Some very basic things that would cause those exact shifts in the data witnessed:

1. Mail vs Drop Box could very easily have a different ratio, which would easily explain a sharp shift if they counted them in different sets for some states when it's a plateau and a hard line jump.

2. Same goes for any sort of discrete shuffled sets - say they are boxed by the week through early voting, or by county early voting, etc. There's literally dozens of explanations for how you can end up with discrete shuffled sets.

3. If mail ballots are counted in roughly the order received, then any gradual shift may represent that one candidate may have early support fade or strengthen over the course of early voting.

4. All of this also ignores that the time series of reporting may not be from the same counties or mix of counties. Change that mix slowly and you also get a gradual shift.

Basically, you need full insight into exactly how ballots are collected (and vary by state) in order to actually have the central hypothesis hold. With no shred evidence of voter fraud at any level and a very clearly biased author, I don't see this holding any merit. While I understand the idea of trying to consider all data, the resounding lack of any evidence should lessen our resources put towards these unlikely theories.

To repeat myself because I think it's worth being clear about this: we have enough insight to know that the premise of ballots being randomized statewide does not hold, at all. The reason we were all hitting refresh on Wasserman and Cohn's feeds on Twitter was that each Wednesday, Thursday, and Friday batch was attributed to a specific county, and we had expectations about what returns in those counties would look like.

This pretty much adds up, and invalidates the whole analysis in my opinion.

But I still believe that the first conspicuous batch of mail-in ballots being dumped in the witching hour, after poll watchers were told the counting was done for the night, is fishy, and I will not be surprised if the court considers it part of a larger pattern of behavior which causes ballots (legitimate but unverifiable, in many cases) to be thrown out.

The first couple days the big counties were counting 24/7. The managers were on TV talking about their teams, and you could watch live streams on many of the counting areas.

The court is likely throwing nothing out. The no poll watcher lawsuit was already laughed out of court when the lawyers had to admit a 'non-zero' number of poll watchers were present the whole time [1]. Trump might be fine lying on Twitter constantly, but in court his lawyers would prefer to not commit perjury.

[1] https://www.cbsnews.com/news/fact-check-trumps-claims-poll-w...

There are many sworn affidavits being filed in court right now, from poll watchers claiming that they were kept too far away to validate a single ballot, or turned away entirely. We’ll see, it will definitely go to court.

Read the fact check. The too far away claim was decided days ago, and people were allowed to be within 6'. There were also live webcams setup. This has literally been one of the most transparent elections in history.

Read through the lawsuits [1]. Most have already been rejected due to no proof. The others outstanding have to do with so few votes it doesn't matter or they actually go against Trump's interests, like stopping the count when he's behind. This is all theater at this point. Maybe they can have a few more press conferences at the garden center next to the sex shop.

[1] https://www.businessinsider.com/trump-campaign-lawsuits-elec...

If one of the new cases filed this week makes it to SCOTUS will you admit you were wrong?

I think I speak for all of us when I say that nobody is going to validate that weird Twitter thread you posted, no matter what happens at SCOTUS.

Of course not, I’ve already said so. Are you being willfully ignorant?

I'm interested in discussing facts, I'm not interested in playing "Time for some game theory" on HN. Sorry.

Note also the extremely telling detail that Trump's WinRed links for fundraising the recounts and legal challenges allocate 60% of those funds to retiring his campaign debt.

Exactly. I was trying to stay away from pointing out that Trump was trying to scam him followers one last time before leaving office, but here we are.

> With no shred evidence of voter fraud at any level

I feel like this is actually the most suspicious thing here. In a country of over 300 million, in one of the most contentious elections in living memory, not one person was found to have committed voter fraud? It seems more likely that we're just terrible at detecting voter fraud rather than that there were literally zero instances of voter fraud anywhere.

Which implies we might need some better measures to detect it. The thing they were trying to do with dead voters is promising but has the obvious problem that there could have been legitimate ballots submitted, or unintentionally illegitimate ballots, if the death was recent etc. But the general idea seems to hold. Maybe we should add a few percent fictitious entries to the voter rolls and then see if any of them votes next time.

Anybody have any good ideas for how we could be detecting this, if it's happening?

If it isn't happening then having some measures in place like this would be good to point to as affirmative evidence that it isn't rather than merely the absence of evidence that it is. And if it is happening, it would be good to know that too.

First, we do detect a small number of voter fraud cases. People have gone to prison for it.

But your premise is also broken. You assume we should see regular attempts at voter fraud, given our enormous sample size. But voter fraud, on the "retail" level, is heavily penalized and extremely unlikely to impact outcomes. There are all sorts of crimes that could occur, but don't, because there's no point.

> First, we do detect a small number of voter fraud cases. People have gone to prison for it.

Fair enough, though now I'm wondering why I keep hearing "no shred of evidence of voter fraud" from so many people.

> But voter fraud, on the "retail" level, is heavily penalized and extremely unlikely to impact outcomes.

There are currently four states (AZ, GA, PA, WI) with less than 1% margin between the candidates. Georgia is at 0.2%. It wouldn't take a high percentage of fraud to flip a state.

But let's suppose that retail voter fraud is rare. How are we even detecting wholesale voter fraud? What's the method to detect when someone with access to the voter rolls slips an extra crate of ballots into the count, or fills out and mails in thousands of ballots in the names of people who never asked them to?

Half a percentage of the AZ vote is over 16,000 votes. See why this crime doesn't make sense?

16,000 votes is only a half a percentage of people in AZ committing voter fraud, and that's assuming they each only submit one fraudulent ballot. Apparently people have been caught submitting dozens of ballots in the names of people they didn't have any right to. If 0.02% of people did that, it flips the state.

You can also make the "doesn't make sense" argument about voting itself, since exactly zero of the states will be decided by a single vote with extremely high probability, but lots of people still voted.

It's not much of a stretch to expect that a fraction of people smaller than the lizardman constant are not doing a diligent cost benefit analysis.

What do you think a fraudulent ballot is? You can't just make a person up; there are voter rolls. Whatever means you find of coming up with a name and address that will be accepted as a vote, you have to do it (here, hypothetically) 16,000 times. If you use real people's names, which you have to because, again, voter rolls, many of them have already shown up to vote. The epollbook shows that on the screen, and you're standing there being asked for ID by a suspicious poll worker. Meanwhile, you would need thousands of coconspirators, and count on none of them ever leaking. You face the prison sentence to end all prison sentences (we put someone in prison in TX for years for casting an invalid vote accidentally) and, if you pull it off, you have... a chance? of altering the outcome if the election is very close?

There's a reason this just doesn't happen, and it's because it doesn't make sense. As a prediction: America is not going to spend a lot of time and energy making you feel safer about it.

My advice to you is, next cycle, sign up as a pollworker. It's so much more of a shitshow than you can possibly imagine, but I came away from it feeling paradoxically better about the process.

It's unfortunate that people will simply not listen to reason. You have gone above and beyond explaining the realities of mass voter fraud, but it goes against the outcome people are searching so they ignore you. You have more patience than I have.

> It's so much more of a shitshow than you can possibly imagine, but I came away from it feeling paradoxically better about the process.

This is the first time I really dug into the process, and I agree. I voted absentee, and that process plus what I saw on TV around the counting process made me much more confident in how things work.

> The epollbook shows that on the screen, and you're standing there being asked for ID by a suspicious poll worker. Meanwhile, you would need thousands of coconspirators, and count on none of them ever leaking.

You're talking about someone showing up at the polls to commit voter fraud in person, when the contention here is primarily about mail-in ballots.

If someone has access to the voter rolls from the previous election, how do you detect if they've submitted mail-in ballots for people who were registered but didn't vote last time, on the theory that most of them still won't? Or who registers new people without their request and then submits ballots in their names? That entire process would be taking place outside of the polling location.

> Apparently people have been caught submitting dozens of ballots in the names of people they didn't have any right to.

You'd think there'd be evidence for that somewhere, but you're just parroting die-hard Trump supporter talking points now off Twitter with no evidence. It's very clear you are not talking about any of this with good faith.

There is no evidence of widespread voter fraud is what’s being said. It’s on the accusing party to provide that evidence and they haven’t. It also strikes me as strange that they’re targeting only the cities that they are. Detroit for example when it was actually Oakland and Kent Counties that most altered the outcome in Michigan for Trump.

Election officials have many ways to audit elections for example knowing exactly who voted in an election how and where.

To fill out and mail thousands of ballots you’d need to have them in the first place. Ballots are tracked from mailing to receipt and processing. How would you know who to pick so that there’s no conflict if they show up at the precinct to vote in person?

> It’s on the accusing party to provide that evidence and they haven’t.

But that's what I'm saying. What measures can we put into place to allow them to do that if it's actually true, or rule it out if it isn't?

I'm asking for a good way that someone could find evidence that someone else e.g. mailed in a large number of ballots fraudulently, and you're telling me that there is no evidence of that happening. Well yeah, that's the issue. What I want is a way to distinguish the case where there isn't evidence because it isn't happening from the case where there isn't evidence because we don't have a good system in place for detecting it.

I want the ability to show conclusively that this isn't happening if it isn't happening, not just a lawyer's argument that absence of evidence is evidence of absence.

Before you ask for people to help you preclude that anyone mailed in a large number of fraudulent ballots, I think it's fair to put to onus on you to describe how that actually could have happened. Start at step 1 and talk us through it: what does the "thousand ballot mail-in attack" look like? Bear in mind voters are registered with ID and proof of address into every state's voter rolls, states have records of when people die, and every election authority watches for duplicate votes; those are the table stakes mitigations you need to get past. Go ahead, I'm interested in hearing this out.

So let's come up with a couple of scenarios here.

Probably the hardest one to catch is where they have access to the voter rolls and submit ballots in the names of people who were legitimately registered but didn't vote. The stealthier way to do that would be to have real-time access and then inject the ballots at the last minute in the names of people they know didn't vote, so there wouldn't be duplicates.

The scattershot method would be to predict who isn't likely to vote, e.g. because they didn't vote last time, and submit ballots in those names. Then they would have some duplicate ballots, but if the predictions about who wouldn't vote were mostly right, they could submit thousands of ballots and only have a small percentage of duplicates. There would be some duplicates, but then what do you do? You could throw out the duplicates, but one of the two was a legitimate ballot, and the majority of fraudulent ballots wouldn't be duplicated and would still get counted. It's also not obvious how to distinguish between whether a small percentage of duplicated ballots out of millions cast are a result of a single act of intentional fraud or many independent acts of incompetence.

Another alternative would be to register people who didn't request to be, e.g. people who are institutionalized and a worker at the institution is in a position to register them and then intercept their ballots and submit them.

Some privileged positions would allow someone to do this at greater scale, e.g. postal workers. A postal worker in a high density city could have many thousands of residences on their route, each with multiple eligible voters. Get a copy of the voter rolls, register every eligible voter on their route who isn't already in there, use their real social security numbers obtained from a copy of the Equifax leak or similar, then request mail-in ballots to take and submit.

1. You have to get the actual ballot for the people who haven't voted (you have to request that ballot, of course). You can't just fill them out from a template.

2. To register, you need enough information to steal someone's identity. In states like PA, the first time you vote after registering, you have to present photo ID.

3. The largest mental institution in the United States houses 1,400 inpatients. In the implausible event that you were able to somehow vote for all of them without being detected, you'd have flipped none of the states this cycle.

And until you have that, you're going to assuming the unlikely, unproven, and unprecedented side from a president literally known for lying and questioning the results of the past election baselessly, finding no evidence in 4 years of time to search and the DOJ at their disposal? You think the conspiracy is so widespread and perfect that there's no trail at all, not even to maybe 20% of the alleged fraud? C'mon now.

Keep in mind there are two kinds of voter fraud: the ones that get caught (and never show up in vote data), and the ones that don't. There are always stories about the person dead for 10 years that gets a ballot that is maybe even turned in - but that vote isn't counted, precisely because it got caught.

While I think you're arguing in bad faith, at least there's some interesting discussion in the replies. I think it's an interesting thought experiment to figure out how to do mass voter fraud though:

> How are we even detecting wholesale voter fraud? What's the method to detect when someone with access to the voter rolls slips an extra crate of ballots into the count, or fills out and mails in thousands of ballots in the names of people who never asked them to?

The mere act of bringing in a "create of ballots" is going to be incredibly tricky. You have to obtain the valid ballots somehow: steal them, get some printed, or print them yourself. If you print them, you have to perfectly replicate the design, type of paper, etc. Then you have to physically get the box into the building, into the right room, with the proper security seals or whatever is used.

But let's pretend security is crazy lax, and/or several people are involved AND somehow not one of them is caught or a whistleblower. This is already a stretch -- as the number of people involved goes up, so do the chances one gets caught and/or one is a whistleblower.

If we add the ballots without it adding up to the number of names crossed off the voter rolls list, that's going to be a giant red flag of voter fraud happening. Either this has happened and no one has noticed (one one has ever thought to check this?), or.. it hasn't happened.

With access to the voter rolls we could inject names, but of course that leaves a paper trail. Thousands of names of people that don't exist or are deceased or are from the wrong place would appear and would have all voted. We could also figure out how many deceased people are on the list and have them vote, but this too leaves a paper trail. While there's some reporting of incidences of deceased people voting, it appears to mainly be clerical errors [1] and certainly doesn't appear to be at the scale of thousands needed to sway the vote, so likely this also hasn't happened.

Casting any significant number of ballots for legitimate voters ahead of time would be likely to result in lots of people showing up to find out they've "already voted" and being turned away. Once again, this either has happened and thousands of people quietly walked away saying "oh well" and not going to the media.. or it hasn't happened.

With access to the rolls, we could look up who didn't vote, and cast votes for them. We'd need to be able to cross out names after the polling has closed, then make sure our crate has a matching number of ballots to the people that didn't vote and physically get it in to the count room. And of course, this would show up in the type of analysis this article is doing: it would look exactly like the Russian election results at the start. This also doesn't appear to be the case, at least for 2016.

All of this work would allow some fraction of voting fraud to happen at ONE polling station. If we pick the right station in the right district (where the security and scrutiny is likely to be higher) we could maybe even have a meaningful effect, but for all that work and risk (of jail time) there's no guarantees. We'd need to do it at multiple polling locations, but unfortunately most of the work involved doesn't really scale: we we just need more and more people involved in the conspiracy.

[1] https://www.cnn.com/2020/11/08/tech/michigan-dead-voter-fact...

The concept of voter rolls kills the "crate of ballots" idea just as dead as the "retail voter fraud" idea, because you need to know who's voting and who isn't to pull it off; otherwise, you make a whole mess of duplicate ballots, which is one of the most basic things the election administrators are looking for.

And, as you point out, that's what you're dealing with after you James Bond the crate of ballots into the counting facility and through the scanners.

Trump also barely won MI, WI, and PA last time. No was nitpicking vote result curves at that time.. I'm sure they were equally as strange.

You were expecting Trump to question the results in states he won?

An accusation of hypocrisy doesn't tell you which time they were wrong to do it.

They are wrong to do it because it convinces millions of uneducated Americans that it's all rigged and their vote doesn't matter. Here are investigators debunking most of the voter fraud allegations thus far. https://twitter.com/Ike_Saul/status/1324435797374808066

https://www.reuters.com/article/uk-fact-check-biden-vote-spi...

I'm absolutely willing to accept that very few people are (e.g.) voting twice in different states, filling out ballots mistakenly sent to their address, and similar small-scale stuff. But I do share your skepticism that when the stakes are this high, nobody's trying to meaningfully affect this. Especially this year when anyone could have told you the election will come down to the same small number of states waiting for days to count ballots in a small number of precincts.

We know how well the TSA does because we actually try to send people through with weapons to see how often they get caught. In 2015, the answer is that 95% of weapons get through.

I think we need the same kind of thing for elections. How tough is it to get / remove a few boxes of ballots overnight from wherever they're stored prior to being counted? When you do, is that immediately detected and reported, or is it presumed to be a mistake and just accepted? How many people does it take?

[1] https://www.nbcnews.com/news/us-news/investigation-breaches-...

> But I do share your skepticism that when the stakes are this high, nobody's trying to meaningfully affect this.

Thing is, it's much easier to impact the result legally through voter suppression.

https://www.bbc.com/news/election-us-2020-54380684

https://www.reuters.com/article/us-usa-election-locations/so...

That's a separate conversation, though. I agree reducing poll locations can make some people less likely to vote, but it's not in the same category as fraud.

To commit fraud at the level required though just isn't plausible in a national election. Even in a close state like PA, we're going to end up talking about 45k+ votes. In MI, 100k+ votes, WI 20k+ votes, etc... All these states have mixes of R/Ds running things, and any fraud claims quickly ends up in grand conspiracy fallacy territory.

This is not some 1000 vote situation like FL in 2000 (which was actually much more complicated than everyone paints it out to be).

20k in WI, 45k in PA.....let's call the baseline as a swing of 20,000 votes in a single state.

That would be 7-20x the impact of election fraud from the Patterson, NJ election case this year (estimates range from 1,000 to 3,000 votes affected). [1] There were four men charged in that incident. [2] So with about 30-100 personnel with malicious intent, distributed across a dozen election districts, you could theoretically swing an entire state's results in the 5-figure ballot range.

While this is not PROBABLE, we should acknowledge that it is POSSIBLE, and take steps to mitigate the risk accordingly. But basic enforcement of election integrity just outright sucks. Why else would someone feel so bold as to video themselves ballot harvesting and thinking it's ok?[3]

[1]https://newjerseyglobe.com/local/evidence-of-massive-voter-f...

[2]https://www.northjersey.com/story/news/paterson-press/2020/0...

[3]https://www.twincities.com/2020/09/28/project-veritas-video-...

Philadelphia received over 350k mail-in ballots [1]. Biden's current winning margin for the whole state is under 46k [2].

Wayne county (Michigan) received over 430k [3]. He's won Michigan by 148k [2].

In Wisconsin, the margin is just 20k [4], with the most populous county representing 459k votes.

In Georgia--same kind of thing.

Normally I agree with the dismissive "it's so decentralized, you couldn't do it." But it's different this year. Anyone could have predicted that the pro-Trump areas would finish counting faster, and then the pro-Biden areas would have multiple days and nights to count, all in staunchly pro-Biden areas capable of swinging the election. Of course we expected mail-in votes to lean Biden, but that doesn't mean anywhere from 51-99% indicates "no fraud."

I'm not saying any fraud happened. All I'm saying is that it's not rational to presume it couldn't have happened without being detected. And it's worth some efforts to see how well we'd catch it if it were occurring.

[1] https://whyy.org/articles/pa-election-2020-results-thousands...

[2] https://www.nytimes.com/interactive/2020/11/03/us/elections/...

[3] https://www.nytimes.com/interactive/2020/11/03/us/elections/...

[4] https://www.nytimes.com/interactive/2020/11/03/us/elections/...

So you're arguing that absentee ballot by mail is the problem? None of the states in question did automatic mail in voting. Each voter had to apply for absentee. And given the politicization of the pandemic, Trump made it a virtue signal to vote day-of in person. So of course the absentee vote was heavily D, and it came in from heavily D areas to start.

I voted absentee and honestly it felt more secure than any in person voting I've ever done. I had to apply, fill out, and then drop off my vote in a box. My wife wasn't even allowed to drop off my vote, and instead I had to do it personally. I was then able to track it online and verify it was counted without an issue. Contrast this to when I hit some buttons in person and have no idea what happens after that. The point is, you have to trust the process at some point, and in this case fraud is so unlikely across multiple states with 10s of thousands of votes.

> So you're arguing that absentee ballot by mail is the problem?

No. It's just one factor.

For the rest, respectfully, you'll have to actually read my comments. Your responses haven't yet taken into account anything in my comments. It's like you saw the word "mail-in" and just typed out a response.

Is there a small amount of fraud? It wouldn't surprise me at all to find some in every election.

How about a thought experiment: If you wanted to do it on a large enough scale to swing an election, How would you do it? Say you knew that there are three states which are going to be very close, and you wanted to add 10000 votes for your preferred candidate in each state. What would you do?

Excellent question. I don't know the answer. That's why I think we need a "red team" or "bug bounty" program to investigate.

How many ballots can fit in a mail truck? What does the chain of custody look like for all the different ways people can vote? What does the process look like when that chain of custody is suspected to have been broken? We've all seen how easily technological systems can be compromised through social engineering. Are there any gaps in the processes that could be exploited?

It is very easy but more costly than a recount to verify the integrity of an election.

Auditing relies on the assumption that real voters can be linked to attestations and the aggregate of attestations can be linked to the aggregate of votes.

In Australia when you vote in person your named is checked off on the electoral roll. This is an attestations. When you submit an absentee ballot you also fill out a form. This is an attestation as well.

The audit does a full recount of attestations and votes and verifies the totals are the same. Duplicate checking is also performed on attestations.

Furthermore, the attestations are randomly sampled and the real voters are interviewed to check if they actually voted. In the case of absentee ballots the details on the form can be checked as well. The random sample is performed so you have an ‘acceptable’ level of fraud and then you sample enough ballots such that if you were higher than the acceptable level it would be likely to sample a fraudulent ballot.

This process breaks down a bit at this point because of the issue of false positives where the voter cannot be contacted. However, with more work from the auditor it is often possible to distinguish between the voter being uncontactable and voting and the voter not existing, or being uncontactable and not voting.

Because the audit process establishes a trail from real voters to real ballots it is very difficult to commit electoral fraud.

There are some forms of electoral fraud this audit will not detect like bribery and coercion. It also doesn’t prevent substitution attacks where someone destroys real ballots and replaces them with dodgy ones.

It might be tougher in my state because we allowed ballots that were received by mail on Election Day, but without a postmark. One of the accusations of voter fraud includes backdating the postmarks on some number of ballots so they would be counted.

Obviously postmarks weren't created or intended to be cryptographic verification of when something was received anyway.

Well for example there are a more than a few definitely dead people who vote each election, that's pretty easy to show. But finding the, uh, living person who did the fraud is not always easy.

The "dead people" vote is a myth spawned as a result of technical issues, and is noted and discredited in voter fraud studies. State run voter rolls often didn't require DOB in past or continue to leave it optional in certain circumstances, causing the DOB to be represented as 0 AD, or 1970 AD, or some other spurious result. Even the Steve Bannon funded GAI study on voter fraud[1], which has been discredited, made this claim and refuted this data from the study. This is disinformation.

[1] http://www.g-a-i.org/wp-content/uploads/2017/07/Voter-Fraud-...

Here’s one with a DOB that matches Joe Frazier’s actual DOB: https://www.bigtrial.net/2020/11/smokin-joe-frazier-among-22...

And look, I’m not saying this is a significant factor or anything, but we’re arguing about it now so whatever.

Dead people will vote in every election that allows absentee ballots, because people die after they mail their ballot but before election day. Atomicity. It's a bitch.

Every state has an electronic pollbook that tracks voter eligibility. Votes for dead people will arrive, but they should not as a rule be counted.

There are definitely a few dead people who register to vote after dying, and successfully vote, in every election too.

Then you can definitely name one.

The funniest one I’ve heard so far is Joe Frazier. The birthday apparently matches, but this may have to wait to be confirmed in court: https://www.bigtrial.net/2020/11/smokin-joe-frazier-among-22...

Never heard of this source - I hope it’s okay in this time of misinformation to ask for a recognizable source?

Edit: I mean I am okay with leaving it as “somebody on the internet says”, but wondered if the evidence was more clear than that.

Not much else for that one online, Giuliani cited it and it’s funny. The other one he cited was a woman who registered to vote the day after she died:

https://www.swgfuneralhome.com/obituary/Denise-Ondick

https://twitter.com/clewandowski_/status/1325256319176224770...

The tweet has a screenshot, and unless they’ve already corrected it, you can probably find it on the PA website yourself.

Okay, so your source is Giuliani. Well that is an answer, I guess. Sorry I was looking for a news source and not a political one. This is not nearly as credible - until evidence reaches a courtroom - as “somebody on the internet”.

How many on average?

Not many, surely. Perhaps even single digits.

It seems disingenuous to give a legitimate reason for an anomaly but then follow it with a claim that it’s fraud:

> By 4am the D to R ratio was all thrown out of whack. That is because these ballots were not sampled from the real Wisconsin voter population, and they were not randomized in the mail sorting system with the other ballots. They inherently have a different D to R signature... than the rest of the ballots quite possibly bc additional ballots were added to the batch, either through backdating or ballot manufacturing or software tampering.

Do we know that mail in ballots are randomly shuffled in the mail? I know some states specifically pull aside ballots cast after certain dates, which could easily explain the differences in the graphs.

> Again, this should not happen, and it is observed almost nowhere else in the country, because all of the ballots are randomly shuffled...in the mail system and should be homogeneous during counting.

No, they aren't. They aren't shuffled by time mailed (which may have significant D/R differences) and they aren't shuffled across counties of the state in places that have them mailed to the voters county election office and counted in that county with the results then reported to the state, the same as other ballots are.

>CulturalHusbandry @APhilosophae

>The following information is provided via an anonymous data scientist and another anonymous individual who wrote a script to scrape the national ballot counting time series data of off the @nytimes website.

>This is based on their proprietary "Edison" data source which would ordinarily be impossible to access for people outside the press.

https://twitter.com/APhilosophae/status/1325592112428163072

Why should we trust these anonymous people (including twitter thread OP)? They could just tamper with the data and we would have no way to know that.

Take into account the pinned thread by @APhilosophae too, they are not impartial at all:

https://twitter.com/APhilosophae/status/1325135291791839232

Well that's why I ported and ran the python script, and put it on GitHub, so you could re-scrape the NYT data too, it will just overwrite what's in the repo. I'm not vouching for any of the corresponding analysis, but the data seems real: https://github.com/guscost/nyt-election-2020-analysis

Ah I misunderstood, I thought one can't repeat the scraping part. But what did they mean by "proprietary "Edison" data source which would ordinarily be impossible to access for people outside the press" then?

Obviously, the bigger problem with trusting this Twitter rando is that they don't know how mail-in ballots are county, came up with weird broken premises, and then accused Wisconsin, Georgia, and Pennsylvania of perpetrating a gigantic fraud.

From my understanding most of the press data sources are just companies that wrote code to scrape the multiple state sources of data and/or interface with archaic fax systems that sent counts from counting centers to central locations. I believe in some cases they had observers who took totals off the counting machines themselves.

I believe the primary reason for them is that they got the data faster than the official websites.

Thanks, now I understood that "this is based on ..." means "this script is based on ..." and I initially parsed that as "this information is based on ...".

This is extremely poor analysis. All the "anomalies" are explained by the fact that the ballots came from Dem heavy areas...

And even if you accept that some shuffling occurs in the mail system, it's not nearly enough to homogenize the entire mail-in ballot intake.

Certain sorting centers (i.e. the first 3 digits of a collection of ZIP codes) are still going to have drastically different demographics than others. So as trays of mail arrive, and are stacked up in the tabulation centers, they're likely to stay clumped together.

It is completely unsurprising that a very large batch of mail-in ballots, arriving in one transport, are going to show a high ratio of D/R votes compared to other batches of mail-in votes. Knowing where the voters in each batch live will explain the differences.

Here's some more of their trenchant analysis:

https://twitter.com/APhilosophae/status/1325148807546376193

They're showing results from Prince William County in Virginia to say, a-ha, a huge number of people voted for Biden but didn't vote at all for a downballot race: clearly a sign of fraud, irregularities, malfeasance, perhaps even skullduggery!

Except the image doesn't show the titles of those races: the second is obviously for President, and the first, with fewer votes, is... for the representative for the first Congressional district. There are three different Congressional districts that are partially within Prince William County.

These aren't nefarious, suspicious Biden-only votes, they're just showing one race that's on everybody's ballot and comparing it to another that isn't. The count of reporting precincts also seems to have been edited for the listing of Presidential votes: it shows 44 in the image instead of the actual number of 94.

You could chalk these things up to mistakes or misunderstandings, but judging by the content of the rest of this person's messaging I think it's pretty clear that assuming good faith isn't warranted here.

(edited to fix Twitter link... I think)

This is a really poor understanding of several things, including, for instance, how the USPS actually handled the ballots. This "randomly shuffled" thing is flat wrong.

There's actually a lot of really good data about how many mail ballots were requested in each state and the return rate. Many of the states even have data about which party did a better job returning the ballots.

https://electproject.github.io/Early-Vote-2020G/index.html

Several of the "modelers" out there used such data to basically know by ~1am Wednesday morning that Biden would likely win PA, for instance. It was just a matter of time.

If states piled up the ballots in the order they arrived, that would mean later arriving votes would be counted before early arriving votes.

Then if one party had a propensity to vote by mail earlier than the other we'd see a pattern where the proportion from that party would increase over time.

That's what we see.

Can we verify this any other way? Not completely, but in Florida survey data shows that the Democrats had an early lead of 462,000 before early (non-mail) voting started and by election day that had decreased to ~100K[1]

[1] https://thehill.com/homenews/campaign/524085-unprecedented-e...

Here's a reuters article that fact checks the weird voter spikes: https://www.reuters.com/article/uk-fact-check-biden-vote-spi...

Nice. You can get GitHub to do the scraping for you using GitHub actions. To better utilize git as a datastore, prettyprint it so each commit is a smaller diff.

https://simonwillison.net/2020/Oct/9/git-scraping/

> https://threadreaderapp.com/thread/1325592112428163072.html

Do you know why the scatter plots for SD and CA are showing a few fractional counts? Shouldn't they all be integers?

Reply on news.ycombinator.com