We really need to get some better election security.
Thats just terrifying. Although, did hacking technology just get way better? or is there some kind of second order protection in place here (IE< maybe hackers just don't know what to hack to modify voting results), because if it genuinely was this trivial why did the Russians bother with all the propaganda?
If you just change the votes without generating a plausible social movement to account for the weird vote totals people start to question the results more readily.
It's why if you're going to fuck with vote totals you do very little manipulation and you pair it with a propaganda campaign that makes the result look like the results of a motivated wild card element.
Because hacking is easy when you have access. And hacking machines, while do-able is rather pointless when you can hack the count hardware.
I mean. If we are going to believe that the votes are hacked* then it’s done in counting.
*and they probably are. The statistical evidence is pretty overwhelming and there is no viable explanation for the correlations seen. Though if there is I would love to hear it because I want to be wrong here
We really need to get some better election security.
Thats just terrifying. Although, did hacking technology just get way better? or is there some kind of second order protection in place here (IE< maybe hackers just don't know what to hack to modify voting results), because if it genuinely was this trivial why did the Russians bother with all the propaganda?
If you just change the votes without generating a plausible social movement to account for the weird vote totals people start to question the results more readily.
Yeah, but couldn't you just hack in, find out the true numbers, and give every seat nationwide like 0.4-.6% more Republican votes and 0.4-.6% less Democratic ones, unless the final number you were giving out would cause a recount. +- a fraction of a percent, even nationwide is within errors for the vote totals. Even with recounts you probably wouldn't get spotted
And now I've convinced myself that they absolutely did that too... Urgh, I bet you could even make it pretty much statistically untraceable.
We really need to get some better election security.
Thats just terrifying. Although, did hacking technology just get way better? or is there some kind of second order protection in place here (IE< maybe hackers just don't know what to hack to modify voting results), because if it genuinely was this trivial why did the Russians bother with all the propaganda?
Almost all of the hacks required physical access to the machine. Some involved unscrewing the back panel.
This does not excuse some of the more egregious security failures. VoteActive Winvote machines in particular were exploited wirelessly and were already running music playing software of Chinese origin. If I recall correctly from prior articles, these machines are still running Windows XP. As far as I can tell, VoteActive Winvote is almost comical how bad they are.
There were other failures on other platforms like storing passwords in cleartext.
However, "hacks" requiring disassembly of a machine should be taken with a grain of salt. Every system can be "hacked" if you have unsupervised physical access to the internals.
every person who doesn't like an acquired taste always seems to think everyone who likes it is faking it. it should be an official fallacy.
Also that article last page about was changing the web site, not the votes.
I really want to see some more details about the r00tz activity. According to the tweet, using a SQL injection on a replica of a state Secretary of State website allowed them to change votes. I really want to see how the replicas are set up and how similar they are to state SoS websites in production.
every person who doesn't like an acquired taste always seems to think everyone who likes it is faking it. it should be an official fallacy.
We really need to get some better election security.
Thats just terrifying. Although, did hacking technology just get way better? or is there some kind of second order protection in place here (IE< maybe hackers just don't know what to hack to modify voting results), because if it genuinely was this trivial why did the Russians bother with all the propaganda?
Almost all of the hacks required physical access to the machine. Some involved unscrewing the back panel.
This does not excuse some of the more egregious security failures. VoteActive Winvote machines in particular were exploited wirelessly and were already running music playing software of Chinese origin. If I recall correctly from prior articles, these machines are still running Windows XP. As far as I can tell, VoteActive Winvote is almost comical how bad they are.
There were other failures on other platforms like storing passwords in cleartext.
However, "hacks" requiring disassembly of a machine should be taken with a grain of salt. Every system can be "hacked" if you have unsupervised physical access to the internals.
Ahh OK, so mainly physical access protection levels. Which are decent 'real' protection.
Urgh, unless the hacks happened at the places where the voting machines are stored.
The voting machines are not really a problem. There is no point to hack them because they're not counting the votes. Its hella inefficient unless you're doing it at the base level (I.E. you're inside the companies that provide them) and even then still pretty inefficient.
Like, vote totals were still hacked in 2016, but they weren't done though the machines.
The voting machines are not really a problem. There is no point to hack them because they're not counting the votes. Its hella inefficient unless you're doing it at the base level (I.E. you're inside the companies that provide them) and even then still pretty inefficient.
Like, vote totals were still hacked in 2016, but they weren't done though the machines.
Vulnerabilities that can be exploited wirelessly are a significant problem. (I'm less concerned about vulnerabilities that require physical access.)
Sure, you're not going to swing a Presidential election through voting machines vulnerabilities but you could swing a local election or a tight Congressional district.
every person who doesn't like an acquired taste always seems to think everyone who likes it is faking it. it should be an official fallacy.
The voting machines are not really a problem. There is no point to hack them because they're not counting the votes. Its hella inefficient unless you're doing it at the base level (I.E. you're inside the companies that provide them) and even then still pretty inefficient.
Like, vote totals were still hacked in 2016, but they weren't done though the machines.
Vulnerabilities that can be exploited wirelessly are a significant problem. (I'm less concerned about vulnerabilities that require physical access.)
Sure, you're not going to swing a Presidential election through voting machines vulnerabilities but you could swing a local election or a tight Congressional district.
You cannot flip or add votes in small precincts. This is because its easy to confirm when small precincts return incorrect numbers. When 50 people vote in a precinct and they all know each other then taking votes there makes it easy to get caught. Similarly its harder to stuff because people will notice variations in turnout when everyone knows everyone else. As such as the precinct size goes up the number of votes you can steal goes up. Such if we expect fraud then we should find that vote % correlates with precinct size. And similarly if we expect no fraud then we should expect that vote % does not correlate with precinct size OR correlates slightly democratic*.
This is because well, aside from fraud, there isn't any reason why the vote % should correlate with anything besides demographic data and so any correlation we find between non-demographic data and vote % should be the result of omitted variable bias if there is no fraud. OVB produces an effect with the omitted variable correlates with the potentially biased variable and the determined variable(and the biased variable is left out of a regression). Which is to say that if precinct sized correlated with whiteness then we would have a reasonable finding for so many republican districts producing such high red shift.
However i sure and shit cant come up with an omitted variable that should go in there.(though i don't have access to the actual data in order to make a proper determination, and the web sites statistical analysis is not particularly thorough**)
*this is because we expect larger precincts to be in more dense locations which should tend to swing democratic.
**Ideally you would want to do two things before declaring an issue 1) You would want to control for demographic data as well as precinct size; this negates OVB 2) You would want to check the distribution of your measurement over the entire sample; this prevents random correlations from showing up due to selection bias.
It reads to me like they were hacking the public facing Sec State website, i.e. a place where vote results are published, not the actual repository of canonical vote tallies.
Still pretty bad, but I wouldn't expect a site like that to have the same level of security as the actual voter database.
the whole article is pretty light on the info
life's a game that you're bound to lose / like using a hammer to pound in screws
fuck up once and you break your thumb / if you're happy at all then you're god damn dumb
that's right we're on a fucked up cruise / God is dead but at least we have booze
bad things happen, no one knows why / the sun burns out and everyone dies
It reads to me like they were hacking the public facing Sec State website, i.e. a place where vote results are published, not the actual repository of canonical vote tallies.
Still pretty bad, but I wouldn't expect a site like that to have the same level of security as the actual voter database.
the whole article is pretty light on the info
Yeah but it's enough. I think that's the info the press reports on, which can have a big effect on people turning out later in the day.
The voting machines are not really a problem. There is no point to hack them because they're not counting the votes. Its hella inefficient unless you're doing it at the base level (I.E. you're inside the companies that provide them) and even then still pretty inefficient.
Like, vote totals were still hacked in 2016, but they weren't done though the machines.
Vulnerabilities that can be exploited wirelessly are a significant problem. (I'm less concerned about vulnerabilities that require physical access.)
Sure, you're not going to swing a Presidential election through voting machines vulnerabilities but you could swing a local election or a tight Congressional district.
You cannot flip or add votes in small precincts. This is because its easy to confirm when small precincts return incorrect numbers. When 50 people vote in a precinct and they all know each other then taking votes there makes it easy to get caught. Similarly its harder to stuff because people will notice variations in turnout when everyone knows everyone else. As such as the precinct size goes up the number of votes you can steal goes up. Such if we expect fraud then we should find that vote % correlates with precinct size. And similarly if we expect no fraud then we should expect that vote % does not correlate with precinct size OR correlates slightly democratic*.
This is because well, aside from fraud, there isn't any reason why the vote % should correlate with anything besides demographic data and so any correlation we find between non-demographic data and vote % should be the result of omitted variable bias if there is no fraud. OVB produces an effect with the omitted variable correlates with the potentially biased variable and the determined variable(and the biased variable is left out of a regression). Which is to say that if precinct sized correlated with whiteness then we would have a reasonable finding for so many republican districts producing such high red shift.
However i sure and shit cant come up with an omitted variable that should go in there.(though i don't have access to the actual data in order to make a proper determination, and the web sites statistical analysis is not particularly thorough**)
*this is because we expect larger precincts to be in more dense locations which should tend to swing democratic.
**Ideally you would want to do two things before declaring an issue 1) You would want to control for demographic data as well as precinct size; this negates OVB 2) You would want to check the distribution of your measurement over the entire sample; this prevents random correlations from showing up due to selection bias.
If there isn't like some counterfactor here (like, a massive correlation between number of votes cast and how urban the district is) then the data shown on that website (if true) is 100% absolutely convincing evidence that the election was stolen. Better evidence than a direct confession from Trump that he stole the election. If districts were fairly assigned and contained random numbers of voters then it is completely impossible that the results would look like that.
The evidence in your link is utterly convincing if they didn't just make it up.
edit - Christ, they did it in all the other 'wow I can't believe the Republicans won that state' states too. Unless they made the data up, or there's some giant correlation they are missing then our democracy has been stolen.
The voting machines are not really a problem. There is no point to hack them because they're not counting the votes. Its hella inefficient unless you're doing it at the base level (I.E. you're inside the companies that provide them) and even then still pretty inefficient.
Like, vote totals were still hacked in 2016, but they weren't done though the machines.
Vulnerabilities that can be exploited wirelessly are a significant problem. (I'm less concerned about vulnerabilities that require physical access.)
Sure, you're not going to swing a Presidential election through voting machines vulnerabilities but you could swing a local election or a tight Congressional district.
You cannot flip or add votes in small precincts. This is because its easy to confirm when small precincts return incorrect numbers. When 50 people vote in a precinct and they all know each other then taking votes there makes it easy to get caught. Similarly its harder to stuff because people will notice variations in turnout when everyone knows everyone else. As such as the precinct size goes up the number of votes you can steal goes up. Such if we expect fraud then we should find that vote % correlates with precinct size. And similarly if we expect no fraud then we should expect that vote % does not correlate with precinct size OR correlates slightly democratic*.
This is because well, aside from fraud, there isn't any reason why the vote % should correlate with anything besides demographic data and so any correlation we find between non-demographic data and vote % should be the result of omitted variable bias if there is no fraud. OVB produces an effect with the omitted variable correlates with the potentially biased variable and the determined variable(and the biased variable is left out of a regression). Which is to say that if precinct sized correlated with whiteness then we would have a reasonable finding for so many republican districts producing such high red shift.
However i sure and shit cant come up with an omitted variable that should go in there.(though i don't have access to the actual data in order to make a proper determination, and the web sites statistical analysis is not particularly thorough**)
*this is because we expect larger precincts to be in more dense locations which should tend to swing democratic.
**Ideally you would want to do two things before declaring an issue 1) You would want to control for demographic data as well as precinct size; this negates OVB 2) You would want to check the distribution of your measurement over the entire sample; this prevents random correlations from showing up due to selection bias.
If there isn't like some counterfactor here (like, a massive correlation between number of votes cast and how urban the district is) then the data shown on that website (if true) is 100% absolutely convincing evidence that the election was stolen. Better evidence than a direct confession from Trump that he stole the election. If districts were fairly assigned and contained random numbers of voters then it is completely impossible that the results would look like that.
The evidence in your link is utterly convincing if they didn't just make it up.
edit - Christ, they did it in all the other 'wow I can't believe the Republicans won that state' states too. Unless they made the data up, or there's some giant correlation they are missing then our democracy has been stolen.
This is something which has been happening in a bunch of precincts the country over, for some time now. First time I saw the analysis was in a primary in Kansas in like 2010 or something like that, if memory serves.
It is built on one fundamental assumption: that precinct size (by population) should have no impact on the percent of votes expected to go to one candidate over another. No defense of this assumption is made, and I'm not convinced that it's a safe assumption to make. Here's a study (warning, PDF) which shows this same bias in 20 different states. It seems more consistent in Democratic precincts than in Republican precincts, but the fact that it seems to be SO universal seems to imply to me that there's a not-completely-understood bias which is introduced via how we tend to draw our precincts.
Precincts don't contain random numbers of voters* and cannot be apportioned unfairly. Precincts do not cross district boundaries but generally only exist for the purpose of making voting easier(everyone in the precinct gets the same ballot) so people don't swamp different voting locations. There are more precincts in high population dense areas and more people in those precincts. Hence we should expect that "large number of people in the precinct" would correlate to "democratic vote %" rather than the other way around if there was going to be a correlation. However the link between precinct size and urbanity isn't as large as we can identify by looking at specific strongholds or demographic effects because even rural areas will have places with large precincts.
Its also possible that such bias could be the result of massive systematic disenfranchisement of people who live in large precinct areas and tend to vote democratic. This could happen in a number of ways which should be fairly obvious. However i don't really see that kind of effect say happening in say Pierce County Washington, which is approximately 150% white.
*not in the colloquial sense or in the sense that matters for statistics. But its not actually necessary that they are for this analysis. Simply that we cannot identify a casual omitted variable such that the effect can be explained by actual voting patterns.
R goes from about 35% to 58% while D goes from 60% to below 40%!
However, registrations are like 52:32 so it's much more likely that the high D results from small precincts are the aberrations. This is roughly true of the other counties they show, the D vote share starts out wildly ahead of registrations in smaller precincts - so we would expect a regression to the expected values quickly as sizes scale up
That pdf Jragghen linked to has an analysis for PA 2008 and it shows almost exactly the same bias, tiny precincts are about 70-80% D while larger ones were nearly even and this is when D won by 10% (see figure 2)
R goes from about 35% to 58% while D goes from 60% to below 40%!
However, registrations are like 52:32 so it's much more likely that the high D results from small precincts are the aberrations. This is roughly true of the other counties they show, the D vote share starts out wildly ahead of registrations in smaller precincts - so we would expect a regression to the expected values quickly as sizes scale up
That pdf Jragghen linked to has an analysis for PA 2008 and it shows almost exactly the same bias, tiny precincts are about 70-80% D while larger ones were nearly even and this is when D won by 10% (see figure 2)
Yes there were irregularities in 2008 as well. It’s still not explainable unless for some reason the cities in PA are surprisingly Republican. While there should be mean regression the mean regression is happening too slowly. If it shot up to the value very fast then leveled off then things would be less indicative of bias.
Counter explanations are unlikely. Either small precincts all happened to vote Democratic randomly or democrats have a rural advantage. Neither make sense.
This is what the graphs should look like
While this produces a slight democratic bias (4000 vote count) it’s pretty close to what you should expect. Washington overall produces a graph like that with a stronger democratic bias. (50k total votes). Which is to be expected since cities are likely to vote democratic and cities dominate the larger precinct areas. But it still would be fair to say that there is evidence of tampering (it’s just much weaker since it’s 1/5th the variation) if you were going strictly on a non-zero delta.
Earlier I mentioned Pierce county I was misremembering it was Kititas that was off in WA
These results are more interesting when compared to normal counties. For example, in Washington State, the county results either have little variation with precinct size, or a slight boost to Democratic votes with increasing size. (Presumably due to hitting urban areas.) All counties except Kittitas County. Kittitas County has the same increase in Republican votes as the precinct size increases.
I checked, and though Washington is vote by mail, the counting is done at the county level, so it's at least possible for a county to be screwing with the vote. However, the precinct lines are also handled at the county level, so it's also possible for this one county to have done something weird.
“Doing something with precinct lines” makes no sense. You gain no benefit unless you’re cramming people into precincts so as to make it impossible to vote at them.
Counting is done by machine at the county level yes. The same way it’s counted in any other place with paper ballots.
R goes from about 35% to 58% while D goes from 60% to below 40%!
However, registrations are like 52:32 so it's much more likely that the high D results from small precincts are the aberrations. This is roughly true of the other counties they show, the D vote share starts out wildly ahead of registrations in smaller precincts - so we would expect a regression to the expected values quickly as sizes scale up
That pdf Jragghen linked to has an analysis for PA 2008 and it shows almost exactly the same bias, tiny precincts are about 70-80% D while larger ones were nearly even and this is when D won by 10% (see figure 2)
Yes there were irregularities in 2008 as well. It’s still not explainable unless for some reason the cities in PA are surprisingly Republican. While there should be mean regression the mean regression is happening too slowly. If it shot up to the value very fast then leveled off then things would be less indicative of bias.
Counter explanations are unlikely. Either small precincts all happened to vote Democratic randomly or democrats have a rural advantage. Neither make sense.
This is what the graphs should look like
While this produces a slight democratic bias (4000 vote count) it’s pretty close to what you should expect. Washington overall produces a graph like that with a stronger democratic bias. (50k total votes). Which is to be expected since cities are likely to vote democratic and cities dominate the larger precinct areas. But it still would be fair to say that there is evidence of tampering (it’s just much weaker since it’s 1/5th the variation) if you were going strictly on a non-zero delta.
Earlier I mentioned Pierce county I was misremembering it was Kititas that was off in WA
There are lots of implicit assumptions in the phrase "This is what the graphs should look like"
I looked up the map of lancaster county which went entirely R except for Lancaster itself - the city. And coincidentally, the city has tons of tiny precincts with <600 votes cast, some are just a few square blocks. So that's a situation where precinct sizes appear to be directly correlated with party affiliation
Its true. We spend all day finding the way this shit goes wrong (or creating those ways). We all know its a bad idea, except maybe the SV tech utopia zealots
We really need to get some better election security.
Thats just terrifying. Although, did hacking technology just get way better? or is there some kind of second order protection in place here (IE< maybe hackers just don't know what to hack to modify voting results), because if it genuinely was this trivial why did the Russians bother with all the propaganda?
I strongly suspect, though the article isn't explicit, that he changed the website not the underlying db
edit
nm they do make it explicit in the 2nd half
“It would be extremely difficult to replicate these systems since many states utilize unique networks and custom-built databases with new and updated security protocols,” it read. “While it is undeniable websites are vulnerable to hackers, election night reporting websites are only used to publish preliminary, unofficial results for the public and the media. The sites are not connected to vote counting equipment and could never change actual election results.”’
Well, you didn't look for one and then dismissed one out of hand when it was provided as "irregularities". That site didn't bother looking either or dismissed it, which makes me skeptical of their other results
And when you go to the larger statewide graph it does look like the one you posted! There are plenty of heavily R small rural precincts, just not immediately outside of the cities so they show up in other counties
By comparison in WA, Lincoln Country, right next to Spokane has something like 25 precincts and 5000 voters, Garfield County to the south has 6 for 1000 voters
That site also has a problem in that they are claiming fraud happened in states that use paper voting so I guess there is a grand conspiracy
Well, you didn't look for one and then dismissed one out of hand when it was provided as "irregularities". That site didn't bother looking either or dismissed it, which makes me skeptical of their other results
And when you go to the larger statewide graph it does look like the one you posted! There are plenty of heavily R small rural precincts, just not immediately outside of the cities so they show up in other counties
By comparison in WA, Lincoln Country, right next to Spokane has something like 25 precincts and 5000 voters, Garfield County to the south has 6 for 1000 voters
That site also has a problem in that they are claiming fraud happened in states that use paper voting so I guess there is a grand conspiracy
I think if you say there are voting irregularities every year.
Well, you didn't look for one and then dismissed one out of hand when it was provided as "irregularities". That site didn't bother looking either or dismissed it, which makes me skeptical of their other results
And when you go to the larger statewide graph it does look like the one you posted! There are plenty of heavily R small rural precincts, just not immediately outside of the cities so they show up in other counties
By comparison in WA, Lincoln Country, right next to Spokane has something like 25 precincts and 5000 voters, Garfield County to the south has 6 for 1000 voters
That site also has a problem in that they are claiming fraud happened in states that use paper voting so I guess there is a grand conspiracy
I think if you say there are voting irregularities every year.
Manor Twp - Washington Boro Dist on the other side. Should be on the far left of the chart but went to Trump close to 5:1.
The chart is far too clean to be raw data, but far too dirty to be some kind of regression function - the combination of these facts is a classic red flag for invented data, as it's exactly the way our brains fail at math.
“Doing something with precinct lines” makes no sense. You gain no benefit unless you’re cramming people into precincts so as to make it impossible to vote at them.
Counting is done by machine at the county level yes. The same way it’s counted in any other place with paper ballots.
The random sort on Lancaster is basically a flat line. Why is so unlikely that the larger districts would be rural?
Further, why are you accepting the conclusion of "vote manipulation" from a source that has not demonstrated the key point that the vote split in those specific precincts has no simpler explanation?
“Doing something with precinct lines” makes no sense. You gain no benefit unless you’re cramming people into precincts so as to make it impossible to vote at them.
Counting is done by machine at the county level yes. The same way it’s counted in any other place with paper ballots.
The random sort on Lancaster is basically a flat line. Why is so unlikely that the larger districts would be rural?
Further, why are you accepting the conclusion of "vote manipulation" from a source that has not demonstrated the key point that the vote split in those specific precincts has no simpler explanation?
Random sort will basically always be a flat line. That is what a random sort does. The fact that you do not have a flat line in an ordered sort is how you see correlation. If there is no correlation then whichever sort you choose should look like a random sort because a random sort can only have correlation randomly*.
Its unlikely that larger precincts would be rural (in general) because population density should tend to follow precinct size because the primary consideration in precincts is minimizing distance to the voting locations.
I am not accepting it because they have not demonstrated there is no simpler explanation but because i cannot find a more correct explanation. If i had a reason to say "ahh yes, it makes perfect sense that small precincts are in the dense urban areas and large precincts are in rural areas" then we would be golden. Or if i had the omitted variable. The inability to make said determination is more or less how all statistical inference works (and really therefore all science**)
We might even be fine if we saw large instances of vote shifting for the dems, as it would make sense if there was some symmetric issue here. But when we see vote shifting for dems its comparatively minor. I am not sure why we should only see red shift when we see shifts and why those aren't symmetric across similarly demographic areas. Why do we have this problem in Kititas and not Spokane? Or Yakima? Or Chelan? Or Skagit? Or Whatcom? They're all pretty similarly geographically structured. (Large rural population with a single small city core)
*Specifically if you take a stationary random variable and correlate it to anything the correlation coefficient will be an random variable with a center of zero and a variance equal to... honesty i forgot and don't want to do the math but i am pretty sure if its not the product of the variance of the random variable and the determined variable its a similar function. A random sort is a stationary random variable.
**The best theory is the one that fits the evidence and for which no better alternate theory exists. You don't "prove" that no better alternate theory exists because doing so is equivalent to proving the theory. It would not longer be a theory at that point but a law or an axiom; it couldn't not be true.
Well, you didn't look for one and then dismissed one out of hand when it was provided as "irregularities". That site didn't bother looking either or dismissed it, which makes me skeptical of their other results
And when you go to the larger statewide graph it does look like the one you posted! There are plenty of heavily R small rural precincts, just not immediately outside of the cities so they show up in other counties
By comparison in WA, Lincoln Country, right next to Spokane has something like 25 precincts and 5000 voters, Garfield County to the south has 6 for 1000 voters
That site also has a problem in that they are claiming fraud happened in states that use paper voting so I guess there is a grand conspiracy
I think if you say there are voting irregularities every year.
Manor Twp - Washington Boro Dist on the other side. Should be on the far left of the chart but went to Trump close to 5:1.
The chart is far too clean to be raw data, but far too dirty to be some kind of regression function - the combination of these facts is a classic red flag for invented data, as it's exactly the way our brains fail at math.
No. The chart has vote% (not count) on one side and precinct size on the other. It should be smooth because even large total count changes at the end of the count produce a very small change in vote %. This is the law of large numbers in effect.
E.G. Suppose you flip two coins one lands heads 50% of the time and one lands heads 100% of the time and you flip them each 50% of the time.
If you randomly choose which coin to flip this coin looks precisely like a single coin that lands heads 75% of the time. Indeed there isn't even a difference mathematically.
If you order the coin flips so that you flip the 50% coin first 500 times and then the 100% coin 500 times then you will see a wiggly line quickly approach 50% and then once you hit the 100% coin it will veer off towards 75%. For an easy way to see this you can look at the 500th flip where you should have about 250 heads/500 flips so lets just say that is what you have. When you get to the 501th flip you will have 251/501 and then and then 252/502... and then eventually 750/1000.
edit: various edits to correct incorrect numbers and add an explainer on why the graphs look like they do
The best theory is the one that fits the evidence and for which no better alternate theory exists. You don't "prove" that no better alternate theory exists because doing so is equivalent to proving the theory. It would not longer be a theory at that point but a law or an axiom; it couldn't not be true.
The best theory that fits the evidence is that these larger districts that went Trump were rural.
Because they are, per the github data source. There are large districts that went Clinton that have been filtered out of the visualization, as well as smaller ones that went Trump. The large district for C are in Lancaster, the ones for Trump are outside the city.
Nothing was filtered out of the visualization. The smoothness is a function of the law of large numbers and would only see a spike if a particular district size had a hilariously large number of people in it (E.G. If you look at the numbers of Washington, King County does not report individual precincts and so produces a hilarious jump at the end of the graph when all 1 million votes got added in, nearly 32% of the total votes in the election)
Nothing was filtered out of the visualization. The smoothness is a function of the law of large numbers and would only see a spike if a particular district size had a hilariously large number of people in it (E.G. If you look at the numbers of Washington, King County does not report individual precincts and so produces a hilarious jump at the end of the graph when all 1 million votes got added in, nearly 32% of the total votes in the election)
Ah, I misunderstood the chart, turned on vote bubbles, saw what I expected.
The rural precincts containing more voters seems the most obvious explanation here. The city of Lancaster comprises ~20% (47/241) of the precincts and 10% (59k/536k) of the county population. It stands to reason that the average population of the outlying precincts is larger.
“While it is undeniable websites are vulnerable to hackers, election night reporting websites are only used to publish preliminary, unofficial results for the public and the media. The sites are not connected to vote counting equipment and could never change actual election results.”’
Yeah but the public-facing preliminary results are what politicians use to decide whether or not to concede...
“While it is undeniable websites are vulnerable to hackers, election night reporting websites are only used to publish preliminary, unofficial results for the public and the media. The sites are not connected to vote counting equipment and could never change actual election results.”’
Yeah but the public-facing preliminary results are what politicians use to decide whether or not to concede...
Or more importantly, some people may decide their vote won't matter based on the preliminary reports, so they don't bother. Change the vote by discouraging the voters with false information.
I am not accepting it because they have not demonstrated there is no simpler explanation but because i cannot find a more correct explanation. If i had a reason to say "ahh yes, it makes perfect sense that small precincts are in the dense urban areas and large precincts are in rural areas" then we would be golden. Or if i had the omitted variable. The inability to make said determination is more or less how all statistical inference works (and really therefore all science**)
But that's the thing, if you actually looked at the NYTimes link Phyphor posted, and the returns for Lancaster county that PantsB posted, you'd see exactly that. The denser urban areas have much smaller, more granular precincts, while the more rural (but not totally rural, so there's a higher population still) areas around the cities had much larger precincts. Which makes sense on an infrastructure scale too. In dense urban areas, it's easy to set up lots of little polling places, especially for a lot of urban city dwellers who might not have access to a car, and rely on walking or public transit to access a polling place. However in the areas surrounding a city that are more rural, everyone has to drive anyways, and what's the difference that an additional 2-3 minute drive would make?
Posts
If you just change the votes without generating a plausible social movement to account for the weird vote totals people start to question the results more readily.
It's why if you're going to fuck with vote totals you do very little manipulation and you pair it with a propaganda campaign that makes the result look like the results of a motivated wild card element.
I mean. If we are going to believe that the votes are hacked* then it’s done in counting.
*and they probably are. The statistical evidence is pretty overwhelming and there is no viable explanation for the correlations seen. Though if there is I would love to hear it because I want to be wrong here
Yeah, but couldn't you just hack in, find out the true numbers, and give every seat nationwide like 0.4-.6% more Republican votes and 0.4-.6% less Democratic ones, unless the final number you were giving out would cause a recount. +- a fraction of a percent, even nationwide is within errors for the vote totals. Even with recounts you probably wouldn't get spotted
And now I've convinced myself that they absolutely did that too... Urgh, I bet you could even make it pretty much statistically untraceable.
Almost all of the hacks required physical access to the machine. Some involved unscrewing the back panel.
This does not excuse some of the more egregious security failures. VoteActive Winvote machines in particular were exploited wirelessly and were already running music playing software of Chinese origin. If I recall correctly from prior articles, these machines are still running Windows XP. As far as I can tell, VoteActive Winvote is almost comical how bad they are.
There were other failures on other platforms like storing passwords in cleartext.
However, "hacks" requiring disassembly of a machine should be taken with a grain of salt. Every system can be "hacked" if you have unsupervised physical access to the internals.
the "no true scotch man" fallacy.
https://www.theregister.co.uk/2018/08/13/defcon_election_vote_hacking/
the "no true scotch man" fallacy.
I really want to see some more details about the r00tz activity. According to the tweet, using a SQL injection on a replica of a state Secretary of State website allowed them to change votes. I really want to see how the replicas are set up and how similar they are to state SoS websites in production.
the "no true scotch man" fallacy.
Ahh OK, so mainly physical access protection levels. Which are decent 'real' protection.
Urgh, unless the hacks happened at the places where the voting machines are stored.
Like, vote totals were still hacked in 2016, but they weren't done though the machines.
Vulnerabilities that can be exploited wirelessly are a significant problem. (I'm less concerned about vulnerabilities that require physical access.)
Sure, you're not going to swing a Presidential election through voting machines vulnerabilities but you could swing a local election or a tight Congressional district.
the "no true scotch man" fallacy.
I mean, yea. But as far as i can tell Trump got about to at least 500,000 votes more than he should have in PA.
A short explainer.
You cannot flip or add votes in small precincts. This is because its easy to confirm when small precincts return incorrect numbers. When 50 people vote in a precinct and they all know each other then taking votes there makes it easy to get caught. Similarly its harder to stuff because people will notice variations in turnout when everyone knows everyone else. As such as the precinct size goes up the number of votes you can steal goes up. Such if we expect fraud then we should find that vote % correlates with precinct size. And similarly if we expect no fraud then we should expect that vote % does not correlate with precinct size OR correlates slightly democratic*.
This is because well, aside from fraud, there isn't any reason why the vote % should correlate with anything besides demographic data and so any correlation we find between non-demographic data and vote % should be the result of omitted variable bias if there is no fraud. OVB produces an effect with the omitted variable correlates with the potentially biased variable and the determined variable(and the biased variable is left out of a regression). Which is to say that if precinct sized correlated with whiteness then we would have a reasonable finding for so many republican districts producing such high red shift.
However i sure and shit cant come up with an omitted variable that should go in there.(though i don't have access to the actual data in order to make a proper determination, and the web sites statistical analysis is not particularly thorough**)
*this is because we expect larger precincts to be in more dense locations which should tend to swing democratic.
**Ideally you would want to do two things before declaring an issue 1) You would want to control for demographic data as well as precinct size; this negates OVB 2) You would want to check the distribution of your measurement over the entire sample; this prevents random correlations from showing up due to selection bias.
It reads to me like they were hacking the public facing Sec State website, i.e. a place where vote results are published, not the actual repository of canonical vote tallies.
Still pretty bad, but I wouldn't expect a site like that to have the same level of security as the actual voter database.
the whole article is pretty light on the info
fuck up once and you break your thumb / if you're happy at all then you're god damn dumb
that's right we're on a fucked up cruise / God is dead but at least we have booze
bad things happen, no one knows why / the sun burns out and everyone dies
Yeah but it's enough. I think that's the info the press reports on, which can have a big effect on people turning out later in the day.
If there isn't like some counterfactor here (like, a massive correlation between number of votes cast and how urban the district is) then the data shown on that website (if true) is 100% absolutely convincing evidence that the election was stolen. Better evidence than a direct confession from Trump that he stole the election. If districts were fairly assigned and contained random numbers of voters then it is completely impossible that the results would look like that.
The evidence in your link is utterly convincing if they didn't just make it up.
edit - Christ, they did it in all the other 'wow I can't believe the Republicans won that state' states too. Unless they made the data up, or there's some giant correlation they are missing then our democracy has been stolen.
This is something which has been happening in a bunch of precincts the country over, for some time now. First time I saw the analysis was in a primary in Kansas in like 2010 or something like that, if memory serves.
It is built on one fundamental assumption: that precinct size (by population) should have no impact on the percent of votes expected to go to one candidate over another. No defense of this assumption is made, and I'm not convinced that it's a safe assumption to make. Here's a study (warning, PDF) which shows this same bias in 20 different states. It seems more consistent in Democratic precincts than in Republican precincts, but the fact that it seems to be SO universal seems to imply to me that there's a not-completely-understood bias which is introduced via how we tend to draw our precincts.
Its also possible that such bias could be the result of massive systematic disenfranchisement of people who live in large precinct areas and tend to vote democratic. This could happen in a number of ways which should be fairly obvious. However i don't really see that kind of effect say happening in say Pierce County Washington, which is approximately 150% white.
*not in the colloquial sense or in the sense that matters for statistics. But its not actually necessary that they are for this analysis. Simply that we cannot identify a casual omitted variable such that the effect can be explained by actual voting patterns.
R goes from about 35% to 58% while D goes from 60% to below 40%!
However, registrations are like 52:32 so it's much more likely that the high D results from small precincts are the aberrations. This is roughly true of the other counties they show, the D vote share starts out wildly ahead of registrations in smaller precincts - so we would expect a regression to the expected values quickly as sizes scale up
That pdf Jragghen linked to has an analysis for PA 2008 and it shows almost exactly the same bias, tiny precincts are about 70-80% D while larger ones were nearly even and this is when D won by 10% (see figure 2)
Yes there were irregularities in 2008 as well. It’s still not explainable unless for some reason the cities in PA are surprisingly Republican. While there should be mean regression the mean regression is happening too slowly. If it shot up to the value very fast then leveled off then things would be less indicative of bias.
Counter explanations are unlikely. Either small precincts all happened to vote Democratic randomly or democrats have a rural advantage. Neither make sense.
This is what the graphs should look like
While this produces a slight democratic bias (4000 vote count) it’s pretty close to what you should expect. Washington overall produces a graph like that with a stronger democratic bias. (50k total votes). Which is to be expected since cities are likely to vote democratic and cities dominate the larger precinct areas. But it still would be fair to say that there is evidence of tampering (it’s just much weaker since it’s 1/5th the variation) if you were going strictly on a non-zero delta.
Earlier I mentioned Pierce county I was misremembering it was Kititas that was off in WA
I checked, and though Washington is vote by mail, the counting is done at the county level, so it's at least possible for a county to be screwing with the vote. However, the precinct lines are also handled at the county level, so it's also possible for this one county to have done something weird.
Counting is done by machine at the county level yes. The same way it’s counted in any other place with paper ballots.
There are lots of implicit assumptions in the phrase "This is what the graphs should look like"
I looked up the map of lancaster county which went entirely R except for Lancaster itself - the city. And coincidentally, the city has tons of tiny precincts with <600 votes cast, some are just a few square blocks. So that's a situation where precinct sizes appear to be directly correlated with party affiliation
But those should go the other way in general.
Its true. We spend all day finding the way this shit goes wrong (or creating those ways). We all know its a bad idea, except maybe the SV tech utopia zealots
QEDMF xbl: PantsB G+
I strongly suspect, though the article isn't explicit, that he changed the website not the underlying db
edit
nm they do make it explicit in the 2nd half
QEDMF xbl: PantsB G+
No/ 1. Each precinct corresponds to a voting location. You vote at your precinct and not at any other
Precinct sizes are completely arbitrary. PA has cities with lots of small precincts - or fewer democrats voted in them making them smaller - relative to the rural ones, that's apparently just something they do, take a look https://www.nytimes.com/interactive/2018/upshot/election-2016-voting-precinct-maps.html
And when you go to the larger statewide graph it does look like the one you posted! There are plenty of heavily R small rural precincts, just not immediately outside of the cities so they show up in other counties
By comparison in WA, Lincoln Country, right next to Spokane has something like 25 precincts and 5000 voters, Garfield County to the south has 6 for 1000 voters
That site also has a problem in that they are claiming fraud happened in states that use paper voting so I guess there is a grand conspiracy
I think if you say there are voting irregularities every year.
Also its pretty straight forward to show that chart is made up. Here are the Lancaster county precinct level results for 2016 Lancaster City - 9th Ward - 4th Precinct alone shows that their data is invented.
QEDMF xbl: PantsB G+
Manor Twp - Washington Boro Dist on the other side. Should be on the far left of the chart but went to Trump close to 5:1.
The chart is far too clean to be raw data, but far too dirty to be some kind of regression function - the combination of these facts is a classic red flag for invented data, as it's exactly the way our brains fail at math.
The random sort on Lancaster is basically a flat line. Why is so unlikely that the larger districts would be rural?
Further, why are you accepting the conclusion of "vote manipulation" from a source that has not demonstrated the key point that the vote split in those specific precincts has no simpler explanation?
Oh...
Random sort will basically always be a flat line. That is what a random sort does. The fact that you do not have a flat line in an ordered sort is how you see correlation. If there is no correlation then whichever sort you choose should look like a random sort because a random sort can only have correlation randomly*.
Its unlikely that larger precincts would be rural (in general) because population density should tend to follow precinct size because the primary consideration in precincts is minimizing distance to the voting locations.
I am not accepting it because they have not demonstrated there is no simpler explanation but because i cannot find a more correct explanation. If i had a reason to say "ahh yes, it makes perfect sense that small precincts are in the dense urban areas and large precincts are in rural areas" then we would be golden. Or if i had the omitted variable. The inability to make said determination is more or less how all statistical inference works (and really therefore all science**)
We might even be fine if we saw large instances of vote shifting for the dems, as it would make sense if there was some symmetric issue here. But when we see vote shifting for dems its comparatively minor. I am not sure why we should only see red shift when we see shifts and why those aren't symmetric across similarly demographic areas. Why do we have this problem in Kititas and not Spokane? Or Yakima? Or Chelan? Or Skagit? Or Whatcom? They're all pretty similarly geographically structured. (Large rural population with a single small city core)
*Specifically if you take a stationary random variable and correlate it to anything the correlation coefficient will be an random variable with a center of zero and a variance equal to... honesty i forgot and don't want to do the math but i am pretty sure if its not the product of the variance of the random variable and the determined variable its a similar function. A random sort is a stationary random variable.
**The best theory is the one that fits the evidence and for which no better alternate theory exists. You don't "prove" that no better alternate theory exists because doing so is equivalent to proving the theory. It would not longer be a theory at that point but a law or an axiom; it couldn't not be true.
No. The chart has vote% (not count) on one side and precinct size on the other. It should be smooth because even large total count changes at the end of the count produce a very small change in vote %. This is the law of large numbers in effect.
E.G. Suppose you flip two coins one lands heads 50% of the time and one lands heads 100% of the time and you flip them each 50% of the time.
If you randomly choose which coin to flip this coin looks precisely like a single coin that lands heads 75% of the time. Indeed there isn't even a difference mathematically.
If you order the coin flips so that you flip the 50% coin first 500 times and then the 100% coin 500 times then you will see a wiggly line quickly approach 50% and then once you hit the 100% coin it will veer off towards 75%. For an easy way to see this you can look at the 500th flip where you should have about 250 heads/500 flips so lets just say that is what you have. When you get to the 501th flip you will have 251/501 and then and then 252/502... and then eventually 750/1000.
edit: various edits to correct incorrect numbers and add an explainer on why the graphs look like they do
the "no true scotch man" fallacy.
The best theory that fits the evidence is that these larger districts that went Trump were rural.
Because they are, per the github data source. There are large districts that went Clinton that have been filtered out of the visualization, as well as smaller ones that went Trump. The large district for C are in Lancaster, the ones for Trump are outside the city.
Re: Election Security? If not I appologize.
Ah, I misunderstood the chart, turned on vote bubbles, saw what I expected.
The rural precincts containing more voters seems the most obvious explanation here. The city of Lancaster comprises ~20% (47/241) of the precincts and 10% (59k/536k) of the county population. It stands to reason that the average population of the outlying precincts is larger.
Yeah but the public-facing preliminary results are what politicians use to decide whether or not to concede...
Or more importantly, some people may decide their vote won't matter based on the preliminary reports, so they don't bother. Change the vote by discouraging the voters with false information.
But that's the thing, if you actually looked at the NYTimes link Phyphor posted, and the returns for Lancaster county that PantsB posted, you'd see exactly that. The denser urban areas have much smaller, more granular precincts, while the more rural (but not totally rural, so there's a higher population still) areas around the cities had much larger precincts. Which makes sense on an infrastructure scale too. In dense urban areas, it's easy to set up lots of little polling places, especially for a lot of urban city dwellers who might not have access to a car, and rely on walking or public transit to access a polling place. However in the areas surrounding a city that are more rural, everyone has to drive anyways, and what's the difference that an additional 2-3 minute drive would make?