#US_Election | Logs for 2020-11-16
« return
[00:10:44] <halibut> A few days ago, aristarchus submitted a link I thought was interesting: https://www.thegatewaypundit.com
[00:10:52] <halibut> Here it is in the logs (in #soylent, but I figured this was a better channel to respond on): https://logs.sylnt.us
[00:11:06] <halibut> To summarize, they took all the vote counts in Pennsylvania and, for each county, computed the fraction of election-day votes that went to Trump, and the fraction of mail-in votes that went to Trump. The difference of those two fractions was about 40 percentage points, for all the counties (except Philadelphia). It varied from around 30 points to 45 points.
[00:11:17] <halibut> They claimed that such small range had virtually no chance to occur naturally (since the range in the values being subtracted was more like 40 percentage points instead of 5 to 10 percentage points observed in the difference).
[00:11:28] <halibut> The 40 percentage point difference they came up with seemed highly unlikely to me, so I decided to look into it. First, I went to https://www.electionreturns.pa.gov and checked some of the numbers, and the 40 percentage point difference appears to be real.
[00:11:47] <halibut> Then, I copied the numbers from the thegatewaypundit.com article into a spreadsheet so I could play with them myself. c0lo posted a video earlier that warned how taking the difference of ratios is rarely useful: https://www.youtube.com
[00:11:51] <halibut> Here it is in the logs: https://logs.sylnt.us
[00:12:10] <halibut> With that in mind, I plotted the fraction of mail-in votes that went to Trump as a function of the fraction of election-day votes that went to Trump. Not too surprisingly, the relationship between the two is close to linear, with a slope very close to 1. That seems very reasonable to me: If more election-day votes in a county went to Trump, I would expect a similar amount more mail-in votes to go
[00:12:16] <halibut> to Trump.
[00:12:19] <halibut> However, just as in the video c0lo posted, if you have a linear fit y = mx + b, and then instead plot y-x as a function of x, the result will still be linear but with a slope of m-1 instead of m. Since m was near 1 here, the slope for the subtracted data here is near 0. Thus, the difference in the two ratios will tend to have almost no variance relative to the variance in the original ratios,
[00:12:25] <halibut> precisely because they are related to each other with a slope near 1.
[00:12:34] <halibut> In fact, the variance predicted by this fit is *less* than the variance in the actual data, even if you include the noise. This is because the data has a little curvature to it (not strictly linear), and a quadratic fit is a little better than a linear fit. The plot of the differences using the quadratic fit has the vertex near the center of the data, which is why there is little variance, and the
[00:12:40] <halibut> curvature means it has more variance than the linear fit.
[00:12:41] <halibut> Even more telling, Philadelphia county, which is the outlier in the 40 percentage point difference case, fits the prediction from the quadratic curve very nicely.
[00:12:53] <halibut> It was predicted before the elction that Trump voters would tend to vote on election day and Biden voters would tend to vote by mail, so the difference being a positive number is to be expected. The fact that the difference changes so little is a consequence of the two ratios having a relationship with a slope near 1.
[00:13:02] <halibut> The fact that the difference happened to be about 40 percentage points is not important. The same article did a similar plot for Arkansas, and there the difference was closer to 25 percentage points instead of 40. It is just going to be some number.
[00:13:12] <halibut> Anyways, that seems to me a perfectly natural explanation of what is shown in that article. Even if it was a sign of fraud, I have trouble imagining why the fraudsters in Pennsylvania would do some complicated gymnastics that would result in a near-constant difference in ratios (unless that near-constant difference is somehow a natural result, in which case it does not imply fraud).
[00:13:18] <halibut> Even more so, it seems strange that fraudsters in Arkansas would independently come up with a similar complicated scheme for fraud, and then choose a different difference.
[00:13:24] <halibut> Anyways, I think the thegatewaypundit.com article is not showing some clear-cut evidence of fraud. To those who will tell me I should have expected that, I will respond that I did expect that. I was just interested in how that result could have come about.
[00:13:32] <halibut> I should note that just because there is a natural reason for these statistics does not mean there is no fraud, it just means they do not imply fraud. That said, I have yet to see a statistical argument that indicates fraud.
[00:13:44] <halibut> c0lo has been submitting articles stating that many other claims of fraud have been investigated and found to either lack actual evidence or be a few isolated instances of clerical errors (some of which had already been corrected).
[00:13:51] <halibut> Furthermore, in case you do not trust the media, the courts (which McConnell and Trump have bragged about packing in their favor, so are unlikely to be biased against the Republicans) seem to be routinely throwing out these voter fraud cases for lack of any convincing evidence. This includes some about excluding watchers, as in the cases I read about, the accusors had to admit, under oath, that they
[00:13:57] <halibut> did have some watchers watching.
[00:14:00] <halibut> I conclude that there has not yet been any significant evidence of fraud in the US elections. Again, not surprising.
[00:14:05] <halibut> FatPhil: I hear voting machine security is poor. I will wait for the lawsuit alluded to in your article, but I am guessing they do not have the evidence they claim to have that voting machines were misued in that way.
[00:33:52] <FatPhil> Any argument which proposes a thesis of "voting machines are secure" has a 1-word killer counter-argument.
[00:34:14] <FatPhil> clue: it begins with 'd' and ends with 'd'.
[00:35:42] <FatPhil> can you show me your plot of trump votes vs. republican votes? I was thinking of just taking a screenshot from the vid, and applying a skew transform!
[00:38:56] <FatPhil> I'd like to see a chart documenting the claims of fraud and the time until debunking/dispriving
[00:45:16] <tedious> Do we even know what the state of the election is?
[00:46:04] <tedious> I heard that "they called Georgia for Biden" but that's impossible because the audit isn't even halfway done yet is it?
[01:28:07] <FatPhil> the state? totally minnesota.
[01:30:58] <tedious> I don't know what that's supposed to mean.
[01:39:02] <halibut> I'm guessing you mean ``Diebold''. I remember how badly their fared when people tried to break them. I should clarify my position. Last I heard, they were terrible. I suspect they have improved some, but would be surprised if they are better than plain bad.
[01:41:04] <halibut> That said, there is a difference between ``hackable'' and ``hacked''. I am guessing the Dominion machines were not adjusting votes, but open to evidence that they have. I am also guessing that the people in the article FatPhil linked do not have the evidence they claim to have, but would be very interested if they did.
[01:43:00] <halibut> As for the lengthy argument, I looked into it because I was curious how such a strange result came about. I wrote about it because now I have a record of it, and can point others to that. It was really more for me than anybody else, although I hope others may find it at least convincing.
[03:05:19] <halibut> Here is the data for the plots. Save as pa_votes.csv.xz.b64, and then run (if on a GNU/Linux-like system): base64 -d pa_votes.csv.xz.b64 | xz -d -c >pa_votes.csv
[03:05:29] <halibut> BEGIN BASE64
[03:05:34] <halibut> /Td6WFoAAATm1rRGAgAhARwAAAAQz1jM4A2rBspdABEQyedTnFLkybcvlwRODbatClYxQgRhjh4tmBTeD7qXrI4C1h7GUdZeX8PhFJK0fiTuFcuANIGZe/Q+KekRzIuLwGa3YSdBcCc4WwluaE0lrVXJZoKwf6k2fAUlq6da8JKgcQdg7HiTxJYoDfL6rB2XsY5RK6Yd71HZT5dMg7EguxYqN5j1JfbmvQnE2Sk4p9j0L571Pp6A/lUXQQZah5oKpA+7HB749AEzD/g+Fy8wHacHQVDp4SNxkCBajvZSts5YhU46j3YW3I5ECAdc6CHraXTVfU5yH1FjNa2LL6VjBPR2MFS4PSucmRLpuOOrCLGWwiPsxNaRmtYwZmHzk85U
[03:05:40] <halibut> CfPc3zizijoCi+UAFwacJDKQkWxc+TEc4iXzat5GxVyWm4WpXF9ytrRzS5UCbO49D5EE76yf/QBjgF2TVdq1UjdzwY5X/Wl2DP6C52FE8b5D24iJV5KHzpMVyIGnPw++7wBwq9C3T7ol16deCXJVXEeTDzmgzhz1orUgMPaQVGgV6otbcB5WyMr3FET1ZGUslzcZ6LmmFwdMksIYlbyItZhARtzOk2JOBq3j0f5Jzzi+2I22ktFca5UkHZTxiYVbLN7K+EHmceb3mxF14opWSdR2UWHeMUjAJPLUK7qFKtaHNKiXcIUMdA68CWKRPWAChuumfidyXl+XtXyHDfw1OX05oVU/NTjZpPw3iN5L5m8uLcNchPG6r7M01LkX3ai7
[03:05:46] <halibut> o0IvLmXUCBsXRzyfrN9LrFyt0QJjpEpZbOy/xfwP6ttqGCVxfD9xL86ZV2RmObfwAbWvhrt7nzUUsCryI0WEbV33vWPZyyhMKgOh++qcJ0NlEDLBK/Ui+L3iMM3ZoF4BAVjoVMncBAWLUNKfiKW6Yc0h4I3puJ4M6aKeKIkeVTLZWI9/xccXFGRLN4oBCB9JkkrULzi7qhc6AJ9ogBj9Qqw+iUfYfpXbKFfVMrAWBZU/k0/pwJpdkVSeoO6L/QoLhDBlXdDjVsfZLZDlRCiNy0WRUxt9pzsM2HCo6fazwvwGdeCee7jLC0RVoMyI5yhjgaKvyfAAhDkbLLJEgWWC7tbgi6xZs1ixEskA/hRUilWnsaQPuYnXcNa/n489vEpj
[03:05:52] <halibut> Ce7DfZnHD177DrFjiPPM2n3CL4bHZJfaTW3Qb+d5FWNFrIdlR/SmzlEP4y2mHFI0hOAVB9QQ1cuJGq/ZDyYOEPlR5kkZKWl8EDgPI7TjOWILnXeURzLtKhyZvF4oSwdGXI6TMG7nE34LyLRNtlATExvJg6YPKIcoKhq+tKzUDgYOZj2lo7twpBe2MRB6HUKBEtcb23B6n5b4wR2AB98tN8k9dIaMAp+MwQ4ttr1K3N4PSg3Be5YGgGXEntdsXeG1Ih8KR+kUtW+UfAsemmCU/1RtY9vadY57m/LGjlCZiI6cSk5RkcDrupLAcnjb3/Oo413wIZsDI8efv9Zz5pI3M3vYtcV0p2VMPWqnko0HdFFEN6hEIl5STUaR86mRoq6S
[03:05:58] <halibut> mOirI7+eAIDX0eyaxM2bNdJghU4RhAueI3nIX7bnGk3OymADwuDRmzFa2csaylENHYazcOHludCduKay2c1/DEpRbj2G43odrY/uQY0IJgTqCnic/8b5igxHIp0+hzuK/NZ+M7dDX8tBJdhPfj2xtxuIycgK+Hi0KM7AW8t6oREH4eAkYeayF5LbSu+RQ3Bqb6LbIsqJSZyVP6PYtjHbKMJXLzbUg4SuZXThY0LbTvzDaDFuy2FbUa0QrS9ygUf9v1kF/S70twSJlfUnZ8AK3suXZcW3yVAM1zQZqsVZyr7nFcmQFC1nq6mxnBS3jKfYulDt5X/Xnk80wXY7pOIV04jXRu6RvbwoPrzMl2LKO0iDYPsSGOYgfyoQSXyP2gP9
[03:06:04] <halibut> B6tkFDpDRl4D70pGsB+w7jtLofGfZQVcV5h6s+1sbmVs9KVFSj0D5vln6c4BIwlb09zSx/7o/eNtGJPvkUv/voPKV+pZ7lCYnRDWCgxhl/fskt9dxDd/Exug1lyah2SlMHKm0gL7JfwWCQ3KP6n1lzIrIoACuWcBDovf3l8Gj2A8AX86NrXglbhuGi0/mbJ5CfBiwcMZk6DtCqER5Yaoa+EttsNhOPUVTk0Gj8CxC2lLplzPK58Uv76NdPvYNP1k8b2N/f5lhWfRAh6zB/OKOaBEEE4jbheuJsIHe30aE4yJ8ZrUQGZV5CX4h6/3zizxj4FPKkrTiPL/d0ASOJdp2ryuNWpZGLrWwDUnoIj0gvKB+AIw+dcpAPcJIaiQassc
[03:06:10] <halibut> JHWgVfNH5tDpTnTsL9NonucRQP0SRKn+4Fs/r6FqisuiVMkmCXfCzKUAAADFIyAMceWlBgAB5g2sGwAA3wIW7LHEZ/sCAAAAAARZWg==
[03:06:13] <halibut> END BASE64
[03:06:25] <halibut> Here are some gnuplot commands to plot the relations:
[03:06:40] <halibut> set datafile separator comma
[03:06:45] <halibut> plot "pa_votes.csv" using ($7)/($3+$7):($8)/($4+$8) with points, f(x)=-0.31295+0.88046*x, f(x) with lines
[03:07:02] <halibut> (Linear relationship with ratios)
[03:07:08] <halibut> plot "pa_votes.csv" using ($7)/($3+$7):($8)/($4+$8) with points, f(x)=0.24859-0.88108*x+1.31804*x**2, f(x) with lines
[03:07:13] <halibut> (Quadratic relations with ratios)
[03:07:19] <halibut> plot "pa_votes.csv" using ($7)/($3+$7):($7)/($3+$7)-($8)/($4+$8) with points, f(x)=x-(-0.31295+0.88046*x), f(x) with lines
[03:07:46] <halibut> (Linear relationship with the difference of ratios, note that one point, Philadelphia county, is pretty far from the 40 point difference)
[03:07:53] <halibut> plot "pa_votes.csv" using ($7)/($3+$7):($7)/($3+$7)-($8)/($4+$8) with points, f(x)=x-(0.24859-0.88108*x+1.31804*x**2), f(x) with lines
[03:08:18] <halibut> (Same, but quadratic relationship, which does a decent job of predicting Philadelphia county)
[03:27:13] <halibut> The .csv comes from copying the data from the spreadsheet screenshot in the image. I may have mistyped some things, but I think I got pretty close. The totals do not add up correctly in three of the rows, but that is how the numbers are in the image.
[03:28:06] <halibut> I just ignored the provisional votes; those numbers are small enough they do not matter very much. Other than that, the ratios are basically (election-day-for-trump)/(election-day-for-trump + election-day-for-biden), and the equivalent for mail-in.
[04:32:05] <pinchy> when in doubt, calculate about
[05:07:20] -!- aristarchus_ [aristarchus_!~c222847a@194.34.gtz.swn] has joined #US_Election
[05:56:22] -!- Runaway1956_ [Runaway1956_!~some@67.223.okg.wi] has joined #US_Election
[05:56:30] -!- Runaway1956 has quit [Read error: Connection reset by peer]
[06:19:47] <aristarchus_> Finally, Runaway has quit. Time to fold up this entire chatroom. It's over.
[06:41:27] <pinchy> get out Runway1956_, its over. the establishment has spoken
[06:45:14] -!- FatPhil has quit [Ping timeout: 265 seconds]
[06:45:39] -!- FatPhil [FatPhil!~phil@22-653-047-30.dyn.estpak.ee] has joined #US_Election
[06:46:36] FatPhil is now known as SoyGuest41274
[07:41:53] <SoyGuest41274> tedious: you bloody youngsters! How can you not get that reference, the song's only 39 years old?!??
[07:50:38] <SoyGuest41274> halibut: data received in the clear: "County","Biden total votes", ... ta!
[07:55:52] <SoyGuest41274> mein got, I hate gnuplot with a vengence
[08:00:40] <SoyGuest41274> "The totals do not add up correctly in three of the rows" - hahah, that's hilarious!
[08:01:28] <SoyGuest41274> oh, shit, I'm going senile, the song's only 29 years old.
[08:07:09] <SoyGuest41274> My main "throw it out it's garbage" argument against shiva is:
[08:07:47] <SoyGuest41274> (1) shiva expects this will be a constant: plot "pv.csv" using ($4+$8)/($2+$6):($6)/($4+$8) with points
[08:08:05] <SoyGuest41274> (2) I would expect the exact opposite, I would expect a negative correlation.
[08:08:40] <SoyGuest41274> (3) He presents no argument at all why he expects what he does. Not even a crap argument, absolutely no argument at all, he simply asserts it.
[08:09:54] <SoyGuest41274> (4) Even my dismissive "trump voters are lazy" predicts a slope.
[08:10:22] <SoyGuest41274> (5) And the data demonstrates a clear slope.
[08:10:28] <SoyGuest41274> With one caveat...
[08:11:13] <SoyGuest41274> Namely, that that data isn't the data I was looking for, it's not the straight ticket vs. individually selected votes breakdown.
[08:11:56] <SoyGuest41274> my thesis is
[08:12:35] <SoyGuest41274> straight ticket <=> lazy <=> republican, for '<=>' = 'correlation'.
[08:13:03] SoyGuest41274 is now known as FatPhil
[08:13:11] -!- FatPhil has quit [Changing host]
[08:13:11] -!- FatPhil [FatPhil!~phil@Soylent/Staff/Editor/FatPhil] has joined #US_Election
[10:23:02] <FatPhil> Ah, this is hilarious:
[10:23:05] <FatPhil> Republicans are also exhibiting some... concerning... attitudes about the franchise, with 46% saying that "some people are not smart enough to vote" (27% among Dems) and 43% saying that people should have to pass a test before voting (15% for Dems).
[10:24:38] <FatPhil> Which proves how dim dems are - it would clearly bias in their favour were there a test of general intelligence in order to vote.
[10:25:39] <FatPhil> '89% of Republicans also think Trump should contest the outcome of the election in court, while 62% of his voters think it will change the outcome.'
[10:26:44] <FatPhil> https://docs.cdn.yougov.com
[10:40:54] -!- aristarchus_ [aristarchus_!~c222847a@194.34.gtz.swn] has parted #US_Election
[11:09:16] -!- pinchy has quit [Quit: Kids come running for the great taste of sampo]
[11:09:32] -!- pinchy [pinchy!~Car@68.66.ysy.qx] has joined #US_Election
[15:32:08] <FatPhil> https://www.theepochtimes.com
[15:33:27] <FatPhil> There certainly are several quite noisy #NeverTrumpers, their adverts (for Lincoln Project) have been hilarious!
[15:33:44] <FatPhil> And yes, they will be interested in throwing the election.
[15:34:39] <FatPhil> However, that can be done in the EC itself, without repercussions in many states.
[15:38:59] <FatPhil> Just remember to vote for "John Ewards"
[15:46:07] <pinchy> muh bawsaq https://cdn.discordapp.com
[21:21:06] <halibut> FatPhil: Yes, the data I posted is not, I think, the data used in the Shiva video. The videos c0lo posted had a pretty thorough explanation of that (or, at least, one plot they created). That was for Kent County in Michigan. The data I worked with, and posted, was for the counties in Pennsylvania, and was the data used in an article making what turned out to be a very similar statistical error.
[21:26:06] <FatPhil> Yup. I'm sure my thesis will be correct, I just don't have the raw data to prove it!
[21:27:47] <FatPhil> Of course, the assumption that the proportions between the two classes of voters will be 50/50 is completely pulled out of shiva's arse too, so there's more than just the lazy<=>trumper hypothesis.
[21:31:55] <halibut> I think the data is available here: https://electionreporting.com
[21:32:33] <halibut> Unfortunately, it is a site that requires Javascript to work. It will let you download the data as a PDF, or you can intercept the JSON file it loads, but that looks a bit messy.
[21:33:26] <halibut> I found it easier to use the drop-down menu to select the party results by precinct, and then the presidential candidate results by precinct, and just highlight the data table and copy/paste it to a text file.
[21:34:05] <halibut> If you select Kent County, there are 252 precincts, if I remember correctly, which makes for a bit more data that is easy to copy/paste with xz and base64 as I did before.
[21:35:10] <halibut> One thing to note: The presidential candidate totals appear to include (as they probably should) both the people who voted for individual candidates and the people who voted for them indirectly by voting a party ballot.
[21:51:17] <halibut> I played around a bit with the Kent County data I got. I can convert it into a .csv file and get it into 11 lines of base64-encoded data. If you are interested, I can transfer that the way I transferred the other file.
[21:56:22] <FatPhil> If you don't have your own webserver to host it on, there's always /dcc send or you can just mail fatphil@fatphil.org to keep the noise down.
[23:32:01] <halibut> https://pastebin.com
[23:32:07] <halibut> Save to votes.csv (or something like that).
[23:32:19] <halibut> This will then recreate the Shiva plot in gnuplot:
[23:32:26] <halibut> set datafile separator comma
[23:32:32] <halibut> plot "votes.csv" using ($3/$2):($6/$5-$3/$2) with points
[23:32:54] <halibut> Note that in this data set, the president columns have the party values already subtracted.
[23:33:19] <halibut> A value in the presidentTot column means that many people in the precinct voted for the president directly, as opposed to indirectly via a party ballot.
[23:33:59] <halibut> Also, it appears that the plot has a single point with a y value below 30% which appears to have been cut off in the video frames I have seen showing the plot.
[23:34:34] <halibut> You mentioned a dislike of gnuplot. Feel free to use some other plotting tool.
[23:34:49] <halibut> CSVs should import into spreadsheets, for example.
[23:35:12] <halibut> Let me know the results of your analysis.