In my last post, I described the aberrant nature of the unofficial election results released by the Antrim County Clerk at 4:09am on the morning of 4th November. 1 Compared to 2016, the Trump vote had collapsed in some precincts, even to single figures in four precincts, but increased in others; and extraordinary numbers of those who had voted in certain precincts had chosen not to cast a vote for President at all. 2
Amended results were released at 4:42pm 5th November. 3 Trump’s vote, expressed as a percentage of votes cast for Trump, Biden or the Libertarian candidate Jorgensen, had recovered from 36% to 57%. 4 In three precincts, however, Custer, Echo and Elk Townships, which he had won in 2016 by between 56 and 69%, he still trailed behind Biden with between 40 and 42% (in all tables, the suffix ‘1’ refers to the results of 4th November, ‘2’ to those of 5th November, and ‘3’ to those of 21st November):
In these same three precincts, the value given for number of ballots cast had greatly increased from the number reported the previous morning, and exceeded the number of registered voters by up to 48%. In Custer and Echo Townships, the values given had exactly doubled, from 776 to 1552, and from 602 to 1204 respectively. In the other thirteen precincts the value given for number of ballots cast was unchanged:
2nd amended results were released on the 21st November, at 3:33pm:
The values given for number of ballots cast in Custer, Echo and Elk Townships had reverted to those given in the initial results of 4th November:
The reduction in number of votes in these three townships in this second revision of the results was taken almost exclusively from Biden, who lost 523,392 and 414 votes in Custer, Echo and Elk respectively, compared to losses of 11, 8 and 14 for Trump. In the other thirteen townships, Biden’s vote remained exactly the same, while Trump’s vote declined by 2 in Central Lake, and remained the same in the other twelve townships:
Trump thus ended ahead in every precinct, as he had in 2016:
Five unaffected precincts
Results for Trump and Biden in five precincts, namely Banks, Central Lake, Mancelona 1, Mancelona 2, and Warner, have remained unchanged, or in the case of Central Lake, only marginally changed, over the course of the two revisions:
Votes for the other Presidential candidates were also unaffected in these 5 precincts, and the same is true for at least the first three candidates in the Senate race (the only non-Presidential results examined by the author so far).
Any explanation for the misreporting of the Antrim County election results must include an explanation for why these precincts were not affected.
Right→left vote transfer in nine precincts
In nine of the remaining precincts, the 2nd amended vote ascribed to Trump was the same or either one or two votes less than the initial vote ascribed to Biden:
Trump, on the other hand, was not initially ascribed the votes that were finally given to Biden. Rather, in each of these nine precincts (but in no others), Trump was initially ascribed less than 20 votes, whereas the final vote ascribed to Biden was in the range 93 to 610:
Instead, Trump appears to have been initially assigned the final vote ascribed to Jorgenson in these nine precincts, with the tallies identical except in the case of Torch Lake, where Trump had one extra vote:
Likewise, Jorgenson appears to have been initially assigned the assumed true vote in these nine precincts of the next candidate on the list, namely Don Blankenship.
The numbers are identical except in Custer Township where Jorgensen had 4 votes to Blankenships’s 2:
Blanken appears to have been initially assigned Hawkins’ assumed true vote in these same nine precincts, now highlighted in blue:
The pattern continues with Hawkins and Fuente:
and with Fuente and ‘Unresolved Write-Ins’:
The initial value given for Unresolved Write-Ins was 0 in all nine precincts. If we conceive of the aggregating program as operating rather like a spreadsheet, then this zero value could perhaps be consistent with an empty column to the ‘right’ of ‘Unresolved Write-Ins’.
Moving to the Senate Race:
the votes initially assigned to the Democrat candidate Gary Peters in the same nine precincts appear to be based upon the assumed true votes for the Republican candidate John James, although with a delta of up to 6, rather than of up to 2 as in the Presidential race:
Given the consistency of the pattern, it seems reasonable to propose that James, likewise, was initially assigned the assumed true votes of the UST candidate Valerie Willis in these precincts, even though in this case, with deltas of up to 4, and Willis’s vote in single figures, the congruence is not quite so striking:
A cursory inspection of the other races contained in the Bailey complaint, that is for Straight Party Ticket, Congress Representative, and Representative in State Legislature, suggests that the phenomenon probably exists in these also, that is of a ‘right-to-left’ 5 vote transfer in the same nine precincts only. See, for example, the Straight Party initial results for Democrat and Republican Parties:
with the collapse in the Republican vote in these nine precincts.
Milton Township was the only precinct in which Biden’s final vote was greater than his initial vote:
Trump’s final vote, meanwhile, was more than twice his initial vote, Biden’s 12% greater. There is no indication of vote-switching, with Biden’s initial tally being 1021 – 686 = 335 votes lower than Trump’s final tally:
The proportion of ballots containing a vote for president was initially only 65.2%, rising to what I would suppose is a normal 99.4% in the amended results.
Elk Township was, with Custer and Echo, one of the three townships that showed an increase in the number of cast ballots in the 1st amended results, and a return to the initial number in the 2nd amended results. Trump’s final vote is 64% higher than his initial vote, while Biden’s is 18% lower. There is no indication of vote switching, with Biden’s initial tally being 1025 – 392 = 633 votes lower than Trump’s final tally:
1) Errors in the initial reporting of 4th November
Let us assume, for sake of argument only, that the 2nd amended results were accurate. Then we can characterise the malfunctioning of the initial reporting system according to its effect on two distinct major groups of the sixteen precincts, along with unique effects on the two other remaining precincts:
A) 5 unaffected precincts.
Votes for Trump and Biden in Banks, Central Lake, Mancelona 1, Mancelona 2, and Warner Townships were unaffected by the malfunctioning of the system. The 2 vote difference in the Central Lake Trump vote was probably related to the retabulation of ballots that took place in this precinct only, and during which 3 ballots are reported to have been damaged, rather than to the initial reporting malfunction. 6
Votes for other Presidential candidates, and for all the other candidates so far examined, were also unaffected in these 5 precincts.
B) 9 precincts with apparent right→left vote-switching
In Chestonia, Custer, Echo, Forest, Helena, Jordan, Kearney, Star, and Torch Lake Townships, Biden initially received the number of votes that had actually (it is being assumed) been cast for Trump, or either one or two votes more than that number. Trump received the assumed true number of votes for Jorgensen, or with one extra vote. This pattern continues all the way through the Presidential race, into the Senate race, and probably throughout those races whose results have been made available in the Bailey complaint. 7
C) Milton Township
Trump initially received about 53% less than what I am assuming to be his true vote, while Biden initially received about 11% less than his assumed true vote.
D) Elk Township
Trump initially received about 39% less than his assumed true vote, while Biden initially received about 22% more than his assumed true vote.
The table below shows the final and initial results, with the precincts arranged into these four groups. The last column shows the difference between Trump’s final vote and Biden’s initial vote, highlighting the evidence for Trump→Biden vote transfer in the nine precincts of Group B:
2) Errors in the 1st amended results of 5th November
Let us assume again, for sake of argument only, that the 2nd amended results were accurate. If the 2 vote delta in Central Lake is set aside, for the reason described above, then the 1st amended results are characterised by large numbers of excess votes for Biden in three precincts, namely Custer, Echo, and Elk, and small numbers of excess votes for Trump in the same townships:
Remarkably, as stated above, the number of cast ballots in Custer and Echo Townships according to the 1st amended results was exactly double the assumed true value, exceeding the number of registered voters by 38 and 48% respectively. The number of cast ballots in Elk Township was 8% greater than the number of registered voters:
Features to be explained
An explanation for the malfunctioning of the Antrim County election system should explain the following characteristic features:
i) Correct functioning in the five precincts of Group A;
ii) in the nine precincts of Group B, for all candidates so far examined, the initial receipt by each of what is assumed to be the true vote of the candidate one column to his or her right in the published results, with sometimes a small and almost always positive delta;
iii) major errors in both Trump and Biden’s votes in Milton and Elk Townships;
iv) abnormally high proportions of cast ballots containing no vote for President in all affected 11 precincts in the initial results;
v) high numbers of votes wrongly assigned to Biden, and low numbers wrongly assigned to Trump, in Custer, Echo and Elk Townships in the 1st amended results;
vi) numbers of cast ballots exceeding registered voters in these same three counties in the 1st amended results.
vii) In addition, the exact doubling of the number of cast ballots in Custer and Echo in the 1st amended results should at least be investigated to determine whether it has a specific cause, or is only a matter of random coincidence.
I intend, in my next post, to raise the question of the identity of the software system which produced such aberrant results. It was supplied to Antrim County by Election Source, but to my knowledge there has so far been no public disclosure as to its provenance, whether it is its own creation, or that of another company.
- See Exhibit C, Bailey complaint. ↩
- Precinct-level data for 2016 was obtained from here. (It may perhaps be helpful to someone to observe that I was able to open the large .csv file with a text editor, but not with spreadsheet software.) ↩
- See Exhibit D, Bailey complaint. ↩
- The reasons for choosing this denominator are a) to enable a comparison with the 2016 results, for which I don’t have precinct level data for number of ballots cast, while b) saving some time by omitting the minor candidates. ↩
- In the spatial sense obviously, not the political sense. ↩
- See ASOG Forensic Report, version 2, D3, p. 7:
- The full 21 November results are available from the Antrim County election web-site. ↩