4
$\begingroup$

Dr Shiva Ayyadurai ran in the Republican primary for Massachusetts senator in 2020, losing this race to Kevin O'Connor. He then filed a lawsuit alleging irregularities in the vote count. The suit is currently pending.

His legal filing is embedded on this page https://vashiva.com/no-one-person-one-vote/ . In the document (on page 21, the previous pages can thankfully be skipped) he sort of makes an actual statistical argument, which I will try to summarize below.

I personally find his arguments extremely questionable, but if anyone with a better statistics background would be willing to tease apart specifically what is wrong, I'd be fascinated to hear it.

Briefly, Ayyadurai provides a histogram counting the number of precincts in which he received N votes vs number of votes:

enter image description here

Then he counts the number of 'High-Low pairs' - by which he appears to mean points where the number of precincts he got N+1 votes is higher than the number of precincts in which he got N votes. He counts 9 instance of this for the first 22 bins (why 22 is not explained)

He then proceeds to argue that:

The question herein is how likely is it that my vote count in Suffolk County could generate 9 or more “High-Low” pairs for the first 11 pairs in the histogram, assuming the reported results are fair and unbiased. To answer this question, I have modeled vote counts by precinct and candidate using a binomial distribution –of 298 precincts x 2 candidates for 596 total distributions.

And concludes:

9 or more “High-Low” pairs should only appear in bins 1-22 of the histogram once in every 741 elections!

So my question is: what statistical fallacies have been committed in this analysis?

And related:

  • What distribution should one expect this chart to follow? Surely it depends on the number of voters in each precinct
  • Is there any legitimate use of this 'high-low pair' concept or is that just something he made up?

Thank you!

$\endgroup$
5
  • 1
    $\begingroup$ Also of interest, a debunking of another of his election analyses: Do these scatter plots reveal fraudulent vote-switching in Michigan? $\endgroup$ – fblundun Dec 6 '20 at 23:48
  • $\begingroup$ What is X axis here? $\endgroup$ – forecaster Dec 7 '20 at 3:27
  • $\begingroup$ The number of votes for the candidate. So first column indicates there are 26 precincts in which he got a single vote. A little surprised there's no column for zero. $\endgroup$ – Edwin Young Dec 7 '20 at 6:09
  • 1
    $\begingroup$ See also his recent analysis purporting to show electoral fraud in Georgia here, and a counterargument here. Lowlight: the choice of y axis in his first graph. He plots (Trump support - Republican support) against Republican support, and then implies that it is suspicious that this graph has a negative slope! $\endgroup$ – fblundun Dec 8 '20 at 13:00
  • 1
    $\begingroup$ I tried to recreate his Suffolk County histogram and ran into an odd thing. As noted, Shiva's histogram has no bin labeled 0. But actually there are 11 precincts where he got 0 votes and 15 where he got 1. His histogram shows 26 for bin 1. For whatever reason he seems to have combined bins 0 and 1. $\endgroup$ – jhughes Dec 28 '20 at 17:16
4
$\begingroup$

Seems to be a lot of cherry-picking in this analysis.

Massachusetts has 14 counties (Wikipedia), of which he has picked just one, Suffolk County, for his analysis. He concludes that the election there, and in 10 other counties which used "predominantly electronic voting systems", was weighted against him. He justifies choosing Suffolk County by saying "Given Boston, the largest city with the largest voter turnout, is in Suffolk County, we began by analyzing Suffolk County." Wikipedia says there are two other counties with higher population (though it's possible they had lower total turnout). As far as I can tell, he never applies the same analysis to any other county.

He says that in Franklin County, which uses paper ballots, the reported data "closely track" with his "de-weighted" results from the 11 electronic voting counties. He doesn't analyze whether the results from the remaining counties (which presumably also don't use electronic voting), Dukes County and Nantucket County, also closely track with his de-weighted results.

The "high-low pair" concept looks like his own invention. I couldn't find a definition. Your definition seems plausible but it doesn't match up with what he says about the histogram associated with O'Connor, his opponent:

In Kevin O’Connor’s histogram and corresponding table chart, the incidence of “High-Low” pairs is much less pronounced

According to your definition, O'Connor's histogram actually has more high-low pairs in the first 22 buckets: I count 11 buckets whose bar is higher than the previous bucket. Hard to say more here without being sure of the definition.

He claims that his votes were reweighted by multiplying by 0.666 and rounding. Since this maps multiples of 3 to even numbers and everything else to odd numbers, he says it explains why among the first 22 buckets, the even buckets are smaller than the odd buckets. But it doesn't explain why the difference between even and odd vanishes for buckets above 22 (which make up more than half the x axis).

Overall, it's a very unconvincing argument.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.