Debunked: Navid Keshavarz-Nia's Claims of "A Sudden Rise in Slope" as Election Fraud Evidence

Mick West

Administrator
Staff member
Dr. Navid Keshavarz-Nia gave an affidavit as part of Sydney Powell's "Kraken" lawsuit. It's largely speculation based on his experience and on media reports and has remarkably little in the way of actual analysis or data.

One of the few direct claims of evidence revolved around the New York Times data feed, which is claims shows things that "demonstrates data manipulation by artificial means,""does not occur in the real world without an external influence," and "are cause for concern and indicate fraud."


g. Beginning on approximately 4:30 AM EST on November 4, 2020, the vote counts favored Vice President Biden by nearly 80% in many jurisdictions. The data distribution is statistically congruent, even when considering a larger number of absentee ballots were collected for Vice President Biden.

h. The data variance favoring Vice President continues to accelerate after 4:30 AM EST on November 4, 2020 and continues until it momentum through November 9, 2020. This abnormality in variance is evident by the unusually steep slope for Vice President Biden in all battleground states on November 4, 2020. A sudden rise in slope is not normal and demonstrates data manipulation by artificial means. For example in PA, President Trump’s lead of more than 700,000 count advantage was reduced to less than 300,000 in a few short hours, which does not occur in the real world without an external influence. I conclude that manually feeding more than 400,000 mostly absentee ballots cannot be accomplished in a short time frame (i.e., 2-3 hours) without illegal vote count alteration. In another case for Edison County, MI, Vice President Biden received more than 100% of the votes at 5:59 PM EST on November 4, 2020 and again he received 99.61% of the votes at 2:23 PM EST on November 5, 2020. These distributions are cause for concern and indicate fraud.
Content from External Source
He concludes:

My analysis of the 2020 Election from NY Times data shows statistical anomalies across the battleground state votes. These failures are widespread and systemic - and sufficient to invalidate the vote counts.

I conclude with high confidence that the election 2020 data were altered in all battleground states resulting in hundreds of thousands of votes that were cast for President Trump to be transferred to Vice President Biden. These alterations were the result of systemic and widespread exploitable vulnerabilities in DVS, Scytl/SOE Software and Smartmatic systems that enabled operators to achieve the desired results. In my view, the evidence is overwhelming and incontrovertible.
Content from External Source
So let's look at one of his data points:
For example in PA, President Trump’s lead of more than 700,000 count advantage was reduced to less than 300,000 in a few short hours, which does not occur in the real world without an external influence. I conclude that manually feeding more than 400,000 mostly absentee ballots cannot be accomplished in a short time frame (i.e., 2-3 hours) without illegal vote count alteration.
Content from External Source
This is based on the NYT data, which was available on the NYT website in JSON format, and can easily be converted to a spreadsheet, although it does require a little code to get the date and time into a usable format. Keshavarz-Nia's example is PA (Pennsylvania), here's the graph of the votes for Trump and Biden.

2020-11-29_11-51-32.jpg

These are not exact figures, because the NYT data only stores percentages of the total votes, and not the individual vote counts, but it generally shows what is going on.

So what of Keshavarz-Nia's claim that "in PA, President Trump’s lead of more than 700,000 count advantage was reduced to less than 300,000 in a few short hours, which does not occur in the real world"?

We can see from this graph that this is simply nonsense. Trump gained that 700,000 vote lead, in six hours of counting. Far less time than it took Biden to get it back (about 12 hours).

To his broader point: "This abnormality in variance is evident by the unusually steep slope for Vice President Biden in all battleground states on November 4, 2020. A sudden rise in slope is not normal and demonstrates data manipulation by artificial means." We can see from the correctly plotted graph that there is no "sudden rise in slope". Other than the occasional jumps from precincts or counting centers reporting a large batch, the slope of both lines gradually decreases. This is hidden slightly by the overnight pauses where the slope is flat, but if we mark those, then we can see each days slopes have no sudden rise. If anything there's a sudden fall from day-to-day.
2020-11-29_11-51-32.jpg


Keshavarz-Nia's argument here seems to be purely on personal incredulity of incorrect data analysis. And since this is one of the few times he uses actual numbers, it raises significant doubts as to the quality of his more speculative testimony.,
 

Attachments

  • gov.uscourts.mied.350905.1.19.pdf
    492 KB · Views: 673
  • penn_dt.xlsx
    73.3 KB · Views: 656
Technical notes. The code to extract the data with Excel format time code
Code:
import json
import sys
from datetime import datetime

def excel_date(date1):
    temp = datetime(1899, 12, 30)    # Note, not 31st Dec but 30th!
    delta = date1 - temp
    return float(delta.days) + (float(delta.seconds) / 86400)

def ffDumpTime(NAME):
    with open(NAME + '.json', encoding="utf8") as f:
        x = json.load(f)
    print("Time,Biden Share,Trump Share,Total Votes,Biden Votes,Trump Votes")
    for i in range(len(x["data"]["races"][0]["timeseries"])):
           curr = x["data"]["races"][0]["timeseries"][i]
# Time format is like 2020-11-11T16:40:52Z
        datetimeObj = datetime.strptime(curr["timestamp"], '%Y-%m-%dT%H:%M:%SZ')
        print ( str(excel_date(datetimeObj))+ ","+str(curr["vote_shares"]["bidenj"]) + "," + str(curr["vote_shares"]["trumpd"])+","+str(curr["votes"])+","+str(int(curr["vote_shares"]["bidenj"]*curr["votes"]))+","+str(int(curr["vote_shares"]["trumpd"]*curr["votes"])))

ffDumpTime(f"{sys.argv[1:][0]}")

This is structured as a python file that takes a command-line argument of the name of the .json file (without .json). Pipe the output to a csv file, like:

python program.py State-Pres-pennsylvania >penn_dt.csv

The time output is in Zulu (GMT) format, as a floating point number of days with the time as the fraction (Excel format.) To convert to EST, you need to subtract 5/24 (as in the second column in the spreadsheet attached above)

To plot the graph with the times correctly spaced, you need to use a line scatterplot in Excel, and not a line graph (that will evenly space the data points, giving a misleading graph).

I also limit the x-axis date range to 44138.5 to 44142.0 to focus on the areas of interest. Ensure that "Smooth Line" is OFF for both data series.
 
So your analysis implies that it takes the same amount of time to sort, verify, open and scan, etc a mail in ballot that it does to process an in person vote?

How long did it take to count in person votes and how many where there? You claim it took 12 hours to overcome Trumps lead when counting mail in votes. Again how many were counted in that group? Keep in mind they had to be sorted, verified, opened, scanned, etc. In person are only scanned.
 
How long did it take to count in person votes and how many where there? You claim it took 12 hours to overcome Trumps lead when counting mail in votes. Again how many were counted in that group? Keep in mind they had to be sorted, verified, opened, scanned, etc. In person are only scanned.
Have a look at the graph again, note how many votes there were at the end of each day (points indicated here are from just before voting resumes the following morning)
2020-11-30_09-15-54.jpg
So on the first day (Nov 3, election day) there are about 5.3 million votes counted. (2,968,494+2,353,206 = 5,321,700)
Nov 4 ends with a bit under 6.3 million. (3,219,679+3,054,567 = 6,274,246), so under 1 million votes counted on day 2. (6,274,246-5,321,700 = 952,546)
Nov 5 ends with under 6.6 million (3,286,296+3,266,379 = 6,552,675), so just 278K additional votes.

So yes, absentee vote counting was much slower to process. The figures above are not clear cut, as there's some overlap, and some preparation of the absentee votes may have been done in advance (like sorting and verifying, although that varied by state). But the day 2 results came in at 1/5th the speed of the day 1 results. Far from a "sudden rise in slope," this is a huge decline in the slope.
 
So your analysis implies that it takes the same amount of time to sort, verify, open and scan, etc a mail in ballot that it does to process an in person vote?

How long did it take to count in person votes and how many where there? You claim it took 12 hours to overcome Trumps lead when counting mail in votes. Again how many were counted in that group? Keep in mind they had to be sorted, verified, opened, scanned, etc. In person are only scanned.
If you live in the US, go to your local election office, and ask them.

How long it takes for a single ballot is immaterial: how long it takes for all of them depends on how much personnel you have to do it.

But even for single ballots: verifying that absentee voters are on the voter rolls is usually done as soon as the ballots arrive; that's why people can check online if their ballot has been received or not. So what you do on election day in the states where the Republicans blocked early-absentee-counting in the state legislature is run the secrecy envelopes through a letter opener and then scan stacks of ballots. After that step, you're in the same place as with the in-person ballots: write-ins need to be processed, overvotes and undervotes need to be adjudicated as to whether voter intent is discernible, stuff like that happens with both in-person and absentee ballots. And there are also provisional ballots to be processed.

It's not just the voters that come to a polling place, are listed on the voter roll, and submit a perfectly machine-readable ballot with no over- or undervotes and no write-ins that have a right to be counted; it's every citizen. That's why the initial provisional incomplete count doesn't decide the election, it's the final certified count that does.
 
It is also important to know when state law allows a count to start. Florida had a very different standard than my home state of Pennsylvania, which did not begin the mail-in tabulation until 7:00 am on Election Day.

Provisional ballots were another story, made more complicated by delays in the mail and processing. I went through training when I volunteered as a poll worker and county officials told me they were constantly short-handed this year.

I was frankly surprised that the state did so well given the volume of total votes.

https://electproject.github.io/Early-Vote-2020G/PA.html
 
Back
Top