michigan

2020 Voting
Analysis
Report
11-27-20


Table of Contents
Executive Overview .......................................................................................3

1 - Analysis of Michigan County Vote Counts ........................................5

2 - Wayne & Oakland Counties: Finding Excessive Votes ..................8

3 - Exploring Michigan Mail-in Ballots Data ......................................14

4 - Absentee Ballots Compared in Select MI Counties ....................16

5 - Irrational MI Absentee Ballot Findings .........................................18

6 - Analysis of a Survey of Absentee Ballot Recipients ...................21

Summary ........................................................................................................ 24

Cover Graphic credit <tinyurl.com/y4wa6k3d>.

Page 2
 Executive Overview
This scientific analysis of the reported Michigan (MI) 2020 Presidential voting
results is a non-partisan effort by unpaid citizens and volunteer experts
(several un-named). Our only objective is to play a small roll in helping assure
that all legal MI votes are counted, and that only legal MI votes are counted.
Whether Donald Trump or Joseph Biden wins is not of concern in this
analysis — the scientists involved with this report just want the election
results to truly reflect the wishes of Michigan voting citizens.
Since there are multiple reports of voting chicanery circulating the Internet, a
collection of statisticians and other scientists volunteered to examine the
reported MI results from a scientific statistical perspective.
We feel that the best way to do this is to start by putting ourselves in the
shoes of bad actors — and then considering how they might go about
changing the wishes of MI citizens, into a different result. Some of the actions
they might take are:
1 - Keep ineligible people (e.g. deceased, moved, etc.) on the voting roles.
(This would disguise actual voter participation rates, allow fabricated votes
to be submitted in their names, etc.)
2 - Get legislation passed that does not require in-person voter identification.
(This would make it easier for non-citizens, felons, etc. to vote.)
3 - Encourage a much higher percentage of voting by mail.
(This would make it much easier to manipulate, as in-person checking is a
more secure way to keep track of actual registered citizens, etc.)
4 - Discard envelopes and other identifying materials from mail-in votes.
(This makes it very hard to check for duplications, etc.)
5 - Count mail-in votes without careful signature or registration verification.
(This makes mail-in an easier choice for manipulators.)
6 - Allow votes to count that are received after Election Day.
(This can direct where mail-in votes are needed to go.)
7 - Stop vote counting for several hours before the final tabulations.
(This allows for an assessment of how many votes are “needed” etc.)
8 - Do not allow for independent oversight of voting tabulation.
(This would make it easier to lose or miscalculate actual votes.)
9 - Connect voting machines or precincts to the Internet.
(This makes it quite easy for third parties to access and change votes.)
10-Distribute vote manipulations over multiple precincts and/or counties.
(This makes the adjustments more difficult to find.)
12-Make most of the manipulations in unexpected districts.
(In other words, don’t do as much manipulation where it’s expected.)
12-Use multiple methodologies to change vote results.
(It requires a much longer investigation to find all the adjustments.)
There are undoubtedly more strategies those who are trying to control our
politics would employ — but this is a representative sample. It should also be
clear that many of these are difficult and time-consuming to find.
Page 3
Frequently there is documented proof of some of these voting actions (e.g.
leaving non-eligible voters on the rolls). However, these are usually dismissed
with cursory responses such as: we’re doing the best that we can, or these
deviations are not statistically significant, or our rolls are as accurate as other
states, or there are some benefits for doing this (e.g. #3 & #6 above), etc.

However, studies like this and reports like this do not instill confidence that
election results actually reflect the wishes of actual citizens.

So what can we do as scientists? Clearly we can’t verify the legitimacy of every

Michigan vote submitted. On the other hand, we can (from a scientific
perspective along with with sufficient data) provide a statistically strong
assessment that reported votes in certain locations are statistically unusual.
Such a determination should be treated as an indication that some type of
accidental or purposeful manipulation almost certainly occurred.

Such a science-based statistical analysis can not identify exactly what

happened — or prove that fraud was involved. Honest mistakes, unintentional
computer glitches, etc. can and do happen.

We approached this project assigning different experts to look at the Michigan

data from different perspectives. By-and-large the experts worked mostly
independently of each other. As a result, there may be some overlaps in the
analyses in the following four “chapters.”

All of the experts agreed that there were major statistical aberrations in some
of the Michigan results that are extremely unlikely to occur naturally.

Using more conventional statistical analyses, we identified nine counties with

abnormal results (see Chapter 1). Due to time, data and manpower
limitations, for this Report we focused on the statistical analysis for the worst
two counties. Our strong recommendation is that both of those Michigan
counties have an audited recount.

If the results of a carefully audited recount are that there is no significant

change in voting results for those two counties (very unlikely), then the
authors of this report recommend that we write off those county deviations as
an extreme statistical fluke, and that the Michigan voting results be certified.

On the other hand, if the results of a carefully audited recount are that there
are significant changes in voting results for either of these two counties, then
the authors of this Report recommend that (as a minimum) that the next
seven statistically suspicious counties also have an audited recount, prior to
any certifying of the Michigan voting results.

See Summary on the final page, for more conclusions.

— Editor, physicist John Droz, jr. 11-26-20

Page 4
 1 - Analysis of Michigan County Vote Counts
S. Stanley Young, PhD, FASA, FAAAS, 11-25-20

Summary: 

People today generally vote as they have done in the past. If a voting pattern
changes, is it a slight shift, or are large changes occurring in a small number of
locations? Our idea is to look at relative vote changes in counties within Michigan.
How does Biden vs Trump2020 compare to Clinton vs Trump2016? There could be
slight shifts that accumulate across the state, or there could be major changes in a
relatively few counties. We use contrasts to examine voting results. We find vote
changes are modest for the bulk of MI counties: less than 3,000± votes. However,
there are nine counties with much larger changes in votes, up to 54,000±.

Item 1 —
Consider Biden vs Trump2020 compared to Clinton vs Trump2016.
Contrast = (Biden – Trump2020) – (Clinton – Trump2016)

Here is the distribution of Contrast:

Examine the left side of the above chart. There we see an approximate bell-shaped
distribution, which is normally what would be expected. The Contrast (change in
votes for Biden vs Trump relative to Clinton vs Trump) for almost all counties is
within the range of plus or minus 3000± votes.

The outliers (numbers unusual relative to the rest of the data) are on the right of
the chart, where Biden bested Trump much more than Clinton bested Trump.
Page 5
Item 2 — 
Here we rank contrasts from largest to smallest for all Michigan counties.

In the above histogram, each dot is one MI county. In 74 of 83 MI counties, the

differential is small (near zero) implying that for the vast majority of counties,
voters considered Biden vs Trump2020 much like they considered Clinton vs
Trump2016. On the left side of the histogram are the nine (9) outliers — i.e.
counties with numbers that substantially deviate from the main distribution.

These nine counties together substantially increase the vote count for Biden. For
instance, in the first two of these counties (Wayne and Oakland), the differential
(contrast) swing for Biden amounts to 96,000± votes.

Page 6
The remainder of the nine outlier counties (ranks 3 to 9 on the spreadsheet above)
represent an additional 95,000± excess votes for Biden, compared to Clinton vs
Trump. (For example, Trump bested Clinton in Kent county by 10,000± votes but
lost to Biden by 22,000± votes, for a net swing of 32,000± votes.) The total
unexpected votes for Biden in the nine Michigan outliers is 190,000± votes.

Item 3 — 
Here is another anomaly that indicates suspicious results. The first set of plots
compare Trump’s election day votes to his mail-in votes, for each county. As would
be expected, the distributions are quite similar. The second set of plots compare
Biden’s election day votes to his mail-in votes, again for each county. As is easily
seen, the distributions are very different. This is a serious statistical aberration.

CONCLUSIONS: The distribution of Item 1, and the magnitude of the

differentials in Item 2, and the statistically deviant patterns in Item 3, are all
statistically improbable relative to the body of the data.

Page 7
 2 - Wayne and Oakland Counties:
Finding Excessive Votes in the 2020,
Well Outside Their Voting History
(condensed version: full version available)
Dr. Eric Quinnell, Dr. Stanley Young
11/26/2020

Contents
Executive Summary 8
Wayne County/Oakland Buck the Trend 8
Wayne County 8
Oakland County 12

Executive Summary
Analysis – A statistical team of unpaid citizen volunteer scientists, mathematicians, and engineers
collaborated in a statistical vote analysis in the Pennsylvania 2020 Presidential Election, after having worked
originally as individuals on various vote analysis across the country. Following the PA report (available on
request), the collaboration team netted steep learning curves in analysis and methods, and produced a
mathematically based predictive model to reverse engineer vote differential signatures. This now much
more robust model is re-applied to Michigan.

Using simple linear regression of unproblematic voting districts, we predict hypothetically problematic
voting districts. Using distributional characteristics within problematic counties, we point to problematic
districts and precincts.

Findings – Two Michigan counties stand out as problematic, Wayne and Oakland Counties, 40,000 and
46,000 estimated excessive votes, respectively. Problematic districts and precincts within these counties
exhibit unusual Democrat/Republican (D/R) ratios relative to their history and excessive vote in favor of
Biden often in excess of new Democrat registrations.

Wayne County/Oakland Counties Buck the Trend

A bi-variate trend-line across all Michigan counties (see next page) identify Wayne County and Oakland
County as behaving well outside the trends of the rest of the state in 2020. Wayne and Oakland counties
also stood out from the analysis done in another section of this report (see Page 6). Thus, these two
counties were selected for deeper analysis.

Page 8
Wayne County
A bi-variate linear fit of the Trump and Biden votes in 2020 Wayne County show major precincts
completely off the charts as compared to the majority of the other precincts in the same county. The
points exceedingly off the fit are mostly those in the Absentee Vote Counting Board (AVCB) districts.
Several others outside of Detroit also buck the trend of the rest of the area.

The AVCB mail-in districts within Detroit have no ability to correlate with the precincts inside the city, so a
historical voting pattern per precinct is not possible. There is also no indication that the AVCB distributions
include the same precincts from year to year, so therefore there is no way to link AVCB in obvious ways.
Instead, we first looked at the remainder of Wayne County. Outside the city we have much more history
and can observe both mail-in votes as well as election day votes correlated to a precinct with history.
Page 9
Outside Detroit, Wayne County shows a significant disruption or new vote distribution well outside the
2016 norm. Specifically, both candidates achieved the total 2016 vote count and added to their sums,
consistent with new turnout. What’s curious is that above the 2016 totals, a new vote ratio appears in
contrast to the history of the area – showing new votes going 70% Democrat vs 30% Republican – a 15-
point mismatch to the same area just in the last Presidential Election.

Voting totals of precincts may presume to follow a semi-normal distribution with enough data points. By
fitting a normal distribution to actual data and taking the difference between the fitted and actual,
potentially anomalous precincts can be identified. Using a per-precinct history, we can take an election
result like this:

And identify anomalous precincts. We forced the anomalous precincts back to their voting history ratios
and adjust to keep pace with the 2020 turnout. This results in this prediction:

Which helps us identify several townships outside Detroit in Wayne County that significantly stick out. A
partial list of main townships that show excessive votes vs a standard normal with reasonable variance:

Townships Excessive Votes

Page 10
As an example of the excess vote gains above the norm, consider the Township of Livonia, broken into
precincts. Nearly every single precinct first achieves the entire 2016 vote total for each party, but then a
new population of votes skews excessively in favor of the Biden camp – resulting in a “new vote
population” that is voting 76 D / 24 R — in a 2016 Republican township!

Additionally, the votes gained by Biden well outpace even the new registrations in the township – gaining
151% of the new registered voters and 97% of the new votes above 2016. This result/example is incredibly
mathematically anomalous.

Page 11
Oakland County
Oakland shares the Wayne County mathematical deviance of being well outside the norm. In Oakland all
votes added by both candidates above the 2016 take show a new vote ratio of 72% Democrat to 28%
Republican – an 18-point mismatch to the same area just since the last Presidential Election.

As mentioned, voting totals of precincts may presume to follow a normal distribution. By fitting a normal
distribution to actual data and taking the difference between the fitted and actual, potentially anomalous
precincts can be identified. Using a per-precinct history, we can take an election result like this

and identify anomalous precincts. Should we peel those anomalies back to the voting history ratios and
keep pace with the 2020 turnout, we get this prediction:

This helps us identify several townships in Oakland County that significantly stick out. This is a partial list of
main townships that show unexpected deviations:

Townships Excessive Votes

Page 12
As an example of the excess vote gains above the norm, consider the Township of Troy, broken into
precincts. Nearly every single precinct first achieves the entire 2016 vote total for each party, but then a
new population of votes skews excessively in favor of the Biden camp – resulting in a “new vote
population” that is voting 80 D / 20 R — in a 2016 almost even split Dem/Rep township.

Additionally, the votes gained by Biden well outpace even the new registrations in the township – gaining
109% of the new registered voters and 98% of the new votes above 2016.

This situation is yet another example that is incredibly mathematically anomalous.

Page 13
3 - Exploring Michigan 2020 Mail-In Ballots Data
Robert Wilgus 11/27/20

The 2020 election data for Michigan mail-in ballots was provided as a large file obtained via an FOIA. The
data was perused for anomalies that stood out. A more comprehensive analysis is appropriate and that is
what has been arranged (see Conclusions).
The data file contains 19 fields for each mail-in application. The fields can be text, numbers, or dates. My
understanding of the process is that certain voters (not sure how they were determined) were sent a
form to request a mail-in ballot.
The data available captures the process from when the application was sent. The total of requested
absentee ballots is 3,507,129. The table below contains measures that merit further investigation:

Ballots did not get sent to about 36,000 of the requests received. It’s not clear what the
reason(s) were for this (e.g. faulty address, etc.). The ballot can be marked as Rejected or
Spoiled. Spoiled ballots (incomplete?) and Rejected ballots (duplicates?) add up to about
135,000 ballots that got tossed. That seems like a lot.
The data also includes the voter’s year of birth. One is 170 years old, likely an error but their
applicahon was not rejected. In total more than 1400 of these absentee voters are over 100
years old. These could well be nursing home pahents.

Page 14
There are 217,271 applicahons without a recorded date (i.e. never received back). More
intereshng is the 288,783 that have the applicahon sent and ballot received on the same day.
Maybe these are one stop vohng and get recorded with the mail in ballots? The table below
contains other date related findings:
.

The ballots rejected doesn’t provide any addihonal informahon for what the reason was. It
does appear that the majority of ballots received ajer Nov-3 did fall into this category.

The last but not least is the spoiled ballots. There is a lot of them. In the first table there are
8,341 duplicate Voter ID. I would expect these were the ‘spoiled’ ones that got new ballots.
There is another column in the table named SPOILED_IND that means spoiled by the
individual. It has values ‘N’ or is not entered.
There is also very small number that are both rejected and spoiled

CONCLUSIONS: There are numerous measures in the mail-in ballot data that warrant further
investigation. This is surprising because there are very few field values with obvious errors. The records
with multiple empty fields are of concern. Additional information is also needed for the high number of
applications and ballots with the same and returned dates
Because of the importance of this file we recently shared it with a firm that specializes in data analytics of
very large databases, to see what they can tease out if it. We are looking forward to some interesting
analyses.
Page 15
 4 - Michigan Absentee Ballots:
Several Key Counties Compared
Dr. William M. Briggs, 11/26/20

Data from counhes in Michigan where absentee votes by candidate were available were
gathered. The counhes were (alphabehcally): (1) Eaton, (2) Grand Traverse, (3) Ingham,
(4) Leelanau, (5) Macomb, (6) Monroe, (7) Oakland, and (8) Wayne.

In Eaton and Oakland votes could be either straight party (e.g. choose all Democrats for all
contests) or variable ballots (e.g. choose candidates individually). These were treated
separately.

The data sources are: Eaton (XML), Grand Traverse (PDF), Ingham (PDF), Leelanau (PDF),
Macomb (HTML), Monroe (PDF), Oakland (XML), and Wayne (PDF).

The percent of the total vote for each candidate (not the over all total, but the candidate total)
that was absentee was calculated across each precinct or district within each county. The data
within a county was sorted by the absentee percentages for Biden, low to high, for display ease.

Next, we plot the percent absentee votes for both Biden (D:blue) and Trump (R:red). See below
for examples of two large counhes. (For the same types of graphs of more Michigan counhes
see here.) The precinct numbers are here arbitrary, and reflect the sorhng of the data.

Page 16
Almost never does the percent of absentee ballots cast for Trump exceed the percent cast for
Biden. There are only rare excephons, such as in very small precincts where we'd expect totals
to be more variable.

If absentee vohng behavior was the same for those vohng for Trump and Biden, the chance
that absentee ballots for Biden would almost always be larger would, given the large number
of precincts here, be vanishingly small.

Thus, either the absentee vohng behavior of those vohng for Biden was remarkably
consistently different, or there is another explanahon, such as manipulahon of totals.

More proof of this is had by examining the rahos of absentee ballot totals in each precinct. See
below for examples of the same two large counhes. (For the similar graphs of more Michigan
counties see here.) Again, the precinct numbers are arbitrary and reflect the same sorting as before.

Only 36 precincts out of the 2,146 examined had 0% absentee ballots. These are obviously not
shown in the figures (because of divide-by-zero possibilihes). As menhoned, the raho of Biden
to Trump absentee votes is astonishingly consistent. The mean raho inside each county is
printed in the figure, along with the number of precincts.

If vohng behavior was similar for both candidates, we'd expect this raho to be 1, with some
variability across precincts, with numbers both above and below 1. Instead, the rahos are
almost always greater than 1, and with a hght mean about 1.5 to 1.6 or so. This indicates the
official tallies of absentee ballots for Biden were about 50-60% higher almost everywhere, with
very litle variahon, except in smaller counhes were the raho was slightly higher.

Such behavior could be genuine, or programmahc changes of the votes could be the
explanation of these unusual results. The data here is more consistent with the later hypothesis.
Page 17
 5 - Irrational MI Absentee Ballots Findings
Thomas Davis, 11/27/20

All American cihzens, regardless of party affiliahon, should be concerned about the integrity of
of our elechon system. If the people no longer determine who their representahves are, the
United States is no longer a Republic. Accordingly, post-elechon scruhny of suspicious results is
not only appropriate, but required.

It is unsurprising that absentee vohng in 2020 occurred at a much higher rate than in previous
years. (In Kent County Michigan, for example, there were 68,967 absentee voters in 2016, and
211,209 in 2020 – a threefold increase.) The COVID-19 virus undoubtedly had a direct impact
on the strong move to absentee vohng across the nahon. In Michigan, there were two
addihonal major contribuhng factors: 1) voters approved a no-reason absentee vohng law in
2018, and 2) Secretary of State Jocelyn Benson sent absentee vohng applicahons to all 7.7
million registered Michigan voters this past summer.

When stahshcs in Michigan showed especially high numbers of absentee votes for Biden, it
didn’t raise many red flags. Ajer all, the Democrahc party had encouraged people to vote
absentee while the Republican party had encouraged vohng in-person since ballots could be
lost in the mail. However, a closer look at absentee vohng (from the select Michigan counhes
that publish detailed vohng stahshcs) appears to tell a different story.

Let’s start by showing what normal (non-manipulated) absentee vohng results should be. The
plot below is the actual percentage of absentee ballots received by each 2020 presidenhal
candidate in a swing state county, by precinct (Red = R and Blue = D). Note the irregularihes
that normally occur: some are higher for R, some are higher for D, and the difference between
the two varies widely — from plus to minus. This is what a normal result looks like!

Page 18
Now back to what happened in Michigan. Specifically, save for outliers, the percentage of
Democrahc absentee voters exceeds the percentage of Republican absentee voters in every
precinct. Even more remarkable – and unbelievable – these two independent variables appear
to track one another. This stahshcal anomaly can be seen very readily in Monroe County:

DEM% = # of absentee votes for Biden / total # of Biden votes

REP% = # of absentee votes for Trump / total # of Trump votes

In Oakland County (with 4x more voters than Monroe County) this same patern can be seen,
albeit somewhat less clearly (as there are more data points — i.e. precincts):

Page 19
Comparing the same county (Oakland) 2020 data with 2016 brings the stahshcal anomaly into
even sharper focus. The 2016 plot is below, and it looks approximately like the inihal plot
shown (page 18), of what a normal situahon would look like:

By examining the absentee vohng in the other Michigan counhes for which these data are
available, similar stahshcal anomalies are observed.

Conclusion: This is very strong evidence that the absentee vohng counts in Michigan have
been manipulated by a computer algorithm.

On the surface it would seem that the tabulahng equipment in each precinct has been
programmed to shij a percentage of absentee votes from Trump to Biden. A simple hand-
count of absentee ballots from a sampling of precincts should be sufficient to determine
whether this asserhon is valid; a forensic analysis of the tabulahng equipment would be
required for definihve proof.

Page 20
6 - An Analysis of Surveys Regarding Absentee
Ballots in Several States (including Michigan)
Dr. William M. Briggs, 11/23/20
1: Summary
Survey data was collected from individuals in several states, sampling those who the
states listed as not returning absentee ballots. Data was provided by Matt Braynard.
The survey asked respondents whether they (a) had ever requested an absentee ballot,
and, if so, (b) whether they had in fact returned this ballot. From this sample I produce
predictions of the total numbers of: Error #1, those who were recorded as receiving
absentee ballots without requesting them; and Error #2, those who returned absentee
ballots but whose votes went missing (i.e. marked as unreturned).
The sizes of both errors were large in each state. The states were: Arizona, Georgia,,
Michigan, Pennsylvania, and Wisconsin.
2: Analysis Description
Each analysis was carried out separately for each state. The analysis used (a) the
number of absentee ballots recorded as unreturned, (b) the total number of people
responding to the survey, (c) the total of those saying they did not request a ballot,
(d) the total of those saying they did request a ballot, and of these (e) the number
saying they returned their ballots.
From these data a simple parameter-free predictive model was used to calculate the
probability of all possible outcomes. Pictures of these probabilities were derived, and
the 95% prediction interval of the relevant numbers was calculated. The pictures for
Michigan appear in the Appendix at the end. (Other states are available on request.)
They are summarized here with their 95% prediction intervals.
Error #1: being recorded as sent an absentee ballot without requesting one.
Error #2: sending back an absentee ballot and having it recorded as not returned.

Page 21
Ballots that were not requested, and ballots returned and marked as not returned were
classified as troublesome. The estimated average number of troublesome ballots for
each state was then calculated using the table above and are presented here:

3: Conclusion
There are clearly a large number of troublesome ballots in each swing state
investigated. Ballots marked as not returned that were never requested are clearly an
error of some kind. The error is not small as a percent of the total recorded unreturned
ballots.
Ballots sent back and unrecorded is a separate error. These represent votes that have
gone missing, a serious mistake. The number of these missing ballots is also large in
each state.
Survey respondents were not asked that if they received an unrequested ballot whether
they sent these ballots back. This is clearly a possibility, and represents a third possible
source of error, including the potential of voting twice (once by absentee and once at
the polls). No estimates or likelihood can be calculated for this additional potential
error due to absence of data.

(See next page for an Appendix to this chapter…)

Page 22
4: Appendix
The probability pictures for Michigan for each outcome as mentioned above.

Probability of numbers of un−requested absentee ballots listed as not returned for Michigan:

Probability of numbers of absentee ballots returned but listed as not returned for Michigan:

Page 23
 Summary
Several nahonally recognized stahshcal experts were asked to examine some 2020
Michigan vohng records, and to idenhfy anything that they deemed to be stahshcally
significant anomalies — i.e large deviahons from the norm.

In the process they basically worked separately from other team members, consulted
with other experts, analyzed the data they were given from different perspechves,
obtained some addihonal data on their own, etc. — all in a very limited hme allotment.

Their one — and only — objechve was to try to assure that every legal Michigan vote is
counted, and only legal Michigan votes are counted.

The takeaway is that (based on the data files they were examining) these experts came
to one or more of the following conclusions:
1) There are some major stahshcal aberrahons in the MI vohng records, that are
extremely unlikely to occur in a normal (i.e. un-manipulated) seyng.
2) The anomalies almost exclusively happened with the Biden votes. Time and again,
using a variety of techniques, the Trump votes looked stahshcally normal.
3) Nine (out of 83) Michigan counhes stood out from all the rest. These counhes (see
p 6) showed dishnchve signs of vohng abnormalihes — again, all for Biden.
4) The total number of suspicious votes in these counhes is 190,000± — which greatly
exceeds the reported margin of Biden votes over Trump. (We don’t know how
many of these are arhficial Biden votes, or votes switched from Trump to Biden.)
5) These stahshcal analyses do not prove fraud, but rather provide scienhfic evidence
that the reported results are highly unlikely to be an accurate reflechon of how
Michigan cihzens voted.

As stated in the Executive Summary, our strong recommendation is that (as a minimum):
the two worst of the nine abnormal MI counMes have an immediate audited recount.

If the results of a carefully audited recount are that there is no significant

change in voting results for those two counties (very unlikely), then the
authors of this report recommend that we write off those county deviations as
an extreme statistical fluke, and that the Michigan voting results be certified.

On the other hand, if the results of a carefully audited recount are that there
are significant changes in voting results for either of these two counties, then
the authors of this Report recommend that (as a minimum) that the next
seven (7) statistically suspicious counties also have an audited recount, prior
to any certifying of the Michigan voting results.
Page 24