Thursday, November 19, 2020

How Trump Lost Michigan

Four years ago, I wrote a long analysis of How Trump won Michigan.  The key points of that analysis are

  • Trump did much better than other republicans in rural and downscale (working class/union) areas like Downriver, Macomb, suburban Genesee, and the northeastern Lower Peninsula.
  • Trump did poorly in upscale (highly educated) suburbs in Oakland, Kent, and Kalamazoo counties.
  • Turnout in Detroit was relatively poor, providing democrats with a relatively small absolute margin there.
Trump's margin of victory in Michigan in 2016 was 10,704 votes.  At the moment, the unofficial totals show Joe Biden with 2,804,039 votes and Trump with 2,649,852 votes.  Biden's margin over Trump is 154,187 votes.  This analysis will consider county vote totals to determine where this shift occurred, and whether there is any reason to be suspicious of the official totals.

County Margin Changes

Trump's performance by county is illustrated in the map below.

Data at the SOS website is broken down by county.  Subtracting 2016 and 2020 margins, we find the biggest (approximate) changes in margin in the following counties.

-54200 Oakland
-32500 Wayne
-30900 Kent
-23000 Washtenaw
-11300 Ingham
-10700 Kalamazoo
  -9400 Macomb
  -7300 Ottawa
  -4600 Grand Traverse

The totals are broadly consistent with the results of the last two elections.  Counties with many white leftists, particularly those featuring large universities (Washtenaw, Ingham, Kalamazoo) saw large increases in D turnout.  Upscale suburbs continued to trend away from Trump.  This is particularly the case in Dutch Reformed Christian areas (Kent, Ottawa), which voted for Ted Cruz in the 2016 primary.

Trump also improved his margin in many smaller counties.  The largest gains were about 3300 each in Monroe and St. Clair, and 3100 in Montcalm.

Local Voter Fraud?

Of course, this is taking the unofficial vote totals for granted.  Since election day, there have been many claims of fraud, and allegations that the election was stolen.  Specific allegations include that late-arriving absentee ballots were back-dated so that they could be counted, and that some poll workers encouraged voters to support Biden.  These allegations are plausible (though unproven) and deserve further investigation.

We should be careful to distinguish between voter fraud that happens a few votes at a time, and mass voter fraud (adding, deleting, or changing thousands of votes at a time).  Only the latter could have swung the election by a large enough margin to give the state to Biden.

Allegations of fraud have centered around the city of Detroit and the counting of absentee ballots at the TCF Center (formerly Cobo Center).  This is understandable, given Detroit's long history of incompetence (or worse) at running elections.  To get a handle on what happened in Detroit, consider the raw totals and percentages for GOP and D candidates for the past six elections.

Year – GOP (percent) and Dem (percent):
2000 – 15688 (5.22%), 282111 (93.88%)
2004 – 19343 (5.93%), 305258 (93.65%)
2008 –   8888 (2.65%), 325534 (96.99%)
2012 –   6018 (2.09%), 281743 (97.63%)
2016 –   7682 (3.11%), 234871 (94.95%)
2020 – 12654 (5.08%), 233908 (94.00%) 

Note that Trump got 5000 MORE votes than he got four years ago, consistent with a modest improvement with black voters nationwide.  Biden got 1000 FEWER votes than Hillary did in 2016.  The data provide no evidence to support mass voter fraud in Detroit.

How then did Biden gain 32000 votes in Wayne County versus 2016?  Trump lost ground in the upscale suburbs of western Wayne County.  For example, Trump's margin declined in the following jurisdictions.
-7500 Canton Township
-5200 Livonia
-2000 Northville Township
-1800 Plymouth Township

Similar patterns occurred elsewhere in the state.  In Kent County, Trump lost 13000 votes in Grand Rapids, and lost the suburbs of Kentwood, East Grand Rapids, Wyoming, and Grand Rapids Township.

Trump's biggest relative declines from his 2016 percentages were in West Michigan, Oakland, and Washtenaw.  Trump's biggest relative increases from his 2016 percentages were in the northern lower peninsula and Wayne County.

Dominion Voting Systems

If fraud in one or two jurisdictions cannot explain the swing to Biden, could fraud have been widely distributed across the state?  Such theories revolve around Dominion Voting Systems, a manufacturer of voting machines and vote counting machines.

On election night, something went wrong with the vote totals in Antrim County in northwest Lower Peninsula.  The reported totals showed Joe Biden handily winning the county, and many people (including me) immediately realized that this could not be correct.  The county clerk retracted the results and ordered a recount, which produced a reasonable result.  (Trump's margin declined by 233 votes compared to 2016.)

It still isn't clear what happened in Antrim County.  The clerk blamed a software error.  The Secretary of State blamed human error (i.e., the clerk).  If Dominion software produced a false result in Antrim, could it have done so elsewhere?

In theory, vote counting software could be programmed to flip a small fraction of votes from one candidate to another, so that the fraud would be difficult to detect.  Some commentators have claimed that Dominion did this.  This cannot be what happened in Antrim, as the results there were absurd, and quickly corrected.  Did Dominion flip votes in other counties?  Dominion machines are used in 65 Michigan counties, and the others are split between two other vendors.  The state of Michigan helpfully maps the vendors by county on a website.  I broke down the presidential vote totals by counties that use Dominion machines and those that don't.

In 2020, the totals in counties that use Dominion machines are
Trump 1410197, Biden 1459230 (49.1% Trump two-party)
In 2016, the totals in the same counties were
Trump 1237598, Clinton 1335566 (48.1% Trump two-party)
In 2020, the totals in counties that don't use Dominion machines are
Trump 1209676, Biden 1212807 (49.9% Trump two-party)
In 2016, the totals in the same counties were
Trump 1069867, Clinton 1056032 (50.3% Trump two-party)

Thus Trump actually improved from 2016 in counties that use Dominion machines and declined in counties that don't use Dominion machines.  The data provide no evidence to support the theory that Dominion machines flipped votes from Trump to Biden.  Indeed, some of the counties where Trump declined the most (Oakland, Macomb, Ottawa, Kalamazoo, Grand Traverse, Livingston) don't use Dominion machines.

Of course, a conspiracy theorist could double down and claim that all three vendor's machines were compromised, not just Dominion.  There is a simple way to test this theory.  Remember that Michigan doesn't have voting machines, it has vote counting machines.  There are paper ballots that can be recounted when desired, as was done in Antrim County.  As far as I know, the Trump campaign has not requested a recount in Michigan.

Comparison to US Senate Results

Another argument casting doubt on the validity of the election results concerns the results in the US Senate race.  The unofficial vote totals give Gary Peters 2,734,559 votes, and John James 2,642,221 votes, a margin of 92,338.  The distribution of votes across the state is similar to the Presidential race, but the relative differences are noteworthy.

Some observers have noticed the vote totals for Trump and James were quite close (only 7631 apart), while the vote totals for Biden and Peters were about 70000 apart.  Indeed, there were about 58000 more votes case in the presidential race than the senate race.  This has lead some to claim that 70000 ballots were cast only for Biden, and that this is proof of voter fraud.

There are several problems with this theory.
1. There is no proof that dropoff is due to votes only for Biden.  Voters could have skipped the senate race but voted in other races downballot.
2. There is nothing unusual about dropoff.  In 2012 (the last year before 2020 with both presidential and senate elections), there were 78043 more votes for president than for senate.  In 2008, there was a dropoff of 153156 between these races.  There is typically more dropoff further downballot, as some uninformed voters choose not to vote in races when they are unfamiliar with the candidates.
3. Someone committing mass fraud would presumably want Ds to win downballot, and they could just as easily vote straight ticket D to do this.
4. There was essentially no dropoff in Detroit (note that 77% of votes in Detroit were straight ticket D).
5. Trump and James did better in different regions of the state.  The map below uses Orange for Trump performing better than James, and blue for James performing better.

Trump performed better in downscale areas, while James performed better in traditional GOP areas in West Michigan.  Note that James won Kent, Muskegon, and Leelanau Counties, which Trump lost.  James got 11000 more votes than Trump in Kent County and 8700 more in Oakland, while Trump got 12000 more votes than James in Macomb County.  James did essentially the same as Trump in Detroit, gaining only 80 more votes there.


There is nothing in the data to indicate mass voter fraud in Michigan of the size that would have to exist to change the result of the Presidential election.  More than 150,000 fraudulent votes should be noticeable in the raw election results, but they are not there.  If there is any doubt about the reliability of the vote counting machines, a manual recount would resolve it.

No comments: