EAST LANSING – There is no evidence that indicates systematic tampering or hacking of voting machines, or conspiracy or fraud among election officials, resulted in vote tallies altered to favor the Republican candidate in Michigan in 2016, experts at the consulting firm Anderson Economic Group LLC reported this week.

Anderson Economic Group also reported there was no evidence that indicates a large number of ineligible foreign-born residents voted for the Democratic candidate in Michigan in 2016, if “large number” means Michigan’s share of “3-million to 5-million” across the United States.

“This round of tests, like the last one, clearly indicates no evidence for systematic fraud, tampering, or hacking of voting machines,” Patrick L. Anderson, Principal and CEO. “Furthermore, it demonstrates that both high and low immigrant-population counties did not vote in the manner suggested by the allegation of ‘3-5 million illegals’ voting for the Democratic Party candidate in 2016.

“After two rounds of election forensics, a multi-million dollar recount effort, and the certification by bipartisan board of canvassers, this is now the most-vetted presidential election in Michigan history,” he added.

The forensic analysis, like the one done in mid-December, involved a battery of statistical and data analysis techniques applied to county-by-county election data for the 2012 and 2016 elections for President of the United States in the State of Michigan.

The previous analysis was summarized in a letter written by Anderson Economic Group CEO Patrick Anderson to the Michigan Secretary of State, Ruth Johnson, in December of 2016. It involved a set of allegations made shortly after the November 2016 election. This analysis extends the prior analysis to new allegations. Both these analyses were conducted using established principles of Election Forensics. We summarize the allegations and results of our analysis below; the data, statistical techniques, evaluation protocol, and limitations of these methods follow.

Allegations Investigated: This analysis attempted to detect evidence of a large share of “3-million to 5- million illegals” across the United States voting for the Democratic Party candidate in the State of Michigan, based on recent statements by President Donald Trump. It also extended tests done earlier on the claims of “fraud or mistake” by county election officials in the Jill Stein recall petition, as well as the repeated assertions that voting machines could have been hacked by Russian or other state actors.

Anderson’s statistical tests indicate no significant deviations that would support the first set of allegations. In particular:

  • The histograms of vote shares across counties, for both 2012 and 2016, indicate the expected pattern across the large majority of counties. In 2016, we can identify 3 outlier counties, all of which are readily explainable (and unsurprising).
  • The pattern of vote shifts is also consistent, showing a gain for “R” vote shares across nearly the entire state.
  • The empirical cumulative distribution functions, for both “R” and “D” votes, are similar in 2012 and 2016.

The second result is there is no evidence that indicates a large number of ineligible foreign-born residents voted for the Democratic candidate in Michigan in 2016, if “large number” means Michigan’s share of “3-5 million” across the United States.

For this analysis, Anderson considered both the prior results, and specific counties in Michigan that are home to large immigrant populations. As the “3-million to 5-million illegals” assertion is not specific to any one state, we projected the possible impact of a hypothetical effort to encourage foreign-born residents to vote across the country, and estimated that this would (hypothetically) involve about 250,000 fraudulent votes in Michigan. Foreign-born residents of Michigan as a whole represented 6.3 percent of its resident population in 2015. However, that share varies widely, from well over 10% in Oakland, Washtenaw, and Macomb; to 2 percent or fewer in over 50 of Michigan’s 83 counties.

“We made use of Census data on immigrants of all status, including naturalized citizens and non-citizens, in specific counties,” Anderson said. “We tested to see if these counties had unusual voting behavior in 2016, after controlling for their historic voting patterns and for the swing in votes in the 2016 election.”

The following are the top Michigan counties for share on foreign-born population:

 

Rank

Michigan County

Total Population

Total Foreign Born Population

Foreign Born Share of Population

1

Oakland County

1,229,503

145,670

11.8%

2

Washtenaw County

354,092

40,920

11.6%

3

Macomb County

854,689

90,243

10.6%

4

Ingham County

283,491

24,761

8.7%

5

Wayne County

1,778,969

144,247

8.1%

6

Kent County

622,590

48,774

7.8%

7

Berrien County

155,565

9,545

6.1%

8

Ottawa County

273,136

15,700

5.7%

9

Oceana County

26,229

1,492

5.7%

10

Kalamazoo County

256,752

12,020

4.7%

Source: Census ACS, 2015.

Evaluating these counties for voting patterns, we found no evidence that they behaved differently than one would expect given the pattern of votes across other counties. Of the top ten high-immigrant population counties, nine saw the share of voters favoring the Democratic Party candidate fall from 2012 to 2016. (Only Washtenaw County saw a slight gain.)

That pattern appeared in counties with high immigrant population as well as low immigrant population. For example, the share of Oakland County (immigrant population of nearly 12 percent) voters for the Democratic Party candidate in 2012 was 46 percent, and in 2016 it fell to 44 percent. The share of Grand Traverse County voters (immigrant population of about 2%) that went for the Democratic Party candidate in 2012 was 44 percent; in 2016 it was 41 percent. Indeed, all counties other than Washtenaw saw a decline-suggesting the alleged large number of ineligible foreign-born voters clearly did not have surface in Michigan.

Data and Election Forensics Techniques: Anderson Economic Group consultants examined the certified election results from the State Board of Canvassers with exploratory data analysis tasks, and two sets of hypothesis tests. The base data included 2012 and 2016 county-by-county election results for all major party candidates, as certified by the Board of State Canvassers, and Census ACS data on foreign-born residents in each Michigan county.

The exploratory data analysis tasks included illustrating the pattern of vote shares by county in two elections, evaluating the empirical distribution functions of vote-shifts among counties across elections, comparing same-county vote shares across two elections, and calculating empirical cumulative distribution functions for the same set of candidates and elections, and comparing them with idealized distribution functions.

The analyses were done by consultants working out of the company’s East Lansing, New York, and Chicago offices, in December and January 2017. Along with the results, the company released new graphics illustrating some of the results and demonstrating specific statistical techniques. These are available at www.andersoneconomicgroup.com.

Evaluation Protocol: These statistical tests attempt to find significant deviations from the patterns of votes in these counties and in the other counties in the state. Significant deviations that were unexplained by other voting patterns, and were consistent with the allegations we investigated, were considered evidence of possible fraud or tampering. A lack of significant unexplained deviations was considered lack of evidence of fraud or tampering.

Limitations of Election Forensics Techniques: These statistical tests cannot detect irregularities in individual precincts that involve small numbers of votes. They also do not characterize patterns of voting that occur regularly in elections, or that occurred around the entire state, as unusual. Finally, election forensics can indicate evidence of fraud or tampering (or the lack of such evidence), but cannot by itself prove or disprove it.

AEG’s initial analysis and results are specific to Michigan. The same methods can be applied to other states, such as Arizona, California, and Texas. However, aside from unreleased results from Wisconsin and Pennsylvania, we have not done this. We caution against extending the analysis beyond what the data actually demonstrate.