No evidence of election fraud or tampering in 2016
The integrity of elections is a bedrock principle upon which our Republic is based. For that reason, Michigan and every other state has a system of careful canvassing of votes and certification of the results. Unfortunately, recent allegations of “fraud or mistake” and “evidence that vote totals could have been manipulated” are causing some citizens to question whether their votes were properly counted.
The atmosphere has become especially charged after President Obama’s press secretary announced that an investigation had been launched into a “pattern of malicious cyber activity related to our presidential election cycle.”
Shortly after a recount petition was filed that alleged “fraud or mistake” by county and state officials, a group of consultants at Anderson Economic Group began a forensic analysis of the 2016 election results. We used statistical techniques to evaluate county-by-county vote data in Michigan and one other state, for each major party candidate and for both the 2012 and 2016 elections. The purpose of such methods is to detect data anomalies that may indicate fraud or tampering. However, these statistical methods cannot, by themselves, actually prove or disprove fraud.
We used both simple and advanced data analytics techniques for this purpose, many of which we also apply for our clients in business and government. Usually, we use these techniques to determine market areas, pick the best location to penetrate a market, or assess how one company is doing selling its products relative to another. This time, we used them to help our fellow citizens evaluate serious charges about the integrity of elections.
Now, assume for a moment that Russian hackers, or a conspiracy of election officials, or a software bug implanted by malicious attackers, had somehow changed the vote totals in several counties by significant amounts. It would have affected the total number of votes and the pattern of such votes across counties. We used statistical methods to look for such anomalies.
The first statistical test was simple enough: we examined a histogram of vote shares for major party candidates in two elections. These created a familiar, lopsided curve centered on the voting preferences for the median county in each election. Of course, we know this curve shifts to the right or left each election, but we expect the shape will remain similar. It did — revealing no anomaly.
We then looked for outlier counties — counties where the results were more than two standard deviations from the mean. Sure enough, we found three counties: Wayne, Washtenaw and Ingham. Statistical outliers, however, are usually not the product of a conspiracy.
We also modeled the statistical relationship between 2012 and 2016 voting preferences by county. This analysis shows a consistent pattern: counties that voted more for the Republican Party candidate in 2016 typically did the same thing in 2012, and vice versa.
Finally, we generated a cumulative distribution, which we consider a “signature” of voting preferences, for both major party candidates in 2016. We compared those distributions with the ones for the same election 4 years prior. Systematic vote tampering would have changed the shape. No such change was apparent.
In a letter to the Secretary of State, we concluded there is no evidence from these results that suggests irregularities or episodic fraud or tampering with the Michigan 2016 general election results for the office of president.
Indeed, what we observed was a relatively smooth distribution of votes across counties; a lack of unexplained outliers; and a reasonably predictable shift in recorded votes in the same counties for major parties across two successive elections.
Patrick L. Anderson is principal and CEO of Anderson Economic Group.