Trump’s curiously high support in certain Wisconsin counties: A statistical analysis

Computer scientists from some of America’s most prestigious universities are confident that electronic voting machines are hackable. Edward Snowden suggests that a post-election statistical analysis may be able to detect such a hack.

Recently, the Green Party filed for a general election recount in the state of Wisconsin, and the Clinton campaign has joined this effort. Do the reported results suggest a hack? We used publicly available data (reported results + Census 2013 data), together with standard data analysis procedures, to begin to address this question.

The Verified Voting organization provides a map that divides Wisconsin’s counties into counties that use paper ballots (22) and counties that use paper ballots AND electronic vote recording systems (50.)

We tested to see whether, on average, these two different types of counties ended up with significantly different shares of their votes to the two major candidates: Hillary Clinton and Donald Trump.

While we found that Clinton and Trump received a roughly equal share of votes in counties that use only paper ballots, Trump received around 20% more of the vote than did Clinton in counties that use electronic voting machines in addition to paper ballots.

While the difference is striking, it is possible that the populations of these counties are just very different. And, even if the populations aren’t different, voters might act in different ways depending on whether they are voting on a paper ballot or on a machine. We do not have the resources to test this second possibility, but we tried to check for the possibility that the characteristics of the peoples of the counties could explain Trump’s performance.

We used a statistical procedure known as “regression,” as it allows us to examine how the demographic profile of the various counties can help to explain Trump’s share of the vote. We decided to examine the roles of the following demographic factors: ethnicity (% African-Americans, Asian-Americans, European-Americans, Hispanic-Americans), household income, and % of people with a college education. These are common explanations for voting behavior.

Here are the statistical results, showing how the different factors are related to Trump’s share of the vote:

First off, these results show some well-known patterns: As the share of European-Americans in a county increases, support for Trump increases; as the share of college-degree holders increases, support for Trump decreases. This is not surprising.

The interesting pattern is that, even when taking factors such as ethnicity and education into account, counties that use electronic voting machines showed higher support for Trump than counties that only use paper ballots (yellow highlight).

What do we make of these numbers?

This research is an initial step in understanding the role of the voting machines in Wisconsin’s general election. One possibility is that a hack truly has occurred, but more research is needed before anyone can reach such a conclusion. And, importantly, even if the patterns of data were caused by a hack, the source of the hack remains unknown (it could be foreign, but it could also be internal to the US, independent of foreign influence.)

As such, we encourage the people of Wisconsin as well as other Americans to examine the results of the election not only in the statistical manner we have done, but also by inspecting the machines, interviewing elections officials, and requesting a formal investigation (independent of the recount). Such a process of engagement with the results can only strengthen the security of America’s voting system.

About: This piece is co-authored with Axel Geijsel. To support our non-funded research on elections, please visit our GoFundMe. You may contact us at and

Acknowledgements: We thank an anonymous scientist for feedback.

Data and Model Selection: We have posted our data online. While there are many more covariates that are worth looking at, we were hesitant in overfitting our model. We encourage others to examine any other possible covariates.

Conflict of Interest Disclosure: Cortes, an American citizen, voted third party. Geijsel is a European citizen.

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.