March 26, 2017

Which States have the Highest Risk of an E-Voting Meltdown?

This post is joint work by Joshua Kroll, Ian Davey, Alex Halderman, and Ed Felten.

Computer scientists, including us, have long been skeptical of electronic voting systems. E-voting systems are computers, with all of the attendant problems. If something goes wrong, can the problem be detected? Can it be fixed? Some e-voting systems are much riskier than others.

As the 2012 Presidential election approaches, we decided to evaluate the risk of a “meltdown scenario” in which problems with electronic voting equipment cause a state to cast the deciding electoral college vote that would flip the election winner from one candidate to the other. We’re interested in the risk of these technological problems, weighted by the relative voting power of each voter. So for example, here in New Jersey we use direct-recording electronic voting machines that have been found by a court to be inadequate, but with Obama polling at +14% it’s not likely that a snafu with these machines could change the entire state’s outcome. But in swing states that poll closer to even, like Virginia (where your voting machines can be modified to play Pac-Man), an electronic voting mix-up could have a much bigger impact. So, which states have the greatest risk of an e-voting meltdown affecting the result of the 2012 Presidential election?

A meltdown scenario is very unlikely, of course. A knife-edge election is highly improbable. Still, we can evaluate the relative risk of a worst-case scenario in each state. Here is how we did it:

First, we created a model of electronic voting risk, using data from the recent Counting Votes 2012: A State by State Look at Election Preparedness report and the Verifier database. Our risk model takes information on every county in the US and combines it into a state-by-state risk score. We took into account which voting technology each county uses, whether paper records are used (and whether those records are marked directly by voters or are machine generated), which procedures are in place if machines fail, how ballots or electronic media are physically protected and accounted for, and what kind of auditing is in place to detect problems after the election. We then weighted our per-state risk calculation by the number of registered voters in the state to estimate the probability, per voter in a given state, that a bad e-voting event will take place.

Next, we combined these risk scores with the meta-polling analysis performed by our Princeton colleague Professor Sam Wang. Professor Wang’s meta-analysis provides a measure of the relative power of a single vote in each state based on the number of electoral votes it affects and the current polls. Professor Wang also provides a prediction of how the election would go were it held today (assuming that the polls are accurate). His predictions are updated four times each day based on the newest polls, so you can follow along. We refresh with his latest vote power data periodically, so the map and top-ten list in this post will update automatically. The most-red states on the map are those most at risk, according to our model. The scores you see are simply a relative measure: the model is normalized so that the highest-risk state scores exactly 100.

Like any model, this one has caveats. For example, we’re not taking complete account of certain kinds of variation between states, such as variations in the probability that a large number of votes could be affected by malfunction or malice in central tallying systems. And this kind of analysis ignores separate voting problems that have historically been important, like the number and location of polling places or even the availability of enough voting machines or ballots. We’re also working hard to get additional resolution for our model so we can more precisely resolve county-by-county risk or risk for other kinds of races (e.g. the Senate).

We certainly hope that electronic voting problems won’t happen at all this year, and especially that they won’t affect the result of the election. Even if a single state is called into question due to malfunctions, public trust in the election system is eroded. But we hope our model can help to fuel the ongoing conversation about reliable voting technology by pointing out the “hot spots” where improvements in the technology and procedures fielded for elections would do the most to ensure a more trustworthy Presidential election.

[Editor’s note: On November 1, the Center for Information Technology Policy is hosting a live-streamed virtual symposium on the state of electronic voting, entitled “E-Voting: Risk and Opportunity”.]


  1. John Millington says:

    I usually cringe whenever my state (NM) makes a top-ten list (it’s usually bad news) but this time, it looks like we made the list because we have “powerful” votes. Whew!

  2. David Hayes says:

    Not on the list because Texas is reliably red, but….

    My county has Diebold electronic voting machines. When I vote, the election officials give me a sticker. There are two choices. One says “I Voted”, the other reads “My Vote Counted”. I won’t accept a “My Vote Counted” sticker because I have no faith that it is correct. I go through the motions but I have no confidence in the accuracy of the tally. I’ve looked into early voting, but that’s still done with the electronic systems. Absentee voting is done on paper, but under Texas law I’m not eligible to vote absentee unless I spend an entire month away from home.

    • I too vote, but I too am pretty confident that most of the time my vote is not counted. Last voting season I think they might have counted, at least they reported in such a way as that they might have counted. In the past, they have never reported votes that didn’t go to the two main parties. It was always one or the other, and I rarely vote for one of those parties. So, in the past, I suspected they just tossed out every vote that wasn’t for the main two.

      Last time they did start reporting other votes, but lumped them all together, so it is hard to really know if my vote really counted or not.

  3. fotis zygoulis says:

    A hell of a project!! Congratulations!! You should probably think of the importance each State has to the voting process.

  4. Jerry Smith says:

    Florida uses paper ballots with an post election audit – I can not see how we are on this list.

    • Joshua Kroll says:

      It looks like the only bad thing our data say about Florida is that its audit procedures are graded “Needs Improvement” by the Counting Votes 2012 report. That, combined with the fact that Florida has a high vote power ranking and a lot of registered voters is what gives it a high score. Remember: the vote power ranking depends on changes in polling data, so it may be that Florida is especially close to tipping one way or the other right now.

    • Jessica Ray says:

      I too was gravely concerned to see Florida placed on the list as it is. Of all the systems we’ve studied my county seems to have the best possible mixture of security and accountability measures. I am participating in poll worker training this morning. I will look out for possible reasons we’re considered insecure. Very interesting. This is quite a project.

  5. Luther Weeks says:

    Strange seeing Connecticut with a 6, with NJ and MA at 0 risk.
    MA has the same optical scanners we do, but without post-election audits.
    NJ has paperless DREs as far as I know, not auditable, not recountable, with documented past errors.

  6. My computer can ‘be modified to play Pacman’. That does not impact on its reliablity.

    • True, but the fact that it can be modified to play Pacman means it can also be modified (programmed) to deliberately miscount things.

  7. Joyce McCloy says:

    I’m confused as to why North Carolina is in the top ten.

    We passed a new law in 2005 to address issues we had in the 2004 election and before.

    Then in 2008, NC cut the under-vote rate for President in half as compared to our 2004 GE.

    The post election pre certification audit of 2008 was good: Obama won by about 14,000 votes out of over 4.3 million ballots cast – and Dr William Kalsbeek, who studied the audit results said that ‘the statistical probability that Obama in fact defeated McCain in the North Carolina election is higher than 99.9 percent’

    North Carolina makes it very easy to vote:
    – NC does not require ID except for first time voters who registered by mail,
    – Voters can cast ballots in person at early voting sites from 19 to 3 days before Election Day,
    – NC residents may register In-Person and vote at a One-Stop Site in the person’s county of residence during the One-Stop Absentee Voting period.
    – polling places that have DREs (that use a paper roll) also have a supply of paper ballots if needed.

    I don’t vote in the early voting period but instead vote in person on election day, and I’ve never had to wait more than about 3 minutes in line to check in and then vote at my neighborhood polling place.

  8. The 2000 election was not a failure of the voting machines of the time; it was a failure of the people in place that did not follow the laws and statutes of Florida. Nothing has changed in Florida, in fact it is worse.

    A new law in Florida prohibits a recount unless the vote difference between 2 candidates for President is within one half of one percent. Based on 2008 that means the 2 candidates statewide must be closer than 42,000 votes. There can not be an audit of one machine, precinct, or county without the state wide difference being within this margin.

    Voting by these machines assumes that the Supervisor of Elections and all employees are honest! The problems of the E-Voting is that there is no auditing or recorded log generated in the machine of the activity and time markers of the programs or of the time to scan and approve the ballot. Without this to compare how the supposedly same ballot was handled by different machines in multiple precincts there is no way to audit the process of voting as a whole. The recount window is has been cut off if the corruption is big; we need to treat the machines a network and audit them before, during, and after an election.

    The DS200 voting machine should verify a single page ballot within 2 to 3 seconds. In the primary election they were taking 8 to 9 seconds for one page. No one is interested in looking into the reason why. They are looking to see what happens at the election but no responsible party and that includes the Federal Government, the media and public interest groups want to touch the problem before the election.

    It is not that we do not have the technical ability to have an honest election but that officials and un-official parties seem to stand back and allow the election fraud to take place!

  9. As a poll captain in Ohio, I know that every voter can verify their vote, both as a computer screeen review and on the hard copy tape as it is printed. Since I know the machines I’m responsible for are in working order and print what is entered, I have as much faith as possible of the results coming out of my polling location. The only thing I can’t control is the software programming. However, with the printout, random checks post election can verify the computer count, which I believe is required in Ohio. I hope this is helpful.

  10. Is the map out of sync with the table? The table suggests Nevada has the highest risk and Ohio is #4 behind Colorado (among others). But the map currently shows Ohio to be the hottest, and Colorado and Nevada are pink.

    • Joshua Kroll says:

      You’re right! Sorry about that – it looks like we had a display problem with the map that should be resolved now. You might have to clear your cache to get the fresh map, though. I certainly did.

  11. Dave Felt says:

    Ohio is #1 ?? Isn’t that tantamount to being hacked ?