The news that Hillary Clinton leads Donald Trump in the popular vote by over two million votes (and counting) is no doubt one of many factors giving rise to demands for recounts in states where the vote totals are very close. Green Party candidate Jill Stein has raised over $3 million so far to finance recounts in Michigan, Pennsylvania and Wisconsin, not to change her own outcomes but because “the unexpected results of the election and the anomolies that have been reported need to be investigated. We deserve elections we can trust.” The total cost of a recount in these three states is $6-7 million.
The unexpected results that require investigation are not the differences between pre-election polls and the outcome. There are many explanations for this discrepancy, including real changes in opinion in the last few days of the campaign and partisan differences in enthusiasm leading to differences in turnout. The real concern of experts in voter research is the discrepancy between the exit poll results and the official vote counts.
Exit polls are interviews with voters conducted, usually by news organizations, as they are leaving the polling place. Representative polling places are chosen, and researchers are given quotas to fill broken down by age, race, gender, etc. These data are then weighted by their best approximation of the demographic composition of the actual voters. In some countries, exit polls results are considered more reliable than official vote totals, and international observers use them to detect voter fraud.
Above are the exit poll-vote count comparisons in four states, including Pennsylvania and Wisconsin, in which the exit polls showed Clinton winning while the official totals gave the victory to Trump. Adding these two percentages together gives you the “red shift,” or the extent to which the official results unexpectedly favored the Republican candidate. In Michigan, the exit polls showed both Clinton and Trump at 46.8%, while the official totals stand at Trump 47.60% and Clinton 47.33%—a miniscule red shift, but worth recounting due to the vote’s closeness. Had Clinton won these three states, she would have won the Electoral College by 288 to 250.
If exit polls are subject only to random error, red and blue shifts should occur with equal frequency. Voting researcher Jonathan Simon, who compiled these data, has found a systematic tendency toward red shifts in all presidential elections since 2004. Blue shifts are almost nonexistent. However, a better name for these discrepancies might be a “conservative shift,” since in this year’s Democratic primaries, exit polls regularly showed higher percentages for Bernie Sanders than the official vote counts.
What, other than fraud, can account for these discrepancies? Here are the most likely suspects.
- Early voting. Early voters are not exit polled. But if a recount is done, it should be easy to separate early voters from those who voted on Election Day.
- Selection bias. The people interviewed may not be representative of the actual voters. This could occur if the interviewer chooses unrepresentative people, such as those who are smiling. A more likely possibility is that Republicans are more likely to refuse to answer. For example, if white male Republicans systematically decline to be interviewed, the white male quota will be filled by more Democrats. Neither the quotas given to the researchers nor the weighting formula totally correct for selection bias.
- Response bias. Some of the respondents who voted for Trump may falsely claim to have voted for Clinton. This is a variation of the Bradley effect, named for Los Angeles Mayor Tom Bradley, in which polls systematically overestimate support for Black candidates and underestimate support for candidates who appeal to voters’ prejudice. In this case, it would mean that some voters are embarrassed to admit to a stranger that they voted for Trump.
Of course, the existence of these possible sources of error doesn’t mean that a recount is not worth doing, only that Democrats should have a low expectation of success.
The main argument for doing these recounts is the apparent ease with which computerized voting systems can be hacked. Alex Halderman, director of the University of Michigan Center for Computer Security, notes that, while he does not personally believe the outcome of the election was tainted by fraud, this year has seen an unprecedented number of cyberattacks apparently intended to influence the election. He explains that, prior to the election, hackers could probe election systems in various localities to see which ones can be hacked easily. Immediately before the election, they could select highly competitive districts and install malware designed to shift the vote totals by a couple of percentage points. The malware would erase itself after the election.
There’s no question that this is possible for technically sophisticated attackers. (If my Ph.D. students and I were criminals, I’m sure we could pull it off.) If anyone reasonably skilled is sufficiently motivated and willing to face the risk of getting caught, it’s happened already.
If the malware is self-erasing, tampering can only be detected when there are backup paper ballots. However, Halderman mentions the possibility that examining the voting equipment might produce evidence of some types of fraud.
Clearly paper ballots are the best technology for preventing and detecting computerized voter fraud. Optical scanning can be used to count the ballots quickly. But many districts, including some in Pennsylvania, continue to use computer systems known to be insecure with no paper ballots as backups. This is unacceptable in 2016.
The deadlines for filing for recounts are November 23 in Pennsylvania, November 25 in Wisconsin, and November 28 in Michigan. Contributions to the Green Party recount effort can be made at https://jillstein.nationbuilder.com/recount. It’s hard to believe the Democratic National Committee isn’t willing to finance this effort.
You may also be interested in reading: