Supporters of former president Donald Trump’s false claim that he actually won the 2020 election are preparing for the next election by attacking the ballot scanners that tabulate the results, arguing that hand-counting paper ballots is more accurate.
Republicans are arguing that humans are more likely than machines to get the count right. Evidence, however, suggests the opposite: Computers — which ballot scanners rely on — are very good at tedious, repetitive tasks. Humans are bad at them. And counting votes is tedious and repetitive.
While scanners haven’t yet been banned in any state, support for hand-counting paper ballots has surged among closely attuned Republicans and plummeted among similarly attuned Democrats, my research finds. With that opinion shift, expect state legislatures to act in 2023.
What recounts in New Hampshire and Wisconsin reveal
Only two published academic studies have compared the accuracy of hand-counting and scanning, each focusing on one state. In each state, ballots in some localities were counted by hand and in others by scanners. Those counts were later examined in recounts, involving a painstaking review of each ballot.
In the first study, political scientists Stephen Ansolabehere and Andrew Reeves examined recounts in New Hampshire in 2000, 2002 and 2004. Most jurisdictions still counted ballots by hand, but a few had begun using scanners. In the 60 races with recounts in those years, the average difference between the original count and the recount was 1.98 percent when originally counted by hand and 0.95 percent when originally counted by machine.
In the second study, Ansolabehere, Barry Burden, Kenneth Mayer and I compared the original counts against recounts in two statewide Wisconsin races: the 2011 state Supreme Court election between incumbent David Prosser and challenger JoAnne Kloppenburg, and the 2016 presidential election, in which Jill Stein requested a recount.
In Wisconsin’s 2011 Supreme Court election recount, the hand-counted paper ballots differed from the recount by 0.28 percent while the difference for scanned paper ballots was 0.15 percent. In its 2016 recount of the presidential election, hand-counted paper ballots were off by 0.18 percent while scanned ballots were off by 0.13 percent.
Although scanning paper ballots was slightly more accurate than hand-counting in Wisconsin, both modes were incredibly accurate — 10 times more accurate than either hand-counting or scanning in New Hampshire a decade before. Why?
While there are many possible reasons, one stands out: The Wisconsin recounts occurred after the nation’s wholesale revamping of voting technologies after the disputed 2000 election. That election led Congress to pass the Help America Vote Act, which appropriated billions of dollars to replace obsolete machines and put officials on notice that vote counting would be closely scrutinized. Although the New Hampshire recounts took place in 2000 and beyond, these votes had first been counted using pre-2000 technologies and procedures.
Thus, these studies show that both hand counting and scanners can be very accurate, but scanners are better. They also found that when rare, large discrepancies do arise, they occur because of poll workers’ decisions, such as stopping counting because they were tired or not counting write-in ballots considered irrelevant.
Public opinion about voting machines has changed
So far, despite the growing Republican push to banish scanners, local election officials, state legislators and county commissioners haven’t taken the bait. Election officials from both parties have pointed out the practical impossibility of counting tens of millions of ballots by hand for hundreds of thousands of individual races.
Nevertheless, the anti-scanner movement appears to be undermining trust in scanners among Republicans, especially among those most highly engaged. That’s what we found when examining answers to questions about voting technologies on a recent survey, which repeated a series of identical questions asked over the past decade in the Cooperative Election Study.
In a 2020 paper James Dunham and I co-authored, we explored nearly a decade’s worth of CES data about attitudes toward voting technologies. This April, I conducted a new online survey of 1,000 adults with YouGov, asking the same questions that Dunham and I analyzed from the 2010s. This sample was conducted online with an opt-in sample by YouGov. Responses were weighted to demographic characteristics of registered voters for each state as measured by the 2020 Current Population Survey and post-stratified to the 2020 presidential vote split.
In this recent survey, the fraction of Americans saying they would most prefer to vote on hand-counted paper ballots ticked up slightly since 2019, from 14 to 16 percent. The preference for scanners dipped a bit, from 36 percent to 33 percent. Neither difference is statistically significant.
The change was more distinct by party — and even stronger among those partisans who reported that they follow what’s going on in government and public affairs “most of the time.” Among all Republicans, the preference for hand-counting jumped from 13 to 20 percent; and among Republicans who said they closely follow public affairs, from 17 to 31 percent. Most-aware Democrats moved sharply in the opposite direction, with their support for hand-counting ballots dropping from 23 to 9 percent.
To be clear, there’s no evidence that the movement to ban ballot scanners caused all these attitude shifts between 2019 and 2022. A lot has happened over that time. But we can see the most active Republicans starting to support scanning bans, suggesting that Republican-led state legislatures may file bills on this issue after the midterms. Bills have already been filed in Arizona, Missouri and New Hampshire.
A better way forward
If the effort to ban ballot scanners succeeds, it will almost certainly not result in more accurately counted elections, as its leaders promise.
But it is indeed difficult to verify that vote counts have been accurate. Over the past 20 years, election specialists have developed a new form of statistically rigorous post-tabulation audits, called risk-limiting audits (RLAs). A few places — notably Colorado and Georgia — have implemented these audits. This is a flexible, powerful tool that could provide evidence that vote counts were correct, regardless of how they were initially done. But RLAs are unlikely to be adopted and funded if political leaders instead focus on changes that are almost guaranteed to lower the accuracy of vote counts.
Charles Stewart III is the Kenan Sahin Distinguished Professor at the Massachusetts Institute of Technology, the director of the MIT Election Data and Science Lab, and the co-director of the Caltech/MIT Voting Technology Project.