In the spring of 2019, Caulfield launched another investigation. This time, he would attempt to determine how much clerks were paying for their own voting equipment. To keep the project simple, he focused only on the cost of machines, ignoring the battery of technology—electronic poll books, centralized ballot counters, computer software—that often come attached. It was on this second go-round, while he submerged himself in ever-more obscure studies and journals, that Caulfield began to suspect that the smartest people in the elections had been living, essentially, in a hall of mirrors.
How much did a voting machine cost? Caulfield chased down numbers with the doggedness of a prosecutor. The most popular number came from a state lobbying group—but it listed no source. That figure had been popularized by another report, which estimated the cost of replacing every voting machine in America at a billion dollars. But deep in the footnotes, Caulfield saw that the number was plucked at random from a comically huge interval: 2.9 billion dollars, give-or-take. And its source, the state lobbying group, would afterwards stop referencing its own figures. The only other sources were a few stray remarks from legislators in New Jersey and North Carolina, who once made some guesses about what their states had paid for machines. Importantly, there had been a scattering of studies that collected a handful of vendor contracts. But those were mostly decades old, and useless for making sense of a national market. Within a matter of weeks, Caulfield had discovered a statistical merry-go-round, in which everyone had cited everyone else, and hard data never really appeared.
The federal government was charged with keeping track of certain voting machine data. But it didn’t have anything concerning the price or value of the technology in use. In 2014, the White House had convened its own report on election administration. But on the question of prices and markets, the information it relied on was pure anecdote. Even Caulfield didn’t appreciate how little anyone knew until he received an email from a private research firm. Caulfield had noticed the firm was updating its financial figures for the voting machine companies, and he emailed the firm excitedly to ask what they’d learned. In response, the research firm sent Caulfield an article from USA Today—referencing Caulfield’s own Wharton study. They hadn’t realized, evidently, that some of the estimates Caulfield included in that study had come, in fact, from the research firm itself.
Within a matter of weeks, Caulfield had discovered a statistical merry-go-round, in which everyone had cited everyone else, and hard data never really appeared.
Around this time, bedraggled and perplexed, Caulfield felt somewhat like the outcast traders of Wall Street during 2008, who went searching for the adults who understood the financial system, and discovered that there were no adults who understood the financial system. “You’d go to presentations. And at these things, people would make claims based on, ‘Oh, we collected contracts from three counties in Virginia!’” Caulfield told me. “People spend years going to conferences making speeches confidently proclaiming the answers to election technology. And then they don’t have an answer to any of these questions.”
It took Caulfield a little time before he figured out what to do. There were estimated to be approximately 3,000 counties in the United States, each with an office responsible for handling or tracking the purchases of its own voting equipment. Why not just ask them?
Caulfield linked up again with his old research partner, Coopersmith, and together they recruited a small team of undergraduates. For a few weeks, the team collected thousands of emails and phone numbers. Then they went about contacting them, one by one. The students’ request was put in simple terms: Did they still have the contract they paid for their voting machine? And would they share it with a gaggle of teenagers from Penn?
The outreach took about eight weeks. That summer, Caulfield was sitting at his desk in Philadelphia when the first responses began to trickle in. Several of the officials wrote back, apologetically, to say that there weren’t any records. They had purchased their machines over a decade ago, some said, or under a different clerk. A few simply relied on word-of-mouth. “One county said something like, “Well, it was back in 1996, and we don’t have the records anymore, but I think we paid X,” Caulfield recalled. “There were a non-trivial number of jurisdictions that literally did not know how much they paid for their own voting system.”
Then there were a few clerks who got angry. One election official in Texas reported Caulfield to the attorney general’s office and alerted the election companies to their mischief. (Caufield described the experience as “a little intimidating.”) One pair of Massachusetts clerks wrote primly that their wooden box had served them just fine for more than a century, “and we don’t have any plans to update our system.”
But there were enough election officials who played along. After several weeks, the data Caulfield’s team had hauled in was significant: 356 jurisdictions agreed to share their vendor contracts, by far the largest such dataset ever assembled. Peering into the computer, Caulfield and his team were the first people to see what America had been paying for its voting machines, all this time. They began analyzing the contracts closely, one at a time, crunching the numbers of thousands of pages of documents. “Once we started doing that, it was really stunning,” Caulfield said. “There was no pattern to it.”
Plotted onto a graph, the variation in what counties were paying for their technology appeared to be practically random. After some analysis, Caulfield began to figure out why. In the first place, unit prices for the machines weren’t uniform. Voting machines had clusters that indicated a mean price; the ES&S DS200, thought to be the most popular voting machine model in use, had a median price tag of $5,750 listed in the contracts. But the variation could be wide. There was another reason for the apparent differences in what counties were paying. Voting companies had been dispensing what they called “discounts” to local governments. These discounts were sprinkled across the contracts and came in dozens of forms. It wasn’t the discounts themselves that startled Caulfield. It was that their widespread use didn’t appear to have a shred of coherence.
I contacted all three major voting-machine companies for this story, asking whether Caulfield’s findings reflected their practice nationally, and why the prices appeared to vary so much from one buyer to the next. Dominion and Hart declined to comment. A spokesperson for ES&S, the country’s largest vendor, said that “all contracts are competitively priced” and reflected a variety of factors, including “early adoption of technology or how the county will pay for the voting system.”
In the data, Caulfield discovered clerks who were buying the same voting machines from the same company, but seeing significantly different discounts than their peers—sometimes in the same state. In California, Mono County and Placer County both purchased orders of Dominion equipment, including the ImageCast Evolution, a voting machine priced at $7,200. But Mono received a 5 percent discount off of its bottom-line order, while Placer, 128 miles away, saw a mark-off of nearly 25 percent. Volume didn’t necessarily matter, either: Dodge County, Wis. purchased the same ES&S machines that Polk County, Fla. did. Even though Polk County bought substantially more machines and equipment, Dodge got a discount seven times larger than Polk’s—a mismatch Caulfield spotted in other states, including Texas and Virginia.
In some cases, the variations veered toward the absurd. The county of Gila, Ariz., received an “Arizona Customer Discount,” while Coconino, Ariz., did not. In Virginia, Arlington County and Spotsylvania County both received what their contracts called a “Good Faith” discount—but one was 20 percent better than the other’s. “Like a Wild West,” Caulfield described it. As far as he could tell, “there was no logic to it.”
Because of this variation, the data made it hard to draw big conclusions about national trends. But if there was a visible pattern, it was a disquieting one. Counties that tended to be smaller and poorer also tended to receive a smaller discount. Caulfield was hesitant to speculate why this might be. But in his initial Wharton study, a picture emerged that offered a plausible explanation: the “buyer power” of a severely fragmented market. In the world of elections, business was so scarce—and the sharks so frenzied—that losing a contract to a big jurisdiction, like a major U.S. city, could potentially pose a serious loss to a company. Almost any discount might be justified, if the jurisdiction were large enough. By the same measure, a sale to a small, rural county with a few thousand residents would not present nearly the same urgency. To Caulfield, and anyone reading the same data, the voting companies threw discounts at rich jurisdictions because they had to. They charged more to the small ones because they could.
But if the prices showed such wild variation, how did the election companies consistently make money? In his massive dataset, Caulfield discovered the answer. The companies didn’t necessarily need to profit from the machines at all. Instead, they appeared to rake in revenue from the significant fees they reaped in afterward. Maintenance, upkeep, annual software licenses—all came with fees that local election officials continued to pay the private companies, year after year. The fees were sometimes so large that they eclipsed the cost of the machines themselves. Caulfield discovered several counties in Colorado, including Boulder, where the cost of fees after 10 years was 140 percent more than the price that taxpayers had footed for the initial machines. In one county in Oregon, it was 200 percent.
a news outlet NEWSLETTERS
Get a news outlet Magazine's newsletter featuring the best stories of the week and our Friday Cover delivered to your inbox.
As Caulfield and Coopersmith wrote, the key to stable revenue meant finding ways “to sustain long-term relationships with counties”—and the key to that was “ongoing annual payments.” Perhaps that was another reason innovation was scarce. A business model in which repair and maintenance costs are the most stable source of revenue might not, to put it delicately, create optimal incentives for designing hassle-free technology. In fact, as Caulfield and Coopersmith suggested, it might not be financially wise to regularly roll out new models at all.
Oftentimes, election clerks were the very last people who cottoned to this screwball model. “How much choice do you have?” Gretchen Reinemeyer, the director of elections for Arlington County, Va., told me. Caulfield’s work had revealed that, in 2015, Arlington and nearby Spotsylvania County both purchased the same voting machines, but that Arlington’s discount was 20 percent lower—even though its purchase volume was essentially double. “We’re such a small department in Arlington County. We only make up 0.1 percent of the total budget,” Reinemeyer explained. In several cases, she went on, “counties are left having to procure equipment without much time and without being able to go through a thorough procurement process.” Nor do they have the wisdom of past experience. “This is the type of purchase we do once a career,” Reinemeyer said. “The equipment lasts ten years, and most people don't make it 20 years in this business.”
Shortly after Bush v. Gore, the Inspector General in Florida’s Miami-Dade County issued a report on the voting system there. It concluded that the relationship between voting companies and the local government “can be summed up in one word: dependence.” Twenty years afterwards, it could be said that, invoking the words of Bollinger’s commission, this assessment remains applicable. Reinemeyer told me, “It's frustrating to know that our taxpayers are paying tons of money for these machines. But we've to run elections. And we don’t have tons of choices.”
The one company that did reply to my questions, ES&S, said it “work[s] with jurisdictions which lack funding to offer discounted pricing” and other incentives, and pointed out that it’s common in business for larger-volume sales, such as to big metro areas, to receive a greater mark-off. (The company didn’t address Caulfield’s finding that volume did not appear to correlate with price.)
The spokesperson for the company also noted that ES&S “devotes significant resources to research and development to ensure the long-term security, accuracy and reliability of voting systems,” pointing out the extraordinary regulatory thicket it must navigate among local, state and federal officials—as well as “the EAC, DHS, law enforcement, voting system manufacturers, and the election community.”
When I asked concerning the charge that voting machine companies don’t innovate at the pace of other technology businesses—ones that, for example, develop smartphones—the ES&S spokesperson said that a slower pace of innovation “is not an indicator of a lack of innovation.” Plus, the spokesperson said, both sectors aren’t really comparable: “Consumer devices are developed and sold based on a rapidly evolving market and a large demand. In the much smaller elections industry, jurisdictions generally purchase new voting technology every 10 or more years.”
To the charge of failing to innovate, ES&S also proposed a solution: actually funding elections more generously. “Providing necessary sustained funding to jurisdictions will allow them to invest in newer technology, additional innovation, [and] security and resources,” the spokesperson said.