Mark Pelczarski was ready to retire. This was 2011; he was teaching computer science in Chicago by then, but that was really just the capstone on a legendary career in software. In 1979, Pelczarski wrote Magic Paintbrush, an artmaking program for the Apple II, the first personal computer capable of color. He started Penguin Software two years later to publish classics like Graphics Magician, and in the late 1980s he went on to develop music software, create a CD-ROM precursor to Google Maps, and play steel drums with Jimmy Buffett. It’s safe to say that computers look and sound the way they do, at least a little bit, because of Mark Pelczarski’s code.
But just when he was about to call it quits, the head of tech for Barack Obama’s reelection campaign called him, asking if any of Pelczarski’s students might have internship potential for their tech team. Pelczarski asked what kind of skills the Obamaites were looking for. “It was a little bit beyond what my students could do, but I was in my last semester at that point,” Pelczarski says. “I said, ‘I might be able to help you a little bit.’”
He ended up working at campaign headquarters for the next year as a kind of deployable asset, running a team that solved the hard tech problems for departments as varied as finance, oppo, and data. That’s when the folks on the campaign responsible for getting voters to the polls—and making sure they didn’t have to wait in long lines—came by. The electoral meltdown of 2000 led to all kinds of changes in the law and the introduction of lots of new voting technologies; long lines in 2004 had again brought chaos. In the name of “voter protection,” this team was working with a balky spreadsheet that was supposed to estimate the capacity of a given polling place based on data like how many check-in stations and voting machines it had and how many people were likely to vote there—the numbers county election workers use to plan for what each station needs so everyone can vote fast.
Keeping people from voting is suppression—whether it comes from literacy tests, men with guns, or just making things damn inconvenient. This is the raw logistics of democracy, the stuff that happens at the county and precinct level that determines the outcomes of American elections. So of course it goes awry. Like, all the time.
The Obama voter protection team showed Pelczarski their spreadsheet and asked him if he could update it. Pelczarski had written his first computer simulation as a high school student in the 1970s, to predict possible outcomes for the World Series. He wrote it in Fortran. On punch cards. So, yeah, he could update the spreadsheet.
“I basically decided to write a simulation program to step little simulated voters through a polling place, based on check-in stations, voting booths, ballot scanners, and all that stuff,” Pelczarski says. “So I spent a lot of time researching what kind of data was available from previous elections. It took a lot of digging.”
Unlike this year, when lines are long again.
But why? Well, that’s both depressingly simple and blindingly complex. Pelczarski’s digging brought him to a specific intersection on the map of research into how democracy functions—a place where political science and sociology cross over with mathematics and operations research. One way to understand why people sometimes wait a long time to vote is by using a field of math called queueing theory. But to understand what causes the queues … ah, well, that fault lies along other lines.
Long lines at a polling place are the same as long lines anywhere else, whether those lines are made of people waiting to buy a movie ticket (wistful sigh) or widgets waiting to be automatically processed into bigger widget-pluses. Fundamentally, the movement of the line is limited by how many resources are available to process the elements in the queue. That’s called a generalized-assignment problem—getting a certain number of jobs done by a certain number of agents in the most efficient way. That problem is NP-hard, which is how mathematicians say “Ooh, yikes.”
Oh, but then an election also has multiple places to vote within a given district. You have a finite number of poll workers, a finite number of voting machines, a finite number of polling stations, and a finite number of voters. If you’re a good election official, you actually know all those numbers except the last one, for which you have a good estimate based on population (assuming the census isn’t screwed up), voter registration, and prior turnout. Now you have to apportion it all. In queuing theory, that’s called a resource allocation problem. It also is not easy.
You can see right away how many variables go into the velocity of the processing and therefore the length of the line of voters waiting to get processed. How many people show up at once? How many agents are there to process them? How long does processing take?
So this problem isn’t just a matter of getting people through the door of the polling place. The number of things that can get processed through a given resource is called throughput, but in addition to checking in—which can go swimmingly fast in places where the laws are light, or drowningly slow if people have to show ID and get checked off against a computer database that might be unfamiliar to the poll worker—then people have to actually vote. If a ballot is long, that takes longer per person. Paper ballots have to get scanned in; that takes time per ballot. Votes registered on some digital voting machines like those that became common after 2000 might be instantly registered or might have paper receipts that have to get processed. “One way you can think about lines and the polling place experience is as a system of inputs and outputs,” says Stephen Pettigrew, director of data sciences at the University of Pennsylvania’s Program on Opinion Research and Election Studies. “The steps in the system in most states are: You have a check-in step where they verify your voter registration status, and then there’s the step of actually voting. Lines out the door can be a consequence of bottlenecks at any of those steps.”
More simply, if enough people arrive at the same time, they’ll swamp the capacity of the polling place. Result: line. And that’s bad for democracy. “Balking” is when people see how long a line is and decide not to bother; “reneging” is getting in line and then bailing out because the wait is too long. It might actually be rarer; time is money, and people treat waiting in line as an investment. Once you’re in for long enough, you’re less likely to get out.
Even after the debacle of 2000 and long lines in Ohio during the 2004 presidential election, most work in this field was theoretical, based on estimates. One of Pelczarski’s main sources changed all that. For the 2008 primaries, researchers based at UC Berkeley sent more than 100 observers to 30 polling stations around California’s Bay Area. Some used classic paper ballots, some used paper ballots that then got put through optical scanners, and some used the then-newish direct-recording electronic voting machines, or DREs. What they found scuttled a bunch of common assumptions.
First, the researchers found that disparities in the experience and age of poll workers didn’t speed up the line much. More poll workers didn’t help; neither did having more experienced ones. Voting method was a problem—back then, people voting on DREs took longer, so those places were more likely to have lines. But the most striking finding codified what observers had long suspected: Rich people and white people don’t have to wait in line. Or, more specifically, people at polling stations in neighborhoods with incomes above the California median voted 32 seconds faster than people at polling stations in neighborhoods where income was below the state median. Put more starkly, they wrote, “for each additional $10,000 of median household income in the Census block group most closely aligned to the precinct, voters could expect to stand at the check-in table for eleven fewer seconds.”
And since income largely tracks to ethnicity in the United States, that means minorities are disproportionately likely to have long lines at the polls. Racist? Sure. But why? Why, as seems to be the case, are poor and minority neighborhoods getting fewer resources—poll workers and polling machines? Even the polling locations themselves tend to be objectively crummier (less nice but also smaller, which lowers their capacity and lengthens lines as well). And some states have switched to having fewer polling places, sometimes for budgetary reasons. (Even if the new polling place is a massive site like a convention center or stadium, where lines to actually vote might be shorter, it still takes longer to get there, park, and vote—a different kind of queue.)
“I hesitate to claim that this is any sort of nefarious form of discrimination, in part because traditionally you want to put the resources where you think the voters are going to be, and up to the point I was looking at the data, [which was] pre-2008, you tended to have turnout higher in white areas than Black areas,” Pettigrew says. “Now that’s changed to an extent. Now Black Americans turn out at a higher rate than white Americans.” That suggests the racism might have been less nefarious than structural, as states and local election officials made decisions about where to send resources based on outdated or prejudiced information. Which maybe makes it worse.
Since 2008, the structural and political problems with apportioning all the stuff that makes lines move fast have only gotten worse. Eight years after the Berkeley team looked at Northern California, an even bigger team fanned observers out across the country—605 polling places in 19 states. In a paper published in 2019, this work, by dozens of authors, set the table for the light chaos already unfolding this year. Race wasn’t the key, they found—at least, not directly. “There is just no evidence that these racial differences pop up. What does pop up were voter ID effects,” says Bob Stein, a political scientist at Rice University who helped lead the study.
More and more elections require people to show ID at the polling place; Democrats think that suppresses turnout, and Republicans think it stops voter fraud. Stein’s team found that neither of those things is true, but that having to show ID and confirm it against a database does add time to the encounter. And that’s the problem. “There’s a difference in African American and non-white populations and the access to IDs,” Stein says. Most people have that ID, but a few don’t—just a few percentage points fewer among residents of mostly non-white neighborhoods than in majority-white ones. But it’s enough. “Like a car crash on a freeway, if one person in the line doesn’t have ID, it creates tremendous backup,” Stein says. “That 5 percent and 10 percent differential just shuts the line down.”
Election officials call the bend in the graph where time spent in line suddenly spikes “the elbow of death.” Every polling place has its own tipping point, a set of conditions that send line length soaring. “You get to a certain number of people coming in and things go smoothly, and then you get this point where things start going bad,” Pelczarski says. “Very quickly, you get a few more people, and things go terrible immediately. And they don’t recover.”
Which brings us to 2020. “This year will be more difficult because of Covid-19. We really don’t know if the Election Day voters will be more or less,” says Muer Yang, an operations and supply chain management professor at the University of St. Thomas who studies poll lines. “We know there’s a lot of mail-in ballots, but will that be an increase in overall turnout? Early voters might be more or less.”
Even as Republican lawmakers and lawyers attempt to limit the count of ballots sent by mail, 80 million Americans have already voted with less than a week to go before the official Election Day—in some places exceeding overall turnout from four years ago. And even though research as recent as last summer suggested no partisan advantage to voting by mail, it seems clear so far that in the face of President Donald Trump campaigning in part on how bad it is, Democrats by far outweigh Republicans in their use of the service. “We’ve polarized everything, from wearing masks to what TV shows you watch. Republicans won’t vote by mail,” Stein says.
So election watchers expect a flood of Republican voters on Election Day—which could mean long lines preferentially affecting one party. Or maybe it won’t. The areas that have had polling place consolidation might not also be heavily Republican. Republican neighborhoods might also be the ones getting more volunteer poll workers. “It could be that the lines will be incredibly short on Election Day, depending on where you go. It’s very hard to figure out. We’re really in uncharted waters,” Stein says. “But if you have to turn out all of your voters on one day, with Covid, that’s hard. If I were a Republican, I’d worry. It could rain.”
Rain? There could be another hurricane. If that keeps just a few thousand Florida or Texas voters home on Election Day, and those voters were going to be almost entirely Republican? Those states are electoral toss-ups.
In fact, the supermodel-thinness of some state margins is exactly what shocked Mike Pelczarksi back in 2012, what kept him from retiring to a life of steel drums. In that paper on the 2008 California primaries, out of 11,858 arrivals at polls, 225 people reneged—that’s 1.89 percent. The observers didn’t count balks at all, so the number of people who didn’t even try to vote because of lines was even higher. But let’s stick with 1.89 percent as a solid estimate. Because as the Berkeley researchers point out, the Minnesota senate race in 2008, the Washington gubernatorial race in 2004, and the 2000 presidential race in Florida were all decided by a margin of less than 0.01 percent. “I realized that it was an issue that could have flipped both the 2000 and 2004 elections,” Pelczarski says, “I thought, ‘Oh, this is a big deal. That election in 2012 was looking like it would be really close again.’”
Donald Trump won Wisconsin, the tipping-point state in 2016, by 0.77 percent. The statistics site Fivethirtyeight puts the polling average between Trump and Joe Biden in Ohio at 0.6 percent, and the likely tipping-point state of Pennsylvania at just 5 percent. So putting a higher likelihood of reneging out of a long Election Day line on one party or the other could have profound, election-changing consequences. “Election officials, they’re a pretty good lot. They really try to make things work for the most part. But if you have a state legislature putting handcuffs on you, limiting what you can do, limiting funds for hiring poll workers or buying voting equipment, there isn’t a lot they can do about it,” Pelczarski says. “I’m encouraged to see all of the lines right now. I’d rather see them now than on Election Day.” He still hasn’t retired. Even the best simulations can’t tell anyone what’s going to happen on the ground on November 3.