Weapons of Math Destruction

American workers have recently coined a new idea: “clopening.” A combination of the words “closing” and “opening,” clopening refers to when an employee works late closing one night and comes in early, sometimes just a few hours later, to open up shop the next morning. While having the same employee or employees close the store one night and open it the next morning makes logistical sense for employers, it can create stress and sleep-deprivation for workers. And because retail and food service schedules often arrive on short notice, some employees might only find out a day or two in advance that they have one or more clopenings coming up.

Irregular schedules and clopenings are both products of the Big Data economy. WMDs that treat workers “like cogs in a machine” create these trends and entrench them into the workplace. Scheduling used to be driven by human observation: if an employee at a family-run store noticed that there were no customers on Tuesday morning but a huge rush on Saturday afternoons, the shop might close Tuesdays and hire additional workers for the Saturday shift. But now, businesses use software to analyze customer traffic, determine exactly how many employees need to be in-store and when, and keep staffing (and spending) at a bare minimum. Gone are the days of student workers studying during downtime on the job—now, every moment of every workday is analyzed and scheduled for maximum efficiency.

U.S. government data shows that over two-thirds of food service workers and over half of retail workers find out about scheduling changes with less than a week’s notice. When The New York Times ran a 2014 article about a Starbucks worker named Jannette Navarro—a single mother struggling to work her way through college—Starbucks promised to change its scheduling practices by eliminating clopenings. But a year later, Starbucks hadn’t made good on their word. Minimal staffing was baked into their company culture and operations. Inefficiency is a huge liability at chains like Starbucks, and individual managers could be punished for a downturn in revenue related to inefficient scheduling.

Modern-day scheduling technology is rooted in the discipline of applied mathematics called “operations research”—OR for short. Mathematicians used to use OR to help farmers plan crop plantings. During World War II, OR was used to help the U.S. and British militaries optimize their resources. After the war, OR was used in manufacturing and supply chain logistics, and now it underpins huge companies like Amazon, FedEx, and UPS. But these models exploit workers, bending their lives to unfair schedules. Optimization programs are everywhere now, and they’ve contributed to the creation of what O’Neil calls a “captive workforce.”

These over-optimized schedules create anxiety, sleeplessness, and stress that keep workers down, no matter how often they switch jobs in search of a better system. And when workers’ schedules are in chaos, their children grow up without routines. In this way, WMDs are seriously affecting people that they shouldn’t even touch. Efficiency outweighs goodness and justice—and this, O’Neil asserts, is the very nature of capitalism. 

In 2008, a company called Cataphora created a software system that used information gathered from employees’ corporate emails and messaging systems to determine what kind of workers they were. People who sent emails that others copied and pasted a lot, for instance, could be seen as ideas generators; other workers were the “neurons” that connected people and only transmitted information.

As this type of analysis became more popular in workplaces around the country, it started to have terrible consequences. Call-center employees were monitored so that their tones of voice and speech patterns could be analyzed for efficiency. And when the 2007 financial crisis hit, less-useful employees across the workforce were laid off based on the software’s determination of their work styles were useful enough. This was the result of “digital phrenology.”

People who get fired because of these kinds of algorithmic metrics don’t always deserve to lose their jobs—but because they do, the algorithm is reassured that it’s working, and its criteria become even more entrenched. Now, tech workers and creative types are often beholden to the same crude analysis and efficiency measures that dictate the lives of overworked retail and service workers.

In 1983, the Reagan administration warned of a “rising tide of mediocrity” in American schools—SAT scores seemed to be plummeting. The administration’s report suggested that underperforming teachers were the cause of the drop, and that they needed to be weeded out. This alarm was the root of programs that would derail the lives of people like Sarah Wysocki. Teachers are workers too, and they are extremely vulnerable to WMDs.

When Tim Clifford, a middle school English teacher in New York City, scored abysmally on a model’s evaluation of his performance, he was devastated—but soon, he found out that many of the educators he’d worked with for years were scoring dangerously low, as well. He had tenure, so his job was spared—and the next year, his score shot from a miserable 6 to a brilliant 96. Clifford now knew for sure that the scores were arbitrary and “bogus”—yet they were threatening to ruin teachers’ lives.

Researchers and analysists following up on the Reagan-era panic in the later 1980s found that the initial outcry was unjustified. Earlier analysts had overlooked the fact that lots of factors (e.g., more students taking the test, and universities opening their doors to more diverse student bodies) were having an impact on changing scores. When these new researchers divided the scores up into subgroups based on economic status, they found that they weren’t dropping that sharply at all. The phenomenon behind this misinterpretation is called Simpson’s paradox, in which a body of data displays one trend when taken as a whole but shows the opposite trend when broken up into groups.

Botched statistics led to Tim Clifford’s struggles, too. His scores were random. In the effort to make sure that teachers were being measured based on how much they were helping their students to improve, the value-added model being used to gauge success tried to predict what a student’s score would be and reward or warn teachers based on the gap between the expectation and the reality. But because teachers’ classes change every year—and because a class of 25 or 30 students isn’t a big enough data set, the result is a model that is essentially “noise.”

The scores Clifford and his colleagues were getting were meaningless—but the “bogus WMD” creating them was still gaining traction. The Obama administration sought to reform legislation that judged school districts based on test scores alone, creating a law that would let states turn around underperforming districts on their own terms. Even though strict correlation between a school’s test scores and its overall health is falling out of favor politically, lawmakers and school board officials still aren’t rejecting WMDs outright (or even recognizing that they’re unfair). Value-added modeling, Clifford grimly predicts, isn’t going anywhere anytime soon. Like inhumane scheduling software at major corporations, it’s just too entrenched.