No matter how they arranged the data, though, it was almost impossible to find patterns—or any evidence that a team’s composition was correlated with its success. “We looked at 180 teams from all over the company,” said Dubey. “We had lots of data, but there was nothing showing that a mix of specific personality types or skills or backgrounds made any difference. The ‘who’ part of the equation didn’t seem to matter.”
Some productive Google teams, for instance, were composed of friends who played sports together outside of work. Others were made up of people who were basically strangers away from the conference room. Some groups preferred strong managers. Others wanted a flatter structure. Most confounding of all, sometimes two teams would have nearly identical compositions, with overlapping memberships, but radically different levels of effectiveness. “At Google, we’re good at finding patterns,” said Dubey. “There weren’t strong patterns here.”
So Project Aristotle turned to a different approach. There was a second body of academic research that focused on what are known as “group norms.” “Any group, over time, develops collective norms about appropriate behavior,” a team of psychologists had written in the Sociology of Sport Journal. Norms are the traditions, behavioral standards, and unwritten rules that govern how we function. When a team comes to an unspoken consensus that avoiding disagreement is more valuable than debate, that’s a norm asserting itself. If a team develops a culture that encourages differences of opinion and spurns groupthink, that’s another norm holding sway. Team members might behave certain ways as individuals—they may chafe against authority or prefer working independently—but often, inside a group, there’s a set of norms that override those preferences and encourage deference to the team.
The Project Aristotle researchers went back to their data and analyzed it again, this time looking for norms. They found that some teams consistently allowed people to interrupt one another. Others enforced taking conversational turns. Some teams celebrated birthdays and began each meeting with a few minutes of informal chitchat. Others got right to business. There were teams that contained extroverts who hewed to the group’s sedate norms whenever they assembled, and others where introverts came out of their shells as soon as meetings began.
And some norms, the data indicated, consistently correlated with high team effectiveness. One engineer, for instance, told the researchers that his team leader “is direct and straightforward, which creates a safe space for you to take risks….She also takes the time to ask how we are, figure out how she can help you and support you.” That was one of the most effective groups inside Google.
Alternately, another engineer told the researchers that his “team leader has poor emotional control. He panics over small issues and keeps trying to grab control. I would hate to be driving with him in the passenger seat, because he would keep trying to grab the steering wheel and crash the car.” That team did not perform well.
Most of all, though, employees talked about how various teams felt. “And that made a lot of sense to me, maybe because of my experiences at Yale,” Julia said. “I’d been on some teams that left me feeling totally exhausted and others where I got so much energy from the group.”
There is strong evidence that group norms play a critical role in shaping the emotional experience of participating in a team. Research by psychologists from Yale, Harvard, Berkeley, the University of Oregon, and elsewhere indicate that norms determine whether we feel safe or threatened, enervated or excited, and motivated or discouraged by our teammates. Julia’s study group at Yale, for instance, felt draining because the norms—the tussles over leadership, the pressure to constantly demonstrate expertise, the tendency to critique—had put her on guard. In contrast, the norms of her case competition team—enthusiasm for one another’s ideas, withholding criticisms, encouraging people to take a leadership role or hang back as they wanted—allowed everyone to be friendly and unconstrained. Coordination was easy.
Group norms, the researchers on Project Aristotle concluded, were the answer to improving Google’s teams. “The data finally started making sense,” said Dubey. “We had to manage the how of teams, not the who.”
The question, however, was which norms mattered most. Google’s research had identified dozens of norms that seemed important—and, sometimes, the norms of one effective team contradicted the norms of another, equally successful group. Was it better to let everyone speak as much as they wanted, or should strong leaders end meandering debates? Was it more effective for people to openly disagree with one another, or should conflicts be downplayed? Which norms were most crucial?
II.
In 1991, a first-year PhD student named Amy Edmondson began visiting hospital wards, intending to show that good teamwork and good medicine went hand in hand. But the data kept saying she was wrong.
Edmondson was studying organizational behavior at Harvard. A professor had asked her to help with a study of medical mistakes, and so Edmondson, on the prowl for a dissertation topic, started visiting recovery rooms, talking to nurses, and paging through error reports from two Boston hospitals. In one cardiac ward, she discovered that a nurse had accidentally given a patient an IV of lidocaine, an anesthetic, rather than heparin, a blood thinner. In an orthopedic ward, a patient was given amphetamines rather than aspirin. “You would be shocked at how many mistakes occur every day,” Edmondson told me. “Not because of incompetence, but because hospitals are really complicated places and there’s usually a large team—as many as two dozen nurses and techs and doctors—who might be involved in each patient’s care. That’s a lot of opportunities for something to slip through the cracks.”
Some parts of the hospitals Edmondson visited seemed more accident prone than others. The orthopedic ward, for instance, reported an average of one error every three weeks; the cardiac ward, on the other hand, reported a mistake almost every other day. Edmondson also found that the various departments had very different cultures. In the cardiac ward nurses were chatty and informal; they gossiped in the hallways and had pictures of their kids on the walls. In orthopedics, people were more sedate. Nurse managers wore business suits rather than scrubs and asked everyone to keep the public areas free of personal items and clutter. Perhaps, Edmondson thought, she could study the various teams’ cultures and see if they correlated with error rates.
She and a colleague created a survey to measure team cohesion on various wards. She asked nurses to describe how frequently their team leader set clear goals and whether teammates discussed conflicts openly or avoided tense conversations. She measured the satisfaction, happiness, and self-motivation of different groups and hired a research assistant to observe the wards for two months.
“I figured it would be pretty straightforward,” Edmondson told me. “The units with the strongest sense of teamwork would have the lowest error rates.” Except, when she tabulated her data, Edmondson found exactly the opposite. The wards with the strongest team cohesion had far more errors. She checked the data again. It didn’t make any sense. Why would strong teams make more mistakes?