On the one hand is the trope of the great scientist: a loner in a rumpled lab coat, content to spend every waking hour chasing the extraordinary from the recesses of his own considerable mind. On the other is the truth about how science actually gets made.

In an influential 2007 paper, Kellogg School professors Brian Uzzi and Benjamin Jones, along with a colleague, analyzed the nearly twenty million research articles in the Web of Science (WOS) database to determine how the production of research has changed over time. “What we saw was that from the fifties up until today, there had been a shift to teams,” explains Uzzi, a professor of management and organizations and the faculty director of the Kellogg Architectures of Collaboration Initiative. “Teams were not only becoming more prominent, but they were becoming bigger each year.”  Teams were also, across a majority of disciplines, increasingly producing the most impactful papers—those capable of setting, or resetting, the research agenda of an entire field. It became clear to us, says Uzzi, that “science had made a fundamental change.”

There are many reasons why the best science may have shifted to teams. Perhaps team research is simply favored by funding agencies, or maybe it plays well at tenure time. But one reason that struck Uzzi and Jones as both plausible and fascinating is the idea that collaborations might foster more creative or novel research.

“You want to be grounded in something that’s well understood and yet be adding in the piece that’s truly unusual.”

Make It New
“There’s a very longstanding idea that the creation of a new thing is about putting existing things together in a new way,” says Jones, an associate professor of management and strategy and the faculty director of the Kellogg Innovation and Entrepreneurship Initiative. “That is, combinations are the key material of creative insight.”

As scientific knowledge has expanded, individuals have had to specialize; teams may allow for depth without sacrificing the ability to combine ideas from different subfields or even disciplines. And there is nothing like the interplay between multiple distinct personalities and perspectives to fuse unlikely but inspired amalgams. Take John Lennon and Paul McCartney. “If it hadn’t been for the two of them bouncing ideas off of each other—always wanting to one up the other—maybe we’d never have Sgt. Pepper,” says Uzzi.

But does the novelty of a research project affect its impact on a scientific discipline? In a paper published this week in Science, Uzzi, Jones, and collaborators Satyam Mukherjee of the Northwestern Institute on Complex Systems (NICO) and Michael Stringer of Datascope Analytics (formerly of NICO) find that novelty does indeed contribute to a research paper’s reception—but only when balanced by highly conventional knowledge. The team also finds that research produced by teams tends to better achieve this balance than research by solo scientists.

Measuring Novelty
In their new study, the researchers measured the novelty of a given research paper by determining whether its influences—the references it cites in its bibliography—go beyond the usual suspects.

Fittingly, they adapted their methodology from an unexpected source. “We wound up building off a system that I heard that the Coen brothers had used to build novelty into their films,” explains Uzzi. After writing a draft of the script for their first film, Blood Simple, Uzzi explains, the brothers found it humdrum. So they cut their script into small pieces and reconstructed it haphazardly—a scene here, a moment there. “And in fact when you watch Blood Simple, you’re like, wow! What a twist in the story. How on earth could that happen? Who would have thought of such a thing? And what you realize is, they did it by building this randomness into it. So we try to recreate something like this in the paper.”

Specifically, the researchers asked whether individual research papers draw largely from combinations of sources that appear together often in the WOS—which spans disciplines as disparate as sociology and nanotechnology—or whether the combinations are closer to what would be expected if citations were drawn randomly, as if from a hat.

The Right Combination
The researchers found that, in actuality, novel combinations of citations are rare, and getting rarer: in the 1990s, just 2.7% of articles had a median citation pairing that was combined less often than expected by chance—down from 3.5% in the 1980s.

They also found novelty by itself was no hallmark of success. “What’s interesting,” says Uzzi, “is most of the work done is conventional. And some of the work is truly novel. And the chances of either one of those classifications of papers being hits is about the same.” Only about 5% of research papers that draw from only very novel or only very conventional sources were among the most highly cited papers in the database.

But there was a third category of research that had nearly twice the likelihood of making it big: papers that relied mostly on conventional combinations of sources but also included a small subset of highly novel ones. “It isn’t all about novelty or conventionality. It’s about both,” explains Jones, who was somewhat surprised by this result. The researchers were also struck by how consistent the trend appears to be, spanning over fifty years and across hundreds of disciplines.

But, says Jones, it also makes sense. “You want to be grounded in something that’s well understood and yet be adding in the piece that’s truly unusual. And if you do those two things [and] stretch yourself in both directions, then you radically increase your probability of hitting a homerun.” Uzzi concurs. “Many of these novel combinations are really two conventional ideas in their own domains,” he says. “You’re taking established, well-accepted ideas, which is a wonderful foundation—you need that in science. But when you put them together: wow. That’s suddenly something really different.”

The Role of Collaboration
Regardless of whether papers were authored by a sole scientist or a team of them, this “virtuous mix”—plenty of conventionality and a dash of originality—was most likely to produce high-impact papers. But teamwork played a role as well. Teams of three or more scientists produced overall more novel work than scientists working by themselves or in pairs. Teams were also overrepresented as authors of papers that hit the conventionality–originality sweet spot. Finally, teams do more with the same mixture of novelty and conventionality. That is, whether a paper’s influences are 90% conventional and 10% novel or 10% conventional and 90% novel, the paper has a better chance of making a splash if it is produced by a team. Teams are better at finding novelty but also, it seems, at assimilating it.

The success of Jones and Uzzi’s own collaboration—this is the pair’s third Science paper in a decade, each produced with contributions from outside colleagues—is itself an endorsement of team-based science. Uzzi is trained as a sociologist, Jones as an economist. “There’s actually pretty big gaps in the way that a sociologist and an economist look at the world,” says Uzzi. Economists are interested in individuals behaving rationally, while sociologists study groups of people behaving in ways that are often not rational at all. “It’s interesting that we study collaboration and one of our findings is that people who bring together expertise from different domains may be an essential feature of progress,” agrees Jones. “We’ve never tested our own papers using these metrics, but maybe we should!” He adds, “It’s clear to me that we both bring important insights and expertise to the problem in a way that makes the whole greater than the sum of the parts.”