Over the last few decades, research has grown ever more international. Big projects, like major astronomical observatories, genome sequencing, and particle physics, are all based on large teams of researchers spread across multiple institutions. And, because of the technology that makes remote work possible, even small collaborations that cross countries or continents have become increasingly commonplace.
In theory, this should make it easier for researchers to build teams that have the right talents to bring a scientific project to completion. But is it working out that way? Some recent studies have indicated that the research we produce may be getting increasingly derivative. And a study released today ties that directly to the growth in what it calls "remote collaboration."
So, is science-by-Zoom at fault? While it's a possibility worth exploring, it's difficult to separate cause and effect at this point.
Measuring collaboration and creativity
The new work was performed by three researchers: Yiling Lin, Carl Benedikt Frey, and Lingfei Wu. It's based on a simple idea, namely that "scientists in on-site teams are better placed to fuse knowledge and conceive the next breakthrough idea." Following up on those ideas, however, may require talents or access to equipment that the local team lacks, so they turn to long-distance collaborations to get the data they need to test their ideas. So, you'd expect that local teams would be behind the most disruptive research and that large, dispersed teams would be performing the incremental work that pushes these disruptive ideas into acceptance.
The challenge of following up on this sort of hypothesis is figuring out how to measure the features of these different types of research. Getting the data is not a problem—scientific developments are cataloged in the peer-reviewed literature, and we have lots of large databases of publications. Figuring out how to identify which ones contain disruptive ideas, and were written by distributed teams, however, is substantially more challenging.
For distributed teams, the researchers focused on city-based proximity. If any two authors of a paper were in the same city, they were considered part of a team that could frequently meet on-site. As soon as a research team included someone from a different city, however, then it was considered a remote collaboration.
Disruptive research is harder to measure, although a number of different methods have been developed for doing so. Most of these methods involve analyzing how future research cites the original work. For this paper, Lin, Frey, and Wu develop what they call a "D score," which is based on a simple rule: If subsequent papers cite both the research paper in question and the papers cited in it, then the work in the paper is incremental—it fits in with the general flow of ideas. If subsequent papers that cite the research in question do not cite its references, then that's a sign that the research paper took a field in a new direction.
So, the Watson and Crick paper on the structure of DNA gets a D score of 0.96 out of a possible 1.0, placing it among the top 1 percent of disruptive papers. By contrast, the human genome paper was built on a lot of earlier work and only gets a D score of -0.017, putting it in the bottom 10 percent of disruptive papers.
The approach was used to evaluate over 20 million papers, with 22.5 million scientists contributing as authors, all published between 1960 and 2020. Separately, a bit over 4 million patents with 2.7 million authors were also considered (with patent data starting in 1976).
Growing collaboration, shrinking impact
One thing that the data clearly shows is that long-distance collaborations are becoming increasingly common. At the start of the study period, the average distance between team members was only 100 kilometers. By the end, it's roughly 1,000 kilometers. The same trend occurred in patents, where the average distance went from 250 km to 750 km.
The possibility of publishing disruptive research moved in the opposite direction during this period. Lin, Frey, and Wu calculate what they call a probability of disruption and find that it drops from 28 percent to 22 percent in papers as collaboration distance rises from local work to 600 km. For patents, the change is from 67 percent to 55 percent.
The authors do a variety of related analyses to confirm these results are robust, including using different measures of disruptive research. "Our findings consistently point to the continued value of geographic proximity for disruptive innovation," they conclude. This holds true regardless of the scientific field or the size of the teams involved.
Many journals require that papers include a mention of the role of each author on the paper—things like designing or performing experiments, writing the paper, and so on. The researchers used those to separate conceptual and practical work and find that scientists who are part of remote teams are more likely to be involved with technical tasks, like performing experiments and analyzing data; meaning, a lot of the conceptual work in science seems to go on locally.
In addition, the researchers look at the role of experience in what researchers are doing, with experience estimated by the number of previously published papers. They find that inexperienced researchers are more likely to be involved in conceptual work when working locally. "We conclude that on-site teams are particularly important as they serve as an escalator for new talent to co-lead in conceptualizing the next breakthrough," Lin, Frey, and Wu conclude.
What are we looking at here?
The researchers pretty clearly feel that these results have important implications. For one, they tie it to the boom in long-distance collaborations more generally, writing, "the post-pandemic shift towards remote work will probably favour incremental innovation at the expense of disruptive discoveries." They also argue that it should have implications for managers and policymakers. Managers, they suggest, should consider specifically assigning creative or disruptive projects to local teams, and incremental work to broader collaborations. Policymakers, by contrast, might consider how to lower the barriers to different types of collaborations.
But that might be a bit premature, as this study is identifying a correlation without really identifying a clear causal mechanism that could mediate the correlation. It's easy to imagine that casual, in person meetings can help foster the development of new ideas. But it's also easy to imagine that disruptive ideas start out somewhat more tentative—supported by a smaller number of experiments that require a lower diversity of expertise. Basically, when trying out a new idea that might be a major leap, scientists are willing to work with the sort of experimental approach that they already know how to manage.
By contrast, filling in the gaps left when a new concept is introduced may require a broader range of experimental approaches, driving people to look further afield to find the expertise they need.
Right now, it's difficult to tease apart these two explanations for the correlation. One key might be in the size of the teams. You might expect that recruiting the diverse expertise needed for follow-on work would lead to larger, more distributed teams, but the correlation held despite team size. Still, until that's investigated a bit more thoroughly, it's probably a bit premature to be setting policy based on this paper.
In any case, it's a bit ironic that, should this idea turn out to be right, the researchers themselves will be undercutting their own data, given that the study's team members work nearly 6,000 kilometers apart.
Nature, 2023. DOI: 10.1038/s41586-023-06767-1 (About DOIs).
