Scientists produce more novel but less disruptive work as they get older, study of millions of academics reveals

Researchers produce more novel work as they age, according to a new analysis of the citation patterns of more than 12.5 million scientists who published three or more scholarly works between 1960 and 2020. But they are less likely to produce disruptive research because they are building on their existing body of work, rather than pursuing new lines of research, says study co-author Lingfei Wu, who is an information scientist at the University of Pittsburgh in Pennsylvania.

The study measured disruption based on whether future references to an analysis only cite the study itself or the papers mentioned in its bibliography, too. That’s because references ‘can be viewed as the foundation of knowledge or dominant ideas that the current focal paper is built upon’, Wu explains. It’s a way to distinguish between two kinds of future citations that recognise a piece of research as original or as incremental contributions, he says.

Both are vital for science to progress, Wu stresses. That’s because if every paper was novel, there would be ‘chaos’, he explains, and if every paper was an incremental advancement, progress would be too slow. ‘We need papers that verify previous understanding,’ Wu adds. Many are likely negative findings, too, he notes, which are important to find blind alleys and dead ends.

Over 40 years of their career, just under 65,000 scientists across all the fields became, on average, significantly less likely to produce highly disruptive studies as they aged, Wu’s analysis found. This suggests that senior academics are less likely to overturn established ideas.

What’s more, older chemistry and biology researchers were among the least disruptive, the analysis revealed, while senior computer scientists ranked the highest. That’s because research methodologies in fields like artificial intelligence often change significantly every few years, Wu says, whereas fundamental processes and principles in biology and chemistry remain largely similar and are therefore less prone to notable short-term disruption. ‘We cannot change the definition of what high blood pressure is every five years,’ he notes.

Wu’s analysis suggests that at earlier stages, scientists chase fundamentally new ideas, which lead to early disruptive works, whereas at later career stages, that innovation switches to finding applications for that earlier disruptive work. ‘You basically establish new linkages between existing concepts,’ he says. ‘You don’t create a new concept.’

Russell Funk, a management researcher at the University of Minnesota who has studied disruption in science, says Wu’s analysis is one of the most important science-on-science studies in recent years. That’s because it shows that the trend of novelty increasing and disruptiveness decreasing with age, and the tension between them, is a ‘system-level’ trend, he explains.

The new study’s most striking finding is that a researcher’s most cited reference is published, on average, two years before their own first paper and persists as an anchor throughout their career, Funk says. ‘That means every scientist’s intellectual identity gets forged during a narrow training window, and they spend decades gravitating back toward it,’ he notes. ‘Now multiply that by millions of scientists, each pulling their students, collaborators and the papers they review toward their own intellectual centre of gravity.’

As for the implications, Funk says they are ‘clear and difficult’. He suggests funding agencies should invest more in early-career researchers, rather than only list those scientists as contributors on senior scientists’ grants. Funk also thinks having multiple corresponding authors should be encouraged and junior researchers should be given ownership over their intellectual direction.

‘The harder question is whether the system has the will to do any of it,’ Funk says. ‘Career advancement in science basically means accumulating more gatekeeping power over time, and this paper shows that [this] has costs.’