What makes a hit? Understanding why things become popular is one of the foundational questions of marketing, but a surprisingly difficult one. It’s easy – too easy – to come up with reasons for success which are satisfying from a narrative and a commonsense perspective, but which are ultimately completely untestable.
This is where large-scale data analysis can really help. In Brian Uzzi, Satyam Mukherjee, Michael Stringer and Ben Jones’ new paper, Atypical Combinations And Scientific Impact, they turn the lens of analysis onto scientific papers themselves. They look at 1.7 million papers to find out what characterises “hit” – i.e. widely cited – works of science across a variety of disciplines, rating them for novelty and conventionality. (Abstract here, PDF presentation here)
Common sense might suggest that novelty is the main factor – a paper that discovers something completely new will surely have greater impact than one which relies on older and more common knowledge. But this turns out not quite to be the case: hit papers actually do both. A hit paper scores high on new knowledge, but also scores unusually high on “conventionality”. As the authors put it, “papers with an injection of novelty into an otherwise exceptionally familiar mass of prior work are unusually likely to have high impact”. It seems Isaac Newton wasn’t just being humble when he said “If I have seen farther than others, it is by standing on the shoulders of giants.”
This is not a new principle. The authors give Darwin’s Origin Of Species as an example – it devotes its first sections to recapping utterly conventional knowledge on cross-breeding dogs and birds before going into novel territory. The rest is history.
But why does new knowledge need such a solid foundation – not just familiar knowledge, but “exceptionally familiar”? My speculative suggestion is that it’s all to do with how people take in knowledge and make decisions about new information.
Psychologists well know that as humans we understand new information by matching it to what we already know. Confirmation bias, for instance, describes the way in which we interpret information in ways that confirm our existing beliefs. We do this partly because going with the flow of existing knowledge is much easier and imposes less cognitive load than trying to work through the implications of new evidence.
And it seems that something similar is happening with scientific papers. New information without the context of the old makes a paper harder to process, which makes it less likely to be remembered and cited. Present old information without enough new combinations and while the paper will be easy to read, it will also just be overlooked in favour of its sources.
In other words, existing knowledge is the spoonful of sugar which helps the medicine of novelty go down.
This might come in useful the next time you’re writing something that you want to be high impact. Being dazzlingly original or contrary won’t help, even if you’re right. Telling people what they already know, but adding new ingredients to the mix, just might.