We rightly admire novelty in scientific research, but also know how hard it is to get novel work published, and funded in the first place. Study sections often decide by consensus, and consensus kills novelty. 1/

An inventive study examined the link between novelty and citation by analyzing 920,000 publications funded by 170,000 grants from the NSF and NIH between 2008 and 2016. It used machine learning to quantify grant novelty and linked grants to subsequent papers. 2/

Wow: all else equal, a scientific article that had arisen from a fully-novel grant had double the citations compared to a fully-incremental grant, and it appeared in higher prestige journals. https://arxiv.org/abs/1911.02712  via Han Zhuang & @daniel_akuna h/t @alexvespi 3/