The Use of Science for Inventions and its Identification: Patent level evidence matched with survey

While backward citation information disclosed in patent documents is often used for tracing the scientific sources of innovations, it is still poorly understood how well the backward citations trace the actual knowledge flow from science. This paper directly evaluates both the completeness and the noise of the inventor citation information, linking the results of an original inventor survey on scientific sources to the dataset of non-patent literatures (NPLs) revealed in the entire patent document. We find that patent citations to NPLs are not only noisy but also highly incomplete. More important science sources are not necessarily more revealed. However, controlling for the propensity to cite NPLs, our estimation results show that the revealed NPLs are more likely to predict the existence of important scientific sources when the inventor refers to highly cited scientific literature early after its publication. We also find that the NPLs revealed at the place where an invention is described provide important additional information in identifying science sources.