In this paper we examine backward patent citations to public research for
possible sources of measurement error as indicators of knowledge flows.
Employing a unique dataset that combines patent citations and a survey measure
of knowledge flows, we find that patent citations to non-patent references
(e.g., scientific publications) are a meaningful—although limited—measure of
knowledge flows while citations to patent references are less informative. More
precisely, our results indicate that patent citations reflect the flow of
codified knowledge through open science and the direct influence of
technological opportunities, yet they also suffer from two sources of
non-classical measurement error. First, citations omit the flow of tacit
knowledge through private interactions and geographic proximity associated with
true knowledge flows. Second, and more critically, they include citations not
associated with true knowledge flows through firm patenting and citing
behaviors such as choosing not to patent the output of basic research and the
academic norm to extensively cite the scientific literature. We estimate the
magnitude and direction of this bias by using our survey measure as an
instrumental variable and find that patent citations understate the impact of
public research by approximately 55%. These results have implications not only
for understanding the full impact of public research on firm innovative
performance, but also on possible sources of measurement error in patent
citations as indicators of intrafirm, interfirm, and cross-national knowledge
flows.