Communication Costs in Science: Evidence from the National Science Foundation Network

 How do communication costs affect the creation of scientific output? This study examines changes in scientific output and citation patterns following an institution's connection to the National Science Foundation Network (NSFnet), an early version of the Internet. Established in 1985 to connect five NSF-sponsored supercomputers, the NSFnet national internet backbone quickly expanded to universities across the United States by linking existing and newly-formed, wide-area regional computer networks. I estimate the effect of connection to the national internet backbone on citations per paper by exploiting plausibly exogenous variation in the connection times of the regional NSFnet networks. Following connection to the national NSFnet, average citations per paper increase by over 10 percent relative to the pre-connection mean. Subgroup analyses reveal that the net effect was driven largely by middle- and top-tier institutions. Finally, I show that NSFnet connection led to a decline in interdisciplinary citations and an increase in within-field citations, but I find no evidence of increases in collaboration patterns.