というNBER論文をサイモン・ジョンソンらが上げているungated版)。原題は「Place-Based Productivity and Costs in Science」で、著者はJonathan Gruber(MIT)、Simon Johnson(同)、Enrico Moretti(UCバークレー)。

Scientists located in areas with a larger stock of related scientists tend to be more productive (Moretti, 2021). This is likely to be an important factor in explaining the geographical concentration of innovative activity in the U.S. and in other developed countries (Kerr, 2020).
Our findings suggest that, on average across U.S. cities, the productivity gains stemming from agglomeration exceed the higher research costs that characterize larger clusters. However, the ratio of productivity gains and costs increases varies significantly across areas. For areas that currently have a small cluster, the productivity gains of adding an additional scientist are much larger than the corresponding cost increases. For the largest innovation clusters (i.e., those with most scientists), the productivity gains of adding an additional scientist are still larger than the corresponding cost increases, but not by much.
A natural question is whether there is a case for place-based government technology policy, even in the absence of costs that offset productivity differentials. A first issue to consider in this respect is equity. In a world of imperfect mobility and imperfect information on underlying needs, place has a role in redistribution (e.g. Gaubert et. al,. 2021). Redistributing by place allows a tool for targeting needy individuals that are missed by other redistributive systems. This is particularly important given the findings of Bell at al. (2019) on the intergenerational correlation of patenting.
A second argument in favor of these policies is robustness to geographic shocks – particularly in a nation as large as the United States. Catastrophes, man-made or natural, would have an outsized effect on the U.S. if they happen to occur in the very few, most agglomerated locations. A broader portfolio of technology centers provides a form of insurance against geographically focused shocks.
A third, and by far the most speculative argument, relates to politics. Gruber and Johnson (2019) point out that one of the reasons for the weak public support in the U.S. for public investment in R&D is the geographical concentration of such investment. In a nation where voting is related to population and geography, not income or productivity, investments that concentrate their benefits in small geographic (even if densely populated) places may suffer from a lack of political support. Gruber and Johnson (2019) argue that even if there is some efficiency loss from redistributing R&D, the rate of return to more R&D is high enough that more geographic dispersion may lead to more overall efficiency by raising the level of support for public science spending.