American Economic Review 2021, 111(10): 3328–3375

The Effect of High-Tech Clusters on the Productivity of Top Inventors^†

By Enrico Moretti*

The high-tech sector is concentrated in a small number of cities. The ten largest clusters in computer science, semiconductors, and biology account for 69 percent, 77 percent, and 59 percent of all US inventors, respectively. Using longitudinal data on 109,846 inventors, I find that geographical agglomeration results in significant productivity gains.

Firms in the innovation sector display a strong tendency to cluster geographically by research field (Carlino et al. 2012). Prominent examples include the internet and software clusters in Seattle anchored by Amazon and Microsoft, respectively; the medical research and biotech clusters in Boston.

The geographic concentration of high-tech sectors is not just a curiosity — it has important implications for cities and states. The presence of a high-tech sector has been shown to be a key driver of local economic growth as innovation-oriented industries have taken on larger roles (Glaeser and Saiz 2004).

3328

3336 the american economic review october 2021

Table 1 lists the largest clusters in each field. I measure cluster size as the number of inventors in a city × field × year, excluding the focal inventor, as a share of all inventors in field × year:

Note: Cluster size is defined as number of inventors in a city × field × year, excluding the focal inventor, as a share of all inventors in field × year.

The numbers point to a remarkable degree of agglomeration: in each field, the top ten clusters account for the majority of inventors. Importantly, the degree of concentration has remained stable over the sample period.

vol. 111 no. 10 moretti: high-tech clusters and top inventors 3337

II. Data and Descriptive Statistics

I use data on US inventors from the Harvard Business School patent database, which covers utility patents granted between 1971 and 2007. Each record reports the inventor's name, city of residence, technology class of the patent, and the number of citations ultimately received.

Inventor productivity is measured by the number of patents in a year and by citation-weighted patents. Following the literature, I assign each inventor to one of three broad research fields — computer science, biology and chemistry, and semiconductors — based on the modal technology class of her patents.

Figure 2. Inventor Productivity and Cluster Size

The figure shows a strong positive cross-sectional relationship between cluster size and the productivity of top inventors in all three fields.

Of course, this relationship is not necessarily causal. More productive inventors may select into larger clusters, and firms in larger clusters may differ systematically in ways that are correlated with patenting. The empirical strategy below addresses both concerns.

To limit the influence of occasional inventors, I restrict the sample to inventors who patent at least twice over the sample period. The resulting panel includes 109,846 inventors observed for an average of 11.1 years.

3338 the american economic review october 2021

A second concern is that moves are endogenous: inventors may relocate precisely when their productivity is about to change. To assess the importance of this channel, I examine the dynamics of productivity around the move year directly.

IV. Event-Study Estimates

I estimate event-study models centered on the year of the move. The specification relates the log number of patents of inventor i in field f, originally in city j, who moved to city c, in year t, to the log size of the destination cluster:

where the coefficients of interest are the betas, which measure the elasticity of inventor productivity with respect to cluster size in the years before and after the move.

The model includes city × field, city × year, and inventor fixed effects. Standard errors are clustered at the city level.