Association for Public Policy Analysis & Management

This paper optimizes cluster-oriented industrial policies in terms of both the targeted industries and geographical sizes, and speaks to how different measurements affect the results.

Industry targeting, despite its disgraceful name for potential unfairness, is practically unavoidable. Policy makers have to either prioritize and focus resources, or dilute endeavors and achieve little. Thus, the question is not whether to prioritize, but how to prioritize.

Recently, policy makers have recognized the role of clusters—geographical concentrations of industries—in elevating innovation. The U.S. Department of Commerce invested enormously to encourage innovation through clusters in high-tech, biochemistry or energy-related industries in large and jurisdictional-based regions, with no evidence showing the desirability of either their industry choices or their geographical choices.

Prior studies fail to sufficiently guide industry targeting. Most studies propose picking highly concentrated or most frequently patenting industries, without acknowledging that the best performers may not be the ones that benefit the most from clusters. A few studies compare the effects of clusters across industries, but either aggregate industries at high levels (three-digit NAICS code or above) or only compare a limited set of industries within one discipline. Moreover, most studies define the geographical scope of clusters based on jurisdictional boundaries and fail to address both the sensitivity of the results to such more or less arbitrary measurement and which (geographical) cluster size best fits an industry.

This paper uses a unique restricted version of the plant-level dataset from the Quarterly Census of Employment and Wages for the state of Maryland from 2004 to 2013, and matches it with the patent application and citation data from the United States Patent and Trademark Office.

It is the first to (a) compare the effects of clusters on innovation across a wide range of six-digit NAICS industries, (b) rank industries in terms of how much they benefit from clusters, and (b) identify industry-specific optimal geographical sizes of clusters. These are achieved through novel quantitative methods, including a continuous quantile analysis method developed by Combes, et al. (2012), ArcGIS python programming and spatial analysis, and the Newton optimization method.

This paper also varies the measurements of the dependent and independent variables and tracks the changes of the results. It (a) continuously changes the geographical size of clusters from 0.5 to 20 miles in radius, (b) changes the measurement of innovation by applying different weighting schemes to the patent data (e.g., patent grants, patent applications, citation-weighted patent applications, etc.), (c) applies various cluster definitions in terms of industry compositions (i.e., which six-digit NAICS industries belong to the same cluster).

This paper will obtain the following results: (a) a ranking of industrial clusters in terms of the additional patents that can be produced by clustering, (b) the optimal cluster size associated with each industry that maximizes patent production compared to other sizes, and (c) the sensitivity of the above results with respect to different measurements. These results can help practitioners to perform knowledgeable industry targeting and grow clusters in the most learning-prone industries at the optimal geographical sizes.