In 2013, The Brookings Institution undertook a study of the automotive industry in the Tennessee Valley region, using a novel establishment-based methodology to generate the sort of granular information necessary for truly understanding the automotive industry as it exists in the real economy.
Measuring the automotive industry is difficult because the industry integrates a great variety of products onto a single platform and into a single system: the automobile. This elongated supply chain includes establishments producing everything from tires (NAICS 3262) to windshields (NAICS 3272) to semiconductors (NAICS 3344). The Bureau of Economic Analysis estimates that in 2011 the motor vehicle industry consumed $393 billion worth of other industries’ output.
Brookings built a dataset from the bottom up by identifying every physical business establishment—defined as a single discrete location of production—involved in producing goods and services for the automotive industry.
To identify establishments in North America producing for the automotive industry, Brookings relied on directories fromMarkLines and ELM Analytics and Hoover's Dun & Bradstreet database.
After merging the two directories, they matched those records to corresponding records from D&B. In addition, they downloaded all D&B records for establishments whose primary activity was classified as one of three auto-exclusive NAICS industry codes: 3361 (motor vehicle manufacturing), 3362 (motor vehicle body and trailer manufacturing, and 3363 (motor vehicle parts manufacturing). They did the same for a larger list of detailed eight-digit Standard Industrial Classification (SIC) codes exclusively related to automotive production. Next, Brookings manually searched D&B by automaker name to ensure that all establishments of major companies had been captured. Finally, Brookings downloaded all establishments in the Hoover’s Auto Family, an automotive industry directory produced by Hoover’s that is similar to MarkLines and ELM.
After duplicates were removed, the final establishment dataset contained 31,605 unique records from the United States, Canada, and Mexico.
Brookings designed its methodology to provide as complete and accurate a picture of the automotive industry in North America as possible. However, this is a snapshot of a dynamic and ever-changing industry. Given the enormity of the task, and the reality that D&B data is compiled via telephonic surveys of companies and reported by humans, the final database inevitably contains error. Brookings took a number of steps to minimize that error.
First, to mitigate selection bias, Brookings utilized the most authoritative third-party industry datasets that had complete North American coverage, but did not augment its database with single-state industry directories—like that kept by the State of Tennessee’s Department of Economic and Community Development—because that would compromise comparability across geographies.
To correct for any employment omissions or potentially faulty estimates in D&B’s data, Brookings substituted ELM Analytics employment numbers—vetted for accuracy by the company—where available. Outlying observations were confirmed or adjusted using company websites, LexisNexis news reports, and satellite imagery to judge establishment size. Brookings imputed average employment in an establishment’s industry code for any remaining establishment whose employment estimates were suppressed by D&B and not available from ELM.