1 For more information on the OECD’s methodology in identifying R&D intensive industries, see Technical Appendix and Galindo-Rueda and Verger (2016). In addition to R&D intensity, KTI industries may be defined using other measures, including high concentrations of workers in STEM occupations (e.g., Wolf and Terrell 2016) or high rates of patenting and innovation activities. However, we are not aware of internationally comparable data for defining KTI industries using these other measures.

2 Specifically, KTI industries are defined as those belonging to the high and medium-high R&D intensity groups (see Technical Appendix and Galindo-Rueda and Verger 2016). A detailed description of each of these industries is provided in the Glossary. The OECD taxonomy also includes the weapons and ammunition industry. This report, however, does not present data on weapons and ammunition, a relatively small industry whose value added accounts for less than 1% of global KTI value added. (See historical data for this industry in the Indicators 2020 edition of this report.)

3 The U.S. production and employment data have been crosswalked from the 2012 North American Industry Classification System (NAICS) to the 4th revision of the International Standard Industrial Classification of All Economic Activities (ISIC, Rev.4) for comparability with the international data. See the Technical Appendix for more details.

4 While some COVID-19 impacts can be explicitly identified in some components of GDP—for instance, the impacts of recovery programs like unemployment insurance or the Paycheck Protection Program on aggregate federal government spending—the total effects of the COVID-19 pandemic cannot be separately identified in the value-added data because the impacts are embedded in source data. For more information, see BEA (2021b).

5 As of the time this report was compiled, the U.S. Food and Drug Administration continues to track shortages in medical devices and supplies throughout the COVID-19 pandemic, primarily in personal protective equipment, testing supplies equipment, and ventilation-related products: https://www.fda.gov/medical-devices/coronavirus-covid-19-and-medical-devices/medical-device-shortages-during-covid-19-public-health-emergency.

6 See Dunn (2021) for a chronicle of supply chain impacts since the beginning of the pandemic.

7 See Technical Appendix for details about the methodological approach to estimating employment in KTI industries.

8 For more information on the definition of the STEM workforce categories, including the STW, see Indicators 2022 report, “The STEM Labor Force of Today: Scientists, Engineers, and Skilled Technical Workers.”

9 NCSES defines underrepresented minorities as Blacks or African Americans, Hispanics or Latinos, and American Indians or Alaska Natives whose representation in S&E education and S&E employment is smaller than their representation in the U.S. population (NCSES WMPD 2021).

10 Some studies have found citizenship to be underreported in U.S. federal surveys like the ACS and the Current Population Survey. In particular, the number of naturalized citizens is found to be overestimated in Census and BLS data, possibly because some noncitizens misreport as citizens. Van Hook and Bachmeier (2013) compared the 2010 ACS PUMS to naturalization records from the Office of Immigration Statistics (OIS). Based on their analysis, the ACS estimates of naturalized citizens were much higher than OIS-based estimates among immigrants from all regions of the world who have lived in the United States fewer than 5 years. Among immigrants residing in the United States for 5 years or more, the OIS-ACS discrepancy is concentrated among those born in Mexico, especially men of all ages and women aged 40 or older. In the 2019 ACS PUMS, the average number of years since naturalization for foreign-born noncitizens ranges from 14 years for workers in IT services to 19 years for those in air and space. Across all KTI industries, 19% of foreign-born workers reported being naturalized with the last 5 years, and 10% reported that their country of birth was Mexico. Hence, some of the KTI citizenship estimates should be considered lower-bound estimates.

11 The COVID-19 pandemic is the latest but not the only unpredictable event to expose the GVC’s vulnerability to disruptions due to unpredictable shocks. It is only the latest in a series of disruptions that includes the 2011 earthquake and tsunami in Japan, the flooding in Thailand the same year, and Hurricane Harvey in Texas in 2017, all of which halted production and created shortages in many sectors (see Lund et al. 2020).

12 The most recent update of the TiVA database includes indicators covering 66 economies including the OECD countries, the European Union (EU) and the Group of Twenty countries (Argentina, Australia, Brazil, Canada, China, France, Germany, India, Indonesia, Italy, Japan, Mexico, Russia, Saudi Arabia, South Africa, South Korea, Turkey, United Kingdom, United States, and the EU), and several East and Southeast Asian economies and South American countries for the years 1995–2018. Indicators are available for 45 industries and provide information on the participation and contribution of industries and countries to various stages in production chains. Additional information and access to the OECD TiVA database is available at https://www.oecd.org/industry/ind/measuring-trade-in-value-added.htm.

13 The investment data include both disclosed and undisclosed transactions. Since a considerable number of transactions are not disclosed, disclosed transactions underestimate the magnitude of total investment. Hence, Arnold, Rahkovsky, and Wang’s (2020) estimates of total investment are, at best, a lower-bound estimate of the actual level of investment.

14 The top 25 AI-related skills identified by Emsi Burning Glass are algorithms, Apache Spark, artificial neural networks, big data, blockchain, computer vision, data science, deep learning, distribute computing, Internet of Things, machine learning, machine learning algorithms, mathematical modeling, natural language processing, Pandas (Python package), predictive analytics, predictive modeling, R (programming language), robotic process automation, Scala (programming language), speech recognition, statistical modeling, TensorFlow, time series, and unstructured data. More information on the Emsi Burning Glass data is provided in the Technical Appendix.

15 The acceleration in growth rate during the pandemic is likely due to several different reasons in addition to the pre-pandemic growth that was already occurring in demand for AI-related skills. According to a 2017 study by the McKinsey Global Institute (Manyika et al. 2017), automation and AI will require an estimated 14% of the global workforce to switch occupations or acquire new skills by 2030. It is possible that the increase in AI-related job postings is driven by the need for a more AI-capable workforce, and employers are seeking to close the projected skills gap by 2030. This is supported by an April 2021 survey by the McKinsey Global Institute (Billing et al. 2021), which found that 58% of executives believe that closing the skills gap in their workforce is a higher priority now than it was pre-pandemic.

16 In recent years, greater attention has been paid to the rise of “superstar” metropolitan areas that have seen their economies grow at rates exceeding economic growth in the United States overall. According to a 2019 study by the Brookings Institution and the Information Technology and Innovation Foundation (Atkinson, Muro, and Whiton 2019), five metropolitan areas in the United States—San Francisco (California), Seattle (Washington), San Jose (California), Boston (Massachusetts), and San Diego (California)—accounted for over 90% of growth in the United States’ “innovation sector” between 2005 and 2017. The study also found that these superstar metropolitan areas have seen a greater increase in employment and jobs that pay higher wages.

17 A recent paper on the development and use of the technology module by the ABS and preliminary findings from the data contain new insights about the level and distribution of GPTs, especially AI, adopted by business (Zolas et al. 2020). The sample size makes the survey one of the largest and most up-to-date data sets available on advanced technology adoption. In particular, small and younger firms are adequately represented. The survey asked three detailed questions: about digitization, cloud computing, and advanced business technologies, including AI-related technologies such as guided vehicles, machine learning, machine vision, natural language processing, and voice recognition software.

18 OECD (2013) also provides a list-based definition (although not exhaustive) to serve as an interpretative guideline to the single definition. It includes DNA or RNA, proteins and other molecules, cell and tissue culture and engineering, process biotechnology techniques, gene and RNA vectors, bioinformatics, and nanobiotechnology. Data on business biotechnology R&D performance for the United States and other countries follow the OECD broad and list-based definition of biotechnology. Other data sources discussed in this section use different definitions. For more information on the different definitions of biotechnology used in each data set, see Technical Appendix.

19 More information on the venture capital data from PitchBook is provided in the Technical Appendix.

20 The 2010–16 data are from the Business R&D and Innovation Survey (BRDIS), and the 2017–18 data are from BRDS. The survey questions regarding R&D performance on biotechnology between the two surveys differ in how they define biotechnology. In BRDIS, biotechnology is “the use of cellular and bio-molecular processes to solve problems or make useful products.” In BRDS, biotechnology is defined according to the OECD definition, that is, “the application of science and technology to living organisms, as well as parts, products and models thereof, to alter living or non-living materials for the production of knowledge, goods and services.” Additionally, the respondent can access a list-based definition (also following the OECD) that states: “The following list provides examples of biotechnology techniques and applications. The list is not intended to be exhaustive, but it is indicative of the types of activities included in the definition of biotechnology, including: DNA/RNA; proteins and other molecules; cell and tissue culture and engineering; process biotechnology techniques; gene and RNA vectors; bioinformatics; and nanobiotechnology.”

21 Powell, Koput, Bowie, and Smith-Doerr (2002) found that venture capital firms were more likely to fund smaller and more science-focused biotechnology companies and that more stable and older biotechnology startups were more likely to secure funding from a nonlocal source. In addition, they found that venture capital investments in Boston and the Bay Area tend to stay local to their respective regions.

22 NIH funding on biotechnology is only a subset of federal funds on biotechnology R&D. Other federal agencies—including the National Institutes of Food and Agriculture, Department of Defense, and Department of Energy—also fund biotechnology R&D, but these other agencies do not produce a comprehensive measure of funding on biotechnology. Arguably, most biotechnology R&D is on medical applications, and examination of NIH funding on biotechnology provides a meaningful indicator of federal support for biotechnology overall.

23 The top 25 biotechnology-related skills identified by Emsi Burning Glass are biopharmaceuticals, biostatistics, case report forms, clinical pharmacy, clinical research, clinical study design, clinical trial management systems, clinical trials, drug development, drug discovery, electronic data capture, good clinical practices, International Council for Harmonisation of Technical Requirement for Pharmaceuticals for Human Use guidelines, key opinion leader development, life sciences, medical affairs, medical devices, medical guideline, non-disclosure agreement (intellectual property law), pharmaceuticals, pharmacovigilance, pre-clinical development, regulatory filings, scientific literature, and Title 21 of the Code of Federal Regulations. More information on the Emsi Burning Glass data is provided in the Technical Appendix.

24 See section “Demand for Workers with AI-Related Skills in the United States” for a discussion of intensity versus concentration.

25 The OECD (2021a) also curates an indicator for biotechnology IP5 patent families based on the World Intellectual Property Organization (see Technical Appendix for differences in biotechnology definitions). Although the absolute number of patent families attributable to the United States varies by definition, the magnitude and trend in U.S. share of total biotechnology IP5 patent families are similar across definitions.