Notes

[8] A similar measure of R&D intensity is the ratio of R&D to value-added (sales minus the cost of materials). Value-added is often used in studies of productivity because it allows analysts to focus on the economic output attributable to the specific industrial sector in question by subtracting materials produced in other sectors. For a more detailed discussion of value-added, see United Nations System of National Accounts 1993 (SNA 1993). For a discussion of the connection between R&D intensity and technological progress, see Nelson (1988).

[9] Industry-level estimates are complicated by the fact that each company's R&D is reported in only one industry (see sidebar, "Industry Classification").

[10] According to NAICS, the utilities industry is limited to establishments engaged in the provision of electric power, natural gas, steam, water, and the removal of sewage. Establishments that provide telephone and other communication services are included in other NAICS industries.

[11] Because federal R&D funding is concentrated among a few companies in a small number of industries, the potential for disclosing information about a particular company is high. Therefore, these data often are suppressed. This prevents the precise tabulation of total R&D performance and the calculation of R&D to net sales ratios for many industries. Appendix table 4-22 presents company-funded R&D to net sales ratios for a wide array of industries.

[12] For a recent study on the role of services industries in R&D and innovation, see Gallaher, Link, and Petrusa (2006).

[13] Suppression of federal R&D funding prohibits the precise tabulation of total R&D performance for some industries (see note 11). Lower-bound analyst estimates are given in cases where potential disclosure of company-reported data or classification issues prevents the publication of total estimates from survey data.

[14] Methodological differences between the PhRMA Annual Membership Survey and the NSF Survey of Industrial Research and Development make it difficult to directly compare estimates from the two surveys. For example, the PhRMA survey definition of R&D includes Phase IV clinical trials (which are trials conducted after the drug is licensed and available for doctors to prescribe), whereas the NSF survey definition does not. Also, the NSF survey sales data may contain income from sources not related to the production of drugs and medicines.

[15] The introduction of a more refined industry classification scheme in 1999 allowed more detailed reporting in nonmanufacturing industries. For the cited 2005 statistic, the R&D expenditures of companies in software, other information, and computer systems design and related services industries were combined. These three industries provided the closest approximation to the broader category cited for earlier years without exceeding the coverage of the broader category.

[16] Suppression of federal R&D funding prohibits the precise tabulation of total R&D performance for some industries (see notes 11 and 13). Lower-bound analyst estimates are given in cases where potential disclosure of company-reported data or classification issues prevents the publication of total estimates from survey data.

[17] NAICS-based R&D estimates are available only back to 1997. Estimates for 1997 and 1998 were bridged from a different industry classification scheme. Total R&D for this sector has grown from $9.2 billion in 1997 to $16.9 billion in 2005.

[18] Because R&D expenses reported on financial documents differ from the data reported on the NSF Survey of Industrial Research and Development, direct comparisons of these sources are not possible. For an explanation of the differences between the two, see Shepherd and Payson (1999).