The National Science Foundation's (NSF) Division
of Science Resources Statistics (SRS) annually
reports data on research and development (R&D)
performed by for-profit businesses in the United States.
The method used to classify these data by industry has
been revised beginning with reference year 2004. This
InfoBrief describes the effect of this revision on the
distribution by industry of business R&D in the United
States—the major impact being a 40% increase in the
amount of R&D classified in the manufacturing sector.
It also provides guidance to users of these data on
interpreting R&D by industry. More information on the
revised methodology and underlying research will be
discussed in a forthcoming working paper.
Background
NSF's R&D estimates by industry are produced from the
Survey of Industrial Research and Development (SIRD).
In the SIRD all of the R&D reported by a company is
assigned to a single industry out of hundreds defined by
the North American Industrial Classification System
(NAICS). The method used to assign companies to
industries has changed relatively little over the past two
decades. However, over this same period the industrial patterns of R&D performance reported by NSF have
changed dramatically. The most notable change has
been the relative growth of R&D in nonmanufacturing
industries. Before 1983, nonmanufacturing industries
accounted for less than 5% of total industrial R&D
performance, but in 2003 they accounted for 40%
(figure 1). This growth can be largely attributed to
three sectors, which together accounted for over 90%
of nonmanufacturing R&D in 2003: wholesale trade
(NAICS 42); information (NAICS 51); and professional,
scientific, and technical services (NAICS 54).
Figure 1 Source Data: Excel file
The growing share of R&D in nonmanufacturing
industries corresponded with the transition of the U.S.
economy from manufacturing to services. However,
the growth in R&D attributed to the wholesale trade
industries (accounting for $25 billion of R&D in 2003)
appeared unusual because these industries were not
commonly thought to perform large amounts of R&D.
A review of the data underlying the R&D estimates
confirmed that almost all of the R&D classified in the
wholesale trade industry was an artifact of the automated
industry classification methodology.
Further research revealed several scenarios that resulted
in companies being classified into nonmanufacturing
industries such as wholesale trade; management of companies and enterprises (NAICS 55); and professional,
scientific, and technical services:
- Companies with global operations. The SIRD
relies on information from the U.S. Census Bureau
to classify a company into an industry. This information
covers only the U.S. operations of a company.
Therefore, a company that performs its
primary business activity overseas is unlikely to be
classified according to that primary business
activity. This company would be classified according
to its primary activity in the United States—often wholesale trade for manufacturing companies.
- Companies that outsource operations. When
classifying companies, the SIRD has no information
on business activities a company may outsource, only
on the activities performed in a company's own
establishments. Therefore a company that relies
entirely on contractors or business partners to manufacture its products would not be classified in
a manufacturing industry. This company would
more likely be classified in wholesale trade; management
of companies and enterprises; or professional,
scientific, and technical services.
- Early-stage companies. Companies are classified
according to their primary activity. For startup companies
or companies in the process of developing their
first products, this activity is often research and
development. Therefore these companies would be
classified in the same industry (NAICS 54) as companies
whose business it is to provide R&D services
to their customers. Currently, the Census Bureau
has no information on the industries in which the
early stage companies' R&D efforts are directed.
The methodology for classifying companies has been
modified by SRS and the Census Bureau to reduce the
impact of these and other related scenarios on the industry-level R&D estimates published by SRS. The
changes to the methodology include the exclusion of
business establishments within a company known to be
sales offices of manufacturers from the automated
algorithm used to determine a company's industry. In
addition, R&D-performing companies classified by the
automated algorithm into NAICS 55 (management of
companies and enterprises), NAICS 42 (wholesale trade),
or NAICS 5417 (scientific R&D services) receive
additional analyst review and may be reclassified based
on information available from public sources such as
financial reports of public corporations and company
websites. This method will also be used to produce
industry-level R&D estimates for 2005, but SRS and
the Census Bureau continue to investigate ways of
improving the quality of these industry-level R&D
estimates. These investigations may result in further
refinements to the methodology in future years.
Impact of Revised Industry Classification
The revised industry classification methodology implemented
for 2004 resulted in a net increase in R&D
reported in manufacturing industries of $37.8 billion
compared to the previous methodology. Less then 1%
of this revision was the result of reclassification of
federally funded industry R&D—the revision resulted
in a net increase of federally funded industry R&D
classified in manufacturing industries of $56 million.
Because the distribution of federally funded industry
R&D was largely unaffected by the reclassification,
the subsequent discussion will focus on the portion of
industrial R&D financed through company and other
funds (excluding federal funding).
Within manufacturing, the new industry classification
methodology resulted in a net increase of more than $1
billion in the 2004 estimates for five industries (table 1).
The largest adjustment occurred in the estimate for
pharmaceuticals and medicines (NAICS 3254), which
nearly doubled from $15.9 billion as estimated using the
original methodology to $31.4 billion using the revised
methodology. The estimate resulting from the revised
methodology is more in line with private estimates such
as those published by the Pharmaceutical Research and
Manufacturers of America (PhRMA). The remaining
four manufacturing industries with net increases over $1 billion were all computer and electronic products
manufacturing industries: computers and peripheral
equipment (NAICS 3341); communications equipment
(NAICS 3342); semiconductor and other electronic
components (NAICS 3344); and navigational, measuring,
electromedical, and control instruments (NAICS 3345).
The R&D of these four industries combined increased
by $18.1 billion due to the new methodology. Within the
manufacturing industries, the R&D of no industry
decreased by more than $1 billion as a result of the
new industry classification methodology.
Table 1 Source Data: Excel file
On the whole, the net increase in manufacturing R&D
for 2004 as a result of the revised methodology was
balanced by a net decrease in nonmanufacturing R&D.
Among the nonmanufacturing industries, five were
revised by more than $1 billion each as a result of the new
classification methodology. Not all nonmanufacturing
industries declined as a result of the reclassification,
and as shown in table 1 both the software industry
(NAICS 5112) and the computer systems design and
related services industry (NAICS 5415) increased by
over $1 billion. Decreases resulting from the revised
methodology were concentrated in three industries:
wholesale trade, management of companies and
enterprises, and scientific R&D services (NAICS
5417). Together, these three industries decreased by
$44.9 billion as a result of the new methodology. Each of
these industries had been susceptible to misclassifications
in the prior methodology when companies fell into any
of the scenarios described in the previous section.
Guidance for Data Users
The changes implemented in the SIRD industry classification
methodology for 2004 attempt to mitigate the
impact of using either incomplete or imprecise information
about a company's operations when classifying a
company. However, data users should be aware that
conceptual errors can also occur when the categories
used in the survey do not match the categories a data
user wishes to employ. If inappropriate categories are
used or survey information is misclassified by a data
user, the resulting conclusions will be inaccurate.
Therefore, it is important to understand that the SIRD
attempts to classify a company's R&D according to the
primary activity of the company—not the industry
primarily served by the company's products. For example,
a company that designs and produces computers and
navigational instruments for use in airplanes would be classified in one of the computer and electronic products
manufacturing industries (NAICS 334) and not in the
aerospace products and parts industry (NAICS 3364). Data
users should be mindful of similar relationships between
industries when interpreting R&D data reported by industry.
In addition, when analyzing data from the SIRD alongside
data from other surveys, data users should note
whether the industry classification methods of the other
surveys produce results comparable to the SIRD's. Data
users are encouraged to contact SRS if they have specific
questions regarding industry classification of R&D data. For further information, contact
Brandon Shackelford
Research and Development Statistics Program
Division of Science Resources Statistics
National Science Foundation
4201 Wilson Boulevard, Suite 965
Arlington, VA 22230
703-292-4685
bshackel@nsf.gov
Footnotes
[1] The manufacturing sector comprises establishments engaged in
the mechanical, physical, or chemical transformation of materials, substances, or components into new products. For the Survey of Industrial Research and Development (SIRD) nonmanufacturing includes all other nonfarm industries. Using this definition, nonmanufacturing is not synonymous with the service sector. For example, mining and construction are included in the nonmanufacturing sector for the SIRD.
[2] The method used to classify companies in 2003 and earlier
years is described in the technical notes of the annual reports in the
Research and Development in Industry series (http://www.nsf.gov/statistics/industry/).
[3] This data anomaly was discussed in two earlier SRS
publications: National Patterns of Research and Development Resources: 2003, NSF 05-308, Brandon Shackelford (Arlington,
VA 2005) (http://www.nsf.gov/statistics/natlpatterns/) and Increase in U.S. Industrial R&D Expenditures Reported for 2003
Makes Up For Earlier Decline, NSF 06-305, Raymond Wolfe (Arlington, VA 2005) (http://www.nsf.gov/statistics/industry/).
[4] Tables describing the impact of the revised industry
classification methodology will be available in the forthcoming R&D in Industry: 2004 Detailed Statistical Tables.