NSF 95-304
There is no single survey of R&D activity in the United States. Rather, NSF sponsors a series of surveys to collect data on the financial and human resources devoted to R&D in the various sectors
of the U.S. economy (defined above). Although these surveys are not designed specifically for this purpose, they provide the primary source material for estimating the national R&D totals. Respondents indicate the amounts they spend on R&D in their
own sector and, generally, the sources of these funds. To the greatest extent possible, national totals are based on data as reported by performers because they are in the best position to (1) indicate how much they spent in the actual conduct of R&D
in a given year; (2) classify their work as basic, applied, etc.; and (3) identify the sector of the economy in which their financing originated. For those reasons, and because the consistent use of performer reporting reduces the possibility of
double counting and conforms to international standards (as outlined by the Organisation for Economic Co operation and Development), R&D data are presented on a performer basis whenever possible.
Separate R&D performance totals are reported for (1) the Federal Government, (2) industry, (3) universities and colleges, (4) university administered FFRDCs, and (5) other nonprofit organizations.
R&D performed by State and local government agencies is not included in the national R&D totals. When State and local governments are listed by a survey respondent as the source of non Federal R&D funds, those amounts are included in the source
totals of the sector reporting the R&D performance.
The remainder of this section and the next section, "Estimating Procedure for 1993 and 1994," outline the approaches and assumptions used for deriving the national totals.
The procedures for estimating R&D expenditures by performer and source for the years 1990 through 1994 are further summarized in table A-5. The structure of the table is the same as those of tables B-2, B-5, B-8, and B-11, in which the expenditure totals are detailed.
Federal agency R&D obligations for intramural performance are treated as the equivalent of R&D expenditures in the National Patterns series. As detailed in the Federal Funds for Research and Development series (Federal Funds), such intramural activities cover costs associated with the planning and administration by Federal personnel of intramural and extramural R&D programs as well as actual intramural R&D performance. (See NSF/SRS, Federal Funds for Research and Development: Fiscal Years 1992, 1993, and 1994, Detailed Statistical Tables, NSF 94-328.) In general, the universe of Federal agencies with R&D programs has been surveyed annually since 1953 for their R&D performance and since 1963 for the distribution by character of work. The most recent survey included R&D funding as reported by 103 Federal agencies or their subdivisions.
In addition to Federal obligation data, National Patterns uses data on Federal R&D budget authority and outlays. The use of each series is clearly noted in the text.
IndustryIn general, the industry sector has been surveyed annually since 1953 for its total R&D performance and since 1956 for the distribution by character of work. The U.S. Bureau of the Census conducts the survey for NSF. The target population of the survey is companies, whether U.S. or foreign owned, that perform R&D in the United States. Prior to the 1992 survey with the introduction of annual sampling, a new sample was drawn and canvassed only every 5 or 6 years (for example, in 1976, 1981, and 1987). In the intervening years, a subset of the last sample-called a panel and including all companies reporting more than $1 million in R&D-was surveyed. Accordingly, data for the years in which a sample was not drawn did not include companies that were new entrants in the R&D field, and such data were generally biased in a downward direction. The Census Bureau, however, did estimate the annual changes in R&D data for companies that reported R&D in the sample year but were not included in the panel. As new samples were drawn, revisions to previous years' estimates were issued-a process called "wedging."
For example, a new sample was drawn for 1987, from which R&D data were collected for 1986 as well as 1987. Data from the new sample were used by Census to revise 1986 R&D performance estimates. The
new 1986 data were also used by Census in combination with data from the last sample year, 1981, to revise estimates for the intervening years. NSF subsequently provided a second round of revisions to the 1982-85 R&D series. In both cases the
revisions were done on an industry basis: an effort was made to apply the overall 1981-86 growth rates while preserving the relative year to year movements in each industry's R&D. This approach resulted in major revisions of the 1982-87 industry R&D
time series previously published in National Patterns, especially of the nonfederally funded component of industry's R&D performance. The revised totals, as well as the industry data reported in National Patterns: 1990
(NSF 90 316), are presented in table A-1. (Some of these data have since been revised, but the revisions were unrelated to the drawing of a new sample.)[18] Details on survey methods,
coverage, concepts, definitions, and reliability of the estimates associated with the R&D expenditure data are contained in the 1990 edition (NSF 94-304) of the Research and Development in
Industry series (Industry R&D).
Sample Design Revision for 1992
This National Patterns is the first to include revised R&D data based on a new industry sample drawn for the 1992 survey year. Data for 1991 were also collected. For the 1987 survey, about 14,000 firms were selected for the sample. For years 1988 through 1991, a panel of approximately 1,700 companies were selected from the 1987 sample and were annually resurveyed. These companies accounted for approximately 90 percent of total industry performed R&D. Other companies in the 1987 sample did not receive another questionnaire, but their R&D data were estimated. Beginning with the 1992 survey, NSF will draw new samples annually, and the sample size is increased to 23,000 firms. The new sample better reflects the widening population of R&D performers among firms in the nonmanufacturing industries and small firms in all industries. As a result, extensive revisions were made in the historical industry and national R&D time series.
The revised 1991 industry R&D performance ($117 billion) is 14 percent higher than was previously reported ($102.2 billion). Of the $14.7 billion revision, $13.7 billion resulted from the new
sample and $1.0 billion from normal data revisions for firms sampled in both surveys. Furthermore, $11.4 billion of the $13.7 billion increase stemming from the enlarged sample design was reported for nonmanufacturing industries, including $2.0
billion of R&D in industries not previously included in the sample frame. The revised U.S. 1991 R&D total ($160.2 billion) incorporating the new industry sample is 10 percent higher than the U.S. performance total ($145.4 billion) based on previous
industry R&D estimates. Complete technical details on industry's new survey methodology are contained in the forthcoming report, Research and Development in Industry: 1992.
Like in previous sampling cycles, this National Patterns includes revisions to the industry data for years intervening the 2 sample years, (i.e., for 1988, 1989, and 1990). The
industry and U.S. time series reported here include the wedged data reported for 1988-90 and the revised data for 1991-92. Since these 1991 data were collected at the same time as the 1992 data, the year to year changes reported for 1991-92 are not
affected by changes in the survey sampling design. For almost all of the aggregate statistics (for example, industry R&D by Federal and non Federal sources of funding), NSF believes that time series comparisons (for example, between 1981 and 1992
data) are still reasonable: Surveys undertaken in both years provided the best estimates of the Nation's industrial R&D performance total by sampling those industries then believed to be conducting R&D. However, wedging does not fully correct all
changes in the survey series; comparisons between some data elements for consecutive years may be problematic.
Not only do the 1987 and 1992 surveys' sample size and frame differ considerably (see above), but $9.2 billion of R&D performed by firms reporting in both surveys was shifted from one industry in
1987 to another in 1992 especially from manufacturing industries in 1987 to nonmanufacturing industries in 1992. Such classification shifts can be attributed to (1) product mix changes of individual firms that occurred some time during 1987-92, (2)
changes in the 1987 SIC that were effected in the 1992 survey, and (3) a change in the methodology used by NSF/Census for classifying companies to specific 2 and 3 digit SIC industries. Wedging smoothes the industry totals over time, but does not
pinpoint the actual year of change.
Appendix table A-2 provides summary statistics for wedged data that appear in this National Patterns with those that were previously published. For example,
the 1990 R&D/GDP ratio is 2.64 percent using the original 1990 industry survey data, but this ratio is 2.73 percent based on the use of wedged data. For 1991, the R&D/GDP ratio is 2.80 percent. The wedged R&D time series smoothes the year to year
changes between the 1987 and revised 1991 survey endpoints. Since NSF is now committed to drawing new samples annually, the issue of wedging is unlikely to recur.
As noted in the National Patterns of R&D Resources: 1992, the procedures used by Census for imputing character of work splits for industry's R&D performance were changed for 1986 and later years; hence, these data are not directly comparable with data for 1985 and earlier years. A full description of the various imputation methodologies and alternatives is presented in the 1988 Industry R&D report (NSF 90-319). Briefly, for 1985 and earlier, for companies that did not report character of work splits, Census imputed the splits based on either (1) the company's percentage distribution reported in its most recent year of available data or (2) in the absence of any prior year breakdown for the company, the average character of work split for the industry to which the company was assigned. For years after 1985 Census does not impute a company's character of work distribution unless the company has reported a breakout within 2 years of the year being imputed. When distributions are not imputed, Census assigns the company's R&D to an "undistributed residual" category. (For example, and as detailed in table A-3, industry reported $96.7 billion in 1992 R&D performance with non-Federal funds; of that, Census did not distribute $16.4 billion.)
To provide character of work estimates for the entire population of firms performing R&D in the United States, each industry's (as contrasted with each individual company's) "undistributed
residual" is allocated to basic research, applied research, and development categories using the average character of work splits reported for that industry. This approach resulted in relatively higher performance shares for basic and applied
research than had been previously estimated and relatively lower estimates for development's share of industry's total R&D performance (table A-3). See the forthcoming report, Research and Development in Industry: 1992
for further discussion of this issue.
The academic sector, including all university administered FFRDCs, has been surveyed annually since 1972 for R&D performance. It was surveyed less frequently before 1972. For 1992, data were collected from a sample of the 594 institutions of higher education in the United States and outlying areas that (1) granted a graduate degree in science or engineering and/or (2) performed activities for which at least $50,000 had been funded from separately budgeted R&D expenditures. The 461 institutions sampled comprised all doctorate granting institutions, all historically black colleges and universities with any R&D expenditures, and a random sample of all other institutions. For 1983-87 and 1989-91, R&D performance was estimated from a census of institutions that granted doctorates in science and engineering plus a sample of all other universities and colleges. For 1988, data were collected from the full population of 556 institutions that met the criteria listed above.
With the exception of 1978, data on the basic research performance of universities and colleges and of university administered FFRDCs have been collected annually since 1972. Since 1979, however,
only the combined total for applied research and development performance has been collected. Furthermore, data on the character of work from individual non Federal sources of funds (i.e., industry, institutional funds, State and local governments,
and other sources) are not surveyed. For the years 1978 to the present, the distribution of applied research and development from Federal sources is based largely on data from Federal Funds; the development split from each of the non Federal funding
sources is estimated at 7 percent of the surveyed non Federal R&D funding totals; and the applied research share from each of the non-Federal funding sources is an estimated residual (total R&D funds, minus basic research funds, minus
development funds).
Revised estimates for Federal funding of applied research and development to universities and colleges and to university administered FFRDCs were first included in the National Patterns of R&D
Resources: 1992. University performers report the amount of R&D and basic research that they undertake with Federal funds. The residual is their combined applied research and development performance. The distribution between applied research and
development is estimated from the percentage shares of Federal obligation data to the academic sector as reported by Federal agencies in Federal Funds.
Although the estimating procedures used previously had been loosely based on the data provided by the Federal funding agencies, the approach adopted here formally links the performer and source
reported survey data. Applied research and development expenditures for universities and colleges were revised for the period 1978 present; for university administered FFRDCs, revisions were made back to 1975. The general result is that the applied
research share is slightly lower and the development share somewhat higher than previously reported. For example, of the $9.0 billion in Federal R&D support to universities and colleges in 1989, $2.9 billion was for applied research and development
and $6.1 billion was for basic research activities. In National Patterns: 1990, $2.5 billion (28 percent of all Federal funds to the sector) was estimated as applied research and $0.3 billion (3 percent) as development. In National
Patterns: 1992 and in this report, 1989 Federal applied research funds are estimated at $2.1 billion (24 percent) and development at $0.7 billion (8 percent). The revised totals, as well as the data reported in National Patterns: 1990, are
presented in table A-4.
Only for the academic sector does R&D performance include research funds subcontracted to outside organizations. (For performance reported by respondents in the other surveyed sectors, R&D
subcontracted to other organizations is excluded.) Details on survey methods, coverage, concepts, definitions, and reliability of the estimates associated with the R&D expenditure data are reported in the FY 1992 report (NSF 94-324) in the Academic Science and Engineering: R&D Expenditures (Academic R&D) series.
It has not been possible to maintain the same survey frequency for other nonprofit institutions; the last complete survey was conducted in 1973. Since then, small and informal surveys of this sector have been undertaken periodically, most recently for 1983. For the years 1984 to the present, estimates for federally funded total R&D and character of work performance by nonprofit institutions-including associated FFRDCs-are derived from Federal obligation data reported in Federal Funds. Industry as a source of R&D funds to this sector is estimated from the average of the annual percentage change in (1) industry's funding of industry performed R&D (from Industry R&D) and (2) industry funding of university performed R&D (from Academic R&D). Nonprofit funding as a source of R&D funds to this sector is estimated from the annual percentage change in nonprofit funding of university performed R&D (from Academic R&D). The character of work splits from the non Federal funding sources that were surveyed in 1983 are carried forward to the present.
18. Note that although Census reestimated 1982-86 R&D totals by funding source, it did not provide a character of work distribution for the revised data. After investigating several possible alternatives, NSF
chose to allocate the revisions (table A-1) on the basis of average character of work distribution published in earlier annual Industry R&D reports. Allocations for the federally funded and nonfederally funded R&D revisions
were applied separately.
