BEA has prepared a satellite account that arrays information about research and development (R&D). First, the satellite account provides estimates of expenditures on R&D that are designed to be used in conjunction with the national income and product accounts measures. Second, it treats R&D expenditures as a form of investment, recognizing the role R&D plays in adding to knowledge and in developing new and improved processes and products that lead to increases in productivity and growth. Third, it provides estimates of the stock of knowledge capital. To focus on R&D and facilitate its analysis, the satellite account changes some definitions and classifications used in the national income and product accounts but otherwise is designed to be consistent statistically and conceptually with those accounts. Thus, the satellite account supplements the existing accounts.
BEA's economic accounts have always benefited from discussion and critique of concepts, source data, and estimating methods. The same is to be expected for the R&D satellite account. Comments are welcome.
Carol S. Carson
Director, Bureau of Economic Analysis
In industrial laboratories, agricultural experiment stations, medical research institutes, and a variety of other settings, the United States undertakes a sizable research and development (R&D) effort. This effort plays a critical role in economic growth and in addressing many specific related concerns. In the words of the National Science Board:
The absolute magnitude of the [R&D] effort and the manifold tasks to which it is directed are indicative of the critical role that R&D plays in addressing such concerns as national defense, industrial competitiveness, public health, environmental quality, and social well-being. Indeed, the long-term importance of R&D expenditures to technological preeminence, military security, and knowledge growth is axiomatic./1/
Ideally, to document this role within the economy and thus lay the foundation for policy and other decisions, one would measure the output of R&Dthe new understanding, or the knowledge, it creates. However, measures of knowledge created, to the extent that they exist, do not share a common yardsticksuch as dollarswith other measures with which they might be used. Almost universally then, analysts turn to expenditures on R&D as a starting point.
Several questions about expenditures on R&D immediately come to mind:
Expenditures on R&D can be viewed as generating future income and product. With this view, a case is made for treating them as investment, paralleling the treatment of business expenditures on structures and durable equipment, and for recognizing a stock of intangible capital, just as there is a stock of tangible capital. Further questions then arise:
Answers to these questions have been less than fully satisfactory. On the one hand, the national income and product accounts (NIPA's) might seem the obvious place to look for expenditure estimates: If R&D expenditure estimates were in the accounts, they could readily be compared with GDP or its components, and models could be constructed to relate changes in R&D to other parts of the economy represented in the accounts. However, only a portion of R&D expenditures are identifiable within the NIPA's, and those identifiedas well as the unidentifiedexpenditures on R&D are treated as consumption rather than as investment. Further, because R&D expenditures are not treated as investment, there is no associated stock of capital./2/ On the other hand, R&D data from other sources are not fully consistent with the NIPA's and with the NIPA-based measures of tangible capital, so they cannot readily be used in conjunction with NIPA estimates in analysis.
This article introduces a satellite account that is designed to provide a view of R&D that has ties to the NIPA's, while also using alternative definitions and providing consistent detail that help to focus on the role of R&D in the economy. BEA began work on the satellite account for R&D in 1992, following a preliminary evaluation of the feasibility and usefulness of such an account./3/
The estimates presented in the satellite account build on data published by the National Science Foundation (NSF), which assembles a wide range of information related to R&D./4/ The estimates extend through 1992, the most recent year for which complete source data are currently available.
The first section of the article defines R&D and describes its role in creating knowledge and then sketches the economic accounting background for the satellite account's investment treatment. The second section provides a methodological overview. The third section presents the estimates of R&D expenditures, investment, capital stocks, and related data. The fourth section discusses future directions that work on the satellite account might follow. A technical note at the end of the article details the construction of the estimates.
R&D is "creative work undertaken on a systematic basis in order to increase the stock of knowledge, including knowledge of man, culture and society, and the use of this stock of knowledge to devise new applications." This definition is from a newly revised manual (the Frascati Manual) of standard practice for surveys of R&D activity, prepared by the Organisation for Economic Co-operation and Development./5/ It is widely used internationally as the basis for R&D statistics, such as those compiled and analyzed by NSF in the United States.
More commonly, R&D is characterized as the sum of three types of activitiesbasic research, applied research, and development. These activities also have been defined in the Frascati Manual, although in practice it is often difficultperhaps increasingly soto establish the boundaries between them:
R&D is to be distinguished from a wide range of related activities that are linked to R&D both through flows of information and in terms of operations, institutions, and personnel. The basic criterion, according to the Frascati Manual, to be used to distinguish R&D from related activities "is the presence in R&D of an appreciable element of novelty and the resolution of scientific and/or technological uncertainty." R&D does not include, for example, the following: Routine activities (such as product testing, quality control, experimental production, routine software development, and monitoring and evaluation of operational programs), patent and license work, final product or design engineering and manufacturing start-up, and training of scientific and technical personnel.
R&D may be viewed as increasing the stock of knowledge that leads to improved understanding or to improved processes or products. Basic research creates a pool of knowledge that can be drawn upon for further basic research or for performing applied research./7/ Applied research draws upon both basic research and earlier applied research to create knowledge that can be used to develop new or improved processes and products. Development draws upon both applied research and earlier development. New or improved processes or products come into being only at the end of the development process. There are lags between the creation of knowledge, particularly that produced by basic research, and its effects on output. The lags reflect both the time needed for R&D to lead to improved processes and products and the time needed for the improved processes and products to be fully adopted throughout the economy.
Neither the creation of knowledge nor the resulting stock of knowledge are measurable directly. Measures of output, such as the number of scientific and technical journal articles published and the number of patents awarded, only roughly approximate the creation of knowledge because they only cover a portion of R&D and because many innovations are not patented. A frequently used method for measuring the output of knowledge is to equate that output with the expenditures required to produce it. Those expenditures can be cumulated over timewith or without depreciationto measure the stock of knowledge.
R&D in standard economic accounts.In accounting for a nation's production by adding up expenditures to derive gross domestic product (GDP), two main issues about the treatment of R&D arise:
In the NIPA's, expenditures on R&D by businesswhether actually purchased from others or carried out inhouseare treated as intermediate rather than final; they are considered as a current expense of production and are not among the expenditures added up in deriving GDP. Treating them as a current expense follows general business accounting practice; the uncertainty about the future benefits of individual projects is a key argument for expensing R&D. Expenditures on R&D by government and by nonprofit institutions are treated as expenditures on final goods and services. All expenditures on R&D by government and nonprofit institutions are treated as part of consumption in the current period, the former as part of government purchases and the latter as part of personal consumption expenditures; none are treated as investment. In the NIPA's, investmentspecifically gross private domestic investmentconsists solely of purchases of structures, durable equipment, and change in inventories by the business sector. Expenditures by a U.S. resident for R&D performed abroad are treated as imports, and expenditures by a foreign resident for R&D performed in the United States are treated as exports. These points are summarized in table A, which also indicates that, with the exception of contractual R&D in Federal national defense purchases, R&D expenditures are not identified in GDP.
The issue of the scope of investment in the economic accounts, and in particular the issue of including R&D in investment, is a longstanding one. John Kendrick, in 1951, identified activities related to improvements in technology and technical innovations as leading to improved productivity; he noted that technological innovations depend on advances in knowledge, and he focused on research as a source of these advances./8/ He viewed research expenditureswhether directed toward improving structures and equipment, raising the level of health, or dealing with problems of land and natural resource useas expenditures devoted to increasing productivity. Accordingly, he proposed that they be treated as investment in the economic accounts. He noted that gross product would be higher by the amount of expenditure by business on R&D, which would be counted as final product rather than expensed; because expenditures by nonprofit institutions and government are already counted as final product, a change to treat them as investment would not change gross product.
The issue was raised again a few years later at a conference that led to the volume A Critique of the United States Income and Product Accounts./9/ The case was made that these expenditures "pay" in terms of yielding future returns and thus fit the general characterization of investment. Although various conceptual and statistical difficulties were identified as obstacles, there were some prospects for better statistics.
The 1968 revision of the System of National Accounts, published by the United Nations as international guidelines for economic accounting, referred to the urgent need to clarify the question of R&D expenditures in dealing with the boundary between current and capital expenditures. It was noted that the clarification could come only on the basis of experience.
R&D in extended economic accounts.A number of analysts, working with the U.S. economic accounts, have proposed systems that expand the boundaries of investment by including R&D and several other categories of expenditures.
Nancy Ruggles and Richard Ruggles, in 1970, proposed a category of "development" outlays defined as those that meet the criterion that the value of the services provided by the outlay must accrue in future periods rather than entirely in the present period./10/ Outlays on education and training and some outlays on health, as well as outlays on R&D, were viewed as meeting the criterion. Their proposed system included stocks of "development" capital. They valued the services of the R&D portion of the capital stock as the amortization adjusted to market prices plus an imputed interest charge on the capital stock.
John Kendrick implemented his view that expenditures on certain intangibles are "made primarily to improve the quality or productivity of the tangible... factors in which they are embodied" and should be treated as investment that creates intangible capital./11/ The intangible investment and capital included R&D, education and training, health and safety, and labor mobility. He created gross stock by cumulating investments over their lifetimes and net stocks by cumulating depreciation on each vintage of investment and subtracting it from the gross stock. He estimated rental values of the capital stocks (referred to by other authors as service values or returns) for the nonbusiness sectors and added them to income and product. His effort, published in 1976, was viewed as a pilot study for determining the feasibility and usefulness of developing estimates of total investment and capital stocks.
Robert Eisner used "include investment in intangible and human capital" as one of the guiding principles for his total incomes system of accounts, published in 1989./12/ His interest in investment stemmed in major part from its relation to productivity and growth. His intangible capital comprised R&D, education and training, and health. His methodology for measuring R&D investment and capitalizing it was essentially the same as Kendrick's.
Satellite accounts.Meanwhile, the possibility of treating R&D and several other activities as investment was discussed at length in preparing for the System of National Accounts 1993./13/ At the outset of the discussion, there was strong support for treating at least some portion of R&D expenditures as investment. Several proposals were made to identify a portion that was most clearly linked to a future returnfor example, the portion of development expenditures in which the expenditures are identifiable and the outcome reasonably certain enough to assure that the costs of the project would be exceeded by the revenue.
In the end, however, no change was made in the treatment of R&D. The explanation of the treatment of business expenditures on R&D noted that they are undertaken to improve efficiency or to derive other future benefits and so are inherently investment-type activities. However, practical difficulties in meeting the accounting requirements for treating R&D and similar activities as investment suggested that they be treated as intermediate:
In order to classify such activities as investment type it would be necessary to have clear criteria for delineating them from other activities, to be able to identify and classify the assets produced, to be able to value such assets in an economically meaningful way and to know the rate at which they depreciate over time. In practice it is difficult to meet all these requirements. By convention, therefore, all the outputs produced by research and development, staff training, market research and similar activities are treated as being consumed as intermediate inputs even though some of them may bring future benefits./14/
Nevertheless, there was strong interest in being able to identify R&D within the economic accounts, and work toward classification systems that would help do so was encouraged. In addition, R&D was recognized as a prime candidate for presentation in a satellite account, an economic accounting tool that achieved international recognition when it was incorporated in the System of National Accounts 1993.
In brief, satellite accounts are frameworks designed to expand the analytical capacity of the economic accounts without overburdening them with detail or interfering with their general-purpose orientation. Satellite accounts, which are meant to supplement, rather than replace, the existing accounts, organize information in an internally consistent way that suits the particular analytical focus at hand, while maintaining links to the existing accounts. In their most flexible application, they may use definitions and classifications that differ from those in the existing accounts; for example, the R&D satellite account uses a different definition of investment, and it classifies transactors into different groupings. In addition, satellite accounts typically add detail or other information, including nonmonetary information, about a particular aspect of the economy to that in the existing accounts; for example, the R&D satellite account includes information about R&D employment.
The advantages of using R&D information assembled along the lines of the Frascati Manual to prepare a satellite account have become increasingly clear. One of the first satellite accounts, prepared in France in the 1970's, built on such R&D information./15/ More recently, a framework for an R&D satellite account for the Netherlands was constructed to use such information./16/ The new Frascati Manual specifically recognizes the connection between the data it describes and economic accounting, and it includes an annex that explains satellite accounts to experts on science and technology who are not familiar with economic accounting.
The R&D satellite account focuses on the value of R&D produced in the United States and the use of that output as investment. Because no direct measure of output is available, R&D produced is measured by summing the costs of its production, a technique of measurement used in economic accounting for most nonmarket production. The resulting total is referred to as R&D expenditures. The expenditure estimates were prepared by starting with the information available from surveys conducted for NSF and adjusting it to statistical and conceptual consistency with the NIPA's. The decision to work with information that is not extensively used to prepare the NIPA estimates was made because the regular source data and estimating methods do not permit the required level of resolution needed to focus on R&D./17/
The satellite account groups organizations in a way that reflects the features of their institutional structures and purposes that are relevant to R&D. In light of the interest in academic R&D, universities and colleges (along with their affiliated institutions, agricultural experiment stations, and associated schools of agriculture) need to be shown separately. Federally funded research and development centers (FFRDC's), which are R&D organizations financed almost entirely by the Federal Government, are shown separately and grouped with the several kinds of entities that administer them. (At present, there are 39 FFRDC's, including the RAND Corporation, Argonne National Laboratory, E.O. Lawrence Livermore Laboratory, and Brookhaven National Laboratory.) The satellite account shows two major groups: "Private" organizations and "government." "Private" organizations consist of business (labeled "industry"); private universities and colleges, private hospitals, charitable foundations, and other nonprofit institutions serving households; and most FFRDC's. "Government" consists of the Federal Government, State and local governments (excluding universities and colleges), public universities and colleges, and FFRDC's administered by State and local government organizations, primarily public universities and colleges.
Constant-dollar R&D expenditures are derived by deflation, the method most often used in the NIPA's. In deflation, constant-dollar estimates are obtained by dividing the most detailed current-dollar components by appropriate price indexes. In the case of R&D, the current-dollar components are its costs of production. The expenditure estimates are treated as investment and cumulated to yield R&D capital stocks using methodologies developed by BEA to estimate fixed reproducible tangible capital stocks.
The most important of the methodological issues encountered in preparing the satellite account are described in this section.
The measure of expendituresreflecting labor costs, the costs of materials and supplies, and overhead costs (including a charge for the capital used in producing R&D)is based on data by performer, when available, from NSF surveys. Only the data by performer provide the cost components needed to construct constant dollars.
BEA supplements the coverage of the survey-based data and extends it back in time. Missing data, primarily for State and local government R&D and the R&D of some types of nonprofit institutions, are interpolated and extrapolated from years for which data are available. Estimates for years prior to the first NSF survey in 1953 (which are needed to estimate stocks and related measures but are not presented in this article) are primarily based on outside studies that estimated R&D for selected years./18/ Some supporting datain particular, for pre-1953 FFRDC'sare estimated by BEA using various sources.
A number of adjustments are made to the NSF survey-based spending data to make them statistically and conceptually consistent with the NIPA's. The statistical adjustments are for timing and geographic coverage and to fill gaps with estimates for some industries in some years. A conceptual adjustment is made to put depreciation of structures and equipment used in producing R&D on a basis that reflects the valuation and consistency appropriate for economic accounts.
BEA has implemented three disaggregations of R&D expenditures for analytical use in the satellite account: By performer, with industry detail; by source of funding; and by type. R&D by performer serves the same purposes for R&D as breakdowns by sector or industry of origin in analyses of production, which are often a first step in studies of structural change. R&D by funder is useful because a substantial portion of R&D is not financed by the performer. R&D by type is useful because the different types interact with the economy in different ways and with different lags. Other disaggregations would also be useful, but are not practical given current resources. For example, R&D disaggregated by purpose, such as defense or health, would help relate R&D expenditures to other issues. Geographic breakdowns would also be of interestfor example, in location decisions, for which proximity to research resources may be a factor.
The R&D satellite account provides estimates of constant-dollar expenditures by performer. In the absence of outputs and output price measures, costs of inputs are deflated by weighted indexes of input prices. The costs of inputs are derived, at the finest level of detail possible, from the limited cost information available from NSF surveys. The cost components are matched as closely as possible with "proxy" prices. The individual constant-dollar cost components are summed to derive constant-dollar expenditures by performer. (Implicit price deflators for R&D by performer are a byproduct of the procedure.)
Constant-dollar estimates derived in this way take into account the changing mix of R&D performers over time. The estimates of constant-dollar compensation of employees, which overall is about 45 percent of inputs, reflect changes in labor productivity only to the extent that the price indexes used reflect a procedure that picks up changes in the mix of employee skills. For example, the indexes that include Federal employee compensation reflect changes in the skill mix estimated by taking into account changes over time in the level of experience and education. Consistent with NIPA practice, the estimates do not include any additional, specific adjustment, such as an assumed rate of increase in labor productivity based on observations in related fields.
Ideally, the same breakdowns available for current-dollar expenditures would be prepared in constant dollarsthat is, by funder and by type as well as by performer. However, because most performers have multiple sources of funding and because all groups of performers do at least some of each type of R&D, more detail on cost components is necessary to deflate R&D by funder or type. It is possible that constant-dollar estimates by funder could be derived by allocating cost components in the cases for which performer and funder do not coincide. Deriving constant-dollar estimates by type of R&D will be more problematic.
It is generally agreed that stocks of intangible capital, such as R&D, are best obtained by cumulating investment flows rather than surveying stocks directly. BEA's review of the methods available led to reliance on the following three elements: (1) The performer breakdown currently available for constant-dollar R&D investment; (2) BEA's current methodology for fixed reproducible tangible capital stocks; and (3) uniform service lives for all R&D capital. The resulting R&D capital stock estimates are the first ones that are fully consistent with BEA's estimates of tangible capital.
Scope of R&D capital.Some researchers have questioned whether expenditures on all types of R&D and in all fields should be treated as capital formation. Some have excluded basic research because they view it as being undertaken for the purpose of improving understanding of the world and not for the purpose of increasing productivity or adding to production./19/ Other researchers have excluded specific fields of researchfor example, defense or space R&Dbecause they view those fields as having little applicability to commercial production./20/ Alternatively, if R&D is regarded solely as an input to the production process, comparable to the blueprints for a new building, only businesses' development expenditures for commercial applications might be included. Some researchers, particularly those who have constructed a broadened view of investment and wealth, have included all R&D./21/ Including all types of R&D in all fields, as the satellite account does, is consistent with a view of R&D as a new kind of wealth. Ideally, an R&D satellite account would publish R&D capital stocks showing detailed information that would allow users of the account to decide which categories of R&D to include or exclude, depending on their use of the estimates, but the satellite account does not yet do so.
Others have questioned whether all R&D, both successful and unsuccessful, should be treated as capital formation./22/ The R&D satellite account is consistent in this respect with the existing NIPA treatment of mineral exploration expenditures, which are all treated as investment in line with the view that returns from the successes are sufficient overall to pay for the failures.
Allocation of R&D capital and consumption of fixed capital.In doing analytical work on R&D, some researchers have allocated most R&D capital financed by government and by nonprofit institutions to the business sector./23/ The Congressional Budget Office allocated all R&D to the economy at large./24/ In a more general setting, most presentations of capital stocks, including BEA's fixed reproducible tangible wealth, are on an ownership basis, allocating stocks to the sectors that own them.
For R&D capital in the satellite account, an allocation by funder would be closer to an ownership basis than an allocation by performer. However, the constant-dollar estimates of R&D expenditures by funder needed to prepare the capital stocks by funder will require additional work (see the section "Future Directions"). The R&D satellite account thus allocates the total R&D capital stock on the basis of performer to private and government components. Similarly, the consumption of fixed capital is allocated on the basis of performer.
Timing.R&D projects typically take more than a year from the time they are started until their results are embedded in new knowledge or in new processes or products. Researchers have identified two types of lags: Gestation lags and applications lags. Gestation lags refer to the time needed to complete an R&D project. Applications lags refer to the time between completion of the R&D and its initial commercial use. The sum of the two lags yields the time needed for R&D investments to increase the stocks of knowledge that are actually being used. Survey-based research has found that gestation lags range from 1 to 2 years and that applications lags range from somewhat less than 1 year to somewhat more than 2 years./25/ Researchers have also studied the total lag between R&D and its peak effects on productivity or profits./26/ They have generally found long lags, particularly for basic research, because most new products and processes that result from R&D are adopted only gradually.
The R&D satellite account only needs to take into account the gestation lag, which is assumed to be 1 year. However, because the U.S. economic accounts measure production at the time that capital and labor are used in the production process, the gestation lag means that another category of outputR&D inventoriesmust be introduced. These inventories are the equivalent of work-in-progress for some tangible fixed capital goods whose production requires more than one time period. R&D inventories are converted to stocks of R&D intangible fixed capital at the end of the gestation lag.
Depreciation patterns and rates.Some researchers have treated some, or all, capital created by R&D as immortalthat is, as a permanent part of the capital stock once it is added./27/ Other researchers have assumed that once R&D capital has entered the capital stock, it is gradually removed by depreciationor, more formally, in economic accounting terms, by consumption of fixed capital. They used a variety of patterns and rates of depreciation./28/ In the satellite account, R&D is assumed to depreciate over a finite lifespan. The depreciation is due to obsolescence as knowledge from newer R&D supplants the knowledge from older R&D, or as applied R&D produces newer processes and products that supplant older ones. (For those who wish to treat R&D capital as immortal, the satellite account includes supplemental series that show cumulative R&D expenditures since 1929.)
The choice of a depreciation pattern for R&D stocks is of necessity somewhat arbitrary. There are no R&D capital markets to provide information on the value of "used" R&D. A study of patent renewal rates in several European countries was inconclusive; its findings could support assumptions about the pattern of depreciation ranging from geometric to slower-then-faster-than-straight-line./29/
In the R&D satellite account, the stock of R&D capital is constructed using the same methodology that is currently used to construct BEA's estimates of fixed reproducible tangible capital: The perpetual inventory method is used with uniform average service lives, straight-line depreciation, and a bell-shaped distribution within each vintage of capital to determine discards. The current-dollar stock of R&D is measured at replacement cost rather than at historical cost./30/
Empirical estimates have been made using geometric depreciation patterns. Among more recent studies, Ariel Pakes and Mark Schankerman found rates of 0.11 to 0.12 per year in some countries, but they reported estimates of 0.17 to 0.26 in the United Kingdom./31/ James Adams estimated depreciation rates of 0.09 to 0.13 for basic research./32/ M. Ishaq Nadiri and Ingmar Prucha estimated a rate of 0.12 for industrial R&D./33/
The R&D satellite account uses the straight-line lifespan that corresponds most closely to a geometric depreciation of 0.11 per year, a rate chosen because it is near the center of a plausible range of rates. This straight-line average service life is 18 years. A study that compared R&D net capital stocks estimated using an 18-year average service life with alternative estimates made using geometric depreciation and a rate of depreciation of 0.11 per year revealed only modest differences that exhibited no particular time trends.
Table B summarizes some results from the R&D satellite account. It shows the following:
The tables that make up the R&D satellite account are in five groups. The tables numbered 1 are summary tables that present expenditures and investment for 195392, and stocks for 195992, in current dollars (or at current cost) and in constant dollars (or at constant cost)tables 1.1 and 1.2, respectively. The tables numbered 2 present expenditures in current dollars by performer, by funder, and by type. The tables numbered 3 present estimates by industry in current dollars. The tables numbered 4 present constant-dollar expenditures, including expenditures by performing industry along with the number of scientists and engineers by industry. The tables numbered 5 present implicit price deflators.
Table 2.1 shows R&D expenditures by performer, and within each performer, by source of funds. Chart 1, which is based on this table, shows shares of R&D expenditures by performer for 1992. Industry, with expenditures of $119.5 billion, was the largest performer of R&D, accounting for 71 percent of total R&D expenditures. It has maintained at least a two-thirds share for most of the last 40 years. Public and private universities and colleges combined, with $18.9 billion, were the second largest performers of R&D. The Federal Government followed with $16.3 billion. Expenditures on R&D activities performed within the Federal Government, which had a share of 23 percent in 1953, had a 10-percent share in 1992. The combined expenditures for R&D performed by State and local governments, nonprofit institutions, and FFRDC'sat $14.6 billionaccounted for the remaining share of 9 percent.
For 1992, the Federal Government, in addition to funding all the R&D it performs, funded a majority of the R&D performed by universities and colleges (59 percent), FFRDC's (99 percent), and other nonprofit institutions (57 percent). However, the share of R&D funded by the Federal Government has declined steadily over time. Industry and State and local governments fund most of their own R&D work, 81 percent and 71 percent, respectively.
Table 2.2 shows R&D expenditures by source of funds, and within each source, by performer. Five sources of funds are shown in the R&D satellite account: Industry, the Federal Government, State and local governments, private universities and colleges, and "other." Because of data limitations, some small flows are combined with the major sources of funding; for example, industry's funding of R&D performed by industry includes funds from the rest of the world, because this funding source cannot be separately identified.
Chart 1 also shows shares of R&D expenditures by source of funds for 1992. Industry, providing $99.4 billion, is the largest source of R&D funds, with a 59-percent share. The Federal Government, providing $61.2 billion, is the second largest source. Together, these two sources provided 95 percent, or $160.8 billion, of the total spent on R&D in 1992. Over the 40-year period covered by the satellite account, these two sources of funds have always accounted for most R&D expenditures, but the shares of the two have changed over time. The Federal Government's share reached a high of 67 percent in 1964 and fell to 36 percent in 1992. By contrast, the industry share of R&D funding has steadily increased over time.
Tables 2.3 and 2.4 show R&D expenditures as the sum of expenditures on the three types of R&Dbasic research, applied research, and development. Chart 1 also shows shares of R&D expenditures by type for 1992. Development, at $99.8 billion, is over half the total (59 percent). Applied research, with $40.2 billion, or a 24-percent share, is less than half as large. Basic research, at $29.3 billion, is 17 percent of the total. The share of basic research has increased steadily: By 1992, it had almost doubled its 1960 share. The increases in the share of basic research were offset by declines in development. The share of applied research has remained steady.
Basic research is increasingly being performed at universities and colleges: In 1953, universities and colleges performed less than 30 percent of basic research; by 1992, they performed over 40 percent of it. Most applied R&D is performed by industry. Industry currently performs over two-thirds of applied research and well over 80 percent of development. The Federal Government performs about 10 percent of applied R&D.
Table 3.1 shows expenditures by major R&D performing industries./34/ Manufacturing industries are shown at the two-digit standard industrial classification, except that transportation equipment is split into "aircraft and missiles" and "other transportation equipment" because the share of federally funded R&D in the former is so large. All nonmanufacturing R&D expenditures are combined.
Expenditures on R&D performed by industry were $122.3 billion in 1992. Until 1992, the aircraft and missile industry consistently had the largest R&D expenditures, with a peak of $24.6 billion in 1987. In 1992, chemicals and allied products took top ranking, with expenditures of $16.8 billion. These two industries were followed by industrial machinery, electronic and other electrical equipment, and other transportation equipment. The nonmanufacturing industries, continuing a sharp uptrend, had R&D expenditures of $30.4 billion. (Nonmanufacturing industries include communication services; computer programming, data processing, other computer-related engineering, architectural, and surveying services; and research, development, and testing laboratories.)
The federally funded share of industry R&D has been steadily declining over the last 30 years, from a high of 59 percent in 1959 to a low of 20 percent in 1992. The Federal share has shrunk rapidly since the mid-1980's, particularly in the aircraft and missiles industry. Federal funding accounted for 61 percent of funding in this industry in 1992, down from 76 percent in 1987. When Federal funding is removed, the aircraft and missile industry drops from second place to sixth place ($6.3 billion). Chemicals and allied products led non-federally funded manufacturing R&D performance with expenditures of $16.5 billion. Nonmanufacturing industries, continuing a sharp uptrend, reached $24.4 billion in non-federally funded R&D.
Table 3.2 shows company-funded R&D expenditures performed outside the United States by U.S. companies and their foreign subsidiaries. The chemicals and allied products industry is the leader, with expenditures of $2.7 billion, in 1992. This industry accounted for 28 percent of the 1992 industry-funded expenditures of foreign subsidiaries.
Table 4.1 shows R&D expenditures by performer in constant dollars. Total R&D expenditures grew at an average annual rate of 7.3 percent in 195368. (Over much of the period, Federal spending on defense-related and space R&D increased sharply.) Expenditures then leveled off for nearly a decade, with an average decline of 1.3 percent in 196875. An uptrend then resumed, but at a more moderate rate; the average annual rate of increase in 197592 was 4.0 percent. (During this period, R&D spent on energy, following the 1973 oil embargo, and on health stepped up.) Chart 2 presents expenditures by performer group. Expenditure patterns have been similar for each of the groups performing R&D. Expenditures increased steadily until the late 1960's, leveled off or declined somewhat for a decade, and, except for Federal performance, then increased through 1992. Federal performance has been flat since the late 1980's.
Table 4.2 shows constant-dollar expenditures by industry (including FFRDC's administered by industry). The recent growth in R&D expenditures for industry has been uneven. From 1987 to 1992, real R&D performed by industry increased at an average annual rate of 2.1 percent; however, 7 of the 14 industry groups had declining R&D expenditures during this period. The fastest rates of decline were in stone, clay, and glass products and in aircraft and missiles. The fastest growth has been in nonmanufacturing, which tripled its R&D expenditures between 1987 and 1992.
Table 4.3 shows the number of R&D scientists and engineers by industry, an additional input series. Like the constant-dollar expenditure series, it abstracts from price change over time, but it is narrower in coverage than expenditures. From 1987 to 1992, the picture of uneven growth across industry groups seen in the constant-dollar estimates is also seen in the number of scientists and engineers. Again, the fastest growth is in nonmanufacturing, but the number of scientists and engineers doubled rather than tripling as the constant-dollar expenditures did.
Tables 5.1 and 5.2 show implicit price deflators (IPD's) for each of the performers. IPD's are constructed to derive constant-dollar estimates; the overall IPD for R&D is a byproduct of the constant-dollar estimates./35/ The IPD's of most performers do not greatly differ from the IPD for total R&D, which grew at an average annual rate of 3.7 percent during 198792. Notable exceptions are universities and colleges and FFRDC's. During 198792, the IPD's for private and public universities and colleges grew at average annual rates of 5.9 percent and 5.8 percent, respectively. On the other end of the scale, the IPD's for FFRDC's administered by nonprofit institutions and by governments grew at average annual rates of 2.5 percent and 2.6 percent, respectively.
Investment, consumption of R&D capital, and gross and net stocks are shown in tables 1.1 and 1.2 in current dollars and in constant dollars, respectively. Stocks and consumption of R&D capital are not shown prior to 1959, because the perpetual inventory method for deriving R&D net fixed intangible capital stock would require additional years of constant-dollar investment data, which are not available.
In 1992, constant-dollar fixed tangible investment was $723 billion; the constant-cost net stock of fixed reproducible tangible capital of government and business (including nonprofit institutions) was $12,020 billion. Adding constant-dollar R&D fixed intangible investment would increase fixed investment by 19 percent, or $140 billion; adding the resulting R&D stock would increase the constant-cost net stock of fixed reproducible capital by almost 9 percent, or $1,049 billion.
Chart 3 compares the constant-cost net stock of R&D fixed capital with the constant-cost net stock of fixed reproducible tangible capital of government and business. The chart shows that while the constant-cost fixed reproducible tangible capital stock grew rather steadily from 1959 to 1992 (left scale), the stock of R&D capital grew rapidly until 1970, slowed sharply from 1970 to 1981, and then grew somewhat more rapidly thereafter (right scale). The ratio of the R&D stock to the fixed reproducible tangible stock increased sharply until 1970, fell until 1981, and has increased thereafter.
The average age of the R&D gross fixed intangible capital stock, a rough indicator of the age of the knowledge in the stock, is a byproduct of the perpetual inventory method. As shown in the addenda to table 1.2, the average age of the total constant-cost gross R&D stock increased from about 6.5 years in 1960 to a high of 8.9 years in 1980, then deceased to 8.2 years in 1992. The age of private stock, which makes up about three-quarters of the total stock, showed a very similar pattern. The age of the government stock started higher, at almost 8 years, decreased about half a year over the decade to 1970, and then increased to 9 years at the end of the 1980's; it stood at 8.8 years in 1992.
The R&D satellite account now presents basic information about R&Dthe value of its production by performer, by funder, and by type in current dollars and by performer in constant dollarsand treats the expenditures that measure that production as investment to obtain a stock of R&D fixed intangible capital. Future work could proceed in several directions: Rounding out the view of R&D within an economic accounting framework, refining the existing estimates and providing additional information, and enhancing the international comparability of the satellite account presentation of R&D.
Rounding out the economic accounting view of R&D.The satellite account presents the total value of R&D produced in the United States by adjusting the best available source of information about R&D to prepare estimates consistent with the NIPA's. The restructured GDP that is implied is shown as follows:
Personal consumption expenditures
Gross domestic investment
Net exports of goods and services
Nonprofit institution and government expenditures on R&D are subtracted from personal consumption expenditures and government purchases, respectively. Those expenditures are added to business expenditures on R&D to obtain total R&D investment, split as described earlier between fixed investment and change in inventories. (The term "tangible" is introduced to distinguish the investment in the existing accounts from R&D.) Consumption of fixed R&D capital is allocated to nonprofit institutions and to government; the consumption of fixed R&D capital allocated to business does not appear because it is intermediate consumption. GDP is increased by the addition of business expenditure on R&D (which had been intermediate consumption) and the addition of the difference between expenditures on R&D and the consumption of fixed R&D capital that is included in personal consumption expenditures and in government purchases.
This view of the production of R&D is useful, but could be more fully rounded out within an economic accounting framework. To do so within the national income and product (NIP) accountthe first of the five accounts in the NIPA summary set of accountswould call for identifying components on the product and income sides of the NIP account that include subcomponents that relate to R&D. For example, tangible investment includes investment that provides the capital used to produce R&D, and identifying that investment within the total may be useful.
Rounding out the treatment in the NIP account would lead to changes in the other four accounts of the NIPA summary set. In particular, the gross saving and investment account would reflect changes in the coverage of investment, consumption of fixed capital, and sector saving.
Further, the investment allocated to government and nonprofit institutions might be expected to have a net return (over and above costs, such as consumption of fixed capital) that would appear both on the product (or expenditure) side of the account and on the income side. At present, the NIPA's do not include such a return for nonprofit institutions' investment, and they do not treat any government expenditure as investment (so that there is no return to be considered). However, BEA, as part of an ongoing modernization of the accounts, is considering treating government expenditures on structures and durable equipment as investment, and a major issue is whether a net return on the capital created should be estimated. Especially if a net return is calculated for government fixed reproducible tangible capital, the R&D satellite account should be brought into line to have a consistent treatment for all government capital.
These points are raised within the context of the existing NIPA's and the associated capital stock estimates. However, BEA is reviewing the appropriateness of the concepts and methods underlying its capital stock estimates, and future estimates may be based on somewhat different concepts and measurement methods./36/ In keeping with the approach of constructing the R&D satellite account measures to be consistent with the more general measures with which they might be used and compared, the satellite account measures could change also. Further, as the U.S. accounts are modernized along the lines of the international guidelines in the System of National Accounts 1993, further modifications might be made.
Refining the estimates.Refining the estimates and providing additional breakdowns would strengthen the satellite account. R&D expenditures funded by the Federal Government by agency would provide a proxy for a partial breakdown of R&D by purpose. This breakdown also could lead to a defense-nondefense split; the defense portion would include both purchases of contractual R&D and R&D performed inhouse. Such additional detail could be expected to be of wide interest and also improve the NIPA estimates.
Sources of data on R&D other than NSF have the potential of improving the accuracy of the estimates and of facilitating the provision of additional information. One possibility is that other sources of R&D data could be used to supplement the information from NSF's surveys. These sources could be useful, for example, in developing the analytically interesting but difficult breakdowns by purpose or region. For example, health-related R&D, by all performers, might be separately identified. Another possibility is that BEA's survey data on direct investmentforeign direct investment in the United States and U.S. direct investment abroadand on international trade in services could be further mined for information about international R&D expenditure flows. Because the knowledge gained from R&D is highly mobile internationally, and because there is some evidence that the internationalization of R&D is intensifying, there is substantial interest in tracing the flows.
Further work to refine the deflation of R&D could yield constant-dollar estimates by source of funding and by type of R&D. These constant-dollar estimates would yield, in turn, R&D capital stocks that would allow users to examine whether R&D capital from different sources of funding have different effects. For example, this refinement would allow further examination of the finding by some analysts that government-funded R&D has different effects on productivity than other R&D. Similarly, stocks of R&D capital by type would allow the examination of whether the different types have effects on productivity with different time patterns.
International comparability.Because of the substantial interest in comparing R&D across countries, several strands of work in the international area could contribute to, and benefit from, the U.S. R&D satellite account. First, further work on R&D and other forms of intangible capital formation and capital stock is on the research agenda that emerged from the preparation of the System of National Accounts 1993. This work might lead to some useful standardization on aspects for which empirical work is not likely to lead to firm answersfor example, on the issue of longevity of basic research capital. Second, several international classifications that identify purpose, or function, are to be completed or updated in the future. One of the specific goals of the work is to include R&Dfor example, in the classification of functions for government and for nonprofit institutions. The classification work is likely to draw upon the Frascati Manual. Third, as noted earlier, several countries prepare or are exploring the preparation of R&D satellite accounts.
In looking to the future of BEA's R&D satellite account, it is especially fitting to note that satellite accounts are sometimes called economic accountants' laboratories. The work in these laboratoriesboth in the United States, reflecting comments from users and BEA's experience, and abroadcan be expected to add to knowledge. This knowledge, combined with resources, would help set a course for future improvements.
This note provides additional information about the construction of the R&D satellite account. It covers the sources of data, estimating methods, and assumptions used to construct the three major segments of the account: Current-dollar expenditures, constant-dollar expenditures, and current-cost and constant-cost gross and net capital stocks. Because the R&D satellite account is designed to supplement the U.S. economic accounts, the methods used to estimate R&D flows and stocks are consistent with those used to construct the U.S. national income and product accounts (NIPA's) and the associated estimates of capital stocks. As it does with other estimates, BEA has modified available source data to tailor them to the statistical and conceptual requirements of the account.
The estimates of R&expenditures are largely based on, or are extensions of, data that began in 1953 from four annual surveys published by the National Science Foundation (NSF): Federal Funds for Research and Development, Federal Support to Universities, Colleges, and Selected Nonprofit Institutions, Academic Science and Engineering: R&D Expenditures, and Research and Development in Industry./37/ The two Federal surveys are universe surveys, the academic survey is close to a universe survey, and the industry survey uses a sample that is redrawn every 5 years./38/ The surveys are intended to cover all formal R&D activities, not just the activities assigned to separate R&D units.
The first survey measures obligations and outlays by Federal agency, and the second survey measures obligations by Federal agency. The last two surveys measure expenditures. These survey measures differ with respect to the kind of information they collect about the tangible capital used in performing R&D. For example, the two Federal surveys include a separate measure of tangible capital spending, but exclude its depreciation. The academic and industry surveys do not include a separate measure of capital spending, but include its depreciation as an unidentifiable part of overhead costs. All of these measures are broken down by type of R&D, and each survey includes at least some geographic detail. In addition, the industry survey provides tabulations of the net sales of R&D performing companies and the full-time-equivalent number of industry R&D scientists and engineers.
Data from surveys of State and local R&D expenditures and of nonprofit institution R&D expenditures, published occasionally by NSF, were also used./39/ These data were interpolated and extrapolated to obtain estimates for missing years.
Performer-based estimates.The R&D satellite account features estimates of R&D expenditures that are largely based on data reported by performers of the R&D rather than by funders of the R&D. This approach attempts to avoid at least two problems. First, the data reported by funders would have to be adjusted to convert them from a time-of-payment-to-the-performer basis to a time-of-expenditure-by-the-performer basis in order to be consistent with the timing with which purchases of goods and services are generally recorded in the NIPA's. Second, the data reported by funders would have to be adjusted to avoid double-counting. Otherwise, R&D that is subcontracted would be counted twiceonce by the primary source of funding and once by the secondary source of funding that subcontracted the R&D.
Expenditures for R&D performed by industry, by public and private universities and colleges, and by academically administered, federally funded research and development centers (FFRDC's) are prepared from data reported by R&D performers. In recent years, these performers have accounted for roughly 85 percent of all R&D expenditures. For the remaining 15 percent, performer reports are not available, and expenditures are based on data reported by funders.
Adjustments to the survey data.BEA adjusts the survey-based data to make them conceptually and statistically consistent with the NIPA's. The adjustments raised the level of current-dollar expenditures, on average, 2.5 percent in 195392; in 198792, the adjustments raised the level somewhat less, an average of 1.5 percent. The four major types of adjustments to the NSF survey data are summarized in table C.
First, BEA adjusts the R&D expenditures from NSF surveys to obtain consumption of fixed tangible capital used in performing R&D. Two methods are used, depending on the handling of this capital in the surveys. For the Federal Government and for State and local governments, BEA removes expenditures on fixed reproducible tangible capitalstructures and equipmentand adds an estimate of the consumption of that capital based on BEA's perpetual inventory methodology. For other performers, BEA converts the depreciation implied in the R&D survey (part of overhead) to a basis that reflects the valuation and consistency (for example, of service lives) appropriate for economic accounts.
To make the conversion for private and public universities and colleges and for FFRDC's administered by universities and colleges, an estimate of expenditures on R&D structures and equipment is made as a first step. Equipment is then split between capital equipment and expensed "research" equipment. (Research equipment is purchased by the academic institution from current fund accounts.) Next, the implied depreciation of structures and capital equipment is calculated using the depreciation patterns and the service lives prescribed for NSF reporting purposes. This implied depreciation and the expenditures on research equipment are then subtracted from reported R&D expenditures. Finally, BEA's estimate of consumption of structures and equipment, which is estimated using BEA's perpetual inventory methodology and the same service lives used in preparing fixed tangible capital stocks, is added back in; it is valued at current cost.
To make the conversion for industry performers, the information on capital expenditures needed to develop an estimate of implied depreciation is not available. BEA developed estimates based on the 1958 survey-based depreciation reported by NSF and the 1982 depreciation charges of R&D auxiliaries from the Bureau of the Census 1982 Enterprise Statistics . For each industry, the depreciation is converted to an economic basis using the ratios of historical-cost to current-cost valuation from the estimates of academic expenditures described above.
Second, two timing adjustments are made. Data from the Federal Government are on a fiscal year basis, and data from universities and colleges are on an academic year basis. These data are converted to a calendar year basis by using weighted averages of adjacent years. Federal obligations by performerfor example, contracts awarded or other binding commitments made that will require outlaysare converted to expenditures using statistically estimated phasing patterns prepared by BEA.
Third, data from the academic surveys are adjusted to exclude R&D performed in geographic areasprimarily U.S. territories and possessionsthat are not included in the NIPA's. Because these academic surveys tabulate expenditures by individual school or FFRDC, these expenditures could be removed.
In the fourth adjustment, BEA developed R&D expenditure estimates for industries and years that had been suppressed by NSF in order to avoid disclosure of confidential information from the industry survey. The BEA estimates are based on statistical techniques (primarily interpolations), on fragmentary data from other sources, and on judgment; they do not disclose confidential company data.
Backward extensions of the survey data.In order to develop the necessary statistical foundation to construct capital stock estimates using the perpetual inventory method, BEA prepared estimates of R&D expenditures for years prior to 1953, when the NSF surveys began. Using various data sources, BEA extended the R&D expenditure estimates back to 1920. Estimates of R&D expenditures are not made for years before 1920, because little information is available; it is assumed that R&D expenditures before 1920 were quite small.
The BEA estimates of expenditures by R&D performers for 192052 are made using a methodology generally similar to that described by John Kendrick./40/ First, expenditures are established for benchmark years. Benchmark years are 1921, 1930, 1940, and 1951 for total R&D expenditures and 1921, 1927, 1931, 1933, 1938, 1940, 1946, and 1951 for industry R&D, and the expenditures are from Nestor Terleckyj./41/ Expenditures in the non-benchmark years in 192041 are estimated by interpolation or extrapolation, using estimates of R&D reported by Vannevar Bush as indicators when available./42/ Expenditures for nonbenchmark years in 194252 are estimated using data published by the Research and Development Board./43/
The BEA estimates for 192052 are supplemented in two ways. Expenditures for FFRDC's, from their inception in 1942, are based on data published in a study by the Denver Research Institute./44/ Expenditures on the Manhattan project (which developed the first atomic bomb) for 194246 are based on data reported by Richard Hewlett and Oscar Anderson, Jr./45/ At its peak in 1944, the Manhattan project accounted for nearly one-tenth of all R&D performed in the United States.
Issues with R&D by funder and by type.As discussed above, the basic framework for the R&D satellite account is R&D arrayed by performer. In addition, R&D is shown broken down by source of funding and by type.
In the R&D satellite account, a maximum of five sources of funding are distinguished: Federal Government, State and local governments, industry, universities and colleges, and other (which includes nonprofit institutions and foreign sources). The satellite account shows less source-of-funding detail for some performers because of varying source data. For example, three sources of funding are distinguished for R&D performed by industry (see table 2.1), although industry source data divide R&D performed into only two funding categoriesFederal funds and all other funds. Within the other funds category, BEA estimated State and local government funding using data from surveys of State and local R&D. BEA assumed that the funding of industry R&D from universities and colleges and from other nonprofit institutions is negligible and that foreign funding is small enough so that the remainder of industry R&D funding could be labeled "from industry."
The breakdowns by type embody substantial uncertainty. Because there are no clear-cut distinctions between the types, uncertainties must be resolved by the judgment of the survey respondents. It is unlikely that these resolutions will be the same among reporting groups. For example, academic respondents may be less likely to report research as applied or development. In addition, the breakdowns by type are voluntary on industry and academic surveys, and not all surveys ask for the full three-way breakdown.
Table D provides an overview of the source data and methods used in deriving constant-dollar R&D expenditures. For each group of performers, constant-dollar expenditures are calculated by dividing current-dollar expenditures by price deflators at the most detailed cost level available. Constant-dollar estimates begin in 1929, the year NIPA price indexes become available. The methodology of using cost components that are matched with existing indexes builds on both existing BEA work and work done by othersmost notably John Jankowski of NSF and Edwin Mansfield of the University of Pennsylvania for industry R&D expenditures and D. Kent Halstead of Research Associates of Washington for academic R&D expenditures./46/
Typically, the cost components are compensation of employees, materials and supplies, and overhead. Data used for deflation include information on prices paid by the Federal Government, the NIPA price index for noncommercial research organizations, the producer price index for industrial commodities published by the Bureau of Labor Statistics (BLS), average hourly earnings and median weekly salaries of managers and administrators from BLS, annual mean salaries of engineers from the Engineering Manpower Commission, component price indexes from the National Institute of Health's biomedical R&D price index, and higher education price indexes published by Research Associates of Washington./47/ The base year is 1987, for which each price index is set equal to 100.
R&D investment is broken into two componentsfixed investment and change in R&D inventories. In the R&D satellite account, it is assumed that expenditures on R&D are inventoried for 1 year before they are included in R&D fixed capital in order to allow for the time needed to complete R&D projects. When measured in constant dollars, R&D fixed investment is equal to the R&D expenditures for the preceding year, and for each year, the change in R&D inventories is equal to the change in R&D expenditures. Because the beginning and ending inventory levels reflect different price levels, current-dollar R&D inventories from the end of the preceding year are revalued to correspond to the prices for the current year by using an inventory valuation adjustment./48/
R&D fixed capital stocks are constructed using the methods BEA uses to construct capital stocks associated with the NIPA's, including the same perpetual inventory method that is used for gross and net stocks of fixed reproducible tangible capital. Thus, the estimates of fixed intangible R&D capital are comparable with those of fixed reproducible tangible capital.
With the perpetual inventory method, the gross capital stock for a given period is obtained by cumulating past investment and deducting the cumulated value of investment that has been discarded, using estimated average service lives and retirement patterns./49/ The gross stock of fixed capital is a measure of the cumulative value of past investment still in existence. The net capital stock is equal to the gross stock less the accumulated depreciation on the assets in the gross stock.
The perpetual inventory method used by BEA is based on uniform service lives, straight-line depreciation, and replacement cost. To adjust for varying retirement patterns, discards (retirement years) are based on a Winfrey S-3 distribution, which is a bell-shaped distribution around the expected service life of the R&D capital. Discards of capital begin as early as 45 percent of, and end as late as 155 percent of, the average lifespan.
In deciding how to apply its methodology to R&D, BEA examined several alternative depreciation patterns and performed sensitivity studies. Geometric depreciation is the pattern typically used in R&D studies, and a rate of 11 percent per year for R&D fixed capital is a plausible midpoint of a range published by academic researchers. (Some recent studies had estimates that ranged from 9 to 13 percent per year and tended to concentrate near 11 percent.) BEA's studies showed that using a depreciation rate for R&D fixed capital of 12 percent yielded a real stock of R&D capital for 1991 that was $65 billion lower (1987 dollars) than a stock constructed using a rate of 11 percent. Using a rate of 10 percent yielded a real stock of R&D capital that was $74 billion higher. Regardless of which rates are used, the general patterns of R&D stock are similar over time.
Because BEA currently uses the straight-line perpetual inventory method for fixed tangible capital, an average service life for R&D capital was chosen that yields a net stock comparable to a net stock from a geometric depreciation rate of 11 percent; an 18-year service life for straight-line depreciation yields the closest match. The gross and net stocks constructed in the account for 195992 are based on current- and constant-dollar R&D investment for 193091.
As with the constant-dollar expenditure estimates, constant-cost net and gross stocks are expressed in 1987 prices. Net and gross stocks valued at replacement cost are constructed by multiplying the constant-cost stocks by the corresponding R&D implicit price deflator.
Tables 1.1 through 5.2 follow.
Comparison of R&D Capital Stock Estimates
1. National Science Board, National Science Foundation , page 89.
2. In an integrated set of economic accounts, flows of fixed investment are viewed as forming stocks of reproducible capital. BEA however, estimates the stocks of consumer durables and of government equipment and structures as if personal consumption expenditures on durable goods and government purchases of durable goods and structures had been treated as fixed investment flows. Flows derived from the NIPA series are used to estimate the stocks, which are, therefore, consistent with the NIPA's.
3. For an early presentation about the preliminary work, see Carol Carson and Bruce Grimm .
4. See, for example, National Science Board, National Science Foundation .
5. See Organisation for Economic Co-operation and Development , page 29. This is the fifth edition of the Frascati Manual. The manual was first prepared in 1961.
6. The definitions of R&D and the three types of activities that are found elsewherefor example, in financial accounting standards and in NSF's specific surveysare similar to these definitions but place emphasis on elements of the definitions that are relevant to the context.
7. See James Adams  and .
8. John Kendrick , pp. 7981.
9. See Eric Schiff , pp. 434435 and George Jaszi , pp. 454455.
10. See Nancy Ruggles and Richard Ruggles , especially page 99.
11. See John Kendrick , especially pp. 121.
12. See Robert Eisner , especially pp. 820.
13. See .
14. See System of National Accounts , paragraph 6.163.
15. See Michael Braibant .
16. See Fritz Bos, et al. .
17. In BEA's input-output accounts, neither current expenses nor receipts for R&D are identified at the published level of detail. A portion of R&D is identified at the level of detail at which the estimates are prepared.
18. See Nestor Terleckyj  and Vannevar Bush .
19. See, for example, Federal Republic of Germany, Federal Statistical Office .
20. See, for example, Zvi Griliches .
21. See, for example, John Kendrick .
22. Eric Schiff  and Fritz Bos, et al. .
23. See, for example, Robert Eisner  and John Kendrick .
24. See Congressional Budget Office .
25. See John Kendrick , John Rapoport , and Lenore Wagner .
26. See, for example, James Adams , James Adams and Leo Sveikauskas , Gellman Associates , Edwin Mansfield , Ariel Pakes , David Ravenscraft and F.M. Scherer , and Nestor Terleckyj  and .
27. See, for example, Zvi Griliches , John Kendrick , David Levy and Nestor Terleckyj , Frank Lichtenberg and Donald Siegel , and Nestor Terleckyj  and .
28. See, for example, Bureau of Labor Statistics , Congressional Budget Office , Robert Eisner , and M. Ishaq Nadiri and Ingmar Prucha .
29. See Ariel Pakes and Mark Schankerman .
30. A full description of BEA's estimates of tangible capital stock may be found in Bureau of Economic Analysis . BEA is now reviewing the appropriateness of the concepts and measurement methods underlying these estimates. Future BEA capital stock estimates may be based on somewhat different concepts and measurement methods.
31. See Ariel Pakes and Mark Schankerman .
32. See James Adams .
33. M. Ishaq Nadiri and Ingmar Prucha .
34. In the tables showing industry detail, FFRDC's administered by industry are combined with the remainder of industry because source data do not provide FFRDC's administered by industry separately by industry classifications.
35. BEA constructed the R&D IPD at the finest level of detail possible. In contrast, NSF and others have used the GDP implicit price deflator or other summary price measures to produce estimates of constant-dollar R&D expenditures. A comparison of the total R&D IPD and the GDP IPD shows that the latter provides a reasonable approximation to the former for deflating total R&D expenditures. Use of the GDP IPD overstates the historical growth in R&D performed in public and private universities and colleges and understates the historical growth in R&D performed in many FFRDC's. NSF views the GDP deflator as an "opportunity cost" of the real resources forgone in engaging in R&D rather than as measuring the costs of doing R&D, and recognizes that the deflator is less useful for calculating finer-level components of R&D. See National Science Board, National Science Foundation .
36. See Jack E. Triplett .
37. See Division of Science Resource Studies, National Science Foundation , , , , , and  for more information.
38. Beginning in 1992, the industry samples will be redrawn annually.
39. See Division of Science Resources Studies, National Science Foundation  and  for more information.
40. See John Kendrick .
41. See Nestor Terleckyj .
42. See Vanevar Bush .
43. See Research and Development Board, Department of Defense .
44. See Denver Research Institute .
45. See Richard Hewlett and Oscar Anderson, Jr. 
46. See John Jankowski , Edwin Mansfield , and Research Associates of Washington .
47. Additional information on BEA's deflators for Federal purchases of R&D may be obtained from the Bureau of Economic Analysis . Additional information on BEA's deflators for higher education and research may be obtained from the Bureau of Economic Analysis . Additional information on the biomedical R&D price index may be obtained from Office of Science Policy and Technology Transfer, National Institutes of Health .
48. See Bureau of Economic Analysis  for more information about the NIPA inventory valuation adjustment.
49. For a more complete description of the NIPA perpetual inventory method, see Bureau of Economic Analysis , page M-3.