Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
The Measurement of Output Any measure of productivity relates output to the input of a single factor or to a weighted average of inputs of several factors. Spurious movements in productivity measures can be avoided by limiting the output in the numerator of the productivity ratio to those activities in which the inputs are used. Measuring output also requires constructing the price indexes needed to adjust the output numerator for the most important changes in the quality of the output. This chapter begins with a discussion of criteria for selecting particular activities for inclusion in the output numerator; it then reviews the procedures now used to measure quality change and considers possible alternative procedures. THE OUTPUT MEASURE DEFINING THE ACTIVITIES TO BE INCLUDED A productivity ratio, whether labor productivity, capital productivity, or multi-factor productivity, is intended to identify shifts in a production function linking output to one or more inputs.) The measurement of output for use in productivity ratios should be based on criteria that minimize spurious changes in productivity not caused by shifts in the production function. Goods and services included in the numerator of the productivity ratio must be those produced by the inputs included in the denominator. To exclude an output produced by an included input would cause a spurious decline in productivity whenever that input increased and 88
The Measurement of Output 89 vice versa. Conversely, to include an output produced by an excluded input would cause a spurious increase in productivity when that output increased and vice versa. The principle is equally valid for the measure- ment of productivity in individual activities or industries and in the whole economy. If one wants to understand technological change in a single activity, one must make sure that the measure of output includes everything produced by included inputs and excludes anything produced by excluded inputs. Once this test of consistency has been applied, a productivity index can be computed for any type of activity. The measurement of productivity for the economy raises a set of issues about the selection of the activities to be included. In recent years, broad welfare-oriented output concepts have been proposed that would exclude some portions of the official GNP because they are "regrettables," goods and services that do not contribute directly to well-being: among these are spending on defense goods, protection against crime, and commuting. Conversely, some activities that contribute to well-being that are not presently part of GNP are proposed for addition to the welfare-oriented output concept: for example, parents' use of time in caring for Clear children and other types of services produced in the home. Whatever the merits of such welfare-oriented approaches to the measurement of total economic well-being, it would be inadvisable to incorporate most of the proposed changes in aggregate productivity measures because of the difficulty of meeting the required standard of consistency in the matching of output and input.2 The point of departure for the measurement of aggregate productivity is a labor input concept that includes total hours sold on the market or worked to produce market goods and excludes time spent at home or time spent in voluntary leisure and charitable activities. By hours sold on the market, we mean the hours of wage and salary workers. By hours worked to produce market goods we mean the hours of proprietors, partners, and unpaid family workers engaged in producing goods and services for sale to others. The two together are referred to here as market-oriented labor. Without a measure of inputs used in non-market-oriented activity, the output of such activity cannot be included in the aggregate productivity measure. At the same time, the exclusion of the output of regrettables would be impracticable in many cases because market-oriented labor input simultaneously produces regrettable (often intermediate) goods and desirable output. For instance, how is the labor engaged in producing automobiles for commuting (a regrettable activity) to be separated from that engaged in producing automobiles for pleasure driving when the same automobiles are often used for both purposes? To limit output to goods and services produced by market-oriented
90 REPORT OF THE PANEL labor may seem to give undue priority to the measurement of market- oriented labor input in principle, we could recommend tailoring input concepts to the desired output universe, rather than vice versa but in fact, most types of non-market-oriented labor, such as parents' time in caring for children, produce non-marketed goods and services. Without a direct measure of the output of these hours, any attempt to measure their productivity will yield a meaningless result. The limitation of output to goods and services produced by market-oriented labor has another advantage. The numerator of the productivity ratio, real output, then applies to the same universe as the numerator of the ratio compensation per hour. The ratio of compensation per hour to output per hour is unit labor cost, a variable of interest in itself for the analysis of the proximate determinants of the rate of inflation. The requirement that output be produced by market-oriented labor is a necessary but not sufficient condition for its inclusion in the numerator of the productivity ratio. A second criterion is that output and inputs be independently measured in a given industry or activity. When output is measured by hours of labor input, as in some categories of government production, not-for-profit activities, and private services, the productivity ratio by definition is constant. Since the use of labor input as an output measure usually reflects practical or philosophical difficulties of measuring output, rather than a "true" absence of productivity change, these activities are best excluded from aggregate productivity measures. Other- wise, the rate of change of aggregate productivity will be biased toward zero, and the level of aggregate productivity will be biased toward constancy. In addition, the growth of aggregate productivity will appear erroneously to slow down whenever there is a shift in the mix of expenditures toward the industries or activities in which output is measured by labor input. Even when the numerator and denominator of a productivity ratio apply to exactly the same universe, there may be shifts in productivity that do not result from changes in efficiency. A cyclical increase in the demand for the products of a high-productivity industry, such as the automobile industry, relative to a low-productivity industry, such as the textile industry, raises average economy-wide productivity when that average is calculated as the ratio of aggregate output to aggregate input. An alternative procedure would be useful for some special purposes, namely, to calculate the change in aggregate productivity as the weighted average, using fixed weights, of the rates of change of productivity in the individual industries, but there may not be enough reliable industry measures for such a procedure.3
The Measurement of Output GUIDELINES FOR INCLUSION OR EXCLUSION 91 Table 5-1 summarizes the current treatment of separate components of the GNP in the broadest official measure of productivity. The first line includes the core activities of the economy in which output is produced by market- oriented labor input and the output measure is independent of the measure of labor input. Ideally, an aggregate productivity measure would be limited to those activities; in fact, the broadest official measure of productivity also includes the activities in the second line, in which output is measured by labor input. This is not a serious problem, however, since 95 percent of the measure consists of activities in which the measurement of output meets our criteria and only the remaining 5 percent consists of activities in which the measurement of output is based on labor input. The activities for which output is measured by labor input are those for which no valid price index is available for the deflation of current-dollar spending and for which an hourly earnings index is used instead. If prices tend on average to rise by an amount equal to the increase in hourly earnings minus the average rate of growth of labor productivity, then the use of hourly earnings as a proxy for a missing price index has the effect of assuming that the change in labor productivity is zero and thus of measuring output by labor input. Among the activities in which hourly earnings play a role are shoe repair, household services, health insurance, the expense of handling life insurance, bank services to individuals, spectator sports, clubs, and the services of proprietary hospitals and schools; these activities all fall within the general area of consumer services. The other major area for which output is partially measured by labor input is construction (see "structures" in Table 5-1~. Although expenditures on single-family houses and public roads are deflated by relatively good price indexes, some of the remaining portions of the construction industry are still deflated by input-cost price indexes, which are a weighted average of wage rates and materials costs that make no allowance for productivity change in construction activity.4 As indicated in the first and second rows of Table 5-2, aggregate productivity growth appears to have slowed down substantially less between 1947-1967 and 1967-1972 for the private business sector, excluding the construction industry than for the total private business sector. The major contribution of construction to the apparent slowdown in the growth of productivity is shown directly in the bottom half of Table 5-2. The contribution of the construction industry is so large relative to its overall share of economic activity because the absolute level of measured productivity in construction actually declined during this period.5
92 REPORT OF THE PANEL TABLE 5-1 Components of Gross National Product and the Measurement of Aggregate Productivity (billions of current dollars, 1976) Component Amount (billions of dollars) 1,700.1 TOTAL: Gross national producta Total included in productivity measure Included in the Bureau of Labor Statistics' productivity measure for the private business sector, output not measured by labor input Included in productivity measure, output measured by labor inputb Personal consumption expenditures Producers' durable equipment Structures Total excluded from productivity measure Private business sector, excluded from productivity measure C (owner-occupied housing) Other sectors excluded from productivity measured Government Rest of the world Households and institutions Statistical discrepancy ,336.1 ,274.2 61.9 43.6 2.1 16.2 364.0 96.4 267.6 92.5 14.4 56.5 4.2 aTable 1 .7 in Survey of Curren t Business July 1 9 7 8. bEstimate provided by the Bureau of Economic Analysis. The figures shown equal 4 per- cent of personal consumption expenditures, 2 percent of producers' durable equipment, and 10 percent of structures. CTable 8.3, line 63 in Survey of Current Business July 1978. dTable 1.7 in Survey of Curren t Business July 1978. SOURCE: Data are from the tables in the U.S. Department of Commerce, Surrey of Cur- rentBusiness, July 1978 and from unpublished data of the Bureau of Economic Analy sis. The components of the line items are detailed in the footnotes. When we examine the slowdown for the next period, 1972-1977, it appears that the sectors in which labor input plays some role in measuring output cannot explain more than a small fraction of the deceleration. Even when construction, financing, insurance, and real estate (FIRE), and services are excluded, the measured productivity growth rate for 1972- 1977 was almost two percentage points less than the 1947-1967 average. In this more recent period, the contribution of construction to the slowdown was much less important than in 1967-1972. Overall, only a small portion of output in the private business sector is measured by labor input, and most of the apparent productivity slowdown between 1947- 1967 and 1972-1977 occurred in the goods-producing sector of the
The Measurement of Output TABLE 5-2 Productivity Growth Rates in the Private Business Sector and Subsectors, 1947-1977 and Subpenods (output per labor-hour in percentage points) 93 Growth Rate 1947-1957 1957-1967 1967-1972 1972-1977 1947-1977 Annual average pro ductivity growth rates Private business sector (PBS) 3.23 3.25 2.08 1.26 2.71 PBS less construc tion 3.20 3.27 2.48 1.41 2.80 PBS less FIREa 3.33 3.42 2.15 1.27 2.82 PBS less services 3.41 3.29 2.33 1.47 2.87 PBS less construc tion, FIRE, and services Total apparent slow down from 1947-1967 to subperiod indicated Contribution of con struction Contribution of FIRE Contribution of ser vice industries Contribution of remaining sectors 3.51 3.53 2.99 1.72 1.16b 1.98b 0.41 -0.07 0.14 0.53 3.13 0.15 -0.13 0.10 1.80 aFIRE = finance, insurance, and real estate. bThe sum of the four percentages below does not equal this percentage because of interaction effects. SOURCE: Unpublished data of the Bureau of Labor Statistics' Office of Productivity and Technology. economy, in which labor input measures play virtually no role in measuring output. It is easier to state the platitude that the government agencies should improve their measures of real output for activities in which labor input currently serves as a proxy than to suggest specific ways to improve the treatment of individual industries. For instance, it can be argued that the output of proprietary schools should be measured by student-hours or number of graduates rather than by the input of teacher-hours and secretary-hours. A shift to number of graduates as a measure of output would show a productivity improvement when class sizes increase. This might be a valid approach when new types of audio-visual aids are introduced, but it would elicit protests from teachers who believe that students receive more educational output per student from small classes
94 REPORT OF THE PANEL than from large ones.6 Similar questions arise in the measurement of productivity in proprietary hospitals.7 At the least, each "problem" industry should be treated separately. Consumer expenditures for the output of some government enterprises, including the postal service, are already measured in the national income accounts on an output basis (e.g., pieces of mail weighted by postage rates in each category), rather than on a labor input basis. Even though a perfect measure of output is unattainable and some questions about the definition of output may be unanswerable, we should not too easily accept the option of using labor input as a proxy for output. A transition to an output basis for measurement for some activities may be possible when research reaches a reasonable consensus on the proper criteria for the measurement of output.8 Productivity measurement would benefit from research on how to replace labor input measures used as output proxies in some sectors of the economy. Until such research yields results, productivity measures might benefit from the exclusion of those activities in which such proxies are used. Nevertheless, current criteria for the exclusion of sectors from aggregate productivity measures are generally sound. Imputed rent on owner-occupied housing is properly excluded from the GNP before productivity ratios are calculated, because there is no associated market- oriented labor input (see Table 5-1~. Measurement of the "true" output of police, firemen, and soldiers raises insuperable problems and justifies the exclusion of the government activities in which they are engaged. Expenditures by consumers that are regrettable intermediate inputs to the achievement of ultimate well-being are properly not excluded (for example, air conditioning, heating, commuting, uniforms, burglar alarm systems, and locks) in part because they are produced by market-oriented labor input, and it would be difficult in practice to exclude from aggregate labor input the workers engaged in producing them. There is growing recognition that the interpretation of productivity measures could be enhanced if more attention were given to quality-of-life issues. Quantitative data from federally funded research by BEA'S Environmental and Nonmarket Division, studies of social indicators, and work on goals accounting are intended to broaden our understanding of many critical aspects of productivity change not fully covered in the existing framework of economic measures. This broadening of understand- ing does not require that data on the quality of life be incorporated into the measures of output and productivity now in use. Recommendation 6. The Panel concludes that many useful analyses of economic and social welfare issues can be undertaken within the
The Measurement of Output framework of output and input used in the current official measures. However, the Panel does agree that for the study of many important social problems for example, improvement of the health status of the population-definitions of output and input that go well beyond those currently used to measure productivity are required. The Panel believes that for each of these special problems, partial and special measures of output and input should be developed when possible. However, the Panel concludes that it is not now possible to add or subtract such special measures from the conventional measures of output and productivity to construct a meaningful single index of overall welfare for the nation. NET VERSUS GROSS OUTPUT 95 At present the output numerator of the BES productivity ratio is defined as gross rather than net of depreciation. If net output and depreciation grow at the same rate in real terms, the choice between a gross and a net output concept makes no difference in the growth rate of the resulting productivity ratio. But if net output and depreciation grow at different rates, there will be a difference in the growth rates of gross and net productivity ratios. There are arguments in favor of both approaches, and it is often useful to present estimates on both bases. Proponents of the net-of-depreciation approach argue that depreciation represents the "using up" of previously produced capital goods. Denison (1967, pp. 1~15) points out: Net product measures the amount a nation consumes plus the addition it makes to its capital stock. Stated in another way, it is the amount of its output that a nation could consume without changing its stock of capital. Insofar as a large output is a proper goal of society and objective of policy, it is net product that measures the degree of success in achieving this goal. Gross product is larger by the value of capital consumption. There is no more reason to wish to maximize capital consumption the quantity of capital goods used up in production than there is to maximize the quantity of any other intermediate product used up in production, such as, say, the metal used in making television sets. It is the television sets, not the metal or machine tools used up in production, that is the objective of the production process.; Proponents of the gross-of-depreciation approach, on the other hand, do not believe that productivity measures should be based solely on a welfare criterion. A portion of the labor input included in the denominator of the productivity ratio is engaged in the manufacture of capital goods to replace those goods that are wearing out or becoming obsolete. There is no more
96 REPORT OF THE PANEL reason to exclude capital consumption from the output numerator of the productivity ratio than there is to exclude regrettable goods and services that do not contribute directly to well-being. According to this view, spending on defense goods, crime protection, commuting, antipollution devices, and on capital goods to replace those used up in production is all similar: they are part of the output of marketed labor input whether or not they are desired for their own sake. Although the Panel does not choose between these two approaches, it notes that gross and net productivity measures show similar secular rates of growth. In analyses of the sources of economic growth, however, use of the gross-of-depreciation approach will show a larger contribution from capital and a smaller contribution from other sources than use of a net-of- depreciation approach. Over short periods, gross-of-depreciation measures of productivity fluctuate less than net-of-depreciation measures in response to output movements because measured capital consumption exhibits little cyclical variation. CAPITALIZATION OF INTANGIBLE INVESTMENT Some researchers have recommended the inclusion in output of some items now treated as consumption or intermediate goods, which would then be capitalized, that is, treated as investment. A prime candidate for addition to a welfare-oriented GNP measure is investment in research and development (see Griliches 1973~. Why should a society be treated as having less real output when it diverts resources from investment in machines to investment in research? Exclusion of research activity from real output means that the change of an hour from conventional production to research will decrease measured productivity in the short run until the fruits of the research begin to raise conventional output. The main problem involved in including research as a part of final output is that no measure of research output is available other than labor input. It would probably be more desirable to exclude research employees from private labor input than to attempt to add research output to aggregate private real output. Education is another prime candidate for capitalization as an investment good. At present, both the output and the labor input of most of the education industry are excluded from productivity measures because all activity in general government and not-for-profit institutions is excluded. In principle, however, education and health should be treated similarly. The main advantage of health over education for productivity measure
The Measurement of Output 97 ment is the existence of discrete "outcomes" or "products" (e.g., a tonsillectomy) that can be included in the consumer price index (thus allowing a genuine rather than a labor-based output measure). In contrast, defining the product of education raises the problem of choosing between units of educational input and the alternative of the number of students "processed" or graduated. But Scott (in this volume) suggests that health output is equally difficult to measure and that alternative concepts of health outcomes and processes yield differing measures of productivity in the health industry. Consumer expenditures on antipollution and safety devices, such as those required by the federal government for automobiles, are currently included in real GNP. This inclusion occurs because in constructing deflators for automobiles and other products a change in value due to the addition of a government-mandated device is counted as an increase in quality rather than as an increase in price. Since the production of pollution is not currently subtracted from the GNP, the current procedure in effect measures the benefits of antipollution devices by their costs and exaggerates their benefits if costs exceed benefits. Applying our previous principles, this approach seems desirable for the purpose of productivity measurement. We do not want the shift of a worker from an assembly line building automobile transmissions to an assembly line building catalytic converters to cause a sudden drop in measured productivity, as it would if all labor input in the automobile and automobile parts industries was included in the input denominator of the productivity ratio but production of converters was excluded from the output.9 Business expenditures for antipollution and other required devices are at present treated in exactly the same way because the price deflators used are developed by the same methods. However, in the case of business, the shift of one dollar of gross investment from normal "productive" types of investment to "unproductive" government-mandated investment will cause future measured output to be lower than otherwise. The measure- ment problems created by business expenditures on pollution control and worker health and safety can be treated in one of two ways. The first is to leave the output definition unchanged and to analyze separately the contribution of government-mandated expenditures to the change in productivity.- The second is to estimate what the contribution of these expenditures to output would have been and add it to output to produce a measure of what output would have been without government-mandated expenditures. We prefer the first way and note the recent efforts of Denison (1978b) to analyze the contribution of environmental legislation to productivity change.
98 REPORT OF THE PANEL DIVIDING VALUE CHANGES BETWEEN CHANGES IN PRICES AND CHANGES IN REAL OUTPUT CURRENT PROCEDURES FOR MEASURING PRICE CHANGES The GNP implicit price deflator is a weighted average of thousands of individual price observations, most of which are collected by the BES for the consumer price index (CPI) and the producer price index (PPI), which was formerly called the wholesale price index (see Early 1978~. The BEA publishes annual deflators for about 60 different categories of consumer spending, with individual item indexes of the CPI averaged together with fixed weights to form each of the group deflators. The implicit deflator for total consumer spending is an average of the 60 separate group deflators with the current level of real spending by category as weights.~° Similarly, annual deflators for producers' durable equipment are published for 22 categories, based almost entirely on individual item indexes of the PP~.~: A variety of sources, both government agencies and private firms, is used to compile the 25 published deflators for purchases of residential and nonresidential structures. The main emphasis of this section is on the adequacy of allowance for changes in quality in the current procedures of price measurement. If changes in the value of products that are actually price changes are treated instead as quantity or quality changes, the growth of productivity and aggregate output are overstated. On the other hand, if value changes that actually represent quantity or quality changes are treated as price changes, the growth in productivity and aggregate output are understated. The basic Bus method of price measurement in both the cat and PP~ is the specification method. BES staff draw up definitions or specifications of the commodities they want to include, as in this example from the apt: "color television receiver, console model, 21", 23", or 25" picture tube, veneer cabinet; manufacturer to dealer or distributor, f.o.b. factory or warehouse, each." For the All, companies provide on monthly mail questionnaires price quotations on one or more particular color TV models that fall within the BES specification. For the UPS, field agents record the retail prices of items that meet the specification. The complications begin when, for example, the TV manufacturer switches to a new model that includes a remote control feature as standard equipment and costs $50 more than the old model. A manufacturer altering a product in any significant manner is supposed to report the change to BES, which then decides how much to adjust the reported price to reflect the change in quality. In the simplest case, a BES official might call the manufacturer and learn that the remote control feature had
The Measurement of Output 99 previously been offered as a $50 option, so that there had been no price change at all but simply the inclusion of a previous option as standard equipment. Unfortunately, matters are not always so simple. Many quality changes do not involve the inclusion of former options as standard equipment, but are entirely new features. In some cases, producers are asked to estimate the added cost of a product change. Automobiles, the most important single type of durable good, are given the most careful attention. Each September, several BES stab members travel to Detroit to consult with the major auto manufacturers and identify specification changes on new models for which adjustments must be made in both the Cal and PA auto indexes. If a manufacturer has introduced a new, heavier bumper to comply with federal safety regulations, the manufacturer is asked to estimate the increased cost of producing the new bumper. This difference in cost (marked up to the retail level for the ceil is then subtracted from the reported price increase on the new-model automobile and thus is treated as a quality change rather than a price change. When consumers spend more dollars on automobiles to pay for the new bumpers, the fact that the price index has been held constant allows the whole change to be reflected as an increase in real GNP. The extent of any change in labor productivity in the auto industry depends on how much additional labor has been used to produce and assemble the new bumpers. Whether such additions or accessories are mandated by the government or are independently included in products by manufacturers, the current treatment of the change in measured prices is consistent with the basic philosophy of quality adjustment in the official BEA deflators. That philosophy is that two different goods are to be considered as identical in quality if they would have cost the same amount to produce in a base year. Quality improvements are only those changes that are "accompanied by an increase in the real cost of production" (see Jaszi 1962, pp. 332-335~. This added-options philosophy of quality adjustment is entirely adequate for many types of product improvement. When there are several models on the market at one time, differing only in options, size, or other attributes that cost real resources to produce, a manufacturer can estimate how much options add to production cost, and thus how much extra quality a new, larger, or more "loaded" model represents by the added-cost criterion.~3 Adjustments for quality change of this type and adjustments due to the appearance of new models are made frequently. From January 1967 to January 1975, the BES made adjustments for roughly 6,400 price observations (a price quotation for a single model for a single month) of a total of 470,000 in the wholesale price index (see Ruggles 1977, Early
100 REPORT OF THE PANEL 1978~. Similar adjustments were made in the construction of the consumer price index. Although the extent of this attention to quality change is impressive, many types of product improvements are completely missed by the current procedures. TYPES OF QUALITY CHANGE The limitations of the present methodology are apparent when it is noted that quality change can take one of three forms. First, a quality change can take the form of a change in the quantity of costly resources used to produce a product. This "type 1" quality change, such as the addition of a remote control device to a television set or a heavier bumper to an automobile, is adequately treated now for many categories of goods. Second, a quality change can be achieved by a technological innovation that raises the quality of a product without any increase in current resource inputs. An important example of this "type 2" quality change is the development of new models of electronic computers that have larger memories and more rapid computational abilities but that cost roughly the same as the models they replace. Third, any consumer or producer durable good simultaneously provides services to its users and imposes costs on them in the form of operating expenses, particularly energy requirements and maintenance. These operating costs may be trivial for some durable goods (furniture may need only an occasional bit of polish), but for some goods, such as electrical generating equipment and commercial jet aircraft, cumulative operating costs over a product's life may be many times larger than the initial capital cost. This "type 3" quality change refers to any design changes in durable goods that result in higher or lower operating costs, holding constant both the quantity of services provided by the good and the wages and prices of the inputs used in its operation. A reduction in the price of gasoline that makes automobiles less expensive to operate is not a type 3 quality change, but the redesign of an engine to improve fuel efficiency iS.~4 Although BES has steadily improved the thoroughness with which it adjusts its price indexes for type 1 quality change, it has devoted little attention to types 2 and 3. If quality changes of types 2 and 3 occur frequently but are not measured by the BES, the aggregate rate of price increase in the private economy is overstated and the corresponding growth rates of output and productivity are understated. Furthermore, if the relative importance of the three types of quality change shifts over time, then the difference between the measured and true rates of productivity growth also changes. If, since the oil price increase of 1974,
The Measurement of Output 101 firms have placed more emphasis on design changes that improve energy efficiency, and if these type 3 quality improvements are unmeasured, some portion of the measured slowdown in the growth of productivity in the past few years may be a result of the failure to take account of these quality changes. CONCEPTUAL PROBLEMS IN THE MEASUREMENT OF QUALITY CHANGE There is no disagreement on the desirability or necessity of achieving adequate corrections for type 1 quality changes, but there has been considerable debate about the treatment of type 2 and type 3 changes. Although the discussions often read as though theoretical or philosophical points were at issue, in fact most of the opposition to suggestions that quality changes of types 2 and 3 be measured appears to boil down to a question of feasibility, not philosophy.~5 The current position of BES begins by distinguishing between input and output price indexes. BES would like to measure prices using the input criterion but is forced by measurement problems to use the output criterion. The Cal is intended to be an input index, in the sense that it approximates changes in the cost of obtaining consumption goods and services that are inputs into consumer welfare or utility. The appropriate quality-measurement criterion for an input price index is value to the user: two goods are treated as having the same quality if their value to the user is the same. If firms discover a way of raising the value of a good to its user without changing its cost (quality change of types 2 and 3), then its quality should be treated as having increased. Examples are the improved picture brightness and reduced repair incidence of color television sets. If a way is discovered to reduce cost without reducing the value to users, then quality should remain unchanged, as in the case of the frequent cost and price reductions for electronic calculators during the past decade. The criterion of user value is equally relevant in measuring the quality of capital goods purchased by firms for use in production: a firm will obviously prefer the more productive or more efficient of two capital goods if their cost is the same. To the extent that they are used to construct measures of real capital input, the components of the PA covering machinery and equipment should also be adjusted for quality change of types 2 and 3 to be consistent with the basic BES position.~7 In the literature the concept of an output price index is often contrasted with that of an input price index. It has been argued that the deflator for national output (and thus the pricing procedure that determines changes in
102 REPORT OF THE PANEL the real output numerator of the productivity ratio) should be based on a resource-cost criterion of quality measurement. By this criterion, two goods are equal if their production would have used the same resources in the base period of the price index (see Fisher and Shell 1972~. For this purpose, any change in a good that raises its cost of production, such as the remote control device on a television set in the earlier example, is treated as a change in quality; any change of type 2 or 3 that does not involve a change in cost is ignored. The terminology of the distinction between input and output price indexes is unfortunate, because there appears to be an implication that the output criterion should be used for the measure of real gross output (real GNP). However, real output actually consists of consumption, an input to consumer utility, as well as investment, changes in the capital stock, an input to the production process. Thus the input or user valuation criterion, which includes quality changes of types 2 and 3, is the appropriate deflator for real GNP (both its consumption and investment components) and for use in productivity statistics. Stated another way, an innovation that raises the user value of a consumer good at no increase in cost raises the input of goods into the consumer's utility, and an innovation that raises the user value of a capital good at no increase in cost raises the quantity of capital input available for use in the production process in the future. It is difficult to find a useful interpretation of an output or resource-cost price index, either for consumption or investment goods. Often there is no practical difference between the user-valuation and resource-cost concepts of quality change because extra resources are normally required to raise the value of a good to its user: for example, improvements in the acceleration of an automobile usually require a larger or more complex engine, and improvements in comfort usually require more expensive padding, upholstery, springs, and shock absorbers. But some changes in products increase user value while leaving cost constant or reduce cost while user value remains unchanged. The most important single example of this type of change has occurred in the electronic computer industry: old models have rapidly been replaced by new models that often cost no more but have memory sizes and computational speeds twice or three times as large as the old models. The steady progress of technical innovation has allowed the price of computer services to decline at an estimated annual rate of 18.4 percent from 1954 to 1965 and of 41.6 percent from 1970 through 1975.~9 The technical advance described in the computer example is sometimes called "costless quality change," a misnomer that appears to violate the economic law that there is no such thing as a free lunch. Most such
The Measurement of Output 103 improvements are not costless, in the sense that they cannot be obtained instantly without cost: rather they are the result of a combination of research and investment and learning-by-doing. At any moment the value of services provided by goods is limited not only by available resource inputs but by current knowledge and the extent of accumulated experi- ence. As time passes, more knowledge from research and additional learning experience permit an increase in output relative to nonresearch inputs. The increased quantity of services produced by the interactive process of research investment and learning-by-doing should be included in GNP for three major reasons. First, the computer manufacturer has invested its own funds in research and development, using a substantial number of employees; the productivi- ty of these research employees would be measured by the increase in real output of computers if the declining price of computer services were adequately measured. To treat the price of computers as unchanged implies that the research effort is wasted, because there is no extra real GNP to show for all the labor hours invested in product improvement in the period during which the computers are produced. The increase in GNP will not show until the effect of computers on the production of consumer goods and services is measured in future periods. Second, industry analysts attempting to explain inter-industry differences in productivity need to measure accurately each industry's input of labor hours and capital services. To understand why industry A outperforms industry B. one needs to know the calculation power of the computers with which their employees work. The resource-cost approach to the measurement of quality change in effect treats as identical two computer boxes having widely differing computational abilities and hence cannot explain why an employee working with a new computer is more productive than an employee working with an old one. Third, business analysts trying to correct company balance sheets for the effects of price change need to know that the replacement cost of a company's 10-year-old computer is less than the original purchase price entered in the company's historical cost-depreciation records. In the same way, the replacement cost of many types of equipment for which inflation has overwhelmed the benefits of technical advance is higher than the original purchase price. But whether the price has gone up or down, the point is the same: analysts need to know how much it would cost to replace an old capital good with one that performs the same services for the company; they have no interest in learning whether technical improvements were achieved by redesign or by a resource-using quality change.
104 THE CURRENT APPROACH OF GOVERNMEI!iT AGENCIES REPORT OF THE PANEL The BES adjustments for quality change at present are limited to type 1, and quality changes of types 2 and 3 are ignored (Bureau of Labor Statistics 1977b): Also, new technology sometimes results in better quality at reduced or no increase in cost. When no satisfactory value has been developed for such a change, it is ignored, and prices are compared directly. The effect of this approach is to use a resource-cost rather than user- value measure of quality change when cost and value diverge, as in the case of recent innovations in the electronics and communications industries. Although the Cal is intended to be an input index based on the concept of user valuation, the BES accepts changes in manufacturing cost as an approximation of what is theoretically appropriate for the index. As indicated in the quotation, when quality has improved but production cost has decreased, the BES ignores the cost decline because a production cost correction would change the index in the wrong direction when quality change is measured by the user-value criterion. The problem is not any disagreement between the basic philosophy of BES and that advocated in the previous section, since both accept the validity of the input approach to measuring prices, but rather the implicit belief of BES that it is infeasible in practice to measure value changes by any method other than changes in manufacturing cost. A similar belief in the infeasibility of measuring the value of quality changes when value and cost diverge appears to lie behind Denison's opposition to the correction of price indexes for type 2 quality changes. Denison (1957, p. 234) has no theoretical objection to adjusting for changes in the quality of durable capital goods on the basis of their contribution to production, because this concept of quality change is coherent and of extreme interest because all changes in real output could be traced to the responsible factor of production or to causes for which the factors were not responsible. Furthermore, it provides a measure of net capital formation which is theoretically meaningful. Why, then, is the extension to type 2 and type 3 quality change opposed? Essentially the problem is a belief that it is impossible to measure type 2 and type 3 quality changes (Denison 1957, p. 234~: Of the . . . methods discussed, the first, which values capital goods at cost, is ... both interesting and feasible.... [The method that would correct for quality changes of type 2 and type 3], which attempts to evaluate the contribution
The Measurement of Output 105 to production of each type of capital good, has great apparent appeal but this is very much dimmed by close examination and is utterly beyond any hope of utilization for reasonably accurate measurement now or in the future. Since it is [the second method] which underlies the classical concepts of"keeping capital intact" and net capital formation, this conclusion amounts to rejecting these concepts, as they have been understood in the past, as measurable entities. It is the suggestion of this paper that, for purposes of measurement, concepts conforming to method 1 be adopted and interpretation of the data be reshaped accordingly. Thus cost of production of capital goods rather than their ability to contribute to production becomes the common denominator enabling the values of different capital goods to be compared and combined. Denison's position has often been cited as supporting the current practices of the BEA, particularly its decision to ignore increases in the computation- al power of electronic computers relative to their cost by setting the deflator for the electronic computer industry permanently equal to 1.0. As Jaszi (1973, p. 203) has written, . . . quality improvements can be quantified only to the extent that they are accompanied by real cost increases. The sits reliance on estimates of the change in manufacturing cost associated with quality changes and Denison's opposition to attempts to measure quality changes of types 2 and 3 thus come down to a question of feasibility. When closely examined, however, the differences between the proponents and opponents of more comprehensive adjustments for quality change involve matters of degree. The proponents cite examples in which quality changes of types 2 and 3 have been measured and argue that these techniques can be extended to a broader range of products. The opponents cite examples in which adjustment for quality change are inherently impossible. It seems clear from these examples that the two sides choose to concentrate on different issues rather than disputing the validity of each other's claims. For instance, a proponent of broader quality adjustments points to the precipitous drop in price involved when the electronic calculator replaced the rotary electric calculator in the late 1960s and early 1970s. In this case, value to the user improved while production cost decreased, causing a divergence between the resource-cost and user-value measures of quality change. The BES, consistent with its general approach, ignored the transitional change in price from rotary electric calculators costing $1,000 to electronic calculators that initially cost $200. Those recommending inclusion of type 2 quality change would have treated the transition as involving an 80-percent reduction in price. It is difficult to see why the decision to treat the two goods as providing the same services is
106 REPORT OF THE PANEL infeasible. One may agree with Denison that we have no simple or reliable way of evaluating the new services that the electronic calculator provided greater speed and less noise. But surely the proposed approach of ignoring the user value of the additional features, of treating the old and new models as the same product, would have been conservative, and recording the transitional 80-percent price reduction would have been feasible. The only subjectivity would have been in the decision to limit the measured price reduction to 80 percent rather than a larger percentage, not in the decision to treat the price change as minus 80 percent rather than zero. Other examples may be cited of innovations that both provided new types of services and allowed the old services to be provided at lower cost. The introduction of the jet airplane improved the quality of services to the user and saved user time, types of quality change that may be difficult to measure. This invention also allowed the old service, a fixed-quality passenger seat-mile, to be produced at a much lower operating cost. This operating saving can be measured objectively, and this can be termed a conservative approach in that the value of greater speed and reduced vibration is ignored. For consumer home appliances, there have been extensive unmeasured type 3 quality changes in the form of improvements in energy efficiency (refrigerators, room air-conditioners) and reduced maintenance and repair requirements (color television sets, automobiles). Data are available to estimate the value of many of these changes in user- cost characteristics. Denison and others are correct in saying that some types of quality change cannot be measured, particularly the introduction of genuinely new characteristics or new products. But the proponents of broader quality adjustments are also correct in saying that some types of changes in the value of products relative to their resource cost, as in the examples of the electronic calculator, the jet airplane, the automobile, and home appli- ances, are susceptible to at least partial measurement. The debate thus comes down to a question of whether half a loaf is better than none: that is, should an attempt to correct for the feasible portion of quality changes of types 2 and 3 be viewed as better than ignoring all such quality changes? We take the view that many of these types of quality change can be measured, even if the invention of genuinely new characteristics or new products will always bring quality changes that cannot be measured. METHODS OF MEASURING QUALITY CHANGES The BLS implementation of adjustments for type 1 quality changes suffered from serious flaws in the past. As BES methodology has improved, there
The Measurement of Output 107 has been an increasing tendency for value changes that would have been classified as price changes 20 years ago to be classified as quality changes now, thus removing an earlier downward bias in the secular growth rates of output and productivity. This section first reviews the present methods of measuring type 1 quality changes and then turns to types 2 and 3 and evaluates the practical obstacles to a comprehensive attempt to account for all changes in quality. PROBLEMS IN MEASURING TYPE 1 QUALITY CHANGES Lateness in Introducing New Products A general phenomenon that affects the accuracy of all price measurement programs, both current and prospective, is the product-price cycle. In the early stages of product development, the price of many products declines gradually, as a result of the improved efficiency of workers through learning-by-doing and of the spreading of fixed cost (both research and administrative) over a larger number of units. Price reductions generate increased sales, which engender a "virtuous circle" of further economies of scale, cost reductions, and price reductions. Eventually a point of diminishing returns is reached at which learning-by-doing and economies of scale no longer yield substantial cost reductions. At this point, prices begin to be pushed up in response to general economy-wide inflation. Ideally, an aggregate price index would weight the price of each component product by its current output so that the price experience of products on the downward-sloping and upward-sloping sections of the product-price curve would be included in proportion to their relative importance in current production. Unfortunately, the current Cal and PA are Laspeyres indexes, which weight each product in proportion to its production in some past period. Until early 1978 the weights of individual products in the Cal depended on the share of each product in consumer expenditures in 1963, leading to an overrepresentation of mature products in the upward-sloping portion of the product-price cycle and an under- representation of new products in the downward-sloping portion. Transistors and integrated circuits are classic examples of products that have a learning curve that allows prices to be reduced as manufacturing technology is developed and the percentage of defective items falls. In 1964, Gordon E. Moore, then director of research at Fairchild, predicted that the complexity of the integrated circuit would continue to double every year. Even in 1977, with circuits containing 262,144 (or 2~) elements available, no departure had yet been witnessed from "Moore's
108 REPORT OF THE PANEL law," although any economist would predict that diminishing returns will begin at some future time (see Noyce 1977~. In addition to the underweighting of new products inherent in any Laspeyres (fixed-weight) price index, the cat and the PA suffer from another problem. The BES begins to include products in its price indexes relatively late in their life cycle, thus often missing entirely the early stages of price decline and including only the later stages of price increase. Even such common products as 10-key adding machines and fluorescent light bulbs did not make their way into the PP~ until 1961, although they were in common use during the 1950s. Assembled electronic products did not begin to be incorporated in the PP~ until the late 1960s. The problem of the late introduction of products might seem to be absent in the deflators used by BEA to convert nominal expenditures into the real output numerator of productivity ratios because BEA'S procedures allow the weight of a major category, e.g., food in the consumer expenditures or engines and turbines in the producers' equipment, to shift in response to changes in current expenditures. However, the deflators for these major categories are made up of individual product-price indexes combined using fixed Laspeyres weights. Thus the engines and turbines category as a whole is given an appropriate weight, but the price deflator for this category tends to be made up of obsolete types of engines and turbines on the upward-sloping section of the product-price curve, with too little representation of new types of engines and turbines on the downward-sloping section of the curve. Obsolete Models Between the late 1940s and the late 1960s the price of electricity remained almost constant. Economists studying the electric utility industry have found that the price of electric generation equipment, primarily steam generators and turbines, remained almost constant over the same period per kilowatt of generating capacity. Over the period 1947-1970, however, the PP~ index for turbine generating sets increased by 144 percent. Why? The PP~ priced only a single model of an obsolete size, 4OsOOO kW, while the size actually purchased and installed increased steadily throughout the postwar era, until in the late 1960s the average size had reached 400~000 kW (see U.S. Federal Power Commission 1969~. The larger units in which manufacturers had invested their research expenditures were apparently ignored. Although it might appear to be extremely costly for the BES to obtain current price information on new models in all product categories, mail
The Measurement of Output 109 order catalogs provide an inexpensive source of information on current shifts in the mix of models offered by manufacturers for many types of consumer and producer goods. In contrast to the BES price indexes, which often keep the same models for a number of years, mail-order catalogs change their mix of models once each year or even more often. In one recent study, an annual series of prices was developed for a number of durable goods from mail-order catalog data. Each price series eliminated the effect of changes in quality by comparing in adjacent years only models with identical specifications. Between 1947 and 1957, the catalog index increased at an annual rate 2.6 percent less than apt indexes for the same products; between 1957 and 1970, the catalog index increased 0.5 percent less.20 Although questions may be asked about the closeness of the correspon- dence between the specifications of the art products and catalog products, the slower price increases of the catalog products is quite pervasive across product classes. These differences might indicate that true catalog prices were declining relative to the true prices charged by the manufacturers reporting to the Pal, but another interpretation is that the BES indexes had an upward bias due to their reliance on obsolete models. This second interpretation of bias in the BES indexes has an important implication for productivity measurement; the apparent finding that the bias was greater in the early postwar years than in the more recent period indicates that the actual growth of output and productivity may have decelerated between the 1947-1957 and 1957-1970 periods as compared with the growth rates recorded in the official series. This implication is valid only if the apparent finding of greater bias in the earlier years, based on research for a relatively small segment of economic activity, can also be assumed to be valid for the economy as a whole. Because BES procedures are continually improving, it is likely that the bias in price measurement is decreasing. This could also lead to understatement in the deceleration of measured productivity growth in the 1970s. For instance, new procedures in the compilation of the Cal have broadened the specifications of individual products in order to allow field agents greater flexibility in pricing products with the greatest sales volume at individual outlets (see Bureau of Labor Statistics 1977b). The current revision program of the PA aims by 1984 to introduce a system by which company and item samples will be revised in one fifth of the industries each year (see Early 1978), a change with a similar eject of increasing the representativeness of its product sample. It is also becoming increasingly feasible for manufacturers to submit data to BES on the prices of a whole range of models rather than a single model.
110 Change in the Prevalence of Discounting REPORT OF THE PANEL The difference in the changes of the catalog and PP~ price indexes discussed above could be partly attributable to shifts in retail markups. If average retail margins had been reduced, particularly during the early postwar period, because of the spread of discounting practices, mail-order firms would have been forced to cut their prices to remain competitive. While the PP~ would have remained an accurate indicator of prices charged by manufacturers, the PA indexes would have been inappropriate as deflators for purchases of durable goods from retailers (and from wholesalers to the extent that their margins were reduced as well).2i The cut may have missed some of this shrinkage of margins because it prices a fixed (or slowly changing) sample of stores; its sample was probably overweighted with older high-margin stores relative to the newer discount operations. A recent study of transaction prices of major appliances during 1947-1970 yields a more rapid decline in price than indicated either by the Cal or by an index of mail-order catalog prices.22 Omitted Products In addition to the absence of adjustments for quality change of types 2 and 3, the BES price index program does not cover some important categories of producers' equipment. The BEA must therefore use inappropriate price indexes as deflators in these categories. An important example is the deflation of aircraft, which is based not on a price index for aircraft but on an average of four unrelated price indexes: those for fabricated structural metal products, metalworking machinery, electrical machinery, and diesel engines. The same is true for ships and boats. There is also no PA component for commercial refrigeration or laundry equipment; indexes for household appliances must be used instead. In the absence of the missing information, of course, it is impossible to determine what bias, if any, is imparted to aggregate measures of price and productivity change. But it is clear that the BEA indexes completely missed the quality improvement associated with the introduction of commercial jet aircraft. By far the most important component of GNP that is not covered by the BES price index program is the output of structures. The BEA deflators for structures rely on indexes from a variety of government agencies and private contractors; of these the most important is the Census Bureau's index of the prices of single-family homes based on a hedonic regression analysis of actual sales price data. The mysterious decline in the absolute level of measured productivity in the construction industry that occurred in the 1967-1972 period appears to account for a substantial share of the
The Measurement of Output declining growth rate of aggregate productivity in that period. A careful study of construction prices and the adequacy of the present deflators should be part of a larger research project needed to identify the possible role of measurement error as a source of the experience of the construction industry. Except for construction, most of the problems of measuring type 1 quality change, as reviewed in this section, are being solved by the improved methods of selecting items and sales outlets incorporated into the latest BES revision programs. The importance of this section relates mainly to the effects of a shrinking measurement bias on historical trends in productivity growth. This leaves the omission of quality change of types 2 and 3 as the main potential source of error in the current price indexes.23 PROBLEMS IN MEASURING TYPE 2 QUALITY CHANGES The extension of quality adjustments to changes of types 2 and 3 does not involve any revolution in existing techniques of measurement but rather different criteria for the application of existing techniques. The basic method of specification pricing used to adjust for resource-using quality changes can be extended to "costless" quality changes. Instead of regarding as equivalent a new and an old model having the same resource cost, BES would regard as equivalent a new and an old model having the same value to the user. If a new model or product replaces an old one but performs the same services for the user (as in the case of the transition from rotary electric calculators to electronic calculators or from piston to jet aircraft), any price reduction involved in the transition would be included in the BES index rather than ignored as at present. If the services perfo~ed by the new model are improved, treating the new and the old models as equivalent will understate the true price reduction. The feasibility of placing a dollar value on the improvement in the service will diner in each individual case. The hedonic regression technique can be useful in measuring any type of quality change except those involving the introduction of products with genuinely new characteristics. It is already used by the Census Bureau to adjust for type 1 quality changes in newly produced single-family houses. The hedonic technique is particularly useful when a product has many attributes and provides a mixture of services. The technique involves applying a statistical regression to the differing available models of a given type of product available in two or more years. The dependent variable in the regression equation is the price of each model, and the independent variables are its measured characteristics and a dummy variable for each year. For automobiles, the product most often studied to date, differences
I12 REPORT OF THE PANEL in prices among auto models available in two adjacent years are explained by size and performance attributes (length, weight, horsepower) and the model's year of production. If 1979 models are higher priced than 1978 models, holding constant their other characteristics, then the estimated coefficient on the year of production will indicate the average increase in quality-adjusted price between 1978 and 1979.24 If the independent variables in the regression equation are the performance attributes of value to the user, then the hedonic technique corrects for quality changes of types 1 and 2 simultaneously. For instance, an improvement in the acceleration of an automobile might be achieved by a resource-using increase in the size of the engine, or by a "costless" replacement of a conventional ignition system by an electronic ignition system made possible by technical innovation. It is important to recognize that the hedonic technique only provides a method of estimating an aggregate measure of performance for a product having many different performance attributes. In principle, the hedonic technique accomplishes nothing that could not be achieved by the conventional specification pricing technique since the latter could compare the price in two years of two models having identical performance attributes; in practice, the hedonic technique is useful for measuring quality change of both types 1 and 2 in the cases of some complex products (autos, refrigerators) because models available in two adjacent years are rarely completely identical. It is sometimes argued that type 2 quality adjustments are not feasible on the grounds that they are inherently subjective and that results may be sensitive to the particular methodology used and the particular perfor- mance characteristics chosen. Yet ambiguity in the form of a range of results from alternative specifications is not a unique feature of attempts to identify type 2 quality change. It also occurs for type 1 quality change. The BEA has accepted the usefulness of the hedonic approach by adopting the Census Bureau's index for the prices of single-family houses; this index is no more or less subjective than the hedonic indexes for the prices of computer services cited above. Nor is the present BES specification method free from subjectivity. Just as hedonic results can vary depending on the particular way in which observations are combined and equations are specified, so are the results of the specification method sensitive to the number of models included and the degree of attention given to adjustments for quality change. In making type 1 quality changes by estimating the cost of new product attributes, manufacturers must make arbitrary judgments on the allocation of joint costs common to several products or attributes. Many of the changes urged here involve applying the common-sense view that electronic calculators provide the same type of service as the older rotary electric
The Measurement of Output ~3 varieties or involve relatively routine applications of well-established econometric methodology whose results may be less controversial and less subject to debate than some decisions already made in the process of measuring type 1 quality change.25 Difficult questions are raised in all areas of national income accounting, e.g., the determination of a proxy for output in such hard-to-measure sectors as health, education, finance, insurance, real estate, and services. There is no reason why a particular class of problems involved in the measurement of type 2 and type 3 quality change should be arbitrarily excluded. A fundamental guideline of national income accounting since its inception has been that products are aggregated using market prices in order to reflect the assumption that the ratios of market prices of two goods measure the ratios of their marginal value to users. In the same way, the fact that users pay more for computers with faster computation speeds or for pocket-size electronic calculators containing a memory should indicate that these attributes are valuable to them. Any technological advance that produces more of these attributes should be reflected as an . . lIlCreaSe ill GNP. PROBLEMS IN MEASURING TYPE 3 QUALITY CHANGES The contribution of durable goods to output should depend on their value to users, which in turn depends on the net stream of benefits they provide: that is, the gross flow of services adjusted for operating costs. A firm should be willing to pay exactly the same extra amount for a new model of a machine that provides $1,000 more per week in revenue than an old machine and costs the same to operate or for one that provides the same revenue per week as the old machine but costs $1,0~)0 less per week to operate. Present BES procedures occasionally capture quality changes of the latter type when a new model is introduced for which a manufacturer can provide an estimate of the extra production cost of providing the buyer with a saving in operating cost. But many improvements in operating efficiency have been introduced on all models simultaneously and have been ignored. Although the improvements in fuel economy on new model automobiles since 1974 have been widely publicized, it is less well known that automobile fuel efficiency (adjusted for changes in the quality of automobiles) improved by 30 percent between 1949 and 1969 and then declined by about half that amount between 1969 and 1974 (see Wilcox 1978~. Consumers benefited during most of the postwar period from substantial reductions in electricity consumption and repair expenses on television sets, air-conditioners, and other home appliances. Similarly,
114 REPORT OF THE PANEL technical improvements achieved a steady reduction in the fuel require- ments of steam turbine generators during most of the postwar period (see Gordon 1974, Chapter 6~. Users should be indifferent to a choice between receiving a $100 price reduction on a durable good and receiving $100 in the form of a reduction in the present value of future operating costs. The basic information required for type 3 quality adjustments may be easier for manufacturers to provide than the estimates of costs of model changes currently required for type 1 adjustments. Changes in fuel and energy requirements can be reported by firms or independent testing organizations. The conversion of these changes in operating cost into a present value (for comparison with the capital cost of the durable good) requires information on average product service lives, future utilization rates, and future prices of energy and other inputs, as well as an assumed interest rate for the purposes of discounting future values. While some arbitrariness is required to choose values for service lives, expected utilization, expected energy prices, and interest rates, this is not different in principle from the difficulty faced by any private business firm or government agency that must calculate future rates of return on private and public investment projects. Fuel and energy are not the only costs of operating a durable good. Design changes that alter maintenance requirements should also be taken into account. The reduced frequency of maintenance and downtime for motor vehicles, electricity generating equipment, color television sets, and other products has been documented, although the information required to adjust for such savings is inherently more difficult to obtain and interpret than engineering data on fuel consumption. Improvements in scheduled maintenance requirements must be evaluated against the claim that assembly defects and recalls have become more common, at least for motor vehicles. Most of the discussion so far has referred to changes in operating costs or durability made possible by manufacturing design changes, but some changes should not be attributed to the industry that manufactures the product. For instance, automobile tires have become more durable over the years partly because of improvements in the quality of road surfaces. The durability of some goods or the costs of repairing them may also be improved by new techniques introduced by users or hired maintenance workers. Careful measurement procedures under constant operating conditions are required to avoid crediting a manufacturing industry with productivity improvements for which it is not responsible. Most new procedures for measuring quality change can be introduced into the price series used to deflate output only after a period of research and experimentation. Introducing changes in the deflators without an
The Measurement of Output ~5 adequate basis in research would not solve the problems we have been . . c Discussing. On the other hand, the present policy of omitting all type 2 and type 3 adjustments represents a defect in official measures of output and productivity growth. If the importance of type 2 change has grown through the increased share in GNP of innovative electronic products and if the importance of type 3 change has increased since the oil price increase of 1974, then part of the slowdown in secular productivity growth observed in the official data in the 1970s may be spurious. Some attempt should be made to quantify those effects. In seeking to measure type 2 and type 3 quality change, statistical agencies may want to maintain alternative deflators for a period before deciding to change (or not to change) those now in use. At times, they may want to publish revised or alternative historical deflators for past periods for use in studies of long-run growth. The development of new or experimental real output measures could be based on price adjustments suggested by outside researchers not only academic researchers but also experts in the BES price research section and in other parts of the federal government. Freed from the need to adhere rigidly to the "type 1 only" criterion of quality adjustment, government price analysts could range more widely than at present both in their sources of data and their techniques of adjustment. Three examples of adjustments that could be implemented quickly at relatively low cost are improved linking, measures of improved performance, and present-value calculations of energy savings. Improved Linking As mentioned above, in the early 1970s the PP~ index for calculating machines was shifted from rotary electric calculators, which then cost about $1,000, to electronic calculators, which then cost about $200, to perform identical operations more rapidly. The old index for rotary electrics was linked to the new index for electronic calculators, so that the PA reflected no price change rather than the 80-percent price decrease that actually occurred. Using the criterion that the old and new varieties should have been treated as the same product because they provided the same services to the user, the 80-percent price reduction could be retrospectively incorporated into an alternative historical price index. Another example of the same phenomenon occurred for diesel engines in the 1950s and 1960s, when technical improvements in high- speed engines allowed many users to shift from low-speed to high-speed engines having a much lower price per horsepower. The alternative deflator could incorporate the effect of the reduction in price per
116 REPORT OF THE PANEL horsepower instead of being based on the present fixed-weight combination of the two types of engines.26 Measures of Improved Performance Given the growing importance of the electronics industry, the most prominent electronics products should be priced on a basis of price per weighted performance attribute. The studies cited above demonstrate that the hedonic regression technique can be used to establish a value for the major performance attributes of electronic computers and turbine generators; the same methodology can be applied to calculators, cash registers, and other products. Previous studies should be used (or new ones commissioned) to provide information on the extent of type 2 improvements for major products during the past three decades of rapid technical progress, especially in electronics. Present-Value Calculations of Energy Savings Probably the most impor- tant adjustments required for productivity analysis are those that reflect changes in energy consumption. A substantial portion of durable goods have been subject to changes (both improvement and deterioration) in energy use including electrical generating equipment, motor vehicles, home appliances, and commercial aircraft. Studies of changes in energy use could be commissioned as a collaborative effort of the BES price research section and the government agencies responsible for estimating future energy demand and the benefits and costs of environmental regulation. It should be noted that many changes in fuel efficiency occurred before the 1974 increase in oil prices. The three kinds of adjustments discussed above involve objective measurement and can be made on a scientific basis. They leave unmeasured, however, quality changes in goods or services that have yielded time savings to users. Such changes include the faster speed of jet aircraft, the electronic calculator, the reduction of scheduled maintenance requirements of motor vehicles, and the provision of free check-cashing services by supermarkets. Given the wide range of estimates of the value of household time that has emerged from recent studies of transportation demand, adjustments of prices for time savings should perhaps be left, at least temporarily, as a subject for further research outside government. It should also be recognized that some other types of improvements can probably never be valued, for instance, the convenience value provided by a typewriter that allows a user quickly to switch typing elements for esoteric symbols and a variety of type faces or the value of a calculator that computes mathematical functions more accurately than a slide rule.
The Measurement of Output IMPORTANCE OF QUALITY ADJUSTMENTS FOR PRODUCTIVITY MEASUREMENT 117 The use of the example of the electronic computer in the discussion above of type 2 quality change might tempt a skeptic to dismiss the importance of the adjustment problem on the grounds that electronic computers represent only a small portion of GNP. If we leave aside services and structures as sectors that raise fundamental questions about the meaning of output, we can attempt to provide a rough estimate of the fraction of goods production for which unmeasured quality change may be important. Unmeasured quality change has probably been limited to a subset of durable goods: motor vehicles and household appliances and other equipment. Among the types of consumer durables that have probably not been subject to unmeasured quality change are jewelry, furniture, kitchen utensils, and books. Many categories of nondurable goods are generally agreed to be relatively free of unmeasured quality change, including most types of food, beverages, shoes, toilet articles, publications, flowers, and petroleum products. Quality changes in the form of new and more durable fabrics have been common in the clothing and semidurable household furnishing categories (with a 1976 value of $74.7 billion), and quality improvements have also occurred in the drug ($10.7 billion) and tobacco ($16.2 billion) categories.27 In the producers' durable equipment (PDE) category, unmeasured quality change is likely to have been relatively more important than in the consumer goods category. Quality adjustments are probably now relatively accurate in the "other" category, which includes furniture, fabricated metal products, agricultural machinery, construction machinery, and . . . mmlug mac emery. Our conclusion, reached by applying 1976 expenditure data to the categories of goods discussed above, is that roughly two fifths of the total production of goods may be susceptible to improved techniques of quality measurement. If unmeasured quality change does turn out to be quantitatively significant for a number of these categories, then aggregate measures of output and productivity change would be altered substantially by the proposed adjustments. Some of the categories are amenable to relatively objective and straightforward adjustments, for example, for the improved performance of electronic computers and energy consumption of vehicles and appliances. But other major categories, particularly consumer nondurables, would require difficult and speculative adjustments to correct for changes for such characteristics as fabric durability and maintenance requirements.
118 SUMMARY AND RECOMMENDATION REPORT OF THE PANEL This chapter has reviewed two sets of issues that bear on the accuracy of output measurement for productivity analysis. First, which components of the aggregate output of goods and services should be included in the numerator or the aggregate productivity ratio? Second, are changes in product quality treated adequately in the national accounts, and what steps should be taken to improve their treatment? The basic answer to the first question is that the present measure of aggregate productivity in the private economy is based on an output concept that is somewhat too inclusive; it contains a relatively small component of activities for which an estimate of productivity change is not meaningful. An analysis of the second question leads to a recommendation that government statistical agencies expand their efforts to deal more adequately with the problem of quality change. Output included in the numerator of the productivity ratio must be produced by the inputs included in the denominator. The basic measure of labor in the denominator is market-oriented labor input. As in the present national accounts, activities produced by non-market-oriented labor input should be excluded from the output universe. Because they are produced by market-oriented labor input, the production of national defense and public safety, along with commuting and other regrettables, should not be excluded from the output universe. The requirement that output be produced by market-oriented labor input is a necessary but not sufficient condition for inclusion in the numerator of the productivity ratio. A second criterion is that output and input be independently measured in a given industry or activity. On this criterion, the present official measure of productivity uses somewhat too inclusive an output concept: its universe includes a number of activities in which output is directly or partly measured by labor input (much of finance, insurance, and a portion of services). About 5 percent of the present output universe is questionable on these grounds. A particular candidate for separate examination is the construction industry, for which the deflation of output has always posed difficult problems and which bears a disproportionate share of the responsibility for the measured deceleration in aggregate productivity in the early part of the past decade. Real output is now measured by deflating the current value of output by price deflators that do not adequately measure all types of quality change. The deflators are quite successful in capturing quality changes that result from new product features produced at additional cost, but they do not attempt to capture quality changes produced by innovations that do not
The Measurement of Output ~9 raise costs or quality changes associated with the lower operating costs or longer useful life of durable goods. Recommendation 4. The Panel recommends that the Bureau of Economic Analysis (BEA) and the Bureau of Labor Statistics (BES) seek to improve their existing price indexes and to develop auxiliary measures of price change. These new auxiliary measures should take into account more adequately the types of quality change that are not now measured. They should be a collaborative effort of BEA and the productivity and price research divisions of BES. Among the adjustments that could be incorporated in the new measures are adjustments (through better linking procedures) for the changes in value to users resulting from the introduction of improved products; estimates of the value to users of improvements in performance that are achieved without increases in real cost; and estimates of the present value of future savings in operating efficiency made possible by design changes and improvements. Since many of the proposed adjustments would require background studies and additional research, they could not all be incorporated simultaneously into a single new output measure to be used for productivity analysis. Instead, the output measure could be revised periodically, perhaps every 5 years, as new research evidence becomes available on the importance of unmeasured quality change both in the current and in previous periods. NOTES 1. The productivity ratio for any input can also change when changes in the relative prices of inputs cause substitutions among them. 2. For further discussion on welfare-oriented output measures, see Moss (in this volume). 3. Such an index would capture the cyclical change in productivity within the automobile industry but would omit the compositional effect of the shift in demand toward the auto industry. 4. The data for "structures" in Table 5-1 reflect a rough estimate by the Bureau of Economic Analysis that 10 percent of construction expenditures are deflated by input-cost price indexes. This estimate may be low, depending on exactly how several of the nonresidential construction deflators are classified. 5. It is not known why an input-cost price index erroneously might yield a productivity decline over the past decade. One conjecture is that there has been a shift in the composition of construction output toward non-union sectors and low- wage regions, which would tend to reduce nominal income originating in the construction industry. If this income series is deflated by an input-cost index based
120 REPORT OF THE PANEL on an index of a fixed mix of wage rates (a Laspeyres index), the deflated real output series would show a spurious decline in real output and productivity. This conjecture can explain only part of the productivity decline, since input-cost measures are used to deflate only part of construction activity. 6. As an example of the current controversy on the productivity aspects of class size, see New York Times (1978~. 7. Scott (in this volume) discusses the measurement of the output of hospitals. 8. Scott (in this volume) poses the question of whether an outcome or process measure for the output of hospitals should be used. Process and outcome measures are not necessarily substitutes (although the former can be useful in analyzing the latter). 9. While the current treatment of government-mandated devices makes sense for productivity measurement, it has the unfortunate consequence of excluding from the price indexes important changes in specifications for which consumers are not generally willing to pay and which they perceive as increasing price rather than quality. There may be a need for an auxiliary price deflator that treats the introduction of government-mandated devices as representing price rather than quality changes for use when that approach is appropriate. The present concept should be retained for the measurement of real GNP and productivity. 10. The annual subgroup deflators for consumption expenditures are published in the July 1978 issue of Survey of Current Business (hereafter referred to as Survey 1978) for the four most recent years (Table 7.12~; historical data back to 1929 can be found in Survey (1976~. 11. The annual subgroup deflators for producers' durable equipment are also published in Survey (1978, Table 7.14) for the four most recent years; historical data back to 1929 can be found in Survey (1976~. 12. Detailed sources are provided in Survey (1974~. 13. The hedonic regression technique has been used by econometricians as a systematic method for measuring the value of added size or other features previously available on at least some models. But, as Jaszi (1962) perceptively points out, the hedonic regression method is not a fundamental break from the traditional method. Rather it is a way of identifying the determinants of price differences across models and of averaging out the contribution to price of added length, weight, and other such attributes. 14. Some commentators have been skeptical of adjustments for type 3 quality changes because of the danger, for example, that shifts in fuel use in response to changes in energy prices might erroneously be treated as attributable to equipment manufacturers. The main measurement task is to identify shifts in the transforma- tion function linking energy efficiency to the performance characteristics of goods. Changes in relative prices simply cause movements along a given function and are not treated as quality change, whereas shifts in the function are legitimately treated as a quality change. For a detailed presentation of this position see Gordon (1979~. 15. An early argument for measuring type 2 quality change is in Griliches (1964~. Defenses of the current practice of limiting measurement to type 1 quality change include those of Jaszi (1962) and Denison (1957~. 16. For an important survey of theoretical and empirical issues in the measurement of prices, see Triplett (1976~. Correspondence with Jack E. Triplett has been helpful in the drafting of this section.
The Measurement of Output 121 17. For the purpose of growth accounting, especially for large aggregates like the private business sector, one may want to keep measures of type 2 and type 3 quality change separate from the capital input measure. This is because these changes are intimately related to advances in knowledge, and it is important to be able to distinguish the effects of these advances from those that flow from savings and investment behavior given the stock of knowledge (see Chapters 3 and 7~. 18. GNP consists of consumption, investment, government purchases of goods and services, and net exports. The statements in the text about consumption and investment apply equally to government purchases. If a manufacturer can improve a machine gun by an innovation that raises the accuracy of its sight without an increase in cost, there has been an increase in the value of that machine gun to its ultimate user the taxpayer who is purchasing defense services. Whether the value of that improvement can actually be measured is another question (see discussion below). 19. See Chow (1967) and Archibald and Reece (1978~. The Archibald-Reece result must be regarded as tentative; its implied 93.2-percent price decline over only 5 years needs to be confirmed by additional research. 20. These figures are the unweighted averages of the difference between the growth rates of the catalog and PP~ indexes for 62 individual product classes (see Gordon 1974~. 21. Deflators for the producers' durable equipment category of GNP are based entirely on the PPi, yet many small businesses purchase equipment from retailers and wholesalers. 22. See Gordon (1974, Chapter 6~. The prices are those paid by Consumers Union shoppers for refrigerators, washing machines, and room air-conditioners. 23. In principle, the price indexes can contain an upward or downward bias because of the measurement of quality change; Triplett (1976) presents examples that go in both directions. 24. For a brief survey of the more important hedonic regression studies of automobile prices, see Triplett (1976, pp. 3~41~. The Triplett paper also reviews studies of quality change, both using the hedonic regression method and other techniques, for a number of other products. 25. For example, the decision to treat safety and antipollution devices on automobiles as a change in quality rather than price and the decision to ignore all quality changes involved in the "downsizing" of automobiles in the period since 1976 are arguable. 26. This reduction in price per horsepower is shown in Census Bureau unit- value data for diesel engines and has been confirmed by conversations with engine manufacturers (see Gordon 1974, Chapter 5~. 27. The changes in tobacco refer to the reduction in average tar and nicotine levels. Different users may evaluate these changes in different ways.