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.
Uses, Misuses, and Limitations of Productivity Statistics . THE NATURE OF PRODUCTIVITY MEASURES Productivity measures are a family of related measures that are useful for a variety of purposes. The choice of a particular member of the family will usually depend on the particular use to which it is being put. The nature of productivity measures is examined in detail in the chapters that follow; this chapter discusses the principal kinds of uses that are made of these measures to furnish a background against which they can be examined. Broadly defined, productivity measures include all measures that relate one or more measured inputs to a measure of output usually in the form of a ratio. Thus, we can measure the yield of a cornfield in bushels per acre, the fuel economy of a car in miles per gallon, or the average labor productivity of a shoe factory in pairs of shoes per employee-hour, and all of these are productivity measures. We are concerned in this report both with measures that use only the basic factors of production- labor, capital, and lanai as inputs and with measures that also use intermediate products or services as inputs. It should be made clear at the outset that changes in productivity measures based on a single input cannot necessarily be attributed solely to that particular input. For example, one cannot infer from a rise in output per employee-hour that employees are more skilled or that they are working harder than they were in the previous period: either or both may be the source of the rise in productivity in whole or in part, but need not be. The entire rise could be attributable to an increase in capital inputs, to 19
20 REPORT OF THE PANEL higher rates of capacity utilization, or to technological change. It would even be possible to have a technological change that raised output per employee-hour that, at the same time, reduced the level of skill or difficulty of work required of employees. The inappropriateness of the contrary inference that is sometimes made can be seen easily with examples of different kinds of productivity measures: one would not be inclined to attribute a rise in the yield of corn in bushels per acre entirely, if at all, to an improvement in the quality of land or to attribute an improvement in automobile fuel economy measured in miles per gallon entirely to an improvement in the quality of gasoline. The Bureau of Labor Statistics (BES), in the U.S. Department of Labor, often takes pains to warn its readers against possible misinterpretation of productivity measures. BES bulletins on productivity routinely include the following (or similar) sentences: Although these measures relate output to employment and employee hours, they do not measure the specific contribution of labor, capital, or any other factor of production. Rather, they reflect the joint effect of a number of interrelated influences, such as changes in technology; capital investment per worker; level of output; utilization of capacity; layout and flow of material; managerial skill; and skills and effort of the work force. However, no such cautions appear in the quarterly press releases on productivity and costs or in the notes to the tables on productivity in BES'S Monthly Labor Review. Nor do such cautions appear when BES productivi- ty series are reproduced in the publications of other government agencies. These omissions may contribute to the frequent misinterpretation of productivity measures in the press. For example, an Associated Press story of April 25, 1978, reporting a decline in output per hour of all persons in the private business sector in the first quarter of 1978 began: "American workers' productivity fell 3.6 percent at an annual rate in the first four months of 1978." Quite apart from the error of giving the length of a calendar quarter as 4 months instead of 3, the language seems to attribute the decline in measured productivity to workers. Readers might well infer incorrectly that workers were less energetic or diligent in the first quarter of 1978 than previously. Recommendation 1. The Panel recommends that the Bureau of Labor Statistics give more prominence in its publications and press releases to cautionary statements warning against misinterpretation of its output-per-hour measures.
Uses, Misuses, and Limitations of Productivity Statistics Our examples so far of output and input have involved simple physical measures. Such measures are best viewed as approximations of more appropriate measures that distinguish quality as well as quantity. For example, a ton of steam coal is less valuable than a ton of metallurgical coal: it will therefore pay to mine metallurgical coal in thinner seams, where output in tons per hour (though not in value per hour) will be lower. Frequently, the output of an establishment or industry is too diverse to measure at all simply by counting or weighing. Such cases are dealt with by constructing a weighted measure of different kinds of output, using values or labor requirements at a base date as weights. Inputs also are sometimes difficult to measure. The problems of measuring output are discussed in Chapter 5 and those of measuring inputs in Chapter 6. The measured economic output of goods and services that enter into productivity measures, even when measured without error, are not always the ultimate outcomes that are of greatest concern. We sometimes lack measures that are directly related to ultimate satisfaction or utility: for example, we may be forced to measure the output of physicians' services by the number of office visits by patients or procedures performed by physicians, when we would prefer to measure the contribution of physicians' services to improvements in health or reductions in morbidity or mortality. However, the available measures may still be useful for improving managerial efficiency in physicians' offices. Shortcomings of economic data in measuring ultimate satisfaction, of course, are not confined to productivity statistics. They are common in all measures of economic activity. It should also be noted that production processes often have undesired effects in addition to the desired one of producing economic output. These undesired effects include the production of substances that pollute air and water, the creation of excessive noise or heat, and the presence of conditions that impair the health and safety of workers. For some purposes, it might be desirable to subtract estimates of these "negative outputs" from the measured positive outputs in calculating productivity changes or making productivity comparisons. 21 PRODUCTIVITY MEASURES AS INDEXES OF PERFORMANCE It is hard to imagine any use for a single measure of productivity. If we are told that the output per employee-hour of a certain coal mine in 1976 was one ton, we have learned very little unless we also know something about the level of this measure in some other year or the output per employee
22 REPORT OF THE PANEL hour of other coal mines. Productivity measures exist largely to be compared we are interested almost exclusively in productivity differences or productivity changes. In this context, productivity measures can be thought of as indicators of relative efficiency in production the efficiency with which inputs are converted into economic output. If we learn that output per employee- hour in a certain coal mine has been declining, we could regard that information as cause for concern. We might ask such questions as: Has the management become lax? Are there new employees who are less skilled than the average employee? Is the mine now working thinner coal seams? Similarly, if output per employee-hour in one mine is lower than that in another, we could ask such a question as whether the equipment in the less productive mine is less modern. The use of productivity measures as measures of performance designed to highlight possible problems and to stimulate action if needed is most direct when data are gathered at the establishment or divisional level within a firm. A firm having several different establishments producing the same product could compare productivity data across establishments in an effort to extend the best practice of production to all of its establishments and thus to raise the average productivity of the firm. Such extension of best practice may not be easy, since it often involves changes in technology requiring new capital equipment. A single-establishment firm could sometimes compare its own measured productivity with that of its industry, but this comparison would be meaningful only in industries with homogeneous products or after appropriate allowance for differences in product mix. When the establishments within an industry produce different products or have different product mixes, simple comparisons will not usually be very instructive. Comparisons of productivity designed to stimulate action are often more meaningful if they involve both levels and rates of change. Thus a firm that discovered that its productivity was rising less rapidly than the industry average would have much more cause for concern if it began the period substantially below the average level for the industry than if it began substantially above. Among the productivity comparisons of great interest are international comparisons, which are considered further in Chapter 9. Perhaps the most basic source of interest in such comparisons is the fact that international differences in productivity play a major role in explaining differences in the standard of living or in per-capita income. Other sources of such differences in living standards would include differences among countries in endowments of natural resources, the size of capital stocks, and the ratio of the economically active population to the dependent population.
Uses, Misuses, and Limitations of Productivity Statistics 23 Recently, there has been particular interest in international comparisons of productivity in the production of internationally traded goods. Such productivity differences are one important determinant of the patterns of exports and imports among industrial countries. For example, if output per employee-hour in the Japanese or European steel industry is higher or is growing more rapidly than that in the U.S. steel industry, this could be one important factor leading to increased imports of steel into the United States. A full analysis of changes in imports would, of course, also have to include many other factors, such as levels of and changes in the money wages of steel workers in the different steel-producing countries, costs other than labor costs (such as capital and raw materials), exchange rates among the national currencies, transportation costs, and tariff and non- tariff barriers to trade. A recent example of the use of international productivity comparisons in this sort of analysis can be found in a report of the Council on Wage and Price Stability (1977a, p. 53) on the steel industry. As part of a comprehensive analysis of differences in the cost of steel production among the United States, Japan, and Europe, the report notes that in 1976 estimated employee-hours per metric ton of raw steel, adjusted to the U.S. product mix, were 8.2 in the United States, 9.6 in Japan, and 11.1 in the European Economic Community.2 These productivity figures were then used as the basis for deriving unit labor costs, using estimates of compensation per employee-hour converted to dollars at the prevailing rate of exchange. This yields estimates of labor costs in dollars per metric ton of raw steel of $100.24 for the United States, $60.58 for Japan, and $96.14 for the European Economic Community. International comparisons of changes in productivity are of as much interest as comparisons of levels. Table 2-1, condensed from a recent BES publication, shows estimates of changes in productivity and unit labor costs in the iron and steel industry for five countries from 1964 to 1974. The estimates indicate that over this period, productivity in the steel industry rose less rapidly in the United States than in any of the other countries except the United Kingdom. However, because the United States had the smallest increase in compensation per hour in national currency and because the value of the dollar declined relative to three of the other four currencies, the United States had the smallest increase in unit labor costs measured in dollars. (The difficulties in comparing measures among nations that have different data collection programs and different methodologies are discussed in Chapter 9.) Regional productivity data would be of similar interest in analyzing the sources of industry shifts within an economy, such as the shift of employment in the United States from the "snowbelt" to the "sunbelt."
24 REPORT OF THE PANEL TABLE 2-1 Estimated Changes in Output per Hour, Hourly Labor Costs, and Unit-Labor Costs in the Iron and Steel Industry, Five Countries, 1964 1974 (index numbers, 1964= 100) United United States Japan France Germany Kingdom Output per hour 134.2 282.9 173.0 197.2 122.0 Hourly labor costs (U.S. dollar basis) 202.1 564.7 348.0 404.7 232.7 Unit-labor costs (U.S. dollar basis) 150.6 199.6 201.2 205.2 190.7 SOURCE: Bureau of Labor Statistics (1977) Comparative Growth in Manufacturing Productivity and Labor Costs in Selected Industrialized Countnes. Bulletin 1 9 5 8. Table 12, p. 20. Washington, D.C.: U.S. Department of Labor. However, no regular regional statistics on productivity are currently published. Measures of output per unit of all inputs, when available, are in principle better guides to comparisons of efficiency than measures of output per unit of labor input alone.3 It is not always efficient, for example, for an employer to invest in labor-saving machinery. If wages are high, the price of new machinery is low, interest rates are low, and the existing machinery is near the end of its useful life, then investment in labor-saving machinery could produce large gains in efficiency. If one or more of these conditions is reversed, however, it might be more efficient to continue to use existing machinery longer. The result is that an establishment with low output per hour of labor may be more efficient than one whose labor productivity is higher, because the first establishment is economizing in the use of other inputs. PRODUCTIVITY MEASURES, LIVING STANDARDS, AND INCOME DISTRIBUTION Interest in measures of productivity changes for the economy or its major sectors arises in large part from the relation between changes in productivity and changes in the standard of living. Productivity growth is one of the most important sources of a rise in aggregate output and income, and it is by far the most important source of the rise in output and income per person employed. Denison (1974, Table 9-4) has estimated that from 1929 to 1969, 47 percent of the growth of real national income can be attributed to increases in output per unit of input rather than to the
Uses, Misuses, and Limitations of Productivity Statistics 25 increases in factor inputs. For the growth of real national income per employed person, he estimates the share of output per unit of input over the same period at 79 percent the rest being accounted for largely by increases in inputs of education and physical capital (Denison 1974, Table 9-7~. Other methods of analysis give somewhat different estimates of the contribution of productivity change to the growth of output. Jorgenson and Christensen (1973, Table 16, p. 308) estimate that 43 percent of the growth of gross private domestic product from 1929 to 1969 is due to the increase in output per unit of input. This estimate is lower than Denison's in part because the measure of output used is gross of depreciation, while Denison's is net.4 Jorgenson and his associates estimate that in the period 1969-1973 the proportion of growth of output attributable to increases in output per unit of input has fallen substantially (see Christensen et al. 1980). The relationship between changes in productivity and improvements in the standard of living suggests why an apparent slowdown in productivity growth, such as that experienced in the United States in the past decade, is a cause for concern. It implies a slower improvement in our ability to raise levels of consumption, to reduce poverty, and to enhance the quality of life. Increases in real income arising from productivity growth are generally distributed as increases in real compensation to suppliers of the factors of production, whose real compensation otherwise can change only through income redistribution. It follows from the definition of productivity that if productivity is constant, the output of goods and services can be increased only by using more units of tangible inputs. All of the added output will be needed to compensate the providers of these additional units of inputs at existing rates of compensation, with no surplus left for raising the level of real factor compensation for the providers of either the original or the added units. In these circumstances, the providers of any one input could increase their real compensation only at the expense of the providers of some other input. In contrast, if productivity is rising, the real income of the providers of some or all inputs can rise without the real incomes of others necessarily falling. Thus the salient use of productivity measures is as an indicator of the extent to which real income and the standard of living (as measured by consumption of tangible goods and services) can rise through time. To choose a particular form of the increase in real income-that in which prices are constant factor compensation per unit of input measured in current dollars can rise at the same rate as productivity without requiring any increase in the average price of output. More generally, the money
26 REPORT OF THE PANEL compensation of factors can rise faster than prices or fall more slowly by the amount of the change in productivity. This discussion can be made more explicit by focusing on a particular measure of productivity and on the compensation of one factor. Let us consider an economy in which all work is done by employees and choose as our measure of productivity the most widely used one, output (Q) per employee-hour (L), Q/L. Assume that over 2 years, output rises by 5 percent with labor input remaining constant. If compensation per employee-hour also rises 5 percent, unit labor costs will remain constant. Unit labor costs are simply the aggregate compensation of labor divided by output, or wL/Q, where w includes both money wages and fringe benefits per hour of work. In our example, Q and w have each risen by 5 percent, canceling each other out, and L has not changed. The ratio wL/Q can most easily be seen to be unit labor cost when output can be expressed in physical units. Thus, if a mine produces 40 tons of coal an hour while employing 30 workers at an average hourly compensation of $8.00, its unit labor cost is $240/40 or $6.00 a ton. If the price of output, p, is included as a factor in the denominator, the ratio becomes a measure of labor's share of the value of output. If the price of coal at the mine is $10 a ton, labor's share of output is $240/400 or 60 percent. For the same reasons as above, an equal percentage increase in compensation per hour and in productivity will not affect labor's share. With L constant, a 5-percent increase in both output and compensation would be 1.05wL/1.05Qp, and again the increases cancel. It would make no difference in the result of this calculation if the rise in productivity took place through a reduction of hours with output constant rather than through a rise in output with hours constant, or if it took place through a combination of an increase in output and a decrease in hours. Thus we see that an increase in the compensation of labor proportional to the trend of increases in the average productivity of labor has two interesting properties. First, because on average the increase in compensa- tion does not raise unit labor cost, increases in the price level are not needed to pay for it; in this sense it can be called noninflationary.5 Second, the increase in compensation does not necessarily alter the distribution of output between labor and non-labor inputs. These properties have led to the frequent use of changes in the average productivity of labor in the private business economy as one standard for judging the appropriateness of wage changes.6 The first major use of this sort in the United States appears to be in the 1948 and 1950 collective bargaining agreements between the General Motors Corporation and the United Automobile Workers (UA W). The 1950 agreement incorporated a wage increase of 2.5 percent in each year of its 5-year term, a figure that
Uses, Misuses, and Limitations of Productivity Statistics 27 was said to be based on the average increase in productivity in the American economy over the preceding 50 years (see Harbison 1950, p. 4~00~. Since the agreement also included a cost-of-living escalator clause, real rather than money wages were being related to productivity trends. The annual improvement factor has remained a prominent feature of General Motors-uAw contracts since 1950 and has been extended to many other firms that bargain with the UAW. Other unions and industries have included similar provisions in multi-year agreements without giving them the same name or an explicit rationale. A more prominent use of the average productivity concept as a guide for wages may be found in the income policies of several national administra- tions since 1960.7 The first important instance of such a use occurs in the wage-price guideposts of the Kennedy and Johnson administrations as set forth in the Economic Report of the President, 1962 (U.S. President 1962~. The report states (p. 189~: The general guide for noninflationary wage behavior is that the rate of increase in wage rates (including fringe benefits) in each industry be equal to the trend rate of over-all productivity increases. General acceptance of this guide would maintain stability of labor cost per unit of output for the economy as a whol~though not of course for individual industries. Nothing was said about cost-of-living increases in wages so that the guideposts applied to money rather than to real compensation. This is presumably because changes in consumer prices at the time were very small. (By 1967, when the wage guideposts were dropped from the Economic Report, the consumer price index had been rising faster than the trend of productivity, and the problem of price changes could no longer be ignored.) The 1962 Economic Report also pointed out circumstances in which one would want to make exceptions to a policy of tying particular wage changes to national productivity trends. Some of these related to shortages and surpluses of labor in particular geographic or occupational labor markets; others related to instances in which wages in the initial period are considered to be inequitably low or inequitably high. As noted above, acceptance of a policy that ties compensation to productivity trends implies that labor's share of output will remain constant. Historically, it has in fact risen, even after corrections to add back the estimated labor share of the incomes of proprietors and partners, which are a declining portion of total income (see Kravis 1968~. This suggests a general objection to the continuation of a productivity-based income policy for any extended period that it would tend to freeze distributive shares that would ordinarily tend to change in response to changing economic circumstances.
28 REPORT OF THE PANEL If more than one kind of labor is used in producing output, which is almost always the case, the common propositions about the relation between increases in the average productivity of labor and in compensation apply only if the mix of labor used has remained constant. If the mix is changing, increasing the compensation of each kind of labor separately by the increase in average productivity per unit of unweighted labor input could raise or lower both labor's share and unit labor cost. Suppose, for example, that in a certain establishment a journeyman can produce twice as much in a day as an apprentice and that in period one, 200 units of output per day are produced by 18 journeymen and 4 apprentices. In period two, 200 units per day are produced by 19 journeymen and 2 apprentices. A measure of output per unweighted employee day would rise because output has remained constant and employment has fallen from 22 workers to 21. But in both periods, each journeyman produced 10 units per day and each apprentice produced 5 units. In erect, a more productive worker, the journeyman, has been substituted for two less productive workers. If apprentices receive less pay than journeymen, as they ordinarily do, average daily earnings of all workers taken together will be higher in the second period, but there is no basis in productivity change for any increase in the compensation of either group separately. When one considers broad national statistics instead of one establishment, a shift of labor between industries, for example, from agriculture to manufacturing, could have the same kind of effect as a shift between occupations. One might wonder why it is seldom suggested that changes in productivity at the industry level, as contrasted with broader measures, may be related to changes in compensation. It is true by definition that if compensation in an industry rises at the same rate as output per employee- hour in that industry, unit labor costs will be constant. However, if this identity were used as a rule to govern changes in compensation in each industry separately, given the great dispersion of productivity trends among industries, the result would soon be very different pay in different industries for work of comparable difficulty requiring comparable skill. These wage differences could be unrelated to shortages or surpluses of labor in these industries. The industries with low rates of productivity growth would find it hard to recruit labor, and workers in these industries would regard their compensation as inequitably low. The industries with high productivity growth would not be able to pass on lower unit labor costs to their customers in the form of lower prices, and their growth would therefore be restricted. During the period of wage and price controls under the Economic Stabilization Act of 1970 (August 1971 to April 1974), measures of the change in output per employee-hour at the industry level, both published
Uses, Misuses, and Limitations of Productivity Statistics 29 and unpublished, were used by the Price Commission and the Cost of Living Council to convert changes in compensation into estimates of changes in unit labor cost, as a means of examining the cost justification for proposed price increases. These measures of unit labor costs were then used in establishing allowable price increases for the products of the industry. Apart from any questions about the wisdom of the controls program as economic policy, serious doubts were expressed as to whether the measures so used were of sufficient quality to justify their use for this purpose. The uses of productivity measures as a guide for setting prices and wages are easiest to document in expressions of public policy; there must be many more instances of their unrecorded use by private decision makers both price setters and wage setters including the parties to collective bargaining. Finally, we should add the important proviso that the use of productivity measures to guide the distribution of economic gains need not be confined to distributions to the providers of tangible factor inputs. Suppose that we consider a policy of improving air and water quality that would use inputs of labor and capital without increasing measured output. If we knew that some accepted measure of productivity was rising at 3 percent per year, we might decide to use 0.5 percent per year to improve air and water quality, leaving 2.5 percent for increases in conventional output per unit of input to be distributed to factor providers. If the increase in measured productivity was only 1.5 percent, we might select a slower or less costly program of improving air and water quality, or perhaps none at all. Thus productivity measures tell us not only what our opportunities are to improve the consumption of food, clothing, shelter, and transportation, but also they tell us more broadly what our opportunities are, without robbing Peter to pay Paul, to improve the quality of life. THE USE OF PRODUCTIVITY MEASURES IN FORECASTING AND BUSINESS CYCLE ANALYSIS Another major use of productivity data is in measuring and forecasting such economic magnitudes as potential gross national product (GNP) and labor requirements. Potential GNP iS defined as the output that could be produced at a high employment level.8 A recent example of the use of productivity data to measure potential GNP may be found on pages 52-57 of the Economic Report of the President, 1977 (U.S. President 1977~. Potential GNP iS estimated from trends in the growth of the labor force,
30 1,600 1 ,400 1,200 o 1,000 an o REPORT OF THE PANEL Seasonally Adjusted annual rates -. 800 _~ 1 1 New potential G NP Old potential GNP Actual GNP 1964 1966 1968 1970 1 972 1 974 1 976 YEAR FIGURE 2-1 Gross national product, actual and potential. (U.S. President  Economic Report of the President. P. 55. Washington, D.C.: U.S. Government Printing Office.) average hours of work per week, and output per labor-hour in the private business economy, together with an assumption about the unemployment rate at a high employment level. Figure 2-1, taken from the Economic Report of January 1977, shows a major downward revision in the estimate of potential GNP. The principal basis for this revision was the lower trend of labor productivity since 1968. Forecasts of potential GNP are made by extrapolating trends such as those shown in Figure 2-1. At a more detailed industry level, labor requirements measured in employee-hours could be forecast by forecasting demand and hence output, then dividing that forecast by the projected trend in output per employee-hour. Of course, even if productivity had been measured with perfect accuracy in the past, there is no assurance that its future changes will coincide with an extrapolation of the past trend. In 1966, no one could reasonably have predicted that measured productivity growth in the ensuing 10 years would fall substantially below that of the previous 10 years, though the portion of the change due to changes in the age
Uses, Misuses, and Limitations of Productivity Statistics 31 composition of the labor force might have been predicted. Forecasting productivity is beset with all the dangers that plague economic forecasting generally. The use of productivity measures in forecasting is closely related to their use in business-cycle analysis and hence in forming monetary and fiscal policy. It is well established that measured output per employee-hour tends to fall, or to increase at a slower rate, when economic activity is contracting, especially in the early stages of the contraction. Some of the more important reasons for this phenomenon are summarized below. One reason output often falls more than employee-hours in a recession is that some employees are needed as long as any output is produced. For example, a freight train may have 100 cars when traffic is heavy and 80 when traffic is light, but it needs an engineer in either case. Such factors led to a drop in railroad car miles per employee-hour from 1973 to 1975, a period of declining output. A second reason for the decline in productivity in a recession is that some employees are kept on the payroll even though they are not needed to produce current output. Some may be engaged in maintenance or in research and development; others will simply have a lighter workload. Employers engage in such hoarding of labor because if a skilled or specialized worker is let go, this worker may not be available when demand recovers and more labor is needed. A third reason for the downturn in productivity in a recession is that it takes time for employers to recognize and adjust to a decline in demand. What is eventually seen as a substantial decline may at first appear to be just a small random disturbance. Employers are reluctant to lay oh workers unless a decline is expected to persist. Once the extent of the decline is fully appreciated, labor inputs are reduced and productivity may . . rise again. There may be some offsetting factors to the tendencies just described. For example, the least skilled or least experienced workers are most likely to be laid off during a recession. It has also been suggested that workers may work harder during a recession, because when jobs are scarce they do not want to risk losing one. It is clear, however, that any such tendencies are not strong enough to reverse the effect of the factors that lead to lowered measured output per employee-hour in a recession. In a business-cycle recovery, the erects of the recession on productivity are reversed; productivity usually rises by more than the long-run trend, at least in the early stages of a recovery. A fiscal or monetary policy deigned to restore full employment after a recession would have to take this tendency into account in determining the size of the stimulus that would be appropriate.
32 REPORT OF THE PANEL ANALYZING THE SOURCES OF ECONOMIC GROWTH The use of productivity measures in analyzing the sources of economic growth is considered in detail in Chapter 7; it is mentioned briefly here. We caution the reader that the attribution of growth to particular causes is a difficult task, and estimates of this sort are subject to substantial uncertainty. Changes in productivity include all of the sources of growth except increases in the quantity of inputs. This means, in particular, that accounting for productivity change involves measuring or finding indica- tors of changes in the quality of inputs, the state of technology, and economies of scale, but such a listing does not do justice to a difficult and important subject. While some measures of change in the quality of inputs or of change in technology are readily available, others are difficult to construct. Moreover, different analysts may not agree on which measures are most relevant or appropriate. Among the well-known estimates of the sources of growth of the U.S. economy are those of Denison. In his scheme, the principal division of the sources is between factor inputs and output per unit of input, although he stresses that the detailed categories are more important than this principal division. The inputs include labor, capital, and land. Labor inputs are divided into number of employees, hours, and education plus some additional factors such as shifts in the age-sex composition of the work force. Capital inputs are divided into inventories, nonresidential structures and equipment, dwellings, and international assets. The portion of output growth not attributed to increases in inputs is the growth in output per unit of input. The major factors accounting for increases in output per unit of input are economies of scale, improved resource allocation, and "advances in knowledge," which is estimated as a residual. Other well- known estimates of sources of growth are those of Jorgenson and his associates. These apportion growth between changes in inputs and changes in output per unit of input, but do not attempt to separate the forces accounting for changes in multi-factor productivity. The purpose of summarizing these approaches here is to show that dividing the sources of growth into increases in inputs and increases in productivity is but an initial step in understanding the varied and complex sources of economic growth. Better understanding of the sources of growth is needed if one is to adopt rational policies for promoting growth.
Uses, Misuses, and Limitations of Productivity Statistics SOME LIMITATIONS OF PRODUCTIVITY MEASURES 33 Many people believe that productivity measures should be extended as rapidly as possible to all sectors of the economy. After all, work units in all sectors use resources; it can be agreed that we should be concerned about their stewardship; that we should take account of the relation of their output to their inputs. In short, we should measure their productivity. But the uncritical extension of productivity statistics to inappropriate areas is not only useless and hence a waste of resources it can also be harmful. To specify which areas are inappropriate is not easy, but two general conditions are relevant: productivity statistics are not helpful and may be harmful when there are no clear measures of output and when there is no well-understood production "technology," that is, no efficacious procedures for reliably converting inputs into output. For example, organizations such as schools, many types of government agencies, and community mental health agencies may lack clearly defined goals or demonstrably efficacious techniques for achieving them. Since they usually do not sell their output in a market, there is no simple way to ascertain the value attributed to their products or services by those who use them. Attempts to apply productivity measures to organizations lacking clear output measures can produce unanticipated and perhaps undesirable effects. To push too hard for the development of clear and measurable objectives can result in a distortion of functions such that less important but more quantifiable objectives drive out more important but diffuse goals. For example, excessive attention to the learning of specific skills in professional schools may allow less time for the development of diagnostic judgment or ethical standards. Goals may be further distorted in situations in which performers or evaluators emphasize maximizing evaluation scores rather than performance quality. For example, pressures on schools to adopt competence-based evaluation systems can lead to the develop- ment of teachers who teach and students who learn "to the tests." Many organizations lacking empirically demonstrable techniques use instead what may be termed socially validated processes. These processes, which include legal, symbolic, and ceremonial activities, depend primarily on faith or belief or trust to achieve their erects. Their efficacy depends on agreements shared by producers, consumers, and the public that the processes work. It depends on the nurturing of a shared belief in the legitimate capacity of the processes to perfo~ certain kinds of work. Thus, ministers baptize converts, lawyers draw up contracts, and schools graduate students using production processes that are, to a large extent, socially defined and validated. To attempt to measure the productivity of
34 REPORT OF THE PANEL such enterprises could sometimes lead to some rather undesirable consequences: an undermining of the confidence of both personnel and clients and, in general, a de-legitimation of the processes and products produced in this manner. Social organizations for performing useful functions come in a variety of forms and guises. Concepts and techniques that may be appropriate or stimulating to certain classes of enterprise may be unsuited for and damaging to others. While organizations like schools and welfare agencies may appear to be similar in form to industrial firms, this common appearance can be misleading. They diner greatly in their work processes. In some settings, techniques for rationalizing work and making it more efficient may undermine the social consensus required to carry it out. NOTES 1. Frequently, land and man-made capital are combined in a single measure and called simply capital. 2. The productivity measure used here, hours per ton, is the inverse of the more usual one, tons per hour. 3. Chapter 3 discusses at length methods of combining factor inputs and the relationship of multi-factor ratios to single-factor productivity measures. 4. For a full discussion of differences between the methodology of Denison and that of Jorgensen and his associates, see the special issue of Survey of Current Business (1972) 52~5) Part II. 5. This statement is not true if the change in the average productivity of labor occurs because more capital is used. In that case, capital costs per unit of output may rise and prices may have to rise to cover those costs. 6. Micro-economic theory concludes that in a competitive labor market, wages will equal the marginal product of labor (the increase in output resulting from adding one more hour of labor to the production process, other inputs remaining constant) rather than the average product of labor. Changes in marginal product will be equal to changes in average product only under special assumptions about the nature of the production process, and it has not been demonstrated that these assumptions hold true. Since there are as yet no time-series data (other than wages themselves) on changes in marginal product, however, marginal product is not a practical guide to wage policy. 7. Our discussion of these policies is not intended as an endorsement of them but simply as an account of the uses to which productivity measures have been put. 8. The concept of potential GNP, although widely used, involves certain arbitrary assumptions that may be open to question (see Denison 1974~.