Computers Can Accelerate Productivity Growth
The evidence can be found by comparing computer-intensive industrial sectors with other sectors.
Conventional wisdom argues that rapid change in information technology over the past 20 years represents a paradigm shift, one perhaps as important as that caused by the electric dynamo near the turn of the century. The world market for information technology grew at nearly twice the rate of world gross domestic product (GDP) between 1987 and 1994, so the computer revolution is clearly a global phenomenon.
Yet measured productivity growth has been sluggish in the midst of this worldwide technology boom. In the United States, for example, annual labor productivity growth (defined as output per hour of work) actually fell, from 3.4 percent between 1948 and 1973 to 1.2 percent between 1979 and 1997. For the period 1979 to 1994, total factor productivity (TFP) growth (defined as output per unit of all production inputs) also fell substantially, from 2.2 percent to 0.3 percent per year. In light of the belief that computers have fundamentally improved the production process, this is particularly puzzling. As Nobel laureate Robert M. Solow has observed, “You can see the computer age everywhere but in the productivity statistics.”
Detailed analysis of the U.S. economy suggests that computers have had an impact, but it is necessary to look beyond the economy-wide numbers in order to find it. New technology affects each business sector differently. For most sectors, the computer revolution is mainly a story of substitution. Companies respond to the declining price of computers by investing in them rather than in more expensive inputs such as labor, materials, or other forms of capital. The eight sectors that use computers most intensively, for example, added computers at a rate of nearly 20 percent per year from 1973 to 1991, whereas labor hours grew less than 3 percent per year. This capital deepening (defined as providing employees with more capital to work with) dramatically increased the relative labor productivity of the computer-using sectors, those with more than 4 percent of total capital input in the form of computers in 1991.
Before 1973, labor productivity in the manufacturing sectors that invested heavily in computers grew only 2.8 percent per year, compared with 3.1 percent for those that did not. After they accumulated computers so rapidly in the 1970s and 1980s, however, labor productivity growth jumped to 5.7 percent per year for 1990-1996 in the computer-using sectors in manufacturing, but declined to 2.6 percent per year for the other manufacturing sectors. Comparison of the relative performance of these sectors over time shows that computers are playing an important role in determining labor productivity.
Computer-related productivity gains in the manufacturing sectors also suggest that measurement errors have been a large obstacle to understanding the economy-wide impact of computers on productivity. Computer investment is highly concentrated in service sectors, but in those sectors there is no clear evidence of the dramatic productivity gains found in manufacturing. Economists, however, have long argued that both output and productivity are poorly measured in service sectors. If one conjectures that the true impact of computers is approximately the same in both manufacturing and services, these results imply an increasing understatement of output and productivity growth in the service sectors.
The computer-producing sector reveals yet another way in which the computer revolution affects economy-wide productivity growth. This sector experienced extraordinary TFP growth of nearly 3 percent per year in the 1980s, reflecting the enormous technological progress that enabled computer companies to churn out superior computers at lower and lower prices. This one sector, despite being relatively small in terms of private GDP (less than 3 percent), was responsible for one-third of TFP growth for the entire U.S. economy in the 1980s.
Moving beyond aggregate data
Computers have experienced dramatic price declines and extraordinary investment growth in the past two decades. The price of computer investment in the United States decreased at the remarkable rate of more than 17 percent per year between 1975 and 1996, whereas the broader investment category of which computers are a part, producers’ durable equipment (PDE), increased more than 2 percent per year. At the same time, and mostly in response to rapid price declines, business undertook a massive investment in computers. Starting near zero in 1975, the computer share in real PDE investment in the United States increased to more than 27 percent by 1996. With cumulative investment in new computer equipment near $500 billion for the 1990s, U.S. companies have clearly embraced the computer. Countries across the globe are also rapidly accumulating computers. Between 1987 and 1994, growth in the information technology market exceeded GDP growth in 21 out of the 24 member countries of the Organization for Economic Cooperation and Development (OECD). These figures present a compelling view of the depth and breadth of the computer revolution. From Main Street to Wall Street, computers appear everywhere, and computer chips themselves can also be found inside automobiles, telephones, and television sets.
Yet aggregate productivity growth remains flat by historic standards. And services-which are the most computer-intensive sectors–show the slowest productivity growth. This apparent inconsistency is at the heart of the computer productivity paradox.
Any attempt to explore this paradox, however, must move beyond the economy-wide data on which it is based. The aggregate data hide many illuminating details. For most companies, computers are a production input they invest in, just like new assembly lines, buildings, or employee training. Not all companies use computers the same way, however. Nor can all companies benefit from computer investment. These important differences are lost in the economy-wide data. Furthermore, computers are also an output from a particular manufacturing sector.
To explore these differences, the U.S. economy was divided into 34 private sectors and ranked according to their use of computers. Eight of these sectors use computers intensively–more than 4 percent of their capital was in the form of computers in 1991–and were labeled computer-using sectors. As shown in Table 1, these eight sectors accounted for 63 percent of total value added and 88 percent of all computer capital input in 1991.
Computers are highly concentrated within three service sectors-trade; finance, insurance, and real estate (FIRE ) and “other services,” which includes business and personal services such as software, health care, and legal services–that account for more than 75 percent of all computer inputs. In manufacturing, only five of 21 sectors used computers intensively enough to be labeled computer-using; they accounted for less than 40 percent of total manufacturing output in 1991.
Computers are not everywhere
This wide variation in computer use is evident in recent surveys of adoption of computer-based technologies. For example, in 1993 a staggering 25 percent of all manufacturing plants surveyed by the U.S. Census Bureau used none of 17 advanced technologies. Moreover, patterns of adoption varied greatly by industry and technology. In fact, very few of the surveyed technologies showed use rates greater than 50 percent, and many (particularly lasers, robots, and automated material sensors, all of which depend on computers) were used by fewer than 10 percent of surveyed plants. The most prevalent technologies are computer-aided design and numerically controlled machine systems. Virtually identical surveys in Canada and Australia confirm the diversity reported by U.S. manufacturers.
OECD surveys also show that computers are highly concentrated in specific sectors. In Canada, France, Japan, and the United Kingdom, for example, information and communication equipment is steadily increasing its share of total investment and is much more highly concentrated in the service sectors. In 1993, OECD estimates indicate that the service sector contained nearly 50 percent of all embodied information technology for the seven major industrial nations and that this capital was concentrated primarily in finance, insurance, services, and trade.
More specific data from France and Germany suggest that computers are becoming universal in some industries. Nearly 90 percent of all workers in the French bank and insurance industry used a personal computer or computer terminal in 1993. This proportion is up from 69 percent in 1987 and substantially exceeds the 30 to 40 percent in French manufacturing industries. In Germany, nearly 90 percent of surveyed companies in the service sector report that computers are important in their innovation activities.
When the price of an input falls, companies respond by substituting the cheaper input for more expensive ones. With the enormous price declines in computers, one would expect to see companies substitute less expensive computers for relatively expensive labor and other inputs. For example, companies might replace labor-intensive written records with computer-intensive electronic records. Detailed analysis of the U.S. sectoral data suggests that this is exactly what happened. The eight computer-using sectors invested in computers rapidly and substituted them for other inputs. From 1973 to 1991, these eight sectors report annual growth in real computer input in excess of 17 percent, with seven out of eight above 20 percent (see Table 1).
When compared with the growth rates of labor and output in these sectors, the swift accumulation of computers appears even more striking. In contrast to the phenomenal growth rates of computer capital, labor hours declined in three sectors and experienced growth rates above 3 percent in only two. Similarly, output growth ranged from -0.4 percent to 4.7 percent per year. Moreover, substituting computers for other inputs is not limited to these computer-intensive sectors; the phenomenon is observed in virtually every sector of the U.S. economy.
Several independent company-level studies from the United Kingdom, Japan, and France also suggest that an important part of the computer revolution is substitution of inputs. The French study, for example, found a strong positive relationship between the proportion of computer-users and output per hour. A survey of Japanese manufacturing and distribution companies finds that information networks complement white-collar jobs but substitute for blue-collar jobs.
Rather than looking at empirical relationships between computers, productivity, and employment patterns, the Australian Bureau of Statistics used a more subjective, although still informative, approach. In a 1991 survey of manufacturing companies, nearly 50 percent rated lower labor costs as a “very important” reason for introducing new technology. A 1994 follow-up study found that almost 25 percent of the companies cited reducing labor costs as a “very significant” or “crucial” objective in technological innovation. These survey results offer still more evidence that companies expect high-tech capital to substitute for other production inputs.
These results suggest that a large part of the computer revolution entails substitution of one production input (computers) for others (labor and other types of capital). But is this just wheel-spinning? The answer depends on how productivity is defined and measured and what one means by wheel-spinning. Economists use two distinct concepts of productivity: average labor productivity (ALP) and total factor productivity (TFP). Although these concepts are related, they cannot be used interchangeably, and TFP is the productivity measure most favored by economists when analyzing the production process.
ALP is defined simply as output per hour worked. A major advantage of this measure is computational; both output and labor input statistics are relatively easy to obtain. Since the 1930s, however, economists have recognized that labor is only one of many production inputs and that labor’s access to other inputs, especially physical capital, is a key determinant of ALP. That is, when their labor is augmented by more machines and better equipment, workers can produce more. This increase in output need not reflect harder work or improved efficiency but is simply due to increases in the complementary inputs available to the labor force.
This key insight led to the concept of TFP, defined as output per unit of total inputs. Rather than calculating output per unit of labor as in ALP, TFP compares output to a composite index of all inputs (labor, physical capital, land, energy, intermediate materials, and purchased services, augmented with quality improvements), where different inputs are weighted by their relative cost shares.
Increased TFP has often been interpreted as technological progress, but it more accurately reflects all factors that generate additional output from the same inputs. New technology is a key source of TFP growth, but so are economies of scale, managerial skill, and changes in the organization of production. Furthermore, technological progress can be embodied, at least in part, in new investment.
ALP and TFP are fundamentally different concepts although TFP is an important determinant of ALP. ALP grows–that is, each worker can produce more–if workers have more or better machinery to work with (capital deepening), if workers become more skilled (labor quality), or if the entire production process improves (TFP growth).
Despite the connection between these two concepts, the trend toward greater use of computers implies different things for each measure of productivity. If investment in computer capital is primarily for input substitution, then ALP should increase as labor is supported by more capital. TFP, however, will not be affected directly; it will increase only if computers increase output more than through their direct impact as a capital input. It is this more-than-proportional increase that many analysts have in mind when they argue that increased investment in computers should result in higher productivity.
It is easier to define these productivity statistics than to measure and apply them. There is a growing consensus among economists that both output growth and productivity growth are poorly measured, especially in the fast-growing service sectors with a high concentration of computers. This measurement problem is part of a more fundamental issue concerning output growth and quality change. Most economists agree that quality improvements are an important form of output growth that need to be measured. The U.S. Bureau of Economic Analysis (BEA) officially measures the enormous quality change in computer equipment as output growth. Based on joint work with IBM, BEA now uses sophisticated statistical techniques to create “constant-quality price indexes” that track the price of relevant characteristics (such as processor speed and memory). These price indexes allow BEA to measure the production of real computing power and count that as output growth. Thus, the quality-adjusted price of computer equipment has fallen at extraordinary rates, while real computer investment has rapidly grown as a share of total investment in business equipment.
For other sectors of the economy, however, output is harder to define and measure. In the FIRE sector, for example, BEA extrapolates official output growth for banks based on employment growth so that labor productivity is constant by definition. Yet most would argue that innovations such as ATMs and online banking have increased the quality of bank services. Because difficulties of this type are concentrated in all service sectors, output and productivity estimates in those sectors must be interpreted with caution.
Labor productivity growth
In the early 1970s, the industrial world experienced a major growth slowdown in terms of aggregate output, ALP, and TFP. Economists have offered many possible reasons for this slowdown–the breakdown of the Bretton Woods currency arrangements, the energy crisis, an increase in regulation, a return to normalcy after the unique period of the 1950s and 1960s, and an increase in the share of unmeasured output–but a clear consensus has not yet emerged. Because the computer revolution began in the midst of this global slowdown, untangling the relationship between computers and productivity growth is particularly difficult. For example, does the drop in U.S. ALP growth from 3.4 percent (1948 to 1973) to 1.2 percent (1973 to 1996) mean that computers lowered ALP growth? Or would the slowdown have been much worse had the computer revolution never taken place? Without the proper counterfactual comparison–what productivity growth would have been without computers–it is difficult to identify the true impact of computers.
Our approach to that problem is to compare ALP growth in the computer-using sectors with the non-users in the 34-sector database before and after the slowdown period. Chart 1 compares growth rates of average labor productivity for five computer-using sectors in manufacturing and 16 other manufacturing sectors for 1960 to 1996. For the early period of 1960 to 1973, labor productivity growth was roughly the same for the two groups-2.8 percent per year for computer-using sectors and 3.1 percent per year for non-computer-using sectors. Both groups then suffered during the much-publicized productivity slowdown in the 1970s as ALP growth rates fell to about 1.5 percent per year during the period 1973 to 1979.
As the computer continued to evolve and proliferate in the 1980s, businesses adapted and their production processes changed. Personal computers-first classified as a separate investment good in 1982-became the dominant form of computer investment, and ALP growth accelerated in the computer-using sectors in manufacturing. Between 1990 and 1996, these sectors posted strong ALP growth of 5.7 percent per year, whereas other manufacturing sectors managed only 2.6 percent per year. Because ALP growth for computer-using sectors prior to the 1970s was lower than in other manufacturing sectors, this analysis strongly suggests that computers are having an important impact on labor productivity growth in U.S. manufacturing.
The same comparison for nonmanufacturing sectors yields quite different results, with no obvious ALP gains for computer-using sectors outside of manufacturing (Chart 2). Rather, the 3 computer-using sectors and the 10 non-computer-using sectors show healthy productivity growth prior to 1973, but sluggish productivity growth thereafter: Labor productivity grew only 0.9 percent per year for computer-using sectors in nonmanufacturing between 1990 and 1996, and 0.8 percent for other nonmanufacturing sectors.
The sharp contrast in productivity growth in computer-using sectors in manufacturing and in services highlights the difficulties associated with productivity measurement. Economists have long argued that output and productivity growth are understated in the service sectors due to the intangible nature of services, unmeasured quality change, and poor data. These results support that conjecture and further imply that measurement problems are becoming more severe in the computer-intensive service sectors. This suggests that much of what computers do in the service sectors is not being captured in the official productivity numbers.
Although measurement errors probably understate output and productivity growth in the computer-intensive service sectors, this does not change the finding of significant input substitution. In the trade and FIRE sectors, for example, the growth of labor slowed while computer inputs increased more than 20 percent per year from 1973 to 1991. Because capital and labor inputs are measured independently of service-sector output, this type of primary input substitution is not subject to the same downward bias as is TFP growth. Whatever the true rates of output and TFP growth, these service sectors are clearly substituting cheap computers for more expensive inputs.
Variation in growth
Estimates of TFP growth for each of the 34 sectors demonstrate no relationship between it and the growth of computer use. TFP grew in some sectors, fell in others, and stayed about the same in others, but there was no obvious pattern relating TFP growth to computer use. Nor was there any relationship evident for just the eight computer-using sectors (see Table 1). These findings suggest that, in contrast to increases in ALP, there have been few TFP gains from the widespread adoption of computers.
Many consider this disappointing. Learning lags, adjustment costs, and measurement error have been suggested as reasons for a slow impact of computers on TFP growth. It is important to remember, however, that this finding is entirely consistent with the evidence on input substitution. If computer users are simply substituting one production input for another, then this reflects capital deepening, not TFP growth. Recall that TFP grows only if workers produce more output from the same inputs. If investment in new computers allows the production of entirely new types of output (for example, complex derivatives in the financial services industry), the new products are directly attributable to the new computer inputs, not to TFP growth.
This conclusion partly reflects BEA’s explicit adjustment for the improved quality of computers and other inputs, but most economists agree that quality change is an important component of capital accumulation. That is, when computer investment is deflated with BEA’s official constant-quality price deflator, the enormous improvement in the performance of computers is then folded into the estimates of computer capital. Quality improvements are effectively measured as more capital, so capital becomes a more important source of growth, and the TFP residual accounts for a smaller proportion of output growth.
So far this analysis has focused on the role of computers as an input to the production process. But computers are also an output; companies produce computers and sell them as investment and intermediate goods to other sectors and as consumption and export goods. Because the observed input substitution in computer-using sectors is driven by rapid price declines for computer equipment, it is important to examine the production of computers themselves and investigate the source of that price decline.
The data show that TFP is the primary source of growth for the computer-producing sector and a major contributor to the modest TFP revival in the U.S. economy, particularly in manufacturing. From 1979 to 1991, virtually the entire growth in output in the computer-producing sector is attributable to TFP growth; that is, output grew much faster than inputs and caused a large TFP residual. In fact, output grew 2.3 percent per year even though labor, energy, and material inputs actually declined. The computer-producing sector is itself also an important user of computers; nearly 40 percent of the growth in output attributable to capital services comes from computer capital over this same period.
Rapid growth in TFP in the computer-producing sector contrasts with sluggish TFP growth in the entire U.S. private-business economy, which fell from more than 1.6 percent per year before 1973 to -0.3 percent for the period from 1973 to 1979. Even the computer-producing sector showed negative TFP growth in that period. After 1979, however, the story is very different. While annual TFP growth for the 35 sectors rebounded mildly to 0.3 percent per year, TFP growth in the computer-producing sector jumped to 2.2 percent for 1979 to 1991.
The aggregate economy consists, by definition, of its sector components. How much of economy-wide TFP growth reflects TFP growth from the computer-producing sector? In the 1980s, it was as much as one-third of total TFP growth. In the 1990s, TFP growth in the sector remained high, but because there were increases in TFP growth in other manufacturing sectors, it accounted for less of the total, about 20 percent between 1991 and 1994.
Recent estimates of TFP growth for manufacturing industries confirm these trends. Of the 20 manufacturing sectors analyzed by the Bureau of Labor Statistics (BLS), “industrial and commercial machinery,” where computers are produced, showed the most rapid annual TFP growth: 3.4 percent per year from 1990 to 1993. Total manufacturing, on the other hand, showed TFP growth of just 1.2 percent for the same period. Although these estimates are not directly comparable to those derived from the 35-sector database, they confirm the importance of the computer-producing sector in economy-wide TFP growth.
Given the substantial work by BEA on computer prices, real output growth in the computer-producing sector is probably among the best measured. Thus, the estimates of rapid output and TFP growth in the computer-producing sector appear sound. Furthermore, these results support the conventional wisdom that computers are more powerful, affordable, and widespread than ever. Recent work at BEA, however, suggests that constant-quality price indexes should also be used for other production inputs. If the quality of these other inputs such as semiconductors is improving rapidly but costing less, TFP growth will be overstated in the sectors that use these inputs and understated in the sector that produces them. This kind of mismeasurement primarily affects the allocation of TFP among sectors, not the economy-wide total TFP.
The substitution of computers for other, more expensive, inputs goes a long way toward explaining the computer paradox. The impact of computers is observable not in TFP, as many observers perhaps expected, but in the accumulated stock of computer capital. This explains why, despite the pickup in labor productivity growth after 1979, economy-wide TFP growth has remained low. For most sectors, computers are a measured input that contributes directly to economic growth. Rapid TFP growth occurs primarily in the computer-producing sector, where faster, better computers are continually offered at ever-lower prices. This reflects fundamental technological advances that are driving the computer revolution and makes a substantial contribution to economy-wide TFP growth.
Moreover, there is little indication that this growth will slow. BLS, for example, projects that labor productivity growth in the computer and office equipment industry will accelerate to 9.9 percent per year through 2005. If these projections are correct and companies continue to substitute relatively inexpensive computers for costlier older models, computers will become an increasingly important source of economic growth.