While the evidence indicates that IT has created substantial value for firms that have invested in it, it has been a challenge to link these benefits to macroeconomic performance. A major reason for the gap in interpretation is that traditional growth accounting techniques focus on the (relatively) observable aspects of output, like price and quantity, while neglecting the intangible benefits of improved quality, new products, customer service, and speed. Similarly, traditional techniques focus on the relatively observable aspects of investment, such as the price and quantity of computer hardware in the economy, and neglect the much larger intangible investments in developing complementary new products, services, markets, business processes, and worker skills. Paradoxically, while computers have vastly improved the ability to collect and analyze data on almost any aspect of the economy, the current computer-enabled economy has become increasingly difficult to measure using conventional methods. Nonetheless, standard growth accounting techniques provide a useful benchmark for the contribution of IT to economic growth.
Studies of the contribution of IT concluded that technical progress in computers contributed roughly 0.3 percentage points per year to real output growth when data from the 1970s and 1980s were used (Jorgenson and Stiroh 1995, Oliner and Sichel 1994, Brynjolfsson 1996).
Much of the estimated growth contribution comes directly from the large qualityadjusted price declines in the computer-producing industries. The nominal value of purchases of IT hardware in the United States in 1997 was about 1.4 percent of GDP. Since the quality-adjusted prices of computers decline by about 25 percent per year, simply spending the same nominal share of GDP as in previous years represents an annual productivity increase for the real GDP of 0.3 percentage points (that is, 1.4 .25 = .35). A related approach is to look at the effect of IT on the GDP deflator. Reductions in inflation, for a given amount of growth in output, imply proportionately higher real growth and, when divided by a measure of inputs, for higher productivity growth as well. Gordon (1998, p.4) calculates that "computer hardware is currently contributing to a reduction of U.S. inflation at an annual rate of almost 0.5 percent per year, and this number would climb toward one percent per year if a broader definition of IT, including telecommunications equipment, were used".
More recent growth-accounting analyses by the same authors have linked the recent surge in measured productivity in the U.S. to increased investments in IT. Using similar methods as in their earlier studies, Oliner and Sichel (2000) and Jorgenson and Stiroh (1999) find that the annual contribution of computers to output growth in the second half of the 1990s is closer to 1.0 or 1.1 percentage points per year. Gordon (2000) makes a similar estimate. This is a large contribution for any single technology, although researchers have raised concerns that computers are primarily an intermediate input and that the productivity gains are disproportionately visible in computer-producing industries as opposed to computer-using industries. For instance, Gordon notes that after he makes adjustments for the business cycle, capital deepening and other effects, there has been virtually no change in the rate of productivity growth outside of the durable goods sector. Jorgenson and Stiroh ascribe a larger contribution to computer-using industries, but still not as great as in the computer-producing industries.
Should we be disappointed by the productivity performance of the downstream firms?
Not necessarily. Two points are worth bearing in mind when comparing upstream and downstream sectors. First, the allocation of productivity depends on the qualityadjusted transfer prices used. If a high deflator is applied, the upstream sectors get credited with more output and productivity in the national accounts, but the downstream firms get charged with using more inputs and thus have less productivity. Conversely, a low deflator allocates more of the gains to the downstream sector. In both cases, the increases in the total productivity of the economy are, by definition, identical. Since it is difficult to compute accurate deflators for complex, rapidly changing intermediate goods like computers, one must be careful in interpreting the allocation of productivity across producers and users.
The second point is more semantic. Arguably, downstream sectors are delivering on the IT revolution by simply maintaining levels of measured total factor productivity growth in the presence of dramatic changes in the costs, nature, and mix of intermediate computer goods. This reflects a success in costlessly converting technological innovations into real output that benefits end consumers. If "mutual insurance" maintains a constant nominal IT budget in the face of 50 percent IT price declines over two years, it is treated in the national accounts as using 100 percent more real IT input for production. A commensurate increase in real output is required merely to maintain the same measured productivity level as before. This is not necessarily automatic since it requires a significant change in the input mix and organization of production. In the presence of adjustment costs and imperfect output measures, one might reasonably have expected measured productivity to initially decline in downstream sectors as they absorb a rapidly changing set of inputs and introduce new products and services.
Regardless of how the productivity benefits are allocated, these studies show that a substantial part of the upturn in measured productivity of the economy as a whole can be linked to increased real investments in computer hardware and declines in their quality-adjusted prices. However, there are several key assumptions implicit in economy or industry-wide growth accounting approaches which can have a substantial influence on their results, especially if one seeks to know whether investment in computers is increasing productivity as much as alternate possible investments. The standard growth accounting approach begins by assuming that all inputs earn "normal" rates of return. Unexpected windfalls, whether the discovery of a single new oil field, or the invention of a new process which makes oil fields obsolete, show up not in the growth contribution of inputs but as changes in the multifactor productivity residual. By construction, an input can contribute more to output in these analyses only by growing rapidly, not by having an unusually high net rate of return.
Changes in multifactor productivity growth, in turn, depend on accurate measures of final output. However, nominal output is affected by whether firm expenditures are expensed, and therefore deducted from value-added, or capitalized and treated as investment. As emphasized throughout this paper, IT is only a small fraction of a much larger complementary system of tangible and intangible assets. However, current statistics typically treat the accumulation of intangible capital assets, such as new business processes, new production systems, and new skills, as expenses rather than as investments. This leads to a lower level of measured output in periods of net capital accumulation. Second, current output statistics disproportionately miss many of the gains that IT has brought to consumers such as variety, speed, and convenience. We will consider these issues in turn.
The magnitude of investment in intangible assets associated with computerization may be large. Analyses of 800 large firms by Brynjolfsson and Yang (1997) suggest that the ratio of intangible assets to IT assets may be ten to one. Thus, the $167 billion in computer capital recorded in the U.S. national accounts in 1996 may have actually been only the tip of an iceberg of $1.67 trillion of IT-related complementary assets in the United States.
Examination of individual IT projects indicates that the 10:1 ratio may even be an underestimate in many cases. For example, a survey of enterprise resource planning projects found that the average spending on computer hardware accounted for less than 4 percent of the typical start-up cost of $20.5 million, while software licenses and development were another 16 percent of total costs (Gormley et al. 1998). The remaining costs included hiring outside and internal consultants to help design new business processes and to train workers in the use of the system. The time of existing employees, including top managers, that went into the overall implementation was not included, although it too is typically quite substantial.
The up-front costs were almost all expensed by the companies undertaking the implementation projects. However, insofar as the managers who made these expenditures expected them to pay for themselves only over several years, the non-recurring costs are properly thought of as investments, not expenses, when considering the impact on economic growth. In essence, the managers were adding to the nation's capital stock not only of easily visible computers, but also of less visible business processes and worker skills.
How might these measurement problems affect economic growth and productivity calculations? In a steady state, it makes little difference, because the amount of new organizational investment in any given year is offset by the "depreciation" of organizational investments in previous years. The net change in capital stock is zero. Thus, in a steady state, classifying organizational investments as expenses does not bias overall output growth as long as it is done consistently from year to year. However, the economy has hardly been in a steady state with respect to computers and their complements. Instead, the U.S. economy has been rapidly adding to its stock of both types of capital. To the extent that this net capital accumulation has not been counted as part of output, output and output growth have been underestimated.
The software industry offers a useful example of the impact of classifying a category of spending as expense or investment. Historically, efforts on software development have been treated as expenses, but recently the government has begun recognizing that software is an intangible capital asset. Software investment by U.S. businesses and governments grew from $10 billion in 1979 to $159 billion in 1998 (Parket and Grimm 2000). Properly accounting for this investment has added 0.15 to 0.20 percentage points to the average annual growth rate of real GDP in the 1990s. While capitalizing software is an important improvement in our national accounts, software is far from the only, or even most important, complement to computers.
If the wide array of intangible capital costs associated with computers were treated as investments rather than expenses, the results would be striking. According to some preliminary estimates from Yang (2000), building on estimates of the intangible asset stock derived from stock market valuations of computers, the true growth rate of U.S. GDP, after accounting for the intangible complements to IT hardware, has been increasingly underestimated by an average of over 1 percent per year since the early 1980s, with the underestimate getting worse over time as net IT investment has grown. Productivity growth has been underestimated by a similar amount. This reflects the large net increase in intangible assets of the U.S. economy associated with the computerization that was discussed earlier. Over time, the economy earns returns on past investment, converting it back into consumption. This has the effect of raising GDP growth as conventionally measured by a commensurate amount even if the "true" GDP growth remains unchanged.
While the quantity of intangible assets associated with IT is difficult to estimate precisely, the central lesson is that these complementary changes are significant and cannot be ignored in any realistic attempt to estimate the overall economic contributions of IT.
The productivity gains from investments in new IT are underestimated in a second major way: failure to account fully for quality change in consumable outputs. It is typically much easier to count the number of units produced than to assess intrinsic quality—especially if the desired quality may vary across customers. A significant fraction of value of quality improvements due to investments in IT—like greater timeliness, customization, and customer service—is not directly reflected as increased industry sales, and thus is implicitly treated as nonexistent in official economic statistics.
These issues have always been a concern in the estimation of the true rate of inflation and the real output of the U.S. economy (Boskin et al. 1997). If output mismeasurement for computers were similar to output mismeasurement for previous technologies, estimates of long-term productivity trends would be unaffected (Baily and Gordon 1988). However, there is evidence that in several specific ways, computers are associated with an increasing degree of mismeasurement that is likely to lead to increasing underestimates of productivity and economic growth.
The production of intangible outputs is an important consideration for IT investments whether in the form of new products or improvements in existing products. Based on a series of surveys of information services managers conducted in 1993, 1995, and 1996, Brynjolfsson and Hitt (1997) found that customer service and sometimes other aspects of intangible output (specifically quality, convenience, and timeliness) ranked higher than cost savings as the motivation for investments in information services. Brooke (1992) found that IT was also associated with increases in product variety.
Indeed, government data show many inexplicable changes in productivity, especially in the sectors where output is poorly measured and where changes in quality may be especially important (Griliches 1994). Moreover, simply removing anomalous industries from the aggregate productivity growth calculation can change the estimate of U.S. productivity growth by 0.5 percent or more (Corrado and Slifman 1999). The problems with measuring quality change and true output growth are illustrated by selected industry-level productivity growth data over different time periods, shown in table 4.2.According to official government statistics, a bank today is only about 80 percent as productive as a bank in 1977; a health care facility is only 70 percent as productive and a lawyer only 65 percent as productive as they were 1977.
Source: Partial reproduction from Gordon (1998, Table 3).
These statistics seem out of touch with reality. In 1977, all banking was conducted at the teller windows; today, customers can access a network of 139,000 ATMs 24 hours a day, 7 days a week (Osterberg and Sterk 1997), as well as a vastly expanded array of banking services via the Internet. The more than tripling of cash availability via ATMs required an incremental investment on the order of $10 billion compared with over $70 billion invested in physical bank branches. Computer-controlled medical equipment has facilitated more successful and less invasive medical treatment. Many procedures that previously required extensive hospital stays can now be performed on an outpatient basis; instead of surgical procedures, many medical tests now use non-invasive imaging devices such as x-rays, MRI, or CT scanners. Information technology has supported the research and analysis that has led to these advances plus a wide array of improvements in medication and outpatient therapies. A lawyer today can access much wider range of information through online databases and manage many more legal documents. In addition, some basic legal services, such as drafting a simple will, can now be performed without a lawyer using inexpensive software packages such as Willmaker.
One of the most important types of unmeasured benefits arises from new goods. Sales of new goods are measured in the GDP statistics as part of nominal output, although this does not capture the new consumer surplus generated by such goods, which causes them to be preferred over old goods. Moreover, the Bureau of Labor Statistics has often failed to incorporate new goods into price indices until many years after their introduction; for example, it did not incorporate the VCR into the consumer price index until 1987, about a decade after they began selling in volume. This leads the price index to miss the rapid decline in price that many new goods experience early in their product cycle. As a result, the inflation statistics overstate the true rise in the cost of living, and when the nominal GDP figures are adjusted using that price index, the real rate of output growth is understated (Boskin et al. 1997). The problem extends beyond new high tech products, like personal digital assistants and handheld Web browsers. Computers enable more new goods to be developed, produced, and managed in all industries. For instance, the number of new products introduced in supermarkets has grown from 1,281 in 1964, to 1,831 in 1975, and then to 16,790 in 1992 (Nakamura 1997); the data management requirements to handle so many products would have overwhelmed the computerless supermarket of earlier decades. Consumers have voted with their pocketbooks for the stores with greater product variety.
This collection of results suggests that IT may be associated with increases in the intangible component of output, including variety, customer convenience, and service. Because it appears that the amount of unmeasured output value is increasing with computerization, this measurement problem not only creates an underestimate of output level, but also errors in measurement of output and productivity growth when compared with earlier time periods which had a smaller bias due to intangible outputs.
Just as the Bureau of Economic Analysis successfully reclassified many software expenses as investments and is making quality adjustments, perhaps we will also find ways to measure the investment component of spending on intangible organizational capital and to make appropriate adjustments for the value of all gains attributable to improved quality, variety, convenience, and service. Unfortunately, addressing these problems can be difficult even for single firms and products, and the complexity and number of judgments required to address them at the macroeconomic level is extremely high. Moreover, because of the increasing service component of all industries (even basic manufacturing), which entails product and service innovation and intangible investments, these problems cannot be easily solved by focusing on a limited number of "hard to measure" industries—they are pervasive throughout the economy.
Meanwhile, however, firm-level studies can overcome some of the difficulties in assessing the productivity gains from IT. For example, it is considerably easier at the firm level to make reasonable estimates of the investments in intangible organizational capital and to observe changes in organizations, while it is harder to formulate useful rules for measuring such investment at the macroeconomic level.
Firm-level studies may be less subject to aggregation error when firms make different levels of investments in computers and thus could have different capabilities for producing higher value products (Brynjolfsson and Hitt 1996, 2000). Suppose a firm invests in IT to improve product quality and consumers recognize and value these benefits. If other firms do not make similar investments, any difference in quality will lead to differences in the equilibrium product prices that each firm can charge. When an analysis is conducted across firms, variation in quality will contribute to differences in output and productivity, and thus will be measured as increases in the output elasticity of computers. However, when firms with high quality products and firms with low quality products are combined together in industry data (and subjected to the same quality-adjusted deflator for the industry), both the IT investment and the difference in revenue will average out, and a lower correlation between IT and (measured) output will be detected. Interestingly, Siegel (1997) found that the measured effect of computers on productivity was substantially increased when he used a structural equation framework to directly model the errors in production input measurement in industry-level data.
However, firm-level data can be an insecure way to capture the social gains from improved product quality. For example, not all price differences reflect differences in product or service quality. When price differences are due to differences in market power that are not related to consumer preferences, then firm-level data will lead to inaccurate estimates of the productivity effects of IT. Similarly, increases in quality or variety (e.g., new product introductions in supermarkets) can be a byproduct of anti-competitive product differentiation strategies, which may or may not increase total welfare. Moreover, firm-level data will not fully capture the value of quality improvements or other intangible benefits if these benefits are ubiquitous across an industry, because then there will not be any inter-firm variation in quality and prices. Instead, competition will pass the gains on to consumers. In this case, firm-level data will also understate the contribution of IT investment to social welfare.
It is worth noting that if the exact quality change of an intermediate good is mismeasured, then the total productivity of the economy is not affected, only the allocation between sectors. However, if computer-using industries take advantage of the radical change in input costs and quality to introduce new quality levels (or entirely new goods) and these changes are not fully reflected in final output deflators, then total productivity will be affected. In periods of rapid technological change, both phenomena are common.