It is becoming increasingly clear that the growth of computing and communications technology is exceeding our understanding of its economic and social impacts (Box 1.1).1 The processing power of microchips is doubling every 18 months. From 1989 to 1995, Internet traffic doubled every 12 months, and it is now doubling every 6 to 9 months. When plots of trends over time require a logarithmic scale on the vertical axis, something interesting must be going on! Accompanying and supporting these dramatic increases in the power and use of new information technologies has been the declining cost of communications as a result of both technological improvements and increased competition in a sector long dominated by monopoly (and in other countries often state-owned) enterprises.
At the same time, there has been comparatively little investment in research to help understand how information technology has affected and will affect our society. The United States, and indeed the world, are facing critical policy issuesinvolving intellectual property rights, privacy, free speech, education, and other crucial concernsarmed with very little understanding and analysis of the consequences of possible choices.
For several reasons, too little social science research has been done to date in this area. Certainly the phenomena associated with information technology have developed rapidly in comparison with academic calendars and funding cycles. In addition, serious investigation of a number of complex technological, economic, and social issues requires interdisciplinary work to which there are currently many barriers. It is rare, for example, to see economists and sociologists collaborating, much less sociologists and computer scientists. One of the reasons is
One of the most common findings in prior studies of information technology's (IT's) impact has been that outcomes are far from uniform across all settings and contexts. In earlier years we looked for the impact of IT on, say, organizational centralization, and scholars tended to hew to one end or the other of a bipolar spectrum: centralization versus decentralization, upskilling or deskilling, job destroying versus job creating. What scholars found, in almost every case, was that this was an unproductive way to conceptualize the issue. One almost always found evidence of both extremes of outcomes or impacts as well as many points in between (see Attewell and Rule, 1989). We finally realized that we were asking the wrong question. We should have asked, In what contexts does outcome A typically predominate, and in what contexts does outcome B tend to prevail, and when does one see A and B in equal measure?
We found that a technology does not usually have an impact. The context or setting in which the same technology is used often produces strikingly different "impacts." This phenomenon has been discussed in terms of "Web models" (Kling), or "structural contingency theory" (Attewell) or Robey's ''Plus Ca Change" model. All imply that we fully appreciate the role of context in technology outcomes and that we therefore expend sufficient research effort to measure the context, and to delineate its interactions with the technology. If we fail to do this, we return to the old "black box" paradigm, that is, attempting to measure only the input (say, a particular software program) and the outcome (say, kids' test scores) without bothering with the context (the classroom, the kids' family backgrounds) or the causal mechanisms.
Black box research on impacts often discovered "inconsistent" outcomes across studies but proved unable to show why there was so much variation, because it neglected to measure the contextual variables that were moderating the effects of the input on the output. For example, the old paradigm would phrase a research question so as to ask whether or not home PCs would improve kids' school performance. In contrast, research within the current contextual paradigm would ask under what conditions having PCs at home affects students' school outcomes. A piece of my own work has indicated, for example, that having a home PC currently has a minimal effect on the school performance scores of poor and minority kids but is associated with substantial positive effects on the school performance of kids with high socioeconomic status (SES), when other factors are controlled for (Attewell and Battle, 1997). Race and class/SES, in this example, prove to be very important contextual features moderating the impact of home PCs on school performance.
It is important to understand that because of the last three decades of research and the importance of context as discussed above, many distinguished scholars of technology avoid the term "technology impact." Using this term in framing the question would be viewed by some of them as indicating an ignorance of the body of scholarship in technology studies. For them, the term "impact" connotes a kind of technological determinism that is very dated and widely discredited. Personally, I am not so averse to the term "impact," but I do agree with their larger point about avoiding models based on simple technological determinism.
Paul Attewell, "Research on Information Technology Impacts"
(see Appendix B of this volume)
the investment of time and energy required by both technologists and social scientists to understand enough of an unfamiliar discipline area to enable serious progress in joint efforts. Also, research funding agencies and programs are generally not organized to exploit interdisciplinary opportunities.
To aid in identifying fruitful approaches to assessment of both the positive and negative impacts of using information technologies, the National Science Foundation asked the Computer Science and Telecommunications Board (CSTB) of the National Research Council to gather perspectives on the problem from experts in several relevant disciplines, such as economics, sociology, and psychology, as well as computer science and engineering. It was thought that a gathering of a group with pertinent experience as well as openness to the benefits of interdisciplinary analysis might suggest new ways of addressing what has proved so far to be a complex and difficult undertakingassessing the diverse outcomes in a variety of contexts of the growing use of computing and communications technology. The results of this exploration are intended to be useful to the National Science Foundation in its efforts to assess the impacts of computing and communications technology, to provide examples of successful research and pose interesting questions to the research community, and to inform policy makers about the nature and utility of such research.
The Steering Committee on Research Opportunities Relating to Economic and Social Impacts of Computing and Communications organized a workshop to explore opportunities for research on the impacts of information technology and ways to assess these impacts. Since this was an endeavor of limited budget and time frame, and to be based in large part on input received at a single workshop, the committee adopted the approach of identifying and developing some significant examples and issues, rather than performing a more comprehensive study of the full range of relevant topics. The content of this report reflects areas of interest and issues raised in the workshop and the position papers submitted by workshop participants and subsequent work by the steering committee.
This chapter illustrates the dramatic increases in computing power and the penetration of new communications technology, and it highlights several key areas where impacts of these trends are being felt. It concludes by discussing the role of social science in characterizing these impacts.
1.1 Growth Trends
1.1.1 Computing Power
Computers have become so affordable and ubiquitous in part because of the remarkable improvements in semiconductors. Since the 1960s, semiconductor chip makers have increased the density of transistor circuits at a rate of about 10 percent a year. Combined with numerous other technological advances, this capability has led to a doubling of microprocessor power every 18 months (Figure
1.1), a trend known in the computer industry as "Moore's Law" after a 1964 prediction by Gordon Moore, a founder of Intel Corporation. Improvements in semiconductors and other components account for the annual 20 to 30 percent decline in the quality-adjusted price for computers (Berndt and Griliches, 1993; Gordon, 1990), even as the costs of other industrial equipment have been increasing steadily (Figure 1.2). At the same time, businesses and consumers have chosen to increase their spending on computers, suggesting that they find new uses of computers worthwhile. As a result, the real quantity of computer power deployed, which reflects both increased spending and increased computer power per dollar spent, has grown tremendously over the past several decades (Figure 1.3).
1.1.2 Demographics of Computer Ownership
A recent survey of more than 50,000 households found that more than 40 million U.S. households now own PCs2 (Computer Intelligence, 1997). Ownership
patterns correlate strongly with age and income. About 20 percent of households with incomes between $10,000 and $20,000, but more than 60 percent of households with incomes of $60,000 to $75,000, have computers. Some of the largest annual gains in computer ownership were found in middle-income groups. For example, households in the $40,000 to $50,000 income group reached the 50 percent level, and there was a nearly 5 percent increase in penetration in the $20,000 to $30,000 income group. Almost 60 percent of households with children own a computer today.
An earlier RAND analysis by Anderson et al. (1995) examined computer ownership data from the 1993 Current Population Survey conducted by the Bureau of the Census. They categorized computer ownership according to income, education, race/ethnicity, age, sex, and location of residence. This study found that income and educational status were associated with significant differences in
computer ownership. For example, about 7 percent of the lowest-income quartile of households had computers at home, whereas nearly 55 percent of the highest-earning quartile had computers at home.3 In 1993, about 13 percent of individuals over the age of 15 who did not have a high school diploma had a home computer, whereas 49 percent of college graduates had a home computer.
1.1.3 Internet Use
The most reliable figure on Internet use is the number of computer and network domains connected directly to the Internet, all of which must have a public Internet address. According to conservative estimates from a recently conducted Internet domain survey (Network Wizards, 1998), the number of Internet host computers grew from 1.313 million in January 1993 to almost 30 million in January 1998.4
The number of individuals who use the Internet is very difficult to measure, in part because the definition of users is somewhat fuzzy. Users might be subscribers to a consumer Internet service or other computer service that includes the Internet as a component, or they might be potential users with access to the Internet at their workplace. Matrix Information and Directory Services (Quarterman, 1997) estimated that as of January 1997 about 71 million people worldwide had access to e-mail. When access to information by file transfer protocol (FTP) or the World Wide Web (WWW) was the criterion used, the figure was about 57 million people worldwide.
Project 20005 at Vanderbilt University has conducted several carefully designed surveys of Internet and Web use. One study (Hoffman et al., 1996) examined Internet use by age, education, and sex using a telephone sample of 3,785 respondents. It found significant differences in usage patterns across these demographic groups. For example, only 13.5 percent of those with a high school diploma or less reported using the Web once or more per week, whereas nearly 56 percent of those with a college degree reported weekly use. Hoffman and Novak's most recent estimates, based on Nielson Media Research's Spring 1997 Internet Demographic Study, showed that 45 million people in the United States 16 years of age and older, or about 22 percent of the population, had accessed the Web at least once (Hoffman et al., 1997a).
Another study (CommerceNet/Nielsen Media Research, 1997), in which 9,000 people were interviewed, estimated that 58 million adults in the United States and Canada used the Internet on a regular basis. More than half of the respondents said that they had been online within 24 hours of the interview.
Although e-mail appears to be the most widely used Internet service, the World Wide Web is one of the fastest growing services. A Baruch College-Harris Poll (1997) survey of 1,000 U.S. households conducted in the spring of 1997 indicated that the number of adult Web users had nearly doubledto 40 million people, or 21 percent of adultsfrom the previous year. About a quarter of these were people in their forties.
According to Matrix Information and Directory Services (Quarterman, 1997), about 36 million individuals worldwide have the technical capability to distribute or publish information via FTP or WWW. The Netcraft Web Server Survey found 525,915 publicly accessible Web servers in an exhaustive search in November 1996; by December 1997 the number had risen to more than 1.6 million (Netcraft, 1996, 1997).
1.1.4 Global Connectivity
According to a study by the International Data Corporation, the surge in Internet use is a global phenomenon (International Data Corporation, 1997a). The study estimated a worldwide total of 91 million users as of November 1997,
with 14 million users in the Asia/Pacific region, 20 million in Europe, and around 1 million in both Africa and South America.
1.2 Some Major Challenges
The studies cited above are, of course, based on particular samples and research methodologies; it is hard to appraise their accuracy without more detailed investigation. Moreover, among the variables that make measurement difficult and contribute to inconsistent results are disparate definitions of "use" and "access." However, these studies clearly indicate that computer and Internet growth is a significant, widespread, and global phenomenon. Although use of information technology is most prevalent among businesses and highly educated, high-income, North American households, all demographic groups in the developed world show a pattern of significantly increasing use.
A phenomenon this ubiquitous and growing this rapidly is clearly important. As the three examples below illustrate, social science research can help policy makers and others to better understand and shape the interactions between technology and society.
1.2.1 Productivity and Organizational Change
In the long run productivity is the primary determinant of our standard of living and the economic resources available to address societal challenges and problems. Understanding whether and how computers affect economic productivity is a critical issue for policy makers as well as business leaders. In 1987, the economist Robert Solow quipped, "We see the computer age everywhere except in the productivity statistics" (Solow, 1987). Solow's comment reflects the fact that, despite the tremendous advances in computer power and affordability shown in Figures 1.1 to 1.3, the aggregate statistics suggest that economic productivity has grown more slowly since about 1973 than it did in the period from 1950 to 1973.
There are many possible explanations for the apparent slowdown in productivity despite the advances made possible by computerization.6 Analysis is complicated by the fact that many of the benefits of the computer age are not reflected in the official statistics on output and productivity. Managers typically cite improved quality, variety, timeliness, and customer service as important reasons for their investments in computers, yet these aspects of output go largely unmeasured.7 Furthermore, the growth of productivity in the service sector is difficult if not impossible to measure. Zvi Griliches, for example, has estimated that the "unmeasurable" sectors of the economy grew from 53 percent of the total in 1948 to about 74 percent in 1994 (Griliches, 1994). Ironically, there is apparently less information in the information age about the value of output than there was in the
industrial era. New and better metrics clearly are needed to help in understanding how productivity in the information age should be gauged and evaluated.8
Another aspect of the productivity paradox may be unrealistically high expectations about the potential of computers to affect output. Investment in computerization is still a relatively small share of the total economy, and so the possible impact of computerization on the whole may be limited. Based on the quantity of computer power purchased each year, it is possible to calculate the expected contribution computers "should" be making to economic growth under the assumption that capital investment in computers is earning a "normal" rate of return. Various researchers have undertaken such an exercise, and the estimates cluster around 0.3 to 4 percent of additional growth in the economy owing to technological advances in computers each year.9 Perhaps, then, the measured slowdown in productivity would have been even more pronounced had it not been for the contributions made by computers.
A constant stream of new discoveries and advances is required just to keep productivity growth from falling to zero, and part of the promise of computers is their ability to engender such new discoveries. In earlier decades, growth in productivity was the result of innovations such as electricity, automobiles, the radio, jet engines, plastics, and even corrugated cardboard (Denison, 1985). Today, new products and services, innovative business processes and organizational forms, and even new industries can be traced to advances in information technology. Moreover, like earlier general-purpose technologies such as the steam engine and electricity, computers have changed the nature of work in myriad minor and major ways, thus magnifying their economic and social impact.
David (1989, 1990) found that the electrification of factories had its greatest effect after factories had been radically reorganized (see section 2.3.3 for more details). Similarly, case studies suggest that realizing the full benefit of computer technologies also often requires a significant rethinking of how business works. As a result, the ways in which computers are used may be more important than the quantity of investment. In one instance, a large medical products manufacturer found that effective use of computerization required changes in two dozen work practices and policies, including those related to inventory, incentive systems, worker training, job responsibilities, and hiring criteria. Initially, productivity fell because the new system did not work smoothly. However, by focusing on the interactions among the work practices, the company was able eventually to reap substantial rewards from use of the technology (Brynjolfsson et al., 1997).
Understanding exactly which work policies and practices need to be changed and which changes are most important in various contextsfor example, what frequency and amount of training are required for workers to adapt to new technologiesare critical research issues. Collection of further case studies can help, but there is a particular need for analysis leading to broader insights that can be generalized across numerous companies and situations. Here again, lack of good
data on how businesses are changing has hampered progress. Section 2.3.3, "Organizations and Processes," discusses these issues in more detail.
1.2.2 Information Technology and Wage Inequality
For more than 20 years, the gap separating high-wage earners from low-wage earners has continued to increase in the United States. For example, the ratio of the wages earned by males at the 75th percentile of the distribution to those at the 25th percentile has grown from 1.75 to 1 in the 1970s to 2.25 to 1 in the early 1990s, and the wages of those at the more extreme percentiles have moved farther from the median (Murphy and Welch, 1993). By some measures, the poorest members of society are worse off than they were a generation ago. Most economists attribute much of the increase in inequality to an increase in the demand for skilled labor and also link that shift in demand to technical change. For example, Autor et al. (1997) showed that the increase in inequality was largest in those industries that were the heaviest users of information technology. However, it remains unclear how use of IT changes labor markets. Thus, it is difficult to predict whether the growth in wage inequality will continue and what policy makers can do about it.
Is the growth of inequality a temporary phenomenon that will correct itself? Should society invest more heavily in education to dampen the negative effects of technological change? What is the role of corporate reorganization in changing the demand for different types of workers? These are some of the important questions that can be addressed only by further research on the economic and social impacts of computing and communications. Section 2.3.2, "Labor and Information Technology," discusses this subject in more detail.
1.2.3 Design of Technology and Standards Setting
Social science has much to contribute to the design of appropriate technology. Today the question facing many technologists is not, How do we do it? but, What should we do? As determinants of technology development, design issues involving human-computer interfaces (see Box 1.2), pricing and "versioning" (the provision of different qualities or versions of a good that sell at different prices), evaluation, and product life cycles have become more important than the traditional engineering concerns. Social scientists can help to design the questionnaires, marketing studies, pricing policies, and interfaces necessary to develop successful information technology.
Standards development is another aspect of technology design in which social science can play a very useful role. Technical standards are the basis for interconnection and communication among information technology systems. However, an important aspect of standards is that once adopted, they can be very difficult to change because of their highly distributed nature and the consequent
As a result of technology advances, interface designers can now attempt to make human-computer interfaces more personable by incorporating human-like attributes such as speech or representations of human faces using pictures, simulations, or animation. When technologists introduce human-like attributes into interfaces, they change the user's experience from one of human-machine interaction to one that approximates some features of human social interaction. Even though such interfaces only partially represent human attributes and clearly are not "real," users still respond somewhat as though they were interacting with a person rather than a machine.
Social science research has explored how people react to more human-like interfaces. For example, Clifford Nass and his colleagues demonstrated that when people interacted with computers that had human-like attributes, they applied such rules as "Be polite when delivering evaluations" and "Praise of others is more accurate than praise of self" when they assessed computer software (Nass et al., 1994).
Lee Sproull and her colleagues demonstrated that people made their own behavior more social when they interacted with human-like interface agents. For example, people reported that they were more altruistic when they answered standard personality-test questions asked by a talking face displayed on a screen than when they answered the same questions displayed in a text window (Sproull et al., 1996).
In neither case did the researchers ask the conventional questions about ease of use: Was the audio easy to hear and understand? Were the visual images sharp and clear? Were the instructions easy to understand? Instead the researchers asked, Did users respond to these computer programs in ways reminiscent of ways they would respond to a human being?
Trying to introduce human attributes into interfaces is not necessarily the right approach.1 However, if technologists do attempt to make computer interfaces more human-like, then social scientists can use theories of social interaction to help technologists understand and predict how people will react to these interfaces, thus increasing the acceptance and utility of the computing technology.
1A recent Computer Science and Telecommunications Board report (CSTB, 1997b) discusses promising research in interface design.
need for broad-range, costly alternations of user behavior and infrastructure. For example, technical standards developed and adopted now for electronic paymentsuch as the proposed Secure Transactions (SET) standard for credit card transactions over the Internetmay be in use for some time, and careful thought should be given to their implications.
Thus far, information technology standards setting has been successful in not hindering the dramatic growth just described. However, technical standards are playing an increasing role in policy considerations, and some current issues require
the development of standards that address social, legal, and policy concerns. See sections 2.4.1, "Protection of Intellectual Property," 2.4.3, "Privacy," and 2.4.9, "Electronic Commerce," for further discussion.
1.3 Role Of Social Science
There is no shortage of pundits who will forecast the future of the information society. Much of their writing is based on extrapolation or slogans, neither of which is an effective guide to crafting public policy and designing technology. One month everyone is talking about "information 'haves' and 'have-nots'"; the next month it is "agent-based commerce." Rarely do articles in the mass media refer to social science research, such as studies of differential access to computers or economic studies of a variety of mechanisms for automated trading, that would be highly relevant to these issues.
For the purposes of this report, the core social sciences are taken to be anthropology, economics, history,10 political science, sociology, and social psychology. Closely allied are the fields of demography, information science, law, and organizational studiesareas that have a significant social science component and are also highly relevant to research on the impacts of information technology. These disciplines use dramatically different methods. Economics, for example, is highly quantitative, whereas anthropology and history are typically much more descriptive.
An advantage offered by social science is that it can help technologists and policy makers to understand how people and social institutions behave in response to current technology developments and what the effects of particular changes might be. Changes might be introduced at the microlevel, for example, by technologists who make computer interfaces more personable, or at the macrolevel, for example, by policy makers who put in place new intellectual property protection frameworks. At either end of and along the entire continuum, social science data and theory can help decision makers make better-informed decisions.
The collection of reliable datasuch as data on the penetration or use of a particular technologyis an endeavor in which the methodology of social science can make an important contribution. For example, as noted in section 1.1.3, measurement of how many people use the Internet is complicated by the current lack of widely accepted, precise definitions of "use." Defining terms appropriately and precisely is basic to social science methodology, as are assessing the reliability of survey data and increasing the reliability by using particular techniques for sampling, increasing response rates, and assessing the impact of missing data.
The technology-related work that has been done by social scientists is not widely known either to policy makers or to technology designers in part because it is published in academic journals, which are relatively inaccessible, and disseminated
only to a relatively small audience. The World Wide Web may allow such studies to become more accessible. However, the typical problem is not in gaining access to scholarly journals and other such sources of pertinent studies, but rather in knowing what to search for. Also, scholarly work is often inaccessible to the lay reader.
Much of the data on the computer and communications industry comes from market research firms whose data complements that generated by social scientists. Both types of data are necessary, and both are worthwhile. One distinction is that the commercial sources make money by selling their data; nobody makes money directly form social science data. But the primary distinction is that social scientists do not simply count things. When they count (to answer how much, how many, how frequently), a theoretical context, involving systematic theories of human behavior, motivates the counts. Thus, for example, a sociologist interested in electronic group dynamics might ask about the relationship between unpleasant events in the group and a decline in group membership. She might count the frequency of ''flames" (hostile messages) or "spams" (electronic junk mail) or flames about spams, or how much time people spend deleting flames and spams from their mailboxes, as indicators of unpleasant events. In assessing how those numbers correlate with changes in the size of group membership, the goal is not to count, but rather to understand how the frequency of one kind of behavior affects the dynamics of a social institution. Such understanding enables more reliable forecasts and more trustworthy inferences about causal relationships.
The value of social science research comes not from tracking the frequency of use of the latest technologies but rather from helping to develop common social and economic principles that can be applied to new circumstances. Those designing or relying on technology and those making policy decisions about the use of technology without reference to systematic theories of human behavior or economics will likely find themselves approaching each new issue in ignorance. Given the rapid pace of technological change, this approach has both economic and social costs.
In summary, the ongoing computing and communications revolution requires serious social science investigation. Such work would be valuable for both social policy and technology design. Many choices being made now will be costly or difficult to modify in the future. To help answer the challenges introduced above, new areas of research, improved collection of data, and new collaborations are called for. The chapters that follow examine these issues.
1. Throughout this report, the term "impacts" is used as a shorthand expression to indicate a complex set of multicausal, multidimensional technology outcomes. Technology does not typically have a single impact, but rather a range of different outcomes depending on the context or settings. For more discussion, see Box 1.1 as well as Attewell's and Kling's papers in Appendix B of this volume.
2. The Computer Technology Index survey used a sample designed to be representative of all U.S. households and had a 66 percent response rate.
3. Evidence is emerging that the advent of PCs priced below $1,000 is allowing lower-income households to enter the market for personal computers (Hoffman and Novak, 1998).
4. Network Wizards changed the method used to collect data starting with the January 1998 survey and cautioned that care is required in comparing figures derived by using the old and new methods. The extent of growth is unmistakable, however. The last survey conducted with the old method found nearly 20 million hosts in July 1997which would correspond to roughly 26 million hosts if the new survey method had been used. See Network Wizards (1998) for a discussion of this issue.
5. Project 2000 publications are available online at ‹http://www2000.ogsm.vandervilt.edu/›.
6. To some, the slowdown in the growth of officially measured productivity seems at odds with the many examples of how computerization has contributed to advances in the past two decades. Automatic teller machines enable banks to handle millions of transactions at all hours of the day and in numerous locations, a capability made possible by computer networking. Volumes of transactions are handled that would be inconceivable without computers. In addition, many modern manufacturers operate with a fraction of the labor previously required while handling far greater product variety. Retailing, medicine, transportation and logistics, communications, and virtually every other industry are in the midst of a computer-enabled transformation.
7. Citing unmeasured improvements in these and other areas, the Boskin Commission recently estimated that the consumer price index overestimates inflation by approximately 1 percent (Advisory Commission to Study the Consumer Price Index, 1996). This implies that growth in productivity has been underestimated by a comparable amount.
8. One tack that several researchers have recently taken is to rely on firm-level data instead of national or industry-level data. For instance, Brynjolfsson and Hitt (1996, 1997) and Lichtenberg (1995) have found that there is, in fact, a strong positive correlation between IT use and productivity at the firm level. See also CSTB (1994a).
9. See, for instance, Brynjolfsson (1996a,b), Oliner and Sichel (1994), Jorgenson and Stiroh (1995), and Lau and Tokutsu (1992).
10. Although history is not usually classified as a social science, historical analysis can play a valuable role closely allied to that of the social sciences in helping to characterize the impacts of computing and communications.