Introduction
The number of individuals who engage in computer-based
communications in the United States has increased
dramatically in recent years and is expected to continue
growing well into the next century. Although its reach at
present is far from universal, information technology is
already "woven into the fabric of the economic and social
life of developed countries," as King and Kraemer (1995) put
it.
The trend toward growing use of computer-based communications stems from two mutually reinforcing influences. First is the often-cited history of improvements in price-to-performance ratios. That is, prices for equal amounts of processing power drop by about half every two years (Tessler, 1991). Having started at least two decades ago, such changes are viewed as the enabling force behind the widespread diffusion of computers to households and offices. Communication technologies are likewise beginning to show price/performance improvements while at the same time shedding their terrestrial and bandwidth constraints (Tessler, 1991; Benjamin and Blount, 1992; King and Kraemer, 1995).
A second and related stream of influence has to do with the convergence of computing and communication technology within an integrated information medium (Eveland and Bikson, 1987). This integration, for instance, permits individuals to communicate information as readily as they create it, to one or many others, sometimes scarcely noticing that generating, editing, storing, and sending are distinct activities (Bikson and Frinking, 1993); conversely, this convergence preserves the "computability" or reusability of what is received via the medium (Steinfield, Kraut, and Streeter, 1993). Steinfield, Kraut, and Streeter contend that this property more than anything else accounts for the benefits of e-mail over other contemporary communication media (e.g., voicemail).
In the past, the advantages of such convergence were generally confined to islands of disconnected interoperability (e.g., within particular organizations). However, the movement toward open systems (discussed further in Chapters Three and Four) has broadened the capability for interconnection, linking larger numbers of geographically dispersed organizations and individuals with heterogeneous hardware and software to one another through a common electronic infrastructure (Bikson, 1994). The result of these lines of influence is perhaps best illustrated by the phenomenal growth of the Internet in the 1990s (see Figure 2.1).
As indicated in the introduction to the report, this chapter provides a detailed look at trends in information and communication technology access for the U.S. population based on CPS data. Specifically, we aimed to learn how evenly computer-based communications capabilities are distributed over the country's varied demographic constituencies and whether those groups exhibit similar trends in access to network services.
Before addressing these questions, however, it is appropriate first to indicate why they are important. That is, are there any reasons to view economic and social stratification of computer and network use differently from the socioeconomic stratification that characterizes the consumption of other goods and services (compare Attewell, 1994). We believe there are at least four important reasons.
Different levels of access to computer-based communication technology, then, may further stratify individuals and create information have-nots alongside the "information elite," as the Times Mirror survey describes those with computers, modems, and access to network services from home. However, the Aspen Institute's Information Bill of Rights (Firestone and Schement, 1995) argues that the emerging information society should have information openly flowing among all individuals and institutions, which in turn requires equitable access to an information and communication infrastructure (see also Denning and Linn, 1994). The policy significance of such potential information stratification can only increase as more public and social information is disseminated electronically.
That networks are not just information technologies but also serve as social technologies (or technologies for affiliation) is supported by the well-established tendency of users to turn what were initially designed as media for accessing and using remote data (e.g., ARPAnet and Minitel) into interpersonal messaging media (see Shapiro and Anderson, 1985, Sproull and Kiesler, 1991b; Lytel, 1992). These observations are corroborated by social science research results indicating that network use creates and sustains both strong and weak social ties (Bikson and Eveland, 1990; Feldman, 1987) and that those who use an organization's network feel more positive about their association with the organization than those who do not (Kraut, 1993; Huff, Sproull, and Kiesler, 1989).
This is not to say the two roles for electronic networks are in conflict; on the contrary, people often use other people as providers of information, pointers to information and filters of information (Eveland et al., 1995; Bikson et al., 1995; Sproull and Kiesler, 1991b; Kraut, Dumais, and Koch, 1989; Bikson, Quint, and Johnson, 1984). Rather, it is to underscore that constraints on network access are at the same time constraints on affiliation. To the extent that civic and social alliances increasingly rely on computer-based interactions, constraints on association represented by less- than-universal access pose policy problems--and especially so if the constraints are unfairly distributed over socioeconomic groups.
As civic and political groups increasingly rely on electronic networks (see Chapter Five), these media could help them overcome status-linked barriers to full participation in social dialog and public life. However, if access opportunities follow traditional socioeconomic lines, these barriers instead will be strengthened by the emerging information infrastructure (see also the Aspen Institute's Information Bill of Rights, Firestone and Schement 1995, section A-II on nondiscrimination in "the information society").
However, most studies of economic benefits associated with use of networked information and communication technologies have been carried out at the level of the enterprise or firm rather than at the individual or household level (see Computer Science and Telecommunications Board, NRC, 1994). At the firm level, the competitive advantages provided by interorganizational networks are widely recognized (e.g., Quinn, 1992; Rockart and Short, 1991; Malone, Yates, and Benjamin, 1987). More important, for purposes of this project, are findings from a well-designed comparative study of data from hundreds of small, medium, and large businesses in the United States and France carried out by Streeter et al. (in press). The study shows that national interorganizational networks (in contrast to proprietary networks) foster the formation of electronic marketplace relationships among businesses. Further, they diminish large organizations' inherent advantages (size, slack resources) in exploiting new technologies, making it comparatively easier for small- and medium-sized businesses to enter the marketplace and benefit from it. While we do not examine organization-level data in this chapter, these findings mirror some of the equalization effects observed at the individual level. They also call attention to the likelihood that policies affecting universal access to computer-based communications technologies are likely to have a disproportionate influence on the economic well-being of small- and medium-sized enterprises. (Economic effects at the societal level are discussed in Chapter Four in relation to critical mass and network externalities.)
There are significant reasons, then, for policymakers to become involved in the debate over universal access to electronic networks. Networks can influence the public's exposure to information. They can create opportunities for individuals and groups to affiliate and to participate in civic affairs. And they can create or shape economic opportunities and advantages. As suggested by King and Kraemer (1995), those who lack access to new communication technologies may be at risk of exclusion from the fabric of the nation's social and economic life.
Computer and Communication Technology In Use
Against this background of converging technology trends and their societal implications, we turn now to a more detailed investigation of access to computers and communication networks in the U.S. population. We rely chiefly on the October 1989 and October 1993 (CPS) data.[1]
The CPS is a large-scale random sample survey of households, conducted monthly by the Bureau of the Census. It is the source for much of the official data published by the Bureau of Labor Statistics. The Bureau of the Census periodically adds supplements to the CPS base questionnaires to gain more insight into topics of interest. In this study, we initially examined the October 1984, October 1989, October 1993, and November 1994 supplements because they include questions on computer use by each individual in the household. The 1984 data are not always comparable to data collected in later years, and the 1994 data were released too recently for careful examination within the context of this project. Consequently we focus here mainly on the 1989 and 1993 data.
CPS data are suitable for analysis at the household level or individual level. This report treats the individual as the unit of analysis. Although some outcomes of interest (e.g., presence of a computer at home) are readily interpretable at either level, others (especially behavioral variables such as use of networked services) are not. Exploratory work suggests that for purposes of this study, where both levels of analysis are appropriate, differences between findings at the individual and household levels are negligible (see, for instance, Figure 2.2).
At the individual level, then, the analyses reported below are based on 289,979 observations (146,850 in 1989 and 143,129 in 1993). The sample consists of noninstitutionalized civilians in the United States living in households. Both adults and children are in the sample, unless explicitly noted otherwise.[2]
Outcome Variables. To represent access to information and communication technology, we employed two binary outcome variables. One, access to a computer at home, is a single-item measure; it receives a positive value if there is at least one computer in an individual's household. At this level of analysis, penetration of computers refers to the percentage of individuals with household access (rather than the percentage of households that have computers).
The other outcome variable, use of network services, represents use of a computer either at home or at work to connect to an electronic network. A derived measure, this variable receives a positive value if an individual uses a computer in any one of the following ways:
An alternative approach would be to define this outcome in terms of having or using a modem (compare Times Mirror, 1994). However, preliminary reviews of CPS data and other studies suggest that individuals do not always know whether they have or are using a modem.[3] Further, we decided to include connectivity in the workplace as well as from home in the definition because it provides a more complete picture of the degree to which individuals use electronic avenues to communicate with others. At present, more people use network services from work than from home, often for both work-related and personal or social purposes; and networks are sometimes used by individuals doing part of their work from home (Figure 2.2). Thus, the distinction between work and home use of network services could not be made reliably in the CPS data.
Unfortunately, no questions were asked of students about how they use computers and networks at school. This implies that our network outcome variable underestimates the actual use of network services among students. For most determinants of interest, this underestimate is inconsequential. However, results on variables that are highly correlated with student status (such as age and, possibly, household income) need to be interpreted with this caveat in mind. As background, Figure 2.2 shows home computers as a percentage of both households and individuals in the CPS data; it also shows network use at home and work.
Predictor Variables. Six predictor variables constitute the core of our study: income, education, race/ethnicity, age, sex, and location of residence. Income, a categorical variable defined by quartiles, refers to the total income of the individual's household. Location, another variable defined at the household level, reflects whether the individual lives in an urban or rural area. Remaining predictor variables refer only to the individual. (Each explanatory variable is further defined in the discussion of results below.)
In investigating the CPS data, our goal was to learn whether and how socioeconomic characteristics are correlated with distribution patterns and diffusion trends in access to computers and electronic networks.
Analysis Plan. Figure 2.2 presents aggregated CPS data representing the two outcome variables of interest for the United States population in 1993 and 1989 (and, for access to a computer at home, 1984). The analysis was designed to answer two questions about these outcomes at the individual level.
Answers to these questions are tested statistically in several ways. First, we examined differences in access to a computer at home and use of network services across socioeconomic groups in the two years separately. These differences follow from cross- tabulations and are shown in bar graphs for each socioeconomic dimension of interest. Because of very large sample sizes, in every cross-tabulation presented, and in both years, the differences between groups are generally statistically significant.[4] (When they are not, we make note of it.)
For purposes of policy analysis and intervention, however, these "gross" differences may be misleading. Socioeconomic status variables are likely to be intercorrelated, meaning that an effort to investigate any one of them should control for the potential influence of all other covariates of interest.[5] Therefore, we held the other socioeconomic variables constant and recalculated computer and network penetration levels to obtain such "net" percentages.[6] Net figures can be interpreted as representing differences between individuals with otherwise equal characteristics (where those equal characteristics are a weighted average of all characteristics found in the data). The same general pattern of findings emerges from the net data but between- group differences are generally reduced.
Appendix A explains the procedure to compute net disparities in detail and provides a table with both gross and net percentage data for purposes of comparison. It also contains tables of the (probit) regression analyses from which the net percentages were derived. The discussion below emphasizes gross results, but we will point out where and to what extent gross results overstate disparities across socioeconomic groups.
In what follows, findings from the data analysis are presented first for each of the six predictor variables. We conclude with a discussion of their combined influence on access to computers and communications technology.
Differences by Household Income. Figure 2.3 presents the percentage of individuals who report that there is a computer in the household and that they have access to network services, as a function of household income category. We distinguish among four quartiles; for example, the bottom quartile includes the 25 percent of the population with the lowest household income. In 1993, the quartile cutoff income levels were $15,000, $30,000, and $50,000 per year.
As is immediately clear in Figure 2.3 (left half), there are very large differences in household computer access across income categories. In 1993, just over 7 percent of the lowest income households had computers, whereas nearly 55 percent of the highest-earning quartile had computers at home. Four years earlier, the respective figures were nearly 6 percent and 35 percent. These data, therefore, reflect highly significant differences in household computer access based on income quartile.
Further, while the income-based gap in household computer access was very large in 1989, it was even wider by 1993. In 1989, individuals in the top quartile were over six times more likely to have access to a computer in the household than individuals in the bottom quartile of the income distribution. By 1993, this gap had widened to well over seven times more likely.
The net disparities, controlling for the other key socioeconomic characteristics, are not quite as large, but they remain substantial. For example, in 1993, on net, individuals in the top income quartile were four and a half times more likely to have access to a computer in the household than those in the bottom quartile (Appendix A, Table A.3). This net income gap is smaller than the gross figure, mainly because low-income individuals tend to have lower-than-average educational attainment. About a third of the income disparity is thus attributable to a concomitant effect from educational differences. The net income gap in 1993 represents a significant widening relative to 1989.
Although use of network services either at home or at work is far less than availability of household computers (see Figure 2.3), generally similar patterns appear for network use as a function of household income level.[7] Again we find large differences between quartiles that are becoming even larger over time. In 1989, close to 2 percent of the lowest income individuals used network services at home or work, whereas over 11 percent of the highest income individuals used them. By 1993, these fractions had increased to nearly 3 percent and 23 percent, respectively. As before, the gross disparities are in part attributable to correlation of household income with other demographic predictors, notably educational attainment. The net gaps remain very substantial, though, and have widened significantly between 1989 and 1993 (Appendix A).
We conclude, then, that there are large differences in both household computer access and use of network services across income categories. These differences are due partly to other socioeconomic characteristics, but they remain highly significant even after controlling for those other characteristics. The gap between high-income and low-income individuals is not only large, it also widened between 1989 and 1993; higher-income individuals appear to be adopting the new technologies at a faster pace. Interestingly, the net differences are smaller for network usage than for household computer access. This may be due to broader access to network services in the workplace, where no investments in hardware on the part of the individual user are required.[8]
These results are congruent with the Times Mirror survey's conclusion (1994) that the spread of technology through American society is quite uneven. While that survey examined data from only one year (1994), it investigated a broad range of technology to find that these disparities are "greatest with respect to computers and on-line capability." Rapidly improving price-to-performance ratios in recent years thus seem not to have narrowed (or even held constant) income-based gaps between information technology haves and have-nots.
Differences by Educational Attainment. Figure 2.4 (left side) shows household computer access fractions for individuals without a highschool diploma, for high school graduates, and for college graduates. (Children under 15 years of age are not included in this part of the analysis.) Persons with some college education, but without bachelor's degrees, are included among high school graduates. As may be expected, there are large differences in household computer access by educational attainment. Among persons without a high school diploma, only about 8 percent had a home computer in 1989. College graduates, by contrast, had a penetration rate of about 32 percent. All groups experienced an increase in home computers between 1989 and 1993, leading to penetration rates of about 13 percent and 49 percent in 1993 for those without a high school diploma and college graduates, respectively. Controlling for other socioeconomic characteristics, the differences are substantially smaller but still highly significant statistically (see Appendix A) in both years. In the interim, moreover, such education- based differences in household computer access have increased significantly. The disparities parallel, but are less sharp than, the income-based differences in household computer access reported above.
Figure 2.4 (right side) also presents differences in network usage by education category. We find that use of network services is dominated strongly by well-educated individuals. In 1989, a mere half percent of individuals without high school diplomas used network services, compared with over 18 percent of college graduates. Both groups strongly increased their use in 1993, to over 1 percent and about 34 percent, respectively. As expected, the net differences are smaller but still substantial and statistically significant (see Appendix A). The net differences in network use by educational attainment have also widened significantly between 1989 and 1993. Interestingly, the divergence is entirely due to an acceleration of the adoption of network services among college graduates; the gap between high school drop-outs and high school graduates did not change significantly.
In summary, we find large differences in access to information and communications technology by educational attainment that are increasing over time. Given the established correlation between use of network services and knowledge of current political, professional, and organizational affairs cited above (see the introductory section of this chapter), these results suggest that disparities in access to electronic networks may well amplify differential knowledge produced by education differences alone.
The Times Mirror survey of technology in American households (1994) yields similar conclusions using 1994 data. It also draws attention to the effect of these patterns of technology access on children's educational opportunities. Among college graduates with children in that sample, almost half reported that the child used a personal computer; but among those with a high school education or less, only 17 percent reported that children used a home computer (Times Mirror, 1994). However, while large income- and education-based differences exist in children's access to computer technology, the survey found "virtually no socioeconomic differences in how often and for what purposes children used computers if present in the home." These findings suggest that effects of parental educational stratification could be at least partially offset if it did not result in differential access to information and communications technology for children.
Finally, it should be noted that if more and more jobs at relatively low levels increasingly make discretionary use of network technologies (such as e-mail) available, differences among adults in access to on-line information based on income and education could decrease in the future.[9] At present, such differences are problematic because they exacerbate differences in earnings as well as differences in general level of knowledge (see the introductory section, above).
Differences by Race and Ethnicity. Black community leaders have recently expressed concern that African Americans are lagging behind in the use of computers (New York Times, 25 May 1995). At least part of the race- based difference is due to lower average household income and lower average educational attainment among blacks as compared with whites. However, our analysis shows that those characteristics do not account for the entire difference in outcome variables. Rather, racial and ethnic characteristics exert an independent influence on home computer access and network use.
For purposes of this analysis, we combine race and ethnicity into mutually exclusive categories. We distinguish between Hispanics, non-Hispanic whites, non-Hispanic blacks, Native Americans (both Indians and Eskimos), and Americans of Asian descent (including Pacific Islanders). In subsequent comments we refer to non-Hispanic whites as "whites" and to non-Hispanic blacks as "blacks." A small fraction of respondents (0.11 percent) are identified as "other" in the CPS data; we do not reflect the "other" category in Figure 2.5 (or in the table in Appendix A). Figure 2.5 portrays the percentage of individuals with a computer in the household and access to network services at home or at work by racial/ethnic categories.
As Figure 2.5 makes clear, the highest penetration rates for household computers are found among whites and Asians. In 1993, over 30 percent of whites and over 37 percent of Asians lived in a household with a computer. Hispanics, blacks, and Native Americans, by contrast, all reported a penetration rate of around 13 percent. As we mentioned above, part of these differences may be due to average differences in other characteristics, notably, household income and educational attainment. Controlling for these characteristics, however, we still find substantial differences. That is, net of other influences, race or ethnicity has a statistically independent and sizable effect on household computer access. In particular, Hispanics, blacks, and Native Americans are currently underrepresented among computerized households. Similar patterns of significant racial/ethnic difference in household computer access are also evident in the 1989 data. However, unlike income- and education-based differences, racial/ethnic gaps in home computer access have not widened over time; instead, they have remained constant.
Differential use of network services as a function of race/ethnicity is also apparent in Figure 2.5. Again, there are significant between-group differences, even when the influence of other socioeconomic characteristics is controlled. Net differences, however, are slightly smaller than for household computer access. Somewhat surprisingly, Asians have the lowest net rate of network use, even though they have the highest net rates of household computer access among the racial/ethnic groups we distinguished. Another striking finding concerns the relatively high use of network services among Native Americans. Controlling for other sources of effect, their net usage rate was not significantly different from the net rate among whites in 1993, despite a much lower penetration rate of computers in their households.[10] Given the results of research on relationships of peripherality to network use (see, for example, Hesse et al., 1990; Huff, Sproull, and Kiesler, 1989), it is worth exploring whether Native American usage rates are at least in part explained by their geographic remoteness. No statistical differences in these patterns emerged from our analyses; that is, the racial/ethnic gap in use of network services has remained constant between 1989 and the present.
In summary, we find rather large and persistent differences across race/ethnicity in both household computer access and network services usage. These findings are partially consistent with racial/ethnic differences reported in the Times Mirror survey (1994), although that study did not include Asians and Native Americans as separate subsamples. There is no generally accepted explanation for these kinds of differences. For Hispanics, it has been suggested that language barriers may be partly responsible for the differences. For Native Americans, our literature review surfaced what seemed to us a comparatively large number of articles describing on-line educational, library and other information-oriented services targeted to this constituency. For instance, in a recent BoardWatch poll of favorite bulletin boards (1994), a Native American entry emerged in the top 20.
In advance, from our literature review, we did not expect to find any race/ethnicity differences other than those that could be explained by differences in income and education and, perhaps, residential location and age. That racial and ethnic differences remain even when the influence of other predictor variables is controlled is a matter that merits further research.
Differences by Age. We now turn to differences by age.[11] For purposes of this analysis we constructed four categories, distinguishing between individuals under 20 years of age, between 20 and 39, between 40 and 59, and 60 years of age and older. Boundaries based on age are admittedly arbitrary, and different studies employ different cutoffs, different numbers of categories or both (see, for instance, Times Mirror, 1994; BoardWatch, 1994). Particular boundary choices do not, however, appear to affect analytic results in ways that would affect most policy decisions.[12] Figure 2.6 displays household computer access and network access as a function of the age categories defined here.
As Figure 2.6 suggests, household computer access is distributed fairly evenly across broadly defined age categories up to age 60, where rates of penetration decline steeply. In 1993, around 30 percent of individuals under age 60 had access to a home computer, whereas only about 10 percent of individuals above age 60 lived in a household with a computer. Even when other socioeconomic variables are controlled, this difference is highly significant. The age gap appears to be headed for reduction, though; compared with the situation in 1989, older adults have higher relative penetration rates for household computers. However, this change is not large enough to reach statistical significance in the net figures; as indicated in Appendix A, Table A.1, gaps between those over 60 and others have remained stable.
Figure 2.6 also reveals the existence of large differences in the use of network services across the age categories defined, with disparities accumulating at both ends of the distribution. In 1993, only 1 percent of children and students under age 20 reported using network services, compared with over 18 percent among 20-39 year olds, over 20 percent among 40-59 year olds, and over 3 percent for people aged 60 and older.[13] In addition, use of network services among very young children (e.g., those too young to read or spell) is likely to be near zero, which further lowers the average estimates for the group under 20 years of age. On the other hand, a special issue of Communications of the ACM (CACM) on education reported very little use of networks at school, even in schools where students had access to computers (Soloway, 1993). On balance, then, those in the youngest age category are disproportionately likely to lack network access.
Older adults likewise make significantly less use of network services than younger adults do; the 3 percent access rate reported here coincides almost exactly with that obtained in the 1994 Times Mirror survey, in spite of differences in how both outcome and predictor variables were defined. But they may be catching up in the future--the data in Figure 2.6 show that in 1993 older adults were about three and a half times more likely to use network services than in 1989 (net figures in Appendix A show similar change). This growth rate is faster by far than that exhibited in the two other adult age groups during the same period, suggesting that age level per se does not determine either adoption or use of these new technologies. Their rapid diffusion among older adults now may be explained in part by the larger proportion of household income available for discretionary spending among older adults (Bikson et al., 1991). Nonetheless this growth rate has not yet produced a statistically significant reversal of trends; that is, while age gaps have not widened, they also have not yet narrowed significantly between older adults and their younger adult counterparts.
Differences by Sex. There is minor variation by sex in access to home computers and use of network services, as Figure 2.7 illustrates. While the gross percentages shown in Figure 2.7 suggest a two-point difference in 1993, that difference disappears entirely when the influence of other socioeconomic variables is controlled (see the net percentages in Appendix A). This was not the case in 1989; controlling for other variables, men were more likely to have access to a computer in the home in 1989. The difference was small (less than a percentage point), but statistically significant (see Appendix A). Between 1989 and 1993, though, the gender gap has closed. It should be noted that this outcome variable--having a computer at home--does not take into account which household member instigated the purchase of the computer. CPS usage data for 1993 (tabled in Appendix A) show that men in fact make more frequent use of home computers than women; on the other hand, women are significantly more likely than men to use a computer at work (Times Mirror, 1994).
Use of network services also exhibits very little variation as a function of sex. We found differences between men and women on this outcome variable to be statistically negligible in both 1989 and 1993 when the influence of other socioeconomic variables is controlled. Paralleling the data for computer use, the Times Mirror survey (1994) reported that men engage in on-line activities from home more often than women; but women are more likely to go on-line at work. The reduction of the gender gap among adults, as evidenced in CPS data, seems also to be reflected in the behavior of children in households with computers. Times Mirror survey data (1994) show that among all households with a computer and at least one child, one or more children are using the computer in 75 percent of them; and there is little difference between boys and girls in either likelihood or frequency of use.
Further, those survey data indicated that girls may be heavier computer users than boys, at least for applications that assist them with their school work (e.g., word processing). Boys and girls were about equally likely to use a home computer for educational games and art, but boys significantly outpaced girls in frequency of use of the home computer for playing noneducational games. On the whole, our analysis of sex differences in access to information and communications technology provides evidence that the gender gap among adults has decreased; and we concur with the Times Mirror conclusion that it could disappear entirely in the next generation.[14]
Differences by Location of Residence. Household computer access and access to network services as a function of residential location, the last predictor variable we explored in detail, are given in Figure 2.8. Location is categorized here as rural or urban, where "urban" characterizes residences within standard metropolitan areas.[15]
Ostensibly, the household computer penetration rate in urban areas is much higher than in rural areas. In 1993, just over 29 percent of individuals living in an urban area had a computer at home, compared with just over 19 percent among rural residents. About half of the difference is due to correlation with other characteristics such as household income or education. The net gap is nonetheless statistically significant and narrowed somewhat between 1989 and 1993 (see Appendix A). There are substantial differences in the use of network services as well. In urban areas, over 12 percent of residents made use of network services in 1993, whereas for rural residents the figure is less than 8 percent. Again, roughly half of the difference is due to characteristics such as income and education. The gap has remained approximately constant between 1989 and 1993.
Interestingly, the use of network services is approximately proportional to the penetration of computers in the household; that is, rural residents show no greater deficit in network use than in home computer access when compared with urban dwellers. This finding conflicts with the expectation that "access ramps to the information superhighway" (Times Mirror survey, 1994) are likely to take longer to diffuse to rural homes. In contrast to urban modem users, who are often able to dial into the Internet or other networks with a local call (at low or even zero marginal cost), rural telephone connections may well involve nontrivial toll costs. Installation of a second line for computer access may also likely be more costly in rural than in urban areas. Nonetheless, we find no greater urban- rural differences in use of network services than we do in household computer access. Motivation to use network services in spite of such obstacles may be partially explained by geographic remoteness (see Hesse et al., 1990).
Besides urban-rural differences, CPS data also show strong regional differences in distribution of information and communications technology (see Appendix A). For instance, in 1993 about a third of the people in New England and in the Pacific and mountain states had a computer at home, and about 13 percent of people in these areas were using network services; in the east south-central states, by contrast, the corresponding figures are 16 percent and 7 percent. A review of specific cities and towns indicates further disparities that reflect other than regional differences (see also Appendix A). For example, in Michigan in 1993, about two- thirds of people in Ann Arbor had a computer at home and 27 percent of Ann Arbor residents were using electronic networks. In neighboring Flint, 16 percent of the residents had home computers and 5 percent of residents used networks. These differences are likely to be at least partially accounted for by the influence of other socioeconomic variables (income and education, especially). Other factors (e.g., proximity to a major research university) need also to be taken into account in explaining location-related gaps in computers and communication technology.
Research reviewed in the introduction to this chapter provides evidence that access to computers and communications technology influences opportunities to participate effectively in a range of economic, social, and civic activities. If so, it is important to find out whether parts of the U.S. population are cut off from the emerging information society on the basis of their socioeconomic status. To address this question, the analyses described above sought to learn whether significant differences in access to these electronic media existed in 1989 and 1993 and, if so, what had happened to the size of the gaps over time. Table 2.1 serves as a score card summarizing the results.
Although this score card does not do justice to specific variations along the socioeconomic dimensions studied, it brings to the foreground the main conclusions: There are information society haves and have-nots; membership in these two classes is significantly predicted by income, education, and--to a lesser extent--race/ethnicity, location, and age. Except for gender gaps, these disparities have persisted over a period when the technologies of interest have decreased dramatically in price and increased markedly in user- friendliness.
Summary of Socioeconomic Findings
________________________________________________________________________
Computers at Home Use of Network Services
____________________________ ___________________________
What Happened What Happened
Major Gaps in to Gap Major Gaps in to Gap
1993 1989 over Time? 1993 1989 over Time?
________________________________________________________________________
Income Yes Yes Widened Yes Yes Widened
Education Yes Yes Widened Yes Yes Widened
Race/ethnicity Yes Yes Constant Yes Yes Constant
Age Yes Yes Constant[a] Yes Yes Constant[a]
Sex No Yes Narrowed No No --
Location Yes Yes Narrowed Yes Yes Constant
________________________________________________________________________
[a]Although the results are not statistically significant, the data
suggest that those age 60 and older may be beginning to narrow the age
gap.
More worrisome still, gaps based on income and education have not merely persisted but in fact have increased significantly. There is nothing in the data, then, to suggest that, without policy intervention, these gaps will close.
These conclusions, drawn from a national sample of the U.S. population, are disturbing because sizable demographic subgroups who remain in the have-not segment may be deprived of the benefits associated with membership in the information society. It is also appropriate, therefore, to inquire whether the quantitative data we have examined provide any evidence that the expected benefits are in fact realized. Most findings about benefits, such as those reviewed in the introduction to this chapter, come either from relatively small experimental research samples or from larger studies carried out at the organizational and international level.
For the most part, CPS data are suggestive rather than conclusive since the survey was not designed to address this question specifically. Times Mirror national survey data (1994) more directly bear on it. Results from studies of both datasets, however, tend to corroborate the view that access to computers and communications technology supports informational, affiliative, and participatory outcomes as prior research studies predict.
With respect to information gains, for example, CPS data from 1993 show that 21 percent of adults with household computers use them to access databases (34 percent do so from computers at work). Additionally, 15 percent of adults use home computers for educational programs, as do 39 percent of children with household computers. Even larger percentages of adults as well as children use computers at home to do school assignments (see Appendix A for a table of activities people do by computer). These findings are corroborated by Times Mirror data, which show information-seeking to be one of the two most common activities pursued by people with computers and modems at home. This accounts in part for why such people scored higher on a political knowledge test embedded in the survey than a demographically equivalent sample of nonusers (see above). It should be emphasized that such benefits appear not to be restricted to upper socioeconomic levels. Further, the Times Mirror survey (1994) also reports that younger children's use is almost entirely a function of access: "Among households with PCs, only modest differences were found across racial or income groups in use of computers by children." Results for the teenage survey were the same. The implication is that provision of the technology could go far toward equalizing information benefits across socioeconomic strata.
A second dimension of interest is affiliation, or the extent to which computer-based media yield social contact and support. Some evidence for this thesis comes from the CPS data, which show that apart from word processing, e-mail is the single activity pursued by the largest proportion of adults with household computers (see Appendix A). More direct evidence on this point comes from the Times Mirror survey, which asked more detailed questions about communicative activities. That survey found communicating with other people constitutes a distinct set of activities carried out independently of information-seeking. While acknowledging the importance of information seeking, the Times Mirror report concludes: "of potentially equal significance to society is the quieter revolution of computers facilitating communication between people." These findings give strong support to the view that electronic networks are social technologies that serve affiliative needs. It follows that lack of access can constitute a barrier to association in the information age, constraining opportunities for social interaction in ways that universal service policies could remedy. Further, that affiliation is demonstrably an independently significant function of these technologies gives rise to a "right to information" often cited in discussions of computers and connectivity.
A third key question is whether participation in civic and social life is likely to increase along with use of information and communications technology. No CPS items address this point. Times Mirror survey data are, however, highly instructive. Comparing computer and modem users with their demographic equivalents, the report notes that, aside from working at home more, the largest difference between them is engagement in groups and organizations. Specifically, the users are significantly more likely than their demographic counterparts to belong to a group in which they regularly take part. Further, controlling for other variables, users belong to more groups and are more likely to have worked for or attended a meeting of a group in the past week. Such measures have long been taken by sociologists (e.g., Havighurst, 1973) as indicators of engagement in civic life. These data are consistent with previous smaller-scale research indicating that computer-based interactions supplement and extend, rather than supplant, social participation (Bikson and Eveland, 1990; Hesse et al., 1990; Huff, Sproull, and Kiesler, 1989; Kraut and Streeter, 1990). Broad access to computers and electronic networks, then, might help reduce if not reverse the trends toward disengagement in civic and political affairs discussed in the introductory chapter.
The congruence of findings between national-sample survey data and social science research studies strengthens the conclusions drawn here. While the social science research projects cited here are well-designed and are better able to support causal conclusions than cross-sectional surveys, questions inevitably arise about whether the results will scale up to the national level. The survey data presented here are national in scope and representative of the U.S. population; but although they can establish clear correlations and reveal strong trends, they shed less light on causal or functional relationships over time. Together, the combined sources should be viewed as a robust policy foundation. Both kinds of information would, however, benefit from studies of real world processes in day-to-day contexts. Understanding how outcomes are distributed over time when entire intact communities are networked with computers and communications technology would richly complement what has been learned from social science and from survey research. Chapter Five represents first steps toward providing that part of the picture.
[2]The analyses are done using individual weights that approximately equal the inverse of the probability of being in the sample, adjusted for interview response rates and normalized to add up to the sample size.
[3]Sometimes when computers have built-in modems, the respondents may not be aware of using a modem to access network services. In other cases, respondents are using computers linked to local area networks or hardwired to organization-wide networks that provide access to broader network services (e.g., Internet) without necessitating their use of a modem.
[4]Statistical significance is determined here on the basis of the Pearson Chi-square test. Note that all weights are normalized to add up to the sample size.
[5]For example, suppose that equal use of network services across socioeconomic groups is a political goal. As we shall see below, black individuals tend to use network services to a lesser extent than whites. This may prompt policymakers to direct efforts to increase use of network services to black communities. However, as we shall also see below, low-income individuals too tend to make less use of network services than high-income individuals. As is well known, the average household income among blacks is lower than among whites. It may well be the case, then, that part or all of the racial difference is due to income differentials. To achieve equal use of network services across socioeconomic groups, public funds may then be more effective when directed toward poor communities generally, rather than to black communities specifically.
[6]The procedure is explained in detail in Appendix A. For example, consider the effect of household income. We distinguish four income categories, corresponding to four income quartiles. First, we estimate a multivariate regression model for, say, presence of a computer in the household. Then, for all individuals in the sample, we predict the probability he or she has a computer in the household, under the counterfactual assumption that everyone falls into the bottom income quartile. The average, over all individuals in the sample, is the predicted fraction of low-income individuals with a computer, controlling for all other covariates. Then we assume that all individuals fall into the second income quartile, again compute predicted probabilities of owning a computer, and average over all individuals in the sample. This is repeated for the third and fourth quartiles, yielding a total of four average probabilities ("net" fractions).
[7]Recall that our measure of network use includes use at home or work, but not at school, since no appropriate questions were asked from students. This implies that we are likely to overstate differences in network use across income categories, because students who live away from their parents tend to have low household incomes.
[8]As may be inferred from Figure 2.2, there is no large overlap between use of network services in the home and at work. Use of network services at work by low-income individuals may partially explain why the net disparity in network service use is smaller than in access to a computer in the household.
[9]To date, e-mail use in general, and discretionary use in particular, has typically been limited to high-level positions in organizations (e.g., Krueger, 1993). The extent to which organizations in the United States permit or support access to external networks via internal e-mail systems is not presently known.
[10]Native Americans are oversampled in the CPS. In 1989, 1,408 Native Americans were interviewed; in 1993, there were 1,703.
[11]The treatment of age is determined by the objective of the study. Obviously, the decision to purchase a computer is in part determined by the size and the age composition of the household, but we wish to document socioeconomic differences in access to a computer, not in personal ownership or usage. The connection between presence of a computer in the household and access to it requires only the assumption that the computer is available to all household members--a relatively plausible assumption.
[12]In preliminary analyses, we distinguished as many as eight different age categories. We decided to collapse them into the four categories presented here, because the patterns that emerged were robust to this more parsimonious classification.
[13]It should be noted, however, that this measure of network use probably underrepresents students' access to on-line services because no CPS questions addressed network use at school (see above).
[14]This is not to suggest that sex differences in usage styles and preferences are disappearing. BoardWatch, for instance, reports that its own subscriber poll indicates bulletin board usership is still overwhelmingly male (1994).
[15]As defined by the Office of Management and Budget's June 30, 1984, definitions.