This paper presents an overview of the global diffusion of the Internet as of the third quarter of 1999. We begin with a survey of several Internet diffusion indicators: connectivity, host count, number of Web sites, language distribution, number of users, and compound indices of pervasiveness, geographic dispersion, sectoral absorption, connectivity infrastructure, organizational infrastructure, and sophistication of use. We observe rapid growth everywhere and significant disparities between developing and industrialized nations. Similar disparities are found between have and have-not groups within nations.
Our review of the current state of Internet indicators is followed by a look at infrastructure that complements the Internet. Again, we observe rapid growth accompanied by disparity and note the relationship between wired and wireless telephony, personal computers, and the Internet. The paper concludes with a discussion of the difficulties of measurement and a proposal for the formation of a decentralized community for regularly measuring the state of the Internet.
Why do we wish to track the global diffusion of the Internet? Researchers like Ernest Wilson worry that the rapid diffusion of the Internet into the organizations, cultures, and societies of industrialized nations may widen the multidimensional gap separating them from developing nations, exacerbating an already significant moral and practical problem. A more optimistic hypothesis is that the Internet's flexible, low-cost communication may lead to improved economic productivity, education, health care, entertainment, awareness of the world, and quality of life in developing nations and pockets of poverty within nations, thus reducing disparity. It is arguable that the low cost of the Internet and its ability to transport a variety of data types will lead to its subsuming other media. The Internet is also a decentralized, two-way medium, therefore, conducive to freedom, which, as Nobel-Prize-winning economist Amartya Sen points out, is both a facilitator and constituent of development. Ongoing tracking of Internet diffusion will allow policy makers to plan and scholars to begin testing these hypotheses.
This paper presents an overview of the global diffusion of the Internet as of the third quarter of 1999. We begin with a survey of several Internet diffusion indicators, and it comes as no surprise that we observe both rapid growth everywhere and significant disparities between developing and industrialized nations. These are illustrated in Figures 1 and 2, which could serve as generic depictions of nearly any indicator of Internet diffusion. Similar disparities are found between have and have-not groups within nations.
Our review of Internet indicators is followed by a look at infrastructure that complements the Internet. Again, we observe rapid growth accompanied by disparity, and we note the relationship between wired and wireless telephony, personal computers, and the Internet. We track the state of the Internet and other technology by estimating the values of various indicators and indices. Our measures shape our worldview, and with it the futures we can imagine, but they are imperfect. The paper concludes with a discussion of the difficulties of measurement and a proposal for the formation of a decentralized community for regularly measuring the state of the Internet.
It is common knowledge that the Internet is growing rapidly, perhaps at an unprecedented rate. Let us look at some of the measures in common use: connectivity, number of hosts, number of Web sites, languages used, number of users, and compound indices. In doing so, we will consider the values of these indicators, trends, and disparities among and within nations.
In tracking the diffusion of the Internet, one must choose a balance between breadth and depth. One of the first to track global Internet diffusion was Larry Landweber of the University of Wisconsin, who simply noted whether or not a nation had an international IP link. Landweber produced well-known maps between 1991 and 1997, graphically showing the Net's progress (see Figure 3). Keeping track of only one easily defined variable allowed him to maintain a global perspective at a reasonable cost, but this system was limited by the fact that differences among and within nations were hidden.
A developing nation might have a very slow connection to the Internet. For example, in the spring of 1999, Cuba's total international bandwidth was 832 kb/s, which is less than a home with high-speed DSL service or cable modem and less than 1/50th of the bandwidth from my campus of the California State University to the Internet. Furthermore, connectivity was concentrated in Havana (though less concentrated than in many developing nations) and limited to relatively few people, almost exclusively through their work.
Africa is an even stronger case in point. Figure 4 shows the speeds of connections from African nations. Note that many nations have international connectivity speeds roughly equivalent to a single analog modem. Figure 5 shows that in many African nations, access is available only in one city (shown in white).
A global view of backbone networks further illustrates connectivity disparities. Figure 6 shows the international backbone network of one major provider, UUNET, and Figure 7 shows a composite of 48 national and international backbone networks. Nearly all global connectivity is between North America and Western Europe, Australia, and narrow parts of Asia.
Another commonly cited measure of the growth of the Internet is the number of hosts, computers with domain names. Network Wizards (NW) has reported the number of Internet hosts since 1981 (though they have twice adjusted the counting method for reasons discussed below). As we see in Figure 8, the number of hosts continues to grow rapidly, and that growth has accelerated somewhat during the latest measurement period.
NW currently reports host counts for 251 top-level domains (7 generic and 244 national). Traditionally, the generic domains have been used by organizations in the United States, but that is increasingly not the case, particularly for the com domain.
Table 1 shows the growth rates of the generic top-level domains, which is an indication of new organizations. The commercial and network domains account for the majority of absolute and percent growth. The others are relatively saturated. For example, nearly all U.S. universities and government agencies are now on the Internet. This rapid growth in hosts and organizations may be good news, but host counts also allow us to see discrepancies among nations.
Over 70 percent of the hosts NW identified are in generic top-level domains (39,414,465 of 56,218,330). The majority, but far from all, of these are in the United States. If we focus on the national top-level domains, we see that the top six nations account for over half of the remaining hosts. The Internet Software Consortium (ISC), www.isc.org, makes these data more meaningful by adjusting them for the location of the organization registering the host. Figure 9 shows these adjusted counts on a per capita basis for the Organization for Economic Cooperation and Development (OECD) nations.
It is clear from this figure that even within the relatively homogeneous OECD nations, there are large disparities in hosts per capita. As of July 1999, OECD nations accounted for 93 percent of the hosts. If we examine hosts per capita globally, we find a highly skewed distribution (Figure 10). Figure 11 shows that Africa accounts for a minute portion of the world's hosts.
The growth of the Internet accelerated with the invention of HTTP, the World Wide Web hypertext transfer protocol, and the number of Web sites is an important indicator. This growth is shown in Figure 12, compiled by Netcraft, www.netcraft.com. The Online Computer Library Corporation (OCLC), www.oclc.org, estimated the number of Web sites as 4,882,000 in June 1999. They also estimated there were nearly 300 million pages and over 500 million files on the Web. The discrepancies between the OCLC and Netcraft estimates (see Table 2) are due to differences in their methodologies and definitions.
In August 1999, the OECD analyzed the Netcraft data and estimated that there were over 46,000 secure Web servers in member nations. They estimated that this represented 95 percent of the global total and a 109 percent increase over the previous year. Figure 13 shows the rapid growth of secure servers in OECD nations and also shows the very large per capita disparities. Secure servers are of special interest because they indicate electronic commerce.
The OECD has studied the distribution of languages used on the Web pages they harvest and found a predominance of English, and the imbalance is even more pronounced when focusing on commerce-oriented secure servers (Figure 14). The OCLC found similar results in their 1988 survey, Table 3. They also found that 85 percent of the sites were in English-speaking countries.
Some find this language disparity a cause for alarm and worry that telecommunication will erode culture and eliminate variety. According to Genesis, we began with one language, "one set of words," but people invented a new technology, bricks, and began building the Tower of Babel. God became worried that "nothing they wish will be beyond their reach," and scattered them about the Earth, speaking different languages. Will another new technology, telecommunication, undo His work?
Broadcast telecommunication erodes Babel. TV, radio, film, music, and commercials all cross borders, and Stephen Pinker warns that many of the world's languages are endangered. However, the Internet cuts both ways. For example, 25 years ago, Yiddish was common on the streets of Venice Beach and other parts of Los Angeles, but today the signs and people are gone. Los Angeles can no longer support Yiddish, but a Yahoo search for "Yiddish" turns up 66 site hits, and Altavista finds 50,670 Web pages. Another example of preservation of a dying language is http://www.rcp.net.pe/QUECHUA, a site devoted to the Inca language Quechua. Expatriates the world over find each other and, using their languages, stay in touch with their homelands on the Internet. There is also the issue of quality rather than quantity. For example, sites like Virtual Cervantes, www.cervantesvirtual.com and 1000 Books, http://som.csudh.edu/cis/lpress/netstate/www.mil-libros.com.mx, present high-quality text, discussion, links, etc. on Spanish literature yet count as only two sites in a survey.
Language homogeneity has pros and cons. It increases international understanding and productivity at the cost of loss of conceptual richness. The dominance of English on the Internet today is decreasing, but it may continue as the "lingua franca" of the Net. Paradoxically, if English turns out to be a universal second language, native English speakers will find themselves cognitive second-class citizens with less language diversity than native speakers of other languages.
The numbers of Internet users in a nation is another common indicator. NUA, Ltd., http://www.nua.ie/, estimates the number of users in each nation by compiling and averaging journalistic and market research reports from around the world. Their current estimate is 201 million users worldwide, as shown in Table 4. Their estimates show the same sorts of disparities we have seen above, with over 50 percent in North America.
Computer Industry Almanac Inc., http://www.c-i-a.com/, estimates and projects the numbers of users in nations as a function of the number of personal computers and other determining factors. As is seen in Table 5, they project 717 million users by 2005, and the same sorts of discrepancies we see today.
There are disparities within nations as well as among nations. For example, a recent U.S. Commerce Department survey shows that the proportion of Internet users in the United States varies depending upon race, income level, education level, residential classification (urban, rural, central city), and state (see Figures 15-1 to 15-7). The survey also showed that the race, income, and education-level gaps grew between 1997 and 1998. Figure 16 illustrates this widening divide between races. Note that these data reflect household connectivity, and, if one were to consider connectivity at the workplace, the differences might be even greater since the same groups tend to be underrepresented among knowledge workers. On the other hand, Internet access is often available outside the home in the United States, at public libraries, schools, Internet cafés, etc. This is generally not the case in developing nations.
The preceding sections focus on specific indicators, but it is also possible to construct indices based on several indicators and other qualitative factors. An index may be more robust than an indicator in trying to measure a qualitative concept. For example, the concept of "development" used by the United Nations Development Program (UNDP) is designed to measure human autonomy and breadth of choice, equity, sustainability, and empowerment as well as economic productivity. In an attempt to capture this complex concept of development, UNDP computes a comprehensive Human Development Index (HDI) as a function of productivity, health, and education.
My colleagues and I have defined six such indices in establishing a framework for conducting case studies on the state of the Internet in a nation. These indices are:
Each index may assume five ordinal values from 0-4, and the definitions of the levels draw broad distinctions, enabling a person familiar with the Internet in a nation to make relatively rapid and reliable classifications.
This framework has been applied in 16 case studies and used in a questionnaire of Internet experts in developing nations. This approach provides a more comprehensive picture of the state of the Internet in a nation than a single indicator and therefore better differentiates among developing nations, but the same general disparities we have noted above are still apparent. Table 6 shows the most recent results, a questionnaire study of attendees of the 1999 Internet Society Network Training Workshops for Countries in the Early Stages of Internet Internetworking. The table shows both the respondents' rating for their nation on our six indices and their predictions (increase or remain the same) for the coming year.
We have seen that the Internet is growing rapidly, but there are disparities globally and within nations. This comes as no surprise, since the same is true of nearly any dimension -- economic productivity, health care, education, capital accumulation, etc. Table 7 shows the global disparities in the UNDP HDI mentioned above, and Figure 17 shows the correlation between the HDI and the number of hosts in a nation.
We have focused on the current state of the Internet, but that is in turn influenced by many determining factors, such as government policy and the strength of the economy in a nation. This section looks at a subset of those determinants, complementary technology. We will look at the telephone network, which is often used for Internet data transport; the growing importance of wireless communication; and personal computers, which play several roles in the Internet.
Telephone infrastructure is critical to the Internet, but telephone companies and governments may also see the Internet as a threat to revenue. This is particularly the case in developing nations, where telephone charges may provide a significant portion of government revenue (Figure 18). In cases where the Internet is seen as a threat, government support may be ambivalent at best, but the global trend is clearly in the direction of privatization and liberalization.
We see the same sorts of disparities in telephony as on the Internet. The International Telecommunication Union (ITU) groups nations by income, and Table 8 shows that the majority of humanity resides in low-income nations. In Table 9, we see that the number of main lines per capita in high-income nations is 34 times that in low-income nations. The situation is a little better with regard to payphones, indicating that the access gap is a bit smaller than the installed equipment gap, and saturation has slowed the expansion rate in high-income nations, but the disparities are enormous.
Further evidence of telephonic disparity is shown in Table 10, which contrasts nations averaging less than 5 minutes of international telephony per capita with those averaging over 100. The numbers of minutes range from .3 in Bangladesh to 597 in Luxembourg. The oft-heard claim that 80 percent of the people on earth have never heard a dial tone may be accurate.
Many hope that wireless technology -- both terrestrial and satellite -- will enable developing nations to modernize quickly. Table 11 looks at cellular telephony, and the picture is somewhat less grim than for wired connectivity. We see the same huge gap in installed base, but the low-income nations are adding new subscribers at a faster rate than the high-income nations, although the growth rate is declining. The cellular boom in low-income nations is fueled by relatively low capital costs for infrastructure and declining handset prices coupled with prepaid billing offerings that enable companies to service poor people without assuming credit risk. We also note that, since they came online later, developing countries have often skipped early analog systems and have a high rate of digitization.
Mobile communication is growing at an increasing rate and represents a growing proportion of global telephony (Figure 19), and the ITU expects this trend to continue. It expects cellular to become predominant within the next decade (Figure 20), and several developing nations are already among the leaders in the ratio of cellular to fixed telephones (Figure 21).
To get a sense of the implications of this for the Internet, we need to consider the evolution of cellular telephony. Table 12 shows three generations of cellular deployment. The current installed base is too slow for interactive Internet access but is sufficient for store-and-forward e-mail (or fax), a very important application. The next generation is fast enough to allow interactive applications, particularly to fixed locations (as opposed to moving vehicles). The standards for digital data at rates up to 2 mb/s are now being defined, and, with the ITU and the European Community taking the lead, rollout is expected to be underway in three years. The United States will be several years behind that and developing nations further behind. Still, it seems market demand and falling prices will eventually attract the investment necessary to build cellular networks in the urban areas of even poor nations.
But what will happen in villages, in remote areas?  Nearly 40 percent of the world population lives in rural areas of nations with low-income economies. Microfinancing, as pioneered by Grameen Bank, may offer a solution. Grameen makes small loans to the poorest people, typically women who do not own land, in villages. Its model has succeeded (Figure 22 shows Grameen's Internet-like growth) and been emulated by many others. One of Grameen's current programs is providing loans for the purchase of cellular telephones. The loan for a telephone is paid off at approximately $3.50 per week from a 2-cent-per-minute profit on calls. This service has not spread as quickly as planned because of the difficulty in funding and establishing wireless cells near villages, but it holds hope. There are other experiments with wireless connectivity to villages, for example, in Pondicherry, India; however, as we have seen, today's terrestrial wireless technology and standards are too slow to support interactive Internet access, and it will be many years before faster equipment is available in poor villages.
Satellite connectivity may offer a solution. Consider, for example, Teledesic's planned network. In 2003, Teledesic, http://som.csudh.edu/cis/lpress/netstate/www.teledesic.com, plans to begin offering connectivity through a network of 288 satellites (Figure 23). These low-earth-orbiting satellites will house IP routers with algorithms that take into consideration their constantly changing locations.
Teledesic ground stations will be capable of up-link speed up to 2 mb/s and down-link speed up to 64 mb/s. Speed selection will be under program control so Teledesic will be able to offer user-variable, guaranteed quality of service (QOS) between any points on earth. If it were operating today, its QOS guarantee would give them a competitive advantage in developed nations, where there is current demand for such service. (It would be a lesser advantage in major metropolitan areas, where the distance to a point of presence capable of guaranteeing QOS would typically be lower than in a more remote area). As terrestrial backbone operators deploy technology and make exchange and settlement arrangements allowing them to guarantee QOS, this competitive advantage will diminish in developed nations, but Teledesic would still enjoy a monopoly in developing nations, where there is little terrestrial connectivity, particularly outside of major cities.
Discriminatory pricing -- having different prices for different classes of customer -- is to a firm's advantage as long as its customer groups have different demand curves and the markets can be kept separate, avoiding arbitrage. Thus we see the prices of children's haircuts or movie tickets are lower than the prices of adults'. The demand curve for communication services is clearly different in a Bangladeshi village than in New York or even Dhaka, but, as the telephone companies have learned from callback operators, arbitrage may be difficult to combat. If Teledesic could keep the markets separated, it would be to its advantage to cut prices in developing nations.
Teledesic could discriminate on the price of ground stations or service or both. There are many joint pricing models, but one option would be to sell the ground stations at a profit and offer unlimited free traffic over low-QOS links (sell the "razor" and give the lowest-quality "blades" away). A kiosk operator in a remote village could then offer unlimited e-mail and file transfer and several interactive sessions at marginal cost. While several years away from setting prices, Teledesic estimates that the price of a ground station will be around $4,000. A microcredit enterprise could amortize such an investment. Various geostationary satellite systems are also hoping to offer prices in this range.
As demand arose for higher QOS, for example in handling voice traffic, a usage-based charge could kick in, but the price in a developing nation would still be below that in a developed nation. Teledesic would be seeding expansion without losing money since the marginal cost of handling low-QOS traffic would be zero. (Presumably the routers over a developing nation at any point in time would have excess capacity.) Low prices would encourage usage, leading to increased productivity and the invention of new applications, which would fuel demand. The value of a network connection would also increase as new users and services were added, forming positive feedback and encouraging growth. Low prices would also discourage telephone companies, railroad operators, and others from entering the backbone market.
Personal computers (PCs) are a strategic, rapidly growing segment of global information technology. They are central to the deployment of the Internet in several ways. First, they are used for access to the Internet. More important, people with PC experience can rapidly become Internet users, and trained, demanding users are perhaps the key driver of Internet growth. Networks require technicians as well as users, and the availability of PCs in universities and other areas is critical to training and development. Finally, in developing nations, PCs are often used as networking equipment. PCs running Unix are often used as time-shared hosts for shell accounts, Web servers, mail servers, routers, etc.
Tables 13 and 14 show the numbers of PCs as reported by the ITU and the Computer Industry Almanac. We see the same sorts of disparities here as in other areas. The top 15 countries in the Computer Industry Almanac study account for nearly 79 percent of the worldwide PCs, and the United States accounts for 35.4 percent. We also see rapid growth. The Computer Industry Almanac reports growth from 98 million PCs in 1990 to 222 million in 1995 and over 364 million in 1998. It should be noted that these growth figures are understatements in the sense that the power of PCs approximately doubles every year, so the processing power of the installed base grows faster than the number of machines.
When we read a gross domestic product or number of Internet hosts in an article or government report, we tend to accept it as a fact. We assume the indicator is well defined and accurately measured. But this is not always the case, and we must accept difficulty and imperfection. This section outlines several difficulties in tracking Internet growth and concludes with a proposal for ongoing measurement.
Automated data gathering techniques are vulnerable to changes in technology. The NW host counts offer an important example because they are often quoted and used in others' research. The NW counting procedure has changed twice, in 1986 and 1995. Until 1986 it was possible to count the numbers of hosts by examining host tables. With the introduction of the domain name system and the spread of the Internet, that became impossible, and a program that "crawled" from network to network issuing a "show all names" command to each domain name server was used until 1995. But increasing numbers of system administrators began configuring their servers to refuse these requests for all names, causing an undercount, and NW switched to a counting program that generates the set of possibly assigned names and tests to see whether they have in fact been assigned. NW adjusted its numbers back for several periods, estimating what the count would have been had the new method been used.
Even this counting method misses the hosts behind organizational firewalls, which are hidden for security purposes. This is a rapidly growing segment of the Internet that is not accounted for by NW. Additionally, the Yankee Group (http://www.yankeegroup.com/), a market research firm, forecasts that there will be 10 million networked homes in the United States by 2003. These too will be invisible to the NW host count. Furthermore, increasing numbers of users and small networks connect to an ISP with a dynamically assigned address that does not have a host name. This is not to criticize the NW host counts but to keep in mind what they do and do not measure.
Only a few years ago, the Internet was used primarily for text-based applications in education and research, and most users connected through accounts on time-sharing systems that were on the Internet. At that time, we speculated that the average number of users per host was around 10. Today, most users connect via a PC that is directly on the Internet, and tomorrow it is expected that a typical user may have more than one Internet access device and homes will have many. These changes make comparing and even estimating user counts very difficult.
The studies mentioned in this paper range in cost. Landweber's data collection costs very little and was therefore able to scale to global coverage. An automated technique like the NW host counts is also relatively low cost. Sampling studies like those done by OCLC are more expensive than these because human follow-up is required to analyze the sites selected for the sample. In-depth case studies of the state of the Internet in a nation, such as those carried out by the MOSAIC Group, or the U.S., Canadian,  British, or Australian governments are probably most expensive, and it would be prohibitive to repeat them on a regular basis for every nation. The same holds for country-specific studies done by commercial market research firms. Government and market research studies are costly and are justified by expected commercial return, and the latter are generally not affordable to the research and policy communities, particularly in developing nations.
With the exception of Landweber's simple noting of connectivity or none, the indicators we have discussed are difficult to define. What constitutes a "user"? Is it someone who reads e-mail once a week at a Hotmail account or someone with DSL at home, a T1 at work, and a wireless connection to a personal digital assistant (PDA) who is never offline? Are we interested only in people who spend a certain amount of time online or use interactive applications in addition to e-mail? Some surveys interview ISPs and count accounts, but how many people share an account? How many people in a nation use schools, work, community centers, public libraries, or commercial cybercafes to access e-mail and surf the Web?
And what of the notion of a Web site? My university has perhaps a dozen computers running Web servers. Several of these have independent sites for hundreds of faculty members, courses, research projects, etc. A corporation may have sites for making new sales, current customers, suppliers and distributors, internal employees, etc. If the sites share a domain name, should they be considered one site or many? How many Web sites are there at GeoCities? A single domain name, for example, yahoo.com, may address a server "farm" with many Web servers.
And what is it that an indicator or index is trying to measure? For example, gross domestic product (GDP) is commonly cited as an indicator of economic well-being in a nation. But, GDP, like any indicator, may miss the mark in what it measures. For example, rising GDP might be accompanied by environmental damage, anger over growing disparity in income distribution, disappointment when expectations rise faster than they are fulfilled, displacement of traditional values and customs, crowded cities, and so forth. GDP also counts many painful transactions like bypass surgery, buying a second home after a divorce, or the paycheck of a housewife who is forced into the labor market to make ends meet, as positive factors. In spite of these limitations, we define and measure GDP as carefully and consistently as possible, and nearly all nations report its value annually.
Indices are also problematic. Ideally, they should define operational, mutually exclusive levels that are sufficiently separated that any person would assign the same value to any index for any nation. Our experience indicates that that is not always the case. Differences lead to discussions among the raters and may result in either revised opinions or adjustments to the definitions of the indices and their levels. A set of Internet indices should also be orthogonal, each measuring an independent aspect of the state of the Internet in a nation, but it is difficult to define indices that are both comprehensive and uncorrelated.
We have seen various methods of conducting studies. Some, for example Landweber, the Cooperative Association for Internet Data Analysis (CAIDA), and our own questionnaire studies rely upon voluntary contributions of data. Others, for example the NW host counts, are automated and exhaustive. OCLC is automated but uses sampling. Case studies require visits to nations, interviews, and gathering of data from operating companies (like ISPs). Each of these approaches has advantages and disadvantages. Our goal is global coverage using precisely defined, consistently applied measures on a regular basis. To achieve or even come close to achieving that goal will require a new approach.
We suggest the establishment of a stable community of experts representing each nation for the ongoing tracking of the state of the Internet. This community would be responsible for the definition of indicators and indices and their periodic measurement. The experts would own the process and the data. Definitions of indicators would be publicly available and open to viewing and discussion, but their maintenance and control over changes would be the prerogative of the tracking community. This is analogous to open source software, in which many participate and contribute but a smaller group considers these contributions and maintains version control.
Community members would also be responsible for periodically reporting the values of these indicators and indices for their nations. We would prefer several respondents per nation, perhaps representing different communities such as the university, government, and industry. If there were, say, three members per nation, they would report their values independently, then resolve differences in a timely manner through discussion among themselves. These small "delphi studies" would lead to more accurate evaluations and point out the need for adjustment of the definitions of the indicators and indices. Once the values were complete, they would be published online.
A decentralized approach with respondents in each nation is well suited to the Internet, which provides low-cost communication and has a culture favoring community control of open standards and content over proprietary solutions. It could scale to cover the globe at relatively low cost and would provide a reasonable base of ongoing measurement to guide policy makers and scholars.
Researchers like Ernest Wilson worry that the rapid diffusion of the Internet into the organizations, cultures ,and societies of industrialized nations may widen the multidimensional gap separating them from developing nations, exacerbating an already significant moral and practical problem. Is the Internet gap truly growing? How does the Internet affect development (in a broad sense)? Can the rapid growth of the Internet be sustained in the face of growing disparity, or will that disparity disrupt the social and institutional environment that enables, among many things, that growth?
A more optimistic hypothesis is that the Internet's flexible, low-cost communication may lead to improved economic productivity, education, health care, entertainment, awareness of the world, and quality of life in developing nations and pockets of poverty within nations, thus reducing disparity. Amartya Sen points out that freedom is both a facilitator and a constituent of development. He argues that many forms of freedom are conducive to economic development and those same freedoms are basic constituents of human development. Perhaps communication is freedom's lubricant. It is arguable that the low cost of the Internet and its ability to transport a variety of data types will lead to its subsuming other media, and, since it is a decentralized, two-way medium, it is conducive to freedom.
Ongoing annual reports on the state of the Internet would help us understand its importance and help answer questions raised by scholars like Wilson and Sen.
The author thanks Bob Anderson of the Rand Corporation for comments and support.
 For an example of disparity within a nation (the United States), see National Telecommunications and Information Administration, Falling Through the Net: Defining the Digital Divide, July 1999, http://www.ntia.doc.gov/ntiahome/digitaldivide/.
 Some claim the Internet is growing more rapidly than earlier information technologies like radio or television, but such claims depend upon the choice of a starting date for the Internet and the definitions of growth measures.
 Martinez, Jesus, The Net in Cuba, Matrix News, Vol. 1, No. 1, January 1999.
 Jensen, Mike, http://www3.sn.apc.org/africa/afrmain.htm.
 These figures were produced by the Cooperative Association for Internet Data Analysis, http://www.caida.org/, and the program that generates them is at www.caida.org/Tools/Mapnet/Backbones/.
 Network Wizards is now a member of the Internet Software Consortium, which conducts the survey and further analyzes the data: www.isc.org/dsview.cgi?domainsurvey/index.html.
 OECD, Internet and Electronic Commerce Indicators Update,
 OECD, Internet and Electronic Commerce Indicators Update,
 This figure was produced by the author using population data from the World Bank and host data from Matrix Information and Directory Services, which, as a member of the Internet Software Consortium, geographically locates the hosts NW discovers. Note that unconnected nations were not considered in computing the percentile cutoff points, and several of the nations that were unconnected in 1998 are now connected.
 Jensen, Mike, http://www3.sn.apc.org/africa/afrmain.htm.
 Netcraft, http://www.netcraft.com/survey/Reports/current/graphs.html. Note that Apache, a nonproprietary, open source program, is the most popular Web server.
 Online Computer Library Corporation, http://www.oclc.org/oclc/research/projects/webstats/statistics.htm.
 OECD, Internet and Electronic Commerce Indicators Update,
 OECD, Internet and Electronic Commerce Indicators Update,
 OCLC Web Characterization Project, http://www.oclc.org/oclc/research/projects/webstats/
 Lavoie, Brian, personal communication.
 Pinker, Stephen, The Language Instinct, MIT Press, 1995.
 NUW, Ltd., http://www.nua.ie/surveys/how_many_online/index.html.
 Computer Industry Almanac, http://www.c-i-a.com/199908iu.htm.
 National Telecommunications and Information Administration, Falling Through the Net: Defining the Digital Divide, July 1999, http://www.ntia.doc.gov/ntiahome/digitaldivide/.
 UNDP, United Nations Development Programme Report on Human Development, Oxford University Press, Oxford, 1999.
 The HDI is an evenly weighted average of three indices, life expectancy, educational attainment, and real gross domestic product (GDP)/capita. Life expectancy is a percent of 85 years, educational attainment is a linear function of adult literacy rate and school enrollment rate, and real GDP/capita is adjusted for purchasing power parity. GDP/capita is also nonlinear -- amounts above the world average ($5,120 in 1992) are deemed to have decreasing utility, up to a maximum of $40,000. While limited, and surely not worthy of the 3-digit significance reported by the UNDP, the HDI is more reasonable than simple GDP/capita.
 Goodman, Sy, Burkhart, Grey, Foster, Will, Press, Larry, Wolcott, Peter, The Global Diffusion of the Internet Project: An Initial Inductive Study. MOSAIC Group, Arlington, VA, March 1998, http://www.agsd.com/gdi97/gdi97.html and Press, Larry, Burkhart Grey, Foster, Will, Goodman, Sy, Wolcott, Peter, and Woodard, Jon, An Internet Diffusion Framework, Communications of the ACM, Vol. 41, No. 10, October 1998, pp. 21-26, http://som.csudh.edu/fac/lpress/articles/acmfwk/acmfrwk.htm.
 For results of earlier case and questionnaire studies and further background on the framework and project, see som.csudh.edu/fac/lpress/gdiff. See http://som.csudh.edu/cis/lpress/netstate/www.agsd.com/gdiff/gdiff2 for recent examples of case studies of India and China. For a comparison of those cases, see Press, Larry, Burkhart, Grey, Foster, Will, and Goodman, Sy, The Internet in India and China, INET'99, San Jose, CA, June 1999.
 Compiled by the author using HDI data from the UNDP and host counts from Matrix Information and Directory Services (adjusted values of the NW counts).
 Primo Braga, Carlos A., Forestier, Emmanuel, Stern, Peter A., Developing Countries and Accounting Rates Reform, A Technological and Regulatory El Niño?, Public Policy for the Private Sector, World Bank, Washington DC, February 1999, http://www.worldbank.org/html/fpd/notes/173/173braga.pdf.
 Press, L., What May the WTO Telecommunication Agreement Mean for Emerging Nations? OnTheInternet, Vol. 3, No. 3, May/June 1997, pp. 36-38.
 Greg Staples, Editor, Telegeography 97/98, Telegeography, Washington DC, 1998, www.telegeography.com.
 ITU, Digital Mobile Growth, http://www.itu.int/plweb-cgi/fastweb?getdoc+view1+www+33368+9++GPRS and ITU, World Telecommunication Development Report, International Telecommunications Union, Geneva, 1999, http://www.itu.int/ti/publications/wtdr_99/wtdr99.htm.
 ITU, World Telecommunication Development Report, International Telecommunications Union, Geneva, 1999, http://www.itu.int/ti/publications/wtdr_99/wtdr99.htm. For more detail on evolving standards, see the Third Generation Partnership Project, http://www.3gpp.org.
 Press, L., Connecting Villages, OnTheInternet, Vol. 5, No. 4, July/August 1999, pp. 32-37, http://som.csudh.edu/fac/lpress/articles/villages.htm.
 World Bank World Development Report, 1997, Oxford University Press, Oxford, 1997.
 For further information on worldwide microcredit, see www.soc.titech.ac.jp/icm/icm.html.
 Press, Larry, A Client-Centered Networking Project in Rural India, OnTheInternet, January/February 1999, pp. 36-38.
 This hopeful scenario for Teledesic or other satellite operators might seem at odds with the recent bankruptcy of the Iridium satellite effort, but Iridium was in the mobile voice business, while Teledesic and others are targeting high-speed IP data.
 For a discussion of PCs, telephone infrastructure, human resources, and other factors required to support networking in a developing nation, see Press, L., Developing Networks in Less Industrialized Nations, IEEE Computer, Vol. 28, No 6, June 1995, pp. 66-71, http://som.csudh.edu/cis/lpress/ieee.htm.
 Press, L., Beyond the Personal Computer, Communications of the ACM, in press.
 Emerging Digital Economy II, http://www.ecommerce.gov/ede/ede2.pdf.
 Pierlot, Paul, Electronic Commerce and Internet Service Providers in Canada, INET'99, http://www.isoc.org/inet99/proceedings/1g/1g_1.htm.
 The Information Society, http://www.isi.gov.uk/isi/isiframe.htm.
 Creating a Clearway on the New Silk Road, http://www.dfat.gov.au/nsr/index.html.
 For discussion of these definitions, see O'Neill, Edward T., McClain, Patrick D., Lavoie, Brian F., A Methodology for Sampling the World Wide Web, OCLC, 1999, www.oclc.org/oclc/research/publications/review97/oneill/o'neillar980213.htm and Lavoie, Brian, Nielsen, Henrik Frystyk, Web Characterization Terminology & Definitions Sheet, W3C Working Draft 24 May 1999, http://www.w3.org/1999/05/WCA-terms/01.
 Cobb, Clifford, Halstead, Ted, Rowe, Jonathan, If the GDP Is Up, Why Is America Down? The Atlantic Monthly, Vol. 276, No. 4, October 1995, pp. 59-78.
 Wilson, Ernest, Meeting the Challenges of Internet Inequality, OnTheInternet, Vol. 5, No. 6, November/December, 1999, pp. 26-30.
 Press, L., The Role of Networks in Developing Nations, Communications of the ACM, Vol. 39, No. 2, February 1996, pp. 23-29, http://som.csudh.edu/fac/lpress/devwins.htm.
 Sen, Amartya, Development as Freedom, Random House, New York, 1999.