See, again, Gort and Klepper (1982) and Agarwal (1998).

Figure 2

Number of FM Radio Stations: 1941-65

Number of Stations 1400

0 1935

1950 Year

SOURCE: Sterling and Kittross (2002).
The Diffusion of FM Radio
For an example of an industry life cycle generated by technological improvements, consider the commercial diffusion of FM radio broadcasting shown in Figure 2. Much like the Internet, FM technology provided a new medium for broadcasting and opened up a business opportunity for both new and existing radio stations, which could make profits by airing advertisements. In 1941, the year of the first authorization for commercial FM stations, only five stations were in operation. But the number of stations increased steeply after World War II, peaking in 1950, as the business opportunity was aggressively pursued by both new FM stations and the established AM stations diversifying into FM broadcasting. By 1949, about 85 percent of FM stations were owned by existing AM stations. The AM stations used FM stations frequently as an insurance against a possible demise of the AM technology and at the same time to deter entry by independent FM broadcasters. A shakeout followed between 1950
and 1957, during which 203 stations, about 28 percent of all stations at the peak, shut down. Thereafter, the number of stations rebounded and continued to grow steadily.6 A similar pattern of early mass entry and shakeout was observed in the diffusion of AM radio and television stations, but the extents of the entry and the shakeout, their durations, and the reasons driving them were not the same. For example, in the case of AM broadcasting, the main force behind the shakeout was the regulation placed on broadcasting frequencies. In the case of FM stations, the reasons were uncertainty about the future of FM technology, lower-than-expected interest in the new medium from advertisers, competition from AM and television stations, and some conflicts arising from joint ownership of AM and FM stations. Such conflicts were also
In many industries, there is no such post-shakeout growth in the number of firms. The growth in the number of FM stations postshakeout is probably a consequence of the fact that FM stations are local in nature, and growth in local population over time may have led to an increase in the variety and number of such stations.
pertinent in the early experience of the Internet. That AM stations embraced FM technology to take advantage of synergies, as well as to deter entry by independent FM stations, is similar to the clash between entirely Internet-based retailers and traditional retailers adopting the Internet as a sales channel.

In a single-product firm, economies of scale indicates declining per-unit costs as the number of units produced increases; in a multiproduct firm, economies of scope indicates cost-saving synergies among different product lines. Dinlersoz and Pereira (2004) provide a theoretical analysis of how these factors may affect adoption incentives for established versus new firms.
See Dinlersoz and Yorukoglu (2004) for an analysis of how improved methods of communication have affected firm and industry dynamics. See Agarwal and Gort (2001) for potential explanations for this phenomenon. For instance, it took approximately 45 years for electricity to reach 20 percent of American households, 35 years for the telephone, 25 years for the television, and 15 years for the personal computer.
beginning, the tendency to adopt was quite different for two groups of retailers: existing retailers with established traditional market functions and facilities compared with entirely new entrepreneurs who had no traditional market presence. Even though the website-design technology was available at a low cost to almost anyone who wanted to start a retail business, the cost of investing in warehousing and distribution facilities, which are required for large-scale retail operations, is high in some sectors. Established retailers in such sectors seemed to have an edge with respect to new entrepreneurs, so it is surprising that they were the latecomers.12 The reluctance of existing retailers to diversify to the Internet market stemmed partly from the potential problems associated with harmonizing traditional and Internet retail channels, giving rise to channel conflict. This conflict comes in many forms, including the resistance of the firms traditional operations and subunits to the possibility of being replaced by the Internet, the incentives for free riding by traditional market rivals on the product information and related services provided directly on the firms website, and the possibility that a firms business on the Internet might compete for its own clientele in the traditional market.13 Nevertheless, channel conflict currently appears to have lost its role as a major concern in deterring existing retailers from diversifying. Eventually, for well-known traditional retailers, their established names, their ability to raise funding to finance new ventures, and their existing warehousing and distribution facilities allowed them to enter the Internet market strongly. In some product categories, however, the largest online sales today are still made by pure online retailers and by manufacturers selling their products directly, rather than by diversified traditional retailers.14

During its emergence and early growth, Internet retailing was largely free of regulation. However, one important and persistent policy has been the absence of taxes. Like catalog retailing, Internet commercial activity is free of tax as a result of a moratorium initiated in 1998 that continues to apply. While there has been no other special infant industry protection program for Internet retailing, the no-tax environment clearly encouraged the growth of the industry by favoring Internet firms over local firms. Goolsbee (2000) provides preliminary estimates that imposing taxes would have reduced the sales on the Internet by 25 to 30 percent.15 The evolution of this industry was therefore positively influenced by the absence of taxes. In addition to aiding the growth of Internet retailing, the tax-free environment had some implications for the location of Internet retailers sales offices and warehouses. Since the shipments within the state where the firm is physically located are subject to local taxes, there are incentives to avoid populous states. However, the tax break neither changed the main course of the industrys evolution nor prevented the shakeout. With taxes, we would have probably observed fewer sales and a smaller number of firms, but no major changes in the trends.
Some Effects of the Internet on Retail Industry Structure
The Internet is a hybrid medium that is capable of combining two basic methods of exchanging information in a market: advertising and shopping around. The reach of the Internet makes these two functions truly global. As a consequence, the location of demand has less influence on retailer location. The geographic separation between the locations of demand and supply can increase the scale and scope of a retailer. Internet retailers that can dominate the market in a certain category of products are also able to easily expand their operations into other categories. Amazon.com is a good example. Amazon started as a book retailer but now sells many different products. This replicability or expandability,
Some Internet-based firms, however, overcame this difficulty by using a method called drop-shipping, which allowed them to use manufacturers to ship products on their behalf. This reduced the investment needed in warehousing and shipping in some cases. See, for example, Carlton and Chevalier (2001), Shaffer and Zettelmeyer (2002), and Dinlersoz and Pereira (2004). For instance, in books, Amazon.com has a much higher share than the traditional retailer Barnes and Noble. See Latcovich and Smith (2001).

Also see Ellison and Ellison (2003) for a smaller-scale, but morerecent, analysis of the effects of sales tax on Internet retailing.
in some cases through linkages with traditional retailers, is due to the fact that adding a new product to the existing set of products is probably much easier and cheaper on the Internet. Basically, all that needs to be done is to create digital space for the new product on the website and physical space in the warehouse. Big Internet firms such as Amazon.com have a much wider range of products than traditional big firms, such as Wal-Mart. In addition to the availability of lower prices, the proliferation of varieties on the Internet is a key feature that increases consumer welfare.16 Besides enhancing search and advertising, the Internet also offers interactivity. Unlike other media, it allows for a two-way exchange of information between consumers and firms and can also be used to record and store this information the various steps of this exchangefor future use. This latter feature of the Internet is especially useful for retailing because it makes it possible for firms to learn about consumers preferences by analyzing their shopping patterns. This type of information extraction works in favor of customization of goods and services to satisfy finer individual tastes. In this respect, the Internet is an advanced form of the scanner technology used at the checkout counter that previously revolutionized retailing by allowing firms to monitor what consumers bought. The Internet also enables firms to target consumers individually or in small groups, unlike other communication tools, such as radio and television, which can at best target large, coarsely defined groups of consumers. The Internet also offers firms the possibility to monitor rival firms strategies more closely, especially their prices and promotional efforts, making it easier for firms to respond quickly to changes in rivals strategies. The costs of pricing products and adjusting prices, referred to as menu costs, appear to be much lower on the Internet.17 This feature is likely to speed up the pace of competition in retail markets.
What will be the main characteristics of retail industries on the Internet in the future? Will the industry structure look more like a competitive industry or a monopolistically competitive one, with many small firms each serving a particular niche in the market? Or will it be more concentrated with a few large firms dominating the market for a particular product type or many product lines simultaneously? It is too early to answer this question convincingly. Clearly, there are features of the Internet that can promote entry, competition, and fragmentation. Initially, it was believed that low entry costs associated with operating a website might foster entry and competition. However, the Internet also provides an environment in which the scale and scope of operations can be expanded at very low cost and information about a firms attributes can be disseminated easily; it also can give rise to firms that can quickly become large. These features can lead to high concentration. While some early findings suggest that industry concentration ratios on the Internet were initially much higher than their traditional market counterparts, there is no overwhelming evidence that this is the case. In one of the earlier studies, Latcovich and Smith (2001) find that industry concentration is much higher on the Internet than in the traditional market in the case of book and music retailing. The authors also report that advertising and promotion efforts are more intense on the Internet compared with the traditional market. Thus, post-entry sunk costs in the form of investment in advertising and customer loyalty programs may be an important aspect of competition. Such investments have the potential to deter entry and lead to a highly concentrated market structure.18 In a more comprehensive study, Noam (2003) also points to high concentration, as measured by the Herfindahl-Hirschman index (HHI), in several industries for the pre-2002 period.19 He finds that the Internet sectors overall concentration was high, and concentration initially declined

The Growth of Retail E-commerce Sales
Despite the shakeout, retail e-commerce sales have been growing at a steady pace over the years, as shown in Figure 3. While the current share of retail sales accounted for by e-commerce is still very low (around 2 percent), its growth rate is very high. As total retail sales grew at an average rate of 1.3 percent quarterly over the sample period, e-commerce sales exhibited an average growth rate of 8.6 percent. The strong seasonality in e-commerce sales is also apparent from Figure 3, with fourth quarters exhibiting exceptional growth, due to the surge in online shopping during holiday seasons. The sectoral breakdown of the share of retail e-commerce sales is shown in Table 1. In almost all sectors, the share in 2002 was less than 1 percent, and the differences across sectors were not highly perceptible. Table 2 presents the percentage of sales accounted by e-commerce by merchandise line, considering only the firms classified as electronic and mail-order houses. The electronic and mail-order houses industry includes all catalog and mail-order houses and other direct retailers, many of which sell in multiple channels, as well as pure Internet-based firms and hybrid brick-and-click retailers, if the e-commerce group operates as a separate unit and is not engaged in the online selling of motor vehicles. The diffusion of e-commerce sales was relatively rapid and widespread among electronic and mail-order houses compared with other retail sectors, and differences across merchandise lines in the share of e-commerce are more visible in this industry. In 2001, the highest shares were observed in books and magazines, electronics, and music and videos. Relatively low shares were observed in food, beer and wine, clothing and apparel, and drugs.21 These observations make clear that the nature of the product matters for the extent of the diffusion. However, the differences across categories are expected to vanish over time as both sellers and buyers experiment with various product types
The n-firm concentration ratio is defined as the market share accounted for by the n largest firms in the market.
Part of the lack of growth observed in beer and wine e-commerce sales is probably related to the restrictions set on interstate shipments of alcohol by many states.

Figure 3

Growth of Total Retail Sales Compared with Growth of E-commerce Sales (millions of dollars)
Total Retail Sales 950,000 E-commerce Sales 18,000 17,000 16,000 900,000 15,000 14,000 850,000 13,000 12,000 11,000 800,000 10,000 9,000 750,000 Total E-commerce 700,000
:Q 20 :Q 20 :Q 20 :Q 20 :Q 20 :Q 20 :Q 20 :Q 20 :Q 20 :Q 20 :Q 20 :Q 20 :Q 20 :Q 20 :Q 20 :Q 20 :Q 20 :Q :Q 2 19

Standard error

1.4 0.9 0.9 0.2 (S) (S) (Z) 0.3 0.8 (S) 0.7 18.7 28.1
1.1 0.6 (S) 0.8 0.2 (S) (S) (Z) 0.2 0.6 (S) 0.5 15.0 23.0
(Z) (Z) (S) 0.2 (Z) (S) (S) (Z) (Z) 0.1 (S) 0.1 0.3 0.3
(Z) (Z) (S) 0.1 (Z) (S) (S) (Z) (Z) 0.1 (S) 0.1 0.2 0.3
100.0 16.3 (S) 1.8 1.4 (S) (S) (Z) 1.1 1.5 (S) 1.5 74.8 72.7
100.0 26.2 2.9 2.8 9.3 15.2 5.6 7.6 5.3 2.5 14.0 3.2 5.5 3.5
Furniture and home furnishings stores (S)
NOTE: Reproduced from Tables 5 and 5A in the U.S. Census Bureaus E-commerce Multi-sector Report. 1Estimates are based on data from the U.S. Census Bureau, 2002 Annual Retail Trade Survey. Sales estimates are shown in millions of dollars; consequently, industry group estimates may not be additive. 2 Estimates include data for businesses with or without paid employees and are subject to revision. 3 Estimates are not adjusted for price changes. For information on confidentiality protection, sampling error, nonsampling error, sample design, and definitions, see www.census.gov/eos/www/restats.html. (S) Estimate does not meet publication standards because of high sampling variability or poor response quality. Unpublished estimates derived from this table by subtraction are subject to these same limitations and should not be attributed to the U.S. Census Bureau. (Z) Sales estimate is less than $500,000 or percent estimate is less than 0.05 percent.
reminiscent of the way local markets were once reshaped by the entry of Wal-Mart stores and other dominant chains.
SERVICES AND THE INTERNET
Services industries have also been embracing the Internet rapidly, even though the overall share of e-commerce in total revenues is still below 1 percent, as shown in Figure 4. In some ways, the affinity between the Internet and services industries is not very surprising. Services industries
in general have been quick in adopting the basic technologies such as computers and Internet access. Moreover, since many service products are essentially information goods that come in digital form, they can be easily traded online. Examples are publishing services, information services, travel reservations, and even mortgage lending and stock trading. Such goods that can be traded in digital form are bound to become dominant categories in online retailing, as argued by Dinlersoz and Pereira (2004), because they can be conveniently delivered and returned by e-mail, they can bypass wholesale and retail layers, they

Table 2

U.S. Electronic Shopping and Mail-Order Houses1Total and E-commerce Sales by Merchandise Line2
E-commerce as % of total sales Percentage3 Merchandise line Total electronic shopping and mail-order houses (NAICS 454110) Books and magazines Clothing and clothing accessories (includes footwear) Computer hardware Computer software Drugs, health aids, and beauty aids Electronics and appliances Food, beer, and wine Furniture and home furnishings Music and videos Office equipment and supplies Sporting goods Toys, hobby goods, and games Other merchandise4 Nonmerchandise receipts5

pij (1 pij ) Ni

Table 5
Ranking of Internet-Based Processes by Their Rates of Adoption in Manufacturing Industries
Process Basic Internet access and degree of access Access to vendors products or catalogs Ordering of materials and supplies Product descriptions or online catalog for external suppliers Ordering from vendors Inventory data for other company units Ordering by customers Order status for other company units Customer support Product descriptions or online catalog for other company units Order status for external suppliers Acceptance of orders for manufactured products Payment by customers Product descriptions or online catalog for external customers Payment to vendors Outsourcing of research and development Bidding Inventory data for external suppliers Electronic marketplaces linking specialized business buyers and suppliers Order status for external customers Inventory data for external customers Average rank 21 Average adoption rate 0.84 0.48 0.41 0.35 0.31 0.30 0.25 0.24 0.22 0.20 0.17 0.17 0.14 0.12 0.11 0.09 0.07 0.07 0.07 0.06 0.04
advanced, such as machinery, electrical equipment, computer and electronic products, and transportation equipment, tend to rank high. These industries are also the ones where computers have traditionally been applied in various ways. Industries that are at the bottom of the list are wood products, nonmetallic mineral products, and furniture and related products. The second summary, shown in Table 5, is the ranking of Internet-based processes based on their rates of adoption in different industries. As in Table 4, we first ranked all processes for each industry in terms of adoption rate and then calculated the average rank for each process across all industries. The most heavily adopted processes are basic Internet access and degrees of access, access to vendors products or catalogs, and ordering from vendors. The least adopted processes are provision of inventory data for external customers
and provision of order status information for external customers. Somewhat surprisingly, the adoption rates of online bidding and use of electronic marketplaces are relatively low. These processes are precisely the ones that were initially thought to be revolutionary. Day, Fein, and Ruppersberger (2003) argue that the limited success of these applications can be attributed to the fact that online exchanges did not dramatically alter the existing way firms manage their supply chains. Firms value obtaining the right combination of products at the right time, and coordinating complex production activities is easier with a dedicated, traditional supply chain. The cost savings offered by online exchanges were simply not enough to convince firms to sacrifice other aspects of production, such as timeliness and access to preferred brands.

Plant Size and Adoption Rate
The increasing use of the Internet for transactions within and across firms has also raised the question of whether the rate of usage is closely associated with firm size. A related issue is how adoption of Internet-based processes affects firm size. As Varian (2002) pointed out, it is not clear in which direction firm size will move as Internetbased transactions continue to replace traditional ones. The answer depends on the relative magnitudes of competing forces. If Internet-based transactions reduce the costs of using external markets by more than they reduce internal transaction costs, then firm size can decrease. The data available are not suitable for a full analysis of the Internets effect on firm size, but they are informative with respect to the role that plant size plays in adoption. We can estimate the rates at which certain Internet-based processes are adopted by plants of different sizes. For 10 employment size groups, the data contain the number of plants that have adopted a certain Internet-based process at the time the survey was conducted.27 We can again assume that the population of plants in size group k is generated by a Bernoulli distribution with parameter pijk , which can be estimated as the ratio of the number of plants in industry i that adopted process j, nijk , to the total number of plants surveyed in this size group, Nik. In other words, a plant in size group k adopts the process with probability pijk independently of other plants in the size group and in other size groups.28 The sampling procedure used by the census is a probability-proportional-to-size sampling scheme in that large plants are sampled with higher frequency and small plants are underrepresented in the sample. Therefore, the standard errors on the estimates for smaller plants are in general higher.29 As an example, consider the estimated rate of Internet access by plant size in Figure 5. The small27
est plant size group has an estimated adoption rate of 48 percent compared with 98 percent for the largest group. For larger size groups, the estimated values are higher and the estimated standard deviations are lower, in part reflecting the sampling scheme mentioned. Consequently, the confidence intervals are narrower for larger size groups and the differences between estimated adoption rates are usually highly significant across size classes, with a few exceptions. The pattern in Figure 5 is generally applicable to a majority of the processes. In some cases, the standard deviations of the estimates increase with plant size, implying that there is much variation in the adoption rate among large plants, after controlling for the fact that they are represented more heavily in the sample. In the following discussion we will focus on characterizing whether the adoption rate generally exhibits a positive and statistically significant relation to plant size. For a compact presentation of the patterns, we aggregated the 10 employee size groups into three size classes: small plants (with 1 to 20 employees), medium plants (with 21 to 99 employees), and large plants (with 100 or more employees). Table 6 confirms that in many cases there is a statistically significant increase in the adoption rate as plant size increases. Exceptions occur for some important processes, however. In the case of placement of orders for materials and supplies, the adoption rate declines with plant size, as shown in Figure 6. A similar pattern is observed for acceptance of orders for manufactured products, as seen in Figure 7. While these exceptions deserve further exploration, lack of plant characteristics prevents us from reaching a definitive conclusion about the adoption rate/firm size relationship.30 Since larger plants are more likely to be vertically integrated, it is quite possible that these plants rely less on the Internet to access outside suppliers. This explanation may also apply to the case of accepting orders online, albeit to a lesser extent.

The size groups are 1 to 4, 5 to 9, 10 to 19, 20 to 49, 50 to 99, 100 to 249, 250 to 499, 500 to 999, 1000 to 2499, and 2500+ employees. For simplicitys sake, we make the assumption that a plants adoption decision is independent of the overall adoption rate in the industry. Externalities in adoption are likely to affect the probability of adoption for at least some processes. The estimated standard deviation of the estimated probability, denoted by pijk , can be obtained as

pijk (1 pijk ) N ik

A 95 percent confidence interval for the true adoption probability, pijk , is then given as
[pijk 1.96 p , pijk + 1.96 p ].

ijk ijk

Plant characteristics are available from the U.S. Census Bureau, but only for on-site usage, as they are classified as confidential data.

Figure 5

Adoption Rates of Internet Access by Manufacturing Plant Size

Probability

0.11 Size Group
Two other processes deserve attention. It appears that plant size has little effect on the adoption rate of online bidding and use of electronic marketplaces, as shown in Figures 8 and 9. While sampling errors may contribute to these two patterns, there does not appear to be a highly statistically significant increase in the adoption rate of these two processes as plant size increases. In fact, both processes are adopted with a rate of less than 20 percent by plants of all sizes. The low adoption rates of these two processes notwithstanding, virtually indistinguishable rates of adoption across a wide range of size classes suggest that large plants may be benefiting from these external market activities as much as small plants are. Obviously, without the intensity of usage of these two processes by plants, a definitive conclusion cannot be reached based on only adoption rates. Nevertheless, one might have expected a priori that small plants adopt these two processes at a higher rate than larger ones, as smaller plants may rely more on these external market activities because of a lack of internal subunits that focus on individual stages of production and procurement.
One of the conjectures about the Internets impact on the organization of production was that it would lead to more vertical disintegration. Along Coases (1937) arguments, if the cost of making transactions outside of the firm declines, firms should have increased incentives to carry out these transactions with outside specialists, rather than within the firm. While our results do not offer any direct evidence on the issue, they suggest that, at least for some stages of production, this may be happening to some extent. Most processes are adopted at a higher rate by larger plants. Some of these processes are those that can induce vertical disintegration, such as placement of orders for materials and supplies online, ordering from vendors, payment to vendors, online bidding, use of electronic marketplaces, and outsourcing of research and development. As such processes are adopted with higher frequency and intensity, plants and firms may reduce the size of internal units undertaking these functions or eliminate them altogether.

Table 6

Adoption Rates of Internet-Based Processes by Plant Size1
Plant size2 Process Basic Internet access and degree of access Product descriptions or online catalog for other company units Product descriptions or online catalog for external customers Product descriptions or online catalog for external suppliers Order status for other company units Order status for external customers Order status for external suppliers Inventory data for other company units Inventory data for external customers Inventory data for external suppliers Access to vendors products or catalogs Ordering from vendors Payment to vendors Bidding Electronic marketplaces linking specialized business buyers and suppliers Ordering by customers Payment by customers Customer support Outsourcing of research and development Ordering of materials and supplies Acceptance of orders for manufactured products

Small Medium Large

0.6072 (0.0071) 0.0759 (0.0039) 0.0620 (0.0036) 0.2117 (0.0061) 0.0927 (0.0043) 0.0304 (0.0026) 0.0896 (0.0043) 0.1314 (0.0050) 0.0115 (0.0016) 0.0244 (0.0023) 0.6620 (0.0068) 0.2491 (0.0077) 0.0558 (0.0041) 0.0776 (0.0048) 0.1862 (0.0069) 0.0640 (0.0044) 0.1663 (0.0067) 0.0686 (0.0045) 0.0658 (0.0044) 0.7371 (0.0129) 0.6174 (0.0154)
0.9585 (0.0017) 0.1368 (0.0029) 0.1147 (0.0027) 0.3496 (0.0041) 0.1622 (0.0031) 0.0467 (0.0018) 0.1410 (0.0030) 0.2064 (0.0035) 0.0217 (0.0012) 0.0484 (0.0018) 0.8565 (0.0029) 0.2724 (0.0040) 0.0666 (0.0022) 0.0816 (0.0025) 0.2090 (0.0037) 0.0919 (0.0026) 0.2021 (0.0036) 0.0678 (0.0023) 0.0818 (0.0025) 0.7517 (0.0063) 0.4572 (0.0077)
0.9406 (0.0017) 0.2717 (0.0033) 0.1540 (0.0027) 0.4108 (0.0036) 0.3127 (0.0034) 0.0779 (0.0020) 0.2192 (0.0030) 0.3782 (0.0036) 0.0595 (0.0017) 0.0926 (0.0021) 0.9502 (0.0016) 0.3714 (0.0035) 0.1292 (0.0025) 0.0833 (0.0020) 0.2846 (0.0033) 0.1639 (0.0027) 0.2443 (0.0032) 0.0724 (0.0019) 0.1159 (0.0024) 0.6551 (0.0051) 0.2036 (0.0045)
NOTE: 1Standard errors in parentheses. 2 Small: 1 to 20 employees; Medium: 21 to 99 employees; Large: 100 or more employees.

Figure 6

Use of Internet to Place Orders for Materials: Adoption Rate by Plant Size
Probability 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.11 Size Group

Figure 7

Use of Internet to Accept Orders: Adoption Rate by Plant Size

in the adoption of Internet-based processes used to facilitate production. Although the most heavily adopted processes include obvious ones (e.g., basic internet access and degree of access and access to vendors products or catalogs), other processes initially thought to thrive on the Internet (e.g., bidding and use of electronic marketplaces) have not been widely adopted. Analysis of adoption rates of several Internet-based processes across plant sizes and manufacturing industries reveals that, generally, there is a positive and statistically significant relationship between adoption rates and firms plant size. As always, the burden of recording the effects of the ongoing technological revolution rests on the shoulders of data collectors. The steps taken so far by the U.S. Census Bureau are encouraging, but much more remains to be done.32 In our view, the collection of data pertaining to e-commerce activity should be taken to the mainstream.33 For instance, new survey questions can be added to the Census of Manufacturers, a quinquennial dataset collected by the Census Bureau that contains information on all active manufacturing plants, to gather detailed information on plants various uses of the Internet. This practice would allow us to understand the importance of digital inputs in the production processes and how the intensity of usage of such inputs compares with traditional inputs of labor and capital. Any substitution among these various inputs that can take place in the medium- and long-run can then also be detected. Furthermore, data on the intensity of the use of Internet-based processes should also be collected, rather than just information on whether a process is adopted or not. Several processes investigated in this paper can be measured in a continuous way, rather than with a discrete adopt versus
Haltiwanger and Jarmin (2000) provide a good list of broad areas in which data collection efforts can be concentrated. There is also some private effort to collect extensive data, especially on prices. See www.nash-equilibrium.com for an Internet price index tracker.

APPENDIX DATA

The data used in this article come from two U.S. Census Bureau reports on electronic economic activity. The first is the E-commerce Multi-sector Report and the second is the E-business Process Use by Manufacturers, Final Report on Selected Processes. Both of these reports are available online at www.census.gov/estats/.
E-commerce Multi-sector Report
The data on e-commerce economic activity for the three industries we analyze are collected in three separate Census Bureau surveys. First, data on retail e-commerce sales are collected in the 2002 Annual Retail Trade Survey, a survey of more than 19,000 retailers. More recent data on retail Internet sales (such as those used in Figure 3) are available as part of a quarterly retail e-commerce series. Revenue data on selected services industries are collected in the 2002 Service Annual Survey, a survey of more than 58,000 firms. Finally, data on the value of manufacturing e-commerce shipments are collected in the 2002 Annual Survey of Manufactures, a survey of more than 55,000 manufacturing plants. The estimates in Figure 3 are reproduced from the August 20, 2004, release, Retail E-commerce Sales in Second Quarter 2004, produced by the Census Bureau. Estimates are not adjusted for seasonal variation, holiday or trading-day differences, or price changes. For additional details, please see www.census.gov/mrts/www/current.html. The estimates of e-commerce shares of total sales or revenues (and their standard errors) in Tables 1, 2, and 3 are reproduced from Tables 5 and 5A, 6 and 6A, and 4 and 4A, respectively, in the E-commerce Multi-sector Report.
E-business Process Use by Manufacturers
This report tabulates the responses of more than 38,000 manufacturing plants to 39 questions about Internet-based processes used at the plant level. These responses were collected in the Computer Network Use Supplement to the 1999 Annual Survey of Manufactures. The estimates of adoption rates of Internet processes reported in Figures 5 through 9 for manufacturing plants were obtained from the authors own calculations based on the tabulations of the E-business Process Use by Manufacturers report. The same tabulations were used to calculate the rates of adoption of Internet processes to rank manufacturing industries in Table 4, to rank Internet-based processes in Table 5, and to contrast the adoption rates of several processes across three aggregate manufacturing plant size classes in Table 6.