A network grows more useful when someone else enters it. A telephone gains another destination, a language another speaker, a common format another reader who can open the file. The newcomer buys an advantage for himself and leaves advantages behind for people he has never met. This is the positive externality in a form that reaches back into the value of the good itself: what one owns is worth more because of what others have chosen. It also changes the character of the next choice. A person considering an unfamiliar system must compare its qualities with the value of everyone already gathered around its rival. The rival's popularity is now among its qualities. Adoption has begun to make the reasons for further adoption, and a market whose prices are perfectly visible can acquire a history that its participants cannot individually undo.
The feedback can arise through several distinct channels, which it is useful to keep apart.1 A communication network gives direct benefits through the number of reachable users. A technical standard can draw complementary products, so that more users attract more software and more software attracts more users. Experience can also improve a technology as engineers and operators learn from using it. The first mechanism requires connections, the second a supporting ecology, the third accumulated practice. All can give an established choice an advantage that grows with its adoption, but they do not make every successful product a network monopoly. Benefits may level off, congestion may offset them, specialized needs may support several systems, and compatibility may allow users of different products to share much of the same network. The fact to establish is the actual source and strength of the feedback. Popularity alone is evidence that many people chose a thing; it supplies no complete explanation of why they did.
W. Brian Arthur showed how increasing returns to adoption can make the order of individual choices determine the eventual shape of the market.2 In the simplest version, some people initially prefer technology A and others prefer B, while each technology becomes more attractive as more people adopt it. Suppose several people who favor A happen to arrive first. Their choices improve A's position for the next arrivals. With sufficiently strong feedback, the accumulated advantage can eventually persuade even those who would have preferred B at equal development to choose A. Once that threshold is crossed, further adoption reinforces the same result. Reverse the early sequence and B may prevail instead. The preferences and the available technologies can be the same in both histories; a different order of arrival carries the market to a different outcome. Each person has chosen sensibly at the time of choosing, while the sequence as a whole has selected one among several possible futures.
The possible loss appears when one compares the path taken with a path that never had the opportunity to develop.3 The unused technology may have improved more rapidly with experience, or served its users better once it reached a comparable scale. An early lead can deny it the adoptions through which those advantages would have emerged. The comparison is therefore difficult in practice: the winner has a real installed base, a body of accumulated knowledge, and products that actually exist, while the alternative includes improvements that have to be estimated. Arthur's model establishes that an inferior path is possible; it supplies no warrant for declaring every defeated technology the better one. The importance of the result is that survival by itself ceases to settle the comparison. What the market selected is observable. Whether another history would have produced greater value requires a further inquiry, one capable of distinguishing the benefits of the chosen standard from the possibilities its success closed off.
Expectations can strengthen the feedback before the anticipated network has even arrived.4 A buyer may choose the system he expects others to choose, because compatibility tomorrow contributes to the value of his purchase today. If enough people reason this way, their expectations help produce the installed base they anticipated. Arthur's expectations extension shows how this can hasten lock-in within the model. Yet lock-in remains conditional upon the forces sustaining it. In his more general account, increasing returns that are bounded need not eliminate every rival; some histories can leave a shared market. An institution that permits interconnection, a technology that lowers conversion costs, or a sufficiently valuable new use can also change the choices people confront. The past constrains the present through particular benefits and costs. To discover those links is to discover where persistence is strong, where it is vulnerable, and which part of it is serving users well.
The policy question follows the same sequence as the market: first the choice of a path, then the cost of changing it.5 Before a standard settles, support for a promising alternative may preserve the chance to learn what it can become. After settlement, interoperability can sometimes retain the reach of a network while allowing competition in the products through which people enter it. Neither move is free of judgment. Supporting alternatives spends resources on possibilities that may disappoint; imposing compatibility can sacrifice features, security, or a design whose value depends upon tighter integration. An authority choosing a standard also faces uncertainty about which path will develop best, and can become attached to an early success for the same reason the market does. The case for intervention therefore rests upon an identified external benefit or impediment to coordination, joined to an explanation of how the proposed arrangement improves it. The existence of feedback opens the question of institutional design. It leaves the work of answering that question intact.
The network gives The Boundary a temporal depth. An adoption affects people outside the transaction, including people who have yet to arrive; some receive the value of a shared system, and some inherit the cost of departing from it. The price the early buyer sees seldom records the whole of either bequest. A standard can be useful because everyone shares it, while the reasons it became the shared standard lie in a sequence too small to have appeared in any contemporary account of economic fundamentals. History has entered value through the ordinary choices of people making the best use of what was available to them. The question is then what arrangements let a society enjoy the gains from coordination while keeping enough room to discover a better way of coordinating. A network joins people to one another. It also joins the present to the path by which it arrived. The argument remains open.