CONTRIBUTIONS TO INTERNET ECONOMICS
Network science approaches offer additional analytical and empirical tools that can be used to examine the interrelations in the digital economy and the effects of increasing connectivity.
These approaches build upon the earlier literature on the economics of network industries (e.g., Economides, 1996; Shy, 2001) but enhance it in several ways.4.5.1 New Analytical and Methodological Frameworks
Economics has established a successful research program by analyzing representative decision-makers in situations that they understand well or that are repeated so that choices can converge to the best alternative in a few iterations. This has resulted in the wide use of rational choice models not only in economics but also in other social sciences (Gintis, 2009). Despite numerous weaknesses and criticism the model remains highly influential. Even the broad critique based on findings in behavioral economics and cognitive science that challenges the ability of individuals to be rational in the model sense have thus far mainly led to modifications at the margin and have not challenged the basic premise that individuals seek to achieve best outcomes. However, these alternative approaches place much greater emphasis on the role of information constraints, the framing of the decision problem, systematic biases especially involving risk, habits, and concerns about others (‘other-regarding’, social preferences) (Kahneman et al., 1982; Simon, 1982-97; Kahneman and Tversky, 2000; Fehr and Fischbacher, 2002; Sunstein et al., 2002; Kahneman, 2011). In a highly connected world, the role and effect of these concerns will require further analysis.
Network science offers a framework and a variety of modeling approaches that can help shed light on these issues with the potential to support the development of overarching new theoretical foundations. Because it makes it possible to integrate micro-level perspectives, such as the examination of individual nodes in a network and the uni- and bi-directional connections to and from this node, with analyses of the emergent behaviors at various levels of aggregation (sub-networks, the entire network), network science has the potential to offer a novel approach to create a microfoundation for macroeconomic processes.
Such an endeavor could build on and integrate with existing efforts in related theoretical and empirical areas. For example, since the 1950s, economics has widely used game theory to study decision-making by individual agents in situations in which the consequences of an actor’s actions are dependent on other actors’ actions and strategies. A rich literature has shed light on non-cooperative games, cooperative games, and evolutionary games, among others (Fudenberg and Tirole, 1991; Fudenberg and Levine, 1998; Camerer, 2003; Gintis, 2009). One of the surprising results from this line of research was that the well-known Prisoner’s Dilemma - situations in which cooperation would make it possible to achieve a better outcome but coordination problems let each agent act selfishly, resulting in a worse outcome for all - can be overcome. Robert Axelrod (1984) found that in simple computer tournaments the simple strategy of ‘tit-for-tat’ (each player does what the other player does) wins. Subsequent research revealed that the slightly more complicated strategy of ‘win-stay, lose-shift’ is even more successful in longer types of tournaments and interactions. Moreover, it is also more stable and leads to successful cooperation. These processes are widely observable in biological evolution and in cultural evolution (Nowak, 2006; Nowak with Highfield, 2011). Network science promises new insights into such interdependent decision-making processes.The fundamental importance of network relations has also been recognized by economic sociology, including the work of Granovetter (1985) on the effects of strong and weak ties on economic activities and outcomes, the work of White (2001), who developed an innovative approach explaining markets from network relations, and the research by Padgett and Powell (2012) on the emergence of organizations and markets. These insights from the fields of game theory and of economic sociology are ripe for further integration with network analysis into new theoretical and empirical frameworks for the study of social and economic processes in multi-layer socio-technical systems, including but not limited to the Internet.
One promising path in this direction is the use of agent-based modeling (ABM) techniques, which are increasingly deployed to model the economy as a dynamic system (Judd and Tesfatsion, 2005). ABM makes it possible to replace the widely used assumption of maximizing agents in equilibrium with boundedly rational agents adapting to other agents and to their environment and studying the emergent economic processes. This could eventually lead to a new paradigm of explaining social and economic processes from the bottom up (Epstein and Axtell, 1996; Epstein, 2006). Despite its potential this approach faces considerable challenges as the options for explaining outcomes multiply greatly so that model validation becomes excruciatingly complicated. Nonetheless, these are promising approaches and the availability of highly granular and detailed network data harvested from the Internet should provide an enormous boost to this research agenda.4.5.2 Dynamic Economic and Social Processes
Coordinating the actions of distributed agents is one of the central challenges of economies (North, 1990). Telecommunication technologies have facilitated such coordination by reducing the costs of message exchange across ever-larger geographic distances. The telegraph allowed communication with distant territories, shortening the time of message delivery from months and weeks to hours and minutes. Telephones, fax machines and e-mail further reduced communication delays. As a platform-independent technology that can provide cheap communications as an application over the top of the network, the Internet has greatly expanded the reach of these instant forms of communications to more than three billion people worldwide. Moreover, it has affordances that support more complex forms of communication than prior telecommunications technologies and services. For example, the ability to interact with a communication partner via video, document sharing and chat simultaneously increases the potential to engage in transactions that are contingent on high levels of relationship-specific investment and historically required repeated personal interactions (Chattopadhyay, 2013).
Consequently, the Internet and the social media it enables have greatly affected economic and social processes, although the welfare outcomes are sometimes ambiguous and not always positive.The economic model of decision-making under information and resource constraints has been regarded as fairly robust in explaining individual- and aggregate-level outcomes. However, how preferences form and how knowledge about constraints and potential outcomes (payoffs) is obtained is rarely modeled (Wildman, 2008). Network analytical approaches have the potential to fill this gap with explicit models of information flows and their influence on individual decisions as well as their consequences for emergent phenomena at an aggregate level. Advances are possible in several areas. First, economists have long had a hunch that preferences are not immutable but are influenced by interactions among agents. Behavioral economics has found strong evidence for social preferences among decision-makers (Fehr and Fischbacher, 2002; Camerer and Fehr, 2006). Individuals regard others, often differentiated by in-group and out-group relations. In a densely networked environment, these groups may change dynamically and they will likely be influenced by the network structure of relations. In highly connected networks these social relations may have predictable consequences for preferences and decisions (Christakis and Fowler, 2009; Gomez Rodriguez et al., 2013). At an abstract theoretical level, preferences may become endogenous and adaptive (von Weizsacker, 2005) requiring a rethinking of the traditional approach of assuming given individual preference orderings.
Second, it is possible to use network theory approaches to gain a better understanding of whether and how an agent’s knowledge, decisions and behavior are related to the information flows it is exposed to. In networked environments with incomplete and asymmetrically distributed information it may be rational to base decisions on the behavior of other agents.
One example is the existence of direct and indirect network effects. In the case of direct network effects, the benefits of a choice such as the adoption of a technology or device are contingent on the number of other adopters. In the case of indirect network effects, the number of adopters influences the availability of complementary technologies and services. A second example is information cascades or herding effects.Even if an agent’s private information were to suggest another choice, in situations with sequential decision-making observing prior choices of other individuals conveys information on what they may know. Following earlier choices may be a rational response but also an outcome of social pressures. Thus, information cascades can be wrong, can be based on very limited information, and are often fragile (Easley and Kleinberg, 2010, p. 442). In contrast to situations in which the aggregation of independent assessment can lead to accurate outcomes at the population level, in information cascades decisions are made sequentially and that aggregation effect may not hold. In highly networked environments, cascade-like behavior may affect many types of economic and social processes from the dynamics of financial markets to fashion, moral panics and other trends.
Network science approaches have stimulated the development of alternative approaches and deeper insights into processes of dynamic change and diffusion. Such processes can be observed at the level of entire populations, subgroup and components, and at the level of individual decision-makers. Pioneering studies by Ryan and Gross (1943), Rogers (1962), and Coleman et al. (1966) examined the factors that facilitated and blocked the adoption of innovations. Among the factors that Rogers (1962) identified as critical for the success of an innovation were its relative advantage, complexity, observability, trialability and the compatibility with the social system in which it was introduced. Network approaches allow the modeling of these aspects in a highly granular fashion.
In such a perspective, nodes can be seen as responding to neighboring nodes either due to informational effects or due to network effects (Easley and Kleinberg, 2010, p. 499). Such processes can be modeled as coordination games that percolate through the nodes of a network. Consequently, the choices of each node are contingent on the choices of all connected nodes. This framework can also be used to understand the extent to which competing technologies can coexist in a network and the conditions under which a network might tip toward adopting just one of the alternatives. Furthermore, network analyses have revealed that clusters of similar nodes (an outcome of homophily in network formation) may form barriers to the adoption of innovations.4.5.3 Network Structure, Governance and Outcomes
An important question pursued by network analysis approaches is whether the structure of the network, reflected in metrics such as the degree and centrality of nodes or the density of the network, has predictable consequences for the outcomes of the interactions at an aggregated level. An increasing number of findings suggest such patterns. Early models studied epidemics, but similar issues arise in the study of product purchases, fashions, and in marketing. The dynamics of contagion processes is related to the node structure and topology of the network. For example, all else being equal, if contagion starts at nodes with high degree then information will spread more rapidly (Duan et al., 2005). In their study of Digg and Twitter, Lerman and Ghosh (2010) found evidence that the structure of the underlying networks influences the information diffusion process. Because Digg networks are dense and highly connected, news initially spreads quickly but the process slows down once a story is exposed to a larger number of unconnected users. In contrast, stories in the less connected Twitter network spread slower than on Digg but continue to spread at the same rate, so eventually reach further than Digg stories. Similar effects of the network structure were found in other areas, such as viral marketing and recommender systems (e.g., Leskovec et al., 2007), social movements (Gonzalez-Bailon et al., 2011), online and local communities (Toral et al., 2009; Yardi and boyd, 2010), and for user-generated content (Susarla et al., 2008).
Network analytical models also offer novel perspectives for the theory of innovation. Evolutionary approaches recognize that innovation is an experimental process of combination and recombination of knowledge (see Antonelli and Patrucco, Chapter 15 in this volume). Consequently, the structure of the pertinent social and economic networks will influence the available innovation opportunities. This is explicitly theorized in Ronald S. Burt’s theory of ‘structural holes’ (Burt, 1992; Borgatti et al., 2009), which examines the role of gaps in networks of complementors as opportunities for entrepreneurs to innovate and develop market niches. Recent studies have found evidence of the role of structural holes as a driver of innovation in information and communication markets (Hitt and Ireland, 2014).
A third area in which network analytical models offer new perspectives to better understand the Internet economy is issues of income distribution and winner-takes-all dynamics. In recent years, concerns have been growing that the unique economic characteristics of Internet-based markets, including high initial costs combined with low incremental costs, have contributed to high income gains for a very small group of companies and individuals. This rich-get-richer phenomenon is related to the preferential attachment dynamics that result in the familiar power law distributions of nodes, popularity, and many other phenomena on the Internet. While the dynamics is well understood once a node has a higher degree, the initial process is highly unpredictable (Easley and Kleinberg, 2010, p. 484). Nonetheless, the effects are visible in increasing industry concentration in many high-tech and digital economy markets as well as mounting evidence of an increasing share of income gains flowing to top earners and entrepreneurs (Bauer, 2015).
Last but not least, network science allows unique insights for network governance. Spulber and Yoo (2005, 2009) offer an integrated approach rooted in network and complexity theory to the regulation of telecommunications. It is a widely shared view that the networked structure of Internet governance has contributed to its global success (Mueller, 2010). While seen as an alternative to traditional forms of state and market governance, the details of how such networked coordination affects outcomes remain often opaque. Recent work in network science is helping to shed light on the underlying processes (Padovani and Pavan, 2011; Pavan, 2012). The influence of the network topology and structure has also been examined in more specific areas such as cybersecurity (e.g., Van Mieghem et al., 2009), where the density of a network was identified as a critical factor in the spread of viruses.
4.6