|
Article Excerpt
The growth of the Internet and assorted technologies has made it possible to collect and process detailed information on social networks. This article investigates how firms (and governments) can harness the power of social networks to promote their goals. We show that the optimal use of social networks leads to higher sales and greater profits. However, an increase in the level and dispersion of social interaction can increase or decrease the optimal influence strategy and profits of the player, depending on the content of the interaction. Optimal influence strategies will target individuals with low or high connections, depending on the content of interaction. Finally, the returns to investing in market research on social networks are greater in more unequal networks.
1. Introduction
The important role of friends, neighbors, and colleagues in shaping individual choices has been brought out in a number of studies over the years. (1) In the past, the practical use of such social influences for effective marketing or diffusion was hampered by the lack of good data on them. The growth of the Internet and the large amounts of data on online social networking along with the other advances in information technology have made it considerably easier to gather data on social networks and have led to an explosion in interest on how firms and governments can harness the power of social networks to promote social and private goals. (2)
Practical interest has centered on questions such as: for which product categories are networks important and when are they unimportant, what are the relevant aspects of networks, how should a firm use social networks to promote its product, and how much should a firm be willing to pay to acquire information about social networks? (3) This article develops a framework within which these questions can be addressed.
We suppose there are two groups of players, M and. N. Every member of group A4 chooses a strategy with a view to influencing members of group N to choose certain actions. The actions taken by members of group M lead to some information (or resources) reaching individuals in N. This information is shared by individuals in N "locally," thereby leading to a new distribution of resources or information. Group N members make decisions based on this distribution, which in turn generates payoffs for members in group M. Our interest is in understanding how social networks affect the design of optimal social influence decisions of player M.
At the outset, it is useful to distinguish the level and the content of social interaction. The level of interaction pertains to the number of people someone talks to or the number of friends she has. The number of connections of an individual is referred to as her degree. The distribution of degrees in the population summarizes a large amount of information about the network in a very simple form. (4) Empirical work over the past decade has generated a great deal of data on degree distributions across product categories as well as their relation to demographic characteristics of individuals which are traditionally used in design of influence strategies. (5) For instance, Leskovec, Adamic, and Huberman (2007) show that degrees--reflected in level of word of mouth communication--are higher in DVDs than in books, and that they are higher in books in certain categories such as religion and technical subjects. Similarly, Keller, Fay, and Berry (2007) find that both the average level and the variance/dispersion of word of mouth communication differ greatly across product categories: food and dining and media entertainment have the highest level of word of mouth communication, whereas children's products have the lowest. They also find that the frequency of word of mouth interaction is much higher among teenagers as compared to the average in the population at large. Finally, an important finding of the empirical work is that demographic characteristics only explain part of the variation in degrees across individuals (see, e.g., Nyilasy, 2006; Oetting, 2007).
The content of social interaction reflects the nature of interaction and the way in which actions of others affect individual incentives. The interaction may involve word of mouth communication about product quality and prices. In this case, the presence of a single informed neighbor leads to product awareness and possibly purchase. (6) Alternatively, the interaction may involve collaborative research work and the learning of a language (natural or computer-based). In this case, a sufficient proportion of neighbors need to choose an action before an individual will switch to this action. This rule of behavior can also be interpreted as a form of social conformism. (7)
Our analysis begins by illustrating how incorporating social network information in the design of marketing and influence strategies can both reduce waste in resources and generate greater sales. The effectiveness of social influence campaigns can be further increased by using more detailed information--such as the degree of different individuals in the social network. This finding provides a foundation for the claims made in the popular press for the advantages of using social networks (and the high valuation of online social networking companies).
Perhaps the best-known use of social networks in marketing is the diffusion of Hotmail accounts. It is estimated that almost 12 million people signed up for Hotmail within 18 months of its launch, and the advertising budget was a mere USD 500,000. Indeed, the relation between the level of social networking and marketing budgets of firms is a major theme in the popular discussion and there is an implicit suggestion that effective use of word of mouth or an increase in word of mouth could lead to major cost savings in marketing. On the one hand, word of mouth makes advertising by firms more attractive as any information sent by a firm is now available to more people. On the other hand, by indirectly informing consumers, word of mouth communication makes direct advertising by firms less necessary. Thus the effects of incorporating word of mouth advertising on optimal strategy are a priori unclear. Our analysis shows that if a firm is already advertising heavily, then word of mouth and advertising are substitutes and an increase in word of mouth lowers optimal advertising; the converse is true if the firm undertakes little advertising to start with.
The effects of social networks on profits has been an important concern. Our analysis reveals that the effects on profits depend very much on the content of interaction. For example, keeping advertising fixed, an increase in word of mouth communication enables greater spread of information and therefore it increases sales and profits. However, when the product of a firm is characterized by adoption externalities, an increase in number of neighbors makes it harder to satisfy the requirement that (say) all of them buy a product. Thus, an increase in social interaction in the presence of adoption externalities lowers profits.
This result suggests a note of caution in the current excitement about social networks. The high connectivity of social networking websites helps the spread of information about goods and therefore it is beneficial for the advertising of restaurants, books, and movies, which are products that a consumer buys once he is aware of them. However, high connectivity may be detrimental for a firm which uses the network to induce adoption of new technologies such as software and alternative operation systems.
We turn next to the subject of targeting individuals. A recurring theme in the popular discussion as well as the academic literature on social networks has been the simple idea that it would be better to focus efforts--send information, coupons, or free samples--on individuals who are influential. Our analysis identifies conditions of the content of interaction under which optimal strategy targets more or less connected individuals. In particular, we find that in the word of mouth application it is optimal to target individuals who get information from a few others (the marginalized consumers). By contrast, in the proportional adoption externalities application, it is optimal to seed the most connected individuals (as they are unlikely to adopt via social influence)! Thus the optimality of targeting highly connected nodes depends very much on the content of social interaction.
The growth of online social networking along with related developments in information technology (which enable the collection of data on mobile and fixed telephones in combination with data on decisions) motivates the question: how much should firms be willing to pay for information about networks? We find that the value of information increases in the dispersion in the degree distribution. Going back to the empirical evidence we reported earlier, our result implies that detailed information on social networks is particularly valuable for product categories such as food and dining, media entertainment, and fashion where the level of word of mouth communication across individuals is very unequal.
All the above results are obtained in a model with one firm/government. The effects of social networks on competitive strategy are important and poorly understood. We present a simple model of competition among differentiated firms. Preliminary analysis of this model suggests that enhanced word of mouth communication lowers profits if products are similar, whereas the converse is true if products are dissimilar.
We now locate our article in the literature. There is a large literature on optimal firm strategies with regard to advertising and the adoption of goods with adoption externalities. (8) Similarly, there is a large literature on local interaction both with regard to word of mouth communication and with regard to adoption in the presence of local externalities. (9) The main contribution of our article is a model which allows us to study optimal marketing and influence strategies in the presence of local interaction. Our article thus bridges these two important and large literatures in economics. Our analysis yields several insights into how the content and level of social interaction jointly shape optimal strategies and profits. In turn, these insights, together with the existing empirical work, have a number of practical implications about optimal strategies for different product categories.
Our work is related to two recent articles which study optimal strategies in the face of local interaction, Ballester, Calvo-Armengol, and Zenou (2006) and Banerji and Dutta (2007). Ballester et al. study a model in which individuals located in a network choose actions (criminal activities) which affect the payoffs of other individuals within the network. They examine the question: which individuals should be eliminated from the network if the objective is to minimize crime? Banerji and Dutta study a setting where firms sell to consumers located on a network and there are local adoption externalities. Their interest is in characterizing networks which can sustain different technologies in equilibrium. There are two key differences between these articles and our article. First, they assume that the network is common knowledge, whereas we suppose that the network is imperfectly known and we model networks in terms of degree distributions. This leads to very different techniques of analysis, and in particular it allows us to investigate the incentive of firms to acquire information about the network. Second, these articles focus on adoption payoff externalities, whereas our model allows for different contents of interaction--word of mouth and adoption externalities--and this leads us to a number of novel results on the interplay between the level and content of interaction.
The article is organized as follows. Section 2 presents a basic model of strategic diffusion. Section 3 develops a number of results on optimal strategy when the external player A4 knows the degree distribution only. Section 4 studies optimal strategy when M knows the degrees of individuals in the population. Section 5 considers a number of extensions such as the value of network information, optimal strategy for influencers and opinion leaders, and the effects of social networks on competition strategy between differentiated firms. Section 6 concludes. All our proofs are presented in Appendix A. Appendix B develops alternative models to illustrate the robustness of our analysis.
2. Basic model
* We study the problem of a player M who exerts costly effort with a view to getting a group of individuals to choose an action. Individual behavior is influenced by social interaction. Social interaction is modelled via a degree distribution and it may involve sharing of valuable information or adoption externalities. We study how the distribution of social connections shapes the optimal strategy of player M and the level of surpluses that she can hope to earn. We now get into the details of the model. (10)
There is a unit measure of individuals N = [0, 1]. Individuals are located in a social network, and in principle the structure of the network can be complex and take on a variety of forms. We assume that M has limited knowledge about this network: she knows the proportions of individuals having different levels of social interaction. We now elaborate on this formulation.
For an individual i [member of] N, the level of social interaction is parameterized by a number k, where k is termed the degree. We will suppose that each individual draws k others with probability P(k) [greater than or equal to] 0, k [member of] {1, 2, ..., [bar.k]} = O, and [[summation].sub.k[member of] 0] P(k) = 1. She uses a (atomless) uniform distribution on the unit interval to pick her sample. So, if she has a k-sized sample, she makes k draws, and each draw is independent. (11) Now suppose that the draw of the sample size is independent across individuals. We can then say, following a standard "abuse" of the law of large numbers, that there is fraction P(k) of individuals who choose a k-sized sample. We will refer to P as the degree distribution. Define [??] = [[summation].sub.k[member of]0] P(k)k as the average degree of social interaction.
We note that when a person i draws j, there is no implication that j draws i as well. Therefore the flow of influences is directed and the degree here refers to the number of people an individual gets information from or the number of people an individual is influenced by. Indeed, from a formal point of view, degree in our model refers to the out-degree of...
|
