To answer these questions, authors Zsolt Katona , PhD candidate, and Miklos Sarvary, associate professor of marketing at INSEAD, develop a model that explains the network formation process and structure of the commercial world wide web.
Virtually no previous research is relevant to www links, where there are large differences between the content and connectedness of sites, and where cost varies depending where a link originates. This paper also studies advertising via network links, where sites are both advertising purchasers and sellers. It asks which activity dominates and whether this depends on the site's content.
The model describes websites and links as a directed graph, with nodes corresponding to sites, and directed edges to links between sites. The number of links going out of a site is the 'out-degree' of the site; the 'in-degree' is the number of incoming links.
To establish the number of visitors to a site in a given timescale, the paper considers how potential customers browse the web. The authors use the solution proposed by Brin and Page (1998), the Page Rank - also used by Google to rank web pages - as they find that this, together with the underlying process, is a consistent descriptor of how traffic is distributed across sites for any given link structure of the network.
To measure a website's value for the public in a particular content domain, the paper assumes that the gross profit of the site is proportional to the number of consumers flowing through the site as well as the site's content. Nodes purchase links (advertisements) at rates quoted as 'per thousand impressions' that point from the seller to the buyer.
The trade-off between keeping consumers and handing them over to another site is captured in the price of the link: the higher the gain from a customer buying its products, the higher the price for handing the customer over to another site.
In equilibrium, higher content sites, often the well-known brands, buy more advertising links but sell fewer links at higher prices. They tend to buy links from sites with lower content. Low content sites benefit from selling advertising links to high content sites even at modest prices, while high content sites benefit more from the sale of content than of advertising.
Therefore the number of in-links correlates with the value of the corresponding site. These assumptions remain true even in an infinite version of the network formation game, where prices and links are both decision variables, and the number of players is infinite.
The paper adds two extensions to the model. It shows how reference links, commonly used in non-commercial sites, also have an important role in forming structures in the commercial web. Higher content sites will still have more in-links, but will also have more reference out-links, especially to other sites of higher content, as they increase the content of referring sites.
It also looks at how search engines measure content using simple heuristics based on in-links to rank sites. Findings show a 'specialisation reward', especially for sites of low content, while high-content sites, such as Amazon.com, can afford to diversify. Sites vie for 'search engine optimisation', especially in the 'Google dance', when the sites are reordered.
These results provide useful guidelines for marketing managers on how to manage their firms' sites in connectedness to the web. The key contribution is that, in explaining what drives the choice of links between web sites, it is consistent with the empirical features of the web.
Miklos Sarvary and Zsolt Katona~