The Personal Network Effect

The presumption in the design of most networks is that the value of the network increases with the number of nodes in the network. This is known as the Network Effect, a term that was coined by Robert Metcalfe, the founder of Ethernet.

Image source

It is therefore tempting to suggest that a similar sort of thing holds for members of the network, that the value of the network is increased the more connections a person has to the network. This isn't the case.

Each connection produces value to the person. But the realtive utility of the connection - that is, its value compared to the value that has already ben received elsewhere - decreases after a certain point has been reached.

The reason for this is that value is derived from semantic relevance. Information is semantically relevant only if it is meaningful to the person receiving it (indeed, arguably, it must be semantically relevant to be considered information at all; if it is not meaningful, then it is just static or noise).

Semantic relevance is the result of a combination of factors (which may vary with time and with the individual), according to whether the information is:

• new to the receiver (cf. Fred Dretske Knowledge and the Flow of Information)
• salient to the receiver (there are different types of salience: perceptual salience, rule salience, semiotic salience, etc)
• timely, that is, the information arrives at an appropriate time (before the event it advertises, for example) - this does not mean 'soonest' or 'right away'
  
• utile, that is, whether it can be used, whether it is actionable
• cognate, that is, whether it can be understood by the receiver
• true, that is, the information is consistent with the belief set of the receiver
• trusted, that is, comes from a reliable source
  
• contiguous, that is, whether the information is flowing fast enough, or as a sufficiently coherent body
  

Because of these conditions, the value of each new piece of information, on average, will decrease relative to its predecessors. At a certain point, the value of the new information will be such that it actually detracts from the value of the information already received (by, say, blocking it, distracting one's attention from it, contradicting it, and the like).

For example, suppose someone tells you that the house is on fire. This is very relevant information, and quite useful to you. Then another person tells you on fire. It's useful to have confirmation, but clearly not as useful as the first notice. Then a third and a fourth and a fifth and you want to tell people to shut up so you can hear the next important bit of information, namely, how to get out.

This is the personal network effect. In essence, it is the assertion that, for any person at any given time, a certain finite number of connections to other members of the network produces maximal value. Fewer connections, and important sources of information may be missing. More connections, and the additional information received begins to detract from the value of the network.

Most people can experience the personal network effect for themselves by participating in social networks. One's Facebook account, for example, is minimally valuable when only a few friends are connected. As the number grows over 100, however, Facebook begins to become as effective as it can be. If you keep on adding friends, however, it begins to become less effective.

This is true not only for Facebook but for networks in general. For any given network, for any given individual in the network, here will be a certain number of connections that produces maximum value for that member in that network.

This has several implications.

First, it means that when designing network applications, it is important to build in constraints that allow people to limit the number of connections they have. This is why the opt-in networks such as Facebook produce more value per message than open networks such as email. Imagine what Twitter would be like is anyone could send you a message! The value in Twitter lies in the user being able to restrict incoming messages to a certain set of friends.

Second, it provides the basis for a metric describing what will constitute valuable communications in a network. Specifically, we want out communications to be new, salient, utile, timely, cognate, true and contiguous.

Third, it demonstrates that there is no single set of best connections. A connection that is very relevant to one person might not be relevant to me at all. This may be because we have different interests, different world views, or speak different languages. But even if we have exactly the same needs and interests, we may get the same information from different sources. By the time your source gets to me, the 'new' information it gave you might be very 'old' to me.

We see this phenomenon is web communities. Dave Warlick today posted a link to a video produced by Michael Wesch's Cultural Anthropology students at Kansas State University. Warlick obviously does not read OLDaily because I linked to the site two weeks ago. Warlick credits John Moranski, a school librarian from Auburn High School and Middle School in Auburn, Illinois (no link, which means he probably told him about it in person or by email). Warlick's link, therefore, is of little value to me; it's old news. However, to many of his readers (specifically, those who don't read me), this will be new. And hence he is a valuable part of their network.

Now here is the important part: the people who read Warlick don't need to read me (at least with respect to this link). They are getting the same information either way. There is no particular reason to select one source over another. Warlick may be part of his readers by accident (he is the first ed tech person they read, for example) or he may be more semantically relevant to them for other reasons: he is a folksy storyteller, he writes in a simple vocabulary, they have met him personally and trust him, whatever.

One final point: if we change the way we design the network, we can change the point of maximal value:



It is toward this effect that much of my previous writing about networks has been directed. How can we structure the network in such a way as to maximize the maximal value? I have suggested four criteria: diversity, autonomy, openness, and connectedness (or interactivity).

For example, networks that are more diverse - in which each individual has a different set of connections, for example - produce a greater maximal value than networks that are not. Compare a community of people where people only read each other. You can read ten people, say, of a fifty person community, and hear pretty quickly what every person is thinking. But reading an eleventh will produce almost no value at all; you will just be getting the same information you were already getting. Compare this to the value of a connection from outside the community. Now you are reading things nobody else has thought about; you learn new things, and your comments have more value to the community as a whole.

It is valuable to have a certain amount of clustering in a network. This is a consequence of the criterion for semantic relevance. This is that people like Clark are getting at when they talk about the need for a common ground, or what Wenger means by a shared domain of interest. However, an excessive focus on clustering, on what I have characterized as group criteria, results in a decrease in the semantic relevance of messages from community members.