Re: Generation of network effects on the Web

Phil Agre (pagre@weber.ucsd.edu)
Mon, 21 Sep 1998 15:44:29 -0700 (PDT)


All of this agonizing about the actual evidence proves that you are not an
economist. Those primitive models of network effects fail to account for
many phenomena that can't be expressed simply in math. For example, if the
value of a given network good really is linear in the number of other people
who use it, then indeed a vendor or vendor coalition faces a major critical
mass problem. In reality, the value of a typical network good to a typical
potential user is equal to the summed value *to that particular user* of
connecting with *certain other particular users*. We could formalize this
with a matrix V[i,j] of the value to potential user i of being able to use
the network to connect with user j. The point of creating this matrix is
one can then model dense submatrices corresponding to social networks of
people who have lots of reason to contact one another. In reality, these
subnetworks correspond to the niche strategies that are now commonplace in
high tech. A vendor needn't subsidize early adopters at random, or perhaps
even at all, so long as enough PR is generated to persuade the members of
particular subnetworks that everyone else in that subnetwork is going to use
the service. Then those networks of users can be employed to leverage the
service to larger networks. That's what really happens.

But your case is fancier in a different way, which is that there are several
distinct categories of players whose decisions need to be modeled. It seems
to me that you need to model users' choices to get online, but also the
choices of the organizations that create Web sites. You could then model
the advertisers. Having done all this, you could then model the vendors who
will benefit if users, site creators, and advertisers all sign up. This is
all complicated, of course, for all of the reasons you said, and for other
reasons, for example the distinction between sites that take advertising and
sites that do not. But if you want to make economic models, you take it all
a step at a time. Start with a simple model and then add complexity to the
model in increments.

Phil