Hey, we sure like Uncle Bert, but we could do without his political rants on FB. This problem that connectivity clashes with preferences gave rise to websites like Flickr or LastFm where we can match preference and connectivity.
One of the key shortcomings of network theory is that it's still largely a theoretical construct, with weak empirical support. Most of our understanding is anecdotal, which is subject to ex post rationalization (like almost everything else in innovation econ).
Indeed since interaction effects undermine revealed preference, it's very tricky to empirically separate preference from influence in observed group behavior.
For instance, if you're going to the movies you're likely going to pick a different movie depending on whether you go alone, with your spouse, your buddies, or your parents. But as an outside social scientist, how to establish causality?
You can't pick your parents, but likely you picked your buddies and your spouse because you share common interests. If you don't you likely set up a complex bargaining system to make sure you still get to do things together.
If we know anything empirically about network effects, they are probably quite weak and preferences tend to prevail. That in itself poses a new problem, as to speak with Mark Granovetter, the strength of weak ties tends to accumulate.
The canonical model of a pairing with positive interaction effect but diverging preferences is the battle of the sexes. That's something we understand pretty well for the two-player but things get tricky when we expand the model to a bigger group.
Especially if the network tying the participants together is sparse, as it commonly is in social networks, the bigger group is much more likely to split into multiple subgroups (coalitions) based on a mix of preferences and (mutual) influence.
And things get even more tricky if we allow that interaction effects can be negative or even asymmetrical (the "pesky little brother" game). But the general a priori assumption in these situations is not that the network converges to a winner-takes-all state.
Finally (I hope) a few words on the role of capacity (i.e. stock) and congestion (i.e. flow). In a world of abundant computing power and storage we kinda stopped paying attention to them as supply constraints (altho blockchain based systems brought that back with a vengeance).
But of course they also exist on the demand side, and often in more subtle ways then expected. Capacity constraints aren't always computational, they can as well be human. We tend to react negatively to information overload, especially if it's conflicting or ambiguous.
That's a big reason why most social networks have switched from a chronological feed to a ranked "most interesting" feed. And of course because they want to control us...
So that was a very cursory run-through touching a couple of reasons why we shouldn't automatically conclude that (positive) network effects means winner-takes-all outcomes. It's much more typical to have a power law distribution. Like in the browser wars.
As usual, I'm gonna fix a few things and post this on Medium for better readability. //FIN
