Some nodes have more connections than others. If most of these nodes send out a signal, it will seem to the other nodes that the majority of ALL nodes sent the signal, even if the highly-connected nodes are actually a minority.

e.g. if the popular kids believe X, then to most people it will seem like the majority believe X, even if there are only a few popular kids.

Say there are 100 people on a social network: 50 with tons of friends ("popular"), and 50 with only a couple of friends ("unpopular").

If you pick a random person out of the social network's user list, there's an equal probability they'll be popular or unpopular. But if you start from your friend list and pick a random user from there, it's more likely that the person will be popular, because popular people appear on more friend lists.

Now say there's some belief that most of the popular people agree with, but unpopular people don't. ("Popular people are better than other people"? ;)) Everyone decides to post their opinion on this topic on the social network. Because most of your friends are popular, it will look to you as though there's majority agreement, even though the opinions are actually split 50/50.

In real life, the numbers are more extreme. The top percentage can have thousands or tens of thousands of friends. If these users hold an opinion, it can skew people's perceptions a lot more easily.