If you model the spread of a rumor as a stocastic process occuring between graph nodes, with nodes having a chance to pass the rumor to neighboring nodes as the product of how interesting (unexpected, novel, prurient) the rumor is and how gossipy the node is, and if each time the rumor is spread a random change is made, the rumors to which errors that increase its interest level occur will spread much further.

This sounds like a model that has some congruence with reality.