I came across an interesting article about Yahoo researcher Duncan Watts and his empirical findings on how trends spread through a modern, networked society. The article tended to pit Duncan against Malcolm Gladwell, author of the Tipping Point, as well as other leading marketers. In essence Duncan debunks the theory that influential members of our society are responsible for trends taking off like wildfire. His research has shown, by exercising both real-world and computer models of mini-societies, that lots of factors contribute to a trend taking off: Joe six-pack is just as likely as a highly social, well-connected and well-regarded citizen to set a trend in motion.
Gladwell and his contemporaries argue that while Duncan’s results are interesting, they stand by their “real-world” results. Duncan, in-turn, states that their results don’t necessarily point to a root cause, with conclusions largely theoretical. Tempers flare.
As I am embarking on a real world modeling effort myself, it struck me as odd that such contentions might really exist: both parties are right, and what the article really is pointing out is a very common dilemma in any engineering endeavor. Models are, by definition, inaccurate.
Models are used as a utility, and well-designed ones make approximations and assumptions that are understood and accepted. I’ll use an example from my old acoustics days. If you wanted to build a sealed wooden box with a speaker stuck in it, you can use the following formula:
This is not as bad as it looks, as 
is the volume of the box, and the other values come with the handy spec sheet of the woofer you buy. A few simple punches in the calculator and some wood glue, and you have a decent sounding speaker!
This is a great model, because it boils down very complicated acoustics into high school algebra. If you studied acoustics you will find more elaborate models at work, but even those are inaccurate. You can involve thermodynamics and even human hearing physiology, but there are diminishing returns for all this work. The equation above gets you 90% of the way…reams of additional study will get you maybe a few points more. That equation bakes in literally decades of work, with its principle authors essentially saying “trust us, it will sound fine.”
So, in the case of studying the viral trends, Duncan’s models probably serve his needs of predicting the outcome of Yahoo’s advertising programs, within a certain ballpark at least; but they are not perfect. Gladwell and others, however, are also correct in basing hypotheses on the observed outcome of various real world studies; but they cannot prove it. There usually exists a gap between observed outcome and predicted outcome in natural systems. I say “usually” as there are exceptions, and these exceptions have largely changed our world — something I’ll have to write more about later!
(Photo from the source article. The pretty printed math courtesy of the great wpmathpub plugin!)