Judging by the cyclists on our roads and the new members at my gym, the Olympics has inspired Rutland's fattest women to take up exercise rather than the milfiest. If this is the case - and what could possibly be wrong with anecdotal evidence? - it raises an issue about the fragility of equilibria in networks.

You might think that cost-benefit thinking predicts that fatter folk would be more likely to take up exercise, because the benefits of moving from (say) a BMI of 30 to one of 25 are greater than those of going from 25 to 20 whilst the costs of doing so are smaller; the first few pounds fall off very easily.

Such thinking, however, fails to explain why folk get fat in the first place. One reason why they do, economists believe, lies in peer (pdf) effects. If our friends and neighbours are chubby, we are more likely to become so, because being surrounded by lardies increases our estimation of what represents a normal weight.

This is where the Olympics enters. Jess and Victoria have given women very salient role models. The comparator for body weight, for some people, has shifted from their fat friends to slimmer Olympians, which has inspired them to lose weight.

This is why I say equilibria are fragile. Our behaviour often imitates others. But which others? Changes in whom we imitiate can change behaviour very radically. Take just two examples:

- Riots. If one or two people follow the first person to cause trouble, a riot can spread - as subsequent people follow the group. But if they don't, the riot won't happen. It's thus possible that the choice of just one person can make the difference between there being a riot and not, as Mark Granovetter explained in his threshold model (pdf).

- Excess volatility in stock markets. Sometimes, investors herd; they buy when others buy (pdf) and sell when others sell. When this happens, we get bubbles and crashes. The shift from one to the other can be triggered by a very small cue, which stops or reverses imitative behaviour. This was the story of the 1987 crash, and it might be story of Facebook now.

What we have in these cases is the same as we have for Rutland's women - a shift in equilibrium, from peaceful behaviour to rioting, from believing an asset is cheap to thinking it's dear, and from being fat to taking exercise.

But here's the key point. Although it's possible to explain such shifts after they have happened as the result of shifts in peer effects, information cascades and social learning, they are pretty much impossible to forecast in advance.As Jon Elster wrote:

Sometimes we can explain without being able to predict, and sometimes predict without being able to explain. True, in many cases one and the same theory will enable us to do both, but I believe that in the social sciences, this is the exception rather than rule. (Nuts and Bolts for the Social Sciences, p8)

The failure of the social sciences to make successful predictions, therefore, might reflect not a failure of science, but instead the fact that social phenomena are just so damned complex.