A recent article in Wired about increases in the placebo effect over time has provoked much discussion.  Here, for example, is a good counterpoint from Mindhacks.

But let’s assume that placebo is indeed a potentially effective treatment for psychological reasons.  When you are a subject in a placebo-controlled study you are told that the drug you are taking is a placebo with probability p.  Presumably, the magnitude of the placebo effect depends on p, with smaller p implying larger placebo effect.

This means there is a socially optimal p.  That is, if doctors were to prescribe placebo as a part of standard practice, they should do so randomly and with the optimal probability p.  Will they?

No, due to a problem akin to the Tragedy of the Commons.  An individual doctor’s incentive to prescribe placebo is based on trading off the cost and benefits to his own patients.  But the socially optimal placebo rate is based on a trade-off of the benefit to the individual patient versus the cost to the overall population.  That cost arises because everytime a doctor gives placebo to his patient, this raises p and lowers the effectiveness of placebo to all patients.

So doctors will use placebo too often.