Over at his other blog Razib talks about the selective pressures that shaped the modern distribution of skin color. One thing he didn't emphasize but which I found illustrative is that loss of function plays a prominent role in this story as a special sort of adaptation: sometimes losing something can be good. But this raises a minor problem for interpreting some kinds of tests of selection: When you run, say, the McDonald-Kreitman test on MC1R in Europeans, the verdict returned is "neutral". But at the same time, the loss of function that turned this locus nearly-neutral is an adaptation!

The fact that a locus has faded into the nearly-neutral background isn't evidence against adaptation -- quite the opposite, in fact. Nothing is metabolically costless, so if the cost is significant and the benefit nolonger exists then there'll be selection for loss of function. Relaxation of constraint is just one way of shifting the fitness peak, and natural selection will respond to that as it always does. This is a convenient example of the general principle that you can't just take the outputs of canned statistical tests at face value: They require interperetation in the light of theory and history.