[verteu]

My interpretation is "If you're unsure enough to flip a coin, then... 1) Being forced to quit your job would make you 5.203 out of 10 points happier(!) 2) Being forced to start your own business would make you 5.203 out of 10 points happier(!) 3) Being forced to propose would make you 5.125 out of 10 points LESS happy(!)"

The size of the effect is much larger than I'd expect.

> Also, based on the GP’s formula, maybe the effects look large because people don’t do what the coin tells them very often so you are dividing by a small number?

I don't think so. The same "small number" exists implicitly in the numerator, since (EV[H | heads] - EV[H | tails]) = 0 for anyone ignoring the coin.

For example, imagine quitting really does make you 5 points happier, but a random 99% of people ignore the coin, and 1% follow it.

The numerator would be (EV[H | heads] - EV[H | tails]) = 0.99*0 + 0.01*5, since happiness is unaffected by coin toss for people ignoring the coin. The denominator would be 0.01, so we'd get B_wald=5 as desired.