Totally agree that experiments have drawbacks. But scarce researcher time has to be allocated across methods. And when most applied economists think of advantages of a typical experiment relative to a typical quasi-experiment, "better statistical power" is rarely on their list and it should be.

What pops out to me is the dramatic difference in power between experiments and non-experiments in Table 2.

But they replace Captcha with the montage at the beginning of "Up" and only let you pass if you cry.

Yeah, that's where my knowledge ends...economics needs a book of counter-example utility functions/preference relations...

(I'm trying to sort out if the symmetry is required by differentiability or just more generally a feature of any preferences that you can represent with a utility function...not immediately obvious to me...)

Could you formalize this with some weird non-differentiable preferences....like two stage utility function that is Leontief over tasks (diapering, baking, etc.) and and tasks are either perfect substitutes or totally irrelevant within task?