I also thought that. But on second thought I agree with her analysis. The reason that y and y' are different is the same that x and x' are different. It is a problem of lacking resolution, so it will always underestimate more on the higher value, at least if you do enough simulations to get enough samples.