Psychological science uses this technique a lot. It's not perfect, of course there's lots of noise because of both of the issues you raise (in addition to other

Psychological science uses this technique a lot. It's not perfect, of course there's lots of noise because of both of the issues you raise (in addition to other issues like how to develop questions to measure nebulous things like "self-esteem" or "extraversion"). But as long as the results aren't pure noise, i.e., there's some signal there, then with an adequate sample size you can do something with it. Lots of fields deal with noisy, imprecise data. It's just conditional on it not being only noise.