[Dennis Campbell]

That's the main reason I take issue with workplace 'personality tests.' They're not based on performance, but on sloppy and largely unsupported generalizations and predictions, which I suspect are used to skirt equal employment regulations.

Depends on the test, the research behind it and the applications. There are well-based tests wrongly applied, poor tests badly applied and some good procedures well applied that are quite accurate. "Quite accurate" meaning a correspondence statistic between say .75 and .85 from test prediction to outcomes in real life.

By the way, “validity” as I use it applies to the statistical correlation of some test result with some other measureable observation, either concurrent or in the future. That is also expressed between 0 and .99.

Perhaps an actual example would help. In the 1960s Peace Corps candidates were assessed prior to placement in some location. They were subject to (l) a psychiatric interview, (2) a series of psychological tests, and (3) ratings from each other, as they lived and trained together. Those results were correlated with actual outcome measures such a dropout rate, terminations, and complaints. The best predictor was peer group ratings, the 2nd best was tests and the worse was psychiatric interviews. So the “goodness” was defined by those statistical correlations.