You are entirely correct in that the crux of the issue is determining which of the initial dimensions are most responsible for the effect you are trying to expl

You are entirely correct in that the crux of the issue is determining which of the initial dimensions are most responsible for the effect you are trying to explain. The trouble is that the the very nature of the subjects of soft sciences is such that it isn't possible (or at least practical) to perform any kind of rigorous principle component or sensitivity analysis. The result is that dimensional reduction intrinsic to the vast majority of studies in economics, psychology, and the like is built purely on institutional bias, subject to fads and dogma. People essentially build careers around how effectively they can peddle their particular flavor of bullshit - this isn't physics or chemistry or engineering where there is a minimally ambiguous answer. We don't have the tooling (or arguably the right institutional culture) to study soft sciences with rigor and that's dangerous because the academic establishment influences social and political policy with comparatively little self awareness of it's vulnerability to bias.