[simiones]

I think an important difference is that physical models usually have a pretty good idea of what details they are ignoring, and thus some quantifiable idea of the amount of error they may introduce.

Meanwhile, sociological models don't really have any idea of how much relevant detail they are ignoring, meaning that it is far more likely than in physics that the model is simply not useful, equivalent to saying that may make unfounded assumptions (like trying to use a spherical frictionless cow model to compute what kind of cow shoes work best).