Economics falls under the second category quite a bit as well. In particular its hard to reason about impacts of individual policies on any kind of real complex economy because there are so many confounding variables. And it's not like you can conduct experiments because time travel isn't possible and no two economies are even similar, let alone the same.
I agree that the empirical part of economics is hard, but manageable in modern times [1]. However, the real problem is that even the theory has so many flaws, yet it is used as if it was evidence, which is why I think the real problem is 1).
[1] Some very popular multi-player games have a quite advanced economy, where the individual actions of each participant are recorded and can be analyzed. While this is not "real", it gives more detailed information about human economic interaction than all the mass surveillance in the "real word" right now. Also, the game developers have god-like powers in the virtual world, so they can run controlled experiments,. As far as I know, they already do so. And and if something goes really wrong, they still have the ultima ratio to reset everything to the state of a few hours or days ago. Or, they can re-interpret the last actions of the players in a more appropriate way. That may still make some players unhappy, but won't crash a real economy.
