This is the example used in methodological discussions of economics. I don't know if astronomers do actually perform experiments on some limited scale, but it seems that classicaly they relied on "luck" to test predictions e. g. testing general relativity through observations of eclipses. In social sciences, there is a somewhat similar thing called "natural experiment", which is the main test for the validity of predictions. In both cases, the researcher takes a more passive role in testing his predictions.
This can be dauting to researchers, but overall it's for the best. There are indeed experiments in economics, specially in game theory. But the majority of possible experiments would cross the boundary of what is morally acceptable, and well into morally abhorrent e. g. randomly restricting a sample's access to education to estimate it's effects on wages.
Well, certainly, natural experiments can lead to knowledge. But that's a much slower and more dangerous path, artificial experiments are better in every sense, except that they are mostly not available for economics. Anyway, I don't think the use of natural experiments are the bottleneck currently holding economics back.
I'm not going to defend sociology specifically because I don't know the field or it's state, but in every science (e.g. chemistry), you make a model for some restricted case to fit some data and you make assumptions on the scope of the model. Otherwise you couldn't say pretty much anything about anything: "We tested that proton and that one has a mass of X units. We can't say anything about this proton, thogh." -- the evidence piles up that the model has widespread vality. In the same sense, we can make assumptions on the scope of models. Of course, I imagine we have no hope for the time being of uniting the basic physical laws to that of sociology, simply because we don't have the power to understand the human brain just yet. I don't think refraining from modelling behavior is useful, and it will probably even help us better model the brain and generalize it's behavior towards "better" intelligence (AI).