well...most of the time its done is to make the “math work out” and in the past its paid dividends! Experimental physics catches up and shows the math intuition was sound (antiparticles, neutrino, etc). Unfortunately nowadays both cosmology & particle physics are at a point where experiments are either very hard & super expensive or (mostly!) there’s no know way to do them. So you make do with the math and hope it tells you something and experimental physics finds a means to catch up. Can you be full of crap in your math jiggles most certainly!
The problem with the pure math modelling going on with things like the mutliverse and string theory is that the theories might begin with math for which there is some empirical basis, and then not just the formula but factors plugged into the formula are invented to make the model produce what the theorists has already imagined. Its as if you could invent factors and math for them, what you have invented MUST be real, somewhere. It’s not science, it’s science fiction.
Experimental physics catches up and shows the math intuition was sound (antiparticles, neutrino, etc). Unfortunately nowadays both cosmology & particle physics are at a point where experiments are either very hard & super expensive or (mostly!) there’s no know way to do them.
...
It does seem to have changed. For relativity and quantum mechanics it was the strange experimental results that led the way.