There's no need for correcting for local variations that are needed in econ. Or from a different perspective the fact that we have simple laws in physics, and chemistry, and engieering, is because we are only considering/using a _very_ limited part of the whole universe.
The classic CH4 + 2 O2 = CO2 + 2 H2O is a gross oversimplification, because there are a few dozen other things going on at the same time, and NASA/SpaceX/BlueOrigin has to consider those when they burn in an athmosphere. But since we rarely encounter these problems we don't often find ourselves lacking enough data to fit a better theory. (But of course that's what's happening in post standard model physics, or in cosmology, or in trying to design fusion plants, we are very early in magnetohydrodynamics, and even designing a better corrosion resistant material for use in molten salt reactors is a challenge.)
So, just as we have models for ant ecologies we have for economics of states/countries, but just as neuroscientists have trouble figuring out how exactly an ant works and when does what and where ants will go to forage next economists have trouble predicting when the next crash will happen. (See also bee hive collapse syndrome, we still don't know what causes it, yet there's a lot of hives and we could even do experiments with them easily.)
That does not mean there aren't theories, hypotheses, experiments (usually natural experiments, or small scale pop psychology demonstrations like the dollar auction), and so there is data and you can fit models.
And in economics the hard part is getting good quality data, knowing what to try to quantize, where to start. Some folks spend decades hunting for signal. And then we get priming, and turns out it was nothing. At the same time there's SBTC (skill-biased technological change) the theory describing what happens as automation progresses. But it takes a lot of work to correctly "apply" that theory, because the effects are complex, and you have to keep in mind what else can also affect your observables. (So confounders has to be managed.)
And in the end we get high quality insights, such as the David Autor paper (Why there are still jobs?).
And there are the long and even deeper dives like the "Why nations fail?" book, which talks about the problem of public choice economics (politics) and how to model good politics, how to measure, how to quantify, etc. It's naturally less dense than a paper, but the problem and the pondering is a very important part of science. (The hypothesis generation, the abductive reasoning part.)