And also, interestingly, climate science, where they face a similar problem in having a single enormously complex system that they can't bifurcate and run experiments on. To be fair, even nutrition and digestion and neuroscience/psychology tend to be a pseudo-scientific shitshow, and we've 6 billion test subjects for those. Or even epidemiology, where we get endless statistical garbage about this or that factor leading to cancer. Cancer is a real physical process, triggered by real physical causes (carcinogens, genetics, etc.)... how about we put down the statistics textbook and focus on finding the empirical causes? But again we struggle to do that with 6 billion experimental test beds. So yeah, it is best to take anything like sociology/economics/planetary climate modelling with a big grain of salt, because the simple truth is that we're nowhere near achieving accurate modelling in subjects where we have six billion verification sets, let alone subjects where we have a handful (societies, economies) or one (Earth).