These comments are all over the place. If you must look at correlations, there are some more rigorous methodologies to study cell phone use or vehicle weight you can use.
My preference is when you're given historical data is to find a natural experiment and use something like an interrupted time series regression. For example, if there are two cities with mostly similar characteristics but one criminalizes cell phone use more explicitly, you can compare statical differences in accidents while controlling for historical trends.
And that's just from the top of my head. That said, there's fairly concrete evidence without looking at correlations... at least, we know heavier cars are much more lethal from physical tests, no correlation needed. The historical trends also have conflicting factors too, like safety features; e.g. mandatory rear view camera in... 2017, I think it was? So yeah, things can also just be murky if you don't isolate your variables.
Well, that's why you hope that gets controlled with a natural experiment. You can't, for instance, use a control group in a country with different medical policies, but you could have comparable cities and neighborhoods that, arguably, should have similar medical advancements over time. The closer the control group is, the better your validity should be (too bad there's no mirror universe!).
There's a couple other tricks you can use, but honestly my expertise is in education so I'm not sure how widely these are used; Instrumental variables (basically proxies for your target) for instance, like adding property value or something to the model can inadvertently control for things associated with that, like better medical care or infrastructure. You risk over-specifying the model but we have diagnostics that help with that.
(Btw, education, the main concern is there's a billion possible factors in the home, classroom, society, etc, we can't directly capture, so that's how it's used there).