Types are just one kind of static analysis. So it makes sense that you'd want the superset rather than the subset.

It's not so clear cut. Types are compositional, by definition.

Static analysis needn't be compositional, and so a small change could cause dramatic and unpredictable long-range effects elsewhere in your code. Figuring out what changed or what went wrong can be a nightmare.

I had the chance to experiment this idea a few years ago. I had a text editor which could only lint Javascript. My project was a haskell+js hybrid. There was no analysis for hs aside from running the compiler and seeing the output.

Writing the hs part was a nightmare at first. Once the project grew and I needed refactoring, js became the main nightmare. Hs simply did its thing which at that point, I already had got used to it. Js couldn't tell me if anything wrong outside of function boundaries.

A smarter js linter may prove me wrong but I believe types serve a long term goal by sacrificing productivity in the beginning.

I don't understand this dichotomy - type checking vs. all other static analysis? Or unchecked types vs. static analysis without types? Or something else?