Maybe it's because I'm lazy and sloppy, but I'd rather the language catch type mistakes than having to write tests for that. Plus, domain modelling with types is nice in ML dialects.
Chances are however, your type error will already occur the first time you run your code.