Take home coding tests as a 0th round interview step can be extremely effective if they are short (30m-1h), relatively easy pulse checks (a step or two above FizzBuzz level). Any more involved than that and they are insulting to candidates, thus driving away talent.

I worked for a company that administered a simple half-hour take home coding challenge before any interviews. The problem was simple but not trivial — print Fibonacci numbers, take a string of brackets and print if they were properly nested (e.g. '()(())' is OK, '(()(' is not).

We had essentially zero attrition of candidates with good CVs who refused to take the test. On the other hand, we disqualified >80% of applicants from the test alone, saving an enormous amount of interviewer time. It would have sucked to have phone screened all those candidates and spent half an hour of company time on each one to discover they can't code at all.