I've done this and then subsequently given a 30 minute coding test. With the coding test I tried to make it match as closely as possible the kind of activities

I've done this and then subsequently given a 30 minute coding test. With the coding test I tried to make it match as closely as possible the kind of activities the candidate would actually engage with (e.g. processing a CSV rather than implementing a merge sort). There wasn't much of a correlation between how well a candidate chatted and how well they coded, and there was some horrendously bad code and terrible habits exhibited by people who talked the talk pretty well. Conversely, some of the less verbose people wrote excellent code and were often cheaper - possibly because they didn't interview well. The downside of this approach is that crafting (and iterating upon) a realistic test that exercises all the necessary skills and can be squeezed into 30 minutes is hard work.