A lot of my interaction between coders is explaining to each other what we've done, I'd prefer someone who might be a worse coder than me but can explain things well (and write good commit messages, etc.) than a guy who writes epic code but can't explain it to me.
We also have a programming test. Everybody is subjected to it. In the beginning (when I had to do it :) ), thought it was stupid, but now I find this a valuable tool to aid with the decision (we frequently hire people who did not finish the test). The problems are relatively simple. You get 2 hours alone in a room, a programming language of your choice, and internet connection (so you can search for stuff).
As for programming tests, I think you're playing into my bias about how everyone thinks the questions are stupid until they pass, at which point the questions become the arbiter of all intelligence and merit. ;) It's that sort of hazing ritual sociology that enshrines questions about moving Mt. Fuji or printing n to the power of 100 in decimal form.
Even just a simple a question as: "We're doing XYZ. To handle this situation what would you do?" is incredibly telling.
To my mind, being able to apply knowledge is the valuable skill, otherwise it is just trivia that a person knows.
I've had to write my own sorting algorithm perhaps twice in the last some-odd years. If you're writing sorting algorithms from scratch, you probably should sit back and consult your standard libraries.
So if I ask a candidate some basic coding question and he can't write a for loop, I know it's a no-hire. But past that basic bar, sifting apart the good candidates from the great ones is actually very hard. Do we continue to ask these questions because we can't do better?
(That's why I ask algorithm questions. I realize they suck, but I just don't have any better ideas. I'm an engineer, not an HR flack.)
So what's a better way to test a programmer candidate?