deadcore

Question: Do time bounded algorithmic coding problems efficiently reduce false negatives as opposed to a technical design interviews (whiteboard & marker), practical coding problem (here's a web-service, add a feature), hiring & exercising probation (this may be a UK thing)?

I know this question has been asked & discussed a million times but I want to take a some what different perspective on the matter other than bashing & moaning

My experience is that the industry norm to ask algorithmic coding problems, for example given two string '1100101' and '101' add together and return '1101010'. In the book 'Cracking the Coding Interview' by Gayle Laakmann McDowell they speak about how these exist to reduce false negatives. However my question is that under the artificial stress of an interview and time pressure, does this actually correlate with ones performance? Such that the classification technique is the best for such screening?

Would be interested to see what other people think?

Plus this is my first 'Ask HN' so go easy one me :)