This paper from 2010, is one of the best papers I have read that describes how choices of metrics that don't sound like a big deal (like ratios, or subtractions), or distance units, encode value judgements, giving a bunch of case studies.