When are models too complex? What counts as independent (and representative) data for model testing?

In a Hindsight blog for Functional Ecologists, of the British Ecological Society, Kai Chan and Edward Gregr tackle some crucial questions about model building and validation. Everyone knows that you need independent and representative data to properly test a model, but what counts as 'independent' and 'representative'? How are these concepts affected by scale? If you want to keep up with evolving norms for proper model building and testing, this post may help.