On Zipf’s law and the bias of Zipf regressions