The method of random sampling has an extremely strong position in official statistics. However, in the past decades, high quality surveys based on random sampling have become more difficult and expensive to conduct due to a hardening survey climate. The issue whether we should prefer random sampling to non-random methods pushes itself to the forefront of the agenda, as it did in the beginning of the 20th century. I will try to clarify what this issue is about.

Recent research on ‘balancing’ (Särndal & Lundquist 2014) or ‘representativeness’ in sample surveys seems promising, although this is still ongoing research. I will argue that this may pull us away from random sampling towards obtaining a ‘good’ set of data that will allow for reliable inference, no matter whether that good data set has been obtained through a random sampling mechanism or not.