Towards a general model of non‐random sampling and the impact on population correlation: Generalizations of Berkson's Fallacy and restriction of range