Answers are grounded in this paper and cite their sources.
The authors show from two concrete examples that when comparing different subgroups attained from different total groups it is in general impossible to draw any conclusions about the difference between the total groups. That will hold, of course, with the assumption that the subgroups are attained by natural self‐selection instead of random sampling as in experimental design. This rather paradoxical situation depends on differences in the strength of selection. This strength of selection is decided mainly on the one hand by the relative distribution over subgroups, on the other hand by the overlapping of subgroups on the investigated variable.
No takes yet. Share an insight, caveat, or question.
Lundgren et al. (1969) studied this question.
Synapse has enriched 3 related papers. Consider them for comparative context: