When other unanalysed effects would dominate. It works fine if everything else can be assumed equal. However usually everything else is not equal, that is there tends to be other meaningful properties other than that one that is apparent or understood.
If one could arrive at a proof that the population is really homogenous in respect to unanalysed properties then slighter effects could be focused on. But there is a big difference between the population appearing homogenous because you can't analyse it sufficently and it actually being homogenous. |