3 ms·
If the confounders are fat-tail distributed etc. then arbitrarily large samples can still be inadequate. The idea that even thousands a data points in subgroup
by mjburgess 2y ago
If the confounders are fat-tail distributed etc. then arbitrarily large samples can still be inadequate.
The idea that even thousands a data points in subgroups are going to be 'well mixed' relies on extremely strong assumptions about the distribution of those traits.
- xapata 2y agoJust assumptions, not extremely strong assumptions. Ordinary least squares performs well in a large variety of cases.