3 ms·
I guess my point is, if outcomes are affected by a couple dozen variables, unless we can control for all those variables, then we might not see any affect from
by goda90 6y ago
I guess my point is, if outcomes are affected by a couple dozen variables, unless we can control for all those variables, then we might not see any affect from this one variable. And there is a limit to how many variables we can reasonably control for.
- zwaps 6y agoControling for variables is important to establish causality in observational studies. Many fields in social sciences are developing increasingly sophisticated methods to tease out causal relationships (see Pearl, Angrist&Pischke etc.). However, many of these methods try to obtain the conditions similar to the gold standard: Randomized Control Trials. In an RCT, if you assign treatment randomly to balanced subgroups of a population you care about (or one that is even representative), then you do not require to controls. To see that, imagine the treatment being independent of the confounding factors and the distributions across group being the same. It then follows that you can estimate the average treatment effect without bias.
- davinic 6y agoAnd also the effects of those couple dozen are nonlinear and the degree of nonlinearity probably varies from subject to subject.