Question

In light of the fact that research always contains the possibility of error, we have to be careful not to present research results as causal relationships when they are only correlations. And yet, sometimes correlation is used. For example, in car insurance, male drivers are correlated with more accidents, so insurance companies charge them more. It would be impossible to actually test this for causation. We cannot change the genders of the drivers experimentally.

Given the above, how do we balance this possibility of error with the need to have statistical results to inform our business decisions?