Would be cool to add a case study that introduces to family wise error rates and how to handle them via pf.bonferroni(), pf.rwolf() or simultaneous confidence bands.
This could be done by simulating a dgp with multiple dependent variables under true nulls (i.e. no treatment effect) and by counting false positives under standard inference, Bonferroni & Romano-Wolf corrections.
In a second step, we could then introduce the concept of "uniformely more powerful" tests by comparing the power of the Bonferroni vs RW method of detecting a true effect.
Would be cool to add a case study that introduces to family wise error rates and how to handle them via
pf.bonferroni(),pf.rwolf()or simultaneous confidence bands.This could be done by simulating a dgp with multiple dependent variables under true nulls (i.e. no treatment effect) and by counting false positives under standard inference, Bonferroni & Romano-Wolf corrections.
In a second step, we could then introduce the concept of "uniformely more powerful" tests by comparing the power of the Bonferroni vs RW method of detecting a true effect.