Pairwise t-test

  • we have $n$ groups, $n > 2$
  • we conduct a series of Two-Sample t-tests to find out which groups are different
  • e.g. in post-ANOVA analysis


Controlling Family-Wise Error Rate

It's important to modify $\alpha$ to avoid Type I Errors

  • when we run many tests, it's inevitable that we make them just by chance

E.g. use Bonferroni Correction

  • use modified confidence level $\alpha^* = \alpha \cdot \cfrac{1}{K}$
  • where for $k$ groups $K= \cfrac{k \cdot (k - 1)}{2}$


Sources

Machine Learning Bookcamp: Learn machine learning by doing projects. Get 40% off with code "grigorevpc".

Share your opinion