Thanks, IppolitoMath, for adding that. It's a subtlety that I should have discussed more in the video. The correction of a/n is the upper bound of the p-value assuming independence.
@@mikexcohen1 Thank you, I misspoke I didn't mean "independence" I meant "disjoint". The probability of a finite union of sets is bounded by them sum of the probability of events, and equal when the sets are disjoint. The sets in this case are the events that a p-value is greater than alpha under the null hypothesis. I've not see disjoint mentioned with regards to Bonferroni before.
Yes, FDR is also possible, but it doesn't provide cluster-level corrections that are useful for spatiotemporally correlated data, which I discuss in later videos.
I don't believe Bonferroni assumes independence, but bounds the Type I family rate for a particular set of comparisons which might be dependent by the special case of independence.
Let pi = P(Ti > t) 0
Thanks, IppolitoMath, for adding that. It's a subtlety that I should have discussed more in the video. The correction of a/n is the upper bound of the p-value assuming independence.
@@mikexcohen1 Thank you, I misspoke I didn't mean "independence" I meant "disjoint". The probability of a finite union of sets is bounded by them sum of the probability of events, and equal when the sets are disjoint. The sets in this case are the events that a p-value is greater than alpha under the null hypothesis.
I've not see disjoint mentioned with regards to Bonferroni before.
GREAT WORK!!!- Saved my aignment
Awesome ;)
Awesome Video. How about the False Discovery Rate (FDR) Method that is a new approach to multiple comparisons?
Yes, FDR is also possible, but it doesn't provide cluster-level corrections that are useful for spatiotemporally correlated data, which I discuss in later videos.
I don't believe Bonferroni assumes independence, but bounds the Type I family rate for a particular set of comparisons which might be dependent by the special case of independence.