Week 8: Feature selection Flashcards
What is the family-wise error rate (FWER)?
It equals the probability to, under the null hypothesis, reject ANY of the null hypotheses (FWER= P_0(reject any true H0).
Why can we not assume the p-values for t- statistics in multiple linear regression to be independent?
Since we cannot bluntly assume the explanatory variables in the model to be independent, due to them being correlated.
What is the Bonferroni correction and how is it calculated?
A method for controlling the FWER in multiple hypothesis testing. It assigns a p-value for each single variable that equals alpha / N. equals the number of tests.
What can we use the Bool inequality for?
To prove that the Bonferroni correction will control the family-wise error rate (FWER) in multiple hypothesis testing.
Why is the Holm method superior to the Bonferroni (for controlling FWER)?
It is able to reject more false H0 than Bonferroni (thus, lower type-II-error).
What is the idea behind the Holm method?
What is beta (hypothesis testing)?
Beta is the probability of a type-II-error; falsely rejecting a true null hypothesis.
Why is it that we measure and talk more about false discoveries, rather than the power? 2 reasons.
Because, 1) generally, in science it is more important to control false discoveries than finding true discoveries, and 2) the distribution “under H0 being false” (P1), is not as easy as P0 to define.
What is the (correct) interpretation of a p-value (post-test)?
The p-value is the fraction of the time that we would expect to see such an extreme value of the test statistic if we repeated the experiment many many times, provided H0 holds/is true.
