1

Can someone explain why intuitively here the p-value for double tail test is $2\min\{\Pr(X \leq x |H),\Pr(X \geq x |H)\}$ rather then $\Pr(X \leq x |H)+\Pr(X \geq x |H)$? The latter would be 1, but what about the strange formula starting with 2?

user122424
  • 3,926
  • 1
    You're testing your hypothesis using some threshold (alpha-level). Roughly speaking, the test is a black box that gives you the value that you are comparing whether it is bigger than threshold or not. So, in case of two-tail distribution you don't know on which tail you are, and the tails can be not equal, so for these two probabilities. So, as far as I understand, you can take 2minimal probability (you have definitely left the middle of distribution) or 2maximal probability (you're on the one of the tails probably). – Slowpoke May 20 '16 at 12:27
  • Here is some more detailed discussion on subject – Slowpoke May 20 '16 at 12:40

1 Answers1

1

The value $x$ may be in the right or left tail, and it is not specified beforehand. The tail probability should always be smaller than $0.5$. So, for example, if $x$ is in the right tail

$$ 2\min\{P(X\leq x \mid H), P(X\geq x \mid H) \}=2P(X\geq x \mid H). $$

Because we want to know the probability that $X$ is as extreme as our $x$ test statistic. Even though our test statistic is in the right tail, we want to allow for a similarly extreme test statistic to be in the left tail also. So we multiply by $2$. The left and right tails (given our test statistic $x$ are of equal area. For the Normal distribution and $x$ in the right tail, we get

$$ \begin{aligned} 2\min\{P(X\leq x \mid H), P(X\geq x \mid H) \}&=2P(X\geq x \mid H)\\ &=P(X\geq x \mid H)+P(X\leq -x \mid H). \end{aligned} $$

Often, it should be clear from the context whether a two or one sided test is warranted.

jdods
  • 6,248