I elaborated on the statement, “One should always use two-sided P values”, which is presented in No. 14 on page 342.
Whereas two-sided P values are commonly used in practice, the advantage of using one-sided P values is that there are fewer subjects and resources to underpin significance. Two-tailed P values divide the significance level and it contributes to both sides. Thus, each side of a two-tailed is only half as strong as those of a one-tailed test, which supports all the significance in one aspect. Although one-tailed tests enable more Type I errors, there are many situations in which a one-tailed test could validate the data while a researcher is fully aware of the drawbacks. When there is a very strong reason to validate that one variable is superior to the other, a one-sided test would be applied.
However, a one-sided test will not measure the hypothesis in the opposite direction, so variation can’t be concluded in that direction. In general, two-tailed P values verify the evidence that the control and variation are not the same, while one-sided P values prove that a variation is stronger than the control.