When is a one-tailed (directional) test appropriate?

When is a one-tailed (directional) test appropriate?

• A. When you have no predictions.
• B. When testing for any difference in either direction.
• C. When you want more power.
• D. When there is a directional hypothesis.

Answer: (D)

Use a one-tailed test when there is a directional hypothesis—that is, you expect the effect to go in a specific direction. When you commit to that direction, all of your alpha level is placed in the single tail where you expect to see the effect, which makes it easier to reach significance if the effect really does go that way. In practice, you would reject the null only if the test statistic is extreme in the predicted direction, increasing your ability to detect the effect you anticipated. This approach, however, loses the ability to detect an effect in the opposite direction, which is why it should only be used when there is a justified prediction about the direction of the effect. If you have no prediction about direction, or you expect possible effects in either direction, a two-tailed test is appropriate. Simply wanting more power isn’t sufficient on its own—the directional hypothesis must be justified by the theory or prior evidence.

Use a one-tailed test when there is a directional hypothesis—that is, you expect the effect to go in a specific direction. When you commit to that direction, all of your alpha level is placed in the single tail where you expect to see the effect, which makes it easier to reach significance if the effect really does go that way. In practice, you would reject the null only if the test statistic is extreme in the predicted direction, increasing your ability to detect the effect you anticipated.

This approach, however, loses the ability to detect an effect in the opposite direction, which is why it should only be used when there is a justified prediction about the direction of the effect. If you have no prediction about direction, or you expect possible effects in either direction, a two-tailed test is appropriate. Simply wanting more power isn’t sufficient on its own—the directional hypothesis must be justified by the theory or prior evidence.