Questions: Hypothesis Construction: Directional and Nondirectional Predictions

Question 1

After collecting data, a researcher notices a significant effect in the opposite direction from what they expected. They decide to report it as a one-tailed test aimed at their original predicted direction to maintain consistency with their hypothesis. What is the problem?

• A. One-tailed tests can only detect effects in the predicted direction, so the significant result would become non-significant — and choosing the test direction after seeing the data is p-hacking
• B. Nothing is wrong; the researcher is being transparent about their original hypothesis
• C. One-tailed tests are never appropriate when an opposite-direction effect is observed
• D. The researcher should have used a chi-square test instead

Answer: (A)

A one-tailed test concentrates all Type I error budget in one tail. If you observe an effect in the *other* tail, a correctly applied one-tailed test treats it as non-significant — that is the literal meaning of the test. More seriously, choosing the test direction after seeing the data is exactly what p-hacking looks like: you are picking whichever tail makes your result significant. The power advantage of one-tailed tests is only legitimate when direction is specified before data collection, because the power gain comes from a genuine prior commitment to ignore effects in the other direction.

Question 2

Compared to a nondirectional (two-tailed) hypothesis test at the same alpha level, a directional (one-tailed) test...

• A. Is always more rigorous because it reflects greater theoretical knowledge about the effect direction
• B. Has less statistical power because it uses only one tail of the sampling distribution
• C. Has more statistical power for detecting effects in the predicted direction, but requires treating opposite-direction effects as non-significant regardless of their magnitude
• D. Produces identical results to the two-tailed test when the observed effect is large

Answer: (C)

The one-tailed test places its entire alpha (e.g., 0.05) on one side of the distribution, making the critical value easier to exceed in that direction — hence higher power for predicted-direction effects. But this gain requires a real commitment: if a large effect in the opposite direction appears, the one-tailed test reports p > 0.05. You've agreed, in advance, that an opposite-direction finding is not what you're testing for. If you would actually act on an opposite-direction finding, you should use a two-tailed test.

Question 3

Directional hypotheses are generally preferable to nondirectional ones because they reflect stronger theoretical knowledge and provide more statistical power.

• T. True
• F. False

Answer: False

Directional hypotheses are only preferable when you have strong prior theoretical or empirical grounds for the direction and you are genuinely willing to ignore opposite-direction effects. Without that prior justification, directional tests provide a false power boost: you are reporting higher statistical precision than your actual state of knowledge warrants. When the literature is mixed, the field is new, or the direction is genuinely uncertain, a nondirectional hypothesis is the more appropriate and scientifically honest choice.

Question 4

A researcher who specifies a directional hypothesis only after observing their data cannot legitimately claim the statistical power advantages of a one-tailed test.

• T. True
• F. False

Answer: True

The power advantage of a one-tailed test comes from the prior commitment to ignore effects in the unpredicted direction. If direction is chosen after data collection, no such commitment was made — you saw which direction worked and then declared it 'predicted.' This is a form of p-hacking that inflates Type I error rates. Preregistration formalizes the necessary commitment: specifying the direction before data collection creates a verifiable record that the prediction was genuinely prior.

Question 5

Explain why the power advantage of a one-tailed test comes with a genuine scientific cost, and describe when a directional hypothesis is scientifically defensible.

Model answer: The power gain comes from concentrating all the Type I error budget in one tail — but this means agreeing to treat effects in the opposite direction as non-significant, no matter how large they are. This is a real cost: if an unexpected opposite-direction effect is scientifically important, the one-tailed test will miss it. A directional hypothesis is defensible when (1) a robust prior literature strongly supports the direction, (2) the researcher is genuinely uninterested in opposite-direction effects, and (3) the direction was prespecified before data collection.

These three conditions correspond to three different types of validity. Condition 1 is theoretical validity — the direction is grounded in evidence. Condition 2 is decision-theoretic validity — the test aligns with what the researcher would actually do with an opposite-direction result. Condition 3 is statistical validity — the power calculation is based on an honest commitment, not post-hoc rationalization. When all three hold, one-tailed testing is a scientifically sound choice. When any fails, a two-tailed test is more honest.