One-Tailed vs Two-Tailed Test
A one-tailed test puts all of alpha in one direction, so its p-value is half the two-tailed p-value for the same statistic. This guide shows the difference on a worked example, explains when a one-tailed test is justified and when it is p-hacking, and gives a decision flowchart.
Every hypothesis test asks the same question β is the result far enough from the null to matter? β but one-tailed and two-tailed tests draw the line in different places. A two-tailed test splits the significance level across both tails; a one-tailed test puts all of it in one. That one choice changes the critical value, the p-value and the power, so it is worth getting right before you run the test.
The Alpha Split: One Tail or Two?
Take Ξ± = 0.05. A two-tailed test puts 0.025 in each tail: the result must fall in the extreme 2.5% on either side to reject the null. A one-tailed test puts the whole 0.05 in one tail: the result only needs to fall in the extreme 5% on the chosen side. The diagram below shows the split at Ξ± = 0.05 for a z-test.
| Test | Where Ξ± = 0.05 goes | Critical value (z) | Rejects when |
|---|---|---|---|
| Two-tailed | 0.025 in each tail | Β±1.9600 | |z| > 1.9600 |
| One-tailed (right) | 0.05 in the right tail | 1.6449 | z > 1.6449 |
| One-tailed (left) | 0.05 in the left tail | β1.6449 | z < β1.6449 |
The one-tailed critical value is closer to zero (1.6449 instead of 1.9600), so it is easier to reject the null β but only in the chosen direction. A large result in the other direction counts for nothing.
Worked Example: The P-Value Halves
Take a z-statistic of 1.96. The two-tailed p-value is 0.0500 β the probability of a result at least that extreme in either direction. The one-tailed p-value is 0.0250 β the probability of a result at least that extreme in the chosen direction only. Same statistic, same data, half the p-value.
The same halving works for the t-distribution. With t(20) = 2.10, the two-tailed p-value is 0.0486 and the one-tailed p-value is 0.0243.
The rule: for symmetric distributions (z and t), the one-tailed p-value is exactly half the two-tailed p-value for the same statistic. One-tailed p = two-tailed p Γ· 2.
Work out any p-value yourself with the p-value calculator β it reports both the one-tailed and two-tailed p-value for any z-score.
When a One-Tailed Test Is Justified β and When It Is P-Hacking
A one-tailed test is justified when the hypothesis is directional and was stated BEFORE seeing the data. “The new drug lowers blood pressure” is directional: only a decrease counts as evidence. If the drug raised blood pressure, that would not support the claim β and a one-tailed test correctly ignores that direction.
Choosing the tail AFTER seeing the result is p-hacking. If you run a two-tailed test, see p = 0.08, and then switch to one-tailed to report p = 0.04, you have doubled your false-positive rate: you picked the tail that makes the result significant. The tail must be chosen before the data are examined, and the choice must be recorded β in a pre-registration, a protocol or the methods section.
Not every test has a one-tailed version. The F-test is two-sided by default: it compares two variances, and either one can be larger, so the rejection region sits in both tails. See the F-distribution calculator for the F-test and its critical values.
Choosing a tail after the fact is also a way of inflating the Type I error rate β the false-positive rate. The Type I and Type II errors guide explains what that rate means and why it matters.
Decision Flowchart: One-Tailed or Two-Tailed?
| Step | Question | If yes | If no |
|---|---|---|---|
| 1 | Is your hypothesis directional β does only one direction count as evidence? | Go to step 2 | Use a two-tailed test |
| 2 | Did you state the direction BEFORE seeing the data? | Go to step 3 | Use a two-tailed test |
| 3 | Would a result in the opposite direction be meaningless for your claim? | Use a one-tailed test | Use a two-tailed test |
If any step answers “no”, the two-tailed test is the right choice. When in doubt, use two-tailed: it is the default in most fields, and no reviewer will fault you for it.
The Power Note
A one-tailed test has more power in the chosen direction than a two-tailed test at the same Ξ± β its critical value is closer to zero, so a true effect in that direction is easier to detect. But it has zero power in the other direction: an effect pointing the wrong way can never be detected, however large it is. That is the trade you make. Work out the power for either choice with the statistical power calculator.
All three of the site’s t-test tools β the two-sample t-test calculator, the paired t-test calculator and the one-sample t-test calculator β offer one-sided and two-sided options, so you can run the test either way and compare.
Frequently Asked Questions
Is a one-tailed test ever wrong?
Not when the hypothesis is genuinely directional and was stated before the data were seen. It is wrong when the direction was chosen after looking at the results, or when the opposite direction would also have counted as evidence. In those cases the one-tailed p-value understates the false-positive rate.
Can I switch to one-tailed after seeing my data?
No. Switching tails after seeing the result is p-hacking: it doubles the false-positive rate, because you are picking the tail that makes the result significant. The tail must be fixed before the data are examined.
Which tail do I use for “greater than”?
The right tail. A “greater than” hypothesis (Hβ: ΞΌ > ΞΌβ) predicts large values, so the rejection region is the upper tail and the critical value is positive (for z at Ξ± = 0.05, +1.6449). For “less than” (Hβ: ΞΌ < ΞΌβ), use the left tail and a negative critical value (β1.6449).
Check Tools
Run the test either way with the site’s free calculators:
Two-Sample T-Test Calculator
Compare two independent means, with one-sided and two-sided options.
Try the two-sample t-test calculator β
P-Value Calculator from Z-Score
One-tailed and two-tailed p-values for any z-score, with the shaded curve.
Try the p-value calculator β
Statistical Power Calculator
Power analysis for t-tests β see how the one-tailed choice changes power.
Try the statistical power calculator β
Explore All Statistical Tools
See every calculator on the site in one place.
Check Articles
Go deeper with these guides:
Type I and Type II Errors
False positives and false negatives β what the Ξ± you just split actually controls.
Hypothesis Testing Explained
The full testing framework: null and alternative hypotheses, p-values and decisions.
Z-Test vs t-Test: When to Use Each
Which test statistic to use β and when the z and t p-values differ.