When you set up a hypothesis test, you decide whether your alternative hypothesis is directional or not. That decision makes your test one-tailed or two-tailed — and it directly affects the p-value you get. Choosing correctly (and honestly) matters.
What the "tails" are
Test statistics like t and z follow a distribution with two tails. The tails hold the extreme, unlikely values. A significance test asks: is my result far enough into a tail to be surprising under the null?
- A two-tailed test looks at both tails — it detects a difference in either direction (H₁: μ ≠ value). Your chosen α (say 0.05) is split, 0.025 in each tail.
- A one-tailed test looks at one tail only — it tests a specific direction (H₁: μ > value, or μ < value). The full 0.05 sits in that single tail.
Why it changes your p-value
Because a one-tailed test concentrates all of α in one direction, it's easier to reach significance in that direction — the critical value is closer, and the one-tailed p-value is exactly half the two-tailed p-value (when the effect is in the predicted direction).
| Test | α = 0.05 critical z | p-value relationship |
|---|---|---|
| Two-tailed | ±1.96 | p (baseline) |
| One-tailed | 1.645 (one side) | p / 2 |
That extra power is exactly why one-tailed tests are tempting — and exactly why they're easy to misuse.
The cardinal rule: choose one- vs two-tailed before you see the data, based on the question. Switching to a one-tailed test after peeking (because it makes p < 0.05) doubles your true false-positive rate. Reviewers and examiners look for this.
When is a one-tailed test justified?
Only when a result in the opposite direction would be meaningless or impossible to act on, and you committed to the direction in advance. Examples:
- Testing whether a new manufacturing process reduces defects — an increase wouldn't change your decision (you'd keep the old process either way).
- A non-inferiority drug trial designed to show a treatment is not worse than the standard.
If you'd care about a difference in either direction — which is usually the case — use a two-tailed test. It's the conservative, default choice, and it's what most journals expect unless you justify otherwise.
Try it
Both the t-test and z-score calculators let you pick the alternative (two-sided, greater, or less) and show the corresponding p-value.
Open the t-test calculator →Choose your alternative hypothesis and see one- vs two-tailed p-values side by side.Related reading: Interpreting p-values · Null & alternative hypothesis · T-test vs z-test · Type I & II errors.