A two-sided p-value tells you the probability of seeing a result as extreme as yours, in either direction, if there were truly no effect. It’s the most common type of p-value reported in research, and it accounts for the possibility that a difference could go both ways: higher or lower, better or worse. If you’re reading a study and it reports a p-value without specifying, it’s almost certainly two-sided.
How a Two-Sided P-Value Works
A p-value is a number between 0 and 1 that measures how strong the evidence is against a starting assumption called the null hypothesis. The null hypothesis usually says “there’s no real difference between these two groups” or “this treatment has no effect.” The p-value answers the question: if there truly were no difference, how likely would it be to get results at least this extreme just by chance?
A two-sided p-value considers extreme results in both directions. Say you’re testing whether a new birth control pill changes blood pressure. You don’t know ahead of time whether it might raise or lower blood pressure, so you need to check both possibilities. A two-sided test flags the result as unusual whether blood pressure goes up or down by a surprising amount. If you only checked one direction, you’d be running a one-sided test, which is a different (and less common) approach.
Two-Sided vs. One-Sided Tests
The key distinction comes down to your question. A two-sided test asks: “Is there any difference?” A one-sided test asks: “Is this specifically higher?” or “Is this specifically lower?” That choice changes how your significance threshold gets applied.
With a standard significance level of 0.05, a two-sided test splits that 5% evenly across both ends of the distribution. That means 2.5% sits in the upper tail and 2.5% in the lower tail. Your result counts as statistically significant only if it falls in the most extreme 2.5% on either end. A one-sided test puts the full 5% in just one tail, making it easier to reach significance in that one direction, but completely ignoring the other.
Because the most common statistical distributions are symmetric, converting between the two is straightforward. A two-sided p-value is typically double the one-sided p-value. So if a one-sided test gives you p = 0.03, the two-sided version would be p = 0.06, which flips the result from “significant” to “not significant” at the 0.05 threshold.
Why Two-Sided Is the Default
Two-sided tests are the standard in most published research for a simple reason: they’re more conservative. They protect against being fooled by an effect in the unexpected direction. If you test a drug expecting it to help and it actually causes harm, a two-sided test can catch that. A one-sided test looking only for benefit would miss it entirely.
The FDA uses two-sided testing at the 0.05 level as its standard framework for evaluating clinical trials. In its guidance, the agency notes that an alpha of 0.05 for a two-sided test means the probability of falsely concluding that a drug differs from a placebo, in either direction, is no more than 5%, or 1 chance in 20. Sponsors can technically choose one-sided tests, but the two-sided approach is far more widely accepted. Confidence intervals reported in trials are also usually two-sided, showing the full range of plausible treatment effects in both directions.
A one-sided test is only appropriate when you have a strong reason to care about just one direction and when a result in the opposite direction would be meaningless to you. In practice, that situation is rare. If your alternative hypothesis simply states that two groups are unequal, without specifying which is larger, a two-sided p-value is the correct choice.
Reading P-Values in Practice
When a study reports “p = 0.03 (two-sided),” it means there’s a 3% chance of seeing a difference that large, in either direction, if the treatment had no real effect. Because 3% is less than the conventional 5% cutoff, the result would be called statistically significant.
The significance level itself isn’t fixed at 5%. Researchers can set it at 1%, 5%, or 10% depending on the stakes. A stricter threshold like 1% reduces the risk of a false alarm but makes it harder to detect real effects. The 5% level is simply the most widely used convention.
One common misunderstanding: a p-value does not tell you the probability that the treatment works or that the null hypothesis is true. It only tells you how surprising your data would be under the assumption of no effect. A very small p-value (say, 0.001) means the observed result would be extremely unlikely if nothing were really going on. That’s compelling evidence, but it’s not the same as proof.
A Concrete Example
Imagine researchers want to know whether a hormone affects mating behavior in rats. They inject ten rats with the hormone and give ten others a placebo, then count a specific behavior. The null hypothesis says the hormone has zero effect. Because the hormone could theoretically increase or decrease the behavior, they use a two-sided test.
They calculate a test statistic from their data and check where it falls on a probability distribution. If it lands in the most extreme 2.5% on either end, the two-sided p-value will be below 0.05, and they’ll conclude the hormone likely has some effect. If it falls closer to the middle, the p-value will be larger, and the data won’t provide strong enough evidence to reject the null hypothesis.
The same logic applies in medical research. Testing whether a pill changes blood pressure, whether a diet affects cholesterol, whether an intervention shifts test scores: whenever the effect could plausibly go in either direction, the two-sided p-value is the appropriate measure. It’s a more honest test because it doesn’t assume you already know which way the result should go.