A critical region (also called a rejection region) is the set of values in a statistical test that leads you to reject the null hypothesis. If your test statistic lands inside this region, you conclude that your results are statistically significant. If it falls outside, you don’t reject the null hypothesis. The critical region is one of the core concepts in hypothesis testing, and understanding it makes the entire framework click into place.
How the Critical Region Works
Every hypothesis test starts with a null hypothesis, which is essentially a default assumption. Maybe it’s “this drug has no effect” or “the average score equals 75.” The critical region is the zone of values that would be so unlikely under that assumption that you’d reject it in favor of an alternative explanation.
The boundary of the critical region is set by a threshold called the critical value. You calculate a test statistic from your data, then compare it to that critical value. If your test statistic is more extreme than the critical value (farther from what the null hypothesis predicts), it falls in the critical region and you reject the null hypothesis. If it’s less extreme, you don’t. That’s the entire decision rule.
For example, suppose you’re testing whether a population mean is greater than 3, using a sample of 15 observations and a significance level of 0.05. The critical value turns out to be 1.7613 on the t-distribution. If your calculated test statistic exceeds 1.7613, it lands in the critical region, and you reject the null hypothesis.
What Determines the Size of the Critical Region
The size of the critical region is controlled by the significance level, denoted by the Greek letter alpha (α). Alpha represents the probability of rejecting the null hypothesis when it’s actually true, which is called a Type I error. You choose alpha before running the test, and the most common choices are 0.05, 0.01, and 0.10.
Setting alpha to 0.05 means the critical region covers 5% of the total area under the probability distribution curve. A smaller alpha, like 0.01, shrinks the critical region to just 1%, making it harder to reject the null hypothesis but also reducing your chance of a false positive. A larger alpha, like 0.10, does the opposite: the critical region expands, making rejection easier but increasing the risk of a Type I error. The tradeoff is straightforward. A bigger critical region catches more real effects but also lets in more false alarms.
One-Tailed vs. Two-Tailed Tests
Where the critical region sits on the distribution depends on the type of test you’re running.
In a one-tailed test, the entire critical region is concentrated in one end (tail) of the distribution. You use this when you only care about detecting an effect in one specific direction. For instance, if you’re testing whether a new drug is less effective than an existing one, you’d place the critical region in the left tail. The full alpha goes into that single tail, so at α = 0.05, the critical region covers 5% of one side. For a standard normal distribution, the critical value is 1.645 (or -1.645 for a left-tailed test).
In a two-tailed test, the critical region is split between both ends of the distribution. You use this when a difference in either direction would matter. Alpha gets divided in half, with 0.025 in each tail when α = 0.05. This pushes the critical values farther out. For a standard normal distribution, the two-tailed critical values at α = 0.05 are ±1.96, compared to 1.645 for a one-tailed test. That means your test statistic needs to be more extreme to land in the critical region.
Choosing between the two isn’t just a technical decision. A one-tailed test is appropriate only when you genuinely have no interest in detecting an effect in the opposite direction, and when missing that effect wouldn’t be irresponsible. Picking a one-tailed test solely to make it easier to reach significance is considered inappropriate practice.
Common Critical Values to Know
For tests using the standard normal (Z) distribution, the most widely referenced critical values are:
- α = 0.05, one-tailed: 1.645 (or -1.645)
- α = 0.05, two-tailed: ±1.96
- α = 0.01, one-tailed: 2.33 (or -2.33)
- α = 0.01, two-tailed: ±2.575
These values assume a large sample or a known population standard deviation. When you’re working with smaller samples and an unknown standard deviation, you use the t-distribution instead, which has heavier tails. This means the critical values are larger (farther from zero) for small samples. With 15 observations and α = 0.05 in a two-tailed test, the critical values are ±2.1448 rather than ±1.96. As sample size grows, the t-distribution converges toward the normal distribution. By around 100 degrees of freedom, the t-distribution critical values are nearly identical to the Z values: 1.984 versus 1.960 for a two-tailed test at α = 0.05.
Steps to Find the Critical Region
The process follows a consistent sequence regardless of the specific test:
- State your hypotheses and choose α. Define the null and alternative hypotheses. Set your significance level, typically 0.05 or 0.01.
- Identify the test statistic and its distribution. Depending on your data and what you know about the population, this might follow a normal distribution, a t-distribution, or another distribution entirely.
- Find the critical value(s). Using statistical tables or software, look up the value that corresponds to your chosen α and the appropriate distribution. The alternative hypothesis tells you whether you need one critical value (one-tailed) or two (two-tailed).
- Calculate your test statistic from the sample data. For a Z-test, this is (sample mean minus hypothesized mean) divided by the standard error.
- Compare and decide. If the test statistic falls in the critical region, reject the null hypothesis. If it doesn’t, you fail to reject it.
Critical Region vs. the P-Value Approach
The critical region method isn’t the only way to make a decision in hypothesis testing. The other common approach uses p-values, and both will always give you the same conclusion for the same data and significance level.
The difference is in how you get there. With the critical region approach, you define a boundary before looking at the data, then check whether your test statistic crosses it. With the p-value approach, you calculate the exact probability of observing a test statistic as extreme as (or more extreme than) what you got, assuming the null hypothesis is true. If that probability is less than or equal to α, you reject the null hypothesis.
Think of it this way: the critical region method asks “did my result land in the rejection zone?” The p-value method asks “how unlikely is my result?” They’re two angles on the same question. The p-value approach gives you more granular information because you see the exact probability, not just a yes-or-no answer. The critical region approach can feel more concrete because you have a clear, pre-set boundary. In practice, most statistical software reports p-values, but understanding the critical region gives you the intuition for why a particular p-value leads to rejection or not.