1 Fundamental concepts

A rejection region is a pre-specified set of outcomes for a test statistic that would be unlikely if the null hypothesis were true. In hypothesis testing, it provides a formal rule for deciding whether observed data are extreme enough to justify rejecting the null hypothesis. The region is determined before the data are examined, using the assumed sampling distribution of the statistic under the null model.

1.1 Hypothesis testing framework

Hypothesis testing compares two competing statements about a population or process. The null hypothesis represents the default position, while the alternative hypothesis expresses the possibility of an effect, difference, or departure from that default. The rejection region is one part of the decision procedure used to evaluate the evidence.

1.1.1 Null hypothesis

The null hypothesis, often written as H0, states that no effect, no difference, or a specified benchmark value holds in the population. It serves as the reference condition for constructing the sampling distribution of the test statistic. A rejection region is defined so that, if H0 were true, only a small proportion of outcomes would fall inside it.

1.1.2 Alternative hypothesis

The alternative hypothesis, often written as H1 or Ha, states what is considered plausible if the null hypothesis does not hold. It may indicate a directional change, such as a larger or smaller value, or a nondirectional difference. The form of the alternative hypothesis influences whether the rejection region is placed in one tail or both tails of the distribution.

1.2 Test statistic

A test statistic is a numerical summary computed from sample data and used to evaluate hypotheses. It converts the raw observations into a scale on which the null distribution is known or approximated. Common examples include z, t, chi-squared, and F statistics. The rejection region is expressed in terms of this statistic rather than the original data.

1.3 Significance level

The significance level, denoted by alpha, is the maximum probability of rejecting the null hypothesis when it is actually true. It is chosen in advance, often as 0.05 or 0.01, and determines how rare an outcome must be to enter the rejection region. Smaller significance levels produce narrower rejection regions and stricter decision rules.

1.3.1 Type I error

A Type I error occurs when the null hypothesis is rejected even though it is true. The significance level is the long-run rate at which this kind of mistake is tolerated under repeated testing. The rejection region is designed so that its total probability under H0 equals alpha.

1.3.2 Critical value

A critical value is the boundary point or points that separate the rejection region from the rest of the sampling distribution. Test statistics beyond these boundary values fall in the rejection region. The critical value depends on the chosen alpha, the tail structure of the test, and the distribution of the test statistic.

2 Definition of rejection region

The rejection region is the set of test statistic values that lead to rejection of the null hypothesis. It is defined in advance so that, under the null distribution, the probability of landing in that set is equal to the significance level or a closely related target. In practice, it identifies the most extreme outcomes relative to what would be expected if H0 were true.

2.1 Decision rule

The decision rule states that if the observed test statistic lies in the rejection region, the null hypothesis is rejected; otherwise, it is not rejected. This rule converts a statistical calculation into a formal conclusion. The rule is deterministic once the statistic and the predefined boundaries are known.

2.2 Relation to critical values

The rejection region is often described by critical values. These values mark the threshold beyond which observations are considered sufficiently unusual under the null model. The exact form of the region depends on whether the test is one-tailed or two-tailed.

2.2.1 Upper-tail rejection region

An upper-tail rejection region includes large values of the test statistic. It is used when the alternative hypothesis predicts values greater than the null value. If the statistic exceeds the upper critical value, the result is deemed extreme in the positive direction.

2.2.2 Lower-tail rejection region

A lower-tail rejection region includes small values of the test statistic. It is used when the alternative hypothesis predicts values less than the null value. If the statistic falls below the lower critical value, the null hypothesis is rejected.

2.2.3 Two-tailed rejection region

A two-tailed rejection region includes extreme values in both directions. It is used when the alternative hypothesis allows for a difference in either direction. The significance level is split between the two tails, so each tail contains a portion of the total rejection probability.

2.3 Rejection region versus non-rejection region

The non-rejection region contains values of the test statistic that are not sufficiently extreme to justify rejecting the null hypothesis. It is sometimes called the acceptance region, although that term may suggest stronger certainty than the test actually provides. A result outside the rejection region does not prove the null hypothesis; it only indicates that the evidence is not strong enough for rejection.

3 Construction of rejection regions

Rejection regions are built from the theoretical distribution of the test statistic under the null hypothesis. The choice of distribution depends on the parameter being tested, the sample size, and the assumptions of the procedure. Once the distribution is specified, cutoff points are selected to capture the desired tail probability.

3.1 Based on sampling distributions

The null sampling distribution describes how the test statistic behaves when the null hypothesis is true. The rejection region is chosen so that only a small fraction of this distribution lies beyond the critical boundary or boundaries. This makes extreme results rare under H0 and therefore informative when they occur.

3.2 Using critical points

Critical points are numerical thresholds derived from distribution tables or software. They identify the point at which the cumulative probability in the tail matches the chosen significance level. Different tests use different critical points because their test statistics have different null distributions.

3.2.1 Z distribution

The z distribution is used when the test statistic follows, or is approximated by, the standard normal distribution. Rejection regions are determined by standard normal critical values. This approach is common in large-sample tests and in tests involving known population variance.

3.2.2 t distribution

The t distribution is used when the population variance is unknown and the sample size is limited, especially in mean comparisons. Its heavier tails reflect additional uncertainty from estimating variability. Rejection regions depend on the degrees of freedom, which affect the critical values.

3.2.3 Chi-squared distribution

The chi-squared distribution is often used for tests involving variance or goodness of fit. Its rejection region may lie in the upper tail, lower tail, or both, depending on the parameter being tested and how the statistic is defined. Because the distribution is asymmetric, critical values are not centered like those of the normal distribution.

3.2.4 F distribution

The F distribution appears in tests that compare variances or examine models with multiple explanatory terms, such as analysis of variance. Rejection regions are commonly in the upper tail because large F values indicate a stronger departure from the null hypothesis. The critical value depends on two sets of degrees of freedom.

3.3 Tail probability approach

The tail probability approach defines the rejection region by allocating a chosen amount of probability to the extreme tail areas of the null distribution. For a one-tailed test, all of alpha is placed in one tail. For a two-tailed test, alpha is divided between the lower and upper tails. This approach links the geometry of the region directly to the desired error rate.

4 Types of tests

The form of the test determines how the rejection region is arranged across the distribution. Directional hypotheses produce one-tailed tests, while nondirectional hypotheses lead to two-tailed tests. Composite hypotheses may require more nuanced rejection rules because the alternative hypothesis includes a range of possible values.

4.1 One-tailed tests

A one-tailed test concentrates the rejection region in a single tail of the sampling distribution. It is used when departures in only one direction are relevant to the research question. Because all of alpha is placed on one side, the critical boundary is less extreme than it would be in a two-tailed test with the same significance level.

4.1.1 Right-tailed tests

A right-tailed test rejects the null hypothesis for unusually large test statistic values. It is appropriate when the alternative hypothesis asserts that the parameter is greater than the null value. The rejection region begins at an upper critical value and extends to positive infinity.

4.1.2 Left-tailed tests

A left-tailed test rejects the null hypothesis for unusually small test statistic values. It is used when the alternative hypothesis asserts that the parameter is less than the null value. The rejection region extends from negative infinity up to a lower critical value.

4.2 Two-tailed tests

A two-tailed test uses rejection regions in both tails of the distribution. It is appropriate when deviations in either direction are meaningful. Each tail receives half of the total significance level, and the observed statistic must fall far enough from the center in either direction to trigger rejection.

4.3 Composite hypotheses

A composite hypothesis includes more than one possible parameter value. In such cases, the null or alternative hypothesis is not a single point but a range. Rejection regions may still be defined using standard critical values, but the exact power and behavior of the test can vary across the parameter values contained in the composite hypothesis.

5 Interpretation and decision making

The rejection region supports a formal statistical decision, but the conclusion must be interpreted carefully. Rejecting or failing to reject the null hypothesis does not by itself measure the size or importance of an effect. The decision should be considered alongside practical context and study design.

5.1 Rejecting the null hypothesis

If the observed test statistic falls in the rejection region, the null hypothesis is rejected at the chosen significance level. This means the observed result would be sufficiently unusual under the null model to warrant statistical doubt. It does not prove the alternative hypothesis, but it indicates that the data are inconsistent with the null assumption.

5.2 Failing to reject the null hypothesis

If the test statistic does not fall in the rejection region, the null hypothesis is not rejected. This outcome means the evidence is not strong enough, given the chosen threshold, to conclude that the null hypothesis is implausible. It should not be interpreted as confirmation that the null hypothesis is true.

5.3 Statistical significance

Statistical significance refers to the outcome of a test when the test statistic enters the rejection region or, equivalently, when the p-value is sufficiently small. It indicates that the observed data are unlikely under the null hypothesis. Statistical significance is a property of the test result, not necessarily of the practical usefulness of the finding.

5.4 Practical significance

Practical significance concerns whether the observed effect is large or important enough to matter in real-world terms. A result may be statistically significant but have a negligible effect size. Conversely, a practically important result may fail to reach the rejection region if the sample is too small or the variability too high.

6 Examples

Examples show how rejection regions operate in common statistical settings. Although the specific formulas differ, the underlying logic remains the same: define a null distribution, choose a significance level, and identify extreme values that trigger rejection.

6.1 Mean testing

In a test of a population mean, the statistic may be a z score or t score comparing the sample mean with a hypothesized value. If the alternative is that the mean is greater than the null value, the rejection region lies in the upper tail. If the alternative allows deviation in either direction, the rejection region is split across both tails.

6.2 Proportion testing

In a test of a population proportion, the test statistic often measures the distance between the observed sample proportion and the hypothesized proportion. Large positive or negative deviations may lead to rejection depending on the alternative hypothesis. The rejection region is commonly based on the normal approximation when sample conditions are suitable.

6.3 Variance testing

In variance testing, the chi-squared statistic compares the sample variance with the null variance. Because the chi-squared distribution is skewed, the rejection region is typically not symmetric. Small or large values of the statistic may matter depending on whether the test concerns a lower variance, higher variance, or a difference in either direction.

6.4 Regression and ANOVA contexts

In regression and analysis of variance, the F statistic evaluates whether a model explains a meaningful amount of variation relative to unexplained noise. The rejection region is usually in the upper tail, since large F values suggest that the null model is inadequate. These tests are widely used when comparing multiple means or assessing whether predictors contribute to a model.

Several closely related ideas are used alongside rejection regions in hypothesis testing. These concepts help interpret test results, compare evidence, and describe the behavior of a test under repeated sampling.

7.1 p-values

A p-value is the probability, under the null hypothesis, of obtaining a test statistic at least as extreme as the one observed. It offers an alternative way to summarize evidence against the null hypothesis. A p-value below the significance level corresponds to a statistic falling in the rejection region.

7.2 Confidence intervals

A confidence interval provides a range of plausible values for a parameter based on sample data. For many standard tests, whether a hypothesized value lies inside or outside the confidence interval corresponds to whether the null hypothesis is rejected. Confidence intervals also convey information about effect size and precision.

7.3 Power of a test

Power is the probability that a test will reject the null hypothesis when a specific alternative is true. It depends on the effect size, sample size, variability, and significance level. A well-chosen rejection region can improve power for the kinds of departures most relevant to the study.

7.4 Acceptance region

The acceptance region is an older term for the set of test statistic values that do not lead to rejection of the null hypothesis. Many statisticians prefer non-rejection region because failing to reject is not the same as accepting the null as true. The two terms refer to complementary parts of the test statistic space.

7.5 Critical region

The critical region is another name for the rejection region. In many texts, the two expressions are used interchangeably. Both refer to the values of the test statistic that are extreme enough, under the null distribution, to justify rejecting the null hypothesis.

</INTERNAL_LINK_CANDIDATES> Null hypothesis (the default statement tested against observed data) Alternative hypothesis (the competing statement suggesting an effect or difference) Test statistic (the numerical summary used to assess the hypotheses) Significance level (the preselected threshold for Type I error) Type I error (rejecting a true null hypothesis) Critical value (the boundary separating rejection and non-rejection) Sampling distribution (the distribution of a statistic under repeated sampling) Upper-tail rejection region (the high-value tail leading to rejection) Lower-tail rejection region (the low-value tail leading to rejection) Two-tailed rejection region (rejection zones in both tails) Non-rejection region (values not extreme enough to reject the null) Z distribution (standard normal distribution used in some tests) t distribution (distribution used for mean tests with unknown variance) Chi-squared distribution (distribution used in variance and fit tests) F distribution (distribution used in variance ratios and ANOVA) p-value (the probability of an at-least-as-extreme result under the null) Confidence interval (a range of plausible parameter values) Power of a test (the chance of rejecting a false null) Acceptance region (the non-rejection part of a hypothesis test) Analysis of variance (a method comparing variation across groups) </INTERNAL_LINK_CANDIDATES>