Null and Alternative Hypotheses for a Mean

Why Hypotheses Come Before a Test

A test about a mean begins with a precise question about a population. For example, a clinic may want to know whether its patients’ average waiting time differs from a stated target. The target is a benchmark, and the question is not whether one particular patient waited longer or shorter than that target. It is whether the population mean is equal to, below, above, or different from the benchmark.

Writing the hypotheses carefully fixes what claim is being assessed before looking for evidence in the sample. The direction and wording matter: “below the target” is a different question from “different from the target.” This tutorial focuses on translating that question into hypotheses for a population mean, not on calculating a test statistic.

Definition: Let \(\mu\) denote the population mean of the quantitative variable of interest. Let \(\mu_0\) denote the stated benchmark value. The null hypothesis, written \(H_0\), specifies the benchmark claim being assessed; the alternative hypothesis, written \(H_a\), states the departure from that claim that the research question seeks evidence for.

The benchmark must have the same units as the mean. If the variable is waiting time in minutes, both \(\mu\) and \(\mu_0\) are in minutes. The parameter \(\mu\) refers to the population mean, not the observed sample mean \(\bar{x}\). A sample provides information about \(\mu\), but the hypotheses themselves are statements about \(\mu\).

The Three Common Alternative Forms

The wording of the research question determines which relationship to the benchmark is represented by \(H_a\). The three standard forms are a lower-tail alternative, an upper-tail alternative, and a two-sided alternative.

Research claim about the population mean	Alternative hypothesis	Common displayed null hypothesis
The mean is below the benchmark	\(H_a: \mu < \mu_0\)	\(H_0: \mu = \mu_0\)
The mean is above the benchmark	\(H_a: \mu > \mu_0\)	\(H_0: \mu = \mu_0\)
The mean differs from the benchmark in either direction	\(H_a: \mu \ne \mu_0\)	\(H_0: \mu = \mu_0\)

The null hypothesis is commonly displayed as an equality at the benchmark, including when the alternative is one-sided. There is a useful formal nuance: the full null region for a lower-tail alternative is \(\mu \geq \mu_0\), and the full null region for an upper-tail alternative is \(\mu \leq \mu_0\). In many introductory test statements, the null is written at its boundary, \(H_0: \mu=\mu_0\), because that is the benchmark value used to specify the test. These conventions express the same testing question when used appropriately; they should not be confused with claiming that every value in the composite null is exactly equal to the benchmark.

Key distinction: For a two-sided question, the benchmark equality is the null and all unequal values are in the alternative. For a one-sided question, values on the direction opposite the alternative also belong to the formal null region: \(H_0:\mu\geq\mu_0\) when \(H_a:\mu<\mu_0\), or \(H_0:\mu\leq\mu_0\) when \(H_a:\mu>\mu_0\). The equality form \(H_0:\mu=\mu_0\) is the conventional boundary statement.

Translating Words Into Symbols

A reliable way to write hypotheses is to identify the quantity, the benchmark, and the exact comparison in the question. First define what population is under discussion and what variable is being averaged. Next write the numerical benchmark with its units. Finally, translate the claim: “less than” becomes \(<\), “greater than” becomes \(>\), and “different from” becomes \(\ne\). The null is stated at the benchmark.

Name the population mean.
State what \(\mu\) averages and for which population, such as the mean waiting time for all patients served by a clinic.

Record the benchmark.
Write the stated value as \(\mu_0\), keeping the relevant units clear.

Translate the research claim.
Use \(<\), \(>\), or \(\ne\) in \(H_a\), according to whether the claim is lower, higher, or different in either direction.

Write the null.
State \(H_0:\mu=\mu_0\) as the conventional benchmark statement, and recognize the appropriate composite null for a one-sided alternative.

Worked Example: A Mean Waiting Time Below a Target

A clinic sets a target mean waiting time of 30 minutes. A hypothetical sample of patients has a mean waiting time of 27.8 minutes. The clinic wants to assess whether the population mean waiting time for its patients is below the target. Write the hypotheses.

Let \(\mu\) be the mean waiting time, in minutes, for the clinic’s patient population. The benchmark is \(\mu_0=30\) minutes. The sample mean of 27.8 minutes is a sample statistic; it does not replace the population parameter in the hypotheses.

The claim is “below,” so the alternative is:

H_a: \mu < 30\text{ minutes}

The conventional boundary null is:

H_0: \mu = 30\text{ minutes}

For a formal one-sided statement, the null region is \(\mu\geq30\) minutes. The observed sample mean does not change the direction of the hypotheses; the direction comes from the clinic’s stated question.

Worked Examples in Different Contexts

Worked Example: Testing Whether a Mean Exceeds a Standard

A hypothetical packaging line is designed to fill bottles to a mean of 500 milliliters. A quality manager asks whether the population mean fill is greater than 500 milliliters. A sample of bottles has a mean fill of 503 milliliters. Write the hypotheses.

Define \(\mu\) as the mean fill, in milliliters, for bottles produced by the line under the conditions of interest. The benchmark is \(\mu_0=500\) milliliters. The question says “greater than,” so:

H_a: \mu > 500\text{ mL}

The conventional boundary null is:

H_0: \mu = 500\text{ mL}

The corresponding formal null region is \(\mu\leq500\) milliliters. The sample mean, 503 milliliters, is not the value to put in the hypotheses: it is an observed summary, while the hypotheses concern the population mean.

Worked Example: Testing Whether a Mean Differs From a Benchmark

A hypothetical laboratory uses a reference value of 7.0 pH units for a process average. Its researchers ask whether the process population mean differs from 7.0, without specifying a direction. A sample has mean pH 6.8. Write the hypotheses.

Let \(\mu\) be the population mean pH for measurements from the process. The benchmark is \(\mu_0=7.0\) pH units. “Differs from” allows either a value below or a value above the reference, so the alternative is two-sided:

H_a: \mu \ne 7.0\text{ pH units}

The null states equality to the reference:

H_0: \mu = 7.0\text{ pH units}

The sample mean of 6.8 pH units is below the benchmark, but it does not turn the research question into a lower-tail question. The stated question allows evidence of a mean either below or above 7.0.

Why the Alternative Determines the Question

The symbols divide possible values of \(\mu\) according to the claim under consideration. For example, a lower-tail alternative \(H_a:\mu<\mu_0\) describes values strictly below the benchmark. Its formal null region contains the benchmark and every value above it. These regions do not overlap, and together they cover all possible values of \(\mu\):

\{\mu:\mu<\mu_0\}\ \cup\ \{\mu:\mu\geq\mu_0\} =\{\text{all possible values of }\mu\}.

The same partition applies in the opposite direction for an upper-tail alternative. For a two-sided alternative, every value unequal to the benchmark is in \(H_a\), while equality alone is in \(H_0\). This is why a hypothesis pair must be logically compatible: a population mean cannot simultaneously equal the benchmark and be below it, and every possible mean must fall in one of the stated regions.

The distinction also explains why the alternative must reflect the question rather than the sample’s direction. If a question asks whether a mean differs from a benchmark, a sample mean on one side does not justify replacing \(\ne\) with \(<\) or \(>\). Doing so would change the claim after observing the data.

Common Mistakes

Putting the sample mean in the hypothesis. Writing \(H_0:\mu=27.8\) because the observed \(\bar{x}\) is 27.8 confuses a sample statistic with the benchmark. Use the stated benchmark in hypotheses about \(\mu\).
Reversing the direction. If the research claim is that the mean is below the benchmark, \(H_a\) must use \(<\), not \(>\). Read the wording and translate it directly.
Using equality in the alternative. A claim of “different from” is represented by \(\mu\ne\mu_0\), not by \(\mu=\mu_0\). Equality is the conventional null statement.
Choosing the direction after seeing the sample. A sample mean above the benchmark does not by itself make the alternative upper-tailed. The question should define the claim independently of the observed result.
Interpreting a mean claim as a claim about every individual. \(H_a:\mu<\mu_0\) concerns the population average. It does not say that every person or measurement is below the benchmark.

Key takeaway: Define \(\mu\) in context, identify the benchmark \(\mu_0\), and let the research wording determine \(H_a\): \(<\) for below, \(>\) for above, and \(\ne\) for a difference in either direction. State the null at the benchmark and keep population parameters separate from sample statistics.

Check Your Understanding

For each question, identify the population mean, the benchmark, and the appropriate hypotheses.

A hypothetical school asks whether its students’ mean commute time is less than a 25-minute benchmark. What are \(H_0\) and \(H_a\), and what is the formal one-sided null region?
A device is expected to have a mean operating life of 1,200 hours. Researchers ask whether the population mean is greater than 1,200 hours. Write the hypotheses.
A hypothetical study asks whether a population mean differs from 4.5 units in either direction. Which symbol belongs in \(H_a\), and what is \(H_0\)?
A sample mean is below its benchmark, but the research question asks whether the population mean differs from that benchmark. Should the alternative be changed to a lower-tail alternative? Explain.
Why is a statement about \(H_a:\mu<\mu_0\) not the same as saying every individual value in the population is less than \(\mu_0\)?