Choose a One-Mean Procedure

Introduction

A health report might ask whether the average resting pulse in a population differs from a clinical reference value. A campus survey might ask whether students sleep an average of seven hours per night. In both cases, the data come from one population, the measurement is quantitative, and the question compares that population’s mean with a specified number.

Recognizing that structure is the first step in choosing a procedure. The key is not simply that the data are numbers: it is that the target is one population mean, rather than a proportion or a difference between populations. Once that target is clear, the main choice is whether the population standard deviation is known.

Recognizing a One-Mean Question

For example, if a random sample of campus residents records hours of sleep, the population might be all campus residents, the quantitative measurement is sleep duration in hours, and the parameter is their population mean sleep duration, \(\mu\). If the question compares that mean with seven hours, then \(\mu_0=7\) hours is the reference value.

The reference value is a benchmark, not a second sample. Comparing a sample mean with a fixed target does not turn the question into a comparison of two groups. By contrast, if data are collected from two separate populations and the question concerns the difference between their means, that is a two-group question. If the outcome is a Yes/No category summarized by a proportion, it is not a mean question.

A useful first pass is therefore to name the measurement, population, and parameter in words. Ask: “What quantitative measurement is recorded for each individual? Which one population does the question concern? Is its mean being compared with a fixed number?” If these answers fit, one-mean inference is a candidate. The design and assumptions must still be appropriate.

Choose Between a One-Sample t and z Procedure

In most applications, the population standard deviation \(\sigma\) is not known. It must be estimated using the sample standard deviation \(s\), so the usual procedure is a one-sample t procedure. A one-sample z procedure is appropriate for a mean when \(\sigma\) is genuinely known. Knowing the reference value \(\mu_0\) does not mean that \(\sigma\) is known: they are different quantities.

For a one-sample t test, the test statistic is

Here, \(n\) is the sample size, \(s/\sqrt{n}\) is the estimated standard error of the sample mean, and \(df\) is the degrees of freedom used with the t distribution. For a one-sample z test with known \(\sigma\), replace the estimated standard error by \(\sigma/\sqrt{n}\):

These statistics express the difference between the observed sample mean and the reference mean in standard-error units. A value near zero means the sample mean is close to the reference relative to its estimated sampling variability. A larger absolute value means the observed difference is farther from the reference in those units. For a test, the p-value is calculated from the relevant t or standard normal distribution using the direction specified by the research question.

Why the t Statistic Has This Form

The standard deviation of the sampling distribution of \(\bar{x}\), when observations are independent and have population standard deviation \(\sigma\), is \(\sigma/\sqrt{n}\). This is the standard error when \(\sigma\) is known. If it is unknown, replacing it with \(s\) gives the estimated standard error \(s/\sqrt{n}\). The observed distance from the benchmark is \(\bar{x}-\mu_0\); dividing by its standard error puts that distance on a common scale:

When observations come from a normal population, this standardized quantity follows a t distribution with \(n-1\) degrees of freedom under the reference-mean model. The t distribution accounts for the extra uncertainty from estimating \(\sigma\) with \(s\). If \(\sigma\) is known instead, standardizing by \(\sigma/\sqrt{n}\) gives the z statistic. In either case, the formulas rely on a sampling design and distributional conditions that make the stated sampling model reasonable.

Worked Examples

A Practical Selection Checklist

Procedure selection becomes more reliable when the question, design, and known information are considered in order. Use this short sequence before calculating a statistic:

The conditions depend on how the data were collected and on the distribution of the measurement. A random or otherwise defensible sampling design supports generalization to the target population; dependence between observations can undermine the standard error calculation. With a small sample, the population distribution should be reasonably close to normal and should not have severe outliers. Larger samples can make inference more robust to moderate nonnormality, but they do not repair biased sampling or dependence.

Common Mistakes and Why the Choice Matters

• Treating the reference value as a second sample. Comparing one sample mean with a stated clinical or policy benchmark is still one-mean inference. A second group is present only if a second population was actually sampled.
• Using a proportion procedure for a quantitative measurement. The question about average sleep in hours is about a mean. A question about the percentage of residents who sleep at least seven hours is about a proportion.
• Confusing \(\mu_0\) with \(\sigma\). The benchmark is the hypothesized or comparison value for the mean; the population standard deviation describes variation among individual measurements. Knowing one does not imply knowing the other.
• Choosing z because the sample is large. A large sample does not make an unknown population standard deviation known. When \(\sigma\) is unknown, the standard one-mean choice remains the t procedure.
• Assuming a nonsignificant result proves equality. A large p-value means the data do not provide strong evidence against the reference-mean model under the specified test. It does not establish that the population mean equals the reference exactly.

Choosing the right procedure keeps the parameter, sampling design, and uncertainty calculation aligned. It also prevents an easy but consequential error: answering a question about an average with a method intended for percentages or group differences. Always state the population mean and reference value in context before reporting the calculation.