What Question Does a One-Sample T Test Answer?
A one-sample t test is designed for a focused question: is the mean of one population different from, greater than, or less than a specified value? For example, a manufacturer may want to assess whether the average battery life for its products differs from a stated target of 10 hours. The population mean is not usually known, so a sample of batteries is measured and its results are used to assess the question.
The test is about the population average, not about whether each sampled battery lasts exactly 10 hours. Individual batteries will vary. The relevant question is whether the observed sample provides evidence about the average battery life in the population from which the sample was drawn.
The word one-sample is important. There is one sample of measurements and one population mean of interest. The reference value is not a second sample mean: it may come from a product specification, a policy target, or a previously stated benchmark. The test assesses a population-level average against that value.
The Population Mean and the Sample Mean
The population mean, \(\mu\), describes the average value of the quantitative variable across the population of interest. It is a fixed quantity, though usually unknown. The sample mean, \(\bar{x}\), is calculated from the observed sample and will generally differ from one sample to another.
For a sample of \(n\) measurements, the sample mean is their sum divided by their count:
A one-sample t test uses the sample information to judge how compatible the observed \(\bar{x}\) is with a proposed value for \(\mu\), while accounting for sample size and observed variability. A sample mean that is not exactly equal to the reference value is not, by itself, enough to establish that the population mean differs from it. The size of the difference must be considered in relation to the uncertainty in the sample.
Why the Sample Mean Provides Information About the Population Mean
If the observations form a random sample from a population with mean \(\mu\), each observation has expected value \(\mu\). The sample mean also has expected value \(\mu\). This explains why repeated samples can provide information about the same population quantity, even though their calculated means vary.
This derivation relies on each sampled observation having population mean \(\mu\). It does not say that every sample mean equals \(\mu\), or that a particular sample gives the exact population mean. It shows that, under the random-sampling model, the sample mean is centered on the population mean over repeated samples.
When Is a One-Sample T Test a Plausible Choice?
Before considering a test, define the population and the measured variable. The variable should be quantitative, and the question should concern its population mean. There should be one sample of observations, not two separate groups whose means are being compared. The reference value should describe the population mean of interest, rather than a desired value for every individual.
A one-sample t test also relies on a sampling model. The observations should come from a suitable random or representative sampling process, and observations should be independent. The distribution of the measurements should be reasonably compatible with the t procedure: for a small sample, the population should be approximately normal, without severe outliers; with a larger sample, modest departures from normality are generally less concerning. The details of checking these conditions matter, so identifying a question as a candidate for this test does not guarantee that using it is appropriate.
The population must also be stated carefully. If a convenience sample contains only volunteers from one clinic, it may not support a conclusion about all people in a region. A test cannot repair a sampling process that does not represent the intended population. Its conclusions are about the population that the sampling design can reasonably represent.
Worked Example: Average Battery Life Compared With a Target
Suppose a quality engineer randomly selects eight batteries from a production population and tests each under the same conditions. Their lifetimes, in hours, are 10.1, 9.8, 10.4, 9.6, 10.0, 9.7, 10.3, and 9.9. The question is whether the population mean battery life differs from the 10-hour target.
Identify the quantity. Battery life is quantitative, and the question is about the population mean lifetime, \(\mu\), for batteries made by this process. The sample mean is calculated from the eight tested batteries; it is not itself the population mean.
Calculate the sample mean. The measurements sum to \(79.8\) hours, so:
The observed sample average is \(0.025\) hours below the target. That describes these eight batteries; it does not establish that the population mean is below 10 hours. A one-sample t test is a plausible type of procedure for the question because there is one sample of quantitative measurements and one reference value for the population mean. Its appropriateness would also require a credible random, independent sampling process and a distribution suitable for the sample size. With only eight observations, it is particularly important that the population not be severely nonnormal or contain extreme outliers.
Questions That Sound Similar but Ask Something Else
The word “average” can hide important differences in the question. A one-sample t test is not a general-purpose test for any numerical-looking data or any comparison with a target. The table summarizes the key distinctions.
| Question | Quantity of interest | Is a one-sample t test the natural match? |
|---|---|---|
| Is average battery life different from 10 hours? | One population mean for a quantitative variable | Potentially, if the sampling and distribution conditions are reasonable |
| Do at least 90% of batteries last 10 hours or more? | A population proportion meeting a criterion | No; the question concerns a proportion, not a mean |
| Is every battery expected to last at least 10 hours? | Individual performance or a guarantee about all units | No; a test of the average does not answer a claim about every unit |
| Do batteries from Brand A have a different mean lifetime from Brand B? | A difference between two population means | No; that is not a one-sample question |
The distinction between a mean and a proportion is especially useful. If each battery is classified as either “lasted at least 10 hours” or “did not,” the data are categorical for that question. The quantity of interest is the fraction of batteries in the first category, not their average lifetime. Conversely, if the actual lifetime in hours is recorded and the target concerns its population average, the question is about a mean.
Worked Example: Average Waiting Time for a Service
A service manager wants to know whether the average wait for customers at one location differs from a 15-minute target. A random sample of 40 customers is selected from the population of customers using that location during the period of interest. The measured waiting times sum to 684 minutes, and the sample standard deviation is 4.8 minutes. Assume the observations are independent and that the sample distribution has no severe skewness or extreme outliers.
Find the observed average. The sample mean is:
The sample average is \(2.1\) minutes above the target. The population mean, \(\mu\), is the average waiting time for the defined customer population at this location; it is not known from the sample alone. Because the question concerns one population mean measured in minutes, compares it with a specified value, and the stated sampling and distribution conditions are plausible, a one-sample t test is a suitable candidate. The mean of \(17.1\) minutes does not alone settle whether the population average differs from 15 minutes: sample size and variability also affect how much evidence the data provide.
Worked Example: A Compliance Percentage Is Not a Mean Question
A hypothetical inspection samples 100 devices and records whether each one passes a safety check. Of the 100 devices, 92 pass. The question is whether at least 90% of devices in the production population pass.
Summarize the sample. The observed passing proportion is:
Thus, 92% of the sampled devices passed. But the question concerns the population proportion that passes, not the population mean of a quantitative measurement. A one-sample t test is not the right match. It would not become a mean question simply because the pass/fail values were coded as 1 and 0: the investigative question asks for a proportion meeting a criterion. The sample percentage is also not proof that the population percentage is exactly 92%; it is a summary of these 100 observations.
What the Test Can and Cannot Establish
A one-sample t test can help assess whether sample data are consistent with a stated value for one population mean, or provide evidence against that value. It does not show that the sample mean and population mean are identical, prove that a target has been met, or guarantee that a specified proportion of individuals meet a standard. Even a result suggesting a difference concerns the average, not necessarily every individual in the population.
The test also cannot establish that a difference matters in practice. For example, an average battery life that differs slightly from a target may or may not have important consequences. Statistical evidence about a mean and the practical importance of the difference are related but distinct questions.
Common Mistake
Wrong reasoning: “The sample mean is 9.975 hours, which is not exactly 10 hours, so the population mean must differ from the target.”
Correction: A sample mean varies from sample to sample, even when the population mean equals the target. A one-sample t test considers the observed difference in the context of sample variability and sample size. The calculated \(\bar{x}\) is a sample summary; it is not a direct observation of \(\mu\). A conclusion about the population mean should follow from the inferential analysis, not from exact equality or inequality of the sample mean and target.
Check Your Understanding
For each question, identify the population quantity of interest and decide whether a one-sample t test is a plausible match.
- A random sample of phones is tested to assess whether mean battery life differs from a stated number of hours. What is the population mean, and what is the role of the sample mean?
- A researcher records the number of minutes each sampled patient waits and asks whether the population average differs from a target. What type of variable and population quantity are involved?
- A factory records whether each product passes inspection and asks whether 95% or more pass. Why is this not a question about one population mean?
- A sample has a mean that exactly matches its reference value. Does that fact alone establish that the population mean equals the reference value? Explain.
- A study compares average scores from two separate populations. What feature makes this different from a one-sample t-test question?