Population Parameters and Statistical Hypotheses

Why the Population Quantity Matters

A statistical test begins with a claim about a population, but the data available for analysis usually come from a sample. Keeping those two levels separate is essential: a test concerns a population parameter, while the observed data produce a sample statistic. For example, a question might concern the mean wait time for all patients served by a clinic, even though the analyst has recorded wait times for only a sample of patients.

The distinction is more than a difference in notation. A population parameter describes the full group specified by the research question. A sample statistic summarizes the particular observations collected. The statistic can provide information about the parameter, but it is not the parameter, and its observed value is not itself the population claim being evaluated.

Define the Population Before Naming the Parameter

A parameter is meaningful only after the population and the variable have been specified. “The average” is incomplete: average of what variable, among which individuals, and over what time or setting? A population may be all patients seen at one clinic this month, all units produced by a process during a specified shift, or all students enrolled at a college this term. Changing the population can change the parameter, even if the variable name stays the same.

For a quantitative variable, the population mean is denoted by \(\mu\). For a categorical variable with a defined success category, the population proportion in that category is denoted by \(p\). Their sample counterparts are \(\bar{x}\), the sample mean, and \(\hat{p}\), the sample proportion. The notation signals whether a quantity describes the population or the observed sample.

The individuals, variable, and population definition determine the target. If a survey samples students but records the number of hours each student studies per week, the parameter might be the mean weekly study time among the population of interest. It is not automatically the mean for every student everywhere: the population must be explicitly bounded by the question.

Population Hypotheses and Sample Evidence

A statistical hypothesis is a statement about a population parameter or parameters. In a test, the data are used to assess how compatible an observed sample result is with a specified claim about that population quantity. The observed statistic is part of the evidence; it is not the population parameter being claimed.

For instance, if a manufacturer claims that 10% of units produced by a specified process are defective, the claim concerns the population proportion \(p\). If inspectors check a sample and find that 14% are defective, \(\hat{p}=0.14\) is the observed sample statistic. The data do not change the definition of \(p\), and observing 14% in the sample does not make the population proportion equal to 14%.

This distinction also clarifies why a test is not simply a comparison between a sample number and a target. The sample statistic is expected to vary from one sample to another. A test accounts for this sampling variation when assessing the evidence. The logic of reference distributions and p-values develops later; for now, the important point is to identify the population quantity to which the claim refers before considering how the sample will be analyzed.

Why Sample Statistics Vary

A parameter is fixed for a defined population, even when its value is unknown. A statistic depends on which individuals happen to be sampled, so it can take different values across samples. Before a sample is selected, the statistic can be viewed as a random quantity; after the data are observed, it has a particular calculated value.

A short derivation shows how a sample mean is centered on the population mean under a standard sampling model. Suppose \(X_1,\ldots,X_n\) are independent observations from the same population, each with finite mean \(\mu\), and define their sample mean as \(\bar{X}=(X_1+\cdots+X_n)/n\). Linearity of expectation gives:

This result says that, over repeated samples under the stated model, the sample mean has expected value \(\mu\). It does not say that a particular sample mean must equal \(\mu\), nor does it prove that any particular sample was representative. Rather, it illustrates why \(\bar{x}\) can serve as a statistic for learning about \(\mu\), while the observed value remains distinct from the parameter.

A Reliable Identification Process

Before interpreting a claim or a sample result, identify the target in a fixed order. This keeps the population question separate from the arithmetic performed on the sample.

In later work, the null and alternative hypotheses will express claims about population parameters. The details of writing those hypotheses are a separate task. At this stage, be prepared to say which population quantity a claim is about and which sample statistic supplies the observed information.

Common Mistakes

• Calling a sample result the parameter. If 17.5% of sampled bottles are underfilled, that is the observed \(\hat{p}\), not necessarily the population proportion \(p\).
• Writing a hypothesis about a statistic. A hypothesis test targets a population quantity. The sample statistic is calculated from the data and used as evidence about that target.
• Leaving the population vague. “Mean waiting time” is not fully specified until the relevant patients, clinics, period, and measurement are clear.
• Confusing units of analysis. A mean across patients and a mean across clinic averages describe different populations and may represent different quantities.
• Assuming the statistic equals the parameter. A statistic may estimate a parameter, but sampling variation means different samples can yield different statistics.