Population Quantities and Sample Estimates

From a Research Question to a Quantity and an Estimate

A regression question can concern a whole population, even when the available data come from only a sample. For example, a health department may want to know the average waiting time for all patients visiting a set of clinics during a month, or how waiting time is associated with appointment time. The question concerns a population quantity; the collected records provide a way to estimate it.

The distinction matters because a calculated value from observed data is not automatically the value for the population. A sample average may be close to the population average, or it may differ. Likewise, a sample summary of an association describes the observed cases directly; using it to learn about a wider population requires the data and study design to support that generalization.

Definition: A population quantity (also called a population parameter) is a numerical summary defined for a specified population and a specified measurement or relationship. A sample statistic is a numerical summary calculated from the observed sample. An estimator is a rule or statistic used to estimate a population quantity; the value it produces for the collected sample is the estimate.

A parameter and its estimate answer related but different questions. The parameter describes the target population. The estimate is calculated from the particular collected data. Before data are collected, the sample statistic that will be used as an estimator can vary from one possible sample to another. Once a sample has been observed, its estimate is a particular number.

To define a population quantity precisely, specify the population, the outcome or variables, and the summary. “The average waiting time” is incomplete unless it is clear which patients, which time period, and how waiting time is measured. This is consistent with the earlier material on target populations and samples: a sample estimate can speak to the intended population only when the target and data collection align.

Population Means and Sample Means

For a quantitative outcome, a common population quantity is the population mean. If the outcome is denoted by \(Y\), its population mean is written \(\mu\). In a sample of \(n\) observed outcomes, the sample mean \(\bar{x}\) is a familiar estimate of that population mean. The notation \(\bar{x}\) represents the average of the observed outcome values in this sample.

\mu = \text{mean outcome in the specified population} \qquad\text{and}\qquad \bar{x} = \frac{x_1+x_2+\cdots+x_n}{n}

The formula for \(\bar{x}\) is arithmetic on the observed sample values; it does not require that the sample mean equal \(\mu\). If all members of the target population have been measured, then the population mean can be calculated directly. More often, only a sample is observed, and the population mean remains unknown.

Worked Example: Estimating a Population Mean

Suppose, hypothetically, that a district has five classrooms and that the number of students absent on a particular day is 18, 20, 22, 24, and 26. The target population is these five classrooms on that day. A sample contains three classrooms with 18, 24, and 26 absences. Find the population mean and the sample estimate.

Population quantity. Because all five classrooms in this hypothetical target population are specified, the population mean can be calculated:

\mu = \frac{18+20+22+24+26}{5} = \frac{110}{5} = 22 \text{ absent students per classroom}

Sample estimate. The sample mean from the three observed classrooms is:

\bar{x} = \frac{18+24+26}{3} = \frac{68}{3} \approx 22.67 \text{ absent students per classroom}

The sample estimate is about 22.67 absences per classroom, while the population mean is 22. The difference is not an arithmetic error: the two numbers summarize different sets of classrooms. In an actual study, the population quantity is commonly unknown, so the sample mean would be reported as an estimate rather than compared with a known population value.

Why the Sample Mean Can Estimate the Population Mean

A useful justification comes from a simple random sample without replacement from a finite population. In this sampling design, every member of the population has the same chance of selection, and every possible sample of the specified size is equally likely. Under these conditions, the sample mean is centered on the population mean across repeated samples. This property is called unbiasedness: the expected value of the estimator equals the quantity it estimates.

To see why, suppose the finite population contains \(N\) values \(Y_1,\ldots,Y_N\), with population mean \(\mu=(Y_1+\cdots+Y_N)/N\). Let \(I_i\) equal 1 if population member \(i\) is selected and 0 otherwise. In a simple random sample of size \(n\), each member has selection probability \(n/N\), so \(E(I_i)=n/N\). The sample mean can be written using these indicators:

\bar{Y} = \frac{1}{n}\sum_{i=1}^{N} I_iY_i

Taking the expected value and using the linearity of expectation gives:

E(\bar{Y}) = \frac{1}{n}\sum_{i=1}^{N} E(I_i)Y_i = \frac{1}{n}\sum_{i=1}^{N} \frac{n}{N}Y_i = \frac{1}{N}\sum_{i=1}^{N}Y_i = \mu

This result concerns the average behavior of the estimator over all possible samples under the stated sampling design. It does not say that the estimate from any one sample must equal \(\mu\), nor does it establish that a particular real-world sample was selected randomly or measured well. An unbiased estimator can still be off target in a particular sample.

Group Differences: The Target and Its Estimate

A research question may compare the mean outcome in two population groups. If group A and group B are the specified populations, a population difference can be defined as the mean in A minus the mean in B. Its sign depends on this stated order: a negative value means the mean for A is lower than the mean for B. The sample counterpart subtracts the sample mean for B from the sample mean for A.

\Delta = \mu_A-\mu_B \qquad\text{and}\qquad \widehat{\Delta} = \bar{x}_A-\bar{x}_B

Here, \(\Delta\) is the population quantity and \(\widehat{\Delta}\) is its sample estimate. Both are measured in the units of the outcome. The sample groups must correspond to the population groups named in the question; changing the population definition changes the target quantity.

Worked Example: Estimating a Difference Between Group Means

Suppose, hypothetically, that a program is interested in the number of practice problems completed by learners in two defined groups. The population values are 2, 4, and 6 problems for group A, and 5, 7, and 9 problems for group B. A sample includes values 2 and 4 from A, and 7 and 9 from B. Calculate the population difference and the sample estimate, using A minus B.

Population quantity. The population means are:

\mu_A = \frac{2+4+6}{3}=4, \qquad \mu_B = \frac{5+7+9}{3}=7

Therefore, the population difference is:

\Delta = \mu_A-\mu_B = 4-7=-3 \text{ problems}

Sample estimate. The sample means are \(\bar{x}_A=(2+4)/2=3\) and \(\bar{x}_B=(7+9)/2=8\). Thus:

\widehat{\Delta} = \bar{x}_A-\bar{x}_B = 3-8=-5 \text{ problems}

In this hypothetical example, the sample estimate indicates that group A’s mean is 5 problems lower than group B’s mean, while the population difference is 3 problems lower. The calculation describes the sample exactly; whether it estimates the defined population difference well depends on how the sample relates to the two populations.

Predictor-Outcome Association: Population Summary and Sample Summary

A regression question can also concern how a quantitative predictor and a quantitative outcome vary together. One possible summary of their linear association is the population correlation, written \(\rho\). It is the population covariance divided by the product of the population standard deviations, provided both standard deviations are positive. The sample correlation, written \(r\), is calculated from the paired predictor and outcome values in the sample and can be used as an estimate of \(\rho\).

\rho = \frac{\text{population covariance of }X\text{ and }Y} {\sigma_X\sigma_Y} \qquad\text{and}\qquad r = \frac{\sum_{i=1}^{n}(x_i-\bar{x})(y_i-\bar{y})} {\sqrt{\sum_{i=1}^{n}(x_i-\bar{x})^2 \sum_{i=1}^{n}(y_i-\bar{y})^2}}

Both summaries are unitless and range from \(-1\) to \(1\) when defined. The population correlation is a quantity for the specified population; \(r\) is the value computed from the observed pairs. Correlation summarizes linear association, not causation. In particular, naming one variable a predictor and the other an outcome does not by itself establish that changes in the predictor cause changes in the outcome.

Worked Example: A Sample Association Estimate

Imagine, hypothetically, that four learners each complete a different number of practice sessions, \(x\), and receive an assessment score, \(y\). Their observed pairs are \((1,2)\), \((2,3)\), \((3,5)\), and \((4,6)\). Calculate the sample correlation \(r\), which summarizes the linear association in these four observed pairs.

First, the sample means are \(\bar{x}=(1+2+3+4)/4=2.5\) sessions and \(\bar{y}=(2+3+5+6)/4=4\) score points. The sum of paired products of deviations is:

\sum_{i=1}^{4}(x_i-\bar{x})(y_i-\bar{y}) = (-1.5)(-2)+(-0.5)(-1)+(0.5)(1)+(1.5)(2) = 7

The two sums of squared deviations are:

\sum_{i=1}^{4}(x_i-\bar{x})^2 =2.25+0.25+0.25+2.25=5

\sum_{i=1}^{4}(y_i-\bar{y})^2 =4+1+1+4=10

Substituting into the sample correlation formula gives:

r = \frac{7}{\sqrt{5(10)}} = \frac{7}{\sqrt{50}} \approx 0.9899

The observed sample pairs have a strong positive linear association: learners with more practice sessions in these four pairs tend to have higher scores. This value is the sample statistic \(r\), not automatically the population correlation \(\rho\). The example does not establish a causal effect of practice, and the four pairs alone do not guarantee that the estimate represents a broader population.

Common Mistakes and Why the Distinction Matters

Population quantities and sample estimates are easily blurred when both are described in the same sentence. Keep the target and the data-based result explicit: name the population parameter, name the statistic calculated from the sample, and use the latter as an estimate only when the target and sample make sense together.

Mistake: Treating an estimate as the population truth. A sample mean of 22.67 does not prove that the population mean is 22.67. It is the calculated sample mean and may differ from the population mean.
Mistake: Saying the population quantity is “random” because the sample varies. For a fixed, specified population and measurement, its parameter is fixed. The estimator can vary across possible samples; the realized estimate is the value obtained from the sample actually observed.
Mistake: Forgetting the order in a group difference. A minus B and B minus A have opposite signs. State the subtraction order before interpreting the result.
Mistake: Assuming a sample statistic is automatically a good estimate. Correct arithmetic does not guarantee a suitable sample, valid measurement, or relevant target population. Those features affect whether the estimate can support the intended generalization.
Mistake: Interpreting association as cause. A sample correlation describes how two variables vary together in the observed data. It does not alone show that the predictor caused the outcome.

This distinction is central because the research question usually concerns a population, while the analysis begins with collected data. Precise language makes clear what is known directly from the sample and what is being estimated about the population. It also prevents a numerical result from being given a broader interpretation than the study design supports.

Check Your Understanding

For each item, identify the population quantity, the sample statistic that could estimate it, or the distinction between them.

A school district wants the mean number of absences per student in all its middle schools this year. What is the population quantity, and what could be its sample estimate?
For two populations, group A and group B, define a mean difference using A minus B. If the estimated difference is positive, which group has the larger sample mean?
Explain the difference between an estimator and the estimate obtained after collecting data.
In one sentence, distinguish the population correlation \(\rho\) from the sample correlation \(r\).
Why does an unbiased sample mean not have to equal the population mean in every particular sample?