Sampling Error Versus Individual Variability

Two Different Questions About Variation

Suppose repeated samples are drawn from a population of patients. Their individual treatment times may differ considerably. At the same time, the average treatment time from one sample may be close to the average from another. These are two distinct kinds of variation: variation among individuals and variation among sample statistics.

The distinction matters because a data set describes the individuals actually observed, while a statistic such as a sample mean or sample proportion is used to summarize a population. The spread of individual observations tells us how different those observations are from one another. The spread of a statistic across repeated random samples tells us how much that statistic would typically change if we repeated the sampling process.

Definition: Individual variability is the spread of values among individual units in a population or sample. Sampling error is the difference between a statistic from one sample and its population target. Standard error is the standard deviation of a statistic’s sampling distribution: the typical scale of its variation across repeated samples taken under the same sampling process.

Sampling error is not necessarily a mistake in data collection. It is the discrepancy that can arise because a sample includes only some members of a population. A particular sample may produce a statistic above or below its target. Standard error is not that particular discrepancy; it describes the typical spread of such discrepancies over repeated samples.

For a Sample Mean

Let $X$ denote an individual quantitative observation, with population mean $\mu$ and population standard deviation $\sigma$. For a random sample of size $n$, the sample mean is $\bar{x}$. Its sampling error is $\bar{x}-\mu$, measured in the same units as the observations.

Result: If the $n$ observations are independent draws from a population with finite standard deviation $\sigma$, then the standard error of the sample mean is $$ SE(\bar{x})=\frac{\sigma}{\sqrt{n}}. $$ The standard deviation $\sigma$ describes variation among individual observations. The standard error describes variation among sample means. If $\sigma$ is unknown, $s/\sqrt{n}$ is commonly used as an estimate of the standard error, where $s$ is the sample standard deviation.

This formula assumes independent observations from the same population. For sampling without replacement from a finite population, the dependence between draws can change the standard error; the finite population correction applies when appropriate. Here, examples using the formula assume independent sampling.

Worked Example: Individual Spread and Mean Spread

Imagine a hypothetical population of appointment durations with mean $\mu=50$ minutes and standard deviation $\sigma=10$ minutes. A clinic independently samples $n=25$ appointments. Compare the spread among individual durations with the standard error of the sample mean.

The individual standard deviation is $\sigma=10$ minutes. It describes how far individual appointment durations tend to vary around the population mean. Under the stated independent-sampling assumption, the standard error of the sample mean is $$ SE(\bar{x})=\frac{\sigma}{\sqrt{n}} =\frac{10}{\sqrt{25}} =\frac{10}{5} =2\text{ minutes}. $$

The value 2 minutes describes the typical sample-to-sample spread of averages, not the spread of individual appointment durations. The average is less variable than an individual observation because each sample mean combines 25 observations.

Suppose one such sample has $\bar{x}=53$ minutes. Its sampling error is $$ \bar{x}-\mu=53-50=3\text{ minutes}. $$ This particular sample mean is 3 minutes above the population mean. The standard error remains 2 minutes: it describes the sampling process, not the error in this one sample. The sampling error can be larger or smaller than one standard error.

Why Averaging Changes the Spread

For independent observations $X_1,\ldots,X_n$, the sample mean is the average of the observations. Subtracting the population mean from that average gives the average of the individual deviations from the population mean:

\bar{X}-\mu = \frac{1}{n}\sum_{i=1}^{n}(X_i-\mu).

This identity explains the difference in focus. Individual variability concerns the spread of the terms $X_i$; sampling error concerns their average. Deviations above and below $\mu$ can partly offset one another when averaged. More precisely, independence makes the variance of the sum equal to the sum of the variances, so dividing the sum by $n$ gives

\operatorname{Var}(\bar{X}) = \frac{1}{n^2}\sum_{i=1}^{n}\operatorname{Var}(X_i) = \frac{n\sigma^2}{n^2} = \frac{\sigma^2}{n}.

Taking the square root yields $SE(\bar{x})=\sigma/\sqrt{n}$. This justification depends on the observations having the same variance $\sigma^2$ and being independent. It also shows why increasing $n$ reduces the standard error: averaging more independent observations reduces the variance of the average. It does not mean that the individual observations themselves become less variable.

For a Sample Proportion

A proportion also summarizes observations, but each individual outcome is binary: it either has the characteristic of interest or it does not. Represent an individual outcome with an indicator that equals 1 if the characteristic is present and 0 otherwise. If the population proportion is $p$, an individual indicator has standard deviation $\sqrt{p(1-p)}$. The sample proportion $\hat{p}$ is the mean of $n$ such indicators.

Result: For independent binary observations with population proportion $p$, the standard error of the sample proportion is $$ SE(\hat{p})=\sqrt{\frac{p(1-p)}{n}}. $$ The individual standard deviation $\sqrt{p(1-p)}$ describes the spread of the 0/1 outcomes; the standard error describes the spread of sample proportions across repeated samples.

Both quantities are measured on the 0-to-1 scale, but they answer different questions. An individual indicator can only be 0 or 1, whereas $\hat{p}$ averages many indicators and can take values between 0 and 1. The formula assumes independent observations from the same population. When $p$ is unknown, it may be estimated with the observed $\hat{p}$ for certain standard-error calculations; the example below uses a specified population value to make the comparison direct.

Worked Example: Binary Outcomes and a Sample Proportion

In a hypothetical population, 40% of patients meet a specified follow-up goal, so $p=0.40$. An independent sample of $n=100$ patients includes 47 who meet the goal. Compare individual variability, standard error, and the observed sampling error.

For an individual binary outcome, the standard deviation is $$ \sqrt{p(1-p)} = \sqrt{0.40(0.60)} = \sqrt{0.24} \approx 0.4899. $$ This describes variation among individual 0/1 outcomes, not variation in the proportion across samples.

The sample proportion is $\hat{p}=47/100=0.47$. Given the independent sampling assumption, its standard error is $$ SE(\hat{p}) = \sqrt{\frac{p(1-p)}{n}} = \sqrt{\frac{0.40(0.60)}{100}} = \sqrt{0.0024} \approx 0.0490. $$ Across repeated samples of 100 patients, sample proportions would have a typical spread of about 0.049, or 4.90 percentage points.

The observed sampling error is $$ \hat{p}-p=0.47-0.40=0.07. $$ Thus, this sample’s proportion is 7 percentage points above the population proportion. The value 0.07 is the realized error for this sample; 0.049 is the standard error describing the typical scale of errors across repeated samples.

Estimating Mean Spread from One Sample

In practice, the population standard deviation $\sigma$ is often unknown. The sample standard deviation $s$ describes the spread of the observed individuals around their sample mean. It can be used to estimate the population standard deviation and, in turn, the standard error of the mean. The two numbers should not be confused: $s$ measures individual-level spread, while $s/\sqrt{n}$ estimates sample-to-sample spread in the mean.

Worked Example: A Sample Standard Deviation Is Not the Standard Error

Suppose a hypothetical independent random sample of $n=16$ recovery times has sample mean $\bar{x}=34$ hours and sample standard deviation $s=12$ hours. Estimate the standard error of the sample mean.

The observed individual recovery times have a sample standard deviation of 12 hours. Using $s$ to estimate $\sigma$, the estimated standard error is $$ \widehat{SE}(\bar{x}) = \frac{s}{\sqrt{n}} = \frac{12}{\sqrt{16}} = \frac{12}{4} = 3\text{ hours}. $$

The 12-hour value describes the spread of individual recovery times in this sample. The 3-hour value estimates the typical spread of sample means from repeated independent samples of 16 patients. The sample mean of 34 hours is one observed average; the standard error does not say that each patient’s recovery time is usually within 3 hours of 34.

Common Mistakes and Why the Distinction Matters

When a study reports a large standard deviation, that does not by itself imply that its sample mean has a large standard error. The standard error also depends on the sample size. Conversely, a small standard error does not mean that individual outcomes are similar; it can arise because a large number of highly variable observations are averaged.

Confusing standard deviation with standard error. A standard deviation describes variation among observations; a standard error describes variation in a statistic across repeated samples.
Calling standard error the error in the observed estimate. The sampling error is the estimate minus its population target. Standard error is the typical scale of that error over repetitions.
Assuming a larger sample removes individual variability. Increasing $n$ reduces standard error under the stated sampling conditions, but does not change the population’s individual-level spread.
Treating sampling error as bias or a data mistake. Random sampling can produce a statistic above or below its target even when data are recorded correctly. A systematic departure in the sampling process is a different issue.

This distinction helps explain why statisticians report both summaries of observed data and measures of uncertainty in estimates. A sample standard deviation describes what was seen among the sampled individuals. A standard error describes how much a sample statistic is expected to vary across repetitions of the sampling process, under its assumptions. Neither quantity alone guarantees that a particular sample statistic equals its population target.

Key takeaway: Individual variability is about how observations differ from one another. Sampling error is the discrepancy between one sample statistic and its population target. Standard error measures the typical sample-to-sample variability of that statistic; it is not the individual standard deviation and not the realized error from one sample.

Check Your Understanding

Use the distinction between individual spread and sample-to-sample spread to answer each question.

A population has standard deviation $\sigma=15$ units. Under independent sampling, what is the standard error of the sample mean for $n=25$, and what does each number describe?
A sample proportion is 0.34, while the population proportion is 0.30. What is the sampling error, and is it the same as the standard error?
For binary outcomes with $p=0.25$ and $n=100$, write the formulas for the individual standard deviation and the standard error of the sample proportion.
A researcher says, “The sample standard deviation is 8, so the sample mean is typically 8 units from the population mean.” Identify the mistake and correct it for a sample of size 16, using $s$ to estimate $\sigma$.
Why can a large sample have a small standard error even when individual observations vary widely?