Why Sample Means Vary
A sample mean is an estimate of a population mean, but it is not usually identical from one sample to the next. Imagine repeatedly taking random samples of the same size from one population and calculating a mean for each sample. The means would vary because each sample contains a different selection of individuals. The standard error of the sample mean describes the scale of that sampling variability.
The standard error is not the spread of individual measurements in one sample. That spread is summarized by the sample standard deviation \(s\). Instead, the standard error concerns the sample mean as a statistic: how much sample means typically differ from one another when the sampling process is repeated. Since the population standard deviation \(\sigma\) is generally unknown, the usual estimate of the standard error uses \(s\) and the sample size \(n\).
Calculating and Interpreting the Estimated Standard Error
The calculation has two ingredients: the sample standard deviation, which describes the spread of observations, and the sample size, which determines how much averaging reduces the variability of the mean. Divide \(s\) by the square root of \(n\), not by \(n\). The result is expressed in the original measurement units.
Worked Example: Mean Daily Steps
In a hypothetical random sample of 36 adults, the sample standard deviation of daily step counts is 2,400 steps. Assume the observations are independent and come from the population of adults of interest. Estimate the standard error of the sample mean daily step count.
Here \(s=2{,}400\) steps and \(n=36\). Applying the formula gives
The estimated standard error is 400 steps. In repeated independent samples of 36 adults, sample mean step counts would typically vary on a scale estimated to be about 400 steps, assuming the sampling model is appropriate. This does not say that individual adults’ step counts are usually within 400 steps of the sample mean; individual observations may vary much more.
It is useful to keep the two measures separate. \(s\) describes observed variation among individuals in the sample; \(SE\) estimates variation among sample means from repeated samples of the same size. If a variable is measured in milligrams per litre, both \(s\) and \(SE\) are in milligrams per litre. Neither is a percentage unless the original variable itself is measured as a percentage.
Why the Square Root of Sample Size Appears
The formula follows from how independent random observations combine in an average. Let \(X_1,\ldots,X_n\) represent independent observations from the same population, each with mean \(\mu\) and variance \(\sigma^2\). The sample mean is their sum divided by \(n\). Independence makes the variance of the sum equal to the sum of the variances. Dividing that sum by \(n^2\) gives the variance of the mean:
Taking the square root gives the standard deviation of the sampling distribution of \(\bar{X}\), which is its standard error:
In practice, \(\sigma\) is usually unknown. Replacing it with the observed sample standard deviation \(s\) gives the estimated standard error \(s/\sqrt{n}\). This substitution estimates the sampling variability; it does not make the unknown \(\sigma\) known exactly.
What Changes the Estimated Standard Error?
For a fixed sample size, a larger \(s\) produces a larger estimated standard error. Greater spread among individual observations means that the sample mean is estimated with more sampling variability, all else equal. For a fixed \(s\), a larger \(n\) produces a smaller estimated standard error because the observations are averaged over a larger sample.
The sample-size relationship is proportional to \(1/\sqrt{n}\). Therefore, multiplying the sample size by four divides the estimated standard error by two, provided the sample standard deviation stays the same. Doubling the sample size does not halve the standard error: it multiplies it by \(1/\sqrt{2}\), approximately 0.707.
Worked Example: Comparing Two Sample Sizes
Two hypothetical independent random samples measure the same physiological quantity in the same units. For the first sample, \(n=16\) and \(s=8.4\) units. For the second, \(n=64\) and \(s=8.4\) units. Compare their estimated standard errors.
For the first sample,
For the second sample,
The second sample has four times as many observations, and its estimated standard error is half as large. This comparison holds the sample standard deviation constant to isolate the effect of \(n\). In actual samples, the observed \(s\) values need not be identical, so the calculated standard errors may not follow this exact ratio.
Calculating \(s\) Before the Standard Error
When raw observations are given rather than a reported sample standard deviation, calculate \(s\) first. The sample standard deviation is the square root of the sum of squared deviations from \(\bar{x}\), divided by \(n-1\). Once \(s\) has been found, divide it by \(\sqrt{n}\) to estimate the standard error.
Worked Example: A Small Sample of Recovery Times
Suppose a hypothetical independent sample records recovery times, in days, for five patients receiving a particular treatment: 12, 14, 15, 15, and 19 days. Treat these values as a random sample from the population of interest. Calculate the estimated standard error of the sample mean.
First, the sample mean is
The deviations from the mean are \(-3,-1,0,0,4\) days, so the sum of squared deviations is \(9+1+0+0+16=26\) days squared. Thus,
Now use \(n=5\):
The estimated standard error is about 1.140 days. It estimates the sampling variability of a mean recovery time based on five observations under the stated sampling model. Because the sample is small, the estimate of variability may itself be imprecise; the formula does not remove uncertainty about how representative this particular sample is.
Conditions and Limits of the Formula
The usual formula is based on independent observations drawn from the same population distribution, with a finite variance. A random sample is a standard way to support this model. If observations are strongly dependent—for example, measurements from the same person treated as though they came from unrelated people—the calculation \(s/\sqrt{n}\) may not represent the actual variability of the mean. The sampling design and the way observations are related matter.
The formula also does not correct a biased sampling method. A very small standard error can indicate that sample means from repeated samples of the same size would cluster fairly closely under the model, but it cannot guarantee that the sample represents the population of interest. For example, a large sample drawn only from volunteers may estimate the mean for that group precisely while still failing to represent everyone the study intends to describe.
The standard error describes variability across hypothetical repetitions of the sampling process, not a visible range around one sample mean. It is also not the standard deviation of the population or a confidence interval. Those quantities answer different questions. Here the target is specifically the estimated spread of sample means, expressed as a single standard error.
Common Mistake: Confusing \(s\) with \(SE\)
A common mistake is to report the sample standard deviation as though it were the standard error, or to say that the standard error describes how far individual observations typically fall from the mean. For example, if \(s=12\) units and \(n=36\), reporting an estimated standard error of 12 units is incorrect. The calculation is
Here \(s=12\) units summarizes the spread of individual observations, while \(SE=2\) units estimates the sampling variability of the sample mean. A related error is dividing by \(n\) rather than \(\sqrt{n}\). The denominator in the standard-error formula is the square root of sample size.
Check Your Understanding
Answer each question using the distinction between variation among individuals and variation among sample means.
- A random sample has \(n=25\) and \(s=10\) kilograms. Calculate the estimated standard error, with units.
- In your own words, what does an estimated standard error of 3 minutes describe when the statistic is a sample mean time?
- If the sample size increases from 20 to 80 while \(s\) stays constant, by what factor does the estimated standard error change?
- Explain why a sample standard deviation of 6 units and a standard error of 6 units are not interchangeable.
- Give one reason why \(s/\sqrt{n}\) might fail to describe the sampling variability of a mean accurately.