Deriving the Standard Error of a Sample Mean

Why Derive the Standard Error?

Individual observations vary from one to another. A sample mean averages those observations, so its variability is not the same as the variability of one observation. To quantify the variability of the mean across repeated samples, we need to derive its variance and then take the square root. The result depends on the population standard deviation and the number of independent observations.

This calculation is about the sampling distribution of \(\bar{x}\): the distribution of sample means that would result from repeatedly taking samples under the same conditions. The standard deviation of that distribution is called the standard error. It describes the variability of the statistic across samples, not the spread of individual observations within a sample.

Assumptions and Notation

Let \(X_1,X_2,\ldots,X_n\) represent the observations in a random sample. For this derivation, assume each observation has a finite mean \(\mu\) and variance \(\sigma^2\), and that the observations are independent. In other words, each \(X_i\) comes from the same population model, and knowing the value of one observation does not change the probability distribution of another. We form the sample mean

Here \(\bar{X}\) denotes the random sample mean before a particular sample is observed; after collecting data, its realized value is written \(\bar{x}\). The population standard deviation \(\sigma\) measures the spread of individual observations, while \(SE(\bar{X})\) measures the spread of the sample mean over repeated samples. The variance of each observation is \(\sigma^2\), in squared measurement units.

The shared mean \(\mu\) is the same for each observation because they follow the same population model. The variance calculation below uses the fact that each observation has variance \(\sigma^2\), and that independence makes the covariance between distinct observations zero.

Deriving the Variance of the Sample Mean

The mean is a sum multiplied by \(1/n\). Two variance rules are needed: multiplying a random variable by a constant \(c\) multiplies its variance by \(c^2\), and the variance of a sum includes both the individual variances and any covariances between terms. Independence makes those covariance terms zero. The derivation follows directly from these facts.

The third line is the variance-of-a-sum formula: it adds each variance and twice each distinct pairwise covariance. To see why the covariance terms vanish, independence gives \(E[(X_i-\mu)(X_j-\mu)]=E[X_i-\mu]E[X_j-\mu]=0\) for \(i\ne j\), since both centered expectations are zero. There are \(n\) individual variance terms, each equal to \(\sigma^2\), so their sum is \(n\sigma^2\). Dividing by \(n^2\) gives \(\sigma^2/n\).

The standard error is the standard deviation of the sampling distribution, so it is the nonnegative square root of the variance:

The variance is measured in squared units, such as grams squared. Taking its square root returns the original measurement units, such as grams. This unit check is useful: a standard error for a mean weight should be expressed in weight units, not squared weight units.

Worked Calculations

What the Formula Depends On

The derivation shows precisely where independence enters: it removes covariance terms from the variance of the sum. If observations are dependent, the variance of their mean generally includes covariance contributions and need not equal \(\sigma^2/n\). The formula also uses a common variance; if observations have different variances, the sum of their variances is not \(n\sigma^2\).

For independent observations with the same population variance, the result applies regardless of the population's shape. A normal population is not needed to derive this variance. The result describes the variance and standard error of the sample mean; other questions about the shape of its sampling distribution require additional considerations.

This derivation also explains why the standard error contains \(\sqrt{n}\), not \(n\), in the denominator. Variances add across independent observations, and then the standard deviation is obtained by taking a square root. Specifically, variance decreases in proportion to \(1/n\), while standard error decreases in proportion to \(1/\sqrt{n}\).

Common Mistake

A frequent error is to write \(SE(\bar{X})=\sigma/n\), reasoning that averaging \(n\) observations divides their variability by \(n\). That confuses the scaling of a sum or mean with the scaling of a standard deviation. Multiplying a random variable by \(1/n\) divides its variance by \(n^2\); but the variance of the sum of \(n\) independent observations has already increased to \(n\sigma^2\). Together, these give \(\sigma^2/n\), whose square root is \(\sigma/\sqrt{n}\), not \(\sigma/n\).

For example, if \(\sigma=10\) units and \(n=25\), the incorrect calculation gives \(10/25=0.4\) units. The correct calculation is \(10/\sqrt{25}=2\) units. Substituting into the variance formula confirms the correct result: \(10^2/25=4\) units squared, and \(\sqrt{4}=2\) units.