Standard Error Versus Population Standard Deviation

Two Different Kinds of Spread

A population standard deviation describes how much individual observations vary around their population mean. A standard error describes how much a statistic—here, the sample mean—varies from sample to sample. The two quantities are related, but they are not interchangeable: the standard error of the sample mean depends on both the population spread and the sample size.

This distinction matters whenever a sample mean is used to summarize a population. A population can have widely varying individual values while the average of a sufficiently large sample varies much less. The standard error measures that sampling variability; it does not describe the spread of the individuals in the sample.

Definition: Let $X_1,\ldots,X_n$ be a random sample from a population with mean $\mu$ and finite variance $\sigma^2$. The sample mean is $\bar{X}=(X_1+\cdots+X_n)/n$. The population standard deviation of an individual observation is $\sigma$. The standard error of $\bar{X}$ is the standard deviation of its sampling distribution, $\sqrt{\operatorname{Var}(\bar{X})}$.

The phrase “standard deviation” can therefore refer to different random quantities. The population standard deviation $\sigma$ describes individual observations $X_i$; the standard error describes the statistic $\bar{X}$. Once a particular sample is observed, its mean is one realized value of $\bar{X}$. The standard error, by contrast, describes how that mean would vary over repeated samples under the stated sampling model.

Why the Sample Mean’s Standard Error Has This Form

Suppose the observations are independent and identically distributed, each with mean $\mu$ and variance $\sigma^2$. Independence makes the covariance between distinct observations zero. Applying the variance formula for a sum gives the variance of the average:

\begin{aligned} \operatorname{Var}(\bar{X}) &=\operatorname{Var}\left(\frac{1}{n}\sum_{i=1}^n X_i\right)\\ &=\frac{1}{n^2}\operatorname{Var}\left(\sum_{i=1}^n X_i\right)\\ &=\frac{1}{n^2}\sum_{i=1}^n\operatorname{Var}(X_i)\\ &=\frac{1}{n^2}(n\sigma^2) =\frac{\sigma^2}{n}. \end{aligned}

The third line uses independence: the variance of a sum of independent variables is the sum of their variances. Identical distributions give $\operatorname{Var}(X_i)=\sigma^2$ for every observation. Taking the square root yields the standard error:

\operatorname{SE}(\bar{X})=\sqrt{\operatorname{Var}(\bar{X})} =\frac{\sigma}{\sqrt{n}}.

This result requires a random sample with independent observations from the same distribution and a finite population variance. It is a statement about the variance and standard deviation of the sample mean. It does not require the population distribution to be normal. Normality is needed for an exact normal sampling distribution in settings such as the second example below, but not for this standard-error formula itself.

Result: For an independent, identically distributed sample of size $n$ with population standard deviation $\sigma$, $\operatorname{SE}(\bar{X})=\sigma/\sqrt{n}$. The population standard deviation measures individual spread; the standard error measures the sample mean’s spread across repeated samples.

The square-root relationship gives a useful check on answers. Multiplying the sample size by four divides the standard error by two, assuming the population standard deviation and sampling conditions remain unchanged. It does not divide the standard error by four, because standard errors are square roots of variances.

Worked Example: Standard Error for Hypothetical Claim Severities

A hypothetical insurer models individual claim severities as independent observations from the same population, with population mean $1,200 and population standard deviation $300. A random sample of 25 claims is selected. Find the standard error of the sample mean severity, and compare it with the population standard deviation. The amounts are stated in dollars.

Method. Apply the independent, identically distributed sample-mean formula. The population standard deviation is $300, and the sample size is $n=25$.

First calculate the square root of the sample size:

\sqrt{n}=\sqrt{25}=5.

Then calculate the standard error:

\operatorname{SE}(\bar{X})=\frac{\$300}{5}=\$60.

Thus, individual claim severities have a population standard deviation of $300, while the sample mean severity has a standard error of $60. These have the same dollar units but describe different distributions. Under the model, repeated sample means typically vary around the population mean $1,200 on a scale of $60.

Check. The variance of an individual observation is $300^2=90{,}000$ dollars squared. The variance of the sample mean is $90{,}000/25=3{,}600$ dollars squared, and $\sqrt{3{,}600}=\$60$. The calculation is consistent.

Common mistake avoided. Reporting $300 as the standard error would confuse individual spread with mean spread. Dividing by $n$ would also be wrong for the standard error: $\$300/25=\$12$, whereas dividing the standard deviation by $\sqrt{n}$ gives the correct $60.

Using the Formula for a Sample Proportion

A sample proportion is also a sample mean. If each observation records whether an event occurs, define $X_i=1$ when it occurs and $X_i=0$ otherwise. If the event probability is $p$, then $E[X_i]=p$ and $\operatorname{Var}(X_i)=p(1-p)$. The sample mean $\bar{X}$ is the proportion of observations for which the event occurs. Under independent sampling with the same event probability, its standard error is therefore $\sqrt{p(1-p)/n}$.

Worked Example: Standard Error of a Hypothetical Event Rate

In a hypothetical portfolio model, each policy independently has probability $p=0.08$ of producing a specified event during a year. For a random sample of 100 policies, calculate the standard error of the sample proportion of policies with the event. Give the probability to four decimals.

Method. Represent each policy by an indicator $X_i$, equal to 1 if the event occurs and 0 otherwise. Then $\bar{X}$ is the sample proportion, and the individual indicator variance is $p(1-p)$.

Calculate the variance of one indicator:

\operatorname{Var}(X_i)=0.08(1-0.08)=0.08(0.92)=0.0736.

Divide by the sample size to obtain the variance of the sample proportion, then take the square root:

\operatorname{Var}(\bar{X})=\frac{0.0736}{100}=0.000736, \qquad \operatorname{SE}(\bar{X})=\sqrt{0.000736}=0.027129\ldots.

Rounded to four decimals, the standard error is 0.0271, or about 2.71 percentage points. This is the standard deviation of the sample proportion across repeated samples under the model; it is not the standard deviation of the 0–1 indicator, which is $\sqrt{0.0736}=0.2713$ to four decimals.

Check. The standard error is one tenth of the indicator standard deviation because $\sqrt{100}=10$. Thus $0.2713/10$ is approximately $0.0271$.

Common mistake avoided. Substituting $p=0.08$ for the individual standard deviation would give $0.08/\sqrt{100}=0.008$, which is incorrect. The indicator’s standard deviation is $\sqrt{p(1-p)}$, not $p$.

When the Simple Formula Does Not Apply

The formula $\sigma/\sqrt{n}$ depends on the sampling assumptions. If observations are independent but have different variances, each variance contributes separately. If observations are dependent, covariance terms also contribute. In either case, the calculation must reflect the actual model rather than assume that the observations behave like an independent, identically distributed sample.

For independent observations with variances $\sigma_1^2,\ldots,\sigma_n^2$, the mean’s variance is $(\sigma_1^2+\cdots+\sigma_n^2)/n^2$. If the observations are dependent, the general expansion includes twice the sum of pairwise covariances, also divided by $n^2$. Positive dependence can make the mean less precise than the formula $\sigma/\sqrt{n}$ suggests.

Worked Example: A Mean of Independent Measurements With Unequal Variances

A hypothetical monitoring system takes four independent measurements, $X_1,\ldots,X_4$, of a process level. Their respective population variances are 4, 9, 16, and 25 units squared. Find the standard error of their arithmetic mean. Do not assume that all four measurements have the same variance.

Method. For independent observations that need not have equal variances, add their variances and divide by $n^2$. Here $n=4$, so $n^2=16$.

\begin{aligned} \operatorname{Var}(\bar{X}) &=\frac{4+9+16+25}{4^2}\\ &=\frac{54}{16}\\ &=3.375. \end{aligned}

Therefore,

\operatorname{SE}(\bar{X})=\sqrt{3.375}=1.837117\ldots.

The standard error is approximately 1.837 units, rounded to three decimals. The corresponding individual standard deviations are 2, 3, 4, and 5 units; there is no single common $\sigma$ to insert into $\sigma/\sqrt{n}$.

Check. The four individual variances sum to 54 units squared. Dividing by 16 gives the variance of the average, and its square root is approximately 1.837 units.

Common mistake avoided. Averaging the four standard deviations and dividing by 2 would use $(2+3+4+5)/4=3.5$, then $3.5/2=1.75$. That is not the correct variance calculation. Independent variances, not standard deviations, are added.

Distributional Claims and a Common Exam Trap

A standard error gives the spread of a sampling distribution, but it does not by itself specify the shape of that distribution. For example, the exact normal calculation below is justified because the individual observations are normal and independent. In other models, knowing the standard error alone does not make the sample mean normally distributed.

Worked Example: Using the Standard Error Under a Normal Model

Suppose individual readings from a hypothetical instrument are independent and distributed as $X_i\sim N(72,10^2)$, where the second parameter is the variance. A random sample of 25 readings is taken. Find $P(\bar{X}>75)$. Give the probability to four decimals.

Method. Because the individual observations are independent and normal, their sample mean is exactly normal. Its mean is 72, and its standard error is the population standard deviation divided by $\sqrt{25}$.

The individual standard deviation is $\sigma=\sqrt{10^2}=10$. Thus,

\operatorname{SE}(\bar{X})=\frac{10}{\sqrt{25}}=\frac{10}{5}=2.

Standardize the sample-mean threshold using the sampling distribution’s standard deviation, 2:

z=\frac{75-72}{2}=1.5.

Therefore,

P(\bar{X}>75)=P(Z>1.5)=1-\Phi(1.5)=1-0.9332=0.0668.

The required probability is 0.0668. The denominator in the standardization is the standard error of the sample mean, not the population standard deviation of a single reading.

Check. The threshold 75 is three units above the sampling-distribution mean, which is $3/2=1.5$ standard errors. The upper-tail probability at $z=1.5$ is 0.0668 to four decimals.

Common mistake avoided. Using 10 as the standardizing denominator would give $z=(75-72)/10=0.3$ and an upper-tail probability of 0.3821. That calculation treats the threshold as a deviation for one individual observation, but the event concerns the mean of 25 observations.

Practical Checks Before Reporting a Standard Error

Before calculating, identify the random quantity whose variability is requested. If the question concerns an individual observation, its population standard deviation is relevant. If it concerns a sample mean, use the sampling variance of that mean under the given sampling design. Check whether the observations are independent, whether they share a common variance, and whether the requested result is a standard deviation or a variance.

Keep the units straight. A standard deviation and a standard error have the same units as the observations; variances have squared units.
Use the square root of sample size. Under independent, identical sampling, the mean’s variance is $\sigma^2/n$, so its standard error is $\sigma/\sqrt{n}$.
Do not infer a distribution shape from a standard error. A standard error is a measure of spread, not a complete description of the sampling distribution.
Check the sampling assumptions. Dependence and unequal variances can invalidate the simple formula, even if the sample size is known.

The central distinction is compact: $\sigma$ describes the population distribution of individual observations, whereas $\operatorname{SE}(\bar{X})$ describes the sampling distribution of the sample mean. Under independent, identically distributed sampling with finite variance, the latter is $\sigma/\sqrt{n}$. This is the standard error calculation aligned with the SOA VEE Mathematical Statistics syllabus.

Check Your Understanding

For each question, identify the relevant sampling assumptions before calculating.

A random sample of 36 independent observations comes from a population with standard deviation 18 units. Find the standard error of the sample mean. (Answer: 3 units)
Independent Bernoulli observations have event probability $p=0.20$. For a sample of 50, find the standard error of the sample proportion to four decimals. (Answer: 0.0566)
Independent observations have distribution $N(72,10^2)$. For a sample of 25, find $P(\bar{X}>75)$ to four decimals. (Answer: 0.0668)
Four independent observations have variances 4, 9, 16, and 25 units squared. Find the standard error of their arithmetic mean to four decimals. (Answer: 1.8371 units)
Ten observations each have variance 25 and every distinct pair has correlation 0.20. Find the standard error of their mean, assuming these pairwise correlations apply to all pairs. (Answer: $\sqrt{7}\approx 2.6458$ units)

What You'll Learn