Standard Error for a Difference in Means

From Sampling Distribution to Standard Error

In “Sampling Distribution of \(\bar{x}_1-\bar{x}_2\),” we described how the difference between two independent sample means varies from sample to sample. Its theoretical standard deviation depends on the population standard deviations, \(\sigma_1\) and \(\sigma_2\). In practice, those population values are usually unknown. We estimate the spread using the sample standard deviations, \(s_1\) and \(s_2\).

This estimated spread is called the standard error of the difference in sample means. It describes the typical sample-to-sample variability in \(\bar{x}_1-\bar{x}_2\), estimated from the data. In this tutorial, we focus on calculating it carefully. It is one ingredient in a two-sample t confidence interval.

Use \(n_1\) and \(n_2\) for the sample sizes and \(s_1\) and \(s_2\) for the sample standard deviations. The formula is:

The expression inside the square root has two parts. The first, \(s_1^2/n_1\), estimates the contribution of sample 1 to the variance of the difference. The second, \(s_2^2/n_2\), estimates the contribution from sample 2. Because the samples are independent, these estimated variances are added. The square root converts the result back to the units of the original quantitative variable.

The subtraction order used to define \(\bar{x}_1-\bar{x}_2\) does not change the standard error: reversing the groups leaves the same two terms being added. However, the order does change the sign of the observed difference and the population difference it estimates. As in “Comparing Two Means: The Parameter of Interest,” state clearly which population is group 1 and which is group 2.

How to Calculate the Standard Error

A reliable calculation keeps the two groups separate until the final addition. First square each sample standard deviation. Divide each squared value by its own group’s sample size. Add those two results, then take the square root. The sample standard deviation is not divided directly by the sample size; its square is divided by the sample size.

The formula resembles the theoretical standard deviation in the previous tutorial, but the symbols have an important difference. The theoretical formula uses population standard deviations, \(\sigma_1\) and \(\sigma_2\); the standard error uses the sample estimates \(s_1\) and \(s_2\). The standard error is therefore an estimate, not a known population quantity.

Worked Examples

What the Standard Error Tells You

The standard error describes the estimated spread of the sampling distribution of \(\bar{x}_1-\bar{x}_2\). It is measured in the same units as the variable: centimeters for heights, minutes for times, and so on. It is not the standard deviation of individual observations pooled across both groups. It is also not the observed difference between the sample means.

A smaller standard error means the difference in sample means would generally be less variable from sample to sample, if the populations and sampling plans stayed the same. Increasing a sample size reduces that group’s variance contribution, since \(s_i^2/n_i\) becomes smaller when \(n_i\) increases and \(s_i\) is held fixed. A larger sample mean, by itself, does not reduce the standard error.

Each sample size matters separately. A large \(n_1\) reduces the first contribution, but it does not automatically reduce the second contribution. The effect of increasing a group’s sample size is greatest when that group’s contribution is a substantial part of the total variance. Do not replace \(n_1\) and \(n_2\) with their sum in the formula.

The standard error is an estimate based on the observed sample standard deviations. A different sample could produce different \(s_1\), \(s_2\), and therefore a different standard error. The calculation alone does not establish whether a two-sample t procedure is appropriate; the samples must also meet the relevant design and distribution conditions. “Identifying Two Independent Samples” discusses the design distinction between independent and paired data.

Common Mistakes and AP Exam Tips

• Forgetting to square the sample standard deviations: The terms are \(s_1^2/n_1\) and \(s_2^2/n_2\). Writing \(s_1/n_1\) and \(s_2/n_2\) gives the wrong calculation and units.
• Subtracting the two contributions: Add the estimated variance contributions for independent samples, then take the square root.
• Pairing the wrong sample size with a standard deviation: Keep each group’s values together. Use \(s_1^2/n_1\) and \(s_2^2/n_2\), not a cross-pairing.
• Taking the square root too early: Find both variance contributions, add them, and only then take one square root. Do not take separate square roots and add them.
• Confusing the standard error with an observed difference: The standard error is an estimated measure of variability, not \(\bar{x}_1-\bar{x}_2\) itself.
• Reporting squared units: The terms under the radical are in squared units, but the standard error is in the original units after taking the square root.
• Claiming that a larger sample mean lowers the standard error: The sample size affects the standard error through the denominator. The numerical value of a sample mean is not part of this formula.

For a clear AP response, show the formula, substitute each group’s \(s\) and \(n\), show the two contributions and their sum, and report the square root with units. A complete calculation makes it possible to see that the correct sample sizes were used and that the variance terms were added. Keep the group order clear when describing \(\bar{x}_1-\bar{x}_2\), even though reversing that order would leave the standard error unchanged.