Standard Deviation vs Standard Error of the Mean
SD describes spread in your data; standard error describes how precisely you've estimated the mean. SE = s / √n, a worked example and which to report.
Standard Deviation vs Standard Error: The Difference
The single most useful fact about standard deviation vs standard error is this: the standard deviation (SD) describes how spread out your individual data points are, while the standard error of the mean (SE) describes how precise your estimate of the mean is. They are not interchangeable, and reporting one when you need the other is the most common statistical mistake in introductory research. Altman & Bland (2005), in their BMJ paper 'Standard deviations and standard errors' (331:903), called this confusion 'almost universal' among new researchers. Use the SD to describe your data; use the SE to describe your estimate of the mean. The rest explains why, how the SE formula works with sample size, and which one belongs in your error bars.
What Each Measures
The standard deviation is a descriptive statistic. It tells you how much individual values in a data set differ from the mean. A small SD means the data cluster tightly around the mean; a large SD means the data are spread out. The SD is in the same units as your original data, which makes it directly interpretable. If you measure height in centimetres, the SD is in centimetres.
The standard error of the mean is an inferential statistic. It estimates how much the sample mean (x̄) would vary if you repeated the sampling process many times. It is not about the spread of your data; it is about the reliability of your mean as an estimate of the population mean (μ). The SE is always smaller than the SD because it shrinks as you collect more data.
Think of it this way: the SD answers 'How much do individual data points vary?' The SE answers 'How much would my mean change if I took another sample of the same size?'
The SE Formula: SE = s / √n and Why It Shrinks With n
The standard error equation is straightforward: SE = s / √n, where s is the sample standard deviation and n is the sample size. This equation is derived from the Central Limit Theorem, which OpenStax Introductory Statistics 2e covers in section 7.2 (The Central Limit Theorem for Sums). The key insight is that the SE gets smaller as n gets larger, and it shrinks by the square root of n, not by n itself.
Why the Square Root Matters
If you quadruple your sample size from 10 to 40, you do not quarter the SE. Instead, you halve it: √4 = 2, so SE becomes s/2. To cut the SE in half again, you would need to quadruple n once more, from 40 to 160. This diminishing return on sample size is why large studies need thousands of participants to achieve tiny margins of error, not just hundreds.
What the Equation Assumes
The SE equation assumes your sample is drawn randomly and independently from the population. If your sampling method is flawed (convenience sample, non-response bias), the SE can be misleadingly small because it does not account for systematic error. The equation only captures random sampling error, not bias.
The Failure Case
A common mistake is to report the SE as if it were the SD, especially when the sample size is large and the SE looks impressively small. A study with n=10,000 might have an SE of 0.1 even though the SD is 10. Reporting the SE as your data's variability makes your results look unrealistically precise and misleads readers into thinking individual observations are clustered tightly when they are not.
Worked Example: SD vs SEM in Practice
Take a small data set: reaction times in milliseconds from five participants: 240, 300, 310, 350, 400. The mean is 320 ms.5 ms.
If you report the SD (62.5 ms), you tell the reader that individual reaction times vary by about 62 ms around the mean. If you report the SE (27.9 ms), you tell the reader that your estimate of the mean (320 ms) has a margin of error of about ±28 ms if you repeated the experiment. Both numbers are correct, but they answer different questions.
Now double the sample size to 10, keeping the same SD. The SE becomes 62.5 / √10 ≈ 19.8 ms. The SD has not changed. The data spread is still 62 ms. Only the precision of the mean has improved. This concrete example makes the standard deviation vs standard error distinction visible: the SD stays put; the SE moves with sample size.
Altman & Bland (2005) recommend always clarifying which measure you are reporting. Their BMJ note advises authors to state 'mean (SD)' for descriptive statistics and 'mean (SE)' only when the purpose is to show the precision of the mean estimate. They warn against the common practice of attaching SE to every mean without explanation.
| Attribute | Standard Deviation (s) | Standard Error of the Mean (SE) |
|---|---|---|
| What it measures | Spread of individual data points | Precision of the sample mean as an estimate of the population mean |
| Unit | Same as the original data | Same as the original data |
| Affected by sample size n | No; s is a property of the data, not the sample size | Yes; SE = s / √n, so SE decreases as n increases |
| When it is used | Descriptive statistics: report alongside the mean to show data variability | Inferential statistics: report to show confidence in the mean estimate, often used in confidence intervals |
| Typical error bar | SD error bars show ±1 SD, covering about 68% of data in a normal distribution | SE error bars show ±1 SE, covering about 68% of sample means in repeated sampling |
| Interpretation | Describes your sample data | Describes your estimate of the population mean |
| Common mistake | Reporting SE as if it were the SD, exaggerating precision | Reporting SD as if it were the SE, understating precision of the mean |
Which to Report: SD vs SE in Tables, Figures and Error Bars
Journal guidelines and statistical best practice are clear: report the standard deviation when you are describing your sample data. Report the standard error only when you are making an inference about the population mean, typically in the context of a confidence interval. The choice depends entirely on what question you are answering.
In Tables
Altman & Bland (2005) recommend the format 'mean (SD)' for descriptive tables. If you must include the SE, label the column clearly: 'mean (SE)'. Do not omit the label and force the reader to guess which measure is reported. A reader who assumes SD when you meant SE will misinterpret your variability.
In Error Bars on Graphs
Error bars are a major source of confusion. SD error bars show the spread of the data; SE error bars show the precision of the mean. The two types of error bars can look very different for the same data set because the SE is smaller. If you use SE error bars without explanation, the casual reader will assume the bars show data spread and will underestimate the variability. A safe convention: always state 'error bars are ±1 SD' or 'error bars are ±1 SE' in the figure caption. If you are comparing groups, consider using confidence intervals instead of SE bars, because confidence intervals account for the t-distribution for small samples and are easier to interpret.
When You Must Use SE
Use the SE explicitly when you are constructing confidence intervals for the mean. The 95% confidence interval is the mean ± (t-critical × SE). The SE is the building block of the interval, not the interval itself. Never report SE as if it were a confidence interval. A confidence interval about ±2 SEs (for large n) is not the same as ±1 SE.
Link to Confidence Intervals
The standard error of the mean is the direct link between a point estimate and a confidence interval. The 95% confidence interval for the mean is calculated as x̄ ± (t × SE), where t comes from the t-distribution with n-1 degrees of freedom. For large samples (n > 30), t ≈ 1.96, so the interval is roughly x̄ ± 1.96 × SE. This interval tells you that if you repeated the sampling process 100 times, about 95 of the resulting intervals would contain the true population mean.
Without the SE, you cannot construct a confidence interval. Without a confidence interval, you cannot make a formal inference about the population mean. This is the single reason the SE matters: it is the denominator of the test statistic for a one-sample t-test and the standard input for every confidence interval equation. If you understand the standard deviation vs standard error distinction, you have the foundation for understanding confidence intervals and hypothesis tests.
The failure case here is using the SD instead of the SE in the interval equation. Doing so produces an interval that is too wide (by a factor of √n) and that has no inferential meaning. The interval becomes a range that covers about 68% of your data, not 95% of possible means. This is a common error in self-taught data analysis and one that invalidates any subsequent inference.
The One Thing That Most Often Goes Wrong
The most frequent error with standard deviation vs standard error is reporting the SE as a descriptive statistic, especially in bar charts. A researcher with n=500 collects reaction time data, computes a mean of 320 ms and an SE of 2.8 ms, then puts 'mean ± 2.8' in the graph. A reader interprets the 2.8 as the SD and thinks the data are very tight, when in fact the SD is 62.6 ms. The graph looks impressive and the reader is misled. Altman & Bland (2005) called this practice 'widespread and potentially dangerous'. The fix is simple: decide whether you are describing your sample (use SD) or estimating the population mean (use SE). If you are not sure, report the SD. You can always add the SE in the caption if needed.
Who the Subject Suits and Who Should Skip
The standard deviation vs standard error distinction suits high-school AP Statistics students who need to interpret data variability versus mean precision, college Stats 101 students writing lab reports that require correct error bars, and analysts who prepare summary tables for publication. It suits anyone who has been told to report 'mean ± something' and needs to know which something to use.
Skip this distinction if you are doing hypothesis testing, regression analysis, or Bayesian inference, those methods use the SE automatically in their calculations, and you rarely need to interpret it separately. Skip it also if you are using descriptive statistics only and never plan to make an inference about a larger population; in that case, the SD is all you need.
Common Questions
Is the standard error always smaller than the standard deviation?
Yes, for any sample size greater than 1. The formula SE = s / √n means the SE is the SD divided by the square root of n, so it is always smaller unless n = 1, in which case they are equal. For n = 4, the SE is half the SD. For n = 100, the SE is one-tenth of the SD.
When should I use standard deviation vs standard error in a table?
Use the standard deviation when describing the spread of your data, in the format mean (SD). Use the standard error only when you are making an inference about the population mean, and clearly label the column as mean (SE). Altman & Bland (2005) recommend defaulting to SD for descriptive tables.
Can I calculate the standard error from the standard deviation?
Yes. The formula is SE = s / √n, where s is the sample standard deviation and n is the sample size. You need both the SD and the sample size to compute the SE. If you know the population standard deviation (σ) instead, use σ / √n, though this is rare in practice.
What does it mean if my standard error is large?
A large standard error means your estimate of the mean is imprecise. This happens when your sample standard deviation is large, your sample size is small, or both. A large SE widens your confidence intervals and reduces the statistical power of any hypothesis tests you run.
Should my error bars show SD or SE?
It depends on what you want the graph to communicate. SD error bars show data spread; SE error bars show mean precision. Many journals require you to specify which one you used. A common compromise is to use confidence interval bars (mean ± 95% CI), which are roughly ±2 SE for large samples and are more informative. Never mix them without a clear caption.
Why does the standard error formula use n-1 for the SD but n for the SE?
The SD uses n-1 (Bessel's correction) to make the sample variance an unbiased estimator of the population variance. The SE uses n in the denominator of the square root because the Central Limit Theorem shows that the variance of the sample mean is σ²/n (population) or s²/n (sample). The correction for the SD is separate from the division by n in the SE formula.
What is the difference between SEM and SD in medical research?
In medical research, SEM (standard error of the mean) is often used in meta-analyses and when reporting treatment effects, because it shows the precision of the estimated effect size. SD is used when describing the range of patient outcomes. Altman & Bland (2005) caution that many medical papers misuse SEM as a descriptive statistic, leading to overconfident conclusions.