If you've ever stared at error bars on a graph wondering what they actually represent, you're not alone. The confusion between standard deviation (SD) and standard error (SE) is one of the most common mistakes in data analysis — and it shows up in published research more often than you'd think.
Both metrics involve variability. Both use similar formulas. But they answer fundamentally different questions. Here's how to keep them straight.
Standard deviation measures how spread out your data points are.
Standard error measures how precise your sample mean is as an estimate of the population mean.
In other words: SD describes your data. SE describes your estimate.
Imagine you're measuring the heights of 30 students in a classroom.
You calculate the average height: 165 cm.
Standard deviation answers the question: “How much do individual students vary from that average?”
If the SD is 10 cm, most students fall within a range around the mean. If the SD is 2 cm, everyone is nearly the same height.
Standard error answers a different question: “If I measured a different classroom of 30 students from the same school, how different would the average be?”
The SE tells you how much your sample mean (165 cm) might differ from the true average height of all students in the school.
Where:
Notice that SE is derived from SD. As your sample size increases, SE gets smaller — but SD stays roughly the same.
This makes intuitive sense: the more data you collect, the more confident you become in your estimate of the mean, but the underlying variability in the data doesn't change.
| Situation | Use | Why |
|---|---|---|
| Describing the spread of your data | SD | You're showing how variable individual observations are |
| Reporting demographics (age, weight, income) | SD | Readers want to understand the range of values |
| Showing precision of an estimated mean | SE | You're communicating confidence in your estimate |
| Confidence intervals around a mean | SE | CIs are built from SE (mean ± 1.96×SE for 95% CI) |
| Comparing means between groups | SE or CI | You're making inferences, not just describing |
| Error bars showing data variability | SD | You want to show the spread of raw data |
| Error bars showing estimate precision | SE | You want to show how reliable the mean is |
Describing your data? → Use SD
Making an inference about a population? → Use SE
A surprising number of published papers misuse SD and SE, particularly in figures with error bars. The problem? They often aren't labeled, leaving readers to guess.
Here's why it matters:
Using SE when you should use SD makes your data appear less variable than it actually is. The error bars look tighter, which can be misleading about the true spread of observations.
Using SD when you should use SE makes your estimate look less precise than it is. This can obscure real differences between groups.
Neither is inherently “wrong” — but they communicate different things. The error bars on a graph showing individual patient responses to a drug (SD) tell a different story than error bars showing how confident we are in the average response (SE).
Always label your error bars. If you see a paper that doesn't, be skeptical.
This trips people up, so let's address it directly.
Standard deviation is a property of your data. If you're measuring human heights, the SD will be around 7–10 cm regardless of whether you measure 30 people or 3,000. The population's variability doesn't change.
Standard error, however, depends on sample size:
As increases, you're dividing by a larger number, so SE decreases.
Intuition: The more people you measure, the more confident you are that your sample mean is close to the true population mean. Your estimate gets more precise — even though the underlying data variability stays the same.
| Sample Size (n) | SD | SE |
|---|---|---|
| 25 | 10 | 2.0 |
| 100 | 10 | 1.0 |
| 400 | 10 | 0.5 |
| 10,000 | 10 | 0.1 |
Picture a dartboard. You throw 50 darts at the bullseye. SD measures how scattered your darts are around where they landed on average. High SD = darts everywhere. Low SD = tight cluster.
Now imagine you repeat the experiment 100 times, each time throwing 50 darts and marking the center of where they clustered. SE measures how much those 100 center-points vary from each other. With more darts per round (larger n), each center-point estimate becomes more stable — so SE shrinks.
The interactive tool below lets you see this for yourself. Adjust the sample size and number of samples, then compare the left panel (individual data points — SD) with the right panel (sample means — SE).
How spread out are individual observations?
Sample Mean: 163.04 cm
Sample SD: 11.68 cm
Each dot is one person's height. The spread of these dots is what SD measures.
How much do sample means vary from sample to sample?
Theoretical SE: 1.83 cm
Each dot is the mean of a sample of 30 people. Notice how tightly they cluster compared to individual data points.
The blue curve shows individual data spread (SD = 10). The green curve shows how tightly sample means cluster (SE = 1.83). Try increasing the sample size — the green curve gets narrower while the blue stays the same.
10
Population SD (fixed)
1.83
SE = SD / √n = 10 / √30
n = 30
Sample size
Only if your sample size is 1 — and at that point, neither metric is meaningful. For any real dataset (n ≥ 2), SE will always be smaller than SD because you're dividing SD by .
It depends on what you're trying to communicate:
Many journals prefer 95% confidence intervals over SE bars because they're easier to interpret: if two 95% CIs don't overlap, the difference is likely significant (though this is a simplification).
Convention varies by field:
When in doubt, check the journal's guidelines — and always label what you're reporting.
This is exactly the problem. Using SE instead of SD isn't lying, but it can be misleading if readers expect to see data spread. A study with high individual variability (large SD) can look artificially clean if you only report SE.
Ethical data presentation means choosing the metric that answers the question your audience is asking.
| Standard Deviation (SD) | Standard Error (SE) | |
|---|---|---|
| Measures | Spread of individual data points | Precision of the sample mean |
| Describes | Your data | Your estimate |
| Changes with sample size? | No (roughly stable) | Yes (shrinks as n increases) |
| Formula | ||
| Use when | Describing variability | Making inferences |
If you've ever stared at error bars on a graph wondering what they actually represent, you're not alone. The confusion between standard deviation (SD) and standard error (SE) is one of the most common mistakes in data analysis — and it shows up in published research more often than you'd think.
Both metrics involve variability. Both use similar formulas. But they answer fundamentally different questions. Here's how to keep them straight.
Standard deviation measures how spread out your data points are.
Standard error measures how precise your sample mean is as an estimate of the population mean.
In other words: SD describes your data. SE describes your estimate.
Imagine you're measuring the heights of 30 students in a classroom.
You calculate the average height: 165 cm.
Standard deviation answers the question: “How much do individual students vary from that average?”
If the SD is 10 cm, most students fall within a range around the mean. If the SD is 2 cm, everyone is nearly the same height.
Standard error answers a different question: “If I measured a different classroom of 30 students from the same school, how different would the average be?”
The SE tells you how much your sample mean (165 cm) might differ from the true average height of all students in the school.
Where:
Notice that SE is derived from SD. As your sample size increases, SE gets smaller — but SD stays roughly the same.
This makes intuitive sense: the more data you collect, the more confident you become in your estimate of the mean, but the underlying variability in the data doesn't change.
| Situation | Use | Why |
|---|---|---|
| Describing the spread of your data | SD | You're showing how variable individual observations are |
| Reporting demographics (age, weight, income) | SD | Readers want to understand the range of values |
| Showing precision of an estimated mean | SE | You're communicating confidence in your estimate |
| Confidence intervals around a mean | SE | CIs are built from SE (mean ± 1.96×SE for 95% CI) |
| Comparing means between groups | SE or CI | You're making inferences, not just describing |
| Error bars showing data variability | SD | You want to show the spread of raw data |
| Error bars showing estimate precision | SE | You want to show how reliable the mean is |
Describing your data? → Use SD
Making an inference about a population? → Use SE
A surprising number of published papers misuse SD and SE, particularly in figures with error bars. The problem? They often aren't labeled, leaving readers to guess.
Here's why it matters:
Using SE when you should use SD makes your data appear less variable than it actually is. The error bars look tighter, which can be misleading about the true spread of observations.
Using SD when you should use SE makes your estimate look less precise than it is. This can obscure real differences between groups.
Neither is inherently “wrong” — but they communicate different things. The error bars on a graph showing individual patient responses to a drug (SD) tell a different story than error bars showing how confident we are in the average response (SE).
Always label your error bars. If you see a paper that doesn't, be skeptical.
This trips people up, so let's address it directly.
Standard deviation is a property of your data. If you're measuring human heights, the SD will be around 7–10 cm regardless of whether you measure 30 people or 3,000. The population's variability doesn't change.
Standard error, however, depends on sample size:
As increases, you're dividing by a larger number, so SE decreases.
Intuition: The more people you measure, the more confident you are that your sample mean is close to the true population mean. Your estimate gets more precise — even though the underlying data variability stays the same.
| Sample Size (n) | SD | SE |
|---|---|---|
| 25 | 10 | 2.0 |
| 100 | 10 | 1.0 |
| 400 | 10 | 0.5 |
| 10,000 | 10 | 0.1 |
Picture a dartboard. You throw 50 darts at the bullseye. SD measures how scattered your darts are around where they landed on average. High SD = darts everywhere. Low SD = tight cluster.
Now imagine you repeat the experiment 100 times, each time throwing 50 darts and marking the center of where they clustered. SE measures how much those 100 center-points vary from each other. With more darts per round (larger n), each center-point estimate becomes more stable — so SE shrinks.
The interactive tool below lets you see this for yourself. Adjust the sample size and number of samples, then compare the left panel (individual data points — SD) with the right panel (sample means — SE).
How spread out are individual observations?
Sample Mean: 163.04 cm
Sample SD: 11.68 cm
Each dot is one person's height. The spread of these dots is what SD measures.
How much do sample means vary from sample to sample?
Theoretical SE: 1.83 cm
Each dot is the mean of a sample of 30 people. Notice how tightly they cluster compared to individual data points.
The blue curve shows individual data spread (SD = 10). The green curve shows how tightly sample means cluster (SE = 1.83). Try increasing the sample size — the green curve gets narrower while the blue stays the same.
10
Population SD (fixed)
1.83
SE = SD / √n = 10 / √30
n = 30
Sample size
Only if your sample size is 1 — and at that point, neither metric is meaningful. For any real dataset (n ≥ 2), SE will always be smaller than SD because you're dividing SD by .
It depends on what you're trying to communicate:
Many journals prefer 95% confidence intervals over SE bars because they're easier to interpret: if two 95% CIs don't overlap, the difference is likely significant (though this is a simplification).
Convention varies by field:
When in doubt, check the journal's guidelines — and always label what you're reporting.
This is exactly the problem. Using SE instead of SD isn't lying, but it can be misleading if readers expect to see data spread. A study with high individual variability (large SD) can look artificially clean if you only report SE.
Ethical data presentation means choosing the metric that answers the question your audience is asking.
| Standard Deviation (SD) | Standard Error (SE) | |
|---|---|---|
| Measures | Spread of individual data points | Precision of the sample mean |
| Describes | Your data | Your estimate |
| Changes with sample size? | No (roughly stable) | Yes (shrinks as n increases) |
| Formula | ||
| Use when | Describing variability | Making inferences |
