Statistics · Foundations
Standard Deviation vs Standard Error: What’s the Difference?
They sound the same, they’re one letter apart in your output, and mixing them up is the most common mistake in undergraduate statistics. Here’s the difference, once and for all.
Two numbers. Almost the same name. Your software prints both. And roughly half of all undergraduate stats errors trace back to confusing them.
Here’s the one-sentence answer, and then we’ll earn it: standard deviation describes your data; standard error describes your estimate. One is about individuals. The other is about how much you’d trust an average.
Get this and a surprising amount of statistics clicks into place, because the standard error is the hinge that connects everything you know about a sample to everything you want to say about a population.
Standard deviation: how spread out are the people?
Standard deviation (SD) measures how much individual data points vary around their mean. Big SD, the values are scattered widely. Small SD, they cluster tight.
If the heights of students in a class have a mean of 170cm and an SD of 8cm, that’s telling you most students fall roughly 8cm either side of 170. It’s a fact about the spread of people.
🔑 Key term
Standard deviation — the typical distance of an individual observation from the mean of the data. It describes variability in the population or sample itself, and it does not shrink as you collect more data.
That last point is the one people miss. Collect more students and your estimate of the SD gets more accurate, but the SD itself doesn’t get smaller. People are as varied as they are. Adding more of them to your sample doesn’t make humans more alike.
Standard error: how much would your average bounce?
Standard error (SE) measures something completely different. It measures how much your sample mean would jump around if you repeated the whole study with a fresh sample.
You calculated a mean height of 170cm from 25 students. If you grabbed a different 25 students, you’d get a slightly different mean — maybe 168, maybe 172. The standard error tells you how much that mean would wobble from sample to sample.
🔑 Key term
Standard error — the standard deviation of the sampling distribution of a statistic (usually the mean). It measures the precision of your estimate, not the spread of your data, and it does shrink as your sample grows.
📐 Formula — the connection
SE = SD / √n
The standard error is just the standard deviation divided by the square root of your sample size. That single formula contains the whole relationship.
Read it: a bigger sample (larger n) shrinks the SE, because your average is built from more information. But the SD sits in the numerator untouched — the spread of individuals never changes. More data makes your average more reliable; it doesn’t make your people more similar.
💡 Insight — the √n tax
Look at that square root. To halve your standard error, you need to quadruple your sample. Going from 100 people to 400 only doubles your precision.
This one fact explains why political polls stall around 1,000 respondents, why clinical trials cost what they do, and why “just collect more data” stops being good advice quite quickly. Precision gets expensive fast.
Side by side
| Standard deviation | Standard error | |
|---|---|---|
| Describes | Spread of the data | Precision of an estimate |
| About | Individuals | A statistic (e.g. the mean) |
| As n grows | Stays roughly the same | Shrinks (÷√n) |
| Answers | “How varied are the people?” | “How much would my average bounce?” |
| Use it for | Describing a distribution | Confidence intervals, error bars, t-tests |
⚠ Common error — the error bar trap
What the marker sees: a graph with tiny error bars and a caption claiming “low variability in the data.”
Those bars are almost certainly standard errors, not standard deviations — and with a big sample, SE bars are tiny no matter how spread out the data is. Small SE bars mean your mean is precisely estimated, not that your data is tightly clustered. Always state which one your error bars show. A paper that doesn’t is hiding something or confused, and reviewers notice.
📊 Case study — the same dataset, two honest stories
A university surveys 2,500 graduates and finds a mean starting salary of £28,000 with a standard deviation of £9,000.
The SD of £9,000 is the true story of the graduates: salaries are all over the place. Plenty earn under £20,000, plenty over £38,000. A prospective student should look at this number, because it describes the range of outcomes they might personally face.
The SE is £9,000 / √2,500 = £9,000 / 50 = £180. Tiny. It says the university knows the average graduate salary very precisely — it’s £28,000, give or take a couple of hundred.
Both numbers are true. But watch what a dishonest prospectus does: it quotes the SE-based precision (“average salary £28,000 ± £180”) to make outcomes look certain, while the SD (“± £9,000”) is the number that actually tells a student what their own salary might be. Choosing which spread to report is a rhetorical decision, and now you can catch it.
Practice questions
Q1. A sample of 100 test scores has a mean of 65 and an SD of 20. Find the standard error of the mean.
Q2. You increase your sample from 100 to 400. What happens to the SD? What happens to the SE?
Q3. A researcher reports “mean reaction time 250ms ± 3ms.” A student concludes “almost everyone reacts within 3ms of 250.” What’s the mistake?
Q4. Two studies estimate the same quantity. Study A: SE = 1.2. Study B: SE = 0.4. Which gives the more precise estimate, and roughly how many times larger was B’s sample, assuming equal SDs?
Worked answers
A1. SE = 20 / √100 = 20 / 10 = 2. The mean of 65 is precise to about ±2; individual scores still vary by about ±20.
A2. The SD stays about the same — roughly 20, because the spread of the underlying scores hasn’t changed. The SE halves: it was 20/√100 = 2, now it’s 20/√400 = 1. Quadrupling n halved the SE. That’s the √n tax in action.
A3. The ±3ms is almost certainly a standard error, describing how precisely the average is known — not how spread out individuals are. Individual reaction times could vary by ±50ms or more. The student has read a statement about the precision of the mean as a statement about the spread of people. Classic SD/SE confusion.
A4. Study B is more precise — smaller SE. Since SE = SD/√n and the SDs are equal, SE is inversely proportional to √n. B’s SE is one-third of A’s (0.4 vs 1.2), so √n_B = 3√n_A, meaning n_B is 9 times larger. Nine times the data for three times the precision — the √n tax again.
The short version
• SD = spread of the data. About individuals. Doesn’t shrink with n.
• SE = precision of the mean. About your estimate. Shrinks as ÷√n.
• SE = SD / √n. That formula is the whole relationship.
• Halving the SE takes four times the data.
• Error bars: always say which one they are. It changes the entire meaning of the graph.
• SE = precision of the mean. About your estimate. Shrinks as ÷√n.
• SE = SD / √n. That formula is the whole relationship.
• Halving the SE takes four times the data.
• Error bars: always say which one they are. It changes the entire meaning of the graph.
Once you’re comfortable here, the standard error shows up everywhere next: it’s the engine inside confidence intervals and every regression standard error you’ll ever read.
References
1. Moore, D.S., McCabe, G.P. & Craig, B.A. (2021) Introduction to the Practice of Statistics. 10th edn. New York: W.H. Freeman.
2. Altman, D.G. & Bland, J.M. (2005) “Standard deviations and standard errors,” BMJ, 331(7521), p. 903.
This is Chapter 8 territory of Statistics Made Simple.
Our supplementary guide for undergraduates explains why every formula has the shape it does — not just what to type into R.