The standard error is the standard deviation of a sampling distribution — the typical distance a sample statistic (like xˉ) lands from the true population value, computed as SE=σ/n (or s/n).
SE=nσ
Quick Answer
The standard error is the standard deviation of a sampling distribution — the typical distance a sample statistic (like xˉ) lands from the true population value, computed as SE=σ/n (or s/n). Use it whenever you report a sample mean or proportion and need to say how precise it is, and as the first ingredient of a margin of error, a confidence interval or a test statistic. For a sample proportion, SE=p^(1−p^)/n. Because of the square root, four times as many observations only halves the standard error.
Why This Matters
A sample average of 72 on a test means little until you know whether a repeat sample would give 71 to 73 or 60 to 84. The standard error puts that wobble into one number, and every margin of error, confidence interval and z or t statistic is built from it. It also prices extra data: halving the uncertainty costs four times the sample, which is why national polls stop near 1,000 people.
Intuitive Explanation
Suppose students at a large school sleep an average of 7 hours a night with standard deviation 1.5 hours. Ask one random student and the answer is typically about 1.5 hours away from 7. Ask 25 students and average: short sleepers and long sleepers partly cancel, and the average is typically only 1.5/25=0.3 hours from 7. Average 100 students and the typical miss is 1.5/10=0.15 hours. That typical miss of an average is the standard error.
Why n and not n? Random errors in a sum do not all point the same way, so the total error grows only like n while the sum itself grows like n. Dividing by n to get the average leaves an error that shrinks like 1/n. The cost is steep: going from 25 to 100 students halves the standard error, and halving it again takes 400.
The standard deviation looks similar but answers a different question. It measures how much individual students differ from one another, and it does not shrink with more data: survey 1,000 students and the SD is still about 1.5 hours. The standard error measures how much the average would change if you repeated the survey, and it does shrink. A quick test tells them apart: if the number gets smaller as you collect more data, it is a standard error.
Think of it like this.A jittery bathroom scale is typically off by about a kilogram per reading. Weigh yourself 4 times and average, and the average is typically off by about half a kilogram; average 100 readings and it is off by about a tenth. The single-reading jitter is the standard deviation; the average's jitter is the standard error. No amount of averaging fixes a scale that always reads 2 kg heavy.
Core idea.An average varies less than the values it is made of: across repeated samples, the sample mean's standard deviation is σ/n. That number measures an estimate's precision, and because of the square root, quadrupling the sample halves it.
When to Use
Use the standard error when you have a sample mean or proportion and must say how far it typically sits from the population value, or when a formula needs the spread of a statistic: a z or t statistic, a margin of error, a confidence interval. You need a spread (σ or s, or p^ for a proportion) and a sample size. If the question asks how much individual values vary, the answer is the standard deviation; if it wants a ± range at a stated confidence level, go one step further to the margin of error.
Tip.Ask: Do you need one number for how far a sample mean or proportion typically misses the true value from sample to sample?
How to Recognize It
Before dividing by n, check whether the spread you need is among single values or among sample results.
1.
Is the spread you need that of a statistic, like the mean of 100 people, rather than of single people?
Yes calls for the standard error. For the spread of individual values, use Standard Deviation and do not divide by n.
2.
Do you know a standard deviation and a sample size, or a sample proportion and a sample size?
Yes gives you what you need: ns for a mean or np^(1−p^) for a proportion. Without n, you cannot find it.
3.
Is a confidence level, such as 95%, part of the question?
If so, the standard error is only the first step; multiplying it by a critical value gives the Margin of Error.
4.
Does the question ask how much bigger a sample must be to cut the uncertainty?
Yes is a standard error question: halving it takes four times the sample, and cutting it to a third takes nine times.
Standard Error vs Standard Deviation vs Margin of Error vs Sampling Distribution
All four measure or describe spread, and two of them share units and look alike. Ask whether you need spread among single values, among sample results, a stretched spread for a stated confidence, or the whole pattern.
Standard Error vs Standard Deviation vs Margin of Error vs Sampling Distribution
Type
Meaning
Key test
Formula
Example
Standard Error
The standard error is the standard deviation of a statistic across repeated samples, the typical miss of an estimate.
Spread of a sample result; divides by n.
SE=ns
64 batteries with s=16 hours give SE=2 hours for the mean.
Standard Deviation
Standard deviation measures the typical distance of single data values from their mean.
Spread of single values, which does not shrink with more data.
s=n−1∑(x−xˉ)2
Single batteries typically last about 16 hours more or less than the average.
Margin of Error
The margin of error is the standard error multiplied by a critical value for a chosen confidence level.
Comes with a confidence level and is about twice the SE at 95%.
E=z∗⋅SE
At 95% confidence, 1.96×2≈3.9 hours for the battery mean.
Sampling Distribution
A sampling distribution is the full distribution of a statistic over all possible samples of a given size.
A whole picture with center, spread and shape, not one number.
—
A histogram of the means of thousands of simulated samples of 64 batteries.
Standard Error
Meaning
The standard error is the standard deviation of a statistic across repeated samples, the typical miss of an estimate.
Key test
Spread of a sample result; divides by n.
Formula
SE=ns
Example
64 batteries with s=16 hours give SE=2 hours for the mean.
Standard Deviation
Meaning
Standard deviation measures the typical distance of single data values from their mean.
Key test
Spread of single values, which does not shrink with more data.
Formula
s=n−1∑(x−xˉ)2
Example
Single batteries typically last about 16 hours more or less than the average.
Margin of Error
Meaning
The margin of error is the standard error multiplied by a critical value for a chosen confidence level.
Key test
Comes with a confidence level and is about twice the SE at 95%.
Formula
E=z∗⋅SE
Example
At 95% confidence, 1.96×2≈3.9 hours for the battery mean.
Sampling Distribution
Meaning
A sampling distribution is the full distribution of a statistic over all possible samples of a given size.
Key test
A whole picture with center, spread and shape, not one number.
Example
A histogram of the means of thousands of simulated samples of 64 batteries.
Formula & Notation
SE=nσ
For the sample mean, SE(xˉ)=nσ. When σ is unknown, estimate with SE(xˉ)=ns, where s is the sample standard deviation.
How to read it:SE is the standard error. σ is the population standard deviation, s is the sample standard deviation, and n is the sample size. SE=σ/n.
Worked Examples
Example 1 — Standard error of a mean, and the cost of halving it
Easy
A random sample of 64 light bulbs has mean lifetime 1,200 hours and standard deviations=80 hours. Find the standard error of the mean. How many bulbs would you need to test to bring it down to 5 hours?
Show the steps and the answer
Use s=80 in place of σ, with n=64.
The SD of all bulbs ever made is unknown, so the sample SD is the best available estimate of the spread.
Compute SE=80/64=80/8=10 hours.
This is the spread of the sample mean, not of single bulbs.
Interpret: across many samples of 64 bulbs, the sample mean would typically land about 10 hours from the true mean lifetime.
So 1,200 hours is a fairly precise estimate even though single bulbs vary by about 80 hours.
Solve 80/n=5: n=16, so n=256.
Halving the standard error from 10 to 5 needs n to double, which means four times as many bulbs.
Answer:SE=10 hours; cutting it to 5 hours needs n=256 bulbs.
Takeaway:Divide the spread by n to get the standard error; to shrink it by a factor k, multiply the sample size by k2.
Example 2 — One person or the average? SD versus SE
Standard
A random sample of 100 men has mean height 175 cm and standard deviation 7 cm. (a) About how far from 175 cm is a randomly chosen man's height, typically? (b) About how far is 175 cm from the mean height of all men in the population, typically?
Show the steps and the answer
Read question (a): it is about one man.
Individual heights vary with the standard deviation, so the sample size does not enter.
Answer (a) with the SD: about 7 cm.
A random man is typically about 7 cm taller or shorter than average, and a bigger sample would not change that.
Read question (b): it is about the sample mean as an estimate.
How far an estimate misses the population value is a standard-error question.
Answer (b): SE=7/100=0.7 cm.
Averaging 100 heights cancels most of the individual variation, leaving a typical miss of under 1 cm.
Answer:(a) About 7 cm (standard deviation). (b) About 0.7 cm (standard error).
Takeaway:Questions about individuals use the standard deviation; questions about how good an estimate is use the standard error, which is smaller by a factor of n.
Example 3 — Find the mistake: doubling the sample
Application
A survey of 100 students finds mean daily phone time 4.2 hours with s=2 hours, so SE=2/100=0.2 hours. Jordan writes: "To cut the standard error in half, to 0.1 hours, we just need to survey 200 students." Find the mistake and correct it.
Show the steps and the answer
Test Jordan's plan: 2/200≈2/14.14≈0.141 hours.
Doubling the sample leaves the standard error well above the 0.1-hour target.
Find the error: Jordan treated the standard error as if it shrinks like 1/n.
It shrinks like 1/n, so doubling n divides it by 2≈1.41, not by 2.
Solve 2/n=0.1: n=20, so n=400.
Halving the standard error requires doubling n, which means four times the original 100 students.
Answer:They need 400 students, not 200; 200 students give SE≈0.14 hours.
Takeaway:Precision gets expensive fast: every halving of the standard error needs four times the data.
Where You'll See It
Asking 10 friends vs 1000 strangers
Estimating the school's favorite lunch from 10 friends gives a jumpy, unreliable estimate; asking 1000 students gives a steadier one. The standard error shrinks as the sample grows because SE=σ/n.
Repeated free-throw percentage estimates
If you estimate a player's true free-throw rate from 20 shots, your estimate bounces around a lot between trials; from 200 shots it bounces far less. That typical bounce is the standard error.
Quadrupling a survey to halve uncertainty
Because of the n in the denominator, going from 250 to 1000 respondents (four times as many) only cuts the standard error in half, so doubling a poll's precision takes four times as many respondents.
Common Mistakes
The slip:Standard error and standard deviation are the same thing.
Why it's tempting:Both measure 'spread' and are reported with the same units, so they look like synonyms.
Instead:Standard deviation measures spread of individual data values; standard error measures spread of a sample statistic across samples, and equals σ/n — always smaller for n>1.
The slip:Standard error stays the same no matter how big the sample is.
Why it's tempting:Students treat SE as a fixed property of the data rather than something that depends on n.
Instead:Standard error decreases as sample size increases, because of the n in the denominator of SE=σ/n.
The slip:Computing the standard error as σ/n, so 64 bulbs with s=80 hours give 80/64=1.25 hours.
Why it's tempting:Finding a mean means dividing by n, so dividing the spread by n feels like the same step.
Instead:Divide by n: SE=80/64=80/8=10 hours. Check it against the rule that four times the data only halves the standard error; dividing by n would cut it to a quarter.
Mini Practice
Try these on your own. Tap Reveal when you want to check.
1.
Test scores have standard deviation 20 points. For a random sample of 25 students, about how much does the sample mean typically vary from the true mean?
Hint:Divide the spread of individual scores by n.
2.
With σ=15, find the standard error of the sample mean for n=25 and for n=100. What does quadrupling the sample size do?
Hint:Plug each sample size into σ/n and compare.
3.
A random sample of 400 voters finds p^=0.55. Compute the standard error of p^ using p^(1−p^)/n.
Hint:Multiply 0.55 by 0.45 first, then divide by 400.
4.
A researcher knows σ=12 minutes and wants the standard error of the sample mean to be 1.5 minutes. How large must the sample be?
Hint:Set 12/n=1.5 and solve for n first.
5.
Adult heights in a city have mean 168 cm and standard deviation 9 cm, and a random sample of 81 adults has mean 167 cm. Which number describes how individual heights vary, and what number tells how precise 167 cm is?
Hint:One number is about single people; the other is about an average of 81 people.
6.
A student writes: "σ=30 and n=9, so SE=30/9≈3.33." Find and fix the mistake.
Hint:Check the denominator in the standard error formula.
Want the full set?
50 practice questions for this concept — free to try, every one with a complete worked
solution showing the why, not just the answer.
Why divide by the square root of n instead of by n?
Because chance errors in a sample partly cancel. When you add n independent values, their variances add, so the total's standard deviation grows like nσ, not nσ. Dividing the total by n to get the mean leaves nnσ=nσ. If errors never cancelled, the mean of many values would be no more precise than a single value.
Is a smaller standard error always better?
For precision, yes, but precision is not accuracy. A huge sample from a biased source, such as an online poll of a million volunteers, can have a tiny standard error and still be far from the truth, because the standard error measures only chance wobble and assumes the sample was random. A small, well-chosen sample with a larger standard error can land closer to the real value.
What do error bars on a graph show?
Error bars mark the uncertainty around a plotted value, but studies draw them differently: some show one standard error, some a 95% confidence interval, which is about twice as wide, and some the standard deviation of the data. Always read the caption. Two means whose one-standard-error bars overlap are not significantly different, but bars that barely miss each other do not by themselves prove a difference.
You need Standard Deviation (spread of individual values) and Sampling Distribution (how a statistic varies over repeated samples). The standard error compresses that sampling spread into one number you can compute from a single sample, σ/n. Margin of Error multiplies it by a critical value, and a Confidence Interval places that ± around the estimate.