The standard error puts a number on how much an estimate would change if the sample were drawn again.
It turns one sample into a statement about precision, which a single number cannot give on its own.
Every estimate in a report should come with a standard error or an interval built from it.
For example, with a standard deviation of 20 and a sample of 400, the standard error is 20 divided by 20, which is 1.
For a proportion, such as a 50% conversion rate from 100 visitors, the standard error is 5 percentage points.
The square root explains why precision is costly: halving the standard error needs 4 times as many observations.
Going from 100 to 10,000 observations makes the standard error 10 times smaller, not 100 times.
The formula assumes the sample was drawn at random, and it says nothing about bias.
It also assumes observations are independent, so clustered data needs a larger, corrected standard error.
Using the sample's own standard deviation in the formula adds a little extra uncertainty in small samples.
Error bars on charts often show 1 standard error, which is narrower than a 95% interval, so labels matter.
Bigger samples always shrink the standard error, but they never shrink the spread of the data itself.
In practice, the standard error is how analysts judge whether a difference between 2 groups is large or small.
A difference of 2 minutes means little with a standard error of 3, and a lot with a standard error of 0.5.
The common confusion is reporting the standard error when readers need the spread of the data, or the other way round.