Standard error tells you how precise an estimate is. If you’ve calculated a sample mean (or any other statistic), the standard error measures how much that estimate would bounce around if you repeated the study many times. A small standard error means your estimate is likely close to the true population value; a large one means there’s more uncertainty.
Standard Error vs. Standard Deviation
These two numbers answer different questions, and confusing them is one of the most common mistakes in data reporting. Standard deviation describes how spread out individual data points are within your sample. Standard error describes how confident you can be in a summary statistic, like the mean.
Here’s a concrete example. Say you measure the blood pressure of 100 people. The standard deviation tells you how much those 100 readings vary from person to person. The standard error of the mean tells you how close the average of those 100 readings is likely to be to the true average blood pressure of the entire population. Because standard error is always smaller than standard deviation, reporting SE when you should report SD makes your data look less variable than it actually is. A review in the British Journal of Anaesthesia found this exact error across multiple research journals, where authors used SE to describe sample variability, misleading readers into thinking individual measurements were more tightly clustered than they were.
The Formula and Why Sample Size Matters
The standard error of the mean is calculated by dividing the standard deviation by the square root of the sample size:
SE = SD / √n
This formula reveals two things that shrink standard error: less variability in the underlying data (smaller SD) and more observations (larger n). The square root in the denominator is important. Doubling your sample size doesn’t cut the standard error in half. To halve the SE, you need to quadruple the sample size. Going from 25 participants to 100 cuts SE by half. Going from 100 to 400 cuts it in half again. This is why large studies are more precise but face diminishing returns.
Practically, this means a study with a large standard error may simply have too few participants to draw firm conclusions, even if the underlying data isn’t particularly noisy.
Using Standard Error to Build Confidence Intervals
The most common way standard error gets used is to construct confidence intervals. A 95% confidence interval, for instance, gives you a range that would capture the true population value 95% of the time if you repeated the study over and over. For large samples, the math is straightforward:
95% CI = mean ± 1.96 × SE
So if your sample mean is 50 and the SE is 3, the 95% confidence interval runs from about 44.1 to 55.9. You can be fairly confident the true population mean falls somewhere in that range. For a 99% confidence interval, you use 2.58 instead of 1.96, which produces a wider range. For 90% confidence, you use 1.65, producing a narrower one.
With small samples (under about 30), these multipliers change. Instead of 1.96 for 95% confidence, you use a slightly larger number from the t-distribution that depends on your sample size. For a sample of 25, the multiplier is 2.064 rather than 1.96. The smaller your sample, the wider the interval needs to be to account for the extra uncertainty.
Reading Error Bars on Graphs
When you see error bars in a chart, the first thing to check is what they represent. Bars showing ±1 SE are much shorter than bars showing 95% confidence intervals, and they mean different things. A quick rule of thumb: if your sample has 10 or more observations per group, doubling the length of SE bars gives you an approximate 95% confidence interval.
If you’re comparing two groups on a graph, the amount of overlap between their error bars hints at statistical significance. When sample sizes are 10 or more and the SE bars have a gap between them (no overlap at all), the p-value is approximately 0.05. If the gap is twice the length of one SE bar, the p-value drops to around 0.01. With very small samples (around 3 per group), the rules shift: you need to double the SE bars and check whether those doubled bars overlap.
These are rough visual guides, not substitutes for actual statistical tests. But they help you quickly scan a figure and identify which comparisons are likely meaningful.
Standard Error in Regression
Standard error isn’t limited to means. In regression analysis, every coefficient (the slope for each variable in the model) comes with its own standard error. This SE tells you how much that coefficient would vary across repeated samples. A regression might tell you that each additional year of education is associated with $5,000 more in annual income. The standard error on that coefficient tells you how precisely that $5,000 figure is estimated.
A small SE relative to the coefficient itself suggests the relationship is reliably detected. A large SE relative to the coefficient means the estimate is noisy and you can’t be confident about the true size of the effect. Most statistical software divides the coefficient by its standard error to produce a t-statistic, which then gets converted into a p-value. So when you see a “significant” result in a regression table, it’s really saying that the coefficient is large enough relative to its standard error that it’s unlikely to be zero.
Standard Error for Proportions
When your data is a percentage or proportion (like the share of patients who responded to a treatment), the standard error formula changes slightly. Instead of needing a separate standard deviation, the variability is built into the proportion itself. If 60% of patients respond to a drug, the standard error depends on that 60% figure, its complement (40%), and the sample size:
SE = √(p × (1 − p) / n)
The maximum possible standard error occurs when the proportion is 50/50, because that’s where uncertainty is greatest. As the proportion moves toward 0% or 100%, the SE shrinks because the outcome becomes more predictable. This is why polling margins of error are largest when a race is close to tied.
What “Large” and “Small” Actually Mean
Standard error has no universal threshold for “good” or “bad.” It’s always relative to the scale of what you’re measuring and the precision you need. An SE of 2 is excellent if you’re estimating a country’s average income in thousands of dollars, but terrible if you’re estimating the average number of children per household.
The most useful way to evaluate a standard error is to convert it into a confidence interval and ask: is that range narrow enough to be informative? If a study reports that a new drug reduces blood pressure by 10 points with a 95% CI of 2 to 18, that’s a wide range. The drug might barely work, or it might work quite well. If the CI runs from 8 to 12, you have a much clearer picture. Both of those intervals are built directly from the standard error, and the width of the interval is what determines whether the finding is practically useful.
In formal terms, the standard error is a measure of precision, not accuracy. Precision means your repeated estimates cluster tightly together. Accuracy means they cluster around the right answer. A biased study design can produce a small standard error while still being systematically wrong. The standard error only captures random sampling variability, not errors introduced by flawed methods or unrepresentative samples.