What Is a Standard Error (SE)? Definition, Formula & Uses

A standard error (SE) is a measure of how much a statistic calculated from a sample, most often the mean, is likely to vary from the true value for the whole population, simply because of the randomness of which units happened to fall into the sample. Every time you measure something from a sample rather than the entire population, your result is an estimate, and a different random sample would give a slightly different one. The standard error quantifies that expected variation: it is, in effect, the typical distance between your sample estimate and the true value. A small standard error means your estimate is precise; a large standard error means it is shaky and could easily have come out quite differently. Because it captures precision, the standard error is the engine behind confidence intervals, margins of error, and statistical significance. The most important practical fact about it is that it shrinks as the sample gets bigger, in proportion to the square root of the sample size, which is why more data buys more trustworthy estimates.

The standard error of the mean is the standard deviation of the sample divided by the square root of the sample size: SE = s / √n, where s is the sample standard deviation (how spread out the individual values are) and n is the number of observations. For example, if a sample of 100 orders has a standard deviation of 40 in order value, the standard error is 40 / √100 = 40 / 10 = 4. The formula makes two things clear. First, the more variable the underlying data (a larger s), the larger the standard error. Second, and more usefully, the standard error falls as the sample size grows, but only with the square root of n, so to halve the standard error you need four times as much data, not twice as much. This square-root relationship is the single most important thing to understand about the standard error.

They are related but answer different questions, and confusing them is a common mistake. Standard deviation describes the spread of the individual data points, how much the actual values vary around their average. It is a property of the data itself and does not shrink just because you collect more of it. Standard error describes the precision of a statistic estimated from the data, typically the mean: how much would this sample mean bounce around if I repeated the sampling? Crucially, the standard error does shrink as the sample grows, because a larger sample pins down the average more reliably even though the individual values are just as spread out as ever. A simple way to hold the distinction: standard deviation tells you how variable your customers are; standard error tells you how confident you can be in the average you calculated from them. The two are linked by SE = s / √n.

In A/B testing, the standard error is what separates a real difference from random noise. Each version's conversion rate is only an estimate based on a sample of visitors, and the standard error tells you how precise each estimate is. If the standard errors are large relative to the gap between the two versions, the observed difference could easily be chance, the kind of gap you would see even if the versions were identical. If they are small relative to the gap, the difference is unlikely to be an accident, which is what statistical significance measures. The confidence interval around each result is built directly from its standard error (roughly the estimate plus or minus about two standard errors for 95% confidence), and a test reaches significance when those intervals separate enough that chance becomes implausible. This is why sample size matters: because the standard error falls with the square root of n, running a test to a large enough sample shrinks the error until a genuine difference stands clearly above the noise.

There are two levers, and one is far more practical. The first and most reliable is to increase the sample size: because SE = s / √n, more data directly shrinks the standard error, though the square root means quadrupling the sample only halves the error, so precision gets progressively more expensive. The second lever is to reduce the variability of the underlying data (the s in the formula), sometimes possible by measuring a more consistent metric, segmenting a heterogeneous audience into more uniform groups, or using techniques that account for known sources of variation. What you must not do is shrink the standard error by cutting corners that introduce bias, such as removing 'inconvenient' data points or stopping collection the moment the numbers look favourable, those make the standard error smaller on paper while making the estimate less trustworthy. The honest summary: a bigger, well-collected sample is the main way to reduce the standard error.

A smaller standard error is better in the sense that it means a more precise estimate, but only if it was earned honestly. A standard error that is small because you collected a large, representative sample is genuinely good news. But it can also be made to look small for the wrong reasons: if you shrink it by dropping data that does not fit, ending collection early at a convenient moment, or working from a biased sample, you get a precise-looking estimate of the wrong number, high confidence in a wrong answer, which is worse than an honestly uncertain one. It is also worth remembering that precision is not the same as correctness: the standard error measures how much your estimate would bounce around due to random sampling, but it says nothing about systematic bias. A biased measurement can have a tiny standard error and still be wrong. So the goal is not simply the smallest possible standard error, but a small one produced by a large, unbiased, representative sample.

Omniconvert Explore is an A/B testing and experimentation platform, and the standard error is part of the statistical machinery it uses to tell you whether a result is trustworthy, so you do not have to compute it yourself. Each version's conversion rate is an estimate with a standard error attached, and Explore uses that to report statistical significance and confidence, telling you when the gap between control and variation is large enough, relative to the standard error, that chance is an unlikely explanation. This protects you from acting on random noise: rather than eyeballing a raw difference, you get a principled read on whether the standard errors are small enough for the result to be believed. Because the standard error shrinks with sample size, Explore also helps you judge when a test has collected enough data to be conclusive. And its segmentation lets you see results, and their precision, for specific audiences, since a small overall standard error can hide a much noisier picture within an important segment. Across more than 70,000 experiments, that discipline turns raw numbers into decisions you can defend.