Sampling error is the difference between a result you get from surveying a portion of a group and the “true” result you would get if you could survey every single member of that group. It happens any time you study a sample instead of an entire population, even when your methods are perfectly designed. It is not a mistake or a flaw in the research. It is a built-in consequence of working with incomplete information.
Why Sampling Error Happens
Imagine you want to know the average height of all adults in your country. Measuring every single person is impractical, so you measure 1,000 people chosen at random. The average height in your sample will be close to the true average, but it almost certainly won’t match it exactly. That gap is the sampling error.
Every random sample is a slightly different slice of the population. If you drew a second sample of 1,000 people, you’d get a slightly different average. A third sample, another slightly different number. None of these samples are wrong. They each capture a real snapshot, but no single snapshot is the full picture. Sampling error is just the natural variation that comes from looking at a part instead of the whole.
How It Differs From Other Errors
Sampling error is often confused with other problems in research, but it’s fundamentally different. Non-sampling errors are caused by flaws in how data is collected or processed: poorly worded survey questions, people refusing to respond, recording mistakes, or surveying a group that doesn’t represent the target population. If a survey about all Australians only includes men, that’s a systematic bias, not sampling error.
The key distinction: sampling error can be measured and predicted mathematically. Non-sampling errors are much harder to detect and quantify. You can also eliminate sampling error entirely, at least in theory, by surveying the entire population (a census). Non-sampling errors, on the other hand, can creep into any study, even a census. Memory lapses, dishonest answers, and data entry mistakes don’t disappear just because you surveyed everyone.
What Controls the Size of Sampling Error
Two main factors determine how large the sampling error will be: the size of your sample and the diversity of the population you’re studying.
Sample size has the biggest practical impact. The standard error, which is the most common way to measure sampling error, equals the population’s standard deviation divided by the square root of the sample size. That square root relationship means there are diminishing returns. Doubling your sample from 500 to 1,000 cuts your margin of error substantially, but going from 1,500 to 2,000 buys you very little improvement. In polling, for example, a sample of 500 people gives roughly a 4.4% margin of error at 95% confidence. Increasing to 1,000 drops it to about 3.1%. Pushing all the way to 2,000 only gets you to about 2.2%.
Population variability matters too. If everyone in a group is relatively similar on whatever you’re measuring, a small sample will represent them well. If the group is highly diverse, with responses spread across a wide range, you need a larger sample to capture that spread accurately.
Margin of Error in Polls
The most visible place you encounter sampling error is in political polls. When a poll says a candidate leads 52% to 48% “with a margin of error of plus or minus 3 points,” that margin of error is a direct expression of sampling error. It means the true level of support could plausibly be anywhere from 49% to 55% for the leading candidate.
This matters more than most people realize. During the 2012 U.S. presidential election, a Gallup poll showed Mitt Romney leading Barack Obama by 7 points, while a University of Connecticut poll conducted in virtually the same time period showed Obama ahead by 3 points. That’s a 10-point gap between two reputable polls. Even accounting for both polls’ stated margins of error (2% and 3%, respectively), there was still a 5-point disparity that sampling error alone couldn’t fully explain. When polls diverge that dramatically, non-sampling errors like differences in how “likely voters” are defined are usually part of the story.
How Researchers Reduce It
The most straightforward way to reduce sampling error is to increase the sample size, though as noted above, each additional person you survey provides a smaller incremental benefit. At some point the cost of surveying more people isn’t worth the marginal improvement in precision.
A more efficient approach is stratified sampling. Instead of drawing one large random sample from the entire population, researchers divide the population into subgroups (strata) based on characteristics like age, income, or region, then sample from each subgroup separately. Because members within each stratum tend to be more similar to each other, this method can produce a smaller sampling error than a simple random sample of the same total size. It’s particularly effective when the characteristic being measured varies a lot between subgroups but relatively little within them.
Sampling error can never be completely eliminated from a sample-based study. But it can be estimated, reported transparently, and factored into conclusions. That’s what makes it different from the harder-to-detect errors that plague research. When you see a confidence interval or margin of error attached to a statistic, you’re looking at researchers being honest about the uncertainty that comes with studying a part of the world instead of the whole thing.