iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Choose an uncertainty measure by first asking what you are uncertain about: how much an estimate would vary across samples, which population values are compatible with a statistical procedure, where a Bayesian model places probability on a parameter, or where one future observation might fall. Standard errors and confidence intervals address common frequentist questions; credible intervals summarize Bayesian posterior uncertainty; prediction intervals concern future individual outcomes. None automatically accounts for every problem in the data or evidence.
First ask: uncertainty about what?
Individual observations can vary widely even when their average is estimated precisely. Conversely, a seemingly simple estimate can be uncertain because it comes from a limited or variable sample. The right measure depends on the target: a statistic across repeated samples, a population parameter, a parameter’s posterior distribution, or one future observation.
These measures are not interchangeable labels for a generic “margin of error.” Each answers a different question and relies on its own framework and assumptions.
1. Standard error: how much an estimate varies across samples
A standard error (SE) describes the sampling variability of an estimate, such as a sample mean or proportion. Imagine repeatedly drawing samples under the same design and calculating the same statistic each time: the standard error describes the spread of that statistic’s sampling distribution.
#1 Best Overall
That is different from the standard deviation, which describes the spread of individual observations in a dataset. The standard error is about the estimate; the standard deviation is about the data.
How it is calculated for a sample mean
Under the usual independent-sampling setup, the standard error of the sample mean is the standard deviation divided by the square root of the sample size:
SE of the sample mean = standard deviation ÷ √sample size
Recommended Free Tools
Rank #2
- This guide is a perfect overview for the topics covered in introductory statistics courses.
It is expressed in the same units as the observations. A smaller standard error generally indicates a more precise estimate in the same context. A larger sample tends to reduce the standard error, while greater underlying variation tends to increase it. Other sampling designs and estimators may require different calculations.
2. Confidence interval: a range from a frequentist procedure
A confidence interval gives lower and upper bounds for a population parameter using a procedure designed to have a stated coverage rate under specified assumptions. It is constructed using an estimate and a measure of its sampling uncertainty, often a standard error.
How to interpret a 95% confidence interval
A 95% confidence level describes the procedure, not a probability assigned to the fixed parameter after a particular interval has been calculated. If the same procedure were repeated many times under its model and sampling design, about 95% of the resulting intervals would contain the fixed population parameter. The Office for National Statistics (ONS) describes its 95% interval interpretation in those repeated-sampling terms.
Rank #3
So, after calculating one interval, it is not strictly correct in the frequentist interpretation to say there is a 95% probability that this particular interval contains the fixed parameter. The parameter is treated as fixed; the interval varies from sample to sample.
How the interval is formed
For a normal-approximation interval when its assumptions are suitable, ONS gives the 95% limits as the sample estimate plus or minus 1.96 standard errors. That 1.96 multiplier is the method stated on the ONS guidance page, which does not display a publication year and was accessed in 2026. It is not a universal multiplier: the appropriate critical value and standard-error calculation depend on the design, estimator, and degrees of freedom.
Increasing the confidence level generally makes an interval wider. ONS notes that a 99% interval is wider than a 95% interval, reflecting the need for the procedure to cover the parameter more often over repeated samples.
As a historical illustration—not a current employment figure—ONS reported a UK employment estimate of 32.75 million with a confidence interval of ±177,000 for July–September 2019. The example shows how an estimate and its interval can be reported; it should not be read as a present-day labor-market estimate.
3. Credible interval: where a Bayesian model places posterior probability
A Bayesian credible interval summarizes uncertainty about a parameter using its posterior distribution, which combines prior information with observed data under a model. A 95% credible interval can be described as containing 95% of the posterior probability for the parameter, conditional on that model and prior.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsThis is a probability statement about the parameter within the Bayesian analysis. It differs from the frequentist confidence interval’s long-run coverage statement. The interval’s interpretation depends on the assumptions, prior, and data used to obtain the posterior.
Best Value
4. Prediction interval: where one future observation may fall
A prediction interval is for a new individual outcome, not merely an unknown population parameter or the average response. In a regression setting, NIST explains that prediction uncertainty combines uncertainty in the estimated average response with the residual variation of an individual observation.
Because an individual outcome varies around the fitted average as well as inheriting uncertainty from that estimated average, a prediction interval is generally wider than an interval for the mean response in the same setting. Use it when the practical question is about a future case—for example, a single future measurement—rather than the underlying average.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How the four measures differ
| Measure | Main question | What it describes | Interpretation |
|---|---|---|---|
| Standard error | How much would this estimate vary across samples? | Sampling variability of a statistic | A smaller SE usually means greater precision in the same context. |
| Confidence interval | What parameter values does this frequentist procedure cover? | An interval estimate with a stated coverage level | The coverage rate is a repeated-sampling property under assumptions. |
| Credible interval | Where does the posterior distribution place probability on the parameter? | Bayesian posterior uncertainty | The probability statement is conditional on the model and prior. |
| Prediction interval | Where might one future individual observation fall? | Uncertainty about an individual outcome | Includes individual residual variation as well as uncertainty in the fitted mean. |
Which measure should you use?
- You want to describe the precision of an estimate across samples: use its standard error, with the sampling design and estimator in view.
- You want a frequentist interval for a population parameter: use a confidence interval and state its confidence level and assumptions.
- Your analysis is Bayesian and you want a probability summary for a parameter: use a credible interval, making clear that the result is conditional on the model and prior.
- You want a range for one future individual outcome: use a prediction interval rather than an interval for the mean response.
What these measures do not capture by themselves
A standard error or confidence interval quantifies sampling uncertainty as represented by its method; it does not automatically account for every way the evidence could be wrong or incomplete. The U.S. Bureau of Labor Statistics (BLS), for example, distinguishes sampling error from nonsampling problems in Current Population Survey estimates, including nonresponse, inaccurate answers, processing errors, and coverage failures.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallThere is also a distinction between uncertainty about a clearly defined quantity and broader uncertainty about whether the evidence is relevant or adequate to answer a wider question. The UK Office for Statistics Regulation advises presenting relevant uncertainty in a concise, accessible way. In practice, explain what the measure covers, what assumptions it relies on, and what data-quality or relevance issues remain outside it.
Do overlapping confidence intervals prove that two estimates are not different?
No. Overlap is not a definitive significance test. BLS cautions that a formal test is needed to determine whether estimates differ statistically; the appropriate test depends on the estimates and design. BLS also warns that its CPS standard errors should not be used to test short-term changes without following the specified CPS documentation.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
