The 10% condition is a rule of thumb for treating observations from sampling without replacement as approximately independent. It asks whether the sample size n is no more than 10% of the finite population size N: n ≤ 0.10N. It is a condition for an approximation, not a rule against collecting larger samples.

Why sampling without replacement matters
After a person or item is selected without replacement, it is no longer available for another draw. The draws are therefore dependent. When the sampling fraction n/N is small, the effect of that dependence on common standard-error calculations is often small enough to ignore in an introductory analysis.
The condition does not make a biased sampling method representative, and it does not make dependent observations exactly independent. Random selection and an appropriate sampling design still matter.
How to check the condition
- Identify the population that the sample was actually drawn from.
- Calculate the sampling fraction n/N.
- If it is at most 0.10, the usual 10% rule of thumb is satisfied. If it is larger, consider the finite-population correction or the appropriate sampling model.
Example: 100 or 300 residents out of 1,000
A simple random sample of 100 residents from 1,000 samples 10% of the population. A sample of 300 samples 30%. Both can provide unbiased estimates of the population mean or proportion under the same valid simple-random-sampling design. Sampling 300 residents does not create over-representation merely because the sample is larger.
The difference is how the standard error should account for sampling without replacement. Ignoring the finite population generally overstates the sampling variance when the sampling fraction is substantial.
The finite-population correction
For a simple random sample without replacement, with n > 1 and sample standard deviation s calculated using n−1, an estimated standard error of the mean is:
\[\widehat{SE}(\bar{x})=\frac{s}{\sqrt{n}}\sqrt{1-\frac{n}{N}}.\]
For n=100 and N=1,000, the correction multiplier is √0.9 ≈ 0.949. For n=300, it is √0.7 ≈ 0.837. The larger sampling fraction makes the correction more important. At a census, there is no sampling uncertainty about the finite population mean, although measurement errors and missing responses can still matter.
When the rule applies
Check the sampling design rather than deciding from a test’s name. Finite-population dependence can matter for estimates of means and proportions and for tests built from those samples. A chi-square or t procedure is not automatically exempt. Sampling with replacement and genuinely independent trials do not need this finite-population approximation.
Some courses use a stricter 5% guideline. These thresholds express different tolerances for ignoring a finite-population correction; crossing one is not evidence that the sample is biased. Neither threshold replaces the other assumptions required by the chosen method.
Related topics and references
Continue with confidence intervals and z-scores. For the underlying sampling formulas, see Penn State’s simple random sampling lesson and Berkeley’s discussion of standard errors.