Sample size determination
Choosing the number of observations for statistical inference.
Sample size determination is the process of deciding how many observations or replicates to include in a statistical sample. This choice is a key part of any empirical study aiming to draw conclusions about a larger population from a smaller group. In practice, the sample size is often set based on the cost, time, or convenience of gathering data, balanced against the need for enough statistical power. More complex studies may assign different sample sizes to different groups, such as in stratified surveys or experiments with multiple treatment arms. In a census, the goal is to collect data from the entire population, so the intended sample size equals the population size. In experimental designs, each treatment group might have its own sample size. There are several ways to choose sample sizes. One is relying on past experience, though small samples—while sometimes unavoidable—can lead to wide confidence intervals and a higher risk of errors in hypothesis testing. Another approach is setting a target variance for the estimate: higher precision (a narrower confidence interval) means a lower target variance. A third method is using a power target, which refers to the statistical test’s ability to detect an effect once the sample is collected. Finally, sample size can be determined by a desired confidence level: the higher the confidence level, the larger the sample size needed, assuming precision stays constant. Larger sample sizes generally improve the precision of estimates for unknown parameters. For example, to accurately estimate the prevalence of a pathogen in a fish species, examining 200 fish is better than 100. This is supported by fundamental statistical principles like the law of large numbers and the central limit theorem. However, in some cases, increasing sample size yields little or no gain in precision. This can happen due to systematic errors, strong dependence in the data, a heavy-tailed distribution, or bias. The quality of sample size can be judged by the resulting estimates. It is usually based on cost, time, or convenience, along with the need for sufficient statistical power. For instance, when estimating a proportion, one might want the 95% confidence interval to be less than 0.06 units wide. Alternatively, sample size can be assessed through the power of a hypothesis test—for example, aiming for 80% power to detect a 0.04-unit difference in support for a candidate between men and women. **Estimation of a proportion** A straightforward case is estimating a proportion, such as the share of residents in a community aged 65 or older. The estimator is \(\hat{p} = X/n\), where \(X\) is the number of positive cases in the sample. With independent observations, this estimator follows a scaled binomial distribution (and is the sample mean of Bernoulli data). Its maximum variance is 0.25, occurring when the true proportion \(p = 0.5\). Since the true \(p\) is often unknown, this maximum variance is commonly used for sample size calculations. If a reasonable estimate of \(p\) is available, the quantity \(p(1-p)\) can replace 0.25. For large sample sizes, the distribution of \(\hat{p}\) is approximately normal. Using the Wald method for the binomial distribution, a confidence interval is given by: \[ \hat{p} \pm Z \sqrt{\frac{\hat{p}(1-\hat{p})}{n}} \] where \(Z\) is the standard Z-score for the desired confidence level (e.g., 1.96 for 95% confidence).
- field
- Statistics
- known_for
- Choosing the number of observations in a sample to ensure sufficient statistical power and precision
Lore & Background
Sample size determination is the process of selecting the number of observations or replicates to include in a statistical sample, a decision that is fundamental to any empirical study aiming to infer population characteristics from a sample. In practice, the chosen sample size is typically governed by practical constraints such as cost, time, or convenience of data collection, balanced against the need for sufficient statistical power. For complex studies, sample sizes may vary across subgroups, as seen in stratified surveys or experimental designs with multiple treatment groups. In a census, the intended sample size equals the entire population. Sample sizes can be selected using several approaches: drawing on prior experience, though small samples risk wide confidence intervals and errors in hypothesis testing; targeting a specific variance for an estimate, where high precision (a narrow confidence interval) requires a low target variance; setting a power target for a statistical test to be applied later; or specifying a confidence level, where a higher confidence level demands a larger sample size for a given precision. Larger sample sizes generally increase precision in estimating unknown parameters, a phenomenon supported by the law of large numbers and the central limit theorem. However, increased precision may be minimal or absent in the presence of systematic errors, strong data dependence, heavy-tailed distributions, or bias. For estimating a proportion, a common scenario, the estimator follows a binomial distribution with maximum variance of 0.25 when the true proportion is 0.5; this maximum is often used conservatively in sample size calculations. The sample size formula for a proportion, derived from the Wald method and normal approximation, incorporates the desired confidence level and margin of error, with the margin of error being half the desired width of the confidence interval.
Reader's Guide
Sample size determination is significant because it directly affects the reliability and validity of study findings. Larger sample sizes generally lead to increased precision when estimating unknown parameters, as described by the law of large numbers and the central limit theorem. However, in some situations, the increase in precision for larger sample sizes is minimal or even non-existent due to systematic errors, strong dependence in the data, heavy-tailed distributions, or bias. Sample sizes may be chosen using several methods: based on experience, a target variance for an estimate, a power target for a statistical test, or a confidence level. For estimating a proportion, a common formula uses a conservative estimate of p = 0.5 to maximize variance, yielding n = Z² / W², where Z is the Z-score for the desired confidence level and W is the desired width of the confidence interval. The legacy of sample size determination lies in its foundational role in ensuring that empirical studies can produce meaningful and trustworthy conclusions.
Did You Know?
- The maximum variance of the estimator of a proportion is 0.25, which occurs when the true parameter p = 0.5.
- Sample sizes may be evaluated by the quality of the resulting estimates, often based on cost, time, or convenience of data collection.
- In experimental design, different treatment groups may have different sample sizes.
- The Wald method for the binomial distribution yields a confidence interval using the formula (p̂ − Z√(0.25/n), p̂ + Z√(0.25/n)).
More in Probability & Statistics 1-24
Spotted an error? Know more?
This is a living reference — every entry is fact-audited, and reader corrections feed straight into our audit queue. Suggest an edit · See this site's audit record