Content
Summary
Binomial estimation infers the population proportion p using the sample proportion p-hat, which is an unbiased estimator. Interval estimation relies on the normal approximation via the Wald method when the validity condition n*p-hat*q-hat is at least 5. For small samples or extreme proportions where this condition fails, the exact Clopper-Pearson method is required to ensure accurate coverage.
Intuition
Think of the sample proportion as a snapshot of the true population rate. Because it is based on independent trials, the Central Limit Theorem tells us that for large enough samples, this snapshot follows a bell curve centered on the true rate. This allows us to draw a symmetric range (confidence interval) around our estimate. However, if the sample is small or the rate is very low or high, the bell curve shape breaks down due to skewness. In those cases, we cannot rely on the simple symmetric formula and must use an exact method that respects the discrete, skewed nature of the data.
Key equations
Sample proportion estimator
Unbiasedness property
Estimated standard error
Wald confidence interval
Derivation
- The sample proportion is the mean of n independent Bernoulli trials.
- By properties of expectation, , making it an unbiased estimator.
- The variance of a single trial is pq, so the variance of the mean is pq/n.
- The Central Limit Theorem states that approximates a normal distribution N(p, pq/n) for large n.
- Since p is unknown, we substitute for p in the standard error formula.
- The Wald interval is constructed by adding and subtracting the critical value times the estimated standard error from the point estimate.
Worked example
Calculate the 95% confidence interval for breast cancer prevalence in a study of 10,000 women where 400 had a family history.
- Calculate point estimate: .
- Check validity: . Since 384 >= 5, the normal approximation is valid.
- Calculate standard error: .
- Determine margin of error: .
- Construct interval: , resulting in [0.036, 0.044].
Answer [0.036, 0.044]
Common mistakes
- Applying the Wald method when , leading to invalid intervals.
- Ignoring skewness when p is close to 0 or 1, which violates normal approximation assumptions.
- Using the exact standard error formula with unknown p instead of the estimated version.
- Assuming the confidence interval is always symmetric around the point estimate.