Confidence Intervals: What They Mean, How to Calculate Them, and What They Do NOT Tell You

A confidence interval is a range of values, calculated from sample data, that is likely to contain the true population parameter. A 95% confidence interval for the population mean does NOT mean "there is a 95% probability that the true mean is in this interval." The true mean is a fixed number — it is either in the interval or it is not. There is no probability about it. What 95% confidence actually means: if you were to repeat the sampling process many times and compute a 95% confidence interval from each sample, approximately 95% of those intervals would contain the true population parameter. The 95% refers to the procedure, not to any individual interval. This distinction matters because the common misinterpretation leads to overconfidence in individual results. A single 95% CI gives you a range that was generated by a method that works 95% of the time — but you do not know whether this particular interval is one of the 95% that captured the truth or one of the 5% that missed. The practical interpretation that is both correct and useful: "We are 95% confident that the true population mean lies between [lower bound] and [upper bound]." This phrasing acknowledges that our confidence is in the method, not in the specific interval.

The general formula is: CI = point estimate ± margin of error = x̄ ± (critical value × standard error). For a population mean with known σ (rare in practice, common on exams): CI = x̄ ± z* × (σ/√n). Where z* is the critical z-value for the desired confidence level: z* = 1.645 for 90% CI, z* = 1.96 for 95% CI, z* = 2.576 for 99% CI. For a population mean with unknown σ (the typical real-world case): CI = x̄ ± t* × (s/√n). Where t* is the critical t-value with n-1 degrees of freedom. For large n, the t-distribution is very close to the normal distribution, so the z and t approaches give nearly identical results. For small n, the t-distribution has heavier tails, producing wider intervals to account for the extra uncertainty from estimating σ. Worked example: A sample of n = 36 has x̄ = 72 and s = 12. Construct a 95% CI. SE = s/√n = 12/√36 = 2.0. For 95% confidence with df = 35, t* ≈ 2.03 (from t-table). CI = 72 ± 2.03 × 2.0 = 72 ± 4.06 = (67.94, 76.06). Interpretation: We are 95% confident that the true population mean is between 67.94 and 76.06.

For a population proportion p, the sample proportion p̂ = x/n (successes divided by sample size). The 95% CI for p is: p̂ ± z* × √[p̂(1-p̂)/n]. The standard error for a proportion is √[p̂(1-p̂)/n]. This formula requires that np̂ ≥ 10 and n(1-p̂) ≥ 10 (the success/failure condition), which ensures the sampling distribution of p̂ is approximately normal (by the CLT). Worked example: In a poll of n = 500 voters, 280 support a candidate. p̂ = 280/500 = 0.56. Check conditions: 500 × 0.56 = 280 ≥ 10 and 500 × 0.44 = 220 ≥ 10. Good. SE = √[0.56 × 0.44 / 500] = √[0.000493] = 0.0222. 95% CI = 0.56 ± 1.96 × 0.0222 = 0.56 ± 0.0435 = (0.517, 0.603). Interpretation: We are 95% confident that the true proportion of voters who support the candidate is between 51.7% and 60.3%. Since the entire interval is above 50%, we can say with 95% confidence that the candidate leads among the population. This is how political polls report their results — the "margin of error" they cite is the ± term (in this case, ±4.35 percentage points).

Three factors control the width of a confidence interval. (1) Sample size (n): Larger n → smaller SE → narrower interval. Doubling n reduces the width by a factor of √2 ≈ 1.41. This is the most common way to get a more precise estimate. (2) Confidence level: Higher confidence → wider interval. A 99% CI is wider than a 95% CI, which is wider than a 90% CI. You pay for more confidence with less precision. The trade-off is fundamental: you can be very confident about a wide range ("the mean is between 20 and 80") or less confident about a narrow range ("the mean is between 48 and 52"). (3) Variability (σ or s): More variability in the data → larger SE → wider interval. A population with σ = 100 produces wider intervals than one with σ = 10, all else being equal. You cannot control population variability, but you can control sample size and confidence level. The relationship between these factors creates practical design questions: "How large a sample do I need to estimate the mean within ±2 units with 95% confidence?" You can solve for n: n = (z* × σ / E)², where E is the desired margin of error. This is called sample size determination and is tested frequently on exams. StatsIQ can help you calculate required sample sizes and construct confidence intervals step by step from your homework or exam problems.

Mistake 1: "There is a 95% chance the true mean is in this interval." Wrong. The true mean is either in the interval or not — it is a fixed number. The 95% refers to the long-run success rate of the method, not the probability for any specific interval. This is the single most tested misinterpretation in introductory statistics. Mistake 2: "95% of the data falls within the confidence interval." Wrong. The CI is about the population parameter (the mean), not about individual data points. A prediction interval estimates where individual observations fall — that is a different calculation and is always wider than a confidence interval. Mistake 3: "A wider interval is better because it is more likely to contain the true value." Partly right but misleading. Yes, wider intervals have higher coverage probability, but they are also less informative. An interval from negative infinity to positive infinity has 100% coverage but tells you nothing. The goal is the narrowest interval that maintains the desired confidence level. Mistake 4: Forgetting to check conditions. For a z-interval, you need the sampling distribution to be approximately normal (CLT: n ≥ 30, or population is normal). For a proportion, you need np̂ ≥ 10 and n(1-p̂) ≥ 10. Exams frequently present situations where conditions are not met and ask you to recognize this. Mistake 5: Interpreting non-overlapping CIs as definitive proof of difference. If two 95% CIs do not overlap, the difference is statistically significant. But if they do overlap, the difference might still be significant — overlapping CIs do not prove equality. The proper way to test for a difference is a two-sample test or a CI for the difference, not visual comparison of individual CIs.