Confidence Interval

A confidence interval is a range of values, calculated from sample data by a method that, across repeated samples, captures the true population parameter at a specified rate called the confidence level. The most commonly used level is 95%, meaning that if the study were repeated many times under identical conditions, approximately 95% of the calculated intervals would contain the true parameter. Confidence intervals provide more information than point estimates alone because they communicate both the magnitude of an effect and the precision with which it has been estimated.

Point estimates such as means, proportions, or odds ratios tell you the best single guess for a population parameter, but they say nothing about how reliable that guess is. A confidence interval addresses this gap by quantifying the uncertainty surrounding an estimate. Wide intervals signal low precision and suggest that larger samples or more controlled designs are needed. Narrow intervals indicate a precise estimate, with little random error, although a narrow interval does not protect against bias. Many journals now require confidence intervals alongside p values because they convey practical significance in a way that a binary significant or not significant verdict cannot.

A confidence interval is constructed by taking the point estimate and adding and subtracting a margin of error. The margin of error depends on the standard error of the estimate and a critical value from the relevant distribution (often the z or t distribution). For a 95% confidence interval around a mean, the formula is: sample mean plus or minus (critical value times standard error). The standard error, in turn, depends on the sample size and the variability in the data. Larger samples and lower variability produce narrower intervals. For proportions, odds ratios, and hazard ratios, analogous formulas exist. Ratio measures such as odds ratios and hazard ratios are analyzed on the log scale, where their sampling distribution is closer to normal, so their intervals are symmetric on the log scale but asymmetric around the estimate once converted back to ratios.

Report confidence intervals whenever you present a quantitative estimate in a research paper. They are essential in clinical trials for showing whether treatment effects are clinically meaningful, not just statistically significant. In meta analyses, the confidence interval around the pooled effect size shows how precisely the average effect is estimated; judging how consistent the evidence is across studies needs measures of heterogeneity and, in a random effects meta analysis, a prediction interval. In epidemiology, confidence intervals around incidence rates or relative risks contextualize the precision of disease burden estimates. Even in exploratory studies, confidence intervals help readers calibrate how much weight to place on the findings.

A common misinterpretation is treating a 95% confidence interval as meaning there is a 95% probability that the true value lies within that specific interval. In frequentist statistics, the true value is fixed; it is the interval that varies across hypothetical repetitions. Another mistake is ignoring confidence intervals that cross clinically meaningful thresholds. A treatment might show a statistically significant effect, but if the lower bound of the confidence interval is close to zero, the clinical relevance is uncertain. Researchers also sometimes report confidence intervals only for primary outcomes while omitting them for secondary endpoints, creating an incomplete picture of precision across the study.

A randomized trial of a new antihypertensive drug reports that systolic blood pressure decreased by an average of 8.3 mmHg more in the treatment group than the placebo group, with a 95% confidence interval of 5.1 to 11.5 mmHg. Because the entire interval falls above the prespecified clinically important difference of 5 mmHg, the investigators conclude that the drug produces both a statistically significant and clinically meaningful reduction. If the interval had been 0.4 to 16.2 mmHg, statistical significance would still hold (since zero is excluded), but the lower bound would suggest that the true effect could be negligibly small.