What Is a Confidence Interval? Definition and Uses

A confidence interval is the range of values that plausibly contains the true process figure — the true average, proportion or standard deviation — given the sample you actually collected. Every estimate computed from a sample carries uncertainty; the interval makes that uncertainty visible. A team that reports “average cycle time: 8.2 days” knows less than it sounds like it knows; one that reports “8.2 days, with a 95 percent interval from 7.6 to 8.8” is telling the truth in full.

How it works

The interval is built from three ingredients: the sample estimate at its center, the variation in the data, and the sample size. More variation widens it; more data narrows it. The confidence level — 95 percent by convention — describes the reliability of the method, not of any single interval: if you repeated the sampling many times and built an interval each time, about 95 percent of those intervals would capture the true value. Any single interval either contains the truth or does not; what the level promises is that the procedure rarely misses.

Width is the message most readers skip. A narrow interval says the estimate is pinned down; a wide one says the honest answer is “somewhere in here.” Wide intervals are not failures — they are early warnings that a decision needs more data than has been gathered so far.

A worked example

Suppose a support desk samples recent tickets and finds an average resolution time of 4.2 hours, with a 95 percent confidence interval running from 3.1 to 5.3. Management’s target is 4.0. The point estimate misses the target, but the interval tells the fuller story: values below 4.0 remain entirely plausible, so declaring the desk out of compliance on this sample would be premature. The team’s options are exactly two — accept the ambiguity, or collect enough additional tickets to narrow the interval until it answers the question. The numbers are illustrative; the logic is the standard one.

• The 95 percent describes the method’s long-run success rate, not the odds that this one interval holds the truth.
• Never report a point estimate without its interval — precision withheld is precision implied.
• A wide interval is information: the data cannot yet support the decision being asked of them.
• Comparing two overlapping intervals is not a substitute for a proper two-sample test.
• Halving an interval’s width requires roughly four times the data — plan sample sizes accordingly.

A confidence interval is honesty in numeric form: here is our estimate, and here is how far it might be wrong.

Confidence intervals run through the Measure and Analyze phases of our Green Belt program ($299, 35 hours, 100-question proctored exam, lifetime access). Black Belt training pairs them with power and sample-size planning — the discipline of deciding how narrow an interval must be before it is worth paying for.