How to Understand Confidence Intervals

The direct answer: a confidence interval is a range built from sample data that is likely to contain the true population value. A 95 percent interval means that if you repeated the sampling process many times, about 95 percent of the intervals you build would capture the true parameter. This guide explains the formula, the critical values, and the right way to phrase the result so you stop saying things textbooks mark wrong.

Confidence Intervals at a Glance

How the Formula Is Built

Every confidence interval follows the same shape:

confidence interval = point estimate plus or minus (critical value times standard error)

There are three moving parts, and most student errors come from mixing them up.

The point estimate

The point estimate is your best single guess from the sample. For a mean it is the sample average x bar. For a proportion it is the share p hat. This is the center of the interval.

The standard error

The standard error measures how much that estimate would bounce around from one random sample to the next. For a mean with known population standard deviation sigma it is sigma divided by the square root of n. Notice the square root of n in the denominator: bigger samples give a smaller standard error, which is the whole reason large surveys report tight ranges.

The critical value

The critical value comes from a distribution and sets how many standard errors you step out from the center. You use z when the sample is large or the spread is known, and t when the spread is unknown and the sample is small. The Boston University School of Public Health walks through this z versus t choice for health data, where small samples are common Boston University.

When sigma is unknown, replace it with the sample standard deviation s and use the t critical value from a table with n minus 1 degrees of freedom. The NIST engineering statistics handbook keeps the full set of t tables if you need values your class does not hand you NIST.

Choosing Between z and t

The z critical values most courses use are fixed numbers worth memorizing:

• 90 percent confidence: z = 1.645
• 95 percent confidence: z = 1.96
• 99 percent confidence: z = 2.576

Higher confidence requires a wider interval. A 99 percent interval is wider than a 95 percent interval because it must cover more possibilities. There is a direct trade off: more confidence costs you precision. You cannot make an interval both wider and narrower at the same time, so the level you pick reflects how much error you can tolerate.

For small samples, the t value is larger than the matching z value, which widens the interval to account for the extra uncertainty of estimating the spread from the data. As n grows past 30, the two values converge, which is why many intro courses let you use z for big samples without much penalty.

A Worked Example

Suppose you measure the heights of 100 students and find a sample mean of 168 centimeters with a known population standard deviation of 10 centimeters. Build a 95 percent interval.

Standard error = sigma divided by sqrt(n) = 10 divided by sqrt(100) = 10 divided by 10 = 1.

Margin of error = 1.96 times 1 = 1.96.

Interval = 168 plus or minus 1.96 = (166.04, 169.96).

Reading the result aloud

You would say: we are 95 percent confident the true mean height lies between 166.04 and 169.96 centimeters. The word confident matters. Say it that way and you avoid the mistake in the next section.

What 95 Percent Really Means

The interval either contains the true mean or it does not. After you compute it, the probability is not 95 percent in the sense of a coin flip on that one interval. The 95 percent refers to the long run method: 95 percent of intervals built this way capture the truth.

The repeated sampling view

Picture building 100 separate intervals from 100 separate samples, each using the same recipe. About 95 of those 100 intervals would trap the true mean. About 5 would miss it. You do not know which group your single interval falls into, which is exactly why you phrase the claim as confidence in the method, not as a probability attached to the number.

A single interval is fixed

Once calculated, (166.04, 169.96) is a fixed pair of numbers. The parameter is also fixed. Attaching "95 percent probability" to that pair treats a fixed quantity as if it were still random. Keep the randomness on the procedure, not the result.

Confidence Intervals and Hypothesis Tests

A confidence interval and a two sided hypothesis test answer related questions. If a 95 percent interval for a mean excludes 500, then a two sided test of H0: mu = 500 at alpha = 0.05 would reject H0. If 500 sits inside the interval, you would not reject it. This link is why many courses teach both together, and why my post on hypothesis testing pairs well with this one.

Sample size and the margin of error

The margin of error shrinks when the sample size grows, because sqrt(n) sits in the denominator. Quadrupling the sample size roughly halves the margin of error, which is why larger surveys report tighter intervals. The flip side is real: a tiny sample gives a wide band, and a wide band may contain so many values that it tells you little.

Interval Versus Prediction

Students sometimes confuse a confidence interval for a mean with a prediction interval for a single new observation. A confidence interval narrows as the sample grows because you are estimating one fixed number, the mean. A prediction interval stays wider because a single future measurement carries its own natural scatter on top of the uncertainty about the mean. Both use the same arithmetic family, but they answer different questions, and mixing them up overstates or understates how precise a forecast really is.

A check with software

When you let a calculator or spreadsheet build the interval, read the output rather than trusting the number. Confirm the sample size, the standard error, and whether the tool used z or t. A wrong choice of distribution is the most common silent error in output, because the interval still prints and still looks reasonable. The formula work by hand is what trains your eye to catch it.

Common Misconceptions

• Saying there is a 95 percent probability the parameter is in the interval. The parameter is fixed; the procedure is what has the 95 percent property.
• Using z when a small sample needs the t distribution. Check whether the spread was given or estimated.
• Dropping the plus or minus and reporting only one number. A point estimate alone hides the uncertainty.
• Thinking a wider interval is wrong. It just reflects more caution or a higher confidence level.
• Confusing standard deviation, the spread of the raw data, with standard error, the spread of the estimate.
• Believing a 99 percent interval is "better" than a 95 percent interval. It is safer but less precise, and the right level depends on the cost of being wrong.

Frequently Asked Questions

Why is 1.96 the z value for 95 percent?

In a standard normal distribution, 95 percent of the area lies between minus 1.96 and plus 1.96, leaving 2.5 percent in each tail.

When do I use t instead of z?

Use t when the population standard deviation is unknown and the sample is small (often n under 30). For large samples, t and z are nearly the same.

Does a 99 percent interval always contain the true value?

No. About 99 percent of such intervals do across many samples; any single interval may still miss.

What is the margin of error?

It is the critical value times the standard error. It is the half width added and subtracted from the point estimate.

How does sample size change the interval?

Larger samples shrink the standard error, which narrows the interval and gives a more precise estimate.

Why does my textbook say fail to reject instead of accept?

Because a sample can only supply limited evidence. Not rejecting is not proof the null is true, just that the data did not contradict it enough.

Sources

• Boston University School of Public Health, confidence intervals module
• NIST/SEMATECH e-Handbook, interval estimation
• Khan Academy, confidence intervals