A confidence interval is one of the most commonly misread pieces of statistics, not because the formula is hard, but because the wording of what it actually promises is easy to get subtly wrong. This works through the margin of error, what the confidence level really means, when to use the t-distribution instead of the normal distribution, and what actually widens or narrows an interval, all with real numbers.
In this guide
What a confidence interval is, and what it is not
A confidence interval is a range of plausible values, built from sample data, meant to contain the true population parameter. A 95 percent confidence level means that if the same sampling method were repeated many times, about 95 percent of the resulting intervals would contain the true parameter, a statement about the long-run reliability of the method, not a probability attached to any one already-calculated interval.
Once an interval is calculated, the true value either is or is not inside it; there is no remaining 95 percent chance about that specific range. The 95 percent describes how often the method succeeds across repeated sampling, before any one interval is built.
The margin of error
The margin of error is the amount added to and subtracted from the point estimate to form the interval’s two bounds, equal to a critical value times the standard error.
| Quantity | Calculation | Result |
|---|---|---|
| Margin of error (mean 50, SE 2.0, Z = 1.96) | 1.96 x 2.0 | 3.92 |
| Lower bound | 50 minus 3.92 | 46.08 |
| Upper bound (mean 100, SE 2.5, Z = 2.576) | 100 + 2.576 x 2.5 | 106.44 |
| Lower bound (same data) | 100 minus 2.576 x 2.5 | 93.56 |
What widens or narrows an interval
Three things mainly affect the width of a confidence interval: the chosen confidence level, the sample size, and how much variability is in the underlying data.
| Confidence level | Z critical value | Total width (SE = 5) |
|---|---|---|
| 90% | 1.645 | 16.4 |
| 99% | 2.576 | 25.8 |
Raising the confidence level from 95 to 99 percent, with everything else unchanged, makes the interval wider: being more confident of capturing the true value requires a larger safety margin. Sample size works the opposite way: increasing it narrows the interval, since the standard error shrinks as the sample grows. But that shrinkage follows the square root of n, not n directly, so multiplying the sample size by 4 only roughly halves the width, not quarters it.
When to use the t-distribution instead of the normal distribution
The t-distribution is used instead of the normal distribution when the sample size is small and the population standard deviation is unknown, its heavier tails account for the extra uncertainty of estimating the standard deviation itself from a small sample. It uses n minus 1 degrees of freedom, one is lost because the sample mean, used to estimate the standard deviation, is calculated from the same data.
| Quantity | Value |
|---|---|
| t critical value, 9 degrees of freedom, 95% confidence | 2.262 (versus the normal distribution's 1.96) |
| Margin of error (n=10, sample SD 4, t = 2.262) | 2.262 x 4 / sqrt(10) = 2.86 |
As the sample size grows larger, the t-distribution’s heavier tails shrink down and it converges toward the standard normal distribution, which is why the two are nearly identical once the sample is large.
Confidence intervals for a proportion
A confidence interval for a proportion uses a different standard error formula: the square root of p times (1 minus p), divided by n, built directly from the sample proportion itself rather than a separately measured standard deviation.
A poll’s margin of error
A poll of 100 people finds that 60 percent support a proposal. The standard error is the square root of 0.6 x 0.4 / 100, which is 0.049; multiplying by the Z critical value of 1.96 gives a margin of error of 0.096, about plus or minus 9.6 percentage points.
Reading an interval correctly
| Question | Correct reading |
|---|---|
| Does a 95% CI for the mean contain 95% of the individual data points? | No, that describes the empirical rule or a prediction interval, not a CI for the mean |
| Does a narrower interval always mean a more accurate estimate? | No, it shows more precision, but a biased sample can still produce a tight interval centred on the wrong value |
| Is the confidence level a property of one calculated interval? | No, it is a property of the method, reflecting how often it succeeds across repeated sampling |
A reported interval of -3 to 8 for the difference between two group means is a good example of reading correctly: because the interval spans both negative and positive values, including zero, zero difference is a plausible value, so there is not strong evidence the two group means actually differ, even though the point estimate itself might sit well above zero.
Common mistakes
- Saying there is a 95 percent chance the true value is in this one interval. The 95 percent describes the method’s long-run success rate, not a probability about this specific, already-fixed range.
- Using the normal distribution’s Z critical value for a small sample with an unknown population standard deviation. That calls for the t-distribution, which has a larger critical value to account for the extra uncertainty.
- Assuming a narrower interval is automatically more accurate. Precision and accuracy are different; a biased sample can produce a tight interval that is confidently wrong.
- Forgetting the square-root relationship between sample size and width. Quadrupling the sample only halves the width, not quarters it.
- Treating a wide interval that includes zero as proof of no effect. It means zero is plausible and the evidence for a real difference is not strong, which is a more careful claim than declaring the effect definitely does not exist.
Cheat sheet
| Question | Answer |
|---|---|
| What does a 95% confidence level describe? | The long-run success rate of the method, not one specific interval |
| What is the margin of error formula? | Critical value times the standard error |
| What three things mainly affect interval width? | Confidence level, sample size, and data variability |
| Does raising the confidence level widen or narrow the interval? | Widen |
| Does increasing the sample size widen or narrow the interval? | Narrow |
| When should the t-distribution be used instead of the normal distribution? | Small sample, unknown population standard deviation |
| What degrees of freedom does a t-interval for a mean use? | n minus 1 |
| What standard error formula does a proportion's interval use? | sqrt(p(1-p)/n) |
| Does a CI for the mean contain 95% of the raw data points? | No, that is a different idea entirely |
| What is a confidence interval always centred on? | The sample's own point estimate |
Try it yourself
Quick check
A sample of size 36 has a standard deviation of 18. What is the margin of error for a 95 percent confidence interval on the mean, using a Z of 1.96 (2 dp)?
Show answer
5.88. First find the standard error: 18 divided by the square root of 36 is 3.0. The margin of error is then the critical value times the standard error, 1.96 times 3.0, which is 5.88.
Frequently asked questions
What does a 95 percent confidence level actually mean?
If the same sampling method were repeated many times, about 95 percent of the resulting intervals would contain the true population parameter. It is a statement about the method’s long-run reliability, not a probability about one already-calculated interval.
What is the margin of error?
The margin of error is the amount added to and subtracted from a point estimate to form a confidence interval’s bounds, calculated as a critical value multiplied by the standard error.
When should you use the t-distribution instead of the normal distribution for a confidence interval?
Use the t-distribution when the sample size is small and the population standard deviation is unknown. Its heavier tails account for the added uncertainty of estimating the standard deviation from the sample itself.
Does a narrower confidence interval mean a more accurate estimate?
No. A narrow interval shows more precision, but it can still be centred on the wrong value if the underlying sample was biased. Precision and accuracy are different properties.
How does a confidence interval for a proportion differ from one for a mean?
It uses a different standard error formula, the square root of the sample proportion times (1 minus the proportion), divided by n, built directly from the proportion itself rather than a separately measured standard deviation.
If a 95 percent confidence interval for a difference includes zero, does that mean there is no effect?
It means zero is a plausible value given the data, so there is not strong evidence of a real difference. It is not the same as proving the two groups are definitely identical.
Test yourself
Timed questions on Confidence Intervals, with an explanation for every answer.