Statistics Reference

Confidence Level vs Confidence Interval

One is a promise about the method; the other is what the method produced from your data. Mixing them up creates the most common misreadings in statistics — “there's a 95% chance the truth is in here” among them. This guide separates the two with one dataset computed at three levels.

Setting vs Output

 Confidence levelConfidence interval
What it isA percentage you choose — the long-run capture rate you demandA range the data produces — estimate ± margin of error
DecidedBefore seeing the dataAfter computing from the sample
Belongs toThe procedure (all its repeated uses)This one sample
Example95%(48.04, 51.96)

The connection runs through the critical value: the level picks it (1.96 for 95%), and the interval is estimate ± critical value × standard error — the anatomy covered in the margin-of-error guide.

One Dataset, Three Levels

Sample: mean 50, standard deviation 10, n = 100 — so the standard error is 10/√100 = 1.

LevelCritical valueIntervalLong-run miss rate
90%1.645(48.36, 51.65)1 in 10
95%1.960(48.04, 51.96)1 in 20
99%2.576(47.42, 52.58)1 in 100

Same data, three different claims. Raising the level never moves the center — it only stretches the range, trading precision for reliability. Reproduce any row with the confidence interval calculator.

The Two Legitimate Ways to Narrow an Interval

  • More data. Width scales with 1/√n: quadrupling the sample halves the interval at the same level — the honest route, planned with the sample size calculator.
  • Less noise. Better instruments or designs shrink the standard deviation itself.

Lowering the level also narrows the interval — but that is weakening the promise, not strengthening the evidence. A suspiciously tight interval quoted at 80% confidence is the classic form of this trick; always check the level before admiring the width. The margin of error calculator makes the level-width dependence explicit.

Getting the Words Right

Correct

“We are 95% confident the mean lies between 48.04 and 51.96” — shorthand for: the procedure that produced this interval captures the truth 95% of the time.

Incorrect

“95% of the data falls in this interval” (that confuses it with a spread statement) or “the mean has a 95% probability of being in here” (the frequentist framework assigns no probability to a computed interval).

The deeper interpretation questions — what the long-run reading really commits you to, and z vs t choices — are covered in the full confidence intervals guide.

Try the Confidence Interval Calculator

Compute the interval at 90%, 95%, or 99% and watch the level-width trade-off on your own data.

Frequently Asked Questions

What is the difference between confidence level and confidence interval?

The confidence level is a percentage you choose before analyzing - the long-run success rate you demand from the procedure (usually 95%). The confidence interval is the specific range the procedure then computes from your sample (like 48.04 to 51.96). Level is the setting; interval is the output. Two studies can share a level and produce completely different intervals.

Does a 95% confidence level mean my interval has a 95% chance of being right?

Not quite - the 95% belongs to the method, not to your one interval. Across many repeated studies, 95% of the intervals built this way capture the truth. Your specific interval either contains the parameter or it does not; there is no probability left once it is computed. The practical reading: you hold one ticket from a batch where 95% are winners.

What happens to the interval when I raise the confidence level?

It widens - certainty is bought with precision. With the same data (mean 50, SD 10, n = 100), the interval is +/-1.645 at 90%, +/-1.96 at 95%, and +/-2.576 at 99%. The width scales with the critical value, so 99% intervals are about 31% wider than 95% ones. An interval wide enough to be nearly certain can be too wide to be useful.

Which should I report - the level or the interval?

Both, always together: '95% CI: 48.0 to 52.0'. The interval without its level is uninterpretable (48-52 at 80% confidence is a much weaker claim than at 99%), and the level without the interval says nothing about your data. Reporting the sample size alongside completes the picture.

Can I narrow the interval without lowering the confidence level?

Yes - collect more data or reduce noise. Width is (critical value) x (standard error), and the standard error shrinks with the square root of n: quadrupling the sample halves the width at any level. Lowering the level narrows the interval too, but by weakening the claim rather than improving the evidence.

Is a 99% confidence interval always better than a 95% one?

No - it is a different trade, not an upgrade. The 99% interval misses the truth less often but says something vaguer about where it is. Fields settle on 95% as a working balance; use 99% when the cost of a wrong interval is high, and 90% when a tighter range matters more than occasional misses. The choice should be made before seeing the data.