90% Confidence Interval Calculator

Article — 90% Confidence Interval Calculator

90% Confidence Interval: When and How to Use It

A 90% confidence interval is calculated as x̄ ± 1.645 × σ/√n, where x̄ is the sample mean, σ is the standard deviation, and n is the sample size. The z-score 1.645 is the cutoff that leaves 5% of the standard normal distribution in each tail. The resulting interval, repeated infinitely, would contain the true population mean 90% of the time.

Confidence intervals are how statisticians answer "where is the true average?" given only a sample. The 90% level is the workhorse for business and industrial decisions, sitting between the looser 80% and the academic 95%. This guide shows the math, the interpretation, the common errors, and when to switch to a different tool.

What is a 90% confidence interval?

A confidence interval is a range of plausible values for an unknown population parameter, computed from sample data. The "90%" refers to the long-run frequency: if you took thousands of samples and built a 90% interval from each, roughly 90% of those intervals would contain the true population mean.

The level is not a probability statement about your specific interval. Once you compute [95.89, 104.11] from a sample, the true mean either lies in that range or it doesn't — there is no "90% probability" attached to your particular numbers. The 90% applies to the procedure, not the result.

The 90% confidence interval formula

The standard z-interval formula for a population mean is simple. Plug in three numbers from your sample and one z-critical value.

Numerical example. A factory measures part lengths from 36 samples. Mean = 100 mm, SD = 15 mm. SE = 15/6 = 2.5. ME = 1.645 × 2.5 = 4.11. The 90% CI is [95.89, 104.11].

90% vs 95% confidence interval

The choice of confidence level is a trade-off. Higher confidence means a wider interval; lower confidence narrows it. 90% versus 95% comes up most often.

If your business is screening prototypes for a yes/no go-decision and you can tolerate slightly more uncertainty, 90% is faster and cheaper. If you're publishing in a peer-reviewed journal or making a medical claim, 95% (or 99%) is expected.

Interpreting confidence intervals

The most common misinterpretation is treating the interval as a probability statement about the unknown mean. It isn't. The correct interpretations are:

• Correct: "If we repeated this study many times, 90% of the resulting intervals would contain the true mean."
• Correct: "We're confident the procedure produces intervals that capture the true mean 90% of the time."
• Incorrect: "There's a 90% probability that the true mean is in [95.89, 104.11]." (Frequentist statistics doesn't attach probability to fixed parameters.)
• Incorrect: "90% of the sample data falls in this interval." (CI is about the mean, not individual values.)

Sample size and CI precision

The width of the CI shrinks with √n. To halve your margin of error, you need to quadruple your sample size. This is the single most important practical fact about CIs.

Working backwards is useful for study design. To achieve a 90% margin of error of ±1 unit with σ = 10, you need n = (1.645 × 10 / 1)² ≈ 271 samples.

z- vs t-confidence interval

The z-formula assumes you know σ (the true population standard deviation) — which you almost never do. When you only have the sample SD s, switch to the t-distribution and Student's t-interval: x̄ ± t × s/√n with t depending on degrees of freedom df = n - 1.

For large n (≥ 30) the t and z values are nearly identical, so the z-interval is fine. For small samples, t is noticeably larger than z. At n = 10, t for 90% is 1.833 versus z = 1.645 — about 11% wider intervals.

Common confidence interval mistakes

The math is easy. The interpretation is where things go wrong.

• Treating CI as a probability statement about μ — see above.
• Using z when σ is unknown and n is small — should use t.
• Assuming the data is normal — z-interval relies on the CLT; for skewed data with small n, use bootstrap or transformation.
• Confusing CI width with effect size — a tight CI around zero doesn't mean a large effect.
• Reporting CI with sample SD instead of σ — clarify which you used.
• Stacking CIs from sub-groups — combining 90% intervals doesn't give a 90% interval for the combined parameter.

Real-world examples

Three contexts where 90% confidence intervals come up regularly.

Industrial quality control. A line produces ball bearings with target diameter 10 mm. Daily sample of 50 parts: mean 10.02 mm, SD 0.08 mm. 90% CI: 10.02 ± 1.645 × 0.08/√50 = [10.001, 10.039]. If the spec is 10 ± 0.05 mm, the line is fine.

Marketing research. A survey of 200 consumers rates a new product feature on a 1–10 scale. Mean = 7.4, SD = 1.5. 90% CI: 7.4 ± 1.645 × 1.5/√200 = [7.225, 7.575]. A score above 7 with 90% confidence is enough to move forward with the launch decision.

A/B testing. A pricing test shows checkout rate increased from 4.2% to 4.6%, n = 1000 per arm. Using a proportion CI, 90% CI for the lift is roughly ±0.5 percentage points — overlap with zero would mean inconclusive, separation means a real signal.

One nuance worth noting about confidence intervals at any level. The interval gives a range of plausible values for the parameter, but it tells you nothing about the practical importance of any value in that range. A 90% CI of [0.001, 0.003] for a treatment effect is statistically distinguishable from zero, but a 0.002 effect may be useless in practice. Always pair confidence intervals with a minimum effect size that would matter to your decision — statistical significance is a starting point, not the endpoint.