What this calculator does
This tool estimates the range in which a "true value" is likely to lie, given a sample mean (\(\mu\)) and a standard deviation (\(\sigma\)). It assumes the quantity follows a normal distribution and reports the symmetric two-sided interval \(\mu \pm z\cdot\sigma\), where z is the standard-normal critical value for the confidence level you choose. Mean and standard deviation are plain numbers, and the interval is returned in whatever units you supplied.
How to use it
Enter your mean, your standard deviation (must be 0 or greater), and a confidence level as a percent where \(50 < \text{confidence} < 100\) (commonly 95). The calculator converts the percent to a probability p, finds the critical value \(z = \Phi^{-1}\!\left(\frac{1+p}{2}\right)\), and returns the lower and upper bounds. At 100% the interval would be infinite, so the confidence must stay below 100.
The formula explained
For probability \(p = \text{confidence}/100\), the two-sided z value is the standard normal quantile at \((1+p)/2\). For 95% this is \(\Phi^{-1}(0.975) \approx 1.95996\), so the interval is \(\mu \pm 1.96\sigma\). Useful landmarks: 68.26% coverage equals \(\mu \pm 1\sigma\), 95.45% equals \(\mu \pm 2\sigma\), and 99.73% equals \(\mu \pm 3\sigma\). The inverse normal CDF is computed with a rational (Acklam) approximation plus a Halley refinement step, so no lookup table is needed.
$$\text{CI} = \text{Mean }\mu \pm z \cdot \text{SD }\sigma \\[1.5em] \text{where}\quad z = \Phi^{-1}\!\left(\frac{1 + \frac{\text{Confidence }\%}{100}}{2}\right)$$
Worked example
With mean = 100, sd = 5 and confidence = 99%: \(p = 0.99\), \(z = \Phi^{-1}(0.995) \approx 2.57583\).
$$\text{Lower bound} = 100 - 2.57583\cdot 5 = 87.121$$$$\text{upper bound} = 100 + 12.879 = 112.879$$The result reads "between 87.12 and 112.88".
Common Confidence Levels and Z-Values
The two-sided critical value is \(z = \Phi^{-1}\left(\frac{1+p}{2}\right)\), where \(p\) is the confidence level written as a proportion. The table lists the cumulative probability \((1+p)/2\) used for each level and the resulting z for a standard normal distribution.
| Confidence level | Tail probability (1+p)/2 | Two-sided z |
|---|---|---|
| 50% | 0.75 | 0.6745 |
| 68.27% | 0.84135 | 1.0000 |
| 80% | 0.90 | 1.2816 |
| 90% | 0.95 | 1.6449 |
| 95% | 0.975 | 1.9600 |
| 95.45% | 0.97725 | 2.0000 |
| 98% | 0.99 | 2.3263 |
| 99% | 0.995 | 2.5758 |
| 99.73% | 0.99865 | 3.0000 |
| 99.9% | 0.9995 | 3.2905 |
The 68.27%, 95.45% and 99.73% rows correspond to the familiar 1, 2 and 3 standard-deviation rule.
Interpreting the Interval
This calculator returns \(\text{Mean} \pm z \cdot \text{SD}\). Under a normal distribution, this is the range expected to contain the stated proportion of individual values. For example, with a mean of 100 and an SD of 15 at 95% confidence, the interval is \(100 \pm 1.96 \times 15 = 100 \pm 29.4\), or 70.6 to 129.4. About 95% of values are expected to fall in that range.
Not the same as a confidence interval for the mean. A confidence interval for a population mean uses the standard error, \(\text{SD}/\sqrt{n}\), in place of SD. That interval is much narrower and shrinks as the sample size \(n\) grows. The interval here describes the spread of individual values, or the likely range of a single measurement. It does not describe how precisely the mean is known.
- Widening confidence widens the interval. The half-width equals \(z \cdot \text{SD}\), so it grows in proportion to z. Going from 95% (z = 1.9600) to 99% (z = 2.5758) makes the interval about 31% wider.
- Symmetry. The interval is centered on the mean, with equal distance above and below.
- Assumptions. The result is valid only if the data are approximately normally distributed and the mean and SD are reliable (treated as known). With skewed data, heavy tails, outliers or a small sample, the actual coverage can differ noticeably from the stated level.
Key Terms
- Mean (μ)
- The average value, which is the center of the distribution and of the interval.
- Standard deviation (σ)
- A measure of how widely values spread around the mean, in the same units as the data.
- Confidence level
- The proportion of values (or the probability) the interval is meant to cover, such as 95%.
- Critical value z
- The number of standard deviations from the mean needed to enclose the chosen confidence level; for 95% it is 1.9600.
- Normal distribution
- The symmetric, bell-shaped distribution fully described by its mean and standard deviation.
- Inverse CDF / quantile (Φ⁻¹)
- The function that takes a cumulative probability and returns the z-value below which that probability lies. For example, Φ⁻¹(0.975) = 1.9600.
- Two-sided vs one-sided interval
- A two-sided interval leaves equal tails on both ends, which is why the tail probability is (1+p)/2. A one-sided bound puts all the excluded probability in one tail and uses Φ⁻¹(p) instead.
- Margin of error (z·σ)
- The half-width of the interval, which is the distance from the mean to either endpoint.
FAQ
Why does it use 1.96 for 95%, not 1.645? A symmetric two-sided interval splits the remaining 5% into two tails of 2.5% each, giving \(\Phi^{-1}(0.975) \approx 1.96\). The value 1.645 is the one-sided 95% quantile and is not correct for a two-sided range.
Should I use the t-distribution instead? When \(\sigma\) is estimated from a small sample, a t-distribution with sample size n (interval \(\text{mean} \pm t\cdot s/\sqrt{n}\)) is more appropriate. This tool deliberately treats \(\sigma\) as a known population value and uses the normal z, so it does not require n.
What if sd = 0? The interval collapses to a single point \([\text{mean}, \text{mean}]\).