Written and maintained by CASRAI Editorial Board
Published
On this page
A 95 percent confidence interval is a range of values, calculated from sample data, produced by a method that captures the true population parameter in about 95 percent of repeated samples. For a mean with known population standard deviation σ, it is the sample mean plus or minus 1.96 standard errors. It is a statement about the reliability of the method, not a 95 percent probability that this one interval contains the parameter.
Definition: what a 95 percent confidence interval means
The NIST/SEMATECH e-Handbook of Statistical Methods states the point directly: a 95 % confidence interval does not mean that there is a 95 % probability that the interval contains the true mean. The 95 percent describes the procedure. If you drew many random samples from the same population and built an interval from each one in the same way, roughly 95 percent of those intervals would bracket the true population parameter. Any single interval either contains the parameter or it does not; you do not know which.
OpenStax Introductory Statistics uses the same reading: the confidence level is the percentage of intervals that contain the parameter in repeated sampling, and a result is phrased as “we estimate with 95 percent confidence that the true population mean is between the two endpoints.” For the broader concept, see the confidence interval dictionary entry and the full confidence interval guide; this page concentrates on the 95 percent and 90 percent cases.
Formula for a 95 percent confidence interval, with every symbol explained
When the population standard deviation σ is known (or the sample is large and σ is well estimated), the interval for a mean is:
x̄ ± zα/2 × σ / √n
- x̄ is the sample mean, the point estimate at the centre of the interval.
- σ is the population standard deviation. When it is unknown, the sample standard deviation s replaces it and the multiplier becomes a t value (see below).
- n is the sample size, so σ / √n is the standard error of the mean.
- α is 1 minus the confidence level: 0.05 for 95 percent, 0.10 for 90 percent.
- zα/2 is the point on the standard normal curve with area α/2 in the upper tail. It is the critical value, or multiplier.
- zα/2 × σ / √n is the margin of error (OpenStax calls it the error bound for a population mean, EBM). The interval runs from x̄ − EBM to x̄ + EBM.
NIST gives the equivalent form for an unknown σ: Ȳ ± t1−α/2, N−1 × s / √N, where t is the 100(1 − α/2) percentile of the t distribution with N − 1 degrees of freedom. See the guide to the t distribution for how the multiplier changes with sample size.
90 percent confidence interval z-score (z value) versus 95 percent
The 90 percent confidence interval z-score is 1.645, and the 95 percent confidence interval z-score is 1.96. OpenStax derives both from the inverse normal function: for 90 percent, 0.05 lies in each tail and z = invNorm(0.95) = 1.645; for 95 percent, 0.025 lies in each tail and z = invNorm(0.975) = 1.96. The “z value”, “z-score” and “critical value” for a confidence interval all refer to this same multiplier.
| Confidence level | α | Area in each tail (α/2) | Percentile used | z value |
|---|---|---|---|---|
| 90 percent | 0.10 | 0.05 | 0.95 | 1.645 |
| 95 percent | 0.05 | 0.025 | 0.975 | 1.96 |
A higher confidence level needs a bigger multiplier, so the interval is wider. Going from 90 to 95 percent widens the interval by a factor of 1.96 / 1.645 ≈ 1.19, about 19 percent, for the same data. For the related one-tailed value and the notation, see z alpha; for lookups beyond these two levels, see how to find a critical value.
Worked example: 90 and 95 percent intervals from the same data
This example follows the OpenStax single-population-mean problem, with the arithmetic recomputed here. A sample of n = 36 has mean x̄ = 68, and the population standard deviation is σ = 3.
- Standard error: σ / √n = 3 / √36 = 3 / 6 = 0.5.
- 90 percent interval. EBM = 1.645 × 0.5 = 0.8225. Interval = 68 − 0.8225 to 68 + 0.8225 = (67.1775, 68.8225).
- 95 percent interval. EBM = 1.96 × 0.5 = 0.98. Interval = 68 − 0.98 to 68 + 0.98 = (67.02, 68.98).
- Width: the 90 percent interval is 1.645 in total (2 × 0.8225); the 95 percent interval is 1.96 (2 × 0.98). The ratio is 1.96 / 1.645 ≈ 1.19.
The 95 percent statement reads: “We estimate with 95 percent confidence that the true population mean is between 67.02 and 68.98.” The data table below sets these multipliers beside the t multiplier from the NIST example.
How sample size changes the 95 percent interval
Because the standard error is σ / √n, the width scales with 1 / √n. Quadrupling the sample from n = 36 to n = 144 halves it: the standard error becomes 3 / 12 = 0.25, EBM = 1.96 × 0.25 = 0.49, and the 95 percent interval narrows from (67.02, 68.98) to (67.51, 68.49). A larger standard deviation widens the interval; NIST makes the same point for the t-based form.
When the multiplier is t, not z
With σ estimated from the data, the t multiplier replaces z. In the NIST e-Handbook example with N = 195 observations, the 95 percent multiplier is t = 1.9723 on 194 degrees of freedom, slightly larger than 1.96. The gap shrinks as the sample grows, which is why z is a reasonable approximation for large samples.
Interpretation: reading a 95 percent confidence interval
| Statement | Correct? | Why |
|---|---|---|
| “The method that produced this interval captures the true mean in about 95 percent of repeated samples.” | Yes | This is the repeated-sampling definition used by NIST and OpenStax. |
| “There is a 95 percent probability that the true mean lies in this interval.” | No | NIST states that the 95 % does not mean a 95 % probability for the interval; the level belongs to the calculation method. |
| “A 95 percent interval is wider than a 90 percent interval from the same data.” | Yes | The multiplier is 1.96 instead of 1.645. |
| “A narrower interval comes from a larger sample or a smaller standard deviation.” | Yes | The standard error σ / √n shrinks as n grows or σ falls. |
| “95 percent of the individual observations fall inside the interval.” | No | The interval describes the mean, not the spread of individual values. |
Assumptions and common mistakes
- Normality or a large sample. The NIST formula for a known σ assumes a normally distributed population; for other populations a large sample is relied on.
- Known versus estimated σ. Using 1.96 with a small sample whose σ was estimated from the data understates the width; the t multiplier is the correct one.
- Treating the level as a probability. Say “95 percent confident” about the method, not “95 percent probable” about one interval.
- Choosing the level after seeing the data. Pick 90 or 95 percent before analysis, as you would a significance threshold; see confidence level versus confidence interval.
- Dropping the unit or the estimate. An interval with no named statistic attached cannot be interpreted.
How to report a 95 percent confidence interval (APA example)
APA style places the level before “CI” and gives the limits in square brackets separated by a comma, for example: M = 68.00, 95% CI [67.02, 68.98]. State the level once if every interval in the paper uses it. Notation differences between APA and AMA, and how to pair intervals with p-values, are covered in how to report confidence intervals.
How to calculate a 95 percent confidence interval in Excel, R or Python
Each snippet reproduces the example above (x̄ = 68, σ = 3, n = 36). Results agree with the hand calculation to two decimals; software uses the unrounded z (1.95996 and 1.64485), so the last digits of the 90 percent bounds can differ slightly from the 1.645 textbook figure.
Excel
=NORM.S.INV(0.975) -> 1.96 (z for 95%)
=CONFIDENCE.NORM(0.05, 3, 36) -> margin of error, about 0.98
=68 - CONFIDENCE.NORM(0.05,3,36) -> lower limit, about 67.02
=68 + CONFIDENCE.NORM(0.05,3,36) -> upper limit, about 68.98
R
xbar <- 68; sigma <- 3; n <- 36
se <- sigma / sqrt(n)
qnorm(c(0.95, 0.975)) # 1.645 1.960
xbar + c(-1, 1) * qnorm(0.975) * se # 67.02 68.98
Python
from statistics import NormalDist
xbar, sigma, n = 68, 3, 36
se = sigma / n ** 0.5
for level in (0.90, 0.95):
z = NormalDist().inv_cdf(1 - (1 - level) / 2)
print(level, round(z, 3), round(xbar - z * se, 2), round(xbar + z * se, 2))
# 0.9 1.645 67.18 68.82
# 0.95 1.96 67.02 68.98
With SciPy, scipy.stats.norm.interval(0.95, loc=68, scale=0.5) returns the same equal-tail interval, because interval gives the confidence interval with equal areas around the median.
Frequently asked questions
What is a 95 percent confidence interval?
It is a range of plausible values for a population parameter built by a method that captures the true value in about 95 percent of repeated samples. For a mean it is the sample mean plus or minus 1.96 standard errors.
What is the z-score for a 95 percent confidence interval?
1.96, the value with 0.025 of the standard normal area in each tail.
What is the 90 percent confidence interval z-score or z value?
1.645, the value with 0.05 in each tail (the 95th percentile of the standard normal distribution).
Is a 90 percent confidence interval narrower than a 95 percent one?
Yes. With the same data and sample size the 90 percent interval is about 16 percent narrower, because 1.645 is smaller than 1.96. It captures the parameter less often in repeated sampling.
Does 95 percent mean there is a 95 percent chance the true value is inside?
No. The level describes the long-run performance of the method, per NIST; a computed interval either contains the true value or does not.
Should I use z or t for a 95 percent interval?
Use z when σ is known; use t with N − 1 degrees of freedom when σ is estimated from the sample, as in the NIST example (t = 1.9723 at 194 degrees of freedom).
Sources
- NIST/SEMATECH e-Handbook of Statistical Methods, section 1.3.5.2, Confidence Limits for the Mean: itl.nist.gov/div898/handbook/eda/section3/eda352.htm (read 2026-11-12).
- NIST/SEMATECH e-Handbook, section 7.2.4, Does the true mean of a population differ from a specified value (confidence interval for a mean): itl.nist.gov/div898/handbook/prc/section1/prc14.htm (read 2026-11-12).
- OpenStax, Introductory Statistics, 8.1 A Single Population Mean using the Normal Distribution: openstax.org (read 2026-11-12).
- SciPy documentation, scipy.stats.norm: docs.scipy.org (read 2026-11-12).
- R documentation, The Normal Distribution (qnorm): stat.ethz.ch (read 2026-11-12).
Data: Critical values used for 90 and 95 percent confidence intervals for a mean
Multipliers used in confidence intervals for a mean, compiled from the OpenStax normal-distribution worked solution and the NIST e-Handbook t-based example: confidence level, tail area, distribution and critical value.
| Interval | Confidence level (percent) | Area in each tail | Distribution | Critical value |
|---|---|---|---|---|
| 90 percent, known sigma | 90 | 0.05 | standard normal (z) | 1.645 |
| 95 percent, known sigma | 95 | 0.025 | standard normal (z) | 1.96 |
| 95 percent, sigma estimated, N=195 (194 df) | 95 | 0.025 | Student t | 1.9723 |
Sources
- OpenStax Introductory Statistics 8.1 (accessed 2026-11-12)
- NIST/SEMATECH e-Handbook 1.3.5.2 Confidence Limits for the Mean (accessed 2026-11-12)
Compiled by CASRAI from the sources above. Free to reuse under CC BY 4.0 with attribution.







