For a two-sided 95% confidence interval based on the standard normal distribution, the critical value is 1.96. But there is no single value for every 95% confidence interval: with a small sample and an estimated standard deviation, use a critical value from the t-distribution, chosen using the degrees of freedom.
What a 95% critical value means
A critical value sets how many standard errors to extend from an estimate when constructing a confidence interval. In the common two-sided normal case, the interval has the form estimate ± critical value × standard error. The value 1.96 leaves 2.5% of the standard normal distribution in each tail, so the central area between −1.96 and +1.96 is 95%.
“95% confidence” describes the long-run performance of the method: if the same sampling process were repeated many times and an interval were built each time, about 95% of those intervals would contain the fixed population parameter. It does not mean that there is a 95% probability that a particular interval contains the parameter, under the usual frequentist interpretation.
How to choose the right critical value
Before looking up a number, identify the confidence level, whether the interval is one- or two-sided, and which distribution applies. The 1.96 value assumes a normal-based calculation; it is not automatically the right choice just because the confidence level is 95%.
- Use z = 1.96 for a two-sided 95% interval when the standard normal method is appropriate, such as when a population standard deviation is known or a suitable large-sample approximation is being used.
- Use a t critical value for a mean when the population standard deviation is unknown and is estimated from the sample. Choose it using the degrees of freedom; for a one-sample mean, these are typically sample size minus one.
- Use the method specified for the statistic for proportions, regression coefficients, variances, or other quantities. Their interval procedures can use different distributions, approximations, or methods.
In a two-sided interval, a 95% confidence level means α = 1 − 0.95 = 0.05, split between two tails: α/2 = 0.025 per tail. A one-sided 95% bound instead puts all 0.05 in one tail, so its normal critical value is about 1.645, not 1.96.
Common 95% critical values
For a two-sided interval using a t-distribution with ν degrees of freedom, the critical value is the t quantile with cumulative probability 0.975. It is larger than 1.96 for finite degrees of freedom and approaches 1.96 as the degrees of freedom increase.
- Standard normal (z): 1.960
- t with 1 degree of freedom: 12.706
- t with 5 degrees of freedom: 2.571
- t with 10 degrees of freedom: 2.228
- t with 30 degrees of freedom: 2.042
These t values are rounded; software or tables may show slightly different last digits. The very large t value for one degree of freedom illustrates why a small sample can produce a wide interval when its variability must be estimated from the data.
Example: building a confidence interval
Suppose a sample mean is 50 and its standard error is 2. If the normal critical value is appropriate, a two-sided 95% interval is 50 ± 1.96 × 2, or approximately 46.08 to 53.92. The margin of error is 3.92.
If these figures come from a one-sample analysis with an unknown population standard deviation and 10 degrees of freedom, use t ≈ 2.228 instead. The interval becomes 50 ± 2.228 × 2, or about 45.54 to 54.46. The larger margin reflects the additional uncertainty from estimating the standard deviation. The sample mean and standard error alone do not tell you which critical value is appropriate; the sampling setup matters.
Why the critical value changes
Confidence level and tail structure determine which part of a reference distribution is used. Raising the confidence level requires capturing more of that distribution, which increases the critical value and widens the interval, all else equal. A two-sided interval also differs from a one-sided bound because its error probability is allocated differently.
The choice of distribution matters too. The t-distribution has heavier tails than the standard normal distribution, especially with few degrees of freedom. As sample information grows, its critical values move closer to the corresponding z value. A larger critical value does not change the estimate or standard error; it changes the multiplier applied to the standard error.
Common mistakes and misconceptions
- Using 1.96 for every problem: This can make a small-sample mean interval too narrow when a t method is required.
- Confusing 1.96 with 1.645: The former is for a two-sided 95% normal interval; the latter is for a one-sided 95% normal bound.
- Thinking 95% means a 95% chance for this interval: In the conventional frequentist framework, the confidence level refers to the repeated-sampling procedure.
- Assuming the critical value alone determines interval quality: The data, standard error, assumptions, and interval method matter as well. A correct multiplier cannot repair a biased sample or an unsuitable model.
For proportions in particular, avoid automatically applying the simple normal formula when the sample is small or the observed proportion is close to 0 or 1. The usual approximation may perform poorly in those settings; a method designed for binomial data, such as a Wilson interval, may be more suitable.
Frequently Asked Questions
What is the critical value for a 95% confidence interval?
For a two-sided interval using the standard normal distribution, it is 1.96. A t-based interval has a different value determined by its degrees of freedom.
Is the 95% z critical value 1.96 or 1.645?
It is 1.96 for a two-sided 95% interval. The value 1.645 is used for a one-sided 95% normal bound, or equivalently a two-sided 90% normal interval.
How do I find a 95% t critical value?
Find the degrees of freedom, then look up the t quantile at cumulative probability 0.975 for a two-sided interval. In software, request the 97.5th percentile of the t-distribution with those degrees of freedom.
Does a higher confidence level make the interval wider?
Yes, when the estimate, standard error, and method stay the same. A higher confidence level requires a larger critical value, increasing the margin of error.
Can I use 1.96 for a sample proportion?
Sometimes, if a normal approximation is appropriate and the chosen interval method supports it. With small samples or extreme proportions, consider a binomial interval method instead.
