As the title implies, why can't $ 1 - \alpha = 1$ , so that we have a $100 \%$ confidence level?

I saw these two answers, but want a more mathematical proof:

How to estimate 100% confidence interval aka. what is the Z value of standard normal distribution at probability of 100%?

100% confidence interval for mean


1 Answer 1


We have such an interval, it is just less than useful.

The confidence interval for the mean is $\bar{x} \pm z_{1-\alpha/2} \sigma/\sqrt{n}$.

As $\alpha \rightarrow 0$, $z_{1-\alpha/2} \rightarrow z_1$ (I'm being a bit fast and loose with my notation).

$z_k$ is the kth quantile of a standard normal. What would be the 100th quantile of a standard normal (equivalently, the point where there is 100% probability to the left of said point). That would be $\infty$ (again, being fast and loose for the purposes of exposition).

So your 100% CI exists and it is the real line. Well...that isn't useful.

  • $\begingroup$ Agreed, with a pedantic generalisation to the entire outcome space, which may not be the real line. E.g. the 100% confidence limits for a proportion are 0 and 1. (The title of the question is general, for all the focus on normal distributions and/or sample means broadly.) $\endgroup$
    – Nick Cox
    Commented Oct 23, 2020 at 14:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.