I am trying to build confidence interval for mean of a metrics. To build confidence interval whether standard error or standard deviation should be used ?


Use the standard error $SE$ (not the standard deviation $s$) to calculate confidence intervals.

The confidence interval for a population mean is equal to:

$$CI=\bar{X} \pm t_{crit}\times SE_{\bar{X}}$$

$$SE_{\bar{X}}= \frac{s}{\sqrt{n}}$$ where $\bar{X}$ is the sample mean and $n$ is the sample size. The assumption for the formula is standard normal distribution.

  • 1
    $\begingroup$ It is really the population mean that you determine confidence intervals for. In your example you have a confidence interval centered at the sample mean and the interval is exact only for iid normals. $\endgroup$ – Michael R. Chernick Dec 15 '16 at 3:42
  • $\begingroup$ Thanks for the response. Let assume the case - I have access to entire population. In this case is it meaningful to use standard deviation ? (inferential vs descriptive) $\endgroup$ – Naveenan Dec 15 '16 at 3:54
  • $\begingroup$ If you have access to the population parameter, there is no need for estimation using sample statistics. You can just present the population mean and standard deviation. $\endgroup$ – Jeffrey Girard Dec 15 '16 at 4:01
  • $\begingroup$ If by access to the entire population you mean that you have say a normal distribution with the mean and variance known you don't have an inference problem. For a confidence interval to make sense you would possibly know the variance but not the mean. Then the true standard deviation replaces the sample mean and the t crit with the zcrit but the standard error is still used and the square root of n remains in the denominator. $\endgroup$ – Michael R. Chernick Dec 15 '16 at 4:08
  • 1
    $\begingroup$ @JeffreyGirard I see you changed t_crit in your answer to z_crit. I had no objection to assuming normality With the change you need to assume large n and enough central moments for the central limit theorem to apply is you want to make the population distribution more general. Here I am stating the conditions for the Lyapunov version of the central limit theorem. $\endgroup$ – Michael R. Chernick Dec 15 '16 at 9:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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