0
$\begingroup$

I want to calculate the N% confidence interval for some time-series data set. I have the standard errors for this data series and the error variance of the time-series is assumed to be Gaussian. I understand that we can use a factor of ±1.96 multiplied by the standard error of each data point to obtain the 95% confidence interval (because it is the quantile 0.975 of the Gaussian distribution) [the exact specifics of the standard errors is not important].

I know I am able to get this value of 1.96 by calculating the Inverse Cumulative Distribution Function for the Normal distribution (`InvCDF(0.975) ~ 1.96), what I don't understand is how to map the quartile value (0.975) to the desired confidence interval. What I want is to be able to say is: "I want the 87% confidence interval" and then calculate InvCDF(X), but what is X in this case? And what function relates the quantile value to the confidence interval?

Thanks for your time.

$\endgroup$
  • $\begingroup$ Are you asking about how one obtains $0.975$ from the 95% specification? $\endgroup$ – whuber Jan 7 '15 at 17:28
  • $\begingroup$ Yes. That is exactly why I am asking... $\endgroup$ – MoonKnight Jan 7 '15 at 17:55
2
$\begingroup$

For a two-sided confidence interval, a N% confidence interval requires the quantile 1-(1-N)/2 for the upper bound of the interval and (1-N)/2 for the lower bound. In this way, the (1-N)% of the mass is split symmetrically in both tails of the normal distribution. For a 87% confidence interval you would have 1.87/2=0.935 and (1-0.87)/2=0.065 and the quantile values are ± 1.514.

$\endgroup$

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.