Skip to main content
8 events
when toggle format what by license comment
Jul 19, 2020 at 20:24 comment added Sextus Empiricus It should be stressed that the tails considered here are the tails of the distribution of the data conditional on the parameter to be estimated. For instance we can use 2.5% tails and symmetry on both ends (north south) like estimating the parameter of a weird distribution here$${\hat\theta \sim \mathcal{N}(\mu=\theta, \sigma^2=1+\theta^2/3)}$$ but it does not necessarily give symmetric confidence intervals. In that particular example of the link one may end up with one of the boundaries at infinity.
Jul 19, 2020 at 0:18 comment added BruceET Addendum to my Answer shows various lengths of the CI for $\sigma$ depending on the part of the 5% 'error probability' is allocated to the lower tail of the chi-sq dist'n.
Jul 18, 2020 at 23:58 history edited BruceET CC BY-SA 4.0
added 664 characters in body
Jul 18, 2020 at 21:42 comment added BruceET I did not mean my examples to give the impression that lengths of CIs are 'not much different' depending on allocation of 5% probability to the two tails. // Roughly speaking, shorter CIs result when cut-off points straddle the part of a dist'n with highest density (clearly the case of the CIs for $\mu.)$ But I guess, for my CIs for $\sigma,$ the relevant dist'n may be inverse chi-sq, not chi-sq.
Jul 18, 2020 at 21:32 history edited BruceET CC BY-SA 4.0
deleted 1 character in body
Jul 18, 2020 at 20:13 comment added Victor Luu Based on @BruceET answer, it seems that there is no much difference between symmetric and non-symmetric tails. But are there some cases when there is a big difference in CI lengths which may call for the need of non-symmetric tails?
Jul 18, 2020 at 19:59 comment added Victor Luu Very detailed and clear answer, thanks.
Jul 18, 2020 at 19:44 history answered BruceET CC BY-SA 4.0