Confidence interval obtained by Hoeffding's inequality is the "worst-case''?

51 Views Asked by At

Hoeffding's inequality can be used to generate a confidence interval.

I wonder if Hoeffding-inequality is the "worst-case" approach (among all other common approaches in textbooks, e.g. assuming the sample mean is a normal distribution etc.) in the sense that the confidence interval is the "widest".

My intutive guess is: since the proof of the Hoeffding-inequality doesn't rely on any thing like the central limit theorem, it should be less useful than other approaches with those extra assumptions.

Am I right? Or being naive here..