Say I'm doing a confidence interval for a proportion $p$. What if I make the assumption that the sample is biased in a systematic way, by at most 10% say (ad-hoc/arbitrary choice) -- i.e. that the sample comes from a population whose proportion is in the range $(p/1.1, 1.1p)$.

Then I think the confidence interval would just be $(l/1.1, 1.1r)$ in place of $(l, r)$.

This feels quite silly. It may even be incoherent (let me know if so).

But yeah, in practice, I might want to use a confidence interval to express uncertainty about some proportion -- and I know that my sample might be slightly biased, but I don't think it's super biased. And I want to make sure this bias uncertainty doesn't get lost when I report an interval.

Is there some better way to go about this?

Really I think my main purpose is to

(1) express an uncertain quantitative estimate of a proportion (so I naturally think conf intervals)

(2) guard against people interpreting narrow confidence intervals as more than they are

Is this a good approach for that?

When comparing multiple groups, reporting effect size maybe accomplishes this kind of thing.



Your Answer

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

Browse other questions tagged or ask your own question.