I have two related questions:

Assuming (as the extreme case) I measured a quantity (A) just twice, but I know that the values would "become" normally distributed if I measured the quantity repeatedly. I can calculate the CI for the mean by multiplying the standard deviation of the mean by 12.71 (see here).

I also measured another quantity (B), this time with a higher number of repeats (e.g. 1000x), so that I don't have to correct for bias and can use the usual factor of 1.96 to get the CI.

What about the CI of a derived quantity A+B (or in general f(A,B))? I know how to propagate the standard deviations, but when I calculate the CI from this propagated standard deviation (let's assume the derived quantity is also normally distributed), what's the sample number I have to use for the Student t to calculate the CI? 2? 1000? 10002? Something else? Do I have to propagate the CIs in another way in such a case (how?)? Is the Student's t method to get the CIs not appropriate in this case?

The second question: If I measured two quantities 1000x, and both are sampled from normal distributions with the same mean but different standard deviations. It's easy enough to calculate the weighted mean & corresponding variance/standard deviation as it is done in fixed effects meta analysis (see here). Again the same question - what sample number do I have to use to calculate the CI from the weighted mean standard deviation? 2? 1000? 2000?



Your Answer

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