2
$\begingroup$

I have two related questions:

a) Is there any other way to calculate a confidence interval for the proportion, in addition to the "classical" form?

enter image description here

b) Can we apply this "classical" method when the proportion was calculated from a weighted data base?

I will put in context these questions so you can better understand me.

Reviewing the methodology applied on a survey (http://fra.europa.eu/sites/default/files/fra-2014-vaw-survey-technical-report-1_en.pdf) I found a table showing proportions and their confidence intervals. I calculated the confidence intervals by myself using the "classical" formula, because they looked excesively wide, and they didn't match. The only explanation I can find is that these proportions have been calculated from a weighted base data. But I'm not convinced with this self-explanation, since the "classical" formula takes into account the size of the sample, and that has not changed with the weighting.

Why my confidence intervals don't match with those shown in the table? This question is what led me to ask the two questions above.

Can please anybody help me?

$\endgroup$
1
$\begingroup$

One explanation could be that the survey uses clusters and weights, so effectively your n is smaller now because of the correlation of people within a cluster. That is they are similar people within a group, so not independent. If n is smaller then your confidence intervals are wider.

One good thing about the survey is it has fairly large sample size, as it is at the fringes (i.e. proportions that are zero or 100% or near there) that problems arise with Confidence intervals.

$\endgroup$

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.