I am working with a weighted survey dataset where some respondents did not answer specific questions. Each respondent has a weight assigned to them. My goal is to calculate weighted percentages for different responses within each district.

Here’s how I am currently calculating weighted percentages:

Group responses by district and response category. Sum the weights of respondents who provided an answer. Divide by the total weight of all respondents in the district to get the percentage. However, I am unsure whether the denominator (total weight in the district) should include only those who answered the question or all respondents in the district, including those who did not answer.

For example, if 10 respondents are from District A but only 7 answered the question, should the denominator be:

(A) The sum of weights of the 7 respondents who answered (B) The sum of weights of all 10 respondents in the district (including those who skipped) This is important for correctly interpreting missing data. Any insights or references on best practices for handling survey weights in such cases would be greatly appreciated!