Nonresponse is a common problem in surveys. For example, if a sample size of 1000 is planned in a particular state but in that state only 600 responses are obtained, if we are prepared to believe that...

Nonresponse is a common problem in surveys. For example, if a sample size of 1000 is planned in a particular state but in that state only 600 responses are obtained, if we are prepared to believe that the responders in the subpopulation are no different from the non-responders, then we can weight the responders to represent the non-responders. In combining data over subpopulations, should larger weight be given to observations that represent the non-responders in the subpopulation or smaller weight?