Say I have a target population of 10,000 for a survey, and only 2,000 respond, because they likely feel strongly about the survey topic (either happy or angry). I have clear, abundant non-response bias.

Say one of the questions is to rate something from 1-5, let's call this random variable $y$. How can I best adjust for non-response bias in estimating the population mean $\hat{\overline{y}}_{pop}$? I am hoping to obtain auxiliary variables, such as age, for both the respondents and non-respondents.

I don't think I want to use weighting-class adjustment because I don't have missing at random. I am trying to figure out a way to use propensity score weights and/or regression, if possible, but am not quite connecting the dots. Could anyone help me out?

  • $\begingroup$ Commenting on my own post - I think that raking is going to be useful. I will probably only have the population totals, and the individual characteristics of the respondents. If anyone is an expert on raking or knows of a newer/better method, please let me know. $\endgroup$ – Alex Jun 25 '20 at 15:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.