Income distributions tend to be highly skewed. Yet the American Community Survey (ACS) provides only two summary statistics for its median income estimates in small areas: the mean estimate and a standard error. I want to build a Bayesian error-in-variables model that uses a latent unknown zip-code-level income estimate as a covariate, with the the latent median income variable estimated from the zip-code-level estimate and standard error from the ACS. I don't want to assume a normal error distribution because, for many small zip codes with low median income, that allows for negative income. I don't want to use a higher level of geographic aggregation because the problem remains anyway and because there is a much more linear (thus more tractably estimated) relationship between zip-code-level median income estimates and the outcome variable of interest, but not so at higher levels of aggregation.


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.