Bayesian inference and sample design for regression analysis when there is nonresponse
Given random variables \((X,Z_ 1,...,Z_ r)\), there may be two multiple linear regressions of X on \((Z_ 1,...,Z_ r)\). The first regression refers to the stratum of potential 'respondents' to a sample survey, while the second refers to the complementary stratum of 'nonrespondents'. Assuming the primary inference is about the relationship of \(Z_ 1\) and X, we give an approximate highest posterior density region for \((\beta_{11},\beta_{21})\), the two regression coefficients relating \(Z_ 1\) and X. Here an initial random sample of size n yields data for the \(n_ 1\) respondents, and for \(n_ 2\) nonrespondents, selected as a random subsample from the initial \(n-n_ 1\) nonrespondents. We then investigate the choice of values for \(n_ 2\) and n to optimize the content of the highest posterior density region. As an illustration, we discuss a survey to estimate the long-run income elasticity of housing expenditures using cross-sectional data.
This page was built for publication: Bayesian inference and sample design for regression analysis when there is nonresponse
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3684979)