Exploring Several Uses for Instrumental-Variable Calibration
Kott, P. S. (2012, July). Exploring Several Uses for Instrumental-Variable Calibration. Presented at JSM 2012, San Diego, CA.
Brewer (Business Survey Methods, 2000) introduced the idea of using instrumental variables in a calibration-weighted estimator of a finite-population total as a way of integrating the prediction form of model-based sampling theory with randomization consistency. By employing the nonlinear weight-adjustment function described in Folsom and Singh (SRMS Proceedings, 2002) to create the calibration weights, this paper will explore three potentially more practical uses of instrumental-variable calibration: creating nearly optimal weights under probability sampling theory that never fall below unity and (if desired) are bounded from above, simplifying the production of replicate weights for variance estimation when the function creating the calibration weights is nonlinear, and adjusting for nonresponse when the variables governing the response/nonresponse mechanism are not always the same as the calibration variables.