Maintaining Precision in Survey Estimates while Adjusting for Conditional Bias at the Subnational Level through Calibration Weighting
Keywords: variance estimation, linearization
AbstractCalibration weighting improves inference by adjusting for observed differences between the realized sample and the population. Unfortunately, a commonly-used linearization-based variance estimator often does not account for the increased efficiency provided by the calibration process. As a result, precision estimates based on calibrated weights can be artificially high. Using a relatively new alternative linearization-based variance estimator allows analysts to utilize calibration-weighting techniques while producing more accurate precision estimates. We use calibration weighting to produce more reliable subnational estimates and assess the differences in point estimates resulting from these weight adjustments in the National Crime Victimization Survey, a nationally representative survey designed to calculate victimization rates solely at the national level. We then assess the estimated precision of these point estimates using a conventional linearization-based variance estimator and the alternative estimator. We find that the calibration adjustments mostly reduced the standard errors in subnational estimates but to successfully measure the reduction required using the alternative variance estimator.
How to Cite
Shook-Sa, B. E., Kott, P., Berzofsky, M., Couzens, G. L., Moore, A., Lee, P., Langton, L., & Planty, M. (2017). Maintaining Precision in Survey Estimates while Adjusting for Conditional Bias at the Subnational Level through Calibration Weighting. Survey Research Methods, 11(4), 405-414. https://doi.org/10.18148/srm/2017.v11i4.6789
Copyright for articles published in this journal is retained by the authors, with first publication rights granted to the journal. By virtue of their appearance in this open access journal, users can use, reuse and build upon the material published in the journal but only for non-commercial purposes and with proper attribution.