Improving Precision by Calculating Estimates During the Calibration Process

• Activity Number: 137
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Survey Research Methods Section
• Author(s): Marcus Berzofsky* and Bonnie Shook-Sa and George Couzens and Phil Kott
• Companies: RTI International and RTI International and RTI International and RTI International
• Keywords: Calibration ; Post-stratification ; Variance reduction ; Precision ; National Crime Victimization Survey (NCVS)

Abstract

Analysts typically employ calibrated weights to improve their inferences. Calibration weighting can adjust for differing probabilities of response or coverage among the elements in the sampled population or it can reduce the impact of purely random error sources. Final weights may have undergone several calibration-adjustment steps. Each adjustment can increase the variability of the weights, which results in a decrease in the precision as conventionally measured. Relatively new software such as SUDAAN's WTADJUST procedure allows analysts to produce more accurate precision measures by calculating estimates during a calibration-weighting step rather than after. This means that calibrated weights need no longer be treated as if they were the original design weights (i.e., the inverse-selection-probability weights) in standard-error estimation. Using recalibrated sub-national estimates of victimization rates from the National Crime Victimization Survey, a nationally representative survey of persons in households, sponsored by the Bureau of Justice Statistics, we show how much this improved measure of precision can reduce estimated standard errors.