65th ISI World Statistics Congress 2025

65th ISI World Statistics Congress 2025

A cautionary tale about the calibration of nonresponse bias

Author

FM
Florian Meinfelder

Co-author

  • R
    Raimund Wildner
  • V
    Volker Bosch

Conference

65th ISI World Statistics Congress 2025

Format: CPS Abstract - WSC 2025

Keywords: calibration, imputation, missing not at random, missing-data, nonresponse

Session: CPS 28 - Nonresponse Bias and Missing Data in Surveys

Wednesday 8 October 4 p.m. - 5 p.m. (Europe/Amsterdam)

Abstract

Calibration is a popular method to adjust for unit nonresponse bias in surveys. Typically, some external information on the true population quantities of margins for some calibration variables is available, and sometimes also of higher-order interactions. Weighting-based raking algorithms try to adjust cell proportions in the sample to these external benchmarks, which might be systematically biased. The underlying assumption in this context can be described as selected at random (which is the counterpart to the MAR assumption in item nonresponse settings). However, It is generally assumed that even if the underlying missingness mechanism of the unit nonresponse is nonignorable, weighting will at least alleviate the severity of the bias. We discuss data situations, where weighting adjusts the sampling cell proportions, but increases the bias problem. This result seems so counter-intuitive that we feel it has been neglected in the literature so far.

Figures/Tables

model