Assessment of calibration weighting methods using incomplete auxiliary information

Overview

In survey sampling, it is often desirable to use available auxiliary information in the weighting as calibration constraints to improve the efficiency of survey estimates.  Such auxiliary information is increasingly available from administrative and big data sources, but this information may have incomplete coverage of the target population.  As such the auxiliary information may be considered as a non-probability sample from the population.

Research engagement

Various methods have previously been considered to address the non-response in the auxiliary data.  When the amount of missingness in the auxiliary data is small, the missing data may be imputed using existing data, and the resulting dataset with imputed auxiliary information is then used in the calibration process.  Recently, other approaches have been proposed which treat the incomplete auxiliary data as non-probability data.  These include the model-calibration idea described in Small area estimation using incomplete auxiliary information, and the QR prediction approach described in QR prediction for statistical data integration.  Again, both approaches involve some modelling of the missing auxiliary data.

Research activities

1. Read and obtain an understanding of the methods described in the two papers above.
2. Undertake a quick search of literature to identify additional methods to address incomplete auxiliary information in calibration
3. Undertake an empirical examination of the methods identified using a synthetic dataset, and compare performance against a 'baseline' approach which first fills in the missing auxiliary information via an imputation method and then proceeds with calibration treating imputed values as real values.
   1. Consider different missingness thresholds for the auxiliary information to examine when certain methods start performing better than others
   2. Consider performance of methods when the relationship between auxiliary variable and variable of interest differs from what is assumed by the method (ie mis-specified model)
4. As part of this work we can also assess ease of implementation of the different approaches and compare the conditions/assumptions required to use them

Research skills

Critical Thinking and Reading, Basic Research Skills,  Advanced Statistical Modelling

Outcomes

This project aims to yield insight into how methods may perform under different scenarios, and how straightforward they are to implement relative to each other.  This will inform what is possible in practice when we want to use incomplete auxiliary information for calibration.

Skills and experience

Statistics or Applied Mathematics  or Computer Science

Start date

2 November, 2026

End date

19 February, 2027

Location

GP Y Block Level 8

Keywords

• Auxiliary Data
• Sampling Theory
• Small Area Estimation
• QR Prediction

Contact

Gentry White

3138 1658

whiteg5@qut.edu.au