Modern large-scale household surveys are generally expected to provide high quality estimates of population
parameters. However, strong signals of bias have occasionally been detected for some South-East Asia
countries. For instance, the World Bank’s Household Survey Development Team found significant discrepancies
between survey-based estimates of age-sex and household size distributions and the corresponding Census
counts for Vietnam, Thailand and the Philippines.
A feasibility study has been carried out to investigate whether this issue could be solved through a preliminary
calibration procedure.
Calibration is a systematic and mathematically rigorous method to achieve higher quality estimates by
incorporating auxiliary information on the target population, available from external sources, into the survey
estimation infrastructure [1]. From an algorithmic standpoint, a calibration procedure minimally adjusts the
survey weights in such a way that the resulting calibration estimates exactly match selected known population
totals [4].
Three Proofs of Concept (POC) have been carried out, adopting as empirical test bed the following household
surveys:
 2008 Vietnam Household Living Standards Survey – VHLSS 2008
 2010 Thailand Household Socio-Economic Survey – SES 2010
 2006 Philippines Family Income and Expenditure Survey – FIES 2006
For each survey, calibration constraints have been imposed on known population totals derived from the
closest Population and Housing Census round:
 2009 Vietnam Census
 2010 Thailand Census
 2007 Philippines Census
To tackle the feasibility study, the ReGenesees system was used: an open source software for design-based and
model-assisted analysis of complex sample surveys [10], based on R [3].
All POCs were successful: calibration algorithms were run without noticeable technical problems, and exact
convergence was always obtained. Overall, the study showed that it is technically feasible to integrate a
calibration procedure in the production workflow of all the household surveys taken into account.
Beyond the feasibility study, two possible implementation lines can be envisioned:
1) A calibration procedure could be executed directly by the National Statistical Institute (NSI) in charge of the
household survey, as a process step to be routinely performed preliminary to estimation. Of course,
enabling NSIs to adopt calibration estimators would require appropriate capacity building actions.
2) A calibration procedure could be executed ex-post for analysis purposes, i.e. after data dissemination and
outside the involved NSIs, in order to increase the quality of the estimates derived from the surveys. For
instance, the World Bank could manage the calibration procedure on its own, and integrate the obtained
calibration weights into its microdata repositories.
In both cases, using calibration weights for estimation would be straightforward, while estimating sampling
errors would require specialized software, like the ReGenesees system.