Discussion of “Issues in weighting household and business surveys”

Calibration in its various forms has become an essential tool in surveys. There are sev-eral reasons why calibration is used, beyond the obvious efficiency argument that states that calibration is a convenient way of improving the efficiency of estimates by exploiting exter-nal information. The other reasons, some of which are also noted by Mohadjer, include the following. • balance (in the sense that after calibration, the sample “looks ” like the population; e.g., the male/female split is correct) • target groups (if we need estimates for a specific subpopulation, we add the appropriate calibration variable) • anchoring of estimates (a variable that is correlated with a calibration variable will tend not to “drift ” too far away, providing stability over time) • convenience (e.g., regression-based calibration provides a simple way of getting a single weight per household in a household survey; see Lemaitre and Dufour (1987)) • consistency of estimates (in production systems, each sampled unit is given a unique final weight as part of the calibration process; as a result, estimates are consistent in the sens