Ricerc@Sapienza

The main national institutes of statistics use calibration estimator (Deville e Särndal, 1992; Särndal, 2007). During the years the numbers of auxiliary totals considered in calibration is strongly increased. Because of this, two main problems arise. Very often there is failure to identify "calibrated" weights and, furthermore, the variability of “calibrated” weights can make the efficiency of estimates worse. Therefore, the choice of auxiliary variables in calibration is of a greatest importance.
In the present paper the Shapley decomposition method (Shapley, 1953) has been adapted to this context. Compared to other methods, the Shapley decomposition enables the evaluation of the impact of different calibration systems as a whole. Moreover, with the Shapley decomposition method is possible to derive the marginal impact, net of others auxiliary variables, on the estimates and on the sampling error of each auxiliary variable used in calibration.
The method has been used on the European Health Interview Survey (EHIS) data and it has shown its applicability and, furthermore, a lot consideration that scholars can drawn from it.