Ricerc@Sapienza

In analytic inference, data usually come from complex sampling designs, possibly with different inclusion probabilities, stratification, of units. The effect of a complex sampling design is that sampling data are not {\em i.i.d.}, even if they are at a superpopulation level. This dramatically changes the probability distribution of usual test-statistics, such as Spearman's Rho. An approach based on a special form of resampling is proposed, and its properties are studied.