Heteroscedasticity

Abstract

(). Bouza (a) proposed to use µ rss(nr) = r k= M(i))r, where M(i) = w(i, ,) m(i,,) u= Y ′ (i, u))m(i, ,) + w(i, ,)) m u= Y * (i, u))m(i, ,)). Deening w(i, t) = m(i, t))m, Y ′ (i, u) as the value of Y in the uth unit of s ′ , and Y * (i, u) = y u(u) if the unit with rank u in the u-ranked set responds and zero otherwise. e use of DS showed that the estimator is unbiased and that the expected variance is EV[µ rss(nr) ] = V +G, where V = σ n + W (− θ)σ nθ and G = ∆ − ∆ , deening ∆ = m j= (µ (j) − µ) m and ∆ = r i= E m(i,,) j= (µ (j) − µ) n. Hence, the use of RSS is more accurate than the use os srswr also when the nonrespondent sample is subsampled for solving the existence of missing observations in the sample. Other results in this line are in progress; see for example Bouza ().