Characterization of triply balanced matrices with applications to survey sampling (Q1820524)

The authors develop an exact algebraic characterization of triply balanced matrices showing a) their number of rows R is a multiple of 8, b) any \(R\times L\) triply balanced matrix is an orthogonal array OA(R,L,2,3;\(\lambda)\) with \(\lambda =R/8\) and vice versa. An existence theorem for a \(R\times 2^{-1}R\) triply balanced matrix is stated in terms of the \(2^{-1}R\)-Hadamard matrix. Next, the non-existence of a sequence of nearly triply balanced matrices (i.e. with the number of triples of columns non-satisfying the balance condition of order \(O(L^{1/2})\) being less than \(R_ L\leq (1+\alpha)L\), \(\alpha\in (0,1))\) is proven. These results have straightforward consequences for computation of mean square errors of nonlinear estimators in stratified sampling from finite populations.