A new method for bounding the distance between sums of independent integer-valued random variables (Q607603): Difference between revisions

Let \(X_1,X_2,\ldots,X_n\) and \(Y_1,Y_2,\ldots,Y_n\) be two sequences of independent integer-valued random variables that have finite second moments. Using the \(\zeta_2\)-metric, the author obtains upper bounds for the Kolmogorov and the total variation distances between the distributions of the sums \(\sum_{i=1}^nX_i\) and \(\sum_{i=1}^nY_i\). The bounds are particularly simple when the distributions of \(X_i\) and \(Y_i\) are comparable in the convex stochastic order for \(i=1,2,\ldots,n\), as well as when \(E(X_i)=E(Y_i)\) and the event \(\{Y_i\neq0\}\) is rare for \(i=1,2,\ldots,n\). As applications the author considers a negative binomial approximation for sums of discrete random variables that are of interest in reliability theory, and Poisson and compound Poisson approximations for sums of independent nonnegative integer-valued random variables.