Normalized matching property for competing urn models

Abstract

We study the competing urn model in which $m$ balls are placed independently into $n$ urns according to (possibly distinct) ball distributions. Kahn and Neiman (2010) showed that, under identical ball distributions, the induced urn measure has \emph{conditional negative association} property and asked whether this remains true without assuming identical distributions. We answer this in the affirmative by showing that the competing urn model satisfies the \emph{normalized matching property}. This, in turn, implies conditional negative association for the induced urn measure with non-identical ball distributions, resolving the question of Kahn and Neiman.