In this paper, we study a network interdiction problem on a multiple allocation, uncapacitated hub network. The problem is formulated as a bilevel Stackelberg game between an attacker and a defender, where the attacker identifies r out of p hubs to interdict so as to maximize the worst-case post-interdiction performance of the system with the surviving hubs. We study three variants of the problem, namely, the r-hub median interdiction problem, the r-hub cen- ter interdiction problem, and the r-hub maximal covering interdiction problem. The bilevel problems are reduced to single-level mixed integer programs (MIP) using dual and penalty- based formulations. We exploit the properties of the models to present tighter single-level MIP formulations. We compare the linear programming relaxations of dual and penalty-based for- mulations to establish the dominance relations between them. Our theoretical analysis shows that the single-level dual formulations of all the three problems are stronger than their cor- responding penalty-based formulations. We validate these theoretical results using extensive computational experiments on moderate to large-scale instances. Our computational results on networks with up to 200 nodes and 15 hubs confirm the strength of the proposed formulations.