Manipulation under k-approval scoring rules

Under a k-approval scoring rule each agent attaches a score of one to his k most preferred alternatives and zero to the other alternatives. The rule assigns the set of alternatives with maximal score. Agents may extend preferences to sets in several ways: they may compare the worst alternatives, or the best alternatives, or use a stochastic dominance criterion. In this paper we characterize the non-manipulable profiles for each of these set comparisons. For two-agent profiles we also determine the value(s) of k for which the number of non-manipulable profiles is maximal.