Isn't the core assumption of MVT that there exists a one dimensional spectrum from which to calculate a median in the first place? Voting for candidates is hardly one-dimensional; even a particular ballot measure may have multiple choices, or another question on the ballot of a related but somewhat conflicting nature.