[EM] Statistical pairwise method

Kristofer Munsterhjelm km_elmet at t-online.de
Mon Feb 20 08:13:11 PST 2017

A few days ago, I happened upon this page:


which seems to describe a pairwise method that's based on maximizing the
MLE of a model where a voter samples some preference for X and for Y on
two Gaussians (with unknown means) and votes X>Y if his sample for X is
greater than his sample for Y.

Is this Condorcet? It might well be (because of the Condorcet jury
theorem). In any case, it's an interesting approach, and reminds me of
Peter Zbornik asking some years ago if it'd be possible to model voter
decisions as uncertain on some statistical basis and take that into account.

More information about the Election-Methods mailing list