Rev 07 Dec 25
We believe that PV, our proposed AV variant, is infinitely better than FPTP, and the very best replacement for it that can be offered
(a) in the short term
(b) while we are restrained to a manual count.
But it is clearly somewhat less than optimal, and would merit being upgraded as soon as is technically possible. Two particular examples of its defects:
(a) The candidate with the fewest top choices can be in fact be explicitly preferred to every one of the other candidates taken individually[1], but will nevertheless be the first to be eliminated.
(b) The candidate with the most top choices can in fact be explicitly less preferred than each of the other candidates taken individually, but will neveetheless be protected from elimination.
So for the long term, when we are longer constrained by the capabilities of a manual count, we should consider an extension of the AV/PV algorithm. E.g.:
(1) Elect any candidate who has >50% of top choices, as in AV.
(2) Failing that, elect any candidate who is explicitly preferred to every one of the other candidates taken individually.[2] This involves the tedious assessment of every possible pairwise head-to-head contest, an excessive task for people to undertake, but easy meat for a computer.
(3) Failing that, eliminate any candidate to whom every one of the other candidates taken individually is explicitly preferred, and repeat from (1).
(4) Failing that, eliminate the weakest candidate on some other criterion (but not necessarily AV’s crude “fewest top choices”), and repeat from (1).
[1] We choose not to mention Condorcet. The word is not that well known, and attempts to define it seem to inevitably increase confusion.
[2] That would of course include the >50% winner if they had not already been selected by a more straightforward calculation.