Say there is a three way split (lib, lab con). Under AV one will go out in the first round, then voters will be choosing between the two remaining parties in the final round.
The possible final rounds and expected results are:-
Probability | Final Round | Expected Outcome |
---|---|---|
quite likely | lib vs con | con win |
quite likely | con vs lab | lab win |
unlikely | lib vs lab | lib win |
But now consider the practicalities:
If too many conservatives put libdems ahead of conservative then they may actually prevent the conservative candidate from even reaching the final round! Or they may even give the libdem over 50% in the first count so winning without another round.
How many is 'too many'? well how good are you at predicting election outcomes? And even if you did know, and it was in the hundreds or thousands how would you coordinate this many people (and no more!) to vote in this precise way? In practice there would probably be a few thousand other votes (for other parties) that you would have to predict too.
And, of couse, could you be sure that the libdems weren't going to try exactly the same and try to lend votes to get labour into the final round rather than conservatives, so needing even more conservatives to lend to the libdems for the tactic to work, and if the libdems suddenly decided not to lend their votes, then they would be in an even stronger position to win in the first round!
So you need quite a specific kind of constituency, you need to predict the outcome with enough precision to form a plan and then you need to control many, many voters with the precision to manipulate the result successfully -- and any error at any point is likely to actually damage your candidates chance of winning!
Is this really practical? If so, forget elections, just use your predictive powers to make a killing at the bookies by predicting the results so perfectly!!