The main selling point of the AV system has been that it is possible to name multiple preferences. The idea that AV offers 'fairer votes' is born out of the intuition that under first-past-the-post, results in marginal constituencies under-represent the preferences of the candidate placed second, because only one MP is elected, on the basis of just first preference votes. The ability to name further preferences under AV means that even if your first preference candidate cannot win, you get to allocate your vote to another preferred candidate.
Yet there are two rather odd ways in which AV deals with second, third and further preferences that should make us doubt that it will be a more representative system than FPTP.
Firstly, later preferences have the same weight as a first preference vote: they all add 'one' to a candidate's pile and toward reaching the 50% threshold. Indeed, it is only by this equal weighting that candidates stand a chance of reaching such a threshold. Yet this means that even fourth or fifth preferences are counted as if on a par with first preference votes, and so not only are voters who put unpopular candidates first effectively over-represented, but the system gives very weak preferences (say fourth) an equal status to stronger preferences (say second) even in the same round of counting. This can give elections under AV very strange results as several rounds of later preferences create surprising swings and make it a particularly unwieldy system to predict.
Secondly, given that the concern for preferences arises from the potential disenfranchisement of voters in marginal constituencies, AV is actually likely to overlook the preferences of these very voters. If a seat is marginal between two relatively equally popular candidates in the first round, it is not the second preferences of these voters that are counted to determine the winner, but of voters for the least popular parties. If two large parties in a constituency accrue most of the vote, it would be quite possible for the second preferences of voters for the second place candidate never to be looked at. What a preference-based system should ask is what the second preferences of these voters are – it could be that in a polarised constituency, a compromise candidate could win the support of a large number of voters, or that Labour voters might be happy to place the Liberals as second preference but not vice versa. This would be useful information in determining the most representative candidate, but would not be used under AV. The very voters that supporters of electoral reform claim are disenfranchised under first-past-the-post would likely be ignored under AV, too.
In a sensible and well-argued piece on this blog, Ryan Bourne has contended that AV might work well as long as there was no tactical voting (and he is entirely right that the claim that AV eliminates the need or incentive for tactical voting is plainly false). I think that even this conclusion is crediting AV with too much –there is good reason to doubt that the possibility of offering further preferences will result in a more representative winner, for weak preferences are counted equally with stronger preferences, and the preferences that we would most want to consider (second-preferences of voters for the two strongest candidates) could likely be ignored.