Voting matters - Issue 3, December 1994

Are better STV rules worthwhile?

I D Hill

David Hill is the Chairman of the ERS Technical Committee.

R J C Fennell's article, in Voting matters issue 2, raises a number of matters that deserve reply.

Taking a voting paper naming ABCD in order of preference, where B has been elected on the first count and A is elected on the second count as a result of transfers from B, he asks whether that paper's surplus should go to B or to C. He appears to have failed to notice that, in the current ERS rules, it is totally immaterial whether the vote is taken as if it were ACD instead of ABCD, because that paper's surplus does not go anywhere. The voter's second and subsequent preferences are completely ignored, the whole paper remains with A, while only the new votes that A has received are redistributed.

Let us look instead at the point that he was trying to make. Suppose, in that same election, that C was also elected on the first count, and we have a paper naming BACD. That paper will pass from B to A and will be further redistributed, at a suitable value. Should it go to the next choice C, or jump over C straight to D as currently happens? He suggests that such a voter with future vision would not have put C into the list, so it is right to jump to D. But all voters ought to be treated alike, and therefore, if we are to treat one as if future vision existed, we must do so to all others too, and most voters would wish to change their votes if they knew what was going to happen; nobody would vote for the runner-up, of course. But such a change would make sense only if nobody else changes; if we treat everybody as though allowing them to change, the assumed future vision would collapse, no individual could then know how to change and the whole system would become wildly unstable. There is only one satisfactory way out, and that is to treat each vote in strict accordance with what it says, and not by what we assume that it might have said if only the voter had known what would happen.

Transferring to a candidate who has already been elected, as in Meek-style STV, does not waste votes, as is suggested, because the same size surplus is passed on in any case. The change is only to whether the surplus is taken fairly, from all relevant groups in proportion to their current totals of votes, or unfairly in some other way. To change the example, suppose that there are 100 AC votes and 10 BAD votes in a situation where the quota is 77. A is elected on the first count giving 77 to stay with A for quota, 23 to be transferred to C. If, later on, B is excluded then in current ERS rules the 10 all pass to D. The Meek alternative is that only 3 pass to D, while 7 more of the original ACs pass to C. The two methods are

ERS    100 votes    77 stay with A    23 go to C
further 10 votes                      10 go to D

Meek   100 votes    70 stay with A    30 go to C
further 10 votes     7 stay with A     3 go to D

The next point addressed by Fennell is the treatment of 'short lists', that is to say votes that would be transferred if they had a next choice, but do not show one. He mentions the two possible treatments discussed in detail by Meek, but says that Meek's papers ignore a third possibility, and it is evident that he is thinking of something like the current ERS rule. He is wrong to say that Meek ignored this; his paper said 'If the difficulty were to be avoided by increasing the proportion transferred of votes for which a next preference is marked, to enable all x votes to be retained by C, this would clearly reintroduce inequities of the kind Principle 2 was designed to eliminate'. I agree with Meek that this possibility does not deserve any more discussion than that, but many people have failed to see that this method is wrong in principle, and a far greater quantity of writing has gone into it in the last few years than can be reproduced here. I can well see that people might take the wrong decision on this at a first quick glance, but the number who continue to do so even after thought and discussion is quite extraordinary.

I disagree with Meek that the voters should be given the choice between the two methods he discusses in detail. This would have to mean explaining to them the different effects of each, a task that I would not wish on anyone. Meek points out that the two can give different results; usually they do not but, in the few cases we know of where they do, to give the relevant surplus to 'non-transferable' and reduce the quota to compensate is always the preferable option.

Fennell suggests that these voters may not wish other candidates to have any part of the vote. I agree with that - indeed I insist that, whatever those voters wish, we have no right to assume what their wishes are, but only to obey what their ballot papers say, namely that if they become entitled to a further choice they wish to abstain from making one. It is true that, in the current ERS (and most other) rules, the ballot papers are not physically transferred to any other candidate, but what matters is not what is done with pieces of paper, but the effect of the rule. Consider the simple case, with 4 candidates for 2 seats, and votes

        40 AB
        17 CD
         3 DC

        20 AB
        20 A
        17 CD
         3 DC

It is sometimes argued that if the AB voters had had pre-vision, they would have gone straight to B but, as argued above, we cannot allow that without allowing pre-vision to other voters too. Given pre-vision, the DC voters would have voted for C. Given pre-vision, the A plumpers need not have bothered to vote at all. The only fair thing to do is to take what the ballot papers actually say, and everything in proportion to the numbers involved. That means that half A's surplus must go to B, and half to non-transferable, which gives C the second seat.

Provided that the quota can be changed to allow for the non-transferables, as in Meek's method, it can be shown that this does not waste any extra votes at all. What one method wastes in non-transferable, the other wastes on leaving more votes with elected candidates than they now need to be sure of election. With hand-counting methods, where true quota-reduction is not practicable, it could be the case that the present rule does more good than harm, but I know of no evidence to support such a view.

Turning to the discussion of whether voters should be allowed to express equality of preference if they wish, rather than a strict ordering, I cannot agree that it ought not to be considered. It is undoubtedly the case that the absence of this feature is regarded by many as a major disadvantage of STV. There are some difficulties of implementing it, and it would complicate the instructions to voters. I believe that it is something that we ought to consider introducing one day, but that there are more important things to be done first.

Fennell then discusses the 'Silly Party candidate' method of tactical voting discussed in Woodall's paper. He is right, of course, that trying to utilise it may be to the voter's disadvantage if a wrong guess is made, and would certainly be troublesome if too many voters tried to do it. It is a bad feature though that it should be possible at all. Furthermore it is not necessary for there to be a silly party candidate or tactical voters for the effect to occur. It always happens to those voters who put as first preference the first candidate to be excluded, no matter how sincere their choice, causing a distortion that should be avoided if possible.

Fennell is correct that it is not practicable, save for very small elections, to use methods such as those proposed without doing the count by computer but, in this electronic age, can that really be an adequate reason for putting up with second-best results? He queries whether computer-generated results would be trusted and this is certainly something to which attention has to be given. There are two distinct ways in which things might go wrong. The first is in the input of the data from the ballot papers, but this could be subject to repetition if a recount is requested.

The second possibility of error is in the program to calculate the results but, in a public election it could be arranged that, once the data input has been agreed as correct, each candidate would be given a copy of the data on floppy disc. Each party would have its own program, each independently written from the rules specified in the Act of Parliament, and its own computer near at hand. Within a few minutes each could have checked that the official result is agreed. Such a system would lead to much greater protection against errors than anything that could be done with hand-counted STV.

The article concluded that 'the currently accepted Newland/Britton hand counting rules ... will produce a satisfactory result. I can see no reason to change the current system of counting'. What is meant by a satisfactory result? In comparing the results of real elections by hand-counted rules and by Meek rules the result is different more often than not, and all the indications are that the result is not merely different but better, in more accurately representing the voters' wishes. Now that the ability exists to do something better than can be done by hand, it would be absurd to try to exist in the past. Does it not matter to the Electoral Reform Society (or others) whether we get the best result or not?