Up: Issue 6 Previous: Paper 4

Voting matters - Issue 6, May 1996

Some comments on Replaying the 1992 general election

I D Hill

David Hill is Chairman of the ERS Technical Committee.

At the time of the 1992 general election, Patrick Dunleavy, Helen Margetts and Stuart Weir conducted research designed to indicate how Britain would have voted under alternative forms of voting. Their report [1],[2] states that the result "poses a problem for STV advocates" in that the allocation of seats is far from proportional by first preferences and severely disadvantages the Conservatives. They are very forthright in their claims that the study shows what would actually have happened. A subsequent letter[3] hoped that the Electoral Reform Society would "address the problems for STV that our ... study identified". It is, of course, not possible fully to address such problems without the data, and I am grateful to the authors for letting me have a copy.

In the comments that follow, I have concentrated entirely on the STV part of the document, ignoring the work that they also did on Alternative Vote, Additional Member, and List PR systems.

The data were obtained from a sample of 9614 people across 13 regions of the UK (excluding Northern Ireland). The sampling and interviewing was done by ICM, using their professional experience of getting a representative sample within each region. The interviewees were given a ballot paper of 17 candidates, in sections by party, their names being those of actual candidates in the general election in that region. They always consisted of 4 each from the 3 main parties, plus 5 others who included the nationalist parties in Scotland and Wales. Within this pattern, the aim was to give a mix of well-known and lesser-known, of men and women, etc.

The country was divided into 5-member constituencies so far as possible, but with some 4-member ones, by combining the actual single-member constituencies within each region. There were 133 such constituencies, consistently misquoted as 123 in their reports.

Much trouble was taken to get representative samples, but for analysis the regional results were reweighted for each multi-member constituency "to produce distinctive local profiles". I do not doubt that this was done with good intentions but, so far as I can see, the anomalous results that "pose a problem for STV advocates" result almost entirely from this reweighting.

My analysis has necessarily had to be slightly incomplete because of some missing files. I am told that some computer discs have become unreadable and these files are going to be difficult to retrieve, so it seems better to go ahead with reporting what I can without them. Those missing concern all four of their East Anglian constituencies, 5.4% of the total data, and three of the Greater London ones (those that they call Richmond and Kingston, Hillingdon, and Central London). We can derive the regional Greater London results, from the other constituencies in the region, but not the reweighting for each of these three missing constituencies.

Of the available files, there are some that show trouble in the data in that some spurious figure zeros appear, that lead much of the data to be ignored by my STV computer program that was used. Luckily only one of these instances leads to a different result by political party from what they found, but there are also four others not suffering from this particular trouble, where the results by party seem to have been incorrectly reported. For the reweighting of the data, the authors say that "the 1992 general election results provide a complete picture of people's first preferences" so they use those to reweight the voting patterns. Even if this actually gave an improvement, I completely disagree with the beliefs behind it. A very important reason for wanting electoral reform is that election results at present do not show people's true preferences. Common observation shows that vast numbers of people vote tactically, not for what they would most like to see but for candidates who, they think, have some chance of success, and trying to keep out the party they most dislike. The squeeze of the third candidate in by-elections is notorious and a similar effect in general elections, to a lesser extent, certainly exists. Whether a better electoral system would make much change in voters' stated preferences or not we simply do not know; until we try the real thing the evidence is not available.

Having done the reweighting, for better or for worse, they report (in their Table 11):

                       Con    Lab     L/D   others
Pure proportionality   273    222     114     25
STV                    256    250     102     26
and this is what they say that STV supporters have to ponder. If we do the analysis by what appear to be the original data for each region, without such reweighting, it means using the same voting pattern for every multi-member constituency within the region, which is a crude model and often unrealistic but is probably the best that can be done with the data available. The results (with some assumptions for missing files) are:
                       Con     Lab     L/D   others
Pure proportionality   273     222     114     25
STV                    274     230     108     22
I think that it is they who have some pondering to do.

I cannot see how the numerical values of their reweighting were derived, but my requests for clarification have not been successful. If, as I believe, it was intended to bring the first preferences, by party, closer to the general election votes, it does not seem to have done so. The results are in the large table.

Percentage share of votes by region

East Anglia missing, italic figures approximate due to missing files

If anything the results after reweighting seem further from the general election results than do the raw ones, and certainly the Conservatives have been marked down.

It might be claimed that it is the individual multi-member constituency figures that matter rather than these overall ones, so I have looked at one constituency in detail to see whether that improves the picture. I chose to do this for my home constituency which, in their scheme, would be the combination of the present constituencies of Herts SW, Herts W, Hertsmere, Watford and St Albans. As an example of their reweighting, in this constituency every vote in the raw data with a Conservative first preference has been treated as 98 identical votes, every vote with a Labour first preference as 121 identical votes, every vote with a Liberal Democrat first preference as 87 identical votes and every other vote as 94 identical votes.

For this constituency I find:

                   Con     Lab     L/D    other
General election   53.3%   25.1%   20.3%   1.3%
STV (raw)          52.7%   23.0%   20.0%   4.3%
STV (reweighted)   51.2%   27.6%   17.3%   4.0%
Apart from slightly reducing the others figure, which is far too big nevertheless, has the reweighting helped? I doubt it.

I am well aware that it is much easier to criticize such a study than to perform one, but it does seem to me that a better scheme would have been to take their 9614 interviews equally from each of their 133 multi-member constituencies, i.e. about 72 per constituency, and then use the results in raw form. It might be argued that 72 votes are not many for electing to 5 seats, but that is all you get with a total of 9614. You do not get any more actual information by using the same figures many times with reweighting.

The authors also comment on "some apparently extraordinary results - as with the election of 5 Green MPs in the south east region", and that only 2 of the 5 would survive if Meek rules were used (I make it only 1 of the 5 actually). In interpreting this we need to remember that it is the same set of votes being analysed over and over again, and the identical person as Green candidate, merely with different reweighting for each constituency in the region. That may have been the best that could be done in the circumstances, but I wish they would not claim that this is what would actually happen in practice. Again it is the reweighting that has produced the odd effect - no Green is elected if the original, unmodified, observations are used.

They seem to think it a disadvantage of STV that it can react with different results when the votes change only slightly. I think it an advantage that most constituencies become marginal for their final seat. At present it is only the marginal constituencies that have any real effect on who wins a general election. Under STV nearly every voter can feel that it is worth voting as it could make a difference.

They also make a point of the fact that "STV is only contingently proportional" if comparing seats with first preferences by party. So it should be. It often helps to explain a point such as this by using an exaggerated example. To repeat one that I have used elsewhere, if we have 9 candidates for 3 seats, A1, A2 and A3 from party A; B1, B2 and B3 from party B; C1, C2 and C3 from party C and the votes are

          20   A1   B1
          20   A1   C1
a party-based proportional system would have to elect A1, A2 and A3 as all first preferences were for party A, whereas STV will elect A1, B1 and C1 and appear to do badly if one insists on comparing seats with first preferences by party, but it has done what the voters have asked for, and that should be the aim of an electoral system.

What is more their data show that, of all interviewees who selected at least two preferences with each of their first two from the three main parties, only 79% of them chose the same party for both choices. If this is nothing like what would happen in practice, then the exercise cannot be quoted as meaningful in this respect. Their report claims strongly that their figures do represent what would happen in practice, but they cannot have it both ways; if they are right in that, then the authors' wish to see party proportionality by first preferences is not shared by the electorate. I believe that the wishes of the electors are what matter.

My overall conclusion is that it has been claimed that STV advocates have some problems to deal with, but in fact it is the authors of the study who need to deal first with the problems that they have created.


  1. P Dunleavy, H Margetts and S Weir. Replaying the 1992 general election (LSE Public Policy paper number 3), 1992.
  2. P Dunleavy, H Margetts and S Weir. Proportional representation in action: a report on simulated PR elections in 1992. Representation, vol 31, issue 113, pages 14-18.
  3. S Weir. Letter to the editor. Representation, vol 32, issue 117, page 20.

Up: Issue 6 Previous: Paper 4