Lib Dem Transition Analysis

This page first posted 14 October 2007

Polls published in October 2007 show the Liberal Democrats on relatively low poll ratings. For instance, an Ipsos-MORI survey published on 12 October, showed Con 41%, Lab 38% and LibDem 11%. This marks a significant reduction for the LibDems from their support of 22.6% at the May 2005 General Election. Using our uniform-swing transition model this translates into zero seats for the Lib Dems. This is obviously too low an estimate, but why does the model say this and is it making an obvious mistake?

Let us start by reviewing some recent elections:

ElectionPrevious electionPrev LibDem VoteNew LibDem VotePrev SeatsNew Seats Vote ChangeSeat Change
19971992 (implied seats)18.3%17.2%1846-1.2%28
20052001 (implied seats)18.8%22.7%5162+3.8%11

Let us have a look at these three elections in detail, including seeing them seat-by-seat.

1997 Election

One striking feature is that in 1997 the Lib Dem vote share went down, but their number of seats more than doubled. Let us look at this in more detail with the following chart. Each yellow dot on the chart represents one seat. Its value on the x-axis represents the LibDem vote in that seat in 1992, and its height on the y-axis represents the seat's Lib Dem vote in 1997. Seats won by the LibDems in 1997 are shown with a red border. The best-fit line, which has a slope of 1.03, is also shown in orange.

We can compare this outcome with what different uniform-swing models predict. The standard additive model predicts that all the yellow dots would lie along a straight line with slope 1.00 (with the line passing through the point (18.3%, 17.2%) which marks the average LibDem support in 1992 and 1997 respectively). The Electoral Calculus Transition Model predicts a straight line with slope 0.94 (=17.2%/18.3%), passing through the same average point. The additive model is closer to the actual outcome, but both models do not predict the cluster of seats around 40% which lie above the line. These seats, which are mostly won by the Lib Dems, are seats where the Lib Dems have done much better than any uniform swing predicts. It is behaviour like this which makes predicting the Lib Dems difficult.

This seat information can also be seen in a table. The table divides seats up into groups, depending on the Lib Dem share of the vote in each seat in 1992. The first group is those seats where the Lib Dems received between 0% and 10% of the vote, the second group goes from 10% to 20%, and so on up to the strongest seats where their support is between 60% and 70%. For each group we can measure the average support level in 1992 (which is around the mid-point of the group's range), and also the average support level for those same seats in 1997. This shows how support changes, depending on the starting support level. Also shown are the number of seats won by the Lib Dems in each category.

LowHighCount1992 Ave1997 AveChange1992 Wins1997 Wins
  641  -1.3%1846

Again we can compare the actual movements with uniform swing predictions. Under the additive model, we should expect that the change will be equal for each group at around -1.2%. Under the transition model, we expect that changes in the weaker seats will be smaller (-0.5% for seats in the 0%-10% range), and greater for stronger seats (-3.1% for seats in the 50%-60% range). Broadly speaking the groups for 0%-10%, 20%-30% and 50%-60% behave more like the transition model and the additive model. However, the groups for 10%-20%, 30%-40% and 40%-50% behave like neither. The first group, which contains the bulk of the seats, performs worse than expected under either model. The other two groups actually gain support, against the national trend, and gain quite a number of seats.

This reflects very well on the Liberal Democrats' electoral strategy which is able to "lose" votes where it does not matter (both in seats where they are weak or very strong), and to "win" votes in target marginal seats. But it does make them hard to predict. The final Electoral Calculus prediction in 1997 for the Lib Dems was too low by 21 seats.

2001 Election

We can draw a similar chart for the 2001 election, for which the x-axis now represents the Lib Dem vote in each seat in 1997, and the height represents the vote in 2001. The best-fit slope in this case is 0.94. Since the Lib Dem support has gone up nationally, both the additive and transition uniform-swing models are basically equivalent.

This election was a little easier to predict. The points are fairly evenly distributed along the best-fit line, apart from a couple of outliers. The table breaking down seats according to their 1997 vote share shows a similar picture. Some groups (10%-20%, 30%-40%, 40%-50%) move similarly to the national +1.5% change. The exceptions this year are in groups which do not change the seat winner, such as the weakest and strongest groups along with the 20%-30% group.

LowHighCount1997 Ave2001 AveChange1997 Wins2001 Wins
  641  +1.5%4652

The error in the Electoral Calculus prediction for the Liberal Democrats in 2001 was only 2 seats. See the 2001 track record.

2005 Election

We can draw another chart for the 2005 election, for which the x-axis now represents the Lib Dem vote in each seat in 2001 (using the new implied boundaries for Scotland), and the height represents the vote in 2005. The best-fit slope in this case is 0.87. Since the Lib Dem support has gone up nationally, both the additive and transition uniform-swing models are basically equivalent.

The best-fit line is relatively good, but there is a small but noticeable set of seats around 25% support in 2001 which lie well above it. The table breakdown shows this clearly as the unexpected gains of seats in the 20%-30% range from 0 seats to 10 seats. The bulk of seats in the ranges from 0% to 30% swing as expected at around +4%. But the strong seats from 30% to 50% gain less, and the strongest seats lose support. This meant it was hard to predict the wins in the 20%-30% range, and easy to overstate the expected gains in the 30%-40% group.

LowHighCount2001 Ave2005 AveChange2001 Wins2005 Wins
  628  4.0%5162

The error in the Electoral Calculus prediction for the Liberal Democrats in 2005 was only 4 seats (too low), though this was made up of eight seats down and four seats up. See the 2005 track record.

Next Election

The drop in Lib Dem support, as measured by the opinion polls, from 23% to 11% is unprecedented in recent years and it is hard to say with confidence what will happen to the 66 Lib Dem seats (based on the implied 2005 results). Looking further back into the past, there is one precedent from election history. Before the October 1931 election, the Liberals enjoyed 23% national support and had 59 seats. In that election their support fell to 7%, but they still managed to hold 37 seats. However, by 1945 their support stabilised at 9% but their number of seats reduced to just 12. This suggests there might be a noticeable incumbency effect, but one which fades away with time.

The Liberals have never won fewer than six seats in any election from 1900 onwards.

1. Transition model

The transition model predicts a best-fit line with a slope around 0.5. This means that in each seat, about half the 2005 Liberal Democrat votes are transferred to other parties. This takes votes away from the Lib Dems where they have votes, but it also means that they lose most votes where they are strongest. This has the following features, some of which are good and some are bad

2. Additive model

The Additive model predicts a best-fit line with a slope around 1.0. In each seat, the Lib Dem vote share is reduced by about 12% and transferred to other parties. In weak seats this can make the Lib Dem vote very small or even negative. This also has positive and negative features: Since Electoral Calculus publishes individual predictions for each seat, there was strong feedback from our readership that negative predicted votes were not acceptable. Most other prediction calculators which use additive swing take the precaution of not publishing individual seat forecasts to avoid the visible appearance of this error.

We can note that the difference between the transition model and the additive model for the Lib Dems is only about 15 seats. Whilst this is a very important difference for Liberal Democrat MPs and supporters, it is not very material in calculating which of the major parties, if either, will have a majority in the House of Commons. More important is whether the Lib Dems will do much better than either uniform-swing model suggests and manage to hold 30 or more seats and have the potential to be power-brokers in a hung parliament.

3. Other factors

Also to be borne in mind are other factors which are often mentioned by both interested and disinterested commentators:
Return to top of this page or home page.