Safe Seats: Scenario Analysis

This page first posted 13 September 2009, updated 6 March 2010, 24 April 2010

This page is part of an analysis about which seats are safe.

1. Poll Analysis

We start by looking at all the published national opinion polls from May 2005 to date (6 Sep 2009). There are 370 relevant polls. To define the edges of our extreme scenarios we look at various percentiles of the distribution of party support. For instance the Conservatives support level at the 25% percentile is 36%. This means that in one-quarter (25%) of those polls the Conservatives had support less than 36%, and in three-quarters of the polls they had support more than 36%. The central 50% percentile is also called the median.

The table shows some key percentiles of party support:

Percentile 5%10%25%50%75%90%95%
Con %32333639414445
Lab %24252932363940
Lib %14141618192122
All Others %78911131518

What does this table tell us? Looking at the Conservatives first, it shows that their central support level is 39%, which happens to be close to their current support level of 40%. It also shows that half of all the polls have them in the range from 36% to 41% (from the 25% to 75% percentiles). And, in the extreme case, nine out of ten polls have them in the range from 32% to 45% (from the 5% to 95% percentiles). We will use this extreme case as the basis for our scenarios of Conservative support. So we assume that 32% is the worst reasonably possible result for the Conservatives and that 45% is the best reasonably possible result.

For Labour, the central support level is 32% and the range of worst to best outcomes goes from 24% to 40%.

For Liberal Democrats, the central level is 18% and their range goes from 14% to 22%. Because Lib Dem support often increases during the election campaign, we will extend their best-case range to 25% to make sure that all (reasonably) possible outcomes are included.

We could instead have just used the simple minimum and maximum of poll support levels. Why did we use this more complicated method instead? The answer is that the minimum and maximum can be skewed by a single rogue poll. Since we are using several hundred polls, the chance of at least one or two bad ones is quite high. We can reduce the risk of such "outliers" by using a percentile range instead of the full range. Almost all polls are included, but we ignore a small number of polls (37) that are very far from all the others.

2. Scenario Definition

Now we have our ranges of plausible party support, we can create some scenarios. We want to have a number of scenarios, each of which describes the support of the three major parties. The aim is to have the complete set of scenarios span the extremes of possible voter behaviour.

We have used six scenarios. Three scenarios correspond to one major party doing well, and the other two doing badly. The other three scenarios correspond to two major parties doing well, and the other one doing badly. The scenarios are not meant to illustrate likely outcomes, but rather to describe the limits of credible outcomes. It is likely that the eventual outcome will be within the limits spanned by the scenarios.

For information the base scenario (general election May 2005) is also shown. The scenarios are:

ScenarioStrongWeakCon %Lab %Lib %SwingOutcomeCon seatsLab seatsLib seatsNat seatsChangedSeats
May 2005  3336230Lab maj 422083466780
OneLabCon, Lib324014-2.5Lab maj 118198384351260
TwoConLab, Lib45241412Con maj 2004251542228238
ThreeLibCon, Lab3224255.5Con short -412842388918111
FourLab, LibCon314025-2.5Lab maj 11017338069739
FiveCon, LibLab45242512Con maj 166408148659217
SixCon, LabLib4439134Con short -18307306106111

(Note: some scenarios have adjusted the ranges slightly to make sure that the total of support, including nationalists, is below 100%.)

The scenarios include a range of possible outcomes

We note that over all these extreme scenarios, the largest number of seats which ever change hands is 238 (less than 40% of the total number of seats). So under each individual scenario, six out of ten seats are safe.

3. Scenarios in Scotland

As part of this analysis we also looked at Scotland separately. Since there are far fewer opinion polls, we used a simple minimum and maximum to get the likely range of support.

Con %141820
Lab %283644
Lib %91423
SNP %163034

Again we adjust the maximum of the Lib Dems range up to 25% since no opinion poll has shown them increase their support since the general election. Under each of the six GB scenarios above, we specify the particular share of the vote that the four Scottish parties receive. We assume that the SNP does well in the first three GB scenarios, and does badly in the other three GB scenarios.

ScenarioStrongWeakScot Con %Scot Lab %Scot Lib %Scot Nat %Con seatsLab seatsLib seatsSNP seatsChanges
May 2005  164023181411160
OneLab, SNPCon, Lib1443933143697
TwoCon, SNPLab, Lib202893482262324
ThreeLib, SNPCon, Lab14282433228151413
FourLab, LibCon, SNP144425160431152
FiveCon, LibLab, SNP2028251663315510
SixCon, LabLib, SNP2044916644638

We see that the scenarios generate a range of outcomes. There is the possibility for any party to increase or decrease its number of seats.

4. Special cases

At the 2005 general election, three seats were won by independent or minor party candidates. Since these seats are hard to predict scientifically, we will assume that they are not "safe" and could change hands. We also include otherwise safe seats which have changed hands at by-elections. These seats are:

5. Total numbers of "safe" and "unsafe" seats

We can define a seat to be "unsafe" if any of these conditions apply There are 306 "unsafe" seats, out of a GB total of 632 seats.

Conversely a seat is defined to be "safe" if none of the conditions above hold. In other words, a seat is safe if it never changes hands under any plausible scenario of public opinion. There are 326 "safe" seats, which is about half of the total.

Caveat: Of course, nothing is entirely 100% safe. It is possible that extreme public opinion swings or strong local factors in a particular seat may make a "safe" seat change hands. But within the bounds of the likely set of outcomes and the law of averages, it is pretty unlikely that more than a handful of safe seats could be captured.

6. Campaign update - 24 April 2010

During the election campaign there was a significant increase in Lib Dem support up to 34% in opinion polls. This caused some seats which were deemed "safe" to be predicted Lib Dem gains. To cope with this unexpected development, we have redefined scenarios Three, Four and Five which have strong Lib Dem performance. These scenarios are now:

ScenarioStrongWeakCon %Lab %Lib %SwingOutcomeCon seatsLab seatsLib seatsNat seatsChangedSeats
ThreeLibCon, Lab3224335.5Con short -6526020714418154
FourLab, LibCon293831-3.0Lab maj 9216037191754
FiveCon, LibLab43223112Con maj 126388135989221

There are now 300 safe seats, and 332 unsafe seats in Great Britain.