Track Record: 2001 Errors

This page first posted 13 March 2005

This page shows the seats which were incorrectly predicted at the 2001 general election.

Prediction errors are caused by two different types of effect: opinion poll error and model error, giving the total error.

In summary, 32 seats were incorrect due to opinion poll bias, and 27 seats were incorrect due to model error. But in total only 35 seats were incorrect because in 12 seats the two errors cancelled each other out.

At the next election about 50 seats will be incorrect.

Opinion poll error

Opinion poll error causes the final opinion polls of the campaign to show different levels of party support than actually happens at the polls in the election. This is in turn will create errors in the prediction.

The causes of opinion poll error include:

In 2001, the opinion poll error overstated Labour's lead by 6.7%, causing 32 seats to be incorrectly predicted.

This figure is calculated by comparing the prediction with two different sets of inputs. The first uses the average of the final opinion polls of the campaign. Those were: Con 30.3%, Lab 46.3%, Lib 17.5%. The second uses the actual support levels in the election of Con 32.7%, Lab 42.0%, Lib 18.8%. Seats which have different predictions under the two sets of inputs are seats predicted incorrectly due to opinion poll error.

In total, opinion poll error understated the number of Conservative seats by 32, overstated Labour by 29, and overstated the Liberal Democrats by 3.

SeatPoll
Pred
Win
Poll
Pred
Maj
Vote
Pred
Win
Vote
Pred
Maj
Altrincham and Sale WestLAB1.1%CON5.6%
Bedfordshire South WestLAB3.7%CON3.0%
Beverley and HoldernessLAB2.4%CON4.3%
BillericayLAB1.5%CON5.2%
Boston and SkegnessLAB2.6%CON4.1%
BosworthLAB2.0%CON4.7%
BraintreeLAB6.6%CON0.1%
Bristol WestLAB6.4%CON0.3%
Bury St EdmundsLAB3.3%CON3.4%
Castle PointLAB6.3%CON0.4%
Chipping BarnetLAB1.9%CON4.8%
Dorset Mid and Poole NorthLIB0.5%CON0.6%
Dorset SouthLAB3.8%CON2.9%
EddisburyLAB1.6%CON5.1%
Harrow WestLAB6.3%CON0.4%
HarwichLAB6.3%CON0.4%
HexhamLAB3.5%CON3.2%
KetteringLAB4.3%CON2.4%
Lancaster and WyreLAB6.2%CON0.5%
LichfieldLAB3.5%CON3.2%
MeridenLAB2.9%CON3.8%
Milton Keynes North EastLAB4.4%CON2.3%
Norfolk MidLAB1.7%CON5.0%
Norfolk North WestLAB6.3%CON0.4%
Northampton SouthLAB5.3%CON1.4%
RomfordLAB5.5%CON1.2%
Rugby and KenilworthLAB4.8%CON1.9%
Suffolk WestLAB0.2%CON6.5%
TotnesLIB0.2%CON0.9%
UxbridgeLAB2.2%CON4.5%
WellingboroughLAB4.3%CON2.4%
WellsLIB0.9%CON0.2%

Since the Conservatives were underestimated by the opinion polls, all these seats are seats the Conservatives should have been predicted to win, but were not.

Model error

Model error is due to the prediction model being only an approximation to reality. All the aspects of reality which are not captured by the model can give rise to errors. These include: We can calculate the size of model error in 2001. We feed the actual levels of party support into the model, rather than the opinion poll levels. Any seat which is still incorrect has been subject to model error.

Predictions are shown as they would have been, had the correct national support percentages been used. That is Con 32.7%, Lab 42.0%, Lib 18.8%. No tactical voting adjustment was used to make the prediction.

In all, 27 seats were incorrect. The net effect was that the Conservatives won 11 seats fewer than predicted, Labour won 7 more, the LibDems won 5 more, and the nationalists (SNP) were 1 seat worse off. The fact that the errors are not symmetric suggests that Labour and Liberal Democrat supports used tactical voting against the Conservatives.

The table below shows the 27 seats, with the predicted winner and majority against the actual results. Important local factors are included as a comment, if applicable.

SeatPred
Win
Pred
Maj
Act
Win
Act
Maj
Comment
BraintreeCON0.1%LAB0.7%Double error
Bristol WestCON0.3%LAB8.0%Double error
CheadleCON5.4%LIB0.1% 
ChesterfieldLAB7.8%LIB5.8%Tony Benn stands down from seat
Dorset Mid and Poole NorthCON0.6%LIB0.9%Double error
Dorset SouthCON2.9%LAB0.3%Double error
GuildfordCON7.7%LIB1.1% 
Harrow WestCON0.4%LAB13.2%Double error
HarwichCON0.4%LAB5.4%Double error
Isle of WightLIB9.5%CON4.5% 
KetteringCON2.4%LAB1.2%Double error
Lancaster and WyreCON0.5%LAB0.9%Double error
LudlowCON12.1%LIB3.8% 
Milton Keynes North EastCON2.3%LAB3.9%Double error
NewarkLAB3.1%CON9.0% 
Norfolk NorthCON1.5%LIB0.9% 
Northampton SouthCON1.4%LAB1.7%Double error
RomseyCON15.9%LIB4.9%LibDems consolidate by-election gain
Rugby and KenilworthCON1.9%LAB5.3%Double error
TattonMIN22.1%CON20.9%Sitting independent MP Martin Bell stands down
TauntonLIB4.7%CON0.4% 
UpminsterLAB4.0%CON3.7% 
WellingboroughCON2.4%LAB4.6%Double error
Wyre ForestLAB9.9%MIN35.9%Strong independent candidate wins seat
Carmarthen East and DinefwrLAB6.3%NAT6.8% 
Ynys MonNAT8.2%LAB2.4% 
Galloway and Upper NithsdaleNAT12.6%CON0.2% 

Total error

Here is a summary of the errors described above:

Error typeSeats
Incorrect
CON LABLIBNAT
Opinion poll error32+32-29-30
Model error27-11+7+5-1
Total error35+21-22+2-1

Paradoxically, the total number of seats mis-predicted (35) is much less than than the sum of the opinion poll and model errors (59). This is because twelve seats suffered mis-predictions in opposite directions by the two errors. For instance, Braintree was predicted to be a Labour hold, which was what actually happened. The prediction's accuracy was specious however. If the opinion polls had been correct, it would have been predicted as a Conservative gain. But the model was as wrong as the opinion polls, so Labour held it (just).

The 2001 prediction result is thus a little less accurate than it appears.

It should be assumed that about 50 seats will be mis-predicted at the next election.


Return to top of page, or track record summary, or home page.