Why did the polls converge?

In the big debate over internet and telephone polling many were hoping the May 5 would resolve the issue once and for all. And in the days leading upto to polling day it did look as though the two competing methodologies were presenting a very different picture of the actual vote with, at one stage, one telephone pollster reporting a 14% Labour lead.

But lo and behold along comes May 5 and the forecasts all looked very similar apart from NOP which managed to get it dead right. Well done NOP and a brilliant performance.

Interestingly not one single entrant in our New Year General Election prediction competition chose the firm as the winner – there’s sweet justice somewhere.

No doubt this heavily polled election will come under serious academic scrutiny and we might start to get a better view in a few months time. My own theory, which I put forward at the end of last year, was that there would be convergence because the quality of the consideration given by telephone respondents to questions in an unsolicited phone call would increase the closer we got to polling day. Internet respondents are volunteers and can respond to the survey in their own time and at their own convenience so, perhaps, give better consideration to the questions polled.

This was Martin Baxter’s verdict on the different performances.

YouGov had a very good election. They consistently showed a Labour lead of around 3%-4%, and their methodology has been vindicated by the actual result.
NOP finished very well with a final poll which was close to perfect.
MORI had a mixed performance. Some polls were quite close to the final result, even though they attracted controversy by focusing on respondents “absolutely certain to vote”. Their last two polls were less accurate and showed leads of 10% and 5% respectively.
ICM showed a lead of 6%-8% in their last two polls, which overstated Labour’s lead by about 4%.
Populus had a final poll lead of 6%, but their penultimate poll showed 14% which was the most inaccurate result of any of the pollsters.
CommunicateResearch showed 8% in their final poll, which was wrong by 5%.

We also had the academic based British Electoral Study and the British Polling Index. In addition Harris polls emerged after many years away from the UK scence with an internet survey on the eve of poll that looked very similar to YouGov.

From a personal persepctive I always believed the phone pollsters were over-stating Labour’s lead in the weeks up to polling day and put my money where my mouth is. My biggest betting success on Thursday was a sell spread on Labour’s vote share.

Finally we ought to thank Anthony Wells for establishing and running UK Polling Report. This proved to be an extraordinary tool and his blog a place of serious debate on methodologies. Well done Anthony.

Mike Smithson