Royal Statistical Society Social Statistics Section
 
9 May 2006, 5.00pm, at the Royal Statistical Society Headquarters, 12 Errol
Street, London, EC1Y 8LX (Tea from 4.30pm)
 
Opinion polling: new methods, old problems?
John Curtice (Strathclyde University)
 
The methodology of opinion polling in Britain has been the subject of
controversy ever since the polls pointed to a narrow Labour lead in votes,
yet in the event the Conservatives emerged with a lead of eight points.
Although the polls avoided picking the wrong winner in the landslide
elections of 1997 and 2001, it still seemed to be the case that the polls
had a tendency to overestimate Labour's strength in the electorate.
 
This apparent problem has persisted despite the introduction of some radical
innovations in the methodology of conducting and weighting polls over the
last decade. Most polls have abandoned the face to face quota poll
methodology that still predominated in 1992 in favour of quasi-random
telephone polls. Meanwhile in reporting their results, many polls also now
attempt to estimate the likely preferences of those who fail to declare
their voting intention, take into reported probability of voting, and weight
their samples by respondent's reported vote at the last election. Inevitably
the apparent persistence of a pro-Labour bias despite these changes raised
questions about their efficacy.
 
The 2005 election, meanwhile, was the first general election test of a yet
more radical change in opinion poll methodology - the introduction of
internet polling. Pioneered by YouGov, the introduction of this technique
has inevitably aroused considerable controversy. On the one hand critics
doubted the ability of a methodology that could at best only access half the
electorate, especially in the absence of a sampling frame for that half.
Advocates on the other hand suggested that internet polling could overcome
some of the problems that apparently still bedevilled 'conventional'
polling, such as 'availability bias' and reluctance to declare a vote
intention.
 
This talk will assess the record of the polls in the 2005 election. First,
it will consider whether the record of the polls in general at that election
suggests that the industry as a whole has overcome the apparent pro-Labour
bias of the three previous elections. Second, it will examine the relative
performance of internet and non-internet polls to assess how well the former
has emerged from its first general election test.
 
The meeting will be chaired by Flavia Jolliffe and Paul Baines (Middlesex
University) will act as discussant. Pre-registration is recommended. You can
register by email: [log in to unmask] or by phone (020) 7638 8998. For a
map and directions see http://www.rss.org.uk/about/direction.html.