Opinion polls
Sensation before accuracy
George Gaskell and Rod Bond
The performance of the opinion polls in predicting the outcome of the last two elections may have led some cynics to believe that the party shown to be ahead will lose, and that the polls are the kiss of death to the party they favour. Unlike their medical counterparts the opinion polls are notoriously unreliable.
Estimations of the support for any party can easily be 5 per cent in error either way, and, since elections are won or lost on small margins, one wonders if the opinion polls serve any useful function at all. In addition to this basic inaccuracy, both pollsters and media are still wedded to the outmoded concept of the overall percentage support for the parties, a statistic which is potentially quite misleading.
Nevertheless it is generally accepted that taken as a whole the average of the opinion polls should give a reasonably accurate reflection of the state of public opinion. This is based on the assumption that the error in individual polls will be due to chance factors which should balance out once several polls are considered — hence the 'poll of polls' in the Sunday Times. As an example of the potential sources of error consider the typical stages in the operation of a poll. We have underlined the problematic areas where inaccuracies can and
do occur. A sample of the electorate is asked various questions by an interviewer, who sends the answers to the polling centre for collation. The overall percentage support for the parties is then given to the media, who often estimate the likely parliamentary outcome.
The pollsters pay almost exclusive attention to the likely error involved in trying to estimate the characteristics of the population from a particular sample. This error is due to chance factors and as such it is an unavoidable part of public opinion polling. But other sources of error are less innocuous and there is little evidence that the pollsters exercise sufficient control over them. For example, the interviewers may contribute unintentionally to polling inaccuracy. It has. been shown that the social class of the interviewer may influence the responses of the interviewee. One study demonstrated that middle-class interviewers obtain more 'Conservative votes' than working-class interviewers. If the interviewees are mainly middle-class housewives, then the resultant bias is obvious. Furthermore, the pressure to collect the full number of interviews from the right people may be too great and lead to unfortunate expedient measures. There is also the problem of those who give a 'don't know' response to the interviewer. Little research has been carried out to investigate who these people are and why they respond in the way they do. Possibly for some sections of the population voting preference, like earnings, is a private matter and not disclosed in public. If this were more frequent in the supporters of one party or affected by regional differences then a considerable bias could be built in to the poll results. Although most polls show that 10 per cent do not know how they will vote, it is notable that between 20 and 30 per cent of the electorate abstain at the election. Who are the 10-20 per cent who indicate a preference to the pollsters yet do not go on to vote? Do they actually have a preference or do they invent one for the pollster's benefit? How many have not yet made up their minds but indicate a preference — however tenuous it may be? We do not know and we think that the pollsters share out ignorance of these crucial factors. Overall there are many uncharted sources of error built into the polls, and notably few attempts have been made to accurately measure their totality.
Despite the inaccuracies, is the final result of the polls as depicted in the media of any real value? The pollsters admit that they predict the percentage support for the parties rather than the number of seats, hence even with accurate polling the actual parliamentarymajority is still questionable. As was seen in the last election, though polling most votes the Conservatives did not hold a majority of seats. Marked regional variations, the strength of the Liberals and the Nationalists mean the distribution of public opinion within different parts of the country will critically affect the election outcome 7 and yet the usual overall percen
Previous page
