HOW MIGHT OPINION POLLS BE IMPROVED - THE CASE FOR PROBABILITY-SAMPLING

A classification of sources of error in opinion polling is presented, and related to investigations of observed error in the polls at the ti me of the 1992 general election. This leads directly to implications f or reducing the potential for error in future. It is argued that quota sampling bias, incorporating both selection bias and unit non-respons e bias, is a major source of error. The adoption of probability sampli ng methods is proposed as a way to reduce this bias, and pollsters and their clients are encouraged to experiment with such methods. An exam ple of a practical probability sampling design is described. Other sou rces of error which have been identified, and for which methodological improvement is suggested, are the treatment of item refusals, the tre atment of those who say they 'don't know' for whom they will vote and the treatment of respondents' predictions of whether they will vote.