Some effects of question wording and question administration on the prediction of voting behaviour : a thesis presented in partial fulfilment of the requirements for the degree of Doctor of Philosophy in Marketing at Massey University
The comparatively short history of polling has seen numerous instances of predictions which failed to match the final election outcome. Of the many possible explanations put forward to explain the divergencies that have occurred, the potential for error introduced by the undecided group has come under increasing scrutiny. A series of surveys conducted prior to the 1993 New Zealand General Election examined the effects of three question wording or question administration variables: contextual questions; an increase in the level of confidentiality, and a probability scale, on the proportion of undecided respondents and on the accuracy of the estimates obtained. In addition, the research examined the effects of various weighting procedures. The surveys also used a probability scale to estimate respondents' likelihood of voting. The findings suggest that, although contextual questions, the use of a secret ballot and the use of a probability scale will reduce or eliminate the undecided group, this reduction is not synonymous with an increase in the accuracy of the estimates obtained. Although proportional allocation of undecided respondents resulted in more accurate voting intention estimates than the treatments examined, some weighting factors appeared to improve the accuracy obtained by proportional allocation alone. While replication research is required, the probability scale provided accurate estimates of turnout.