Comparison of election predictions, voter certainty and candidate choice on political polls

Previous research has demonstrated the value of using 0-10 probability scal es to predict election results. Three experiments are presented that confir m the effectiveness of probability scales for political palling, and compar e their predictions with those based on forced-choice scales. The results o f these experiments show that using probability scales in telephone surveys reduces the number of undecided responses by half, compared with forced-ch oice scales, regardless of whether the surveys are conducted two weeks or s ix months before an election. They also show that responses to the question 'Who would you vote for if the election was held today?' contain a degree of uncertainty about future behaviour that is a source of error in politica l polls which is not accounted for by sampling error. The results suggest t hat this uncertainty increases the farther away in time the poll is conduct ed from the day of the election.