Abstract: | A referendum question at the 1999 New Zealand general election resulted in a 92% 'Yes' vote in favour of reform of the New Zealand justice system. However, even supporters of the referendum conceded the question was poorly designed and open to different interpretations. To examine these interpretations, we deconstructed the referendum question into five constituent questions and tested these on a sample of potential voters. Support for the constituent questions ranged from 70% to 95%, and subsequent probing revealed that respondents' understanding and interpretation of the key concepts varied considerably. These findings emphasise that survey professionals need to be involved in designing referendum questions and that proposed questions need cognitive pre-testing to ensure they convey their intended meaning and can be understood by voters. A poorly designed question reduces the credibility of the referendum process and threatens the outcome of specific policy initiatives. |