Election forecasting: Development of the Constant Sum Scale to be used in telephone surveys

University of Otago


Probability-based scales have been used successfully to estimate future behaviour in marketing. The probabilistic nature of these scales allows the means produced to be interpreted in proportionate terms. At least three forms of probability scale – Juster Scale, Verbal Probability Scale and Constant Sum Scale – are used in the marketing literature. The Juster Scale, originally implemented in face-to-face surveys, was developed to estimate the adoption rate of durables (Juster 1966). Researchers subsequently developed this scale for implementation in self-completion questionnaires applied via mail (Gendall et al. 1991) and internet-based surveys (Parackal & Brennan 1999). The Verbal Probability Scale is an adaptation of the Juster Scale for implementation in telephone surveys (Brennan et al. 1995a). This scale was tested successfully for estimating adoption rates, purchase frequencies and voting behaviour. In recent years, the Verbal Probability Scale has been tested to benchmark public issues in a social marketing context (Sharp & Riebe 2000; Parackal et al. 2007). The Constant Sum Scale was specifically developed for collecting data for mutually exclusive behaviours (Metefessel 1947). This scale is recommended to estimate switching behaviour, market shares and customer preferences (Metefessel 1947; Reibstein 1978; Hoek & Gendall 1997). It uses an approach that requests respondents to distribute a set number of tokens (usually ten) across mutually exclusive options. The scale produces data that logically represent the adoption or preferences of the mutually exclusive options. The summation of the means across the options always adds up to 10 or 1. Consequently, the means can be interpreted in terms of proportions of the sample that favour the respective options. This latter feature of the Constant Sum Scale is something that the other probability scales have failed to achieve when used on mutually exclusive options (Hoek & Gendall 1993, 1997; Parackal & Brennan 1999; Flannelly et al. 2000a, 2000b).