Further evidence on the predictive accuracy of the verbal probability scale: The case of household bill payments in Australia

This study examined how well the verbal probability scale predicted shortterm penetration levels for methods of paying household bills. The overall error of prediction was 17 per cent (not 17 percentage points but 17 per cent of the estimated value). This is a good result in comparison to previous research that has used attitudinal, intentions or probabilistic measures. The verbal probability scale was markedly inaccurate for one particular bill payment method, but this is attributed to respondents' misinterpretation of the de.nition of this payment method when providing responses. The major source of error was respondents who gave a zero probability for using a particular method, but then did use that payment method in a subsequent fourweek period. This source of error is consistent with previous research on the prediction of future behaviour. The study also found that this particular source of error was not independent among respondents. Respondents who made this particular error of prediction for one bill payment method (ie gave a zero probability but then used that payment method) were more likely to do the same for another bill payment method. Overall, the results support the aggregate-level predictive ability of the verbal probability scale to estimate penetration levels. The study, however, highlights that all methods, whether attitudinal, intentions or probabilistic estimates of future demand, suffer from a degree of measurement error. If probabilistic estimates are to be used as an intermediate variable against which the impact of a marketing intervention is judged, the planned impact of that intervention would need to be large enough not to be confounded with measurement error.