The Accuracy of Self-Reported Probabilities of Giving Recommendations

Abstract

This paper shows that respondents are better at predicting when they won't give a recommendation than when they will. The main reason for inaccuracy was an over-reliance on past circumstances (past receiving or giving of recommendations) in making future predictions of their own behaviour. Therefore, self-report probabilities are best used as measures of the potential or desire to give a recommendation, rather than predictions of future behaviour. The translation of this potential to behaviour will depend largely on the external environment, which is outside the respondent's control. To improve the accuracy of aggregatelevel predictions of how many people will give recommendations, we suggest reducing the number of those with a high self-reported probability to around 30% of survey estimates.