Survey Accuracy as a Function of Usage Rate

Using a large-scale AT&T database, we decompose usage levels into frequency and duration, and then investigate how they affect survey response accuracy. Results show that more accurate surveys are obtained from those in the high frequency group (possibly because of the increased use of the more accurate rate-based method of recall and higher degrees of regularity), and from those in the longer duration group (possibly because of increased involvement). We therefore suggest that the survey method is less suited for light users because of higher error, and that alternative methods such as diaries might be more appropriate. We also found similarly consistent results with respect to self-perception of usage rate and the accuracy of survey recall.