Revisiting activity sampling: a fresh look at binomial proportion confidence intervals
Abstract
The Wald interval is typically used to assign confidence to the accuracy of activity sampling studies. It is known the performance of the Wald interval is poor, especially when the observed probability is near zero or one. The suitability of the Wald interval for activity sampling is not often discussed in the operations management literature; if it is, this is usually followed by inappropriate and incorrect advice. Herein, a range of alternative binominal confidence intervals for activity sampling is reviewed. A number of selection criteria are considered including achievement of the target nominal coverage probability, size of the interval, and ease of use and presentation. It is recommended that the Clopper-Pearson interval is used for activity sampling. A table of confidence intervals and sample sizes that is specifically designed to be used within a new activity sampling procedure based on the Clopper-Pearson interval is developed. Finally, pedagogical issues are considered.