Recent publications [31] employ a variety of methods for calculating upper limits and there is no universally accepted procedure [27,32,33]. We choose an approach similar to that first advocated by Feldman and Cousins [27]. This method has been since extended by Conrad et al. [34] to incorporate uncertainties in detector sensitivity and the background estimate based on an approach described by Cousins and Highland [35]. A further refinement of the Conrad et al. method by Hill [36] results in more appropriate behavior of the upper limit when the observed number of events is less than the estimated background, as is the case for the present measurement. We have adopted this method but note that Table 2 contains all of the numbers needed to calculate an upper limit using any of the methods in the papers cited above. We assume that the probability density functions of Fsens and background estimates are Gaussian-distributed.
