Bayesian Estimation of Undetected Errors

摘要

An unknown number, N, of errors exist in a certain product, for example, defects in a production lot, errors in a manuscript, or bugs in a computer program. I inspectors with possibly different competencies are to be put to work to find the errors. How should the inspection be organized, and what is a good estimate of the undetected errors (or of N). This problem is similar to the capture-recapture sampling problem of population biology, assuming a closed population and a parallel search effort, for which many classical results are available. Apart from an elementary analysis of the I = 2 case by Gaskell and George and some sequential sampling plans by Freeman, the only Bayesian approach to this problem appears to be by Castledine, who obtains rather complicated results appropriate to the population biology model. This paper, develops the model in a manner more related to error detection problems.