Professional judgment in the Data Quality Objectives process: A Bayesian approach to screening assessment

摘要

The Data Quality Objectives (DQO) process provides a logical planning structure for specifying the optimal sample allocation for defensible decision making, depending on acceptable levels of decision uncertainty and anticipated sampling and measurement errors. These planning inputs must be established prior to designing the data collection activity. Application of the DQO process has traditionally been performed under the framework of Classical statistical theory; elicited decision errors have been interpreted as Classical Type I and Type II errors; mean and variance constraints have been incorporated based on historical information; and, Classical statistical testing methods have been used to determine optimal sample sizes. However, decision errors are usually stated, for dichotomous hypotheses, in terms of the probability of making a false positive or false negative decision; these probabilities, at best, relate loosely to probabilities of Classical Type I and Type II errors. Statements of Classical error types are couched in the language of the probability of rejection of hypotheses as opposed to the probability that a hypothesis is correct. Also, historical or archival data are often insufficient to adequately support prior judgments of means and variances. In many circumstances, however, expert knowledge and opinion is not only available, but is substantial. Finally, a paradigm that provides solutions for other than dichotomous decision problems offers greater diversity for solving real world problems.