laboratory interaction (pl) (Error Type 3). To achieve enhanced prediction/accuracy through the use of serial testing, a single occasion would likely suffice as long as the specimen was sent to multiple laboratories. For Out-come 4, the three way interaction term (pol) explains most of the variance and illustrates a possible failure to specify and sample across relevant sources of error. Since the three-way interaction contains un-modeled error as well as the three-way interaction, this outcome may indicate that the variables investigated are not re-lated to observed variation in test results.
4. Discussion
Although the testing problem presented within this hy-pothetical example focused on the interpretation of a hypothetical diagnostic biomedical test, G theory meth-odology coupled with Bayes serial estimations has much broader application. For example, many concerned con-stituents are currently attempting to assure the accurate and fair use of tests in employment, sports eligibility, and in making sanction decisions. In all of these contexts, issues of fairness have arisen due to the large proportion of false positive results and the high stakes nature of the test results. There is considerable interest in increasing the accuracy of test evidence for making important deci-sion or a diagnosis. In addition, in many instances the data for such analyses may already exist since medical testing companies when seeking FDA approval for a par-ticular test must submit the test to a series of trials.
It is obvious from governing board recommendations and from published legal advice that test users are aware that retesting might reduce error. However, recommen-dations for retesting are usually made without statisti-cally estimation of the utility of retesting. Suggestions that samples be divided into multiple collection tubes, or that the test be repeated implies an expectation of in-creased precision with repeated testing. Unfortunately, when the sources of error are not systematically esti-mated, the usefulness of a particular retesting protocol is currently unknown.