Self-Critical or Generalized Likelihood Analysis of Variance with Application to Prothrombin Times

摘要

The purpose of this work was to develop procedures in the hemotology proficiency testing program of the New York State Department of Health. Approximately 400 laboratories 400 laboratories in New York participate in a program of clinical laboratory proficiency testing, with each laboratory using one of eight methods and one of nine thromoplastins. The responses of each testing program may be modeled as a two-way layout, some cells of which are empty. A model-critical analysis of variance technique was used to determine simultaneously, and in the presence of out-of-control laboratories i.e., outliers, the effect of method and thromboplastin on prothrombin time. The word model-critical indicates that, according to model, set up initially on a tentative basis so as to allow for further evolution, the responses should have the structure of a two-way layout with interaction, and that the responses should have a common error distribution. This tentative framework is examined critically by varying the way observational information is processed to produce parametric summaries. If the summarizations are insensitive to the variation in information processing, then the tentative model stands; if not, the tentative model must be evolved. We provide an objective means of statistically assessing variations in summarization.