ABSTRACT
ROC methodology evolved from practical considerations within the field of radar signal-detection theory (Peterson et al., 1954). Although it drew upon theoretical developments in statistical quality control (Dodge and Romig, 1929; Shewhart, 1931) and statistical inference (Neyman and Pearson, 1933), it has always been firmly rooted in practical applications. Early uses came in psychological testing, but it rapidly became espoused within the field of medical test evaluation where it has found an enormous number of applications over the past twenty years. This wealth of medical examples has already been reflected by many of the illustrative analyses described in earlier chapters, and many more can be found in journals devoted not only to medicine but also to medically-related fields such as radiology, clinical chemistry, cardiology, or health and behavior studies (see, e.g., Collinson, 1998; Kessler, 2002; Obuchowski, 2003; Obuchowski et al. 2004). It ought perhaps to be added, though, that while ROC methodology has provided some enormous benefits in these areas, not all ROC studies have been adequately conducted so the interested reader must always be prepared to exercise critical judgement when considering them. In this regard, Obuchowski et al. (2004) provide a useful service. They conducted an exhaustive survey of all articles involving ROC curves that appeared in the journal Clinical Chemistry in 2001 and 2002, and summarized a number of shortcomings to look out for in similar studies.
