Effective Methods for Analyzing Digital Data

The digital data that are produced and stored by computers can not only be used to generate analog displays that detect phenomena of interest, but can also be used to analyze with great precision the patterns that have been revealed. In the testing of hypotheses, e.g., that a particular phenomenon of interest is present, one might deal with either continuous (smoothly varying) or dichotomous (yes or no) data. The use of receiver operating characteristic (ROC) curves is excellent for evaluating the former and tests that compare proportionalities (such as the chi square test) are appropriate for the latter. The choice of optimal diagnostic criteria for various phenomena of interest can be guided by comparing ROC curves and by calculating such parameters as positive and negative likelihood ratios and relative odds.