Circular Statistical Methods: Applications in Spatial and Temporal Performance Analysis.

摘要

This report surveys methods used for analyzing data generated from circular scales. The focus is on dimensions of simulation network training and development data that may be inaccessible to linear-based analyses. Several problem areas are identified that may lead to erroneous conclusions and/or loss of information when applying traditional analytic procedures to spatial and temporal performance measures. For example, the arithmetic mean and standard deviation are shown to be inappropriate descriptive measures for circular data. In addition, the use of average absolute deviations to measure directional errors of judgment can lead to loss of directional information. Finally, the usual methods of statistical inference are shown to fail in accounting for circularity when it exists. Therefore, these methods are subject to serious, often unknown and unrecognized errors in stated probabilities associated with Type 1 error rates, loss of statistical power, or both. Statistical methods are presented that help circumvent these problems.