Pattern-analytic approach to analysis of accidents due to human error: an application of the ortho-oblique-type binary data decomposition.

An ortho-oblique-type binary data decomposition based on data matrices providing relevant conditions of accidents is proposed as a means of classifying patterns of human error, and the computing procedures are described. The usefulness of the technique is shown by a numerical example of accidents in freight-car classification yard work. Forty-two severely injured victims were interviewed by psychologists and the contents of case reports were rearranged into a reliable binary data matrix which indicated the presence or absence of an interrelationship between each sample and each of 42 items of operational, environmental, and psychological conditions. Three patterns of error leading to an accident were identified by interpreting the orthogonally rotated results. They were 1) failure of strenuous performance in relation to fatigue and poor communication, 2) veterans' mistakes in teamwork, possibly due to hasty operation or distractions, and 3) errors caused by certain defects of machines or inappropriate work space. Recommendations were thus made on training plans for beginners. The importance of using non-scaled accidental data in binary form and allowing rotation for data decomposition is discussed.