Computer Programs For Statistical Graphics In Safety Analysis

This paper describes three programs, currently in use within MSHA, for graphic analysis of accident frequency data. The first program, TREND, plots confidence bands for curved trends in data compiled over a period of time. In addition to detecting statistically significant upward or downward trends, the program can be used to determine whether an observation at a particular time deviates significantly from the historical pattern. The second program, POIPORP, plots interval estimates for the risk of accident in each of several environments, given an observed number of incidents and exposure time at each environment. POIPORP is designed to quantify the uncertainty implicit in incidence rates based on limited information and to facilitate statistically valid comparisons among the rates. The third program, CINDY, plots dynamically adaptive control charts for accident or injury experience at a given location. These charts are used to detect statistically significant departures from a changing standard or a previously established trend. All three programs are written in FORTRAN IV, utilizing DISSPLA and IMSL. They exemplify forms of statistical analysis that would not be practical without a computer. Use of the programs is illustrated with applications to coal mine fatality and injury experience.