Proper Use of Accident Statistics

Incident rates should be plotted on a control chart in order to see if the process is changing over time. The first step is to establish a process mean, which would be the mean recordable rate over the past 12 months or so. The next step is to calculate control limits. The upper control limit is equal to the mean plus 3 standard deviation units as calculated by the U statistic. The lower control limit is the mean minus 3 standard deviation units. Of course it cannot go below zero. The mean and control limits are laid out on a chart, and each month the recordable rate is plotted. The following events indicate that the process has changed (Nelson 1986)  

Should any of these criteria be met, it is appropriate to try to find out what happened. If you have a run of points above or below the mean or several excursions outside of control limits, it is appropriate to compute a new 

It is critical that normal fluctuations of the measure, within the control limits, not be treated as signs that the process has changed. Acting on these random fluctuations as though they represent process changes is called tampering. Tampering not only wastes resources but usually makes the performance of the system worse. 

Upper control limit Lower control limit 

Referring again to our chart, we see another process shift beginning at month 33, with seven consecutive points below the previous mean. Again, the objective would be to have kept good records of actions taken and analyze those and any other potential causes for the mean to have taken another step down. Whether this is a continuing effect of the mitiatives described. 

---