Wheeler appears to be talking out of both sides of his mouth here. He prefers to call control charts process behavior charts, because their primary purpose is to show if the process is performing in an unpredictable manner — special cause, or predictably — common cause. But yet he is dead against transforming data WHEN APPROPRIATE to ensure they do just that.Cycle time data by its nature follows a highly skewed distribution, which when plotted on an ImR chart typically generates many data points above the UCL, even though the process is performing predictably. Since the chart has a number of points that have crossed above the “action limit”, following Wheeler’s approach would lead to tampering and an inevitable increase in variation.If the Cycle Time data is predictable, transforming it typically leads to more than a 10X reduction in the number of points above the UCL, allowing signals of true special cause to be seen above the noise of the distribution’s natural shape. Like most other things, I believe the correct answer lies somewhere in the middle. Data shouldn’t be blindly transformed, but if there is a solid existing understanding of its natural distribution, why shouldn’t that knowledge be applied to help provide a more accurate picture of process behavior?
.
