Dr Wheeler is the world's greatest living Quality expert. He was mentored for 2 decades by Professor Deming. Con men hawking fads, farce and fraud steer their victims away from Dr Wheeler, because he tells the truth. For example, Dr Wheeler calls Six Sigma "GOOFY" and "the stuff of the tooth fairy". He exposes OEE as nonsense.
"MODERN" APPROACH
One outspoken antagonist of Dr Wheeler claims there are "better tools" than histograms which he claims are "old hat". He criticises the old "jagged" histogram and talks about "fitting a probabilistic model" to them.
Instead of histograms, he recommends kernel density estimation (KDE). He claims these are "richer" and "easier". The idea is that because you might have many data points and a computer, you should torture the data before plotting. Without even asking anything of your data, expect it to confess to anything.
KDE's plot an artificial distribution for each point, instead of the actual data for each point. The many distributions are then added. What distribution you might ask. Take your pick. Any height and width will do (bandwidth selection). Depending on what you choose, you can make your KDE look like just about anything.
The image shows 2 different bandwidths, compared to a histogram for the same data. Key information in the histogram is lost.
OPERATIONAL DEFINITION
Like many others, the antagonist does not understand the purpose of the histogram and is unaware of Professor Deming's Operational Definition.
The first step before collecting data or plotting anything, is to ask step 1 of the Operational Definition, "What are you trying to achieve?" Suppose for example, we are trying to choose between 2 suppliers. The data for supplier No1's parts show a histogram that is truncated at one side. Supplier No2 shows a narrow, skewed plot.
We immediately conclude from the histogram that supplier No1 is meeting our spec by inspecting out bad product. Supplier No2 has a far better controlled process. We choose supplier No2 because less variation would be introduced to our process and better quality will result. This is exactly analogous to the Ford transmission study, when Ford engineers were stunned by the quality of Mazda parts. https://lnkd.in/gksGSURX
KEEP IT SIMPLE
By contrast, if we had used the "modern" KDE's, the truncation of supplier No1 data would have been smeared. We would see nice smooth curves ... that failed to answer our question.
It is pointless to attempt to "fit a probabilistic model" to the data. This provides no benefit. It does not answer our question. Many blindly try to fit a normal distribution over the histogram. This is utterly meaningless. It hides the story the histogram is telling.
NEVER draw distributions over a histogram.
Keep it simple. Let the histogram tell its story.
by Dr Tony Burns BE (Hon 1) PhD (Chem Eng)