At the 2018 Magnet Nursing conference in Denver, I saw many improvement posters using line or bar charts with an added trend line to show improvement. Unfortunately, few of the trends were valid. Here’s why:
Latest "Statistics" Posts
Too many quality professionals are clinging to the way things have always been done. There’s some sort of taboo about doing things quickly and easily. What are the differences between 20th and 21st Century Quality? Watch and find out:
Statisticians say “correlation does not equal causation,” but what if we can prove cause and effect using new tools? The Book of Why, by Judea Pearl explains how this new science of cause and effect has shifted statistics over the last 30 years.
Another great webinar where Jay Arthur shows some of the capabilities of QI Macros to a group of folks who signed up and attended. Some were familiar with the software and already use it, and some had not yet begun to use it; all were interested in learning new ways that QI Macros can help them with their Quality Improvement efforts. (You can hear him answering questions typed in by webinar attendees.)
I’ve noticed a disturbing trend. Customers call us wanting to know all of the background about the hows, whys and formulas of a chart. I think they are afraid someone will challenge or ridicule their analysis. Here’s what I think.
People have been trying to make statistics simple and easy to understand for decades.
But statistics aren’t simple. Maybe we should change how we teach them?
I’m here at the IISE (Institute of Industrial and Systems Engineers) conference in Pittsburgh.
One professor had been teaching students how to use Excel to create control charts, but he was beginning to feel like that was a waste of classroom time (duh!).
I beat him up a little for teaching DIY Excel stuff to students. If the professor does it, they think that’s how it’s done. With QI Macros he can get them right into analysis.
I feel the same way at ASA (American Statistical Association) when they use “R” to do statistics. Sure it’s free, but should statisticians be programming in “R” or just using software to achieve the same result.
I have found that an XmR chart is the easiest way to display attribute data. Simply convert the numerator/denominator into a ratio and plot the ratio.
- defects per day could be a c chart, but an XmR chart works just as well
- defects/samplesize could be np, p or u chart, but XmR chart works just as well using the ratio
Almost two decades ago, Tom Pyzdek said: X chart provides an excellent approximation to the p chart.
More recently, Donald Wheeler noted that XmR chart limits will be very close to c, np, p or u chart limits if the underlying distribution is correct.
Customer asked me what seemed like a strange question: What specification limits do I need to get a Cp greater than one? Usually her customer should set specification limits, but her boss wanted to know what they could deliver. Hmmmm!
Then I realized that since QI Macros templates (e.g., XmR chart) calculate the average and sigma estimator, the LSL/USL for Cp = 1.0 would be:
LSL = Average – 3*SigEst USL = Average+ 3*SigEst
For Cp = 1.33, just change the 3 to a 4; Cp = 1.66, change the 3 to a 5. Here’s an XmR chart template with some sample data and calculations to reverse engineer spec limits:
At IHI, I spoke with one Quality Improvement Adviser who was getting blow back from certain statisticians in her organization. They were challenging her methods and tools and analysis.
As I have written before, I call these folks “Stat Bastards” because they belittle others rather than help them.
As a member of the American Statistical Association, almost every statistician I meet is incredibly kind, generous and helpful. The few that flaunt their training rather than helping are impeding the forward progress of their organization.
I say, if they have better data, bring it. If they have better analysis, bring it. If they have better tools, bring it.