Thanks to ASQ for asking me to write this article for their “Statistics Digest” newsletter in October of 2019 – click here to access the whole issue if you're a member. I'm posting the article here with their permission.
Readers of this newsletter most likely understand Statistical Process Control. When I teach workshops on the “Process Behavior Chart” method that I learned from Donald J. Wheeler, PhD, the classroom population is quite often evenly divided. The common theme is that attendees want to improve the way they manage their organization. But, half of those in the class have never really even heard of SPC and the other half has learned, used, and even taught SPC through Six Sigma Black Belt or Master Black Belt programs.
Learning a new method, like Process Behavior Charts, can be challenging to some people (although they usually learn the method is not nearly as complicated as they might fear). But I've learned it's often more challenging for people to unlearn aspects of what they have previously been taught.
Those of us who are involved with Lean (or other improvement methodologies) are supposed to be on the leading edge of the battle against “the way we've always done it.” But, it's easy for people to get stuck in the trap of continuing to do things the way they've always done it (or the way they've always taught it). This can be the case even if “always” is only a year since their training and/or certification.
SPC is Only Applicable to Manufactured Parts Dimensions?
For example, those who learn classical SPC (in an engineering, statistics, or Six Sigma course) might be taught how to use control charts for the monitoring of the key dimensions of manufactured parts. Case in point, my first professional application of SPC was looking at key automotive engine block characteristics, like the machined diameter of a cylinder bore. SPC was already being used when I arrived at GM in 1995. SPC long pre-dates Lean (and it was used at GM long before their attempts to copy Toyota).
SPC is incredibly helpful for making sure we have (and continue to have) a predictable process that will make (and continue to make) parts with dimensions that will remain well within specification limits. This needs to be done without mistakenly confusing “control limits” (the “voice of the process”) with specification limits (the “voice of the customer).
I've heard recent pushback from some who say, “SPC is only appropriate for measured dimensions on manufactured parts. It's inappropriate for business metrics or other situations.” They're not making up a statement like that to be difficult – they're probably repeating something that was taught to them as fact or truth. Once somebody is anchored in the idea that SPC can't be (or shouldn't be) used for business metrics, it's understandably difficult for some people to give that up.
SPC Isn't Part of Lean Management?
Another example, from a Lean context, occurs when an organization is taught by a consultant that “Lean visual metrics” means posting a grid of monthly numbers on a wall. This is sometimes called a “bowling chart” since it looks a bit like a bowling scoresheet (often with 12 months instead of 10 frames).
These monthly scores are usually compared against a target (a type of one-sided specification limit) and they are often color coded as red (worse than target) or green (better than target). These bowling charts are often introduced in the context of methods with names like “Lean Daily Management” or “Strategy Deployment.” Sometimes, experts teach methods that aren't necessarily best practices
Often times, the expert who teaches methods like bowling charts has gotten stuck in their own form of “the way I've always done it” instead of challenging themselves to learn and include SPC in their methodology. Again, SPC pre-dates the Toyota Production System by decades and Toyota has incorporated SPC into their system, so why shouldn't others include SPC in their Lean practice?
If SPC was not mentioned in the Lean book(s) they have read, they might incorrectly reject SPC as being something that Toyota doesn't do (and, again, they do use SPC). To help prevent that problem, I included at least a brief mention of Process Behavior Charts in my first book, Lean Hospitals.
People who read my most recent book (Measures of Success) or attend my workshops can usually understand that, from a rational and logical perspective, that Process Behavior Charts are far more helpful than a bowling chart. Process Behavior Charts are far more visual, and the calculated limits and a few heuristic rules can allow us to detect statistical “signals” that show a system has improved (or gotten worse) in a meaningful way. Process Behavior Charts help us avoid reacting to “noise” in a metric. Looking for a “special cause” explanation for “common cause” variation is a waste of time that hampers our ability to actually improve the system.
Learning the method (how to create, interpret, and use the charts) is relatively easy, but people are complicated – and organizations are more so, since they are comprised of many naturally-complicated people. It's often difficult for people to admit that something they recently learned (bowling charts) might not really be the best way to do something.
They might say, “We can't admit that bowling charts aren't ideal. I mean, we just rolled those out across the entire organization this year… we can't throw another new method at people, no matter how much better it is. People have change fatigue!” That's all very well true… and it's understandable how that dynamic can get in the way of an organization adopting and embracing Process Behavior Charts.
One important lesson that I've learned in my career as an engineer and change agent is that the most-technically-correct solution doesn't always win. Having the “right answer” or a “best practice” isn't always enough to ensure the adoption of a technique or a technology. That's one reason why we probably need a formal change-management methodology instead of just throwing a bunch of people into training courses and then labeling people as “resistant to change” – blaming them as individuals instead of looking at the organization as a system. For more on this, look up a methodology called “Motivational Interviewing” – another method that Toyota teaches managers that doesn't appear in any of the books about Toyota or Lean.
Unlearning Some Six Sigma Training: Not Just Simpler, but Better
In a third scenario, the Six Sigma belts that I reach have previously been taught a number of different control chart methodologies, including the u-chart, c-chart, p-chart, and np-chart, along with a complex flowchart that guides you about which chart to use in which situation.
As I've learned from Professor Wheeler, the Process Behavior Chart (the XmR-chart) is a more recent invention compared to those other four charts. Wheeler calls the Process Behavior Chart the “Swiss Army knife” of control charts, which implies that it's useful enough for almost every purpose. I've heard some people point out that Swiss Army knives aren't effective for all jobs (like cutting down a tree), but the XmR-chart is arguably more applicable for our workplace purposes. Using the XmR-chart isn't “dumbing down” SPC, nor is trading off simplicity for validity.
The older four control charts are based on an assumption of binomial or Poisson distributions – an assumption that doesn't hold up in the real world. If we're charting the number of patients (or the proportion of them) that get hospital-acquired infections, we'd have to believe the probabilistic assumption that every patient is equally likely to get an infection. To chart the percentage of patients who leave the emergency department without being seen, we'd have to believe that each and every patient is equally likely to face a frustrating delay that would cause them to leave. My own personal experience and queuing theory tells me that is an unrealistic assumption.
One benefit of XmR-charts is that they are robust in the real world. These charts are not concerned about the underlying distribution of our data – they just work.
So, the Six Sigma belts (or some classically-trained statisticians) have previously learned about XmR-charts, but they are faced with the challenge of unlearning what they were previously taught, which is, again, a challenge.
Thankfully, when I teach Process Behavior Charts, I'm generally thanked by the Six Sigma belts. They sometimes admit they struggled with my challenging their previously-taught beliefs. But, they come around to expressing gratitude that having a single methodology (the XmR-chart) is not just more applicable, but is also easier to teach. “My executives' eyes glaze over by the time I start explaining the second type of control chart they'll need to learn. I think I can gain better acceptance around a single type of chart.”
In conclusion, the best technical solution doesn't always win in the marketplace of business ideas. Nor is the most complicated method (involving all of those different control charts) necessarily the best. Unlearning can be the most difficult part of the learning process, so it's important to expect that and to be empathetic and patient with those who are going through that process.
VISIT THE ASQ STATISTICS DIVISION WEBSITE:
What do you think? Please scroll down (or click) to post a comment. Or please share the post with your thoughts on LinkedIn. Don't want to miss a post or podcast? Subscribe to get notified about posts via email daily or weekly.