In the 1980s, demand for SPC classes outstripped the supply of competent instructors. Novices were teaching neophytes, and misinformation could be found everywhere. Out of this chaos, many incorrect ideas about SPC became widely circulated. These ideas continue to be spread by those who don’t know any better and by those who use them to try to discredit SPC and sell more-complex techniques. Since these ideas can lead to the misuse of data, they’re hazardous in practice.

My colleague Allen Scott provided me with a list of many of these hazardous ideas that he found currently in circulation. Eight of these ideas are listed below, along with a brief explanation of the fallacies contained therein.

‘SPC is not needed with stable process’

It’s been said that once we have reengineered our process and gotten things working smoothly we no longer need to use SPC to monitor the process.

This hazardous idea combines a naive view of production with a misunderstanding of what SPC does. Real-world processes do not remain unchanged. Over time, entropy will change every process. This undermines the notion that a process that has been “fixed” will stay fixed. And while SPC may be used as a monitor to detect when to adjust a process, adjustment isn’t the main purpose of SPC. Many of these process changes will represent the effects of unknown causes. As such, they will be opportunities to discover new variables that, when controlled, will allow you to operate on target with reduced variation. So, contrary to the hazardous idea above, you need SPC more than ever when you have supposedly “fixed” your process.

If you’ve ever had this hazardous idea, you need to read about a process that was improved while being operated predictably in my Quality Digest article, “How Do You Get the Most Out of Any Process?”¹

‘SPC requires perfect data’

It’s also been said that you need to wait until your process is “well-behaved” before computing limits.

This hazardous idea is the opposite of the preceding one. It’s based upon a complete lack of understanding of how process behavior charts work. The computations in SPC are based on the same foundation as all other modern statistical analysis techniques. The correct computations, pioneered by Walter Shewhart in the 1920s, allow us to get good limits from bad data. In this way we can begin to learn from our data before we perfect our process.

If you’ve ever had this hazardous idea, read “When Can We Trust the Limits on a Process Behavior Chart?”², “Good Limits from Bad Data”³, and “Separating the Signal From the Noise.”⁴

‘SPC ignores mathematics’

Some say that SPC glosses over mathematical theory and gives inexact answers.

The many versions of this hazardous idea are built on what are perceived gaps in the development of SPC. The perception of these gaps begins with a failure to understand how SPC uses a different approach to the decision problem inherent in all data analysis.

When we analyze data, we have to filter out the probable noise to detect any potential signals that may be present. A traditional statistical analysis does this in the following way: A test statistic that is a function of the original data is computed. Using an appropriate probability model for this test statistic, we compute decision limits that correspond to some fixed risk of a false alarm (usually 5%). These decision limits are then used to evaluate the observed value of the statistic. If the observed value is outside the critical values, we say we have detected a potential signal.

However, Shewhart noted that when working with the original data we will never have enough data to fully specify a particular probability model. Without a probability model, the traditional approach above will not work. So Shewhart chose to use a different approach. He turned things around by starting with generic, fixed-width decision limits which would always result in a suitably small probability of a false alarm (less than 2%), regardless of which probability model might be appropriate.

To E. S. Pearson, and countless others since, this seemed like heresy. The gospel of statistical inference requires a fixed risk of a false alarm and customized decision limits. Shewhart proposed fixed-width decision limits and a small, variable risk of a false alarm. As Shewhart said, as long as we know the risk of a false alarm is small, we really don’t need to know the exact risk—we will still end up being right most of the time when we identify a potential signal.

To those with a traditional mindset, the notion of an unknown, variable risk of a false alarm challenges their very presuppositions about how to analyze data. Immediately, they seek to fill in the perceived gap to reconcile SPC with the traditional approach. On this journey, they usually start with a normal distribution, compute the risk of a false alarm for Shewhart’s limits, and assume that this was what Shewhart was trying to do. From this flawed beginning, they race off in various directions rather than pausing to examine their presuppositions and assumptions.

Pearson did this in 1935 when claiming that we needed the original data to be normally distributed to use a process behavior chart. Others have sought to compute probability limits based on an assumed probability model for the data. Still others talk about starting with Phase One charts and then progressing on to Phase Two charts. All of these efforts at reconciliation fail miserably, because Shewhart’s approach is the exact opposite of the statistical approach. Moreover, these “reconciliations” make the charts less useful rather than more useful because they change the objective. They turn the exercise into one of estimation before we even know if estimation makes sense. The purpose of Shewhart’s process behavior chart is the characterization of a process as being either disorganized (unpredictable) or predictable. Estimation makes no sense with a disorganized process.

Shewhart’s approach has been found to work in 100 years of practice. It doesn’t need to be reconciled to someone’s misunderstanding. It has the same mathematical foundation as other statistical analysis techniques. The rigor is there. Just because we don’t have to master the theory to use the technique doesn’t mean that the theoretical foundation is missing.

So if you’ve had any of the hazardous ideas that come from the tension between SPC and traditional statistics, or if you’ve been taught that SPC ignores mathematics, read “The Secret Foundation of Statistical Inference.”⁵

‘SPC is obsolete’

Some may say that SPC is a World War II technique in need of updating.

This hazardous idea ignores the very nature of mathematics. Calculus is a 17th-century idea. Does it need updating? The Pythagorean theorem is 2,500 years old. Does it need updating?

In fact, the essential concept behind SPC dates back to Aristotle, who told us that we can discover the causes that affect a system by looking at those points where the system changes. It is this ancient idea that Shewhart’s process behavior charts formalize by using the foundations of modern statistical analysis. Rigorous mathematical ideas do not become obsolete.

My books are full of examples from my clients where they’ve reduced variation, increased quality, reduced costs, increased productivity, and improved their competitive position using SPC. So SPC works. It’s been proven to be as powerful as any statistical technique can be. And it’s easier to use and understand than the various alternatives that have been developed over the past 100 years. When you have maximum power combined with ease of use, why go anywhere else? While there are those who are afraid of clarity because they fear their ideas may not seem profound, that’s no reason to use complex techniques when simple techniques will do the job. The best analysis is always the simplest analysis that provides the insight needed to take appropriate action.

For an illustration of a complex alternative to SPC read “The Cumulative Sum Technique.”⁶ For the basic mathematics behind all analysis techniques, read (again) “Separating the Signal From the Noise.”⁴

‘The range is inefficient’

It’s been said that the range statistic is obsolete and inefficient, and that modern measures of dispersion are more efficient.

This hazardous idea takes a statement that’s true for large data sets and applies it incorrectly. In SPC, we work with small subgroups where the range is essentially equivalent to the standard deviation statistic. This equivalence holds for subgroups of 15 or fewer values.

The graphs shown in Figure 1 superimpose the distributions of both statistics (after adjusting for their different biases). In each case, the differences between the two distributions shown for each value of n are too small to be of any practical interest.

Figure 1: Distributions of range and standard deviation statistics

Not only are the distributions essentially the same, but the individual statistics for a given subgroup are highly correlated with each other when n is less than 15.

Figure 2 shows the correlations between the bias-adjusted ranges (on the x-axis) and the bias-adjusted standard deviations (on the y-axis) for each of four sets of 100 subgroups.

Figure 2: Scatter plots of ranges and standard deviations

The correlation is 1.00 when n = 2, and it’s 0.996 when n = 3. Then, as n increases from 3 to 15, the correlation between the range and the standard deviation drops approximately 0.01 with each unit change in n. Thus, as long as your subgroup size is less than 12, the range and the standard deviation will have a correlation greater than 90%.

The fact that these statistics are equivalent for small subgroup sizes means that either one will work the same in process behavior charts. Those who prefer complexity may use the radius of gyration (which is what the standard deviation statistic is), while those who prefer transparency may continue to use the range.

So if you’ve been taught this hazardous idea, you can relax. The simplicity of the range may irritate those who love complexity, but it doesn’t degrade the quality of the charts. The requirement of rational subgrouping pushes us toward using small subgroups, and in this context the range is efficient, sufficient, and easy to use.

‘SPC replaces specifications’

Some have asked, “How can a process be ‘out of control’ when everything is conforming?” or “How can we be ‘in control’ when we’re making nonconforming product?”

This hazardous idea confuses the predictability of the process with the conformity of the product stream. It comes from the way people use the word control. To many, out of control means out of specification, and in control means within specifications. This is one reason I dropped the use of the term control chart in favor of the more descriptive process behavior chart.

Specifications are for sorting the good stuff from the bad stuff at the end of the line. Process behavior charts characterize the process as either being operated predictably or being operated unpredictably. They tell when and how to take action on the process, rather than telling us how to sort the product.

If you have used in control to refer to conforming product, read “Two Definitions of Trouble.”⁷

‘Only reengineering prevents defects’

It’s been said that a process that produces too much nonconforming product needs to be reengineered.

This hazardous idea implicitly assumes that once a process is fixed it will stay fixed. It also assumes that the process engineers will do a better job with a remedial approach than they were able to do initially.

Nonconforming product is primarily a result of having too much variation in the product stream. Reduce the variation and you will tend to reduce the amount of nonconforming product. But the variation in the product stream is a function of the uncontrolled process inputs. Some of these uncontrolled variables may be known, while many others will be unknown, forgotten, or overlooked. Reengineering projects can study the uncontrolled variables that are known, but they can’t experiment with the unknown variables. So, once again, we need to turn the traditional approach around by 180 degrees. Rather than trying to study specific process inputs (i.e., reengineering), we need to let the process guide us to discover the dominant uncontrolled variables (SPC). Process behavior charts help you to identify the presence of assignable causes of excessive variation by tracking the process outcomes rather than experimenting with specific process inputs. So if you have had this hazardous idea, read “When Things Go Wrong.”⁸

‘Inspection removes the need for SPC’

Some say that if they do 100% go/no-go gauging they don’t need to waste time measuring the parts.

This hazardous idea confuses the role of inspection with what is needed to operate the process.

One of my clients had a bearing that was produced on any one of three machines. Problems with nonconforming parts led to the following study. Fifty bearings were made on each machine and labeled with a paint pen. After inspecting them with the go/no-go gauge, the diameter of each bearing was measured. Each machine had seven out of 50 bearings fail the go/no-go gauge, for an observed 14% nonconforming. However the measurements produced the histograms shown in Figure 3.

Figure 3: Bearing diameters for three machines

While the three machines have the same fraction nonconforming, the parts they produced are not all alike. The three machines are operating differently. Moreover, when these data are placed on XmR charts, we see that while they’re all being operated unpredictably, they’re unpredictable in different ways.

Figure 4: X charts for bearing diameters

After looking at the charts in Figure 4, the supervisor said that he was taking his critical jobs off Machine 3 until he could get it overhauled. Measurements always contain much more information than counts, and process behavior charts reveal things that no other analysis will reveal.

Summary

It’s easy to find software that will dazzle you with all the barnacles of new graphs and complex computations that have been placed on the back of SPC. Yet the fundamentals are still the key to process improvement. Collect your data. Organize them rationally based on the principles of rational sampling and rational subgrouping. Place them on a process behavior chart and start learning what your process can tell you. As you identify assignable causes of exceptional variation and make them into controlled inputs for the process, you’ll find your process operating on-target with minimum variance. And that is world-class quality.

Donald J. Wheeler’s complete “Understanding SPC” seminar may be streamed for free; for details, see spcpress.com.

Footnotes

¹ “How Do You Get the Most Out of Any Process,” Quality Digest, Nov. 2016, or spcpress.com/pdf/DJW300.pdf

² “When Can We Trust the Limits on a Process Behavior Chart?” Quality Digest, May 2009, or “Don’t the Outliers Distort the Limits,” spcpress.com/pdf/DJW198.pdf

³ “Good Limits From Bad Data,” Quality Digest, July 2009, or spcpress.com/pdf/DJW199.pdf

⁴ “Separating the Signal From the Noise,” Quality Digest, Oct. 2013, or spcpress.com/pdf/DJW260.pdf

⁵ “The Secret Foundation of Statistical Inference,” Quality Digest, Dec. 2015, or spcpress.com/pdf/DJW288.pdf

⁶ “The Cumulative Sum Technique,” Quality Digest, Oct. 2022, or spcpress.com/pdf/DJW405.pdf

⁷ “Two Definitions of Trouble,” Quality Digest, Nov. 2009, or spcpress.com/pdf/DJW203.pdf

⁸ “When Things Go Wrong,” Quality Digest, April 2026