Variation is the enemy! It can introduce waste and errors into a process. The more there is, the more errors. The more errors, the more waste.
What is Variation?
Quick answer: it’s a lack of consistency. Imagine that you’re manufacturing an item. Say, a certain sized screw. Firstly, you want the parameters to be the same in every single screw that you produce. Material strength, length, diameter and thread frequency all need to be uniform. Secondly, your customers want a level of consistency. They want a certain size of screw to all be the same. Using a screw that’s the wrong size might have serious consequences in a construction environment. So a lack of consistency in our products is bad.
We call the differences between multiple instances of a single product variation.
(Note: in some of Game Change Lean Six Sigma’s videos they misstate six sigma quality levels as 99.999997 where it should be 6 sigma = 99.99966 % )
Why Measure Variation?
We measure it for a couple of reasons:
- Reliability: We want our customers to know that they’ll always get a certain level of quality from us. Also, we’ll often have a Service Level Agreement or similar in place. Consequently, every product needs to fit specific parameters.
- Costs: Variation costs money. So to lower costs, we need to keep levels low.
Measuring Variation vs Averages
Once, companies tended to measure process performance by average. For example, average tensile strength or average support call length. However, a lot of companies are now moving away from this. Instead, they’re measuring variation. For example, differences in tensile strength or in support call lengths.
Average measurements give us some useful data. But they don’t give us information about our product’s consistency. In most industries, focusing on decreasing fluctuations in processes increases performance. It does this by limiting factors that cause outlier results. And it often improves averages by default.
How Do Discrepancies Creep into Processes?
Discrepancies occur when:
- There is wear and tear in a machine.
- Someone changes a process.
- A measurement mistake is made.
- Material quality or makeup varies.
- The environment changes.
- A person’s work quality is unpredictable.
There are six elements in any process:
- Mother Nature, or Environmental
- Man, or People
In Six Sigma, these elements are often displayed like this:
Discrepancies can creep into any or all elements of a process.
To read more about these six elements, see 5 Ms and one P (or 6Ms).
For an example of changing processes contrarily causing variation, see the Quincunx Demonstration.
Process spread vs centering
Types of Variation
There are two basic types which can occur in a process:
- common cause
- special cause.
Common cause variation happens in standard operating conditions. Think about the factory we mentioned before. Fluctuations might occur due to:
- metal quality
- machine wear and tear.
Common cause variation has a trend that you can chart. In the factory mentioned before, product differences might be caused by air humidity. You can chart those differences over time. Then you can compare that chart to weather bureau humidity data.
Conversely, special cause variation occurs in non standard operating conditions. Let’s go back to the example factory mentioned before. Disparities could occur if:
- a substandard metal was delivered.
- one of the machines broke down.
- a worker forgot the process and made a lot of unusual mistakes.
This type of variation does not have a trend that can be charted. Imagine a supplier delivers a substandard material once in a three-month period. Subsequently, you won’t see a trend in a chart. Instead, you’ll see a departure from a trend.
Why is it Important to Differentiate?
It’s important to separate common cause and special cause because:
- Different factors affect them.
- We should use different methods to counter each.
Treating common cause as special cause leads to inefficient changes. So too does treating special cause like common cause. The wrong changes can cause even more discrepancies.
How to Identify
Use run charts to look for common cause variation.
- Mark your median measurement.
- Chart the measurements from your process over time.
- Identify runs. These are consecutive data points that don’t cross the median marked earlier. They show common cause variation.
Meanwhile, use control charts to look for special cause variation.
- Mark your average measurement.
- Mark your control limits. These are 3 standard deviations above and below the average.
- Identify data points that fall outside the limits marked earlier. In other words, above the upper control limit or below the lower control limit. These show special cause variation.
Variation is the square of a sample’s standard deviation.
Variation = SD2
How to Find the Cause of Variation
So far, you’ve found any significant variation in your process. However, you haven’t found what its cause might be. Hence, you need to find the source.
You can use a formal methodology like Six Sigma DMAIC, Use a multi vari chart to identify the source of variation.
How to Find and Reduce Hidden Causes of Variation
DMAIC methodology is the Six Sigma standard on how to identify variation in a process, analyze the root cause, prioritize the most advantageous way to remove a given variation, and test the fix. The tools you would use depend on the kind of variation and the situation. Typically we see either a “data door” or a “process door” and use techniques that are most appropriate.
For a smaller, shorter cycle methodology you could try Lean tools like Kaizen or GE’s Work Out.
How to Counter Variation
Once you identify its source, you need to counter it. As we implied earlier, the method you use depends on its type.
Counter common cause variation using long term process changes.
Counter special cause variation using exigency plans.
Let’s look at two examples from earlier in the article.
- Product differences due to changes in air humidity. This is common cause variation.
- Product differences due to a shipment of faulty metal. This is special cause variation.
Countering common cause variation
As stated earlier: to counter common cause variation, we use long term process changes. Air humidity is a common cause. Therefore, a process change is appropriate.
We might subsequently introduce a check for air humidity. We would also introduce a following step. If the check finds certain humidity levels, then change the machine temperature to compensate. The new check would be run several times a day. Whenever needed, staff would change machine temperature. These changes then lengthen the manufacture process slightly. However, they also decrease product differences in the long term.
Countering special cause variation
As we said earlier: to counter special cause variation, we need exigency plans. These are extra or replacement processes. We only use them if the special cause is present, though. A large change in metal quality is unusual. So we don’t want to change any of our manufacture processes.
Instead, we implement a random check of quality after every shipment. Then, an extra process to follow if a shipment fails its quality check. The new process involves requesting a new shipment. These changes don’t lengthen the manufacture process. They do add occasional extra work. But extra work happens only if the cause is present. Then, the extra process eliminates the cause.
Rather than finding variation in a single sample, you might need to figure out combined variance in a data set. For example, a set of two different products. For this you’ll need the variance sum law.
Firstly, look at whether the products have any common production processes.
Secondly, calculate the combined variance using one of the formulas below.
No shared processes
The two products don’t share any production processes? Great! Then you can use the simplest version of the variance sum law.
Variance(X + Y) = Variance(X) + Variance(Y) Variance(X - Y) = Variance(X) + Variance(Y)
The two processes do share some or all production processes? That’s OK, you’ll just need the dependent form of the variance sum law instead.
Variance(X + Y) = Variance(X) + Variance(Y) + Covariance(X,Y) Variance(X - Y) = Variance(X) + Variance(Y) - Covariance(X,Y)
Calculate covariance using the following formula.
Cov(X,Y) = Σ ((X-μ) * (Y-ν)) / n-1
- μ is the mean value of X.
- ν is the mean value of Y.
- n = the number of items in the data set.
What You Need to Know for Your Six Sigma Exam
Combating variation is integral to Six Sigma. Therefore, all major certifying organizations require that you have substantial knowledge of it. So let’s walk through how each represents what they expect.
ASQ Six Sigma Green Belt
ASQ requires Green Belts to understand the topic as it relates to:
Exploratory data analysis
Create multi vari studies. Then interpret the difference between positional, cyclical, and temporal variation. Apply sampling plans to investigate the largest sources. (Create)
IASSC Six Sigma Green Belt
IASSC requires Green Belts to understand patterns of variation. Find this in the Analyze Phase section.
Villanova Six Sigma Black Belt
Villanova requires Black Belts to understand the topic as it relates to:
Six Sigma basic premise
Understand the difference between assignable cause and common cause variation along with how to deal with each type.
Multi vari studies
Create and interpret multi vari studies to interpret the difference between within piece, piece to piece, and time to time variation.
Calculate, analyze, and interpret measurement system capability using repeatability and reproducibility, measurement correlation, bias, linearity, percent agreement, precision/tolerance (P/T), precision/total variation (P/TV), and use both ANOVA and control chart methods for non-destructive, destructive, and attribute systems.
ASQ Six Sigma Black Belt
ASQ requires Black Belts to understand the topic as it relates to:
Use and interpret multivariate tools such as principal components, factor analysis, discriminant analysis, multiple analysis of variance etc to investigate sources of variation.
Use and interpret charts of these studies and determine the difference between positional, cyclical and temporal variation.
Attributes data analysis
Analyze attributes data using logit, probit, logistic regression, etc to investigate sources of variation.
Define and describe the objectives of SPC, including monitoring and controlling process performance, tracking trends, runs, etc and reducing variation in a process.
IASSC Six Sigma Black Belt
IASSC requires Black Belts to understand patterns of variation in the Analyze Phase section. It includes the following:
- Multi vari analysis.
- Classes of distributions.
- Inferential statistics.
- Understanding inference.
- Sampling techniques and uses.
Candidates also need to understand its impact on statistical process control.
ASQ Six Sigma Black Belt Exam Questions
Question: A bottled product must contain at least the volume printed on the label. This is chiefly a legal requirement. Conversely, a bottling company wants to reduce the amount of overfilled bottles. But it needs to keep volume above that on the label.
Look at the data above. What is the most effective strategy to accomplish this task?
(A) Decrease the target fill volume only.
(B) Decrease the target fill variation only.
(C) Firstly decrease the target fill volume. Then decrease the target fill variation.
(D) Firstly decrease the target fill variation. Then decrease the target fill volume.