Introduction
In our previous lesson, we established the fundamental principles of Design of Experiments (DOE), covering concepts like factors, levels, responses, and the critical roles of randomization and blocking. You now understand why DOE is a more powerful and efficient method for experimentation than the traditional One-Factor-At-a-Time (OFAT) approach.
Now, we'll put that theory into practice. Imagine you've just completed a brainstorming session with your team using a Fishbone diagram to identify potential causes of a persistent defect. You likely have a long list of potential factors—perhaps 8, 10, or even more. Testing all of them in a full experiment would be incredibly time-consuming and expensive. This is a common challenge in the Analyze and Improve phases of a DMAIC project.
This lesson addresses that exact problem. Our learning outcome is to design and analyze a screening experiment (e.g., Plackett-Burman) to identify the vital few factors from a large list of potential variables. These highly efficient experiments act as a filter, allowing you to quickly and economically determine which factors have the most impact on your process output, so you can focus your efforts where they matter most.
1. The Need for Screening: Finding the "Vital Few"
In your work with Lean and Six Sigma, you're familiar with the Pareto principle (the 80/20 rule). It suggests that in most processes, a small number of factors (the "vital few") are responsible for the majority of the variation or effect, while most factors (the "trivial many") have little to no impact.
The primary goal of a screening experiment is to identify this vital few from a long list of candidates. Let's consider a practical manufacturing scenario.
This article from JMP, a statistical software company, introduces a great example. An engineering team wants to understand what influences the hardness of a new polymer material. They've identified 10 potential factors.
Please read the section titled 'Example of a 10-factor Plackett-Burman design in 12 runs'. Notice the calculation for a full experiment (2^10 = 1,024 runs) and why that's unrealistic. This clearly illustrates the problem that screening designs solve.
As you saw, running over a thousand experiments is not feasible. Screening designs offer a solution by drastically reducing the number of required runs, making experimentation practical even with many factors.
2. Understanding Plackett-Burman Designs
One of the most common and effective types of screening experiments is the Plackett-Burman design. It's a classic tool that every Green and Black Belt should know.
The key features of a Plackett-Burman design are:
- Efficiency: It allows you to study a large number of factors (k) in a very small number of runs (N), where N is a multiple of 4.
- Focus on Main Effects: Its primary purpose is to estimate the main effect of each factor—the change in the response when a factor is moved from its low to its high level.
However, this efficiency comes with a critical trade-off.
The Trade-Off: Resolution and Confounding
In our last lesson, we briefly mentioned interactions. A Plackett-Burman design is a Resolution III design. This is a technical term with a very important practical meaning:
- Main effects are not confounded (mixed up) with each other. You can tell Factor A's effect apart from Factor B's.
- However, main effects are confounded with two-factor interactions. This means the effect you see for Factor A might actually be a combination of Factor A's true effect plus the effect of the B*C interaction, the D*E interaction, and others.
For a screening experiment, we make a strategic assumption: we assume that at this stage, any two-factor interactions are small and negligible compared to the main effects. The goal is simply to find the factors with the biggest signals. If a factor shows a large effect, we will investigate it—and its potential interactions—in a later, more detailed experiment.
The JMP article you just read has an excellent, clear explanation of this trade-off. It's important to grasp this concept fully.
Please reread the sections 'What is a Plackett-Burman design?', 'When should you use a Plackett-Burman design?', and 'Why (and how) do you use a Plackett-Burman design?'. Pay close attention to the paragraphs on 'Resolution' and 'Confounding'.
3. How to Design and Analyze a Screening Experiment
Let's get practical. How do you actually set up and analyze one of these experiments? We'll use a common manufacturing problem as our guide. Most statistical software packages (like Minitab or JMP) make this process straightforward.
Creating the Design
First, you need to define your factors and levels and have the software generate the experimental plan, or "design matrix." This matrix tells you exactly which combination of high (+1) and low (-1) levels to use for each run.
Create Screening Design : Design of Experiment (DOE) With Example (Definitive & Plackett-Burman)
This video from 'LEARN & APPLY' demonstrates how to create a screening design in Minitab. While it also covers a more modern design type (Definitive Screening), we'll focus on the classic Plackett-Burman approach, which is fundamentally similar and widely used.
Watch the following segments: Example Context (01:34 - 02:03): This sets up a relatable manufacturing problem: optimizing the power output of an ultrasonic cleaner by studying 7 factors. Creating the Design (03:27 - 04:46): This shows the step-by-step process for creating a Plackett-Burman design in older Minitab versions. The logic applies to any software: select the design type, specify the number of factors, and define the factor names and levels. Notice how the software automatically generates the run sheet.
The output is a randomized run sheet that you use to collect your data. Once you've performed the experiments and recorded the response for each run, you're ready for analysis.
Analyzing the Results
The goal of the analysis is to separate the "vital few" from the "trivial many." Statistical software gives you several tools to do this.
Before analyzing, it's critical to ensure your data is valid. As we discussed in previous modules, this includes having a capable measurement system (Gage R&R) and ensuring data is collected properly.
Analyze Screening Design: Design of Experiment (DOE) With Example (Definitive & Plackett-Burman)
Now, let's analyze the data from the ultrasonic cleaner experiment. This video walks through the analysis and interpretation, which is the most important part.
Watch the following segments: Analysis Steps (04:07 - 05:29): Follow the steps for analyzing the design in Minitab. The key action is entering your collected response data (e.g., 'Power Output'). Interpreting Results (05:29 - 07:14): This is the crucial part. Focus on how the Pareto chart is used to visually identify the significant factors (those that cross the red line). The narrator also explains how to use the p-value (which you'll remember from our hypothesis testing lessons) to statistically confirm significance.
As the video showed, the analysis boils down to identifying factors with statistically significant effects. In the example, the factors 'Train' (A) and 'Quiet' (D) were significant. You started with 7 potential factors and have now successfully screened them down to just 2.
4. Interpretation and Next Steps
You've run your screening experiment and identified the vital few factors. What now?
It's important to remember what a screening design doesn't tell you. Because of confounding, it doesn't give you a clear picture of interactions. Therefore, the result of a screening design is not a final predictive model, but rather a focused list of factors for further study.
The next logical step is to conduct a follow-up experiment using only the significant factors. This could be a full factorial or fractional factorial design, which we will cover in our next lesson. These designs are specifically created to help you understand not only the main effects but also the critical interactions between them, allowing you to truly optimize your process.
When and How to Use Plackett-Burman Experimental Design
This article provides a concise summary of the purpose and limitations of Plackett-Burman designs, along with a great comparison table.
Please read the following sections: 'Comparison of the DOE Results': Look closely at Table 5. It perfectly illustrates the trade-off: the Plackett-Burman design used far fewer runs but could not detect the interactions that the full factorial design found. 'When to Use Plackett-Burman Design': This provides a handy checklist for when this tool is appropriate. 'Conclusion': This reinforces the idea that screening is the first step, and factorial designs are what you use next to understand interactions.
Conclusion
In this lesson, you've added a powerful and practical tool to your Six Sigma toolkit. You now know how to tackle the common problem of having too many potential factors to test. By using a screening design like Plackett-Burman, you can efficiently and scientifically narrow your focus to the variables that truly drive your process performance.
Key Takeaways:
- Screening experiments are used to efficiently identify the "vital few" significant factors from a long list of potential variables.
- Plackett-Burman designs are a classic type of screening design that tests many factors in a small number of runs.
- The key trade-off is that Plackett-Burman designs are Resolution III, meaning main effects are confounded with two-factor interactions. You must assume interactions are negligible during this phase.
- Analysis focuses on identifying statistically significant main effects using tools like the Pareto chart and p-values.
- The output of a screening experiment is a list of significant factors that will be studied in more detail in a subsequent experiment.
Preview of the Next Lesson:
Now that you know how to find the vital few, what do you do with them? In our next lesson, we will dive into designing a full or fractional factorial experiment (e.g., 2^k). These experiments are the next step in your investigation, allowing you to not only confirm the main effects but also to understand the crucial interactions between your most important factors.