Green Belt · Analyze phase
Yellow Belt covered Define, Measure, and the basics of Control: framing a problem, collecting reliable data, and holding a gain once it is achieved. Green Belt is where the middle of DMAIC opens up. Analyze is the search for root cause using statistical evidence rather than opinion, and Improve is where a team tests and confirms a fix before it is rolled into a control plan. Both phases lean on inferential statistics: using a sample to say something trustworthy about a whole process, rather than eyeballing a chart and guessing. The IASSC Certified Lean Six Sigma Green Belt (ICGB) exam covers this expanded scope with 100 graded multiple-choice and true/false questions, roughly 20 drawn from each of the five DMAIC phases, in a closed-book, proctored, 3-hour session with a 70 percent passing score, the same pass bar as Yellow Belt. This lesson opens the Analyze phase with the first tool a Green Belt reaches for when a process shows more variation than it should: figuring out where that variation is actually coming from.
Before running any statistical test, a Green Belt first asks a cheaper question: which of the three basic families does most of the observed variation belong to?
These three families are not competing theories; a real process usually has some of each. The question a multi-vari study answers is which family carries most of the total spread, because that tells the team where to spend its investigation budget next.
A multi-vari chart plots a measured characteristic on the vertical axis against nested groupings on the horizontal axis: several measurements on one unit, several units within one batch, and several batches across time. The result is a picture, not a single number, and it is read by comparing the size of three kinds of gaps: the spread of the short bars (positional), the spread between the bars within one time group (cyclical), and the spread between the time groups themselves (temporal). Whichever gap is largest is where the dominant source of variation lives, and it is almost always obvious by eye once the chart is built correctly, which is the tool's main advantage over guessing from a summary average.
Worked example. In the chart above, the within-wall (positional) bars span about 0.04 inch, the house-to-house (cyclical) spread within a single time batch spans about 0.09 inch, and the morning-to- afternoon (temporal) spread is about 0.40 inch, roughly ten times the positional spread and more than four times the cyclical spread. A Green Belt reading this chart would not investigate wall-to-wall technique differences first; the data says to investigate what changes about the crew, materials, or equipment between morning and afternoon.
Why this matters
Multi-vari analysis is cheap: it uses data a team is often already collecting, reorganized into a chart, before anyone commits to a designed experiment or a formal hypothesis test. Exam questions in this section usually describe a scenario and ask which variation family it illustrates, or what a multi-vari chart is used for before choosing a statistical test.
Once a Green Belt knows roughly where variation is coming from, the next question is what shape that variation's data follows, because the shape determines which statistical tools are valid. Data is first split into two broad kinds. Continuous data can take any value on a scale, such as a measured gap in inches or a cycle time in hours. Discrete data can only take specific values, usually counts, such as the number of defects on a punch list or the number of homes inspected in a day.
| Distribution | Data type | Describes |
|---|---|---|
| Normal | Continuous | Symmetric, bell-shaped spread around a mean; the default assumption for many measured process characteristics until proven otherwise. |
| Uniform | Continuous | Every value across a range is equally likely, with no peak, such as where within a scheduled window a delivery truck happens to arrive. |
| Exponential | Continuous | Time between independent events that occur at a constant average rate, such as the time between successive maintenance calls on a rental unit. |
| Binomial | Discrete | Number of successes (or failures) in a fixed number of independent trials with a constant success probability, such as how many of 20 inspected homes pass on the first try. |
| Poisson | Discrete | Number of rare, independent events occurring in a fixed interval of time or space, such as the number of punch-list defects found per house. |
The normal distribution is the familiar symmetric bell curve, fully described by just two numbers: its mean (μ), which sets its center, and its standard deviation (σ), which sets its spread. Many process measurements approximate a normal distribution closely enough for the standard tools, covered in the lessons that follow, to apply directly. The uniform distribution looks nothing like a bell; it is flat, because every outcome in its range is equally likely, and it shows up wherever a process has no natural tendency to cluster near a center, such as a random arrival time inside a fixed window. The exponential distribution is used for the gaps between events rather than the events themselves; it is always right-skewed, has no negative values, and carries the "memoryless" property, meaning the chance of waiting an additional hour for the next event does not depend on how long the wait has already been.
The binomial distribution applies whenever there is a fixed number of independent trials (n), each with the same probability of success (p), and the outcome of interest is how many of those trials succeed. Inspecting 20 completed rehabs and counting how many pass a first NSPIRE-style inspection is a binomial situation as long as each home's pass probability is roughly constant and independent of the others.
The Poisson distribution instead counts how many rare, independent events occur across a fixed span of time, area, or units, when there is no natural upper limit on the count the way there is with a binomial's fixed n. The average rate is usually written λ (lambda). Punch-list defects per house, work-order requests per week, or safety incidents per month are typical Poisson situations in a housing rehab program.
Worked example. Suppose a program's punch-list defect rate averages λ = 3 defects per completed house. The Poisson formula gives P(X = k) = (e-λ × λk) / k!. The probability of a house coming back with exactly 2 defects is: P(X = 2) = (e-3 × 32) / 2! = (0.0498 × 9) / 2 ≈ 0.224, or about 22.4 percent. A Green Belt uses this kind of calculation to judge whether a particular house's defect count is ordinary variation or unusual enough to investigate on its own.
1. Compared with Yellow Belt, what does Green Belt training add in terms of DMAIC scope?
2. How is the IASSC Certified Green Belt (ICGB) exam structured?
3. A rehab crew measures the same wall's insulation gap at the top, middle, and bottom and finds small differences among the three spots. This is an example of which variation family?
4. Three houses inspected the same morning show consistently different average gap measurements from each other. This pattern illustrates which variation family?
5. A program's average work-order response time is noticeably higher in July and August than the rest of the year. This is best classified as which variation family?
6. What is the main purpose of building a multi-vari chart early in the Analyze phase?
7. In a multi-vari study, the within-wall spread is 0.04 inch, the house-to-house spread within a morning batch is 0.09 inch, and the morning-to-afternoon spread is 0.40 inch. Which source should the team investigate first?
8. Which statement correctly describes the normal distribution?
9. A program inspects 20 completed rehabs and counts how many pass a first NSPIRE-style inspection, where each home's pass probability is roughly constant and independent of the others. Which distribution best models the count of passes?
10. Which distribution models the count of rare, independent events, such as punch-list defects per house, over a fixed interval with no fixed upper limit on the count?
11. If a program's punch-list defect rate averages λ = 3 defects per house, what is the probability that a given house has exactly 2 defects? (P(X = k) = e−λ × λᵏ / k!)
12. Which distribution describes the time between successive, independent events occurring at a constant average rate, and carries the "memoryless" property?
13. A delivery is scheduled to arrive at some random point within a fixed two-hour window, with no time in that window more likely than any other. Which distribution describes the arrival time?
14. Cycle times for closing out work orders are measured in hours and can take any value, such as 4.3 or 4.35 hours. This kind of data is best described as which of the following?