Lean Six Sigma Improve Phase
The fourth phase of DMAIC: modelling the relationship between inputs and output, and designing experiments instead of guessing
Improve is where you change something — but deliberately, and with evidence that the change will work. Having identified the causes in Analyze, you now model how the inputs relate to the output and use designed experiments to find the settings that give the result you want. The discipline is in changing several factors in a planned way rather than one at a time by intuition, which is both slower and less informative.
Below is what the Improve phase covers in the Lean Six Sigma syllabus, section by section.
📌 Quick navigation
What you actually do in Improve
| Step | Output |
|---|---|
| Model the relationship | A regression model linking the inputs to the output, with residuals checked |
| Decide what to test | Experiment objectives and the factors and levels worth varying |
| Design the experiment | A balanced design — full factorial where affordable, fractional where it is not |
| Run and fit | A fitted model, diagnosed, with centre points to check for curvature |
| Choose the settings | Input settings that deliver the target output, with the trade-offs stated |
The phase ends with a change that has been demonstrated, not argued for. Improvements adopted without this step tend to be reversed the first time the process behaves unexpectedly.
4.1 Simple Linear Regression
- 4.1.1 Correlation
- 4.1.2 Regression Equations
- 4.1.3 Residuals Analysis
4.2 Multiple Regression Analysis
- 4.2.1 Non-Linear Regression
- 4.2.2 Multiple Linear Regression
- 4.2.3 Confidence & Prediction Intervals
- 4.2.4 Residuals Analysis
- 4.2.5 Data Transformation, Box Cox
4.3 Designed Experiments
- 4.3.1 Experiment Objectives
- 4.3.2 Experimental Methods
- 4.3.3 Experiment Design Considerations
4.4 Full Factorial Experiments
- 4.4.1 2k Full Factorial Designs
- 4.4.2 Linear & Quadratic Mathematical Models
- 4.4.3 Balanced & Orthogonal Designs
- 4.4.4 Fit, Diagnose Model and Center Points
4.5 Fractional Factorial Experiments
- 4.5.1 Designs
- 4.5.2 Confounding Effects
- 4.5.3 Experimental Resolution
Which belt level covers this
Design of experiments is largely Black Belt territory. Green Belt covers regression and the reasoning behind experimental design; Black Belt is expected to design a factorial experiment, understand confounding and resolution, and interpret the fitted model. Yellow Belt covers the concepts without the modelling.
| Level | Course | Mock exam |
|---|---|---|
| Yellow Belt | Course | Practise |
| Green Belt | Course | Practise |
| Black Belt | Course | Practise |
Frequently asked questions
Why not change one factor at a time?
Because it costs more runs and misses interactions. If two inputs only matter in combination, changing them one at a time will never reveal it — which is precisely what factorial designs exist to detect.
What is confounding, in plain terms?
In a fractional design you deliberately run only part of the full experiment, which means some effects become indistinguishable from others. Knowing which effects are confounded, and at what resolution, is what makes a fractional design usable rather than misleading.
Correlation is not causation — so why regression?
Regression describes a relationship; the causal claim comes from the experiment, where you set the inputs yourself rather than observe them. That is the difference between sections 4.1–4.2 and 4.3–4.5, and it is examined.
Is a designed experiment realistic in a live process?
Not always, and part of the skill is knowing when it is not. Where the process cannot be perturbed, the project falls back on modelling observational data — with the weaker conclusions that implies, stated as such.
Design of experiments is the part of the syllabus candidates most often underestimate. Try the mock exam for your belt level — it scores by topic, so you can see whether Improve is where the marks are going. Contact us for advice on which level to sit.
The other DMAIC phases
- Define — deciding which problem is worth solving.
- Measure — establishing the baseline and trusting the data.
- Analyze — finding and proving the causes.
- Control — making the improvement stay.
- Lean Six Sigma — the section overview.
