
Lean Six Sigma Refresher
Lean Six Sigma Refresher: The Improve Phase
The Improve Phase is where insights become action. After identifying root causes in the Analyze Phase, practitioners now design, test, and implement solutions that eliminate variation and drive measurable gains. This phase emphasizes experimentation, statistical validation, and practical deployment.
4.1 Simple Linear Regression
4.1.1 Correlation
Measures the strength and direction of the relationship between two variables.
Correlation coefficients (r) range from -1 to +1.
Helps identify whether changes in one variable are associated with changes in another.
4.1.2 Regression Equations
Regression models quantify the relationship between inputs (X) and outputs (Y).
Provides predictive capability for process outcomes.
4.1.3 Residuals Analysis
Residuals are the differences between observed and predicted values.
Analysis ensures model validity by checking for randomness and absence of patterns.
Detects issues like heteroscedasticity or non-linearity.
4.2 Multiple Regression Analysis
4.2.1 Non-Linear Regression
Models relationships that are not straight-line.
Useful when processes exhibit curvature or diminishing returns.
4.2.2 Multiple Linear Regression
Extends regression to multiple predictors.
Identifies the combined impact of several critical inputs.
4.2.3 Confidence & Prediction Intervals
Confidence intervals estimate the precision of regression coefficients.
Prediction intervals forecast the range of future outcomes.
Both provide context for decision-making.
4.2.4 Residuals Analysis
Same principles as simple regression, but applied across multiple predictors.
Ensures assumptions of linearity, independence, and normality are met.
4.2.5 Data Transformation, Box-Cox
Transformations stabilize variance and improve model fit.
Box-Cox method identifies optimal power transformations.
4.3 Designed Experiments
4.3.1 Experiment Objectives
Define clear goals: optimize performance, reduce variation, or identify key drivers.
Objectives guide design and analysis.
4.3.2 Experimental Methods
Common methods: factorial designs, response surface methodology, Taguchi methods.
Each balances efficiency with insight.
4.3.3 Experiment Design Considerations
Factors: number of variables, levels, interactions, and resource constraints.
Proper planning ensures meaningful results.
4.4 Full Factorial Experiments
4.4.1 2k Full Factorial Designs
Explore all possible combinations of factors at two levels.
Efficient for identifying main effects and interactions.
4.4.2 Linear & Quadratic Mathematical Models
Linear models capture straight-line relationships.
Quadratic models capture curvature and complex interactions.
4.4.3 Balanced & Orthogonal Designs
Balanced: equal representation of factor levels.
Orthogonal: independent estimation of effects.
Both improve statistical validity.
4.4.4 Fit, Diagnose Model and Center Points
Model fit assessed via residual plots and R².
Center points detect curvature in factor-response relationships.
4.5 Fractional Factorial Experiments
4.5.1 Designs
Use a fraction of full factorial combinations.
Reduces resource requirements while retaining insight.
4.5.2 Confounding Effects
Some factor effects may overlap (confound).
Must be carefully managed to avoid misinterpretation.
4.5.3 Experimental Resolution
Resolution defines the clarity of main effects vs. interactions.
Higher resolution designs provide more precise insights but require more runs.
Final Thoughts
The Improve Phase is the engine of transformation. By applying regression, designed experiments, and factorial methods, practitioners move from analysis to action—testing solutions, validating improvements, and optimizing processes. This phase demands both statistical rigor and creative problem-solving, ensuring that changes are not only effective but sustainable.