Analyze Phase
Purpose and deliverables
Analyze identifies and verifies root causes of the Y using data - not opinion. Deliverables: verified root causes (the vital few x's) and a data-supported cause-and-effect story.
Core concepts
- Cause-and-effect (fishbone) diagram: 6M categories - Man, Machine, Material, Method, Measurement, Mother Nature (environment).
- 5 Whys: drill to root cause; stop when further "why" yields no actionable answer.
- Graphical analysis: Pareto, scatter, histograms, box plots, run charts, multi-vari.
- Hypothesis testing: t-tests (means), ANOVA (multiple means), chi-square (categorical association), regression/correlation (relationships).
- p-value logic: p < α (typically 0.05) → statistically significant; correlation ≠ causation.
- FMEA may be used to prioritize failure modes (RPN = severity × occurrence × detection).
Traps in questions
- Accepting a fishbone "cause" without data verification.
- Misreading p-values: failing to reject the null is not proof of no effect.
- Treating correlation as causation.
- Using the wrong test: t-test for two means, ANOVA for three or more, chi-square for counts/proportions.
- Confusing common cause (inherent variation) with special cause (assignable) - special causes get investigated, common causes get process redesign.
Rule of thumb
Every root cause claim needs evidence: a graph, a test, or a designed comparison. If it is only on the fishbone, it is a hypothesis.
Worked example
Defects differ across three shifts. Because there are three groups, use ANOVA (not a t-test). If p = 0.01 < 0.05, shift is associated with defect rate - then investigate what differs by shift (training, material, setup) before concluding.