Analyze Phase
Purpose
Analyze converts a validated baseline into verified root causes. The output is a short list of x's with evidence behind them, not a list of suspicions.
Tools
5 Whys drills from a symptom to a cause by asking "why" repeatedly. Stop when you reach something you can act on; stopping at "operator error" is the classic failure.
Fishbone (Ishikawa) organizes candidate causes into the 6M's: man/mindpower, machine, material, method, measurement, and environment (mother nature). It generates hypotheses; it does not prove them.
Pareto applies the 80/20 idea - a few categories produce most of the defects. Use it to focus, then verify with data.
Scatter plots show relationships between two continuous variables. Correlation is not causation; you need a plausible mechanism or a designed experiment.
Hypothesis testing. Set H0 (no difference) against Ha. Compare the p-value to alpha (often 0.05). If p ≤ alpha, reject H0. If p > alpha, you fail to reject - that is not proof H0 is true, only that evidence was lacking. Type I error (alpha) is rejecting a true null; Type II error (beta) is failing to reject a false null. Power = 1 − beta.
FMEA. Score Severity, Occurrence, and Detection, usually 1–10; RPN = S × O × D. Act on high RPNs, but never ignore a high Severity score just because the RPN looks moderate.
Worked example
S = 8, O = 4, D = 3 gives RPN = 96. A second failure mode with S = 9, O = 3, D = 3 gives 81 - lower RPN, but the safety-critical severity usually earns attention first.
Traps
Treating a failed hypothesis test as proof of equality and confusing alpha with beta.