Analyze: From Candidate Causes to Verified Root Causes
What Analyze must produce
Analyze narrows the long list of x's from Define to the few verified root causes. Concepts: stratification and graphical analysis (histogram, box plot, run chart, scatter plot, multi-vari study) to see where variation lives; Pareto to prioritise; Ishikawa / fishbone using the 6M categories (man, machine, material, method, measurement, mother nature/environment) and 5 Whys to drill down; FMEA prioritising by RPN = Severity × Occurrence × Detection; and hypothesis testing to confirm causes - null and alternative hypotheses, alpha (Type I error, producer risk), beta (Type II error, consumer risk), power = 1 − beta, and the p-value defined as the probability of data at least this extreme if the null were true. Test selection follows data type and group count: 1-sample t, 2-sample t, paired t, ANOVA, chi-square, regression. Always check assumptions - normality, equal variance, independence - and inspect residuals.
Traps questions exploit
- "p = 0.03 means a 3% chance the null hypothesis is true" - false. It is a statement about the data, not the hypothesis.
- "Accept the null" instead of "fail to reject the null."
- Statistical significance mistaken for practical significance; with a huge sample, trivial differences turn significant.
- Correlation presented as causation, or a paired test used on independent samples.
- Re-scoring only the RPN after countermeasures, instead of re-scoring Severity, Occurrence and Detection.
Worked example
"Line stopped 14 times this month." Why? Jam at the capper. Why? Caps misaligned. Why? Chute guide worn. Why? No scheduled maintenance. Why? The PM schedule omits the chute. The countermeasure targets the PM schedule, not the operator.
Rule of thumb: a cause is verified when a statistical test or designed change shows the x moves the Y - not when the team agrees it feels right.