Root Cause Analysis with Hierarchical Taxonomy
Organise causes and activities into a useful taxonomy and understand the additional checks required for causal analysis.
Updated:
A hierarchical taxonomy organises states from summary to detail. It guides investigation, but a “cause” label does not establish root cause. Distinguish observed state, reported reason and confirmed cause.
Build levels without double counting
One person-instant structure is:
- Primary state: direct work, support, waiting, break or other.
- Subtype: within waiting, material, authorisation, equipment or unconfirmed reason.
- Evidence: an incident or record supporting verification.
Assign each record to one primary leaf if the proportions are to sum to 100%. Multiple attributes such as line and order may be retained, but their percentages need not be additive. A hierarchy is not required for binomial validity: a well-defined flat classification can also work.
A parent's count is the sum of its leaves; do not add parent and leaves as separate time. Check denominators for conditional percentages.
Teaching case: 1,250 readings
| Primary state and subtype | Count | Percentage |
|---|---|---|
| Direct work | 562 | 44.96% |
| Tool preparation | 113 | 9.04% |
| Material waiting | 188 | 15.04% |
| Forklift/AGV waiting | 87 | 6.96% |
| Confirmed mechanical stop | 125 | 10.00% |
| Confirmed electrical stop | 63 | 5.04% |
| Recorded breaks | 112 | 8.96% |
| Total | 1,250 | 100% |
This is simulated, not a real deployment. Supply waits total 275/1,250=22%, and mechanical/electrical stops 188/1,250=15.04%. Together they are 37.04%. The 22% does not prove a forklift route is the root cause: stock, scheduling, quality or equipment availability may be involved.
The 8.96% breaks does not measure physiological fatigue or establish unjustified personal delay. The 113 preparation records are support, even if included in a broader activity grouping.
Verify before acting
Check waits against stock, requests, material arrivals and production events. Ask what evidence contradicts each hypothesis. A confirmed breakdown may explain the stop, but competent staff must still investigate its technical cause.
A Pareto chart prioritises frequency or estimated cost; it does not automatically identify the highest-return project. Do not justify cancelling or making major investments without cost, risk and alternative analysis.
Precision at each level
Under independent Bernoulli records, the approximate normal 95% margin for p̂=0.22 and n=1,250 is 2.3 percentage points. The maximum at p=0.5 is 2.8 points. Choosing 1,250 does not itself establish a 3.5-point margin: state formula, assumptions and design.
Rare leaves may have little information even when their parent is well estimated. Distinguish the share of the whole study from the share within waiting, showing both denominators. Clusters and weights require adapted intervals.
Scope and rights
Random schedules do not eliminate reactivity or anonymise data. Define minimal identification, access and retention according to purpose and law. Sampling is not a unique mandatory energy-audit protocol or an employment-law compliance certificate.
OEE needs compatible time data, ideal cycle time and total/good units. The “mechanical failure” proportion is not intrinsic availability or enough to reconstruct all performance.
WorkSamp supports category recording and distribution studies. Check the supported structure rather than assume unlimited trees, SCADA integration or automatic root-cause analysis.