ORDREL / LEARN · ENGLISH
Fishbone Analysis: From Suspected Causes to Evidence
A fishbone diagram organizes possible causes of a defined effect. It is a map of hypotheses, not proof. Its value comes from the evidence checks and decisions that follow the diagram.
Define one effect before listing causes
Specify the observed outcome, process boundary, period and comparison. “Seal failures” is broad; “18 leaking packs in the 600 packs inspected from line B during the stated run” makes the sample and location visible. Do not imply that this inspection represents every pack shipped.
Invite people who know the actual work and distinguish their observations from interpretations. A team can agree on a story that is wrong. Record what is known, what is suspected and what evidence would change the conclusion.
Organize a fictional sealing problem
Fictional manufacturing teaching example. These figures are not ORDREL customer results.
Eighteen findings in 600 inspected packs represent 18 ÷ 600 = 3% of that sample. The team organizes possible explanations using familiar manufacturing categories. These categories help coverage; they do not require an equal number of causes in every branch.
| Category | Possible mechanism | Evidence to seek |
|---|---|---|
| Machine | Temperature indication differs from the sealing surface. | Approved measurement comparison across the operating range. |
| Method | Setup dwell setting is not matched to the current material. | Current instructions, settings and controlled qualification records. |
| Material | A particular film lot behaves differently. | Lot traceability and comparable samples under the same conditions. |
| Measurement | The leak test produces inconsistent classifications. | Repeatability checks using known reference samples. |
| People / work system | The changeover instruction is ambiguous. | Observe how trained users interpret the same instruction. |
| Environment | A relevant ambient condition changes during the run. | Time-aligned records and a plausible physical mechanism. |
Do not substitute category labels for explanations. “Machine” does not tell you how a leak occurs. Avoid collecting identifying employee details in a shared diagram.
Choose checks that can discriminate
For each priority hypothesis, write a predicted observation, a safe check, an owner and a decision rule. For the measurement hypothesis, repeat the classification using approved reference samples. If the test itself is unreliable, apparent relationships with machine settings may be misleading.
Compare like with like. If one film lot only ran on one shift, the lot and shift are confounded. Collect evidence that separates them before naming either as the cause. Do not change several process settings at once and then attribute the result to a single setting. Any controlled trial must use the site’s existing safety, quality and change controls.
Move from a diagram to an investigation record
Mark each hypothesis as untested, supported, weakened or unresolved, and link the status to evidence. “Ruled out” should state the conditions tested; a narrow check may not rule out the mechanism everywhere. Use 5 Whys for a supported branch when the enabling conditions still need explanation.
Prioritize by potential consequence, evidence and feasibility—not by the number of votes a branch receives. A rare severe outcome may deserve attention before a frequent nuisance. Assign the next action even when the answer is “obtain missing evidence.”
Use ORDREL without confusing the map with proof
Use Fishbone to organize causes, evidence and actions. Link the investigation to 5 Whys or an A3 workspace when a branch needs a fuller causal argument. The tool does not automatically verify causes or certify a process.
- Must I use the six manufacturing categories? They are prompts; choose categories that fit the process.
- Should every idea stay on the final diagram? Preserve useful investigation history, but clearly distinguish unsupported ideas from current findings.
- Can a Pareto prove which cause matters? It ranks the measured categories; it does not establish the mechanism.
Sources and method notes
Sources support the method definitions. Examples, interpretation and tool instructions are ORDREL teaching material, not an endorsement by the referenced organizations. See methods and corrections.