Root-Cause Analysis & Competing Hypotheses

The Scientific Approach to Root Cause

A credible root-cause analysis does not jump to the first plausible explanation. It systematically forms multiple candidate hypotheses, predicts the evidence each would produce, compares the predictions with the observed evidence, and converges on the hypothesis that best explains all the important evidence. This is the scientific method applied to failure investigation — and it protects against the most common error: confirmation bias.

A STRONG ROOT-CAUSE CONCLUSION EXPLAINS THE EVIDENCE BETTER THAN THE COMPETING EXPLANATIONS. A conclusion that fits the evidence but has not been challenged against alternatives is not yet a conclusion — it is a hypothesis.

Forming Candidate Hypotheses

The first step is to enumerate the candidate failure hypotheses. Based on the observed damage, the operating history and the structural configuration, what mechanisms could have produced the failure? At this stage, the goal is breadth — include all plausible mechanisms, even those that seem unlikely. An unlikely hypothesis that is systematically rejected is more valuable than an unconsidered hypothesis that turns out to be correct.

• H1 — Overload: a single load event exceeded the structural capacity
• H2 — Fatigue: cyclic loading initiated and propagated a crack to failure
• H3 — Inadequate preload: insufficient clamp load caused joint slip and bolt fatigue
• H4 — Material defect: a manufacturing defect reduced the structural capability
• H5 — Thermal restraint: constrained thermal expansion generated unexpected load

The Hypothesis Matrix

For each hypothesis, the investigator predicts what evidence should be present, compares this with the observed evidence, identifies contradictory evidence, and determines what analysis or test is needed to resolve the question. The hypothesis matrix is a structured way to track this comparison.

ILLUSTRATIVE EXAMPLE — hypothetical content for educational purposes. Not a real investigation.

Hypothesis Matrix Example

The following matrix illustrates how competing hypotheses are evaluated against evidence. Each row represents a hypothesis; each column represents a different aspect of the evaluation. The matrix is filled in as the investigation progresses.

Evaluating the Matrix

The hypothesis matrix is not a voting system — the hypothesis with the most "supported" entries does not automatically win. The investigator must weight the evidence by importance. A hypothesis that is contradicted by the most important evidence — the fracture surface, the failure location, the load history — is rejected regardless of how many minor points it matches. A hypothesis that explains the most important evidence but leaves minor points unexplained may still be the best explanation — with the unexplained points noted as limitations.

Levels of Support

As the investigation progresses, each hypothesis moves through levels of support. Recognising the level helps the investigator communicate the strength of the conclusion appropriately.

Confirmation Bias

Confirmation bias is the tendency to seek evidence that supports a preferred hypothesis while overlooking evidence that contradicts it. It is the most dangerous error in failure investigation. The investigator forms an initial impression — perhaps from the most visible damage, from a previous similar case or from a client's suggestion — and then unconsciously filters the evidence to support that impression. The scientific approach — forming multiple hypotheses and actively seeking contradictory evidence — is the primary defence against confirmation bias.

COMMON MISTAKE: Forming an initial impression from the most visible damage or a client suggestion, then interpreting all subsequent evidence to support that impression. Actively seek evidence that would contradict your preferred hypothesis.

Active Disconfirmation

The investigator should actively try to disconfirm each hypothesis. Rather than asking "does the evidence support H?", ask "what evidence would contradict H, and is that evidence present?" If the contradicting evidence is absent, the hypothesis gains support. If the contradicting evidence is present, the hypothesis is weakened or rejected. This approach — seeking disconfirmation rather than confirmation — is the most effective protection against bias.

1. For each hypothesis, ask: what evidence would contradict this hypothesis?
2. Search for that evidence specifically — do not wait to stumble upon it
3. If contradicting evidence is found, weaken or reject the hypothesis
4. If contradicting evidence is absent, the hypothesis gains support
5. Repeat for each hypothesis until only the best-supported remains

When No Hypothesis Is Fully Supported

Sometimes, after systematic evaluation, no hypothesis fully explains all the evidence. This is not a failure of the method — it is a result. It means either that a hypothesis has been overlooked, that the evidence is incomplete, or that multiple factors combined in a way that no single hypothesis captures. In this case, the investigator should report the best-supported hypothesis with its limitations, note the unexplained evidence, and recommend further investigation — rather than forcing a conclusion that the evidence does not fully support.

FORENSIC CHECK: If no hypothesis fully explains all the evidence, do not force a conclusion. Report the best-supported hypothesis with its limitations and recommend further investigation.

Key Takeaways

• Root-cause analysis is the scientific method: form hypotheses, predict evidence, compare, converge
• Form multiple candidate hypotheses — breadth at this stage is more valuable than selectivity
• Use a hypothesis matrix to track predicted vs observed vs contradictory evidence
• Weight evidence by importance — contradiction on the most important evidence rejects a hypothesis
• Levels of support: possible → plausible → supported → demonstrated
• Confirmation bias is the most dangerous error — actively seek disconfirming evidence
• If no hypothesis fully explains the evidence, report the best-supported with limitations