Deviant Case Analysis¶
When a case violates what the comparison set led you to expect, analyze the violation as evidence for theory refinement rather than dismissing it as noise or treating it as a story by itself.
The Diagnostic Story¶
Symptom: Reports contain footnotes about exceptions that are never analyzed or integrated into conclusions. Analysts describe a case as weird, special, or not fitting the model without specifying why it deviates. A high- or low-performing outlier is admired, blamed, or dismissed without comparison to matched cases. Theory revisions are made from dramatic examples without first checking measurement quality or case-selection bias.
Pivot: Install a deviant-case learning loop: verify the focal case truly deviates, compare it with matched or contrast cases, trace its internal sequence and mechanisms, test rival explanations, and identify whether the deviation indicates error, omitted variable, new subtype, boundary condition, or theory failure.
Resolution: Fewer unexplained exceptions are dismissed or buried. Theories, models, and decision rules gain sharper boundary conditions and better variable coverage. Unexpected successes and failures become reusable learning cases, and operational systems gain a path from anomaly detection to rule revision rather than merely alerting.
Reach for this when you hear…¶
[quality improvement] “We keep calling that site an outlier and moving on — at some point we have to actually look at what they do differently.”
[epidemiology] “Every exposure model has a few people who got the disease without the exposure and a few who did not get it with — those are not noise, they are the edge of our theory.”
[venture investing] “We explain our losses by saying the market was bad, but if the macro was bad for everyone and only our companies failed, that is not a macro story.”
Mechanisms / Implementations¶
- Boundary Condition Revision Workshop
- Deviant Case Selection Protocol
- Deviation Residual Table
- Follow-Up Case Sampling Plan
- Matched Case Pairing Matrix
- Omitted Variable Probe
- Rival Explanation Elimination Table: Lays every candidate explanation for an outcome side by side and rules each out by the evidence it would predict but the cases do not show.
- Within-Case Process-Tracing Memo
Related Abstractions¶
Abstractions this archetype builds on — directly (a source ingredient) or as a related pattern. Links follow the typed catalog namespace.
Built directly on (1)
- Comparative Method: Systematically juxtaposing selected cases so that their similarities and differences do the causal-inference work that controlled experiments cannot.
Also references 22 related abstractions
- Abstraction: Focus on core elements.
- Black Swan (High-Impact, Low-Probability Events): High-impact unexpected events.
- Boundary: Defines system limits.
- Causality: Cause-effect relationships.
- Classification: Sorting entities into discrete categories by explicit rules, turning unbounded variation into a finite, reusable map for downstream reasoning and action.
- Confirmation Bias: Favor confirming evidence.
- Counterfactual Reasoning: Hypothetical alternatives.
- Deductive Reasoning: General to specific conclusions.
- Feedback: Outputs influence inputs.
- Hypothesis Testing (Null vs. Alternative): Null vs alternative evaluation.
Variants¶
Narrower or domain-specific specializations that share this archetype's core structure. Recognized variants are established; candidate variants are provisional.
Negative Case Validation · subtype · candidate
Uses cases that contradict an expected claim to validate, qualify, or revise the claim.
Deviant Success Case Analysis · subtype · recognized
Explains cases that succeed despite conditions under which failure was expected.
Deviant Failure Case Analysis · subtype · recognized
Explains cases that fail despite conditions under which success was expected.
Boundary Condition Discovery · subtype · candidate
Uses deviant cases to specify where a theory, model, or rule stops applying.
Omitted Variable Recovery · subtype · candidate
Uses a deviant case to discover variables, mechanisms, constraints, or interactions missing from the original explanation.