A method for evaluating an explanatory claim, model result, or empirical finding by testing whether the result obtains across a diverse set of plausible assumptions, model structures, parameter values, or experimental conditions; robustness analysis assesses whether a conclusion is an artifact of particular choices or reflects a stable feature of the system or phenomenon.

Principle

If a result persists under substantial, theoretically relevant variations in assumptions, structure, or conditions, its evidential support increases because it is less likely to depend on idiosyncratic model choices.

Demonstration

Illustrative scenario: Economists obtain a prediction that a policy raises aggregate savings (Situation). They re-run models with alternative utility functions, different market frictions, and varied calibration datasets (Recognition). Finding the qualitative effect across these variants (Action) increases confidence that the effect reflects a structural mechanism rather than a modeling artefact (Consequence).

Misapplication

Concluding that observed robustness guarantees truth or treating limited sensitivity checks (small parameter tweaks) as full robustness; the error is failing to vary assumptions or structures that are theoretically plausible and materially consequential.

Consequence

Robustness analysis can identify candidate core mechanisms, prioritize research, and reduce model-dependence in inference; however, apparent robustness can mask shared assumptions or biases across models and may provide false confidence when key alternatives are not considered.

Reversal

When competing models share a hidden common assumption or data source, robustness across that ensemble is misleading; similarly, for tasks requiring precise numerical prediction, robustness of qualitative pattern may be insufficient and even irrelevant.

Boundary

Clearly within: comparing results across structurally different models and parameter regimes. Boundary case: a single model with extensive local sensitivity analysis. Clearly outside: repeating identical model code or datasets without substantive variation in assumptions or structure.

Semantic Tension

Robustness ↔ Precision/Specificity: robustness emphasizes stability of qualitative features across variation, whereas precision demands accurate, narrowly specified quantitative predictions that may not be robust across structural changes.

Synthesis

Robustness analysis treats stability across plausible alternatives as corroborative evidence; it strengthens inferences by exposing model-dependence but must be designed to vary the assumptions that matter, since apparent robustness can arise from shared, unexamined commitments.