Sensitivity Analysis Workshop¶
Parametric analysis — instantiates Assumption Stress Testing
A working session that systematically varies a model's numeric inputs to measure how far the conclusion moves with each — and how assumptions compound — ranking which quantitative premises the answer actually hangs on.
A Sensitivity Analysis Workshop takes a quantitative model — a financial pro forma, a demand forecast, an engineering estimate — and, with the modelers in the room, systematically moves its numeric inputs to measure how much the bottom line swings in response to each. Its one defining move is quantified influence ranking: it does not ask whether a premise is true, but how much the conclusion depends on it, producing an ordered picture of which assumptions the answer is actually hostage to and which barely matter. Its second, subtler contribution is watching for compounding — moving inputs in pairs to catch the interaction where two individually tolerable errors multiply into an intolerable one. It works only on what can be parameterized, and it measures leverage, not likelihood; a highly influential premise is a candidate for scrutiny, not a proven weakness.
Example¶
A transit authority's model for a proposed light-rail line reports a healthy farebox recovery ratio, and the board is ready to commit to the numbers. The sensitivity workshop puts the model on the screen and starts moving inputs across their credible ranges, one at a time. Ridership growth swings the ratio hard — a few points of assumed annual growth is the difference between viable and not. Fare elasticity swings it moderately. The assumed construction contingency barely moves it at all. That alone reorders the board's attention: the whole case rests on the ridership-growth premise, which had been entered as a confident single number. Then the workshop moves inputs together and finds the compounding case — if ridership growth comes in low and fare elasticity is higher than assumed, the two interact and the recovery ratio collapses far below the sum of the individual effects, because low ridership makes any fare increase bite harder. The output is a ranked tornado of which premises the answer hangs on, plus a flagged interaction the one-at-a-time view would have hidden — the map of where the model is truly load-bearing.
How it works¶
- Parameterize the premises. Restate the model's assumptions as inputs with credible ranges, not point values — the workshop can only test what is expressed as a number that can be moved.
- Sweep one at a time. Vary each input across its range while holding others fixed, and record how far the output moves — the basis of the influence ranking.
- Move inputs in pairs. Deliberately co-vary the influential inputs to surface interactions, where combined effects diverge from the sum of the individual ones.
- Rank and flag. Order the premises by how much they swing the conclusion, and flag the interactions — so scrutiny and evidence-gathering go to the inputs the answer is genuinely hostage to.
Tuning parameters¶
- Variation range — plausible bounds versus extreme stress ranges. Wider ranges expose more fragility but blur the line between sensitivity analysis and worst-case scenario work.
- Interaction depth — one-at-a-time only versus paired or full multi-way variation. Deeper analysis catches compounding but multiplies the runs combinatorially.
- Influence cutoff — how large a swing marks a premise as "load-bearing." A low cutoff floods attention; a high one may drop a moderate premise that compounds dangerously with another.
- Model-fidelity trust — how much weight to place on the model's structure itself. Sensitivity within a wrong model measures the wrong thing precisely.
When it helps, and when it misleads¶
Its strength is precision about leverage: it tells you, with numbers, exactly which few quantitative premises deserve real evidence and which are safe to leave roughly guessed — and its paired-variation step catches the compounding interactions a naive check misses.[n1] For a model-driven decision it is the most rigorous way to find where the case is thin.
Its failure mode is that it can only see what is parameterized, so it quietly reassures a team about the numeric assumptions while the fatal premise — a qualitative, behavioral, or structural one outside the model — goes untested and unmentioned. The one-at-a-time habit is its specific trap: varying inputs singly hides exactly the correlated failures that sink real plans, which is why the paired step is not optional. And a sensitivity analysis run inside a mis-specified model measures its own internals with false confidence. The guarding discipline is to treat the influence ranking as a map of where to look, always test the top-ranked premises in pairs, and hold in view that the model's own structure is itself an assumption the workshop cannot check.
How it implements the components¶
assumption_interdependency_map— its paired- and multi-way variation maps how the model's premises depend on and compound one another, surfacing the interactions where two tolerable errors combine into an intolerable one.criticality_filter— the influence ranking is a quantitative criticality filter: it orders the numeric premises by how much the conclusion depends on each, so effort concentrates on the load-bearing few.
It does not construct a coherent alternative-future narrative or drive a premise to failure by a bundled shock (stress_scenario, assumption_break_test) — that is Scenario Stress Test — and it does not assign a confidence grade or record an accepted risk (confidence_and_evidence_rating, accepted_assumption_risk_note) — that is Stress-Test Scorecard. This workshop measures how much the answer moves, not how confident we are in the inputs.
Related¶
- Instantiates: Assumption Stress Testing — the workshop is how the archetype finds which quantitative premises the model's conclusion is hostage to.
- Sibling mechanisms: Scenario Stress Test · Stress-Test Scorecard · Failure Mode and Effects Table · Premortem · Red-Team Future Challenge · Resilience Tabletop Exercise · Trigger Dashboard
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Sensitivity Analysis Workshop operates as an analytical, modeling, inference, comparison, or optimization procedure that derives insight or a solution because it a working session that systematically varies a model's numeric inputs to measure how far the conclusion moves with each — and how assumptions compound — ranking which quantitative premises the answer actually hangs on.
Independent corroboration: The frozen evidence defines Sensitivity Analysis Workshop as 'A working session that systematically varies a model's numeric inputs to measure how far the conclusion moves with each — and how assumptions compound — ranking which quantitative premises the answer actually hangs on', so its operative form is Analysis, Modeling & Optimization.
Nearest alternative: Communication, Facilitation & Learning — Sensitivity Analysis Workshop includes features of a designed message, facilitated interaction, ritual, or learning activity that changes shared understanding, but its defining operation is an analytical, modeling, inference, comparison, or optimization procedure that derives insight or a solution.
Review outcome: Independent reviewer agreement; medium confidence.
Origin Attribution¶
Primary origin: Operations Research
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: Collaboratively varying quantitative premises and ranking decision leverage applies operations-research sensitivity analysis in a facilitated working format.
Related originating lineages:
- Engineering & Design — Design reviews use collaborative trade studies to expose fragile requirements.
- Mathematics — Mathematical modeling, proof, and abstract-structure practice supplies a parallel or contributing lineage for the mechanism's defining operation: a working session that systematically varies a model's numeric inputs to measure how far the conclusion moves with each — and how assumptions compound — ranking which quantitative….
- Organizational & Management Science — Workshop facilitation surfaces owners, tacit assumptions, and compound dependencies.
- Statistics & Experimental Design — Designed parameter variation and uncertainty accounting make the comparisons credible.
Review resolution: The blind reviewers agree that operations_research is the primary origin and differ only on alternate origin disagreement, origin mode disagreement, encyclopedia synthesis disagreement. I preserve every independently explained alternate from both records rather than imposing a numeric cap. I retain cross_disciplinary_synthesis because the combined record shows material contributions from several lineages. The broader reach of multi_domain records portability separately from historical provenance, and encyclopedia_synthesis=true preserves the affirmative synthesis judgment where either reviewer identified one.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
Sensitivity analysis and scenario stress testing are often conflated, but they answer different questions and should not be substituted for each other: sensitivity asks how much does the answer move per input, sweeping a controlled range; the scenario asks what happens in this specific coherent future, moving many inputs together for a reason. A model deserves both — the ranking to find the load-bearing premises, the scenario to see them fail in concert.
[n1] The one-at-a-time (OAT) method — varying a single input while holding the rest fixed — is the default form of sensitivity analysis and its central weakness: it cannot detect interaction effects and explores only a thin slice of the input space. Global sensitivity methods (e.g., variance-based Sobol indices, associated with Andrea Saltelli's work) exist precisely to capture the compounding that OAT misses. ↩