One-way Sensitivity Analysis¶
Analytical procedure — instantiates Sensitivity Analysis Protocol
Moves one input at a time across its range while holding everything else fixed, then ranks assumptions by how far each alone swings the outcome.
One-way sensitivity analysis moves a single input at a time across its plausible range while every other input stays pinned at its baseline value, records how far the watched outcome swings, and repeats the test across the candidate inputs to produce a ranked list of which assumptions individually move the answer most. Its defining move is isolation: because only one thing changes per test, any movement in the outcome is unambiguously attributable to that one input — which is exactly what makes the method cheap, explainable, and completely blind to effects that appear only when inputs move together. It is the first-pass driver screen, the thing you run before you know which assumptions deserve heavier machinery.
Example¶
A mid-sized city is deciding whether to fund a bus rapid-transit line, and the pitch rests on a projected net operating cost that looks affordable. Before the council votes, staff run a one-way sweep. They fix a baseline — projected ridership, fuel price, driver wages, maintenance rate, farebox recovery — and then, one at a time, push each input from its low plausible value to its high while holding the other four at baseline. Ridership alone swings annual net cost by roughly ±$3.2M; fuel price by ±$0.9M; wages by ±$1.6M; maintenance by ±$0.4M; farebox recovery by ±$1.1M. Sorted, the ranking is unambiguous: ridership dominates, wages and farebox recovery matter, fuel and maintenance are close to noise. The council now knows that the whole affordability case hangs on the ridership forecast — so that is the number to pin down before committing, and the two trivial drivers can be set aside.
How it works¶
- Freeze a baseline. Fix every input at its reference value; this is the point every single-input test departs from and returns to.
- Sweep one input. Move exactly one parameter from its low plausible value to its high, holding all others at baseline, and read the outcome at each end.
- Record the swing. The outcome's low-to-high travel is that input's solo effect; log it with the range that produced it.
- Repeat and rank. Do this for each candidate input, then sort the inputs by swing magnitude — the ranking is the product.
Tuning parameters¶
- Range basis — whether each input's low/high is a fixed percentage, an absolute band, or a confidence interval. Percentage bands are uniform but ignore that some inputs are far more uncertain than others.
- Sweep breadth — how many candidate inputs enter the screen. More inputs catch surprises but dilute attention and invite parameter theater.
- Endpoint definition — min–max versus P10–P90 versus ±x%. Extreme endpoints exaggerate a driver's apparent importance; timid ones hide fragility.
- Baseline anchor — which reference case the sweeps depart from; a biased baseline tilts every swing measured against it.
When it helps, and when it misleads¶
Its strength is clarity and speed: for a handful of inputs it produces an honest, defensible ranking of drivers that anyone can follow, and it is the cheapest way to decide where deeper analysis is worth spending. Its failure mode is structural — because it varies inputs one at a time, it is systematically blind to interaction and correlation, so when two inputs move together or amplify each other, the one-way ranking can put the wrong driver first.[n1] The classic misuse is one-way blindness: treating the ranking as the finished analysis when the inputs are in fact coupled. The guarding discipline is to use one-way as a screen, not a verdict — confirm the top two or three drivers with a joint test before betting the decision on their order.
How it implements the components¶
baseline_case— establishes and holds the fixed reference every single-input sweep is measured against.key_parameter— its unit of work is selecting the candidate inputs and moving them one by one.outcome_response— records the low-to-high outcome swing each parameter produces in isolation.parameter_prioritization_rule— sorts those swings into the ranked driver list that is the mechanism's deliverable.
It does not vary parameters together or capture how they interact — correlation_or_dependency_note is Two-way or Multi-way Sensitivity Analysis's, which sweeps parameters jointly. It also does not bundle inputs into coherent worlds (scenario_set, Scenario Variation) or solve for the value where the answer flips (decision_flip_point, Threshold Analysis).
Related¶
- Instantiates: Sensitivity Analysis Protocol — this is the protocol's first-pass single-driver screen.
- Sibling mechanisms: Two-way or Multi-way Sensitivity Analysis · Scenario Variation · Threshold Analysis · Tornado Chart · Sensitivity Table · Probabilistic Sensitivity Simulation · Assumption Stress-test Workshop
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: One-way Sensitivity Analysis operates as an analytical, modeling, inference, comparison, or optimization procedure that derives insight or a solution because it moves one input at a time across its range while holding everything else fixed, then ranks assumptions by how far each alone swings the outcome.
Independent corroboration: The frozen evidence defines One-way Sensitivity Analysis as 'Moves one input at a time across its range while holding everything else fixed, then ranks assumptions by how far each alone swings the outcome', so its operative form is Analysis, Modeling & Optimization.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Operations Research
Origin pattern: Convergent development
Present-day reach: Multi-domain
Rationale: Decision analysis and operations research developed one-at-a-time sensitivity analysis for ranking which input assumptions most move an objective.
Related originating lineages:
- Economics & Finance — Financial modeling institutionalized one-way sensitivity tables for investment assumptions.
- Statistics & Experimental Design — Experimental-design traditions supplied controlled variation and the warning that one-factor changes cannot reveal interactions.
Review resolution: Both independent reviews agree on primary origin operations_research; reconciliation resolves alternate_origin_disagreement. Formative alternate lineages retained: statistics_experimental_design, economics_finance. The broader reach of later applications is kept separate as domain_reach=multi_domain; origin_mode=convergent describes the historical relationship among lineages. Confidence is conservatively reconciled to high, and encyclopedia_synthesis=false preserves the reviewers' boundary judgment.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] One-at-a-time (OAT), also called one-factor-at-a-time, is the class of methods that perturb a single input around a baseline while holding the rest constant. It is simple and interpretable but cannot detect interaction effects, which is why coupled models are re-tested jointly rather than one input at a time. ↩