Counterfactual Sensitivity Probe¶
Sensitivity probe — instantiates Multiple Causation and Explanatory Pluralism
Removes, delays, or intensifies each factor in turn and asks whether the outcome would still hold, ranking causes by how much the result depends on them.
The surest way to learn how much a cause mattered is to imagine the world without it. Counterfactual Sensitivity Probe turns that intuition into a discipline: it takes each candidate factor and asks a structured "what if" — remove it, delay it, or intensify it — and estimates whether the outcome would still occur, occur later, or occur differently. Factors are then ranked by how much the result swings under their manipulation: a factor whose removal erases the outcome is a bottleneck; a factor whose removal changes nothing is either redundant or genuinely minor. Its defining move is counterfactual manipulation of factors, not inspection of the real record — where a process-tracing table weighs what actually happened, the probe reasons about worlds that didn't, and reads causal weight off the size of the difference.
Example¶
A large city's homicide rate falls by roughly 40% over a decade, and five constituencies each claim their factor did it: tougher policing, an improving economy, a shifting age structure (fewer young men), reduced childhood lead exposure, and expanded social services. The probe takes them one at a time. Remove the policing changes — given everything else, would the decline still have landed? Delay the economic upturn by five years — does the timing of the drop move with it? Intensify lead abatement — is a larger effect plausible on dose-response grounds?
Reasoning through each counterfactual with comparison cities, dose-response plausibility, and timing, the analyst finds the decline is over-determined: no single removal erases it, because other factors compensate. But the swings differ. Removing demographics-and-economy together shrinks the decline sharply, while removing policing barely moves it. The probe's output is not "the cause" but a sensitivity ranking — a map of which factors the outcome truly leans on, and a warning that the loudest claimant was the one it could most do without.
How it works¶
- Enumerate factors and choose a manipulation for each. Remove (set it to absent), delay (shift its timing), or intensify (increase its magnitude) — the manipulation matched to what's plausibly variable about that factor.
- Estimate the outcome under each counterfactual. Use comparison cases, dose-response reasoning, or disciplined judgment to project what would have happened; record whether the outcome flips, shifts in timing, or holds.
- Rank by swing. Order factors by how much the outcome moves under their manipulation — that swing is the estimated causal weight.
- Flag the structure. Distinguish a bottleneck (one factor's removal is decisive) from over-determination (several factors are each sufficient, so no single removal bites).
Tuning parameters¶
- Manipulation set — whether each factor is probed by removal, delay, intensification, or all three. Removal answers "was it necessary"; intensification answers "how much leverage"; using only one can miss the factor's real role.
- Estimation method — formal model, comparison case, or expert judgment. Formal models are checkable but data-hungry; judgment is fast but soft.
- Manipulation magnitude — a marginal nudge or total removal. Marginal probes reveal leverage near the actual state; total removal tests necessity but drifts far from anything observable.
- Joint versus single probing — varying factors one at a time or in combination. One-at-a-time is simpler but blind to configurations where two factors are only fatal together.
When it helps, and when it misleads¶
Its strength is that it separates the necessary bottleneck from the over-determined crowd, and it directly deflates "the salient cause": a factor everyone names can be shown, by its removal, to have barely moved the result. Because the swings translate straight into intervention leverage, it is the sibling that most directly tells you where pressure would pay.
Its failure mode is built into counterfactuals: there is no actual world to check the estimate against, so a confident "it would have happened anyway" can be pure assertion. Worse, one-factor-at-a-time probing systematically misses interactions — remove A alone and B compensates, so A looks harmless, yet removing A and B is fatal. This is the classic over-determination trap of but-for reasoning: when two causes are each sufficient, removing either one changes nothing, tempting the false conclusion that neither mattered.[n1] The guard is to probe factors jointly as well as singly, to state the basis for each counterfactual estimate, and to carry the swing forward as a bounded range rather than a point verdict.
How it implements the components¶
counterfactual_probe_set— the mechanism is the set of remove / delay / intensify probes, one designed per factor, that interrogate the outcome's dependence on each.contingency_register— the delay probes and "would it still have happened" questions record exactly the timing sensitivity, path-dependence, and plausible alternative histories the register is meant to hold.causal_weight_scale— the size of the outcome's swing under each manipulation is read as that factor's estimated causal weight, producing a ranked scale.
The probe does not grade the within-case evidence trail (evidential_weighting, the Process-Tracing Evidence Table — its nearest twin, which weighs the actual record where the probe reasons about imagined alternatives), draw the feedback structure (interaction_map, the Causal Loop or Influence Diagram), or inventory factors by family and scale (causal_family_inventory, scale_partition, the Multicausal Factor Matrix).
Related¶
- Instantiates: Multiple Causation and Explanatory Pluralism — the probe is the archetype's tool for testing dependence and preserving contingency.
- Consumes: Multicausal Factor Matrix supplies the factor list whose members the probe then manipulates.
- Sibling mechanisms: Multicausal Factor Matrix · Cause-Role Worksheet · Causal Loop or Influence Diagram · Process-Tracing Evidence Table · Cross-Disciplinary Causal Review
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Counterfactual Sensitivity Probe operates as a computation, comparison, model, or analytic representation used to infer, estimate, or choose because it removes, delays, or intensifies each factor in turn and asks whether the outcome would still hold, ranking causes by how much the result depends on them.
Independent corroboration: The frozen evidence defines Counterfactual Sensitivity Probe as 'Removes, delays, or intensifies each factor in turn and asks whether the outcome would still hold, ranking causes by how much the result depends on them', so its operative form is Analysis, Modeling & Optimization.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Philosophy
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: Counterfactual causal analysis cohered judging causal importance by how an outcome changes when a candidate factor is removed, delayed, or intensified.
Related originating lineages:
- Political Science — Process tracing and comparative explanation operationalize counterfactual tests of necessary, sufficient, and redundant causes.
- Statistics & Experimental Design — Sensitivity analysis supplies one-factor and joint-factor perturbation with uncertainty calibration.
Review resolution: Both reviewers agree on counterfactual causal philosophy and statistical sensitivity. Political-science process tracing is also retained because the mechanism ranks causal importance under redundancy and overdetermination rather than merely varying a numeric input.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; medium confidence.
Notes¶
[n1] But-for (counterfactual) causation asks whether the outcome would have occurred but for the factor. Its known weakness is over-determination: when two causes are each independently sufficient, removing either alone leaves the outcome unchanged, so a naive but-for test wrongly clears both — which is why the probe must also vary factors jointly. ↩