Uncertainty Taxonomy
Not all unknowns are the same kind of unknown. Collapsing them is how confident-sounding nonsense gets made.
Five kinds of not-knowing
| Type | Meaning | Example in this report |
|---|---|---|
| Risk | Outcomes and probabilities both known | Sampling error on correlation coefficients; computable from n |
| Uncertainty | Outcomes known, probabilities not | Whether 2024 patterns persist into 2026 |
| Ignorance | Outcome space itself incomplete | Influence channels nobody has thought to measure |
| Ambiguity | Terms admit several defensible readings | What "control" of a government means at all |
| Undecidability | No possible observation settles it | Whether the electorate has a single collective interest |
Classifying this report's open questions
Risk — quantifiable
The correlation in H2 (r = 0.195 / 0.267, n = 262 / 253) carries ordinary sampling error. With samples this size the estimates are reasonably precise; the refutation of H2 is not an artefact of small n. This is the best-behaved uncertainty in the report.
Uncertainty — not quantifiable, but bounded
Whether the 8.25:1 independent-expenditure ratio is stable is unknown. It has grown since 2010, but a single cycle cannot establish a trajectory, and we decline to project one.
Ignorance — the honest gap
The refutation of H2 pushes the question into genuine ignorance. We know money does not predict roll-call defection. We do not know what it does instead, and we cannot enumerate the mechanisms we have failed to consider. Agenda control (H5) is one candidate; listing it does not exhaust the space.
The tempting move after H2's refutation is to assert agenda control as the "real" mechanism. That would convert ignorance into a confident claim without measuring anything. Agenda control is a hypothesis awaiting data, not a fallback explanation.
Ambiguity — resolved by stipulation, not discovery
"Control of a government" has no unique meaning. The methodology stipulates authorisation-versus-finance. That stipulation is defensible and explicit, but it is a choice; other choices give other numbers. Readers who reject the stipulation should reject the resulting scores, and that is a legitimate disagreement rather than an error.
Undecidability — beyond evidence
Whether "the public interest" names a determinate object is not settled by more data. Arrow's theorem implies that no aggregation rule satisfies all reasonable fairness conditions simultaneously. Claims of the form "money works against the public interest" therefore rest on a normative commitment about which interests count, not on a measurement. This report does not resolve that and should not be read as having done so.
Confidence statement
| Claim | Confidence | Basis |
|---|---|---|
| IE exceeded direct contributions ~8:1 in 2024 | Very high | Direct arithmetic on complete filings |
| Political finance is highly concentrated | Very high | Complete-population measurement |
| PAC share does not predict roll-call voting | High | Pre-registered test, adequate n |
| Money influences legislative outcomes by some other route | Low | Plausible, unmeasured |
| Observed divergence constitutes democratic illegitimacy | Not an empirical claim | Normative; see Ethics |