# Statistical Impossibility Arguments The form: compute the probability of some observed pattern under a deliberately naive model, obtain a number like one in a quadrillion, and present it as proof that the pattern was engineered. The arithmetic is usually correct. The model is the problem, and it is always the problem in the same three ways. ## The three failures **Independence assumed where it does not hold.** The most common. Multiplying probabilities of related events treats them as independent, which collapses the result toward zero. *Election example:* the probability that a candidate leads in four states simultaneously, computed as the product of four separate probabilities, is astronomically small. But state outcomes are heavily correlated — they share national conditions, demographic patterns, and polling error (see [[Poll Aggregation]]). Correctly modeled, the joint event is ordinary. The impossibility is manufactured by the independence assumption alone. **The wrong reference class.** Computing the probability of *this exact outcome* rather than *an outcome at least this unusual.* Every specific outcome is astronomically improbable — that is what "specific" means. Any particular shuffle of a deck has probability about 1 in 8×10⁶⁷, and you have just produced one. The relevant probability is always of a *class* of outcomes, and choosing the class after seeing the data is the Texas sharpshooter fallacy: firing at the barn, then painting the target around the holes. **Base rates ignored.** The probability of the evidence given fraud is presented without the probability of the evidence given no fraud, and without the prior probability of fraud at the relevant scale. This is the prosecutor's fallacy — inverting a conditional — and it is the single most consequential probabilistic error in public argument. ## Why the big number persuades A very small probability produces an immediate intuitive sense that something must be wrong, and that intuition is largely correct in ordinary life: genuinely improbable things usually do indicate a flawed assumption. The move exploits this by directing the conclusion. **When a model assigns near-zero probability to something that observably happened, the correct inference is that the model is wrong.** The impossibility argument instead concludes that reality is wrong. It is the right instinct pointed backwards, which is why it survives contact with numerate audiences who have not slowed down. ## Where it shows up beyond elections **Anti-evolution argument.** Probability-of-a-protein calculations assuming random assembly, ignoring selection, intermediates, and parallel trials. The number is enormous and the model omits the mechanism. **Vaccine adverse-event claims.** Computing the probability of an event cluster while ignoring the population size and the base rate of the event. **Conspiracy pattern-matching generally.** Any set of coincidences is astronomically improbable once you specify it after the fact and search a large space of possible patterns. See [[QAnon]] for the industrial-scale version. ## The diagnostic Three questions dismantle almost every instance: 1. **What model generated this probability, and does it assume independence?** 2. **Was the pattern specified before or after looking at the data?** 3. **What is the probability of this evidence under the innocent explanation?** — the question the argument is constructed to prevent you from asking. The third is decisive. An impossibility argument is a likelihood claim presented without its denominator, and demanding the denominator ends it. ## Related [[Benford's Law and Election-Fraud Claims]] · [[Curve-Fitting and Cherry-Picked Windows]] · [[Pseudo-Precision as Rhetoric]] · [[Election Denialism]] · [[QAnon]] · [[Manufacturing Doubt]] · [[Mathematics and Power Hub]]