Probability Illusion

A few similar examples in a row make a rare event feel far more likely than it is. People then act as if the chance is higher than it really is.

This is a cognitive bias where sequential observations lead people to overestimate an event's probability, driven by a perceived reinforcement of pattern that isn't really there. The misestimation comes from inferential heuristics that overweight whatever evidence is most recent or salient, relative to the actual base rate.

A restaurant has a short line one morning, so a few people join assuming it must be popular. As the line grows, passersby assume the food is excellent purely because others are waiting — even though no one in line has actually eaten there yet, and the original line formed by pure coincidence.

In an equity market microstructure setting, a thinly traded asset receives three consecutive buy orders from distinct accounts. Algorithmic agents watching only the public order flow, with no visibility into the private signals behind those initial trades, run iterative Bayesian-like updates that treat each predecessor's action as independent evidence. Because signal diversity never gets computed and base-rate neglect goes uncorrected, posterior estimates for continued price appreciation inflate well past what the underlying fundamentals actually warrant. A hub-seeding actor who controls even one early node can trigger a self-reinforcing cascade this way, with prior compression amplifying the distortion as it deepens and counterevidence suppression keeping mean-reversion signals invisible to agents arriving late.

Seeing several similar cases makes people assume a trend has started. That mistaken belief is exactly what pulls others in behind it, making the trend look real.

Agents run iterative Bayesian-like updates in which predecessors' observed actions function as strong social signals; a structural asymmetry then emerges because early movers' choices end up weighting subsequent beliefs far more than they should. Limited private signal variance combined with informational opacity produces a weighting asymmetry that favors public actions over the private evidence actually driving them.

Ask for more independent proof before trusting the pattern. Checking how many genuine examples exist against the normal rate keeps the judgment honest.

Corroborating independent signals, plus posterior beliefs adjusted with explicit base-rate priors, correct for the inflation directly. Signal diversity metrics that downweight early public actions during updating help too.

Overconfidence in small samples; Ignoring base rate information; Perpetuation of initial error

An adversarial actor can seed a cascade by orchestrating a small cluster of early, highly visible adopters or endorsers, exploiting the structural asymmetry whereby initial public actions disproportionately weight subsequent beliefs—making a marginal claim appear to have overwhelming evidential support. By controlling the observation window and suppressing counterevidence, the actor can manufacture apparent consensus before independent private signals can correct the distortion. This is especially potent in partially observable networks where downstream agents cannot audit the provenance or independence of upstream signals.

Analysts and decision-makers should explicitly elicit and weight base-rate priors before updating on observed social signals, using structured Bayesian frameworks that penalize correlated evidence chains. Signal diversity audits—tracking the independence of upstream actors' information sources—can flag cascade-driven distortions before they propagate. Institutional protocols that require corroborating independent evidence (e.g., pre-registered replication, multi-source triangulation) directly interrupt the sequential observational updating loop that sustains the illusion.