🧪 This platform is in early beta. Features may change and you might encounter bugs. We appreciate your patience!
base_rate_fallacy
The base rate fallacy occurs when people ignore or underweight the prior probability (base rate) of an event when evaluating conditional probabilities. Instead, they focus on specific, often vivid information about the individual case. This is a violation of Bayesian reasoning, where the posterior probability must account for both the likelihood of the evidence given the hypothesis and the prior probability of the hypothesis itself.
A medical test for a rare disease (prevalence 1 in 10,000) has a 99% accuracy rate. A patient tests positive and believes they almost certainly have the disease. In reality, with 10,000 tests, roughly 100 false positives occur versus just 1 true positive, giving only about a 1% chance of actually having the disease.
An airport security algorithm flags a passenger as a potential threat with '95% accuracy.' The security officer treats this as near-certain danger. But if only 1 in 50,000 passengers is actually a threat, the vast majority of flagged passengers are false positives — innocent travelers caught by the algorithm.
A hiring manager uses a personality test the vendor claims is '90% accurate' at identifying high performers. She assumes a high-scoring candidate will almost certainly excel. But if only 10% of applicants are genuinely high performers, most high scorers are still average employees who happened to test well.
Binary (yes/no) questions an LLM must answer to identify this aspect:
Is a probability or likelihood estimate being made?
Type: binaryIs specific individual-case information being given more weight than general base rates?
Type: binaryWould incorporating the base rate significantly change the conclusion?
Type: binaryThe base rate fallacy occurs when people ignore or underweight the prior probability (base rate) of an event when evaluating conditional probabilities. Instead, they focus on specific, often vivid information about the individual case. This is a violation of Bayesian reasoning, where the posterior probability must account for both the likelihood of the evidence given the hypothesis and the prior probability of the hypothesis itself.
Specific, concrete information (the test result) feels more relevant than abstract statistical information (the disease prevalence). Humans are intuitively poor at integrating base rates into probabilistic judgments.
Always ask 'How common is this condition or event in the first place?' Use natural frequencies rather than percentages (e.g., '1 out of 101 positive tests is a true positive') to make base rates intuitive.
This fallacy is critical in medical screening programs, criminal forensics (DNA match probabilities), and airport security (flagging false positives among millions of travelers).
Forming worldview based on examples that come most easily to mind.
Drawing broad conclusions from limited, unrepresentative, or anecdotal evidence.
Probability-based belief revision using Bayes' theorem.
Diagnostic test accuracy varies when evaluated across different disease severity levels.
The tendency to draw strong conclusions from small samples, failing to recognize that small samples are more variable and less reliable than large ones.
Believing that small samples accurately represent the underlying population distribution.
The tendency to overestimate the accuracy of one's judgments, especially when available information is internally consistent, even if the information is limited or unreliable.
On average, people's friends have more friends than they do, due to sampling bias toward popular nodes.
Random observation of a process is more likely to catch long-duration events than short ones.
A model with higher accuracy can have worse predictive power than a less accurate one on imbalanced data.
Bayesian and frequentist approaches yield contradictory conclusions with large sample sizes.
Use these tools to detect, analyze, or train this aspect.