Base Rate Fallacy
What you'll learn: How ignoring the underlying frequency of events (base rates) can lead to wildly incorrect probability judgments, especially for rare occurrences.
The Core Problem
The base rate fallacy happens when you focus too much on specific new evidence and ignore how common or rare something is in the first place. You've already learned Bayes' Theorem, which correctly combines prior probabilities (base rates) with new evidence. The base rate fallacy is what happens when people skip that first part.
A Classic Example
Imagine a disease that affects 1 in 10,000 people (base rate = 0.01%). A test for this disease is 99% accurate. You test positive. Should you panic?
Many people think: "99% accurate test + positive result = I'm 99% likely to have the disease!"
Wrong. This ignores the base rate.
Using Bayes' Theorem properly: because the disease is so rare, even with a positive test, you're far more likely to be a false positive than to actually have the disease. With these numbers, you'd only have about a 1% chance of truly being sick—not 99%!
Why This Matters in Data Science
When building models for rare events (fraud detection, disease screening, equipment failure), the base rate is crucial:
- A model that's "95% accurate" might simply predict "no fraud" every single time if fraud only occurs 2% of the time
- Always consider P(event) alongside P(positive test | event)
Key Takeaway: Never evaluate probabilities from new evidence alone—always anchor your reasoning in the base rate (prior probability) of what you're trying to detect, especially when dealing with rare events.