Mathwords logoMathwords

False Positive Paradox — Definition, Formula & Example

The false positive paradox is the surprising fact that a very accurate test can produce positive results that are mostly wrong. This happens when the thing being tested for is rare.

Let DD be an event with prior probability P(D)=pP(D) = p, and let a test have sensitivity P(+∣D)=sP(+ \mid D) = s and false positive rate P(+∣Dc)=fP(+ \mid D^c) = f. The paradox is that P(D∣+)P(D \mid +) can be small even when ss is close to 1 and ff is close to 0. In fact, P(D∣+)<12P(D \mid +) < \tfrac{1}{2} exactly when sp<f(1−p)sp < f(1-p), which happens whenever the base rate pp is small compared with ff.

Key Formula

P(D∣+)P(Dc∣+)=p1−p⋅sf\frac{P(D \mid +)}{P(D^c \mid +)} = \frac{p}{1-p} \cdot \frac{s}{f}
Where:
  • pp = Base rate, the prior probability that the condition is present
  • ss = Sensitivity, the probability of a positive result when the condition is present
  • ff = False positive rate, the probability of a positive result when the condition is absent
  • sf\frac{s}{f} = Likelihood ratio, which says how strongly a positive result favors the condition

Worked Example

Problem: A bank's fraud detector flags 99% of fraudulent card transactions and wrongly flags 1% of legitimate ones. Group A is every transaction the bank processes, and 1 in 10,000 of them is fraud. Group B is only transactions that customers have reported as suspicious, and 1 in 10 of them is fraud. In each group, what is the probability that a flagged transaction is really fraud?
Step 1: Find the likelihood ratio. It is the same detector in both groups, so a flag multiplies the odds of fraud by the same factor each time.
sf=0.990.01=99\frac{s}{f} = \frac{0.99}{0.01} = 99
Step 2: Group A has prior odds of 1 to 9,999. Multiply by 99, then convert the odds back to a probability.
19999⋅99=1101⟹P(fraud∣flag)=11+101=1102≈0.0098\frac{1}{9999} \cdot 99 = \frac{1}{101} \quad\Longrightarrow\quad P(\text{fraud} \mid \text{flag}) = \frac{1}{1 + 101} = \frac{1}{102} \approx 0.0098
Step 3: Group B has prior odds of 1 to 9. The same factor of 99 now produces strong evidence.
19⋅99=11⟹P(fraud∣flag)=1111+1=1112≈0.917\frac{1}{9} \cdot 99 = 11 \quad\Longrightarrow\quad P(\text{fraud} \mid \text{flag}) = \frac{11}{11 + 1} = \frac{11}{12} \approx 0.917
Answer: In Group A, a flagged transaction is fraud only about 1% of the time. In Group B, it is fraud about 92% of the time. The detector is identical in both groups. Only the base rate changed, and that alone makes a flag nearly meaningless in one group and highly reliable in the other.

Why It Matters

AP Statistics questions on conditional probability can give a test's accuracy rates and ask for the probability of a condition given a positive result. The paradox is also why doctors confirm a positive screening result with a second, independent test, and why screening is usually aimed at higher-risk groups instead of the whole population.

Common Mistakes

Mistake: Reading "the test catches 99% of cases" as "99% of positive results are correct."
Correction: The first statement is P(+∣D)P(+ \mid D) and the second is P(D∣+)P(D \mid +). These are different conditional probabilities. To get P(D∣+)P(D \mid +) you must use the base rate, through Bayes' theorem or the odds form above.

Related Terms