False Positive Paradox Calculator

How to use the false positive paradox calculator

1. Enter prevalence. This is the percentage of the tested group with the event or condition before testing, sometimes called the base rate.
2. Enter sensitivity. For actual positive cases, enter how often the test catches them.
3. Enter specificity. For actual negative cases, enter how often the test correctly rejects them.
4. Set the population size. Use 10,000 or 100,000 to make the expected counts easy to read. This changes the counts, not PPV.
5. Read the PPV first. Then compare true and false positives in the bar and 2 × 2 outcome table. Switch the chart to the 0–5% range when studying rare events.

The calculator updates automatically as you type. Use the presets to compare cases with exactly the same test characteristics but different base rates, or export the expected counts as a CSV for your notes.

What is the false positive paradox?

A test may identify nearly all genuine cases and reject nearly all non-cases, yet most of its positive alerts can still be wrong. This happens when the thing being detected is rare. The large group of actual negatives supplies enough opportunities for false alarms to outnumber the relatively small group of correct detections.

For example, imagine screening 10,000 cases when only 1% truly have the target condition. A test with 99% sensitivity and 99% specificity finds 99 true positives, but it also marks 1% of the 9,900 actual negatives as positive: another 99 alerts. A positive result is therefore right only half the time under these assumptions.

The word paradox describes a surprising result, not a mathematical contradiction or evidence that the test has failed. The right question is not simply how frequently the test is correct in each actual-state group. It is how to interpret the group of positive results after accounting for prevalence.

Positive predictive value: formula and calculation

Positive predictive value (PPV) is the probability that an item truly belongs to the positive class given that the test reports positive. It is a conditional probability based on Bayes' theorem.

Let p be prevalence, Se sensitivity and Sp specificity. The numerator, Se × p, is the probability of a true-positive outcome. The second term in the denominator, (1 − Sp) × (1 − p), is the probability of a false-positive outcome. Their sum is the probability of any positive test.

The same formula can be written as P(actual positive | +) = P(+ | actual positive) × P(actual positive) / P(+). This is different from the more general inputs accepted by the site's Bayes Theorem Calculator, which handles arbitrary Bayes problems.

Other metrics shown in the result

When a denominator is zero, its corresponding predictive value or ratio is undefined. The calculator tells you that rather than showing an invalid percentage.

Turn probabilities into expected counts

For a population of size N, the expected number actually positive is N × p and actually negative is N × (1 − p). Apply sensitivity and specificity within those two groups:

These are expected outcome counts. A result such as 4.75 means the mathematical expectation is 4.75 across comparable populations, not that a literal person or transaction has been split into fractions.

At what prevalence do false positives overtake true positives?

The break-even point is where TP and FP are equal, so PPV is 50% whenever positives occur. Rearranging the two terms gives:

When sensitivity is 99% and specificity is 99%, break-even prevalence is 1%. Below that rate, the false-positive count exceeds the true-positive count. If sensitivity is zero and specificity is 100%, no positive results occur at any prevalence, so a 50% PPV break-even point does not exist.

Worked examples: from a 50% PPV to one in eleven

Example 1: A 99% sensitive and 99% specific test

Consider 10,000 tests in a population with 1% prevalence. The condition is present in 100 cases and absent in 9,900 cases. With 99% sensitivity, 99 of the 100 actual positives are flagged; one is missed. With 99% specificity, 9,801 actual negatives are correctly cleared and the remaining 99 receive a false-positive result.

There are 198 positive results. Only 99 are genuine, so PPV = 99 ÷ 198 = 50%. The other 50% are false alarms. Sensitivity and specificity are still both 99%; the low base rate explains the apparently surprising answer.

Example 2: Make the event ten times rarer

Now take 100,000 cases, of which only 0.1% are actually positive. Keep sensitivity and specificity at 99%. There are 100 actual positives, yielding 99 true positives and one false negative. Of the 99,900 actual negatives, 999 are flagged as positive and 98,901 are correctly cleared.

PPV = 99 ÷ (99 + 999) = 9.0164%. Approximately 90.9836% of positive alerts are false. There are about 10.09 false alarms per genuine detection. Nothing about the test's sensitivity or specificity changed; prevalence did.

Example 3: A more common event

With a 10% prevalence, 95% sensitivity, 95% specificity and 10,000 cases, the test produces 950 true positives, 450 false positives, 50 false negatives and 8,550 true negatives. PPV is 950 ÷ 1,400 = 67.8571%. In this case true positives outnumber false alerts.

Sensitivity vs PPV, and specificity vs NPV

Sensitivity looks at people or events that actually are positive and asks what percentage the test finds. PPV begins with positive results and asks what percentage are genuine. Reversing the condition changes the calculation.

Specificity and NPV have the same difference in perspective. Specificity is P(test negative | actual negative). NPV is P(actual negative | test negative). A test can have high specificity without its positive results having high PPV, and high sensitivity does not guarantee the reverse.

For a closer look at these definitions, see the sensitivity and specificity guide. If you already have TP, FP, FN and TN counts and want to calculate test performance from them, use the Sensitivity and Specificity Calculator.

Why the base rate changes the meaning of a positive result

The base rate is information available before observing a test outcome. In a screening problem it is the prevalence within the population being screened. Applying a false-positive rate to a very large negative group may generate more alerts than applying sensitivity to a very small positive group.

Reading a test result without accounting for prevalence is often described as base-rate neglect or the base-rate fallacy. The false positive paradox is the numerical result you may encounter when you ignore that information. You can explore the underlying probability rule in the conditional probability guide or in the site's Conditional Probability Calculator.

When you lower prevalence in the chart, watch PPV fall for an informative but imperfect test. Better specificity sharply reduces false alarms among a large actual-negative group. Increasing population size, by contrast, multiplies all four expected outcome counts while preserving the PPV ratio.

Base rate must describe the people or items actually tested. A referral clinic, general-population screening program and already-filtered dataset may have quite different prevalence. The model assumes all cases share the rates you enter; it does not model uncertainty, dependence between tests, or changes in performance across subgroups.

The false positive paradox beyond medical testing

Exactly the same reasoning applies to a spam filter. Suppose 2% of incoming messages are spam, the filter catches 98% of spam, and it correctly lets through 99% of legitimate messages. Out of 10,000 hypothetical messages, 196 real spam messages are flagged but another 98 legitimate messages are mistakenly flagged. Its precision for spam alerts is 196 ÷ 294, or approximately 66.67%.

A fraud-monitoring system may flag very few real fraudulent transactions even when it performs well within each class, because legitimate transactions vastly outnumber fraudulent ones. A security alarm can face the same problem when actual threats are uncommon. These examples are illustrations, not estimates of real-world filter or detection performance.

In machine-learning terminology, the share of reported positives that are correct is often called precision. For a general classification-metrics workflow built from outcome counts, use the Confusion Matrix Calculator.

Common mistakes when interpreting positive results

• “99% sensitivity means a 99% chance my positive result is correct.” No. Sensitivity conditions on the actual state, while PPV conditions on a positive result and must incorporate prevalence.
• “False positive rate equals the share of positive results that are false.” FPR uses all actual negatives in its denominator. False-positive share uses all positive test results. They can differ dramatically.
• “The test is 99% accurate.” A single accuracy percentage hides class imbalance. Sensitivity, specificity and base rate give a more informative account of positive alerts.
• “More testing makes a positive more trustworthy.” Testing more cases increases expected alerts but does not change PPV when all input rates remain the same.
• “A second positive automatically squares the error rate.” Repeated tests may be dependent. Without a justified model of that dependence, simple multiplication can exaggerate certainty.

Related calculators and learning guides

Each link below leads to an existing page with its own purpose. For the broader foundation, return to the Statistics and Probability main category or browse all statistics calculators.

Method: standard Bayes' theorem for two mutually exclusive actual states and two test outcomes. Outputs use the point estimates you enter and do not express confidence intervals for those estimates.