Why Are False Positives Common? Base Rates Explain

Why are false positives common?

A false positive occurs when a test reports a positive result even though the condition, signal, or event being tested for is absent. This can happen in medical screening, spam filtering, fraud detection, quality control, security alarms, and scientific research. A false positive is not necessarily evidence that a test is badly designed. It may arise because the target event is rare, because the test has imperfect specificity, or because many opportunities exist for an alert to occur.

The central idea is the base rate: how common the condition is before the test result is known. When a condition is uncommon, even a small fraction of incorrect positive results can outnumber the correctly identified cases. This is why a positive result does not automatically mean that the condition is likely. The result must be interpreted alongside the starting rate and the test’s performance.

Accuracy has more than one meaning

Two common measures describe different parts of a binary test. Sensitivity is the proportion of people who truly have the condition and receive a positive result. Specificity is the proportion of people who do not have the condition and receive a negative result. The complement of specificity is the false-positive rate: the proportion of unaffected people who nevertheless receive a positive result.

A test can have high sensitivity and high specificity while still producing many false positives in a low-prevalence population. This is not a contradiction. Sensitivity describes performance among affected people, and specificity describes performance among unaffected people. Neither measure alone tells us the probability that a person has the condition after receiving a positive result.

That after-test probability is often called the positive predictive value. It asks: among all positive results, what proportion are true positives? Positive predictive value changes when the base rate changes, even if the test itself has not changed. This distinction between test characteristics and result interpretation is an application of conditional probability and Bayes’ theorem.

A synthetic screening example

Consider this clearly synthetic illustration. Imagine a screening program with a hypothetical group of exactly one hundred thousand people. Suppose the condition has a synthetic base rate of one tenth of one percent, meaning that one hundred illustrative people in this group truly have it. These figures are invented for explanation and do not represent a real disease, test, or population.

Now suppose the hypothetical test identifies ninety-five percent of affected people, so it gives ninety-five illustrative true positives and misses five illustrative cases. Suppose it also has ninety-nine percent specificity. That means one percent of the ninety-nine thousand illustrative people without the condition receive false-positive results. One percent of that unaffected group produces nine hundred ninety synthetic false positives.

The total number of positive results is therefore one thousand eighty-five in this illustrative calculation: ninety-five true positives plus nine hundred ninety false positives. The positive predictive value is ninety-five divided by one thousand eighty-five, or about eight and eight-tenths percent. In this synthetic scenario, a positive result is more often a false positive than a true positive, despite the test’s high sensitivity and specificity.

The arithmetic is not a claim about any real screening program. It demonstrates the effect of unequal group sizes. The unaffected group is much larger than the affected group, so a modest false-positive rate applied to the large group creates many more incorrect positives than the high sensitivity creates correct positives among the small group.

The same test can give a different answer

Keep the test performance in the previous illustration unchanged, but imagine a second clearly synthetic population in which the condition’s base rate is ten percent. This is another invented example, not an estimate of a real population. In a hypothetical group of one thousand people, one hundred would have the condition, ninety-five would be synthetic true positives, and five would be missed.

The remaining nine hundred illustrative people would not have the condition. With the same one percent false-positive rate, nine would receive false-positive results. The total would be one hundred four positive results, of which ninety-five would be true positives. The positive predictive value would be about ninety-one percent.

Nothing about the test changed in these two illustrations. Only the composition of the tested population changed. In the low-base-rate setting, false positives dominated the positive results. In the higher-base-rate setting, true positives dominated. This is the base-rate effect.

The practical implication is that a test result should be interpreted for the population and situation in which it is used. A performance claim from a study in a high-risk group may not transfer directly to routine screening of a lower-risk group. The same warning applies outside health care: a fraud detector used on a population with very few fraudulent transactions can generate mostly false alerts even when its error rate is small.

Why repeated testing needs care

People sometimes respond to an unexpected positive result by repeating the same test. A follow-up test can be useful, but its meaning depends on how the tests are related. If the errors are independent and the follow-up test has comparable performance, two positive results can provide stronger evidence than one. However, real tests may share the same source of error, sample problem, measurement limitation, or decision threshold.

For example, two alerts generated from the same flawed data or the same sensor may not be independent pieces of evidence. Repeating a procedure without changing the source of error can repeat the false positive. A confirmatory test based on a different method may provide more informative evidence than simply running the same test again.

Repeated testing also creates a multiple-testing problem. When many tests, screens, comparisons, or alerts are performed, even a small false-positive rate can produce some positives by chance. This is especially important in scientific research and automated monitoring. The more opportunities there are for an alert, the more carefully the process must control and interpret false discoveries.

Reading a positive result responsibly

Start by asking what population was tested and what the condition’s approximate base rate is in that population. Then ask how sensitivity and specificity were estimated, whether they apply to the current setting, and whether the test threshold or data quality is comparable. These questions help distinguish a general accuracy claim from the probability relevant to the individual result.

It is also useful to separate three statements that are often confused. “The condition is present” and “the test is positive” are different events. “The test is positive given that the condition is present” describes sensitivity. “The condition is present given that the test is positive” describes the interpretation people usually want. Conditional probability changes when the order of those statements changes.

A clear explanation should report counts or probabilities for both true positives and false positives whenever possible. Natural frequencies, such as the number of positive results out of a specified synthetic group, can be easier to understand than isolated percentages. They make the base-rate effect visible and show what the positive results are actually made of.

Practical checklist

• Identify the base rate in the population being tested.
• Distinguish sensitivity, specificity, and positive predictive value.
• Ask whether the reported test performance applies to this setting.
• Look for the expected counts of true positives and false positives.
• Treat repeated results as dependent unless independence is justified.
• Use confirmatory evidence when the cost of a false positive is substantial.