When a Test Says You're Sick, How Worried Should You Be?

A Frightening Test Result

Imagine you are twelve-year-old Mira. You have a routine checkup, and the doctor runs a test for a rare illness that affects only 1 out of every 100 kids your age. A few days later the result comes back: positive. Your stomach drops. You assume you almost certainly have the disease.

But the doctor says, “Hold on. The test isn’t perfect.” It catches the disease 90% of the time when you really have it—that’s called the sensitivity. But it also wrongly says “positive” for 5% of healthy kids—that’s the false positive rate. So if 1% of kids actually have the disease, what are the odds that a positive result means you are truly sick?

The answer surprises most people: only about 15%. How can that be? The key is that the disease is very rare in the first place. When you do the math, you are far more likely to be a healthy kid who got a false alarm than a sick kid who tested positive correctly. This kind of reasoning is the heart of a powerful rule called Bayes’ Theorem.

The Clergyman Who Changed How We Think

Bayes’ Theorem is named after Thomas Bayes (c. 1701–1761), a British clergyman who worked out a clever way to flip probabilities around. His idea: if you want to know the probability that you have a disease given a positive test—written P(disease | positive)—you can figure it out by starting with the base rate of the disease and how strongly the disease predicts the test result.

At its core, Bayes’ Theorem is a formula:

P(H | E) = [P(H) × P(E | H)] / P(E)

Don’t let the letters scare you. H is your hypothesis (e.g., “I have the disease”). E is the evidence (e.g., “the test is positive”). P(H) is your prior probability—how likely you thought the disease was before seeing the test, maybe just 1%. P(E | H) is the likelihood: how probable the evidence is if the hypothesis is true (the sensitivity, 90%). And P(E) is the overall chance of getting a positive test no matter what. When you plug in Mira’s numbers, the probability she is sick given a positive test comes out to a surprisingly small 15%.

Bayes’ Theorem didn’t stay in the 1700s. In the twentieth century a group of philosophers called subjectivists argued that all our beliefs come in gradations of strength, and we should model them as probabilities. When you learn a new fact for certain—like a positive test result—a rational person should condition on that evidence. That simply means your new belief is your old conditional probability: P_new(H) = P_old(H | E). This is called simple conditioning, and it says you update your confidence by exactly the amount the math demands.

Learning When You’re Not 100% Sure

But real life often hands you evidence that isn’t rock-solid. You might see a blurry figure in a photo that could be your friend, but you aren’t certain. Your confidence in “that’s my friend” might shift from, say, 0.3 to 0.7—not all the way to 1.

The American philosopher Richard Jeffrey (1926–2002) showed how to handle such cases with what is now called Jeffrey conditioning (he preferred “probability kinematics”). Instead of setting the probability of the evidence E to 1, you move it to some new value q that is less than certain. Your belief in any hypothesis H then becomes a weighted average: P_new(H) = q × P_old(H | E) + (1 − q) × P_old(H | not E). It’s like blending two pictures: one where you assume E is true, and one where you assume it’s false, in proportion to how confident you are in E.

Jeffrey’s rule reduces to simple conditioning when q = 1, so it’s a more general tool for a messy world. Both forms of conditioning follow directly from the insight that your beliefs should be proportioned to your total evidence. That is, your confidence in a claim should rise or fall only as much as the new evidence warrants. Bayes’ Theorem guarantees that whenever you condition on new information, you update in exactly this careful way.

Measuring the “Oomph” of Evidence

If a positive test cranks your disease probability from 1% to 15%, you might say the test gave you a lot of reason to worry. But what if a different test moved a hypothesis from 50% to 90%? Which piece of evidence is stronger?

Subjectivists say evidence supports a hypothesis when conditioning on it raises the probability. This is incremental confirmation. Yet philosophers can’t agree on how to measure the size of that boost. Some use the simple difference: P(H | E) − P(H). By that measure, going from 50% to 90% (a 40-point jump) is stronger than going from 1% to 15% (a 14-point jump). Others prefer the probability ratio: P(H | E) / P(H). That tells a different story: 15%/1% = a 15‑fold increase, versus 90%/50% = only a 1.8‑fold increase. Now the rare‑disease test looks more powerful.

Which one is right? The disagreement runs deep. Many statisticians and philosophers put their faith in the likelihood ratio, which bypasses your prior altogether. It compares how well the hypothesis predicts the evidence versus how well its negation does: LR = P(E | H) / P(E | not H). For Mira’s test, the likelihood ratio is 0.90 / 0.05 = 18. This tells you that a positive result is 18 times more likely if you have the disease than if you don’t—a strong clue, but it still leaves you to combine that number with your prior to get a final probability.

Bayes’ Theorem can be written in several forms that highlight these different ways of thinking. Whether you prefer differences, ratios, or likelihoods, the theorem always says the same thing: your new belief is your old belief times the predictive power of the evidence.

Why This Old Idea Still Matters Today

You might never do the arithmetic by hand, but Bayes’ Theorem shapes how doctors interpret tests, how forensic scientists weigh DNA evidence, and how weather forecasts are updated. It also exposes a common mistake called the base‑rate fallacy: people panic at a positive test result because they forget how rare the condition was to begin with. Learning to think in terms of priors and likelihoods can protect you from that kind of emotional error—and help you ask smarter questions.

The subjectivist approach also raises a huge philosophical question: where do our prior probabilities come from? Two rational people with different priors can look at the same evidence and end up with different conclusions, and both can be obeying Bayes’ Theorem perfectly. Some thinkers argue that science can eventually force everyone’s beliefs to converge, but others worry that this subjectivity means our beliefs are never fully objective. That debate is still alive.

For now, the key lesson is that changing your mind isn’t guesswork. Bayes’ Theorem gives you a rule: start with what you already think, see how strongly the new information points toward or away from your hypothesis, and adjust—no more and no less. Next time you get surprising news, you’ll know how to weigh it.

Think about it

1. If a disease is extremely rare but a test is nearly perfect, a positive result might still be wrong most of the time. Could a jury in a criminal trial make a similar mistake if they ignore how rare the crime is?
2. Suppose two friends hear the same rumor. One thinks gossip is usually true, the other thinks it’s usually false. Can both update their beliefs rationally and still disagree? Should we expect them to eventually agree?
3. If a weather app says there is a 70% chance of rain and it stays sunny, was the app wrong, or did you just get unlucky? What would a good rule for updating your trust in the app look like?