Probability in science is a way of measuring how likely something is to happen, expressed as a number between 0 (impossible) and 1 (certain). Scientists use probability to quantify uncertainty, whether they’re predicting the chance a patient carries a genetic trait, determining if a drug actually works, or describing the behavior of subatomic particles. It’s not just a math concept borrowed from textbooks. Probability is woven into how scientific knowledge is built, tested, and communicated.
Two Core Meanings of Probability
For centuries, the word “probability” quietly carried two distinct meanings that weren’t clearly separated until relatively recently. The first is physical probability: the actual frequency of an event when you repeat an experiment many times. Flip a fair coin 10,000 times and you’ll get heads close to 50% of the time. The second is epistemic probability: a measure of how confident you are about something given what you currently know. A doctor estimating a 70% chance that a patient has a particular condition is using this second type, even without running the scenario thousands of times.
These two meanings show up across every scientific discipline, sometimes blending together. A physicist calculating the likelihood of a particle’s position is working with physical probability baked into the laws of nature. An epidemiologist estimating the risk of a disease outbreak is working more with epistemic probability, updating estimates as new data arrives.
How Scientists Test Ideas With Probability
Most scientific studies rely on a tool called the p-value to judge whether their results are meaningful or just a fluke. A p-value answers a specific question: if there were truly no effect (no difference between a drug and a placebo, for example), how likely would it be to see results at least this extreme? A p-value close to 0 means the observed results would be very unusual under the “no effect” assumption. A p-value close to 1 suggests the results are easily explained by chance alone.
For decades, researchers treated p = 0.05 as a magic cutoff. Results below that line were labeled “statistically significant,” and everything above it was dismissed. The American Statistical Association issued a formal statement pushing back on this practice. Their key points are worth knowing: a p-value does not tell you the probability that your hypothesis is true, it does not measure how large or important an effect is, and decisions should never hinge on whether a number clears a single threshold. A p-value near 0.05, taken by itself, offers only weak evidence. Treating it as a bright line leads to distorted science, including a well-documented problem called p-hacking, where researchers run multiple analyses and selectively report only the ones that cross 0.05.
Frequentist vs. Bayesian Approaches
Scientists broadly fall into two camps when they use probability, and the distinction matters because it shapes how studies are designed and how results are interpreted.
The frequentist approach treats probability as the long-run frequency of events. It asks: if I repeated this experiment an infinite number of times, how often would I get this result? Frequentist methods don’t assign a probability to a hypothesis itself. Instead, they calculate confidence intervals and p-values, always framing probability in terms of the data rather than the underlying truth.
The Bayesian approach works differently. It starts with a prior probability, an initial estimate of how likely something is before you collect new data. Then, as evidence comes in, that estimate gets updated to produce a posterior probability. This approach directly answers questions like “given the evidence, what’s the probability this treatment works?” Bayesian methods require you to specify what you believed before the experiment, which some scientists see as a strength (it forces transparency) and others see as a weakness (it introduces subjectivity).
Neither approach is universally better. Frequentist methods dominate clinical trials and regulatory science. Bayesian methods are increasingly used in fields like genetics, artificial intelligence, and astrophysics, where prior knowledge is abundant and updating beliefs with new data is natural.
Probability in Genetics
Gregor Mendel’s pea plant experiments in the 1860s were, at their core, probability problems. When two parents each carry one copy of a dominant gene and one copy of a recessive gene, the math predicts that 3 out of 4 offspring will display the dominant trait and 1 out of 4 will display the recessive trait. This 3:1 ratio is one of the most famous predictions in biology, and it holds remarkably well across thousands of observed cases.
Modern genetics extends these simple ratios into far more complex territory. Genetic counselors use probability to estimate a couple’s chance of having a child with a specific inherited condition. Genome-wide studies calculate the probability that a particular gene variant is associated with a disease, often requiring p-values far more stringent than 0.05 (sometimes below 0.00000005) because so many comparisons are being tested simultaneously.
Probability in Quantum Physics
At the scale of atoms and subatomic particles, probability isn’t just a tool for handling incomplete information. It appears to be fundamental to how nature works. In 1926, physicist Max Born proposed that the wave function in quantum mechanics doesn’t describe a particle’s definite position. Instead, the square of the wave’s amplitude at any point gives the probability of finding the particle there. This idea, now called the Born rule, earned him a Nobel Prize and remains one of the pillars of modern physics.
This was a radical departure from classical physics, where everything was, in principle, predictable if you had enough information. In quantum mechanics, even with perfect knowledge of a system’s state, you can only predict probabilities of outcomes. Measure the spin of an electron and you might get “up” 50% of the time and “down” 50% of the time, with no hidden factor determining which one you’ll get on any single measurement. The randomness isn’t due to ignorance. As far as physicists can tell, it’s built into the fabric of reality.
Why Sample Size Matters
One of the most practical ideas connecting probability to real science is the law of large numbers. It states that as you repeat an experiment more times, your observed results converge toward the true underlying probability. Think of it this way: if you flip a coin 10 times, getting 7 heads wouldn’t be shocking. But if you flip it 10,000 times and get 7,000 heads, something is almost certainly wrong with the coin.
This principle drives sample size requirements in research. Comparing the average height of 3 randomly chosen students from a class of 100 gives a rough estimate at best. Measuring 10 students gets you closer. Measuring 99 gives you an answer nearly identical to the true class average. Small samples produce wildly variable estimates, which is why a single small study rarely settles a scientific question. Larger samples reduce the noise and let the real signal emerge.
How Probability Communicates Risk
Outside the lab, probability becomes a communication tool, and how it’s framed changes how people understand it.
In medicine, risk can be expressed in absolute or relative terms. If a treatment reduces the chance of a heart attack from 4% to 3%, the absolute risk reduction is 1 percentage point. But the relative risk reduction is 25%, because 3 is 25% less than 4. Both numbers are accurate, but they feel very different. Studies show that absolute risk is more effective for helping patients actually understand their options. That’s why guidelines increasingly recommend presenting both measures side by side. A related concept, the number needed to treat, flips absolute risk into an even more intuitive form: it tells you how many people need to receive a treatment for one person to benefit.
Climate science offers another example of structured probability language. The Intergovernmental Panel on Climate Change uses a standardized scale where “virtually certain” means 99 to 100% probability, while “exceptionally unlikely” means 0 to 1%. These terms translate complex model outputs into language that policymakers can use, attaching specific probability ranges to what might otherwise be vague words like “likely” or “unlikely.”
Probability vs. Certainty
Science rarely deals in certainties, and probability is the language it uses to be honest about that. A weather forecast giving a 30% chance of rain isn’t wrong if it rains. It means that in similar atmospheric conditions, rain occurs about 3 times out of 10. A clinical trial showing a treatment works with p = 0.01 isn’t guaranteeing the treatment works for everyone. It’s saying the observed benefit would be very surprising if the treatment did nothing at all.
This is often where public frustration with science originates. People want yes-or-no answers, but probability gives them ranges and likelihoods. Understanding that scientific claims are probabilistic, not absolute, is one of the most useful things you can take away from how science actually operates. The strength of a scientific finding isn’t that it eliminates uncertainty. It’s that it quantifies uncertainty precisely enough to act on.