Science involves uncertainty because we are always working with incomplete information about an enormously complex world. Every measurement has limits, every sample is smaller than the whole population, and every model simplifies reality to make it understandable. Uncertainty isn’t a flaw in the scientific process. It’s a built-in feature that reflects honesty about what we know, what we don’t, and how confident we can be in our conclusions.
Two Fundamentally Different Kinds of Uncertainty
Not all uncertainty is the same, and understanding the distinction helps clarify why it can never be fully eliminated. Scientists recognize two broad categories. The first is uncertainty from incomplete knowledge. We simply don’t have all the information yet. A doctor examining a patient with vague symptoms faces this kind of uncertainty: the answer exists, but the available clues aren’t enough to pin it down. In principle, more data, better instruments, or deeper investigation could reduce this type of uncertainty over time.
The second kind comes from genuine randomness in nature. Even if you knew everything about current conditions, some outcomes are inherently unpredictable. Flip a coin a thousand times and you can predict roughly 500 heads, but you cannot predict the result of any single flip. At the cellular level, the number of proteins in a cell fluctuates randomly over time due to the unpredictable timing of individual molecular events. Weather, radioactive decay, and genetic mutations all contain this irreducible randomness. No amount of additional knowledge eliminates it.
Measurement Is Never Perfect
Every scientific instrument introduces some degree of error, and these errors come in two flavors. Random errors are the unpredictable fluctuations that happen even when everything is working correctly. Electronic noise in a circuit, slight air current changes in a lab, or tiny vibrations in equipment all cause readings to scatter slightly above and below the true value. You can reduce random error by taking many measurements and averaging them, but you can never eliminate it entirely.
Systematic errors are more insidious because they push every measurement in the same direction. A thermometer that doesn’t make proper contact with the substance it’s measuring will consistently read too low. A scale that isn’t zeroed correctly will add the same offset to every reading. These errors don’t cancel out with repeated measurements, and they can go undetected for a long time. Scientists use calibration, cross-checking with independent methods, and peer review to catch systematic errors, but the possibility that one is lurking unnoticed is itself a source of uncertainty.
Samples Can’t Capture Everything
Scientists almost never measure an entire population. Instead, they study a sample and use statistics to estimate what’s true for the broader group. This is where tools like confidence intervals come in. A confidence interval gives a range of values likely to contain the true answer, rather than a single number. A 95% confidence interval means that if you repeated the same study many times with new random samples, about 95% of those intervals would capture the true value. The remaining 5% would not.
The width of that interval reflects how precise the estimate is. Small studies produce wide intervals, meaning more uncertainty. Larger studies narrow the range. But no study, regardless of size, produces a confidence interval of zero width. There is always some gap between what a sample tells you and what is true for the whole population.
This sampling variability also explains something that confuses many people: why two well-conducted studies on the same question can produce different results. In a simulation where researchers drew small samples of 10 individuals from two populations with a known, real difference, the observed difference between samples ranged wildly, from 1.46 standard deviations in one draw to nearly zero in another. With small sample sizes, which remain common in published research, results bounce around considerably from study to study even when there is no error in the methods at all.
Statistical Thresholds Are Conventions, Not Certainties
You may have heard that a result is “statistically significant” when it passes a threshold called p = 0.05. This convention, originally proposed as a rough guide rather than a rigid rule, means roughly that there’s a 5% chance of seeing a result this extreme if nothing real is going on. But a 5% false alarm rate is not trivial, especially across thousands of studies published every year. Some researchers have proposed lowering the threshold to 0.005 or 0.01 to reduce false positives.
The American Statistical Association issued a formal statement in 2016 cautioning against mechanical reliance on p-values, noting that they are widely misinterpreted. A p-value does not tell you the probability that a finding is true. It does not tell you the size of an effect or whether it matters in the real world. Many methodologists now argue that science should focus less on binary “significant or not” declarations and more on reporting the size of effects and the precision of estimates. This shift reflects a broader recognition that certainty is a spectrum, not a switch.
Models Simplify a Complex Reality
When scientists build mathematical models to predict things like disease spread, economic trends, or climate change, they face a choice. Deterministic models treat the system as if every outcome follows predictably from the inputs: if you know the starting conditions, you know the result. These models are useful but can be misleading because real systems contain random variation that deterministic equations ignore.
Stochastic models incorporate randomness directly. Rather than producing a single predicted outcome, they generate a range of possible outcomes with different probabilities. Weather forecasting is a familiar example: meteorologists run many slightly different versions of their model to see how sensitive the forecast is to small changes in initial conditions. When all the versions agree, confidence is high. When they diverge, uncertainty is high, and the forecast reflects that with wider probability ranges.
Climate science has developed one of the most explicit systems for communicating this kind of uncertainty. The Intergovernmental Panel on Climate Change uses a structured vocabulary that maps words to probability ranges. “Virtually certain” means 99 to 100% probability. “Extremely likely” means 95 to 100%. “Exceptionally unlikely” means 0 to 1%. This scale allows scientists to be precise about how uncertain they are, which sounds paradoxical but is one of the most useful things science does.
The Replication Crisis Revealed Hidden Problems
Starting around 2010, researchers in psychology, biomedicine, and other fields began systematically re-running classic studies and finding that many results didn’t hold up. This “replication crisis” exposed several factors that had been inflating false confidence in published findings. Publication bias meant that studies with dramatic, positive results were far more likely to be published than studies finding nothing, creating a skewed picture of the evidence. Lack of transparency in methods made it difficult to evaluate or repeat experiments. And the misuse of statistics, including practices like running many analyses and reporting only the one that crossed the significance threshold, artificially boosted the appearance of certainty.
But even after correcting all of those problems, replication would still not yield identical results every time. Random variation guarantees that two well-run studies will produce somewhat different numbers. The realistic expectation is comparable results, not identical ones. Recognizing this has pushed science toward valuing the overall pattern across many studies rather than treating any single result as definitive.
Uncertainty in Medicine
Perhaps nowhere is uncertainty more personally felt than in a doctor’s office. Patients often present with symptoms that overlap across many conditions and change over time. At least 1 in 20 outpatients experiences a diagnostic error, meaning a missed, delayed, or incorrect diagnosis, each year. This isn’t primarily due to incompetence. It reflects the genuine difficulty of applying population-level knowledge to a single individual whose biology, history, and circumstances are unique.
Researchers have identified several layers of medical uncertainty: technical uncertainty from limited data about how diseases progress, conceptual uncertainty in translating general research findings to a specific patient, and personal uncertainty from the complex dynamics between patient and clinician. Doctors manage this by ordering additional tests, scheduling follow-ups, consulting specialists, or sometimes simply waiting to see how symptoms evolve over time. The “test of time” is itself a diagnostic tool, one that explicitly uses uncertainty as a reason to gather more information before acting.
Why Communicating Uncertainty Matters
There’s a common worry that admitting uncertainty will erode public trust in science. The reality is more nuanced. Research published in PNAS Nexus found that the effect of communicating uncertainty depends on whether the evidence aligns with what people already believe. When scientific findings matched a person’s existing views, adding uncertainty language (“we’re not completely sure”) actually made people trust the information less, as though the hedging seemed unnecessary. But when findings contradicted someone’s prior beliefs, communicating uncertainty increased trust, making the information feel more honest and credible.
Across the full spectrum of prior beliefs, communicating uncertainty caused trust ratings to converge rather than diverge. People at opposite ends of a debate moved closer together in their trust levels when uncertainty was openly acknowledged. There was no evidence that transparency about uncertainty contributed to polarization. If anything, it served as a signal of intellectual honesty that people on all sides could recognize.
This is ultimately why uncertainty exists in science and why it’s worth understanding. It isn’t a weakness to be hidden. It’s the mechanism by which science stays self-correcting, prevents overconfidence, and earns the trust that comes from saying clearly: here is what we know, here is how sure we are, and here is what we’re still figuring out.